Obsidian/Zotero/001_artiklid/intelligenttransportsystemsesafety.md

---

category: literaturenote citekey: intelligenttransportsystemsesafety title: Intelligent transport systems - eSafety - eCall for automated and autonomous vehicles authors: "" doi: 10.3403/30387362u url: "https://doi.org/10.3403/30387362u" zotero_key: QASDQKKR zotero_storage: ZT2U8C5N collections: transpordi_planeerimine folder: 001_artiklid

status: converted

ENGINEERING TOOLS, TECHNIQUES AND TABLES

AUTONOMOUS VEHICLES

INTELLIGENT TRANSPORT SYSTEMS AND SMART TECHNOLOGIES

No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in rendering legal, medical or any other professional services.

ENGINEERING TOOLS, TECHNIQUES AND TABLES

Additional books in this series can be found on Nova"s website under the Series tab.

Additional e-books in this series can be found on Nova"s website under the e-book tab.

AUTONOMOUS VEHICLES

INTELLIGENT TRANSPORT SYSTEMS AND SMART TECHNOLOGIES

NICU BIZON LUCIAN DASCALESCU AND NASER MAHDAVI TABATABAEI EDITORS

Copyright © 2014 by Nova Science Publishers, Inc.

All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without the written permission of the Publisher.

For permission to use material from this book please contact us:

Telephone 631-231-7269; Fax 631-231-8175

Web Site: http://www.novapublishers.com

NOTICE TO THE READER

The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers' use of, or reliance upon, this material. Any parts of this book based on government reports are so indicated and copyright is claimed for those parts to the extent applicable to compilations of such works.

Independent verification should be sought for any data, advice or recommendations contained in this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to persons or property arising from any methods, products, instructions, ideas or otherwise contained in this publication.

This publication is designed to provide accurate and authoritative information with regard to the subject matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other professional services. If legal or any other expert assistance is required, the services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.

Additional color graphics may be available in the e-book version of this book.

Library of Congress Cataloging-in-Publication Data

Autonomous vehicles : intelligent transport systems and smart technologies / editors, Nicu Bizon, Lucian Dascalescu, and Naser Mahdavi Tabatabaei (Faculty of Electronics, Communication and Computers, University of Pitesti, Romania).

pages cm

Includes index.

ISBN: (eBook)

1. Autonomous vehicles. I. Bizon, Nicu. II. Dascalescu, Lucian. III. Mahdavi Tabatabaei, Naser. TL152.8.A88 2014

388.3'12--dc23

2014022215

Dedicated to all our teachers and colleagues who enabled us to write this book, and our family and friends for supporting us all along.

CONTENTS

Foreword		ix
Preface		xi
	Part I. Intelligent Transport Systems	1
Chapter 1	Radar Sensor Detectors for Vehicle Safety Systems Modar Safir Shbat and Vyacheslav Tuzlukov	3
Chapter 2	Airborne Ultrasonic Imaging: SONAR Based Image Generation for Autonomous Vehicles Dorel Aiordachioaie, Laurentiu Frangu, Silviu Epure, Radu Belea, Marius Mazarel, Constantin Miholca and Claudiu Chiculita	57
Chapter 3	Autonomous Underwater Vehicle Dynamics Mustafa Dinc and Chingiz Hajiyev	81
Chapter 4	Advances in Environment Sensing and Perception Technologies and Algorithms for Autonomous Ground Vehicles Constantin Ilas, Irina Mocanu and Mariana Ilas	113
Chapter 5	Intelligent Highway Surveillance and Safety Systems Valentin Alexandru Stan, Radu Serban Timnea and Razvan Andrei Gheorghiu	147
Chapter 6	Cyber Security Evaluation of Critical Infrastructures Systems Alecsandru Patrascu and Emil Simion	185
Chapter 7	Sustainable Information Systems Management for Spatial Governance Angela Ionita, Marcel Foca and Maria Visan	207
Chapter 8	Standards of Communications in the Intelligent Transport Systems (ITS) Maaruf Ali	235

viii Contents

	Part II. Smart Technologies Applied in the Electrical and Hybrid Vehicles	247
Chapter 9	Power Electronics and Drives Used in Automotive Applications Ersan Kabalci	249
Chapter 10	Power Control of Plug-In Electric Vehicles in Smart Grids Amin Hajizadeh, Amir Hossein Shahirinia and David C. Yu	275
Chapter 11	Optimal Short-Term Scheduling of Aggregators for Plug-In Electric Vehicles Mohammad Moradi-Dalvand and Behnam Mohammadi-Ivatloo	293
Chapter 12	Energy Efficiency of the Hybrid Power Source Used in the Plug-In Fuel Cell Vehicles Nicu Bizon, Marian Raducu, Mihai Oproescu and Luminita Mirela Constantinescu	313
Chapter 13	Dynamic Response of Electric Machines for Electric Vehicles/Hybrid Electric Vehicles (EV/HEV) Robert Beloiu, Mariana Iorgulescu and Corina Savulescu	353
Chapter 14	Use of the External Magnetic Field for Supervision and Diagnosis of Electrical Traction Motors Remus Pusca, Raphael Romary and Andrian Ceban	379
Chapter 15	Batteries for Electrical Vehicles: A Review Bogdan Enache, Emilian Lefter and Costin Cepisca	409
Chapter 16	Computational Study of Li-Ion Batteries for Automotive Applications Inigo Aramendia, Unai Fernandez Gamiz, Jose Antonio Ramos, Javier Sancho and Ekaitz Zulueta	431
Chapter 17	Ni-Cd Batteries for Automotive Applications: The ac Electrical Data of the Ni-Cd Battery in the Havriliak-Negami Dielectric Representation Mohammad A. Alim, Sudip Bhattacharjee, Sanjida Khanam, Nasiha Muna and Sabiha Runa	453
Chapter 18	Advanced Drive Architectures and State of the Art In-Wheel Motors Blaz Stefe and Gorazd Lampic	485
Chapter 19	The Future of Electrical Vehicles Wim J. C. Melis	509
Index		531

FOREWORD

The Electric Vehicles (EVs) use electric power source for transportation capabilities. Currently electric cars are feasible alternative rather than to the conventional gasoline powered cars because they are significantly less complex, more reliable/efficient, and environmentally friendly, and cost less to operate.The success of extensive electric vehicle carrying depends on the reliability of both the electric grid and the charging equipment. In car accidents, EVs are considered with standard impact a lot better than most gasoline-powered vehicles.

The fossil fuel powered vehicles are generally non-friendly regarding to the environmental aspects. In the past, the vehicles with lower consumption have been presented. After that, the research direct towards more environmentally friendly vehicles which hybrid and electrical vehicles have been developed. The hybrid electric vehicles (HEVs) and developed plug-in hybrid vehicles (PHEVs) are improved for the integration of electrical vehicles which should allow for the vehicle to be charged by green power.

Electric Vehicle components include electric motor, battery pack, battery charger, motor controller, contactor, fuse and circuit breaker, vacuum pump, power steering pump, DC/DC converter, and the instrumentation which each of them has individual technology to obtain the EV efficiency improvement. Also, many EVs have been specialized with special features including resealing tires, programmable charging systems, and regenerative breaking, low maintenance and fairly cost efficient vehicles, with long life batteries.

The current book concerns in development of the advanced autonomous electric vehicles applied in intelligent transport systems and using the smart technologies. Although electric tractions are not new subjects, but their application to electrical automotive vehicles and technological developments for responding to the modern life requirements are discussed.

The above and also advanced aspects related to electric vehicles including the safety systems, plug-in electric vehicles, autonomous ground and underwater vehicles, intelligent transport systems, automotive applications, smart grids, hybrid power source, plug-in fuel cell vehicles, electrical traction motors, and in-wheel motors are illustrated in this book by the editors and authors, in the following topics: radar sensor detectors; sonar based image generation; dynamics analysis and responses; advanced sensing technologies and algorithms; intelligent highway surveillance; cyber security evaluation of critical infrastructures systems; sustainable information systems management for spatial governance; standards of communications; power electronics and drives; power control of smart grid; optimal shortterm scheduling of aggregators; energy efficiency of the multiport power converters; external magnetic field for supervision and diagnosis; batteries for electrical vehicles; computational study of Li-ion and Ni-Cd batteries; advanced drive architectures; and the future of electrical vehicles.

The book chapters and materials are very efficient in theoretical and application issues and are highly recommended for studying and considering in educational and research fields.

February 15, 2014

Academician Arif M. HASHIMOV Institute of Physics Azerbaijan National Academy of Sciences Baku, Azerbaijan

PREFACE

Environmentally friendly, high efficiency, green power charging, renewable sources storage, low maintenance cost, and hybrid integration of electrical vehicles (EVs) over current fossil fuel powered vehicles are generally considered for excellent characteristic features and solutions in transportation development and marketing. The electric vehicles with smart metering, intelligent control and communication interfaces are developed as environmentally friendly cars which use the energy efficiently and minimize the pollution problems caused by burning of the fossil fueled automobiles.

The topics of autonomous ground vehicles, fuel-cell electric vehicles, battery-electric vehicles, automated driver assistance systems, transportation interface equipment, traffic management system, infrastructure elements development, traveller information system, sustainable information systems management, and inter-vehicular communications have grown during the expanding of the electric vehicle systems.

The autonomous vehicles as the intelligent transport systems have recently become more attractive options over combustion engine based vehicles, due to the high fuel price and environmental concerns. The electric vehicles act as vehicle to grid (V2G) systems, which discharge their batteries and send electricity back to the power grid when they are parked and connected to an electrical outlet.

The plug-in electric vehicles (PEVs) are defined as electric vehicle conversions of hybrid electric vehicles (HEVs) and conventional internal combustion engine vehicles which have battery configuration to obtain the required recharged energy for propulsion over the electric grid, the conventional internal combustion engine and regenerative braking. It is essential that the hybrid electric vehicles cannot be recharged from any off-vehicle electric energy sources that are not categorized in plug-in electric vehicles.

Plug-in hybrid electric vehicles (PHEVs) are more flexible loads comparing to traditional transportation systems to increase the power system loading which the large amount of charging power will take place on peak hours, because of stopping at evening and charging in the morning.

The successful operations of extensive electric vehicles depend on the consistency of the electric grid and the charging equipment as well as the cost of the batteries which imply the robust and reliable charger, high power efficiency and minimum impact on the power quality of smart grid distributions systems. The electric vehicles act as regulated consumers and production units as well as the distributed sources to restore the variable energies coming from distributed generation systems.

The hybrid power sources (HPS) combine energy storage devices (batteries, ultracapacitors, etc.) in a hybrid stack (passive, semi-active or active controlled mode) to be used in hybrid electric vehicles (HEVs) and fuel cell vehicles (FCVs) to manage the power flow between the energy sources, and the storage devices and loads based on an energy management unit.

One of the essential parts of the EV/HEV is electric drives in a variety of configurations with more specific characteristics with sufficient reliability and autonomy as well as including faster control of drive torque and improved dynamic performance.

Radar sensor technologies and ultrasound imaging are including high information and accurate measurements from the target vehicle and speed recognition system as well as weather conditioning which are widely used in electric vehicle transportation systems. The technologies are very useful in widespread range of marine systems and unmanned underwater vehicles.

Batteries as the chemical-electrical energy converters through electrochemical discharge reactions should be designed to reduce the possibility of any leakage problem. Two types of available batteries including the primary type discharged after single usage used as the automotive starting battery, and the secondary type such as Hg-Zn, Hg-Cd and Li-Cd batteries are designed for frequently usage which are rechargeable by reversing the chemical reactions used as power of the microelectronic devices sealed to not allow venting of gas to the atmosphere.

The certain problems related to the vehicle"s batteries are the long recharging period, reduced reliability, a limited autonomy, high cost price, high energy density, long life, rechargeable capability, safe handling, and flexibility in applications which should be noted in the research and industry phases.

Autonomous Vehicles: Intelligent Transport Systems and Smart Technologies is a book aimed to highlight the above concepts and challenges by a systematic approach and, therefore, to provide understanding on how the electrical vehicles influence the modern transportation systems, and the ways of controlling grounded and underwater navigation systems by using different research methods.

A large number of specialists are joined as authors of the book chapters to provide their potentially innovative solutions and research related to advanced researches in the electric vehicles systems, in order to be useful in developing new ways in transportation systems, design and operational strategies. Several theoretical researches, case analysis, and practical implementation processes are put together in this book that aims to act research and design guides to help the postgraduates and researchers in electrical power engineering and energy systems. In particular, the specific purposes of the subjects are also including as knowledge base and applications used by undergraduates regarding to indicating research fields and innovative solutions as the challenges and opportunities for solving the problems.

The book presents significant results obtained by leading professionals from the industry, research and academic field, can be useful to the variety of groups in specific areas analyzed in this book. All works contributed to this book are new, previously-unpublished material or extended version of published papers in the proceedings of international conferences and transactions on international journals.

The nineteen chapters present the advanced theories and applications in electrical vehicles systems and their peripherals subsystems and hybrid conditions considering different operating cases. The topics in these chapters are including the advanced methods and Preface xiii

applications in intelligent electrical vehicles and automotive applications in smart grids related to the fields of protection systems, plug-in electric vehicles, autonomous ground and underwater vehicles, hybrid power source, fuel cell vehicles, electrical traction and in-wheel motors.

Chapter 1 introduces the signal detection performance evaluation of detectors employed by radar sensor systems used in vehicle safety structures and discusses about the generalized approach to signal proceeding and noise detection performance such as use of alarm rate and generalized detectors. The chapter includes some important objects related to definition method of the target vehicle parameters. Besides, the chapter introduces middle range radar and short range radar applications.

The applications of ultrasonic waves image diagnosing is given in Chapter 2. The various and important applications in industry, for fault detection and localization, and medicine, for non-invasive diagnostic techniques, are discussed and compared with each other. The algorithms in the chapter are related to ultrasonic images and sonar image generation and processing which the airborne ultrasound images and raw processing methods for navigation purposes are investigated and interpreted.

Chapter 3 presents detailed explanations on the theory and design of the dynamic mathematical modeling of autonomous underwater vehicles (AUVs). The chapter also investigates the simulation model and also implement control and navigation algorithm for the underwater vehicles. The parameter identification of hydrodynamic coefficients based on a least square estimation algorithm for the nonlinear mathematical modeling of autonomous underwater vehicles is also studied in the chapter.

The environment sensing and perception technologies and algorithms used in autonomous ground vehicles (AGVs) are described in Chapter 4. The environmental sensing technologies are also presented in the chapter, focusing mostly on LIDAR, video camera and radar. The chapter shows the increased usage of video cameras, and video camera plus LIDAR or video camera plus radar employed in AGVs. The chapter also shows the implementation of the SVM algorithm based on video camera images and compares the performance with HOG and DSIFT.

The concept and the implementation solution for the integrated highway traffic surveillance system with complex safety integrated systems are introduced in Chapter 5 in order to increase safety and security on European highway. The application of traffic surveillance, traffic re-routing, emergency services announcement, driver informing on modern highway traffic conditions are introduced as well. The chapter also discusses about the highway traffic issues related to traffic video surveillance, automatic incident detection, automatic adaptation of traffic signalization, traffic statistics, speed detection, automatic car identification, and law enforcement system.

Chapter 6 is entirely highlighted on the issues of cyber security, criminals and attacks in cyber systems. The chapter denotes to the attack types such as social engineering, port scanning, packet sniffing, password cracking, denial of the service attacks, machine bugs, and server bugs. Then the cyber threats on SCADA based critical infrastructure like transportations systems, automotive factories, telecommunication, water supply and electricity generation are also discussed. The chapter also covers some attack scenarios explained in detail for in-depth understanding of the concept involved in cyber security.

The complexities of sustainable information systems management in the framework of spatial governance are focused in Chapter 7 in the areas of sustainability based on information technology and communication (ICT). The chapter shows a reorientation proposal of the sustainability for positioning spatial governance based on human behavior and the processing of spatial knowledge. The chapter also describes the requirements for a sustainable information systems management in spatial governance focusing on intelligent transportation systems.

Chapters 8 provides the current standards for communication in intelligent transportation systems (ITS), particularly related to inter-vehicular statement. The chapter presents an overview of the IEEE protocol standard relevant to ITS communication. The latest standardizations in the dedicated short range communication, and the IEEE wireless access in vehicular environments also conferred in the chapter. The infrastructure based and centralized cellular networks are also demonstrated.

Chapter 9 presents power electronics and drives used in electric vehicles. The chapter is also describes electric motors selecting views based on construction structure, efficiency, maintenance requirements, electromagnetic interference, reliability, and speed range parameters. The power electronics devices and motor drives applications used in electric vehicles systems are also described.

In Chapter 10, the plug-in electric vehicles (PEVs) with dual energy storage sources are presented. The chapter shows PEVs dynamic state space modeling containing combined battery systems with supercapacitors. The Lyapunov based Neuro-Fuzzy control strategy to stabilize the DC-link power and distribute the power between dual energy storage sources in plug-in electric vehicles system are analyzed in the chapter. The robust sliding mode control strategy for voltage source converter based on positive and negative symmetrical components are also shown.

Chapter 11 deals with market prices modeling for plug-in hybrid electric vehicles (PHEVs) using stochastic programming technique. The chapter introduces objective of the PHEV aggregator to maximize its profit by charging the PHEVs on the low price hours. The chapter also presents electricity market and PHEVs characteristics constraints incorporating into the proposed method considering an illustrative example.

A mathematical analysis of the energy efficiency for the series and parallel multiport power converter (MPC) topologies is proposed in Chapter 12 to be used in the plug-in fuel cell vehicles (PFCVs). The chapter also defines the electrical motors and inverter systems, and the DC input sources parameters for electrical vehicles, and the strategies for the energy management unit. The other aspect related to choosing the optimal parameters of the MPC control for an efficient use of the PFCV is also introduced. The operation regimes of the energy storage system (ESS) including charge-sustaining, charge-depleting, and chargeincreasing as well as the MPC energy efficiency of the PFCVs in the ESS charging/discharging period from/to the smart grid have been studied.

Chapter 13 provides the dynamic response and the behavior of three types of electric motors for using in the electric vehicle/hybrid electric vehicle (EV/HEV) implemented by different car manufacturers. The chapter shows a testing regime based on the proposed drivers considering all the encountered functional situations. The chapter illustrates the behavior of the rapid and slow start-stop regime, and speed change regime of the DC, induction and brushless machines used in EV/HEV.

The Chapter 14 aims to provide a supervision method for detection of AC electrical machine faults, used in electrical transportation, considering the external magnetic field information nearby the machine may cause the failure responses. The chapter studies two Preface xv

methods based on the analysis of the magnetic field outside the machine which are reliable, original, inexpensive and simple to implement in electrical automotive vehicle. The theoretical analysis and experimental validation are resulted regarding to the low frequency components of the magnetic field which is not influenced by the presence of electronic converters.

The Chapter 15 introduces a review of electrical vehicles batteries as an important factor in development of the automotive industries. The chapter also presents new developments and solutions for creation of batteries applicable to experimental research. The chapter emphasizes that the environmental care completion regarding to reduce the automobiles impact on the environment and also on people"s health which is carried out by accelerating development process of batteries for the electric and hybrid vehicles.

Chapter 16 discusses about designing of a lithium ion cells batteries with the performances of consisting an optimum cooling and heat transfer evaluation along the testing time. The chapter analyzes the battery model based on cold plates based on the geometry and mesh of the elements. The physical properties and initial conditions as well as the convergence of waste products, air speed, cells temperature and heat transfer are also discussed in the chapter.

In Chapter 17 the secondary type of the batteries are discussed with rechargeable performance by reversing the chemical reactions. The chapter shows the AC small signal electrical data of the Ni-Cd battery in the impedance plane and discuss about the various analytical aspects using the Davidson-Cole (D-C) representation. The results of the chapter also show that the possible meaning of the depression parameters extracted from the breadth and skewness as well as deviation of the semicircular Debye or Cole-Cole (C-C) dielectric relaxation loci.

The major mechanical design possibilities of the in-wheel electric motors, including direct drive motors and gears are studied in Chapter 18. It is noted that in-wheel motors increase the total unsprung mass of the vehicle, and therefore the theoretical model of the motors is analyzed in the chapter. The chapter also discusses about the designing topics of inwheel motors for different kinds of vehicles considering different mechanical constructions.

Chapter 19 focuses on the future prospective of the electrical vehicles and their efficiency, comparing with combustion engine (CE) vehicles. The chapter also provides some alternative fuels and storage mechanisms for improving the electrical vehicles performance. The use of batteries as well as alternative fuels such as fossil fuels and fuel cells are studied in the chapter regarding to the environmental impact and also energy storage mechanisms.

The editors recommend book as suitable for dedicated and general audiences that includes the power system professionals, as well as researchers and developers of energy sectors and electrical power engineering community. It is expected that readers to be graduates of energy and power engineering degree programs having a basic mathematical background.

As a whole, the book covers both theoretical background and application examples in transportation systems, and special and professional fields of electric vehicle systems in altogether. Because the subjects" area approached in this book is vast in itself, it has been a little bit hard to balance the theoretical and applicative aspects in each of the chapters, so efforts of editors have been made to well cover the essential topics of the book. Specific indepth further studies are pointed to the dedicated intensive resources of the book subjects for interested readers. Meanwhile, the application and study cases are possibly selected with as much real implications.

Finally, the editors hope that this book will be helpful to undergraduate and graduate students, researchers and engineers, trying to solve energy and power engineering problems using modern technical and intelligent systems based on theoretical aspects and application case studies.

Pitesti, Romania Nicu Bizon

Professor Faculty of Electronics Communication and Computers University of Pitesti Arges, Romania E-mails: nicubizon@yahoo.com, nicu.bizon@upit.ro

Poitiers, France Lucian Dascalescu

Professor PPRIME Institute, UPR 3346 CNRS University of Poitiers-ENSMA, France E-mail: lucian.dascalescu@univ-poitiers.fr

Tabriz, Iran Naser Mahdavi Tabatabaei

Professor Electrical Engineering Department, Seraj Higher Education Institute Tabriz, Iran

E-mail: n.m.tabatabaei@gmail.com

February 20, 2014 Editors

PART I. INTELLIGENT TRANSPORT SYSTEMS

In: Autonomous Vehicles ISBN: 978-1-63321-324-1

Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 1

RADAR SENSOR DETECTORS FOR VEHICLE SAFETY SYSTEMS

Modar Safir Shbat and Vyacheslav Tuzlukov†

School of Electronics Engineering, College of IT Engineering, Kyungpook National University, Sankyuk-dong, Buk-gu, Daegu, South Korea

ABSTRACT

This chapter is devoted to a target return signal detection performance evaluation of detectors employed by radar sensor systems used in vehicle safety systems, for example, safety driving applications. Detection performance and characteristics of several types of the constant false alarm rate (CFAR) detectors and the generalized detector (GD) constructed based on the generalized approach to signal proceeding in noise are compared under the same conditions for the cases of presence and absence of interfering vehicles. The method of definition of the required target vehicle parameters for the middle range radar (MRR) and short range radar (SRR) applications is also discussed.

Keywords: Radar Sensor System, Vehicle Safety System, Constant False Alarm Rate (CFAR) Detector, Generalized Detector (GD), Noise Power Estimation, Multilateration, Attenuation Factors

NOMENCLATURES

ACC Adaptive Cruise Control

ACGO-CFAR Adaptive Censored Greatest of CFAR

AF Additional Filter

Email: modboss80@knu.ac.kr.

Email: tuzlukov@ee.knu.ac.kr.

AWGN Additive White Gaussian Noise

BSD Blind Spot Detection CA-CFAR Cell Averaging CFAR CFAR Constant False Alarm Rate CVD Closing Vehicle Detection

DB Decision Block

FFT Fast Fourier Transform

FMCW Frequency Modulation Continuous Wave

FSK Frequency Shift Keying

GCML Generalized Censored Mean Level

GD Generalized Detector GO-CFAR Greatest of CFAR

i.i.d. Independent and Identically Distributed

LFMCW Linear FMCW LRR Long Range Radar MRR Middle Range Radar MSG Model Signal Generator NP detector Neyman-Pearson detector OS-CFAR Ordered Statistic CFAR pdf Probability Density Function

PF Preliminary Filter RCS Radar Cross Section

RLC Resistor, Inductor, and Capacitor oscillatory circuit SGSA Signal Model Generator Switching Apparatus

SNR Signal-to-Noise Ratio SO-CFAR Smallest of CFAR SRR Short Range Radar THRA Threshold Apparatus UWB Ultra Wide Band

1.INTRODUCTION

Recent investigations discussed in [1] demonstrate a superiority of implementation of radar sensor systems for vehicle collision avoidance and safety driving applications in comparison with other technologies. Radar sensor systems offer a high information content providing accurate measurements of the target vehicle range and relative velocity in addition to ability to operate under any weather conditions and harsh environments.

In this chapter, we evaluate the signal detection and signal processing algorithms used in radar sensor system for vehicle collision avoidance and safety driving applications. Performance metrics and simulation conditions are defined with the purpose to obtain an impartial comparison of different detectors. Under comparison we define the detection performance of the target return signal as the main characteristic.

Detection performance of the target return signal is the function of the signal-to-noise ratio (SNR) and detection threshold and, consequently, the noise power sensitivity. Detection threshold should be defined based on the observed in real time noise power if the noise variance is varied. Thus, the adaptive detection threshold is very important parameter to maintain the constant probability of false alarm PFA . For any radar sensor system with high duty cycle, for example, the linear frequency modulation continuous wave (LFMCW) radar sensor system, a sensitivity of the noise power is an important problem. The constant probability of false alarm PFA is a desirable requirement for most radar sensors applications. By this reason, there is a need to define the detection threshold of target return signal based on the noise power or noise variance estimation if the noise power or variance is varied as a function of time. The noise power estimation is a vital factor allowing us to define adaptively the detection threshold in radar sensor systems and controller area networks used to design the radar sensor systems for safety driving applications, collision avoidance systems, and vehicle tracking systems.

The appropriate signal detection and signal processing algorithms for radar sensor systems employed in numerous vehicle safety and collision avoidance applications are investigated for two different schemes. The first scheme is the constant false alarm rate (CFAR) detector, and the second scheme is the generalized detector (GD) constructed based on the generalized approach to signal processing in noise. GD is a combination of the correlation detector that is optimal in the Neyman-Pearson criterion sense (NP detector) for detection of signals with a priori known parameters and the energy detector that is optimal in the Neyman-Pearson criterion sense under detection of signals with a priori unknown parameters. This combination allows us to formulate a decision-making rule about the presence or absence of target vehicle based on definition of the jointly sufficient statistics of the likelihood function mean and variance with complete compensation of the noise component that is specific both for the NP detector and the energy one. The target return signal processing technique is introduced to define the required target vehicle parameters, namely, the range, relative velocity, and position or coordinates in order to distinguish the threaten and non-threaten detected targets.

Several CFAR detectors are discussed, namely, the cell averaging CFAR (CA-CFAR) detector, ordered statistic CFAR (OS-CFAR) detector, generalized censored mean level (GCML) detector, and adaptive censored greatest of CFAR (ACGO-CFAR) detector. All these detectors can be employed in the linear frequency modulation continuous wave (LFMCW) radar sensor system, which is widely used by the vehicle collision avoidance and safety driving systems for the middle range radar (MRR) and short range radar (SRR) applications such as the closing vehicle detection (CVD) and blind spot detection (BSD).

All the mentioned CFAR detectors and GD are compared by detection performance of the target return signal following the proposed framework and under the same initial conditions. We consider two main cases. The first case – there are no any interfering target vehicles. The second case – there are one or more interfering target vehicles. Based on the theoretical analysis, a simulation is carried out at the conditions closed to practice that allows us to evaluate the above-mentioned detectors employed by LFMCW radar sensor system in the case when the interfering targets are present or not.

The implementation of GD in radar sensor systems used for vehicle safety applications demonstrates the better detection performance in comparison with the CFAR detectors at the same conditions and promises the high probability of detection. This result is confirmed by implementation of the noise power estimation for GD based on sliding window procedure with the same number of reference cells used by CFAR detectors in order to perform the fair comparison.

2.INTELLIGENT VEHICLE AND TOTAL SAFETY DRIVING APPROACH

A lot of technologies are developed by researches and companies as a consequence of great growth of automotive industry. Until recent time, the automotive industry has been dominated by mechanical engineering. Nowadays this industry needs the efforts to signal processing, automatic control, electronic engineering, communication systems, and computer hardware and software in order to construct and develop the modern vehicle intelligent systems, for example, the self-driving and self-parking, vehicle to vehicle communications, navigation systems, and safety driving applications.

The theoretical, experimental, and operational aspects of electronic engineering and information technologies are integrated to enrich the vehicle by required ability to drive safety, avoid accidents, and protect the people inside and outside the vehicle form serious injuries in the case of accident occurrence [2]. The total safety driving approach means that the vehicle should be equipped by specific systems to avoid accidents and prevent injuries. This approach can be achieved by installing different kinds of technologies such as the cameras with image processing, navigation and information systems, and sensors, for example, the laser, ultra-sound, infrared, and radar sensors to form a network of active and passive safety driving applications.

The safety driving applications based on radar sensors can be categorized by three main parameters: the range, sensor type, and location of radar sensor system around the vehicle (frond, side, and rear). Radar sensors offer high information content and provide good ranging ability and relative velocity measurement in addition to other advantages, namely, the operation under any weather conditions and harsh environment (dust, dirt, light, rain, snow, and so on). Radar sensor systems are classified into three types based on the range, namely, the long range radar (LRR), MRR, and SRR [3]. Many safety driving applications are developed using the radar sensor systems, for example, the adaptive cruise control, BSD, front and rear collision warning and avoidance, parking aid, lane change assistant and CVD [4-6]. Figure 1 presents many examples of safety driving applications classified according to the range and radar sensor location.

Figure 1. Safety driving applications.

Radar sensor systems operate typically in pulse (pulse radar sensor) or continuous wave mode, namely, the LFMCW radar sensor, the frequency shift keying (FSK) radar sensor, and the combined LFMCW and FSK radar sensor systems [7]. The coded radar sensor, spread spectrum radar sensor, and random noise radar sensor systems are not widely used in safety driving applications. Table 1 present the most popular types of radar sensor systems employed by various safety driving applications at 24 and 77 GHz operation frequencies [8].

From Table 1, it can be noticeable that LFMCW, FSK, and pulsed radar sensor systems are the most used technologies implemented in various safety driving applications. The LFMCW radar sensor system is considered as the best decision, especially for the MRR applications, such as CVD [9], and SRR applications, such as BSD, owing to comparison of the main features of the radar sensor systems [1] presented in Table 2.

Table 1. Typical radar sensor systems used in safety driving applications

Application	Range/Velocity/View Requirements	Sensor Category	Radar Sensor Type	Operation Frequency
Parking Aid	0.2~5 m 0~30 Km/h Full vehicle width	SRR	UWB/Pulsed	24 GHz
BSD	0.5~10 m Reasonable velocity interval Two side regions	SRR	LFMCW/ Pulsed/ FSK/	24 GHz
Adaptive Cruise Control	1~150 m Reasonable velocity 3 lanes in front of the vehicle	LRR	LFMCW/ Pulsed/ FSK/	77 GHz
Stop & Go System	0.5~150 m/ 0.5~40 m Reasonable velocity 3 lanes in front of the vehicle	LRR/MRR	LFMCW/ Pulsed/ FSK/	77 GHz 24 GHz
CVD	0.5~30 m Any velocity About 45 degree	SRR/MRR	LFMCW/FSK	24 GHz
Collision Avoidance	0.5~150 m/ 0.5~40 m Any velocity 3 lanes in front of the vehicle	LRR/MRR	LFMCW/FSK	77 GHz 24 GHz
Pre-Crash Reversible Restraints	0.5~10 m/ 0.5~30 m Any velocity Full vehicle width	SRR/MRR	LFMCW/FSK	24 GHz

Table 2. Comparison of the main features of radar sensor systems

Criteria	Radar Sensor System
	Pulsed	FSK	LFMCW
Range Resolution	Good	Average	Good
Measuring Ability	Average	Good	Average
Fixed Obstacle Detection	Good	Poor	Good
Robustness to Jamming	Poor	Good	Good

Under design of any safety driving systems, many important practical issues related to the radar sensor network design requirements and controller area network should be taken into consideration. These practical issues can be summarized as follows:

• The time synchronization of radar sensor networks that is important for target vehicle state estimation. In many cases, the data transmission rate and delay are critical factors, especially, for SRR systems.
• The interface for communication between the distributed radar sensor networks.
• The position or location of every radar sensor. For example, the radar sensor location behind the vehicle bumper affects the performance and should be defined very accurately to guarantee a precise determination of azimuth angle estimation.
• The possible cross talk and undesired microwave propagation behind the vehicle bumper must be avoided.
• The computation complexity is increased in the case of radar sensor networks. All radar sensor signals have to be evaluated and the data association and fusion have to be performed and examined.
• The optimal allocation of digital signal processing blocks or units in the radar sensor network is an important problem.
• The structural complexity should be as low as possible to reduce the average number of failures and the cost.
• The number and size of components have to be small to solve the problem of integration space.
• The radar sensor systems must possess the same quality; otherwise, mismatches under signal detection and signal processing should be taken into consideration.

More information and details concerning about these practical aspects and the design of radar sensor networks can be found in [10]. Appropriate signal detection and signal processing procedures for FMCW radar sensors are used to design the MRR and SRR safety driving systems.

3. OVERVIEW OF MRR AND SRR SYSTEMS

There are the following main parameters of the target vehicle, namely, the target vehicle range R , the relative velocity between the subject vehicle and target vehicle Vr , and the target vehicle position defined by the coordinates. These parameters are required to be used by safety driving systems and can be obtained under the signal detection and signal processing procedures as shown in Figure 2.

The frequency modulated signal, i.e. the transmitted signal, is reflected from the target vehicles. The target return signal is multiplied by the transmitted signal copy (the transmitted signal model) at the receiver. The difference in frequency between the transmitted signal and target return signals can be obtained using the mixer. This difference in frequencies is called the beat frequency. If the LFMCW radar sensor generates a triangle waveform (see Figure 3), the up- and down-beat frequencies are distinguished using the fast Fourier transform (FFT) at the mixer output. If the target vehicle is stationary the beat frequency is a constant value.

Figure 2. Overview of MRR and SRR systems.

Figure 3. LFMCW radar sensor waveform.

However, if the target vehicle is moving there is the Doppler frequency and, as a consequence, the beat frequency is changed. To detect the target return signal, the CFAR detector and GD are employed and their detection performances are compared. Using the beat frequencies, we can determine the target vehicle ranger R and relative velocity between the subject and target vehicles in the following form [11, 12]: Vr

$$R = \frac{c(f{bu} + f{bd})T_S}{8B} \tag{1}$$

$$Vr = \frac{c(f{bd} - f_{bu})}{4f_c} \ , \tag{2}$$

where bu f is the up-beat frequency, bd f is the down-beat frequency, c is the velocity of light, B is the bandwidth of the LFMCW radar sensor, c f is the centered or operation frequency, and TS is the period of up-chirp and down-chirp waveforms. The relative velocity and range of the target vehicle are functions of the up- and down-beat frequencies [13]. After definition of the target vehicle range the target vehicle position or coordinates can be determined as an intersection point of circles with the radius equal to the measured range (multilateration process) using two or more radar sensors and the centers of these circles are the radar sensor positions.

The LFMCW radar sensor system changes linearly the transmitted waveform frequency as a function of time. The modulating waveform consists of two sweeps: the up-sweep or upchirp and down-sweep or down-chirp. The instantaneous transmitted frequency f (t) ^T can be presented in the following form:

$$f_T(t) = \begin{cases} f_c + \frac{B}{0.5T_S}t , \ 0 < t < 0.5T_S \ f_c - \frac{B}{0.5T_S}t , \ 0.5T_S < t < T_S \end{cases}$$ (3)

The delay time between the transmitted and target return signals is given by

$$\tau = \frac{2R}{c} \tag{4}$$

The target return signal delay in the case of moving target with the Doppler shift in frequency takes one of the following forms:

$$\tau_m = \begin{cases} \frac{2(R - V_r t)}{c} & - \text{ for approaching target vehicle;} \ \frac{2(R + V_r t)}{c} & - \text{ for receding target vehicle.} \end{cases}$$ (5)

The LFMCW sinusoidal transmitted signal can be written in the following form:

$$S_T(t) = A_T \sin[\phi_T(t)] , \qquad (6)$$

where AT is the amplitude, and (t) ^T is the instantaneous phase defined as

$$\phi_T(t) = 2\pi \int0^{t{in}} f_T(t)dt . (7)$$

From (3), (4) and (7), the sinusoidal transmitted signal can be rewritten in the following form:

$$S_T(t) = A_T \sin \left[ 2\pi \left( f_c t + \frac{B}{T_S} t^2 \right) \right]. \tag{8}$$

The target return signal coming at the detector input can be written in the following form:

$$S_R(t) = A_R \sin[\phi_R(t)] + w(t) = A_R \sin\left{2\pi \left[f_c(t-\tau) + \frac{B}{T_S}(t-\tau)^2\right]\right} + w(t) , \qquad (9)$$

where AR is the target return signal amplitude, (t) ^R is the target return signal phase, and w(t) is the additive white Gaussian noise (AWGN) with zero mean and variance 2 n .

If the Doppler shift in frequency is taken into consideration (9) takes the following form:

$$S{R}(t) = A{R} \sin[\phi{R}(t)] + w(t) = A{R} \sin\left{2\pi \left[f{c}\left(t - \tau - \frac{2V{r}t}{c}\right) + \frac{B}{T{S}}\left(t - \tau - \frac{2V{r}t}{c}\right)^{2}\right]\right} + w(t) .$$ (10)

In practice, under detection of the target return signal there is a need to estimate the noise variance or noise power in real time or, in other words, to apply a procedure for definition of adaptive detection threshold. Noise power estimation techniques are widely employed by wireless communication, cognitive radio, speech recognition, radar sensor, remote sensing, and other systems. In the case of radar sensor system, the noise power estimation is needed to define the adaptive detection threshold owing to expected changes in the noise power, especially, under the non-homogenous noise.

4. CFAR DETECTOR

Using the Neyman-Pearson criterion, the false alarm probability must be minimal, while the probability of detection must be maximal. The probability of false alarm PFA is defined as the probability of event that the observed sample of signal sequence at the radar sensor system output exceeds the detection threshold when the noise only is coming in at the radar sensor system input. The probability of false alarm PFA under employment of the square law detector can be expressed as [14]: PFA PD

$$P_{FA} = \exp\left{-\frac{THR}{\sigma_n^2}\right},\tag{11}$$

where 2 n is the noise variance and THR is the detection threshold. The detection threshold THR can be defined based on the probability of false alarm PFA given before in the following form:

$$THR = \sigman^2 \ln \left( \frac{1}{P{FA}} \right). \tag{12}$$

It is important to note that the probability of false alarm PFA is very sensitive to any variations of the detection threshold value. Figure 4 shows the normalized detection threshold THR ⁿ as a function of the probability of false alarm PFA . The CFAR detection or, or in other words, the adaptive threshold and automatic detection are a set of signal detection techniques providing the predictable detection and false alarm rate behavior [15]. Under CFAR detection, the noise power must be estimated in order to achieve the predictable and consistent probability of false alarm PFA . By this way, the detection threshold can be adjusted to maintain the desired and fixed probability of false alarm PFA .

Figure 4. Normalized detection threshold versus the probability of false alarm.

For any CFAR detector the noise power is estimated after processing the number of reference cells using the sliding window technique. The required threshold is defined by multiplication between the estimated noise power and scaling factor. This threshold is compared with the signal power in the test cell or the investigated cells. CFAR detectors are differed based on the reference cell processing procedure and scaling factor. The threshold of any CFAR detector can be defined in the following form [16]:

$$THR = \alpha \sigma_n^2 , \qquad (13)$$

where is the scaling factor. Figure 5 presents a simple functioning diagram for CFAR detector. CFAR detectors use the estimation technique based on arithmetic averaging. The square law detector shown in Figure 5 generates the output signal amplitude proportional to the squared input signal amplitude.

Figure 5. CFAR detector functioning diagram.

We consider the following CFAR detectors: CA-CFAR, OS-CFAR, GCML, and ACGO-CFAR. All these detectors can be employed by LFMCW radar sensor system. The abovementioned detectors are compared by detection performance of the target return signal under the same initial conditions. We analyze two cases:

• There are no any interfering target vehicles.
• There are one and/or several interfering target vehicles.

4.1. CA-CFAR Detector

The CA-CFAR detector [16] has an optimal performance under the homogeneous noise conditions when the neighboring reference cells contain the noise samples with the same statistical characteristics as the test cell. The CA-CFAR detector estimates the noise power by averaging the data samples in the reference cells of sliding window and after that defines the detection threshold. Basic CA-CFAR detector structure is shown in Figure 6.

Figure 6. The main structure of the CA-CFAR detector.

The noise in the cells of leading and lagging windows and in the test cell is the independent identically distributed (i.i.d.) process. If the target return signal reading is in the test cell, the cells of leading and lagging windows do not contain the target return signal readings from other targets that can bias the detection threshold. CA-CFAR detector uses the maximum likelihood estimator with respect to readings in the cells of leading and lagging windows with the purpose to form the noise power estimation.

The noise power estimation can be presented in the following from [17]:

$$\sigma{n{CA}}^2 = \frac{1}{N} \sum_{i=1}^{N} X_i$$ , (14)

where Xi ,i 1,,N are the square law detector output samples stored by the reference cells. The signal amplitude at the square law detector output is proportional to the squared input signal amplitude and N is the number of reference cells. The probability of detection can be determined in the following form: CA ^PD

$$PD^{CA} = \left[1 + \alpha{CA}(1 + \overline{SNR})\right]^{-N} , \qquad (15)$$

where SNR is the average SNR. The scaling factor for CA-CFAR detector is a function of the number of reference cells N and the probability of false alarm PFA

$$\alpha{CA} = (P{FA})^{-1/N} - 1 . {16}$$

The CA-CFAR detection threshold is defined as follows

$$THR{CA} = \alpha{CA} \sigma{n{CA}}^2 \tag{17}$$

4.2. OS-CFAR Detector

In many practical cases, the non-homogeneous noise is a result of spatial and temporal variations in the noise power or closely spaced target return signals resulting in a bias in the noise power estimation and, consequently, the detection threshold. The OS-CFAR detector is employed under the non-homogeneous noise conditions [18]. In the multiple target case, the detection performance of OS-CFAR detector is better in comparison with the detection performance of the mean level CFAR detector family including the CA-CFAR detector, greatest of CFAR (GO-CFAR) detector, and smallest of CFAR (SO-CFAR) detector. The OS-CFAR detector rearranges the reference cell data samples { , , } X¹ X ^N to form a new sequence according to the increasing power { } X(1) X(2) X(k) X(N) , where X(^k ) is called the *k*th order statistics selected as the noise power reading (see Figure 7).

This approach is based on the fact that the interfering target signal power is usually higher than the noise power. Thus, the OS-CFAR detector is able to reject the number N k of the target return signal readings caused by the interfering target vehicles, which are stored in the reference window cells. Minimal OS-CFAR detector losses are achieved if k has a reasonable value defined by

$$k = 0.75k$$ . (18)

Figure 7. The main structure of the OS-CFAR detector.

The probability of detection for OS-CFAR algorithm is determined by the following from [18]: OS PD

$$PD^{OS} = \prod{i=0}^{k-1} \frac{N-i}{N-i + \frac{\alpha_{OS}}{1+SNR}}$$ (19)

The scaling factor in the case of OS-CFAR procedure is given by

$$\alpha{OS} = (P{FA})^{-1/k} - 1 . {(20)}$$

Finally, the OS-CFAR detection threshold is determined as:

$$THR{OS} = \alpha{OS} \sigma{n{OS}}^2 = \alpha{OS} X{(k)} . \tag{21}$$

4.3. GCML Detector

GCML detector is employed under the non-homogeneous noise [19]. The GCML detector defines the number and location or coordinates of interfering target vehicles by data stored in the reference window cells and discards them prior to define the noise power and threshold. Thus, there is no need to know the number of interfering target vehicles under noise power estimation (Figure 8).

The reference window is divided on two windows equal by length or number of cells, namely, the leading window cells with data { , , , } X¹ X² X ^M , and the lagging window cells with data { , , , } X ^M1 X ^M2 *X2^M* where M N 2 . The data stored in the cells of these two windows are processed independently in a parallel way by censoring processors. The censoring algorithm ranks the data stored in the cells of these two windows in ascending order of the amplitude as { } X(1) X(2) X(^M ) and { } X(*M*1) X(*M*2) X(2^M ) , respectively. The algorithm considers the lowest order X(1) as the noise power sample and defines the threshold 1 (1) (1) THR^X X , where n is the scaling factor or constant chosen to obtain a specific probability of false censoring and n 1,,M 1 . The probability of false censoring can be determined in the following form [19]:

Figure 8. The main structure of the GCML detector.

$$P_{FC} = \frac{M - n + 1}{(1 + \alpha M - n)^{M - n}} \ . \tag{22}$$

The threshold X(1) THR is compared with X(2) and if X(2) exceeds the threshold X(1) THR the censoring processor decides the data { , , , } X² X³ X ^M are the target return signal samples from the interfering target vehicles. If X(2) THR^X the algorithm decides that X(2) is the noise sample without interference. After that the censoring algorithm generates a sum of two lower ordered samples X(1,2) X(1) X(2) , defines the threshold X(1,2) THR , and compares it with X(3) . The censoring process is stopped when the hypothesis is true. The noise power estimation nGCML is scaled by the scaling factor GCML to achieve the desired probability of false alarm PFA . The GCML detection threshold is defined as follows: *H*1

$$THR{GCML} = \alpha{GCML} \sigma{n{GCML}}^2$$ (23)

After definition of the threshold THRGCML , the test cell data are compared with it and a decision a "yes" or a "no" target vehicle is made.

4.4. ACGO-CFAR Detector

The ACGO-CFAR detector can be considered as a modified version of the GCML detector. The statistics stored in the cells of the leading and lagging windows are not the same [20]. The detector suppresses the clutter edge false alarm by calculating the average noise power in the leading and lagging window cells separately after applying the censoring process and selects the highest averaged value considering it as the required noise power estimation. The structure of ACGO-CFAR detector is shown in Figure 9.

Figure 9. The main structure of the ACGO-CFAR detector.

Assume that after applying the censoring process to the leading and lagging window cells *n*1 samples are censored from the leading window cells and 2 n samples are censored from the lagging window cells. The remaining samples from the leading window cells m¹ M n¹ and from the lagging window cells m² M n² are used to estimate the noise power as follows [20]:

$$\begin{cases} U = \frac{1}{m1} \sum{i=1}^{m1} X{(i)}, \ W = \frac{1}{m2} \sum{i=1}^{m2} X{(i)}. \end{cases}$$ (24)

The noise power estimation is set to be a maximum of U and W

$$\sigma{n{ACGO}}^2 = \hat{\sigma}_n^2 = \max(U, W) . \tag{25}$$

The noise power estimation nACGO is scaled by the scaling factor ACGO to achieve the desired probability of false alarm PFA . Then, the output of test cell is compared with the ACGO-CFAR detector threshold to make a decision whether the test cell contains the target return signal sample or not. The ACGO-CFAR detector threshold is given by

$$THR{ACGO} = \alpha{ACGO} \sigma{n{ACGO}}^2 . {26}$$

5. CFAR DETECTION PERFORMANCE

The detection performance of all discussed above CFAR detectors is evaluated under the same initial conditions for two cases:

• There are no interfering target vehicles;
• There are interfering target vehicles.

Figure 10. Probability density functions of RCS fluctuations described by the Swerling models when 1.5 av .

We consider the following scenario. There are the subject vehicle and the only target vehicle approaching from the rear side of the subject vehicle with relative velocity 10m/s V^r and 70m away, i.e. the target vehicle range R 70m . The interfering target vehicles are located at different distances away from the subject target with various relative velocities. The radar cross section (RCS) fluctuations of all targets are modelled by the Swerling 2 model [21]. The statistics associated with the Swerling 2 model of RCS fluctuations is applied to the targets consisting of small scatterers with the comparable RCS, i.e. the independent fluctuating point scatterers with approximately equal physical dimensions. The probability density function (pdf) of the target RCS fluctuations described by the Swerling 1 and 2 models (see Figure 10) is defined as:

$$f(\rho) = \frac{1}{\rho{av}} \exp\left{-\frac{\rho}{\rho{av}}\right} , \quad \rho \ge 0$$ (27)

where av denotes the average RCS fluctuations of total number of targets.

The target RCS fluctuations described by the Swerling 2 model are rapid in comparison with the target RCS fluctuations described by the Swerling 1 model. The losses of target RCS fluctuations described by all Swerling models at the probability of false alarm PFA equal to 4 10 as a function of the probability of detection are presented in Figure 11. In practice, if PD

the probability of detection PD is greater than 0.5, i.e. P^D 0.5 , the losses of RCS fluctuations described by the Swerling 1 and 2 models are higher than the losses of RCS fluctuations described by the Swerling 3 and 4 models. For example, when the probability of detection PD is equal to 0.8, the losses of RCS fluctuations described by the Swerling 3 model are approximately 2.3 dB and the losses of RCS fluctuations described by the Swerling 2 model are 5 dB.

Figure 11. Losses of the fluctuations described by the Swerling models.

In practice, the LFMCW radar sensor system parameters are chosen to get the specific range and velocity resolution that can be defined using the following forms [22]:

$$\Delta R = \frac{c}{2B} \tag{28}$$

$$\Delta V = \frac{\lambda}{T_S} \ , \tag{29}$$

where R is the range resolution, V is the velocity resolution, and is the wavelength. If there is a requirement to achieve the range resolution equal to R 0.75 m, the required bandwidth must be equal to B 200 MHz . When the velocity resolution is equal to V 0.2 m/s, the required modulation time TS must be 0.0625 sec for the radar sensor operation frequency c f equal to 24 GHz.

The simulation for the discussed CFAR detectors is carried out for the ultra-wide band (UWB) LFMCW radar sensor system with the bandwidth , operation frequency 24GHz ^c f , modulation time 0.0625sec T^S that is equivalent to the up-sweep waveform period equal to 0.03125sec and the probability of false alarm equal to 4 10 PFA . Under simulation, the probability of detection is defined after K observations. In this case, the probability of detection PD is determined as the ratio of the number of overshoots Ko with B 600MHz PD

respect to the detection threshold (the detection threshold exceeding) to the total number K of observations [14]:

$$P_D = \frac{K_o}{K} \tag{30}$$

Table 3 presents the main parameters of the investigated LFMCW radar sensor system. The LFMCW radar sensor system parameters are chosen for the MRR and SRR sensor systems implemented in safety driving applications [23-25].

Table 3. Main simulation parameters

Simulation Parameters	Values
Operation Frequency	24 GHz
Bandwidth (B)	600 MHz
Modulation Time (Ts)	0.0625 s
Probability of False Alarm (PFA)	10-4
Reference Cells (N)	20
Number of Observations (K)	1000
Signal Power	100 mW
Selected Cell (k)	15 (for OS)
First Target Range (R1)	60 m
First Target Relative Velocity (V1)	10 m/s
Probability of False Censoring (PFC)	10-3
Second Target Range (R2)	63 m
Second Target Relative Velocity (V2)	10 m/s
Third Target Range (R3)	66 m
Third Target Relative Velocity (V3)	10 m/s
Fourth Target Range (R4)	70 m
Fourth Target Relative Velocity (V4)	10 m/s
Fifth Target Range (R5)	73 m
Fifth Target Relative Velocity (V5)	10 m/s
Sixth Target Range (R6)	76 m
Sixth Target Relative Velocity (V6)	10 m/s
No Guide Cells
Subject target vehicle is simulated as Swerling 2 model
Interfering target vehicles are simulated as Swerling 2 model

Figure 12 demonstrates the additive mixture of the up-chirp target return signal and noise at the mixer output (see Figure 2) after applying FFT to obtain the up-beat frequency. The CA-CFAR processing can be applied to the mixer output using 20 reference cells in order to make a decision about the presence or absence of the target return signal. The detection threshold is set according to the noise power estimation and the system can detect the target return signal if the detection threshold is exceeded, see Figure 12, SNR 9.21dB.

Based on simulation, the probability of detection is determined after K observations, as shown in Figure 13 when the number of overshoots Ko with respect to the detection threshold is counted. PD

Figure 12. Mixer output process: a) after FFT and b) CA-CFAR detection, SNR 9.21dB.

Figure 13. The number of overshoots for CA-CFAR detection with respect to the threshold: a) counting the number of overshoots; b) the number of overshoots for the CA-CFAR detector in the up-chirp target return signal.

Based on simulation, the probability of detection is determined after K observations, as shown in Figure 13 when the number of overshoots Ko with respect to the detection threshold is counted. PD

Figure 14 presents a comparison between the detection thresholds of the CA-CFAR and OS-CFAR detectors in the up-chirp signal when the up-beat frequency is equal to 7360 Hz. In a general case, when the noise is presented only in the reference cells of sliding window, the OS-CFAR detection threshold is higher than the CA-CFAR detection one.

Theoretical probability of detection as a function of SNR for the Neyman-Pearson, CA-CFAR, and OS-CFAR detectors is presented in Figure 15. The CA-CFAR detector has the better detection performance in the case of homogeneous noise in comparison with the OS-CFAR detector. Detection characteristics of the CA-CFAR and OS-CFAR detectors are very close to the detection performance of the Neyman-Pearson detector. In practice, the multiple target scenario, i.e. there are several vehicles in the observation area, is widely used. Assume, there are the primary target vehicle, i.e., the subject vehicle, with the specific range and relative velocity, and the interfering target vehicle with the same relative velocity but its range is different from the primary target vehicle one. The target return signal samples from the interfering target vehicle are stored in the cells of the reference window where the target return signal data from the primary target must be stored. For the case of two target vehicles, namely, the primary target and one interfering target, the up-chirp signals are shown in Figure 16.

Figure 14. Comparison of the CA-CFAR and OS-CFAR detector thresholds: a) enlarged snapshot and b) comparison of the CA-CFAR and OS-CFAR detector thresholds.

The target masking phenomenon occurs when there are several target vehicles and the target return signal sample belonging to a specific target vehicle is stored in the test cell of sliding window, while the target return signal samples from other target vehicles are stored in the reference cells of sliding window. In the case of the CA-CFAR detector, the target return signal samples stored in the reference cells can mask the target return signal stored in the test cell of sliding window. The target masking phenomenon is considered as a disadvantage of the CA-CFAR detector, since the noise power estimation is based on determination of the arithmetic mean or averaging the data stored in the reference cells of sliding window.

Figure 17 demonstrates a comparison between the detection thresholds of the CA-CFAR and OS-CFAR detection procedures for two targets. As shown in Figure 17, the target return signal of the first target vehicle exceeds the CA-CFAR and OS-CFAR detection thresholds. The target return signal of the second target vehicle exceeds only the OS-CFAR detection threshold. Thus, the OS-CFAR detector is able to detect two target vehicles, while the CA-CFAR detector can detect only the first target vehicle owing to the target masking phenomenon. In other words, this is the target masking case for the CA-CFAR detector.

Figure 18 presents a detection of the target return signals for two target vehicles by the OS-CFAR detector at k 15. The noise power estimation procedure under OS-CFAR detection technique allows the detector to be insensitive with respect to the target vehicles closely spaced between each other or to be robust relative to the interfering target vehicles.

Figure 15. Theoretical detection performance of the NP, CA-CFAR, and OS-CFAR detectors.

Figure 16. The up-chirp signals in the case of two target vehicles.

Figure 19 demonstrates a comparison of detection performances between the CA-CFAR and OS-CFAR detectors for three cases:

• There are no any interfering target vehicles;
• There is one interfering target vehicle;
• There are two interfering target vehicles.

We assume that the interfering target vehicles have the same radar cross section (RCS) as the primary target. If there is no interfering target vehicle, in this case, the CA-CFAR detection procedure is the best by performance. If we observe one or two interfering target vehicles, the OS-CFAR detection is the better and demonstrates an advantage in robustness with respect to the interfering target vehicles. At the same time, the CA-CFAR detection is very poor by performance owing to the target masking problem.

Figure 17. The target masking phenomenon of the CA-CFAR detector.

Figure 18. OS-CFAR detection of two closely spaced target vehicles.

Figure 19. Detection performance comparison of the CA-CFAR and OS-CFAR detectors.

The GCML detection procedure allows us to estimates the number of interfering target vehicles. In this case, the target return signals of interfering target vehicles are stored in the reference cells of sliding window. Using cell by cell technique, the censoring procedure allows us to accept or reject the data stored in the cells of the reference window. Thus, to make a decision about detection of the subject vehicle there is no need to know a priori the number of interfering target vehicles. Under employment of the GCML detector, the scaling factors n ,n 1,,M 1 are chosen to get the given before probability of false censoring PFC in (22). For the case when the number N of reference cells equals to 20, i.e. M N / 2 10, and the probability of false censoring PFC is equal to 3 10 , the related scaling constant values are presented in Table 4. Detection performance of the GCML detector is similar to the CA-CFAR detection performance if there are no interfering target return signals.

Figure 20. False alarm for GCML detector caused by clutter edge transition: a) the ACGO-CFAR and GCML detectors clutter edge case and b) false alarm for the GCML detector.

Table 4. The scaling constants for censoring procedure and 3 10 PFC

nth step	1	2	3	4	5	6	7	8	9
n	1999	53.7723	14.8740	7.4090	4.6968	3.3737	2.6106	2.1209	1.7826

The ACGO-CFAR detector can suppress the clutter edge false alarm by independent processing the data stored in the cells of leading and lagging windows with the purpose to obtain two tentative noise power estimations. In the case of GCML detector, there is a false alarm in the clutter transition region. The ACGO-CFAR detector with the high detection threshold is less susceptible to false alarms within the limits of the transition regions (see Figure 20).

Figure 21. Comparison of detection performance of the OS-CFAR, GCML, and ACGO-CFAR detectors: no interfering target vehicles.

Figure 22. Comparison of detection performance of the OS-CFAR, GCML, and ACGO-CFAR detectors: one interfering target vehicle.

Figure 23. Comparison of detection performance of the OS-CFAR, GCML, and ACGO-CFAR detectors: two interfering target vehicles.

Figure 21 demonstrates a comparison between the OS-CFAR, GCML, and ACGO-CFAR detection performance when there are no interfering target vehicles. The curves are very close to each other with slightly vantage to GCML detector. Although the detection performance of the ACGO-CFAR detector is inferior to the GCML detection performance, the ACGO-CFAR detector exhibits the robust false alarm control property and has a capability to censor the undesired interfering target vehicles. Figures 22 and 23 demonstrate the cases when one interfering target vehicle and two interfering target vehicles are introduced, respectively. The OS-CFAR, GCML, and ACGO-CFAR detectors demonstrate robustness and the same detection performance ranking shown in Figure 21.

Figure 24. Comparison of detection performance of the OS-CFAR, GCML, and ACGO-CFAR detectors: five interfering target vehicles.

Figure 24 demonstrates the detection performance of the OS-CFAR, GCML, and ACGO-CFAR detectors when there are five interfering target vehicles. In this case, all detectors show degradation in the detection performance and also keep similar ranking as before excluding the OS-CFAR detector that has the worst detection performance at the high SNR values (more the 20 dB).

Figure 25. Comparison of detection performance six interfering target vehicles.

Figure 25 indicates the fact that the OS_CFAR detection performance is the worst case when there are six interfering target vehicles. The OS-CFAR detector presents a good detection performance if there are five or less interfering target vehicles at N 20 and k 15 . In this case, the OS-CFAR detector can handle the number of interfering target vehicles that is not more than N k 5. Thus, under the condition N 20 five interfering target vehicles are a cutoff case for the OS-CFAR detector and any additional interfering target causes a serious degradation in the detection performance.

The simulation confirms robustness of the OS-CFAR, GCML, and ACGO-CFAR detectors in multiple target scenario. A prior knowledge about the number of interfering target vehicles is needed in the case of the OS-CFAR detection to maintain a good detection performance and determine the cutoff interference conditions. The GCML and ACGO-CFAR detectors are robust with respect to the interfering target vehicles independently of their number.

6. DEFINITION OF TARGET VEHICLE COORDINATES

There are several procedures to define and estimate the target vehicle position or coordinates using the radar sensor systems. Some of these procedures are based on the mechanical and/or electronic scanning radar sensor systems that can be classified as an expensive solution for large-scale production. The use of multiple radar sensors to form a radar sensor network system is a practical solution to define the target vehicle position or coordinates employing the multilateration method based on definition of target vehicle ranges only. The case when there is a network of two radar sensors is shown in Figure 26. For each radar sensor system or network, the target vehicle position or coordinates can be determined as an intersection point of circles with the radius equal to the determined target vehicle range, the centers of which are the radar sensors on the subject vehicle. Each radar sensor measures the target vehicle range with error. An iterative approximation algorithm based on the least square estimation is applied to determine the target vehicle position or coordinates [26].

Figure 26. Definition of the target vehicle coordinates using network of two radar sensors.

Assume the actual position of the target vehicle in the Cartesian coordinates can be defined as

$$\mathbf{p} = [x_t, y_t] . \tag{30}$$

The two radar sensor network positions in the Cartesian coordinates are known and given by

$$\mathbf{s}1 = [x{\mathbf{s}1}, y{\mathbf{s}_1}], \ \mathbf{s}2 = [x{\mathbf{s}2}, y{\mathbf{s}_2}]$$ (31)

Based on (30) and (31), the difference in positions or coordinates between the target vehicle and the first and second radar sensors of the subject vehicle defined by the vectors l and l can be presented in the following form, respectively:

$$\mathbf{l}{1} = \mathbf{s}{1} - \mathbf{p} = [x{\mathbf{s}{1}} - x{t}, y{\mathbf{s}{1}} - y{t}], \quad \mathbf{l}{2} = \mathbf{s}{2} - \mathbf{p} = [x{\mathbf{s}{2}} - x{t}, y{\mathbf{s}{2}} - y{t}],$$ (32)

$$r_{m1} = \sqrt{(x{\mathbf{s}_1} - xt)^2 + (y{\mathbf{s}_1} - y_t)^2} + w1$$ , $r{m2} = \sqrt{(x{\mathbf{s}_2} - xt)^2 + (y{\mathbf{s}_2} - y_t)^2} + w_2$ , (33)

where , 1,2 rm i is the target vehicle range defined by the radar sensor, and wi ,i 1,2 is the additive noise.

The equation (33) shows that a relation between the measured range of the target vehicle and coordinates of the target vehicle is expressed by the non-linear function. This non-linear function can be presented in the following form:

$$\mathbf{z} = \mathbf{h}(x) + \mathbf{w} \,, \tag{34}$$

where z is the vector of target vehicle range, h(x) is the non-linear function vector, and w is the additive noise vector. The non-linear equation can be converted into the linear equation by linearization procedure near the target vehicle coordinates as follows:

$$r_{mi} = r{mi}^* + \frac{\partial r{m_i}^}{\partial xt} \bigg|{p^} (x_t - xt^*) + \frac{\partial r{m_i}^}{\partial yt} \bigg|{p^} (y_t - y_t^*) + F(e),$$ (35)

where , 1,2 r i ^mi is the distance between the target vehicle coordinates p and the radar sensor network, and F(e) is the error term. The target vehicle coordinates are given by

$$\mathbf{p}^* = [x_t^, y_t^] . \tag{36}$$

Neglecting the error term F(e) in (35) we obtain the following equation:

$$\Delta \mathbf{z} = \mathbf{H} \times \Delta \mathbf{x}^* = \begin{bmatrix} r_{m1} - r{m1}^* \ r{m2} - r{m2}^* \end{bmatrix} = \begin{bmatrix} \frac{\partial r{m_1}^}{\partial xt} \Big|{p^} & \frac{\partial r_{m_1}^}{\partial yt} \Big|{p^} \ \frac{\partial r_{m_2}^}{\partial xt} \Big|{p^} & \frac{\partial r_{m_2}^}{\partial yt} \Big|{p^} \end{bmatrix} \times \begin{bmatrix} x_t - x_t^* \ y_t - y_t^* \end{bmatrix},$$ (37)

where z is the vector of difference between the real target vehicle range and the expected target vehicle range, x is the vector of errors in determination of the target vehicle coordinates, and H is the normalized distance between the radar sensors and target vehicle in the vector form determined as follows

$$\mathbf{H} = \begin{bmatrix} -\frac{x{\mathbf{s}{1}} - x{t}}{\sqrt{(x{\mathbf{s}{1}} - x{t})^{2} + (y{\mathbf{s}{1}} - y{t})^{2}}} & -\frac{y{\mathbf{s}{1}} - y{t}}{\sqrt{(x{\mathbf{s}{1}} - x{t})^{2} + (y{\mathbf{s}{1}} - y{t})^{2}}} \ -\frac{x{\mathbf{s}{2}} - x{t}}{\sqrt{(x{\mathbf{s}{2}} - x{t})^{2} + (y{\mathbf{s}{2}} - y{t})^{2}}} & -\frac{y{\mathbf{s}{2}} - y{t}}{\sqrt{(x{\mathbf{s}{2}} - x{t})^{2} + (y{\mathbf{s}{2}} - y{t})^{2}}} \end{bmatrix}.$$ (38)

The least square solution of linearized equations is obtained by minimizing the weighted sum of squared deviations [17]:

$$\frac{\partial \mathbf{J}}{\partial \Delta \mathbf{x}} = 0 , \tag{39}$$

where J is given by

$$\mathbf{J} = [\Delta \mathbf{z} - \mathbf{H} \Delta \mathbf{x}^]^T \mathbf{Cov}^{-1} [\Delta \mathbf{z} - \mathbf{H} \Delta \mathbf{x}^], \tag{40}$$

where Cov is the covariance matrix of noise given by the following form:

$$\mathbf{Cov} = E[\mathbf{ww}^T] = \begin{bmatrix} \sigma_1^2 & 0 \ 0 & \sigma_2^2 \end{bmatrix}. \tag{41}$$

In (40) and (41), T is denotes the matrix transpose. The solution of (39) gives us the error in determination of the target vehicle coordinates:

$$\Delta \mathbf{x}^* = [\mathbf{H}^T \mathbf{Cov}^{-1} \mathbf{H}]^{-1} \mathbf{H}^T \mathbf{Cov}^{-1} \Delta \mathbf{z}. \tag{42}$$

When the radar sensor networks operate under the same noise conditions, i.e., the variance of noise is the same for all radar sensors, (42) can be simplified and rewritten in the following form:

$$\Delta \mathbf{x}^* = [\mathbf{H}^T \mathbf{H}]^{-1} \mathbf{H}^T \Delta \mathbf{z} . \tag{43}$$

Using the error in determination of the target vehicle coordinates given by (43), the process of definition of the target vehicle coordinates is stopped when the error becomes negligible (the error is converged to infinitesimal value after the definite number of iterations). The convergence of error is usually observed after four or five iterations. All previous steps are presented in Figure 27.

The steps of target vehicle coordinates definition can be simply generalized by the following way:

Figure 27. Steps of iteration under determination of the target vehicle coordinates.

• Initiate the iteration procedure assuming the target vehicle coordinates are given;
• Define the matrix H and linearized equations;
• Define the error in determination of the target vehicle coordinates;
• Correct the selected target vehicle coordinates at the first step using the obtained vector of errors x ;
• Repeat the steps from the starting point if the error is not converged.

The calculation of the target vehicle coordinates using the multilateration technique is carried out for the radar sensor network consisting of two radar sensors under the effect of the additive Gaussian noise (see Figure 26). The vectors 1 s and 2 s of coordinates of the two radar sensors in the Cartesian coordinates take the following form: 0 , 1 s¹ and 0 ,1 s¹ . The vector p of the real target vehicle Cartesian coordinates takes the following form: p 50,5. The calculation is carried out for the following initial target vehicle coordinates x and y: 1) x 1, y 1; 2) x 5, y 5; 3) x 10, y 10 .

Figure 28 presents the calculation results using the multilateration technique when the three initial target vehicle coordinates are assigned before. The calculated coordinates of the target vehicle are converged to the real coordinates of the target vehicle after five iterations for all cases. The multilateration technique based on the least square estimation is considered as a robust estimation procedure for definition of the target vehicle coordinates.

7. GD MAIN FUNCTIONING PRINCIPLES

7.1. General Statement

The simple GD model constructed based on the generalized approach to signal processing in noise [27] is represented in Figure 29. We use the following notations: MSG is the model signal generator (local oscillator), the AF is the additional filter (the linear system), and the PF is the preliminary filter (the linear system). A detailed discussion of the AF and PF can be found in [27, 28].

Consider briefly the main statements regarding the AF and PF. There are two linear systems at the GD front end that can be presented, for example, as the band-pass filters, namely, the PF with the impulse response and the AF with the impulse response ( ) hAF . For simplicity of analysis, we think that these filters have the same amplitudefrequency responses and bandwidths. Moreover, a resonant or centered frequency of the AF is detuned relative to a resonant or carrier frequency of the PF on such a value that the target return signal cannot pass through the AF, for example, on a value that is more the signal bandwidth. Thus, the target return signal and noise can be appeared at the PF output and the only noise is appeared at the AF output. It is well known, if a value of detuning between the AF and PF resonant frequencies is more than ^a 4 5f , where ^a f is the signal bandwidth, the processes forming at the AF and PF outputs can be considered as independent and uncorrelated processes. In practice, the coefficient of correlation is not more than 0.05. ( ) hPF

Figure 28. Definition of target vehicle coordinates applying the multilateration technique: a) x 1, y 1; b) x 5, y 5; c) x 10, y 10

Figure 29. Principle flowchart of GD.

In the case of a "no" target return signal in the input process, the statistical parameters at the AF and PF outputs are the same, because the same noise is coming in at the AF and PF inputs (the ideal case). We may think that the AF and PF do not change the statistical parameters of input process, since they are the linear GD front end systems. By this reason, the AF can be considered as a generator of reference sample with a priori information a "no" signal obtained in the additional reference noise at the AF output.

There is a need to make some comments regarding the noise forming at the PF and AF outputs. If the Gaussian noise w(t) comes in at the AF and PF inputs (the GD linear system front end), the noise at the AF and PF outputs is Gaussian, too, because the AF and PF are the linear systems and, in a general case, take the following form:

$$\xii = \sum{i=1}^K h_{PF} w_i$$ and $\etai = \sum{i=1}^K h_{AF} w_i$ , (44)

where i is the sample of observed noise at the PF output, i η is the sample of observed noise at the AF output, and K is the sample size (i 1,K) . If AWGN with zero mean and twosided power spectral density 5 ⁰ 0. N is coming in at the AF and PF inputs (the GD linear system front end), then the noise at the AF and PF outputs is Gaussian with zero mean and variance given by

$$\sigma_n^2 = \frac{N_0 \omega_0^2}{8\Delta_F} \tag{45}$$

where, in the case if AF (or PF) is the RLC oscillatory circuit (resistor, inductor, capacitor), the AF (or PF) bandwidth and resonance frequency 0 are defined in the following manner F

$$\Delta_F = \pi \beta, \omega_0 = \frac{R}{\sqrt{LC}}, \beta = \frac{R}{2L}$$ , where R is the resistance, L is the inductance, and C is the capacitance, respectively.

The main functioning condition of GD is the equality over the whole range of parameters between the model signal forming at the GD MSG output and the detected target return signal at the GD input liner system (the PF) output. How we can satisfy this condition in practice is discussed in detail in [27, 28]. More detailed discussion about a choice of PF and AF and their impulse responses is given in [29].

7.2. GD Structure

Under conditions closed to practice, we can use the GD flowchart presented in Figure 7.1, see [28], in which the threshold apparatus (THRA) device defines the GD threshold, and the signal model generator switching apparatus (SGSA) is used to switch on the MSG to define the unknown parameters of the detected target return signal. The decision block (DB) with the decision function employs the decision-making rule of the GD. The switch K1 takes the position "1" to define the GD detection threshold, and takes the position "2" after definition of the GD threshold and after that a detection procedure of the target return signal

is carried out. The switch K2 works to put the THRA in and out of service.

Let in GDi Z be the GD input stochastic process observed within the limits of the time interval [0,T] ; ai is the sample of target return signal; wi is the sample of the observed additive Gaussian noise with zero mean and known variance 2 n . The detection problem of the target return signal from the target vehicle can be presented in the following form:

$$Z_{GD_i}^{in} = \begin{cases} a_i + w_i & \Rightarrow H_1; \ w_i & \Rightarrow H_0; \end{cases}$$

$$(46)$$

where *H*1 is the hypothesis about the presence of the target return signal from the target vehicle; and *H*0 is the alternative hypothesis. The signal at the GD PF output in the time domain can be defined as Yⁱ a^{i i} . The GD AF output is the reference noise i η given by (44). The model signal is defined as:

$$a_i^* = \gamma a_i \tag{47}$$

where i a is the sample of generated model signal, is the factor of the proportionality, and i 1,K.

The use of the generalized approach to signal processing in noise assumes some modifications concerning the initial premises of classical and modern signal detection theories. The signal detection algorithm constructed based on generalized approach to signal processing in noise can be presented in the following form [27, 28]:

$$\sum_{i=1}^{K} 2Y_i ai^* - \sum{i=1}^{K} Yi^2 + \sum{i=1}^{K} \etai^2 \ge{H_0}^{H1} THR{GD},$$ (48)

where Yi is the stochastic process sample observed at the PF output and THRGD is the GD threshold. The first term in (48) corresponds to the correlation receiver, which is optimal in the Neyman-Pearson criterion sense, when the target return signal parameters are known a priori, with twice the gain and is considered as the sufficient statistics of the likelihood mean. The second term in (48) corresponds to the energy detector, which is optimal in the Neyman-Pearson criterion sense, when the target return signal parameters are unknown a priori, and is considered as the sufficient statistics of the likelihood variance [27, 28]. The third term in (48) presents the reference noise power at the AF output generated according to the main functioning principles of the generalized approach to signal processing in noise [27, 28]. Equation (48) represents the decision-making rule under implementation of the generalized approach to signal processing in noise in any signal processing system.

In the case of target return signal presence (the hypothesis ), Yⁱ a^{i i} , the left side of (48) takes the form K i i K i i K i a^{i 1} 1 ² where a K ^{i ai E} 1 is the signal energy, and K i i K i i 1 1 ² is the background noise at the GD output. The background noise is a difference between the noise power at the GD PF and GD AF outputs. In the opposite case, *H*1

i.e. the target return signal is absent (the hypothesis *H*0 ), Y^{i i} and the left side of (48) is the background noise K i i K i i 1 2 1 ² only. In the ideal case as K the GD background noise K i i K i i 1 1 ² tends to approach zero in the statistical sense.

7.3. GD Threshold for Radar Sensor Systems

According to the main functioning principles of GD, the target return signal must exceed the GD threshold by absolute value before the target return signal parameters would be defined. The fixed GD threshold can be used if the noise power or variance at the GD input is constant. Thus, setting the GD threshold by absolute value more than the background noise power or variance at the GD output may result in the acceptable probability of detection with the low false alarm rate. This condition cannot be guaranteed in the case of unexpected external sources of noise in different radar sensor systems. The main parameters used to determine the GD threshold are the background noise power or variance at the GD output and the probability of false alarm PFA . Applying the Neyman-Pearson criterion to GD, the probability of false alarm PFA is determined in the following form

$$P{FA} = \int{THR{GD}}^{\infty} f{Z_{GD}^{H_0}}(x) dx, \qquad (49)$$

where ( ) f ^H x GD Z is the background noise pdf at the GD output corresponding to the hypothesis *H*0 .

Under consideration of the hypothesis in the narrowband noise case with the Rayleigh amplitude envelope and the phase uniformly distributed within the limits of the interval [0,2] , the pdf ( ) f ^H x GD Z can be presented in the following form: *H*0

$$f{Z{GD}^{H_0}}(x) = \begin{cases} \frac{1}{4\sigma_n^2} \exp\left{-\frac{|x|}{2\sigma_n^2}\right}, & x \ge 0\ 0, & x < 0. \end{cases}$$ (50)

The background noise pdf given by (50) is defined by the exponential-type law when the observation time interval [0,T] is infinitesimal, i.e. T 0 . Based on (49), the probability of false alarm PFA can be presented in the following form [29]:

$$P{FA} = \int{THR{GD}}^{\infty} f{Z_{GD}^{H0}}(x) dx = \frac{1}{2} \exp\left{-\frac{THR{GD}}{2\sigma_n^2}\right}.$$ (51)

According to (51), the GD threshold can be determined in terms of the probability of false alarm PFA in the following form:

$$THR_{GD} = -2\sigman^2 \ln(2P{FA}). \tag{52}$$

Introducing the GD scaling factor 2ln(2 ) GD PFA , the GD threshold can be written as

$$THR{GD} = \alpha{GD}\sigma_n^2. \tag{53}$$

Taking into consideration a definition for the scaling factor, the probability of false alarm PFA can be presented in the following form:

$$P{FA} = \frac{1}{2} \exp\left{-\frac{\alpha{GD}}{2}\right}. \tag{54}$$

The last equation can be used to determine the probability of false alarm PFA for a given scaling factor GD or to determine the required scaling factor GD for the assigned probability of false alarm PFA .

7.4. Noise Power Estimation for GD

In order to maintain the constant probability of false alarm PFA required to achieve the acceptable detection performance stability, the GD threshold should be a function of the noise power or variance at the GD AF output, i.e. the GD threshold should be adaptive. Based on the previous discussion, we see that there is a need to determine the noise power in real time. Thus, there is a need to apply the noise power estimation technique in order to define the GD threshold. For this purpose, the estimation procedure should be valid even in the case of high noise power depending on the noise standard deviation. Owing to the fact that the bandwidths of the expected target return signals of interfering target vehicles are matched with the GD PF bandwidth, the noise power estimation is not affected by these target return signals of the interfering target vehicles. In the case when the bandwidth of interference from jammer or any other sources is matched with the GD AF bandwidth, the noise power estimation procedure will be affected.

Figure 30. Noise power estimation for GD.

Different noise power estimation procedures can be modified to be implemented in GD [30, 31]. The procedures based on the sliding windows techniques allowing us to estimate the noise power by processing a group of reference cells would be suitable solutions in the case of GD employment in radar sensor systems. For sliding window technique, the noise power estimation is performed assuming that the noise samples stored in the reference cells are i.i.d. Cell averaging procedure is a simple and effective method to process the reference cells of sliding window with the purpose to estimate the noise power using the noise sample N , , ¹ (the reference noise) at the GD AF output (see Figure 30).

The noise power estimation using the cell averaging by N reference cells of sliding window can be presented in the following form:

$$\hat{\sigma}n^2 = \frac{1}{N} \sum{i=1}^n \eta_i^2 \ . \tag{55}$$

The GD threshold can be defined adaptively based on the noise power estimation substituting (55) into (52). The number of reference cells N of sliding window is chosen in accordance with tradeoff between the SNR losses, processing complexity, and noise power estimation accuracy.

8. GD PERFORMANCE

The basic GD output signals are presented in Figure 31. The diagram shown in Figure 31a is the background noise at the GD output when there is no target return signal at the GD input (the statistical mean is equal to zero). The diagram shown in Figure 31b presents the GD energy detector output when there is the target return signal at the GD input. We can observe a shift toward the negative region of amplitude of the GD output signal, i.e. the statistical mean is not equal to zero that is a characteristic of a "yes" target return signal, but we do not know this target return signal is caused by the target vehicle at the GD input or not. To answer on this question there is a need to define the parameters of the target return signal. For this purpose we use the GD correlation receiver. The diagram shown in Figure 31c demonstrates the amplitude of the GD output signal when a "yes" target return signal at the GD input and the GD correlation receiver and energy detector operate simultaneously. The process observed at the GD output is a characteristic of the target return signal power, i.e. we are able to define some parameters of the target return signal and make the following decision: this target return signal is caused by the target vehicle at the GD input or not.

The GD detection performance is compared with the CA-CFAR detection one when there is no any noise power estimation procedure applied to GD under assumption that the noise power is known. The CA-CFAR detector uses the sliding window technique with the number of cells in the reference window equal to 20, i.e. N 20 , Figure 32. The CA-CFAR detector has the better detection performance at SNR 2 dB that is a region of the very low probability of detection PD which cannot be used in practice for any signal processing application. At SNR 2 dB, the GD demonstrates the better detection performance. We obtain the critical SNR equal to 2 dB under the simulation parameters mentioned in Table 3.

Figure 31. Basic signal at the GD output: a) the background noise; b) the GD energy detector output; c) the GD output under simultaneous operation of the correlation receiver and energy detector

Figure 32. Comparison of the CA-CFAR detector and GD by detection performance: a "no" noise power estimation for GD.

Figure 33 demonstrates the GD and CA-CFAR detection performance when the noise power estimation procedure based on the sliding window technique with the cell averaging and the number of reference cells equal to 20, i.e. N 20 is applied to both detectors. At the low probability of detection , for example, P^D 0.1 that is not a practical case for all types of signal processing systems the CA-CFAR detector overcomes the GD by detection performance. If P^D 0.5 , the acceptable case in practice, the GD demonstrates superiority in detection performance for wide range of SNR in comparison with the CA-CFAR detection. For example, at P^D 0.8 , the required SNR for CA-CFAR detection is 15 dB, while the required SNR for the GD is 13 dB. Based on Figures 32 and 33, we can see that the GD performance is deteriorated using the noise power estimation technique. The GD performance PD

degradation is predictable since the estimated noise power is not exactly the same by value as the real noise power owing to noise power estimation errors.

Figure 33. Comparison of the CA-CFAR detector and GD by detection performance: a "yes" noise power estimation for GD.

According to the conditions of GD PF and AF choice, the reference noise at the GD AF output is used to estimate the noise power and define the GD threshold. Thus, the target return signals from the desired and/or interfering target vehicles cannot pass through the GD AF [27–32]. The signals reflected from the interfering target vehicles (the interference signals) have the same statistical characteristics as the target return signals from the desired target vehicles, see (10). The bandwidth of interfering signals is matched with the GD PF bandwidth.

If there are no interfering target vehicles, the GD output takes the following form:

$$Z_{GDi}^{out} = \sum{i=1}^{K} (2Y_i a_i^* - Y_i^2 + \eta_i^2)$$ (56)

The observed sample at the GD PF output when there is an interfering signal I(t) can be presented in the following form:

$$Y_i = a_i + I_i + \xi_i, \tag{57}$$

where i I is the interfering signal sample at the GD PF output. If the condition aⁱ aⁱ is satisfied, the GD output takes the following form:

$$Z_{GDi}^{out} = \sum{i=1}^{K} (a_i^2 - 2I_i\xi_i - I_i^2 + \eta_i^2 - \xi_i^2),$$ (58)

where I is the interfering signal energy and i i I is the component caused by the interaction between the noise and interference.

Additional comparison between the GD and CA-CFAR detector by detection performance is performed for the following cases:

• There is only one interfering target vehicle;
• There are two interfering target vehicles.

As before, we assume the bandwidth of interfering signals is matched with the GD PF bandwidth. This comparison is presented in Figure 34. The GD noise power estimation procedure is based on the sliding window technique with the cell averaging for cells of the reference window as in the case of CA-CFAR. The GD demonstrates a good robustness with respect to the interfering signals matched by bandwidth with the GD PF, because these interfering signals do not effect on the noise power estimation procedure and, consequently, on the definition of the GD detection threshold THRGD . Thus a superiority of GD in detection performance over the CA-CFAR detector is evident in the case of one and/or several interfering target vehicles.

Figure 34. Comparison of the CA-CFAR detector and GD by detection performance for two cases: there are no the interfering target vehicles and there are the interfering target vehicles. The bandwidth of interfering signals is matched with the GD PF one.

9. ATTENUATION EFFECT ON THE RADAR SENSOR PERFORMANCE

This section is about how some specific attenuation exerts on the radar sensor detection performance. In practice, under implementation of any radar sensor system, the empirical detection performance will be different from the theoretical detection performance owing to many reasons. One of them is a signal attenuation caused by various sources. The main goal of this section is to clarify the attenuation effect on the detection performance of the target return signal from target vehicles. This effect is analyzed theoretically and the simulation is made for CA-CFAR detection scheme. The obtained results can be generalized to other CFAR detectors and GD because the detection performance degradation caused by attenuation is the same regardless of the detector type.

9.1. Signal Attenuation and Radar Range Equation

If the electromagnetic signal is propagated through any environment, b for example, air or atmosphere, an interaction with molecules of environment absorbs the signal energy. Thus, the signal power is decreased with increasing in the target vehicle range. Introducing the attenuation factor af per meter, the relation between the received signal power Pr and transmitted signal power Ptr can be written as follows:

$$Pr = P{tr} \exp(-2\alpha_{af}R) \tag{59}$$

where R is the target vehicle range. We can express the attenuation factor af in dB/km using the basic logarithmic statement given as

$$\log_a X = \frac{\log_b X}{\log_b a} \tag{60}$$

It is easy to find that lg X 0.43429ln X . Thus, the linear attenuation factor af can be presented in the following form:

$$\alpha{dB} = 4342.9\alpha{af} \text{ dB/km.} \tag{61}$$

From (57) and (59), we can find that:

$$10\lg Pr = 10\lg P{tr} - 2\alpha_{dB}R, (62)$$

where R is target vehicle range expressed in km.

The general attenuation factor gen af is the sum of several attenuation factors representing different attenuation effects such as the atmosphere, rain, fog, dust, snow, sand, and others. We consider the following representation of the general attenuation factor:

$$\alpha{af}^{gen} = \alpha{atm} + \alpha{rain} + \alpha{fog} + \alpha{dust} + \alpha{etc}, \qquad (63)$$

where atm is the atmospheric attenuation factor; rain is the rain attenuation factor; fog is the fog attenuation factor; dust is the dust attenuation factor; and etc is the attenuation factor caused by other sources.

9.2. Atmospheric Attenuation

The atmospheric attenuation factor atm is a function of the operation frequency of the radar sensor system. Figure 35 demonstrates the atmospheric attenuation factor atm as a function of specific atmospheric parameters, namely, the pressure, temperature, and density of water vapour, i.e. humidity. As we can see from Figure 35, we observe the smallest atmospheric attenuation factor at the operation frequency of the radar sensor system equal to 24 GHz in comparison with radar sensor systems functioning under the operation frequencies equal to 60 GHz, 70 GHz, and 120 GHz.

Figure 35. The atmospheric attenuation factor values as a function of the operation frequency.

Figure 36. Specific attenuation due to atmospheric conditions (ideal parameter).

Figure 36 presents a comparison between the total atmospheric attenuation factor and the dry air and water vapour attenuation factors as a function of the operation frequency of radar sensor systems under the ideal parameters at the sea level, i.e. the pressure, temperature, and density of water vapour [33, 34]. We can see that at the operation frequency of radar sensor system equal to 24 GHz, the total atmospheric attenuation factor is lesser 0.2 dB/km.

As we mentioned before, the atmospheric parameters effect on the attenuation factor. By this reason, we present in Figures 37 and 38 the same comparison as in Figure 36 for different

Figure 37. Specific attenuation due to atmospheric conditions (temperature =273.15 K).

Figure 38. Specific attenuation due to atmospheric conditions (air pressure = 898.75 hPa).

values of air temperature and pressure, respectively. We see that the total attenuation factor is independent of air temperature when the weather is cold (the low air temperature).

Figure 38 demonstrates the total atmospheric attenuation factor and the dry air and water vapour attenuation factors as a function of the operation frequency of radar sensor systems when the air pressure is changed. In reality, this situation can be happened at high altitudes above the sea level where the air pressure is low. We observe that the total atmospheric attenuation factor is not changed considerably when the air pressure is varied in comparison with the ideal case.

9.3. Rain Attenuation

The rain attenuation effects on the radar sensor detection performance are strongly related to rain intensity that is called the rainfall rate. For example, that light rain having a uniform spatial distribution can be predicted by relatively easy way. However, the heavy rain comes jointly with squall and it is impossible to predict the rain intensity and drop size distribution [33]. Mie scattering theory [35] is generally applied to frequencies up to the submillimeter wave band. In this case, the rain attenuation factor can be determined using the following form:

$$\alpha_{rain} = \mu L^{\beta} , \qquad (64)$$

where the rain attenuation factor rain has a dimension of dB/km, L is the rain rate expressed in mm/hour, and are the coefficients depending on the operation frequency of radar sensor system. Table 5 presents the values of the coefficients and at the operation frequency of radar sensor system equal to 24 GHz.

Figure 39 presents the rain attenuation factor for different kinds of rain (the rain density) as a function of the operation frequency of radar sensor system.

Figure 39. Rain attenuation factor.

Table 5. Empirical coefficients used to determine the rain attenuation

Coefficient	24 GHz
	0.19
	0.905

The rain attenuation factor at the operation frequency of radar sensor system equal to 24 GHz is presented in Table 6 for different types of the rain, namely, the drizzle rain, the light rain, the medium rain, the heavy rain, the downpour rain, and the tropical rain.

Rain Type	Attenuation Factor [dB/Km]	Rainfall Amount [mm/hr]
Drizzle	0.019	0.25
Light rain	0.089	1.25
	0.3	2.5
Medium rain	1.5	12.5
Heavy rain	3.5	25
Downpour	7	50
Tropical	19	100

Table 6. Rain attenuation factor at 24 GHz

Table 7. The CA-CFAR detector probability of detection under atmospheric and rain attenuation

Type	Probability of Detection (SNR=9.2 dB)
No Attenuation	0.7123
Atmospheric	0.6871
Atmospheric + Drizzle Rain (0.25 mm/hr)	0.6834
Atmospheric + Light Rain (1.25 mm/hr)	0.6821
Atmospheric + Medium Rain (2.5 mm/hr)	0.6800
Atmospheric + Medium Rain (12.5 mm/hr)	0.6135
Atmospheric + Heavy Rain (25 mm/hr)	0.5452
Atmospheric + Downpour (50 mm/hr)	0.3681
Atmospheric + Tropical (100 mm/hr)	0.0827

Figure 40. Rain attenuation factor action on the CA-CFAR detection performance.

Simulation results given in Table 7 demonstrate the probability of detection for CA-CFAR detector under various values of the atmospheric attenuation factor and combination between the atmospheric and rain attenuation factors. Simulation is carried out at the specific SNR 9.2 dB and the target vehicle range equal to 140 m. From the results presented in Table 7, we see that the effect of the atmospheric and low (the drizzle, light, and medium rain when the rainfall amount is less than 12.5 [mm/hr]) rain attenuation does not change considerably the probability of detection of radar sensor system. Starting from the case of heavy rain density, a degradation of the detection performance is evident, especially, in the case of the downpour and tropical rain.

Parameter	Value
Carrier Frequency	24 GHz
Probability of False Alarm PFA	10-3
Number of Reference Cells N	20
Target Range (one way)	70 m
Temperature	20◦C
Moisture Content	7.5 g/m3

Table 8. The general simulation parameters

The simulation results concerning the rain attenuation effects on the CA-CFAR detection performance are shown in Figure 40. These results are obtained based on the simulation parameters presented in Table 8. As follows from Figure 40, the detection performance of the CA-CFAR detector suffers from serious degradation in the case of downpour and tropical rain. Additionally, we can conclude that the attenuation caused by the rain exerts more influence on the detection performance of radar sensor system in comparison with the atmospheric attenuation. The curves in Figure 40 corresponding to the attenuation factors for the case of ideal, atmospheric, atmospheric plus light rain, and atmospheric plus medium rain are very close to each other and it is very difficult to distinguish a difference between them.

9.4. Fog Attenuation

The fog is created by the atmospheric water vapour condensation into water droplets suspended in the air. Thus, the fog can be defined as the water droplet density that restricts or reduces a visibility to less than 1 km where advection of the fog is caused by horizontal movement of a warm air mass. Another type of the fog called the radiation fog is caused by air cooling overnight. This type of the fog is often occurs around the rivers, lakes, and swamps. An empirical equation that relates a visibility of liquid water content in the case of advection fog is given by [34, 36]

$$LW = 1660v^{-1.43}, (65)$$

where LW is the liquid water content expressed in g/m³ , and is the visibility expressed in foot, 1 foot is equal to 0.3048 m. The Rayleigh approximation can be used to evaluate the fog attenuation factor for frequencies that are specific for micro- and millimeter waves [36]. Thus, the fog attenuation factor can be presented in the following form [33], [35]:

$$\alpha_{fog} = \varepsilon LW$$ , (66)

where fog is the fog attenuation factor expressed in dB/km, and is the attenuation coefficient expressed in dB/km per g/m³ at a specific operation frequency of radar sensor system, Figure 41.

Figure 41. Fog attenuation factor as a function of operation frequency.

Figure 42 presents the detection performance of the CA-CFAR detector under the atmospheric plus fog attenuation factors. As we can see from Figure 42, the heavy and serious types of fog are the reason of dramatic degradation in the detection performance. Other fog types have not considerable effects on the detection performance. It is difficult to distinguish the detection curves for the ideal case, the atmospheric attenuation, and the light, thick, and dense types of fog. The visibility and attenuation factors are presented in Table 9 for different types of fog.

Fog Type	Visibility [m]	Attenuation [dB/Km]
Serious Fog	3 – 10	40 – 7.21
Heavy Fog	10 – 40	7.21 – 0.99
Dense Fog	40 – 70	0.99 – 0.45
Thick Fog	70 – 250	0.45 – 0.07
Moderate Fog	250 – 500	0.07 – 0.02
Light Fog	500 – 1000	0.02 – 0.01

Table 9. Fog densities with related visibility and attenuation factors

Figure 42. Fog attenuation factor action on CA-CFAR detection performance.

9.5. Dust Attenuation

A widely used method to characterize the density of aerosols suspended in the atmosphere or air is based on visibility of any target vehicle. In this case, the visibility can be defined as the target vehicle range if a high contrast target vehicle would be observed clear. Taking into consideration the Rayleigh approximation, the dust attenuation factor of radar sensor systems can be presented in the following form [34, 37]:

$$\alpha{dust} = \frac{2.317 \times 10^{-3}}{[(\delta{\gamma} + 2)^2 + \delta_{im}^2]\lambda} \times \frac{1}{\nu^{1.07}} \text{ [dB/km]},$$ (67)

where is the real term of permittivity of the dust particle, im is the imaginary term of permittivity of the dust particle, and is the wavelength. Figure 43 presents the relationship between the visibility and attenuation factor in the case of coal dust [34].

Figure 44 demonstrates a comparison between the water attenuation factor and coal dust attenuation factor. From Figure 44, we can see that an action of the dust attenuation factor is very weak in comparison with the water attenuation one [34].

Figure 45 presents the detection performance of the CA-CFAR detector for the ideal case, the atmospheric attenuation factor, and the atmospheric plus dust attenuation factors. The dust attenuation factor is varied in accordance with variations in visibility from 0.5m to 50m. As follows from Figure 45, the high dust density conditions are the reason of small visibility ranges (less than 1 m) causing a high deterioration in the detection performance.

Figure 43. Visibility as a function of the attenuation factor for coal dust.

Figure 44. Visibility versus the attenuation factor for coal dust and water droplets.

Figure 45. The dust attenuation factor action on the CA-CFAR detection performance.

In conclusion, we can state that the rain and fog attenuation factors, especially the high rainfall and heavy or serious fog, should be considered as a serious reason of deterioration in the detection performance of the CFAR detectors and any other ones. On the other hand, the atmospheric and the dust attenuation factors have less effect on the detection performance. In practice, the worst attenuation conditions should be considered. Thus, a whole set of attenuation factors must be taken into consideration, even the attenuation factors with weak action, because an accumulation of all weak attenuation factors can generate a considerable degradation in the detection performance.

CONCLUSION

Recently, the total safety driving approach has been become very popular. Any safety driving application can be implemented by installing radar sensors to build a system with passive and active vehicle safety elements. This system compensates the errors made by drivers and avoids an occurrence of accidents. Implementation of the controller area network and radar sensors to construct safety driving applications and collision avoidance systems is an effective solution in the automotive industry.

The FMCW radar sensor system is chosen to be used for both the MRR application such as CVD and the SRR application such as BSD to provide the controller area network with the required data, for example, the range, relative velocity, and target vehicle azimuth angle or position.

The main problem in any radar sensor system is to detect the target return signal within the limits of system scanning or observation area. The target return signal detection techniques depending on detection threshold in radar sensor systems have a major disadvantage with respect to sensitivity of noise power. The detection threshold should be defined based on the observed in real time noise power when the noise variance is varied as a function of time. Thus, the adaptive detection threshold is essential in order to maintain the constant probability of false alarm and stable detection performance with the predictable probability of detection. Simulation and detection performance analysis are carried out under conditions closed to practice with the purpose to evaluate the mentioned CFAR detectors employed by LFMCW radar sensor system when the interfering target vehicles are absent or present.

The CA-CFAR detector is the optimal among all other CFAR detectors in the case of homogeneous noise and absence of interfering target vehicles (no target masking phenomenon). The OS-CFAR detector suffers a minor degradation in the detection performance in comparison with CA-CFAR detector. In multiple target scenarios and in the case of non-homogeneous noise the OS-CFAR detection performance is superior in comparison with the CA-CFAR detector when we know a priori the number of interfering target vehicles. The GCML and ACGO-CFAR detectors maintain a good robustness when the number of interfering target vehicles is known a priori. Additionally, the ACGO-CFAR detector is capable to suppress the clutter edge false alarm. The GCML detector has the best detection performance in non-homogenous noise, but the main disadvantage of this detector is the great complexity and computation cost, as in the case of ACGO-CFAR detector. Thus, the OS-CFAR detector can be considered as the best tradeoff between the detection performance and complexity (low computation cost).

The GD is constructed based on the generalized approach to signal processing in noise. The engineering interpretation of the GD is a combination of the correlation receiver that is optimal in the Neyman-Pearson sense under detection of signals with a priori known parameters and the energy detector that is optimal in the Neyman-Pearson sense under detection of signals with a priori unknown parameters. This combination allows us to take into consideration a very important statistical parameter such as the likelihood function variance and formulate a decision-making rule with respect to the presence or absence of the target return signal based on definition of the jointly sufficient statistics of the mean and variance of the likelihood function.

Noise power estimation procedure based on the sliding window technique with cell averaging is presented under the GD implementation in radar sensor systems. In order to evaluate a feasibility of GD implementation with the proposed noise power estimation technique, the GD detection performance is compared with the CA-CFAR detector which is widely used in practice and demonstrates the best detection performance among whole CFAR detector family in homogeneous noise. GD overcomes the CA-CFAR detector by the detection performance under both the homogeneous and non-homogeneous noise conditions. Detection performance of GD is also better than that of other CFAR detectors in nonhomogeneous noise when the bandwidth of interfering signals is matched with GD PF one. The GD implementation allows us to reduce the SNR losses required to compensate the effect of inaccurate noise power estimation (the estimation error) and achieve the desired constant probability of false alarm PFA . The improvement in detection performance demonstrates an effectiveness of the proposed solution based on the GD implementation. Employment of the noise power estimation techniques can improve the noise power estimation quality (to reduce the estimation error) and allows the GD to achieve the best detection performance. The problem dealing with the interfering signals matched by bandwidth with the GD AF one leads us to employment of an interference cancellation approach such as the beam forming and antenna array with or without direction of arrival estimation to eliminate the interference action on the GD performance.

Under construction of driving safety systems based on radar sensor network a whole set of attenuation factors must be taken into consideration, even the attenuation factors with weak action, because an accumulation of all weak attenuation factors can generate a considerable degradation in the detection performance.

ACKNOWLEDGMENTS

This research was supported by SL Corporation and Industry Academic Cooperation Foundation of KNU within the limits of the research project on design of signal detection algorithms for car applications, Grant No. 20101459000 and Kyungpook National University Research Grant 2012.

REFERENCES

• [1] Shbat, M. S.; Khan, Md. R.; Yi, J. H.; Lee, I. B.; Tuzlukov, V.; Modern Radar Systems and Signal Detection Algorithms for Car Applications. Fizika, vol. XVII (1-en), pp. 29- 60, 2011.
• [2] Hosaka, A.; Mizutani, H.; Improvement of Traffic Safety by Road Vehicle Cooperative Smart Cruise Systems. IATSS Research, vol. 24, no. 23, pp. 4-42, 2000.
• [3] Bloecher, H.; Sailer, A.; Rollmann, G.; Dickmann, J; 79 GHz UWB Automotive Short Range Radar - Spectrum Allocation and Technology Trends. Advances in Radio Science, vol. 7, pp. 61-65, 2009.
• [4] Mende, R.; Rohling, H.; New Automotive Applications for Smart Radar Systems. German Radar Symposium (GRS), Bonn, Germany, pp. 35-40, 2002.
• [5] Schneider, M.; Automotive Radar Status and Trends. German Microwave Conference (GeMic), Ulm, Germany, pp.144-147, 2005.
• [6] Australian Communications Authority; A Review of Automotive Radar Systems Devices and Regulatory Frameworks. Document: SP 4/01, 2001.
• [7] Kulpa, K.; Continuous Wave Radars Monostatic, Multistatic and Network. NATO Security through Science Series, Warsaw, Poland, vol. 2, pp. 215-242, 2006.
• [8] Wenger, J.; Automotive mm-Wave Radar: Status and Trends in System Design and Technology. IEE Colloquium on Automotive Radar and Navigation Techniques, London, England, Ref. No. 1998/230, pp. 1-7, 1998.
• [9] Draft International Standard ISO/DIS 17387. Intelligent Transportation Systems Lane Change Decision Aid Systems - Performance Requirements and Test Procedures. International Organization for Standardization, 2006.
• [10] Etschberger, K.; Controller Area Network, Technical Report. 2nd Ed., Munich, IXXAT Automation GmbH, 2011.
• [11] Skolnik, M.; Radar Handbook. 3rd Ed., New York, USA, McGraw Hill, 2008.

• [12] Skolnik, M.; Introduction to Radar Systems. 3rd Ed., USA, McGraw-Hill Book Company Wiley-Inter-Science, 2001.

• [13] Isakov, V; Microwave Circuit for 24 GHz Automotive Radar in Silicon Based Technologies. New York, USA, Springer, 2010.

• [14] Mahafza, B. R; Elsherbeni, A. Z; Matlab Simulation for Radar Systems Design. Chapman & Hall/CRC CRC Press LLC, 2004.

• [15] Richards, M. A.; Fundamentals of Radar Signal Processing. New York, USA McGraw-Hill, 2005.

• [16] Finn, H.; Johnson, R.; Adaptive Detection Mode with Threshold Control as a Function of Spatially Sampled Clutter-Level Estimates. RCA Review, pp. 414-464, 1968.

• [17] Kay, S.; Fundamentals for Statistical Signal Processing, Vol. 1: Estimation Theory. NJ, USA, Prentice Hall, Englewood Cliffs, 1993.

• [18] Rohling, H.; Radar CFAR Threshold in Clutter and Multiple Target Situations. IEEE Transactions on Aerospace and Electronic Systems, vol. 19, no. 4, pp. 608-621, 1983.

• [19] Himonas, S. D.; Barkat, M.; A Robust Radar CFAR Detector for Multiple Target Situations. IEEE National Radar Conference, Dallas, Texas, USA, pp. 85-90, 1989.

• [20] Himonas, S. D.; Adaptive Censor Greatest-of CFAR Detection. IEE Proceedings-f, Radar and Signal processing, vol. 139, no. 3, pp. 247-255, 1992.

• [21] Swerling, P.; Probability of Detection for Fluctuating Targets. IRE Transactions on Information Theory, IT-6, pp. 269-308, 1960.

• [22] Rohling, H.; Some Radar Topics: Waveform Design, Range CFAR and Target Recognition. Advances in Sensing with Security Applications NATO Security through Science Series, vol. 2, pp. 293-322, 2006.

• [23] Yi, J. H.; Lee, I. B.; Shbat, M. S.; Tuzlukov, V.; 24 GHz FMCW Radar Sensor Algorithms for Car Applications. International Radar Symposium (IRS 2011), Leipzig, Germany, pp. 465-470, 2011.

• [24] Shbat, M. S.; Tuzlukov, V.; Generalized Approach to Signal Processing in Noise for Closing Vehicle Detection Application Using FMCW Radar Sensor System. International Radar Symposium (IRS 2011), Leipzig, Germany, pp. 459-464, 2011.

• [25] Shbat, M. S.; Tuzlukov, V.; Signal Processing in Automotive Controller Area Network Based on Radar Sensors. 11th International Conference on Control, Automation, and Systems (ICCAS 2011), Gyeonggi-do, South Korea, pp. 616-620, 2011.

• [26] Klotz, M.; An Automotive Short Range High Resolution Pulse Radar Network Aachen. Germany, Shaker Verlag GmbH, 2002.

• [27] Tuzlukov, V.; A New Approach to Signal Detection Theory. Digital Signal Processing: A Review Journal, vol. 8, no. 3, pp. 166-184, 1998.

• [28] Tuzlukov, V.; Signal Detection Theory. Boston, Springer-Verlag, 2001.

• [29] Tuzlukov, V.; Signal Processing Noise. Boca Raton, London, New York, Washington D.C CRC Press, Taylor & Francis Group, 2002.

• [30] Shbat, M. S.; Tuzlukov, V.; Noise Power Estimation under Generalized Detector Employment in Automotive Detection and Tracking Systems. 9th IET Data Fusion and Target Tracking Conference (DF&TT"12), London, UK, doi:10.1049/cp.2012.0416, 2012.

• [31] Shbat, M. S.; Tuzlukov, V.; Generalized Detector with Adaptive Detection Threshold for Radar Sensors. 13th International Radar Symposium (IRS 2012), Warsaw, Poland, pp. 91-94, 2012.

• [32] Tuzlukov, V.; Signal Processing in Radar Systems. Boca Raton, London, New York, Washington D.C CRC Press, Taylor & Francis Group, 2012.

• [33] ITU-R P.676-6 Recommendation, Attenuation by Atmospheric Gases, 2012, http://www.itu.int/rec/R-REC-P.676/en.

• [34] Brooker, G.; Sensors and Signals. Australian Centre for Field Robotics, 2006.

• [35] Mie, G.; Contributions to the Optics of Turbid Media, Especially Colloidal Metal Solutions. Annals of Physics, vol. 330, no. 3, pp. 377-445, 1908.

• [36] Zhao, Z.; Wu, Z.; Millimeter-Wave Attenuation due to Fog and Clouds. International Journal of Infrared and Millimeter Waves, vol. 21, no. 10, pp. 1607-1615, 2000.

• [37] Goldhirsh, J.; Attenuation and Backscatter from a Derived Two-Dimensional Dust Storm Model. IEEE Transactions on Antennas and Propagation, vol. 49, no. 12, pp. 1703-1711, 2001.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 2

AIRBORNE ULTRASONIC IMAGING: SONAR BASED IMAGE GENERATION FOR AUTONOMOUS VEHICLES

Dorel Aiordachioaie* , Laurentiu Frangu, Silviu Epure, Radu Belea, Marius Mazarel, Constantin Miholca and Claudiu Chiculita

Electronics and Telecommunications Department, "Dunarea de Jos" University of Galati, Galati, Romania

ABSTRACT

Ultrasound imaging refers to images obtained by ultrasonic waves. An emitted pulse is reflected back by the environment and the pixel"s intensity is imposed by the amplitude of the reflected pulse at the place where the image is constructed.

Ultrasound images have various and important applications both in industry, for fault detection and localization, and medicine, for non-invasive diagnostic techniques. A new area is coming now in robotics for artificial vision, for solving some general tasks as recognition and navigation.

In robotics, navigation based on ultrasonic images is a new field of research, especially when discussing about airborne ultrasound images, i.e. images obtained by ultrasound waves propagating in air.

The reason of such delay in using airborne ultrasound images in robotics is explained by the high absorption of ultrasound frequencies in air, where frequencies over 200 kHz are not often used. In this front line, new technologies for image generation and processing are expected and welcomed by the market and users of mobile robots.

We have used simple sonar heads, which have a biomimetic functional behavior with one emitter and two receivers. The sonar heads have six degrees of freedom and are named ROVIBAT (ROmanian VIrtual BAT).

This chapter presents some results concerning sonar image generation process by using a fixed ultrasonic frequency. The chapter is organized in sections. Section 1

^* Corresponding author: dorel.aiordachioaie@ugal.ro.

presents the rationale to use ultrasonic images in the context of smart vehicles. Section 2 describes the structure of sonar heads. Section 3 presents the electronics of transmitter and receivers. Section 4 describes the effect of external pinnae mounted on the front of receivers. Section 5 deals with the image generation and raw processing methods for navigation purposes, including detection and recognition tasks.

Keywords: Sonar, Ultrasonic Imaging, Navigation, Signal Processing, Image Generation

1.INTRODUCTION

Ultrasound imaging points refers to 2D or 3D images obtained by ultrasonic waves. The generation technology uses an emitted pulse through target and which is reflected back. The pixel"s intensity is imposed by the amplitude of the reflected pulse at the place where the image is constructed.

Ultrasound images are used both in industry, for fault detection and diagnosis, and medicine, in diagnosis procedures. All these applications do not use air as main propagation medium.

Applications that use air as propagation medium have a major difficulty; the absorption of ultrasounds waves is very high. This makes inefficient frequencies over 200 kHz.

Ultrasound imaging devices uses arrays of ultrasound transducers, in order to control the beam width and thus to select the details in the explored environment. The ultrasound transducers are accompanied by powerful signal processors. This is an expensive and complex solution, and – some time – for simple applications discrete ultrasound transducers could offer an acceptable and still efficient solution based on classical method of image generation, based on pixel-by-pixel generation. The price for simplify is the appearance of the uncertainty in ultrasound imaging and the generation time could be huge for many applications.

The generation of airborne ultrasonic images is a challenge, the scientific literature and practical experiments are almost missing. This work uses a simple method for obtaining airborne ultrasound images, under the objective of target detection and classification. The objective requires mainly the existence of information about the shape of the objects. The methodology of using ultrasonic images for navigation is not dealt with here, by considering that any classical method of image processing from computer vision field could be a valid solution.

1.1. The Context of Smart Vehicles

In the framework of smart grounded vehicles (or cars), ultrasound imaging could be used in common applications, e.g. navigation and self-orientation, [1], and special conditions (e.g. dark or smog environments), i.e. where other well-known and used methods fail. Ultrasound imaging could also be used in special applications related to pattern recognition and classification, e.g. texture of surfaces. These are also valid for robots driven inside closed or open environments, but with a finite set of trajectories, e.g. robots working in warehouses.

Reference [2] presents another point of view, related to the use of imaging as part of the parking building for parking monitoring. "Detection methods based on cameras and image processing suffer from a lack of accuracy and can be affected by environmental or weather circumstances", [2]. The ultrasonic image system could improve the performances of camerabased systems and, in bad environmental conditions, to solve detection problems.

During the past ten years, many research projects on smart cars were considered worldwide. Most of them approach new technologies to implement automation and real time software in the components of the car. We here mention two projects: (i) The City Car of Massachusetts Institute of Technology (MIT) [3], a lightweight electric vehicle. This car employs fully integrated in-wheel electric motors and suspension systems, which are selfcontained, digitally controlled, and reconfigurable; (ii) The Cybercars project develops road vehicles with fully automated driving capabilities. [4]. More details could be found in [5] and [6], which introduce the structure of the smart car and discuss the obtained major results during projects, worldwide.

For safety reasons mainly, many cars have smart (intelligent) systems in order to assist driver in driving, e.g. collision avoidance system (CAS), side blind spot warning, lane departure control [7], automated switching of lights, park assistance systems, radar [8], pedestrian detection [9], etc. All these systems solve specific tasks, independently or driver enabled.

A smart car means something more, a car with systems, which can solve, in an automated way, various common tasks from its life, such as automated choice of the right road to reach an imposed destination (autopilot), antilock braking system (ABS), automated parking, etc. These tasks are performed independently and in an automated way.

Autonomous vehicles refer mainly to vehicles without any driver. All necessary actions, beginning with motor starting and continuing with safely driving on roads, are solved by the on-board systems of the vehicle.

Finally, it is useful to say that all sensors, inside and outside the vehicles, take into account both to the behavior of the car and driver, and - more – there it is an interaction with players of the roads, i.e. pedestrians and other cars. A short presentation of these aspects is made in [10].

Taking into account the various atmospheric phenomena during driving and the necessity to cover almost all driving conditions imposed by environment, i.e. environment with a very high intensity of light or – the other way around – with a very low light intensity (dark environments), special techniques for orientation and navigation must be considered. Airborne ultrasonic imaging could be one of them, which helps directly the car if the sonar head (which generates the ultrasonic images) is installed on it. The help can be indirect if the sonar head is working outside the car, in fixed positions, as in parking buildings.

2. THE SONAR HEAD: STRUCTURE AND BASIC ELECTRONICS

The common sonar heads used in robotics, mobile or fixed, use frequency in the range of 40 to 50 kHz and with ultrasonic transducers in fixed positions. These have relatively poor performances in the navigation of the robots especially in environment with walls and complex obstacles.

The biomimetic sonar heads mimic the anatomy of bats: the transmitter corresponds to mouth and the two ears correspond to the two receivers, left and right. The biomimetism is double, in the sense the behavior of bats is translated to similar functions of the biomimetic sonar head.

Biomimetism is used in many navigation and optimization algorithms, particularly in the navigation of robots as in [11], [12] or [13]. Biomimetic sonar heads are already available, as in [14], [15], and [16]. The basic hardware contains servomotors for the orientation of the transducers and signal processing boards, with parallel and synchronously inputs. Classical solution use external (outside of the sonar head) signal processing and control boards, which could generate constraints on the quick movement of the robot.

Figure 1 presents the structure of the used biomimetic sonar head. The biomimetism refers to the level of functions and to the degree of freedom, i.e. one emitter and two receivers. The investigated sonar head has all the processing boards (signal plus control) inside the sonar head. The sonar head is called ROVIBAT, an acronym from ROmanian VIrtual BAT. Figure 2 shows sample pictures of two versions of the sonar head.

Four main blocks can be seen in the picture: (1) the mechanics of the head, i.e. mainly bearings and servomotors (SONAR head); (2) the position control board. It computes the driving signals for the orientation of each transducer; (3) the signal processing board. It makes analogous and digital filtering, sampling and analog to digital conversions; (4). a standard computer for coordination and supervised control.

Figure 1. The structure of the sonar heads.

2.1. The Position Control Board

The Voltage Regulator protects the equipment from over voltage and provides steady regulated power. The weight is 1.4 oz.; the working voltage range is 4.8-8.0 V, and the current up to 15A. Input voltage must be at least 1.0 V greater than the intended output voltage. The regulator is ACC234 rated for 20 Watts of power dissipation. Other key features are available on technical sheet [17].

Common pulse proportional servos are used for positioning the shaft. For a range of 90^o motion pulses in the range of 0.9 to 2.1 ms are transmitted, with repetition every 20 ms. a specialized electronic board (SSC-32), [18], is used for servomotors driving. From technological point of view, SMT (Surface Mount Components) were used. The electronics of the receiver (amplifier) is placed under the transducers of the receiver. The gain of the amplifier is around 50-55 dB and the frequency cut-off around 50-55 kHz.

Figure 2. Overall pictures of the ROVIBAT_01 and ROVIBAT_02 sonar heads.

2.2. The Signal Processing Board

The signal processing board has ADC (Analog to Digital Converter) (AD7266). The sampling frequency is 2MS/s and 12 bits for resolution were used. The circuit is interfaced to the Blackfin processor through a SPORT (Synchronous Serial Peripheral Port) port. The digital signal processing is performed by specialized module based on dual core BF561 processor from Analog Devices. The working frequency is 600 MHz and provides 1200 MMACs (Million Multiply Accumulate Cycles per Second).

Figure 3 shows a block diagram of the board and an image of the signal processing board. A special module for ultrasound transducers (or ultrasound sensor array) is considered outside of the signal processing board in order to assure the modularity structure of the sonar head.

Figure 3. Block diagram and image of the signal processing board.

The boards of signal processing and position control are driven by MATLAB software. It receives commands from computer regarding the number of emitted pulses for transmission and the sampling rate for acquisition. The board then drives the emitter and acquires signals from ultrasonic receivers for a limited duration. Data from ADC circuits is transferred to SDRAM (Synchronous Dynamic Random Access Memory) using DMA (Direct Memory Access). Currently, data processing is made by PC with MATLAB software. After the algorithms are developed and tested, the software can be embedded into the board for real time processing.

3. ULTRASONIC TRANSDUCERS AND CONTROL CIRCUITS

Wide frequency ranges are more suitable for exploration of the environment, mainly because of the variable shape of the transmitted pulses, adapted to transmission channels, and - later on - it is possible to combine (aggregate) discrete ultrasonic images at various frequencies. We are interested in wide range frequency transducers, namely capacitive or piezoelectric transducers, which could be used up to 300 kHz.

Based on results reported in literature of wide range ultrasonic sensors, e.g. [19, 20, 21] for general sensing application and [22, 23, 24] for ultrasonic transducers, the EMFi (Electro Mechanical Film) material, [25], was considered.

Experiments with sensors based on these new materials show a strong relation between the frequency range, including the frequency resonance, and the mechanical structure of the ultrasonic transducer. Constraints are imposed also by the connecting (gluing) technology of the terminals. Now of working, this technology is not mature to provide reliable ultrasonic transducers.

By considering the parameters of the EMFi material, [25], as sensitivity of S^q = 25-250 [pC/N] and the specific capacitance of 40 [pF/cm² ], and by using small loads for transducer, i.e. resistors with resistance greater than 1-5 MΩ, the output voltage for a P [N/cm² ] pressure variation is:

$$\Delta V = (1/C_0) \cdot S_q \cdot \Delta P \tag{1}$$

The minimum necessary pressure at the reception point is

$$\Delta P = \Delta V \cdot C_0 / S_q \tag{2}$$

In terms of radiation, as a model, the simplest type of source is the pulsating sphere. Assuming that the radius a of the sphere is small compared to the wavelength λ of the emitted sound wave, the pressure field can be expressed as

$$p(r,t) = j \cdot \rho_0 \cdot c \cdot U_0 \cdot \frac{a}{r} \cdot k \cdot a \cdot e^{j(\omega \cdot t - kr)}$$ (3)

where: U⁰ is the speed amplitude of the vibration; an - the average radius of the sphere; 0 the mass density of an undistributed fluid element.

A source of particular interest is the rigid piston of radius a mounted flush with the surface of an infinite baffle, and vibrating with a time harmonic motion. The final expression for the pressure of a single source at the point placed of distance r and θ angle is:

$$p(r,\theta,t) = j \cdot \frac{\rho_0 \cdot c \cdot U_0 \cdot ka^2}{2r} \cdot e^{j(\omega \cdot t - kr)} \cdot \left[ \frac{2J_1(ka\sin\theta)}{ka\sin\theta} \right]$$ (4)

where J¹ represents the Bessel-function of the first kind of order 1.

An ultrasonic transducer with EMFi material was built. Final and intermediary stages of the transducers are presented in Figure 4. A double-sided PCB is used. Both sides are covered by EMFi. One side is covered by graphite (Figure4.b). By connecting it to a ground plane, an electromagnetic shield is obtained. There are some advantages as minimizing the capacitive coupling between active electrode of the transducers and receiver, and – another advantage – the user is protected over high voltage of the active electrode.

Figure 4. Custom ultrasonic transducer: (a) PCB; (b) PCB and EMFi; (c) Final product.

EMFi based transducers with different active areas have been built and tested for transmitter (4 cm² ) and for receiver (1.5 cm² ). The main diagrams for these transducers are presented in Figure 5. A wide frequency range centered on 250 kHz and a low width of the main beam were obtained. The diagram was measured at 280 kHz.

Figure 5.(a) Frequency response of the transducer-air-transducer path; (b) Directivity diagram at 280 kHz.

3.1. The Receiver

Reference [26] reported a structure of the receiver designed for a sensibility of 100 V and composed of two blocks: a preamplifier with fixed gain (1897 amplification) followed by a variable gain stage.

In this work, the designed receiver is built on three layers, 20x20mm PCB, as presented in Figure 6. One layer is used as electrode for EMFi foil, one for input preamplifier components, and the middle one is used as a shield. The resulting receptor is small, compact and robust to external perturbations.

Figure 6. Layout of the receiver: (a) components layout; (b) EMFIT sensor layout; (c) cross-section of PCB.

An octagon EMFi foil is glued on the PCB, which forms the bottom electrode, and the top electrode is obtained with a layer of graphite, sprayed over the foil. The top electrode is connected to the ground so that electrostatic and capacitive perturbations are minimized. Figure 6 presents some details from the receiver circuit and ultrasonic transducer.

Test version of the receiver circuit, Figure 7, is based on AD8022 dual high speed, low noise operational amplifier from Analog Devices, which offers a frequency of 130 MHz, at unity gain. Input bias currents, typical 2 μA, are compensated with R8, R4, and C3, under classical configuration. Resistor R1 can be used to disconnect EMFi sensor, to separately test the amplifier. A value of 170 kΩ is obtained as input impedance, by considering R3 = 2.2 M Ω, R2 = 220 kΩ, R4 = 680 kΩ, and 20 kΩ the differential input impedance for AD8022.

A bigger input differential impedance Zidif will conduct to higher input impedance, and necessarily to a better sensibility of the receiver amplifier. The preamplifier gain is about 660. The first Operational Amplifier (OA) circuit amplifies signal by 11, and the second one by 60. This will lower the upper frequency of the bandwidth to 2 MHz. The capacitor C2 is used to limit this value further, to 500 kHz.

By testing the preamplifier, it results flexibility in gain and bandwidth around the necessary values and robustness to noise signals. The input impedance could be higher and a pin-compatible op-amp with JFET inputs will be tested in a further version.

From technological point of view, the electronic circuit is mounted on dual-layer PCB, one layer used as ground plane. The transducer and the preamplifier of the receiver have the same size and placed back-to-back.

Figure 7. Preamplifier schematics.

3.2. The Transmitter

The EMFi based transducer needs high voltages to achieve high Sound Pressure Levels (SPL). For example, in [22], signals of about 300 Vpp were used. Three circuits have been tested: (1) a hybrid power amplifier with transformer; (2) a bootstrap high voltage amplifier; (3) an active load A-class H bridge amplifier.

The first one is a circuit that includes an operational amplifier and other active components connected in such a way that the resulting circuit will exceed one or more opamp IC (Integrated Circuit) limits. Largely, the operational amplifier entity refers to a general-purpose DC voltage amplifier that may be configured by a passive network.

The schematic in Figure 8 consists of IC (Integrated Circuit) and a configuration network from feedback resistor R2, ground resistor R1 and in-loop compensation condenser C2. The circuit amplifier can drive a 150 [pF] EMFI transducer using a 1:30 rising voltage transformer. The IC is a 3 MHz bandwidth op-amp (TL071). The power transistors T1 and T2 are BD439/BD 440. The amplifier output voltage V⁰ is applied to the load capacity C^L via the rising voltage transformer, which generates a voltage up to 700-750 Vpp. Even with no problems of stability, a capacity value C2 must be considered by experimenting.

Figure 8. High voltage amplifier with transformer.

The second solution, the bootstrap high voltage amplifier has feedback loops for supply voltages, in order to increase the range of the output voltage. There have been used high voltage transistors (ex. STN83003, [27], STP 93003, [28]), faster op-amp (ex. LM7171, [29]), the supply voltages can be increased at ±180 V and the output signal will vary up to 336 V.

There are many sonar systems with exploring signals as short bursts, i.e. rectangular pulses. These signals need high quality amplifiers at the output stage. The last circuit considers a high quality amplifier. The circuit operates between "stand-by" and "burst emission" modes.

The considered circuit is presented in Figure 9. Four stages could be considered: (1) voltage-current converter: the op-amp, T2, D2, and R2; (2) active load: T4, R8, and R4; (3) voltage divisor: RB1, TB and RB2; (4) electronic switch K that connects the resistor R6, in parallel with R2 when ultrasonic burst is emitted We reduce the power dissipation between two successive bursts, by imposing amplifier work in two modes: (1) Burst-Emission: when the contact K is close; (2) Stand-By: when the contact K is open.

Figure 9. (a) The high quality amplifier; (b) the output characteristics of T2 transistor.

4. NARROW DIRECTIVITY BY EXTERNAL PINNAE

Biological studies show the effect of the pinnae in environment living of animals, e.g. the bats. The external shape of the ear has many non-uniformities and the surface is slightly mobile and flexible. Such a set of features provides a sensibility of hearing and the possibility to precisely localize various targets on the hunting field. This is an adaptive behavior of bats, which requires, by mimetism, an adaptive working regime for the biomimetic sonar heads, related to frequency and shape of the transmitted signals. From the receiving point of view, adaptation means the possibility of slightly changing the shape and the orientation of the external ears.

In the context of the autonomous vehicles and sonar-based systems, a first point to do is to implement adaptation of the beam of directivity to the details of the environment. There are some examples, where [30] and [31] made simulations and experiments related to the optimum shape of the pinnae. Simulation based studies show that the best shape of artificial made pinnae is a kind of paraboloid, in fact a portion of the paraboloid. One of the success examples was also within CIRCE research project, [14].

This section describes some results from the measurements processes of pinnae. Ultrasonic transducers working on fixed frequency (40 kHz) were used, by removing the housing of the transducer (MA40S5), [32]. Data are recorded by an automated ultrasonic measurement system, as it is described in [33].

Figure 10, [33], shows an example of directivity charts, in horizontal plane for: (1) original transducer with no pinnae; (2) transducer with no housing or pinnae; (3) transducer with small pinnae; (4) transducer with big pinnae. Clearly, the bigger the ratio between the size of the paraboloid and the size of the transducer, the better directivity is.

The explored ears use one-half of the paraboloid. The effect is an asymmetry in vertical plane, as presented in Figure 11: (1) transducer without parabolic surface; (2) transducer with small pinnae; (3) transducer with big pinnae. In this last case, a secondary lobe is appearing, which later will introduce distortions in ultrasonic images. More details are in [33]. The experimental measurements show good estimations of theoretical computed values. By using such reflectors, the width of the main lobe of 20° is available. Also, the sensitivity is improved with an order of 3 dB.

Figure 10. Directivity charts, horizontal plane, various configurations.

Figure 11. Directivity charts, vertical plane, various ear sizes.

5. ULTRASONIC IMAGE GENERATION AND RAW PROCESSING

Airborne ultrasound imaging is difficult to develop because of the high absorption of ultrasound waves in air. Thus, the generation of such images is a very difficult task and there is a need for new methodologies and technologies that could solve the above problem. Accordingly, this section introduces airborne ultrasonic image generation and processing methodology.

In the field of mobile robots, map building for navigation purposes is solved by using sonar systems, and implementing signal processing and pattern recognition algorithms, e.g. [34, 35, 36] for detection and localization of reflectors, or [37, 38] for feature extraction. In medicine, ultrasound images are the result of echography, providing an efficient method for diagnosis and right treatments.

Industrial applications are mainly used to detect non-uniformities and fabrication faults inside materials, e.g. [39, 40]. A quite new application is through-wall sensing, which helps to see inside structures of buildings, to identify the materials within building, e.g. [41, 42, 43].

The modern way for generation of ultrasound imaging is based on an array of ultrasound transducers, e.g. [44, 45]. The solution is quite complex, requiring high power signal processors and memory.

Sometimes, solutions using discrete ultrasonic transducers must be considered, at least from the point of view of cost and reliability. The approach of generating ultrasound images is based on pixel-by-pixel basis, each pixel being associated with a pair of signals (transmitted and received). The main problem is related to the processing time necessary to obtain the echoes signals. The greater the image"s size, the greater the generation and processing time.

The following sub-sections focus on the main functions in order to generate all signals involved in ultrasonic image generation process.

5.1. Control Functions

The sonar head works by executing specific tasks, which are described by a set of general functions. These are used to orient the sonar head to target"s position in the 3D space. The basic function must orient the emitter and the receiver to the position of estimated or expected target. Depending on the type of the sonar, active or passive, the orientation function must be synchronized with other tasks, e.g. for transmitting signals in the case of active sonar. In the case of passive sonar, the maximum amplitude of the received signal will impose the orientation of the sonar head.

Let target point be denoted by T x y z T r ( , , ) ( , , ) . The pseudo code for target catching is described below and contains all commands and parameters to properly drive the servomotors.

Function Orient_To_Target (FOT)
   #1: Initial data (arguments): 
   - Target_Azimuth; 
   - Target_Elevation;
   - Initial positions of servos: // the middle of the range
          P0_ini ; P1_ini;
          P4_ini ; P5_ini ;
   #2: Initializations: //all servos 
                  #0 P P0_ini #1 P P1_ini
                  #4 P P4_ini #5 P P5_ini
   #3: Positioning: // emitter and receivers to target
                  #0 P<Target_Azimuth> // set the azimuth of transmitter
                  #1 P<Target_Elevation> // set the elevation of transmitter
                  #4 P<Target_Azimuth> // set the azimuth of left receiver
                  #5 P<Target_Azimuth> // set the azimuth of right receiver
End FOT.

In order to use the right notations and usage of the involved signal, Figure 12 introduces the model of the active sonar system. A digital signal source generates the signal xt() . This is amplified and - at the output of the ultrasonic transducer- the signal st() is obtained. The medium between target and emitter, considered time-invariant and linear, is described by impulse response functions () ^d h t and () ^Rh t . If ¹n t() and ² n t() are the specific noises, then the processed signals are 1 s t() and 2 s t() :

$$s_1(t) = s(t) + n_1(t)$$ (5)

$$s_2(t) = s(t) \otimes h_D(t) \otimes h_T(t) \otimes h_R(t) + n_2(t)$$ (6)

Figure 12. The model of the sonar system.

The received signal, 2 s t() , has two representations obtained by the left and, respectively, the right ear: ( ) ^L s t and ( ) ^R s t . The signals ¹n t() and ² n t() are noise or assimilated as noise signals (e.g. quantization errors). The operator " " means convolution and it is accepted as long as blocks from transmission path are linear.

The basic function involved in signal acquisition is considered. The initial parameters are related to the transmitted impulses, i.e. the number of pulses. This is proportional with the distance to target. Common values could start from 4 to 8 pulses for distances up to 1 m, and might go to 32-50 pulses for longer distances (2-3 m). The acquisition and storage of raw samples start after a delay interval, D, which depends of distance to target.

Function Get_Data(FGD)
   #1: Initial parameters (arguments): 
   - width TX of the impulse x(t); // pulse wave
   - delay D; // depends of distance to target
   #2: Generate x(t);
   #3: Transmit s(t); // output of transducer
   #4: Wait(D);
   #5: Read signals s(t), s1(t) and s2(t). // sampling and recording
End FGD.

Data stored locally, in the local memory of the sonar head, is transmitted via USB (Universal Serial Bus) port to a central computer. Here, the digital signal is arranged in specialized arrays and processed by the high-level simulation languages, e.g. MATLAB. Below, the pseudo code for transmitting data from sonar head to central (personal) computer is presented.

Function Transmit_Data (FTD) 
   #1: Initial parameters (arguments): 
   - Speed; // USB port
   - Protocol; // USB port
   - Data // USB port
   #2: Open port; 
   #3: Make connection; 
   #4: Data Transmission; 
   #5: Close port; 
EndFTD.

The final function will link the previous described function. This is presented below as a procedure:

Procedure SONAR 
    #1: Target_Orientation(); 
    #2: Get_Data(); 
    #3: Transmit_Data(); 
    #4: Process_Data(); 
End SONAR.

A small remark must be discussed, finally. Depending on the objective of environment exploring, some different trajectories could be considered. In the context of image generation, these could be considered as grids. Each grid is defined by points with angular coordinates and . Figure 13 presents three of such possible grids used in the exploration of the environment or image generation task.

Figure 13. Examples of pattern motions, with fixed points, variable points and uniform grid.

5.2. Description of the Experiments

The objective of the experiments is to build airborne ultrasonic images, by considering various types and sizes of targets. In this chapter, three objects are considered: (1) - a ball of 20 cm diameter; (2) - a box with an edge oriented to the sonar head; (3) - a box (the same as in case #2) with a face oriented perpendicularly to the sonar head. The considered objects are presented in Figure 14. The used sonar head is ROVIBAT-02, which has a Pan & Tilt structure, and commercial Murata ultrasonic transducers. The distance between receivers was d = 4.5 cm.

Image generation is made pixel-by-pixel. For each explored direction, the echoes are stored and processed. The values of the main parameters involved in the experiments are: sampling frequency, 450 kHz; exploring distance, 3 m; target distance, 1m; resolution of the image, 20 by 20; azimuth range, 0 20 ; elevation range, 40 . The value of 0⁰ corresponds to the middle of the target. Each pixel is quantized with 8 bits.

As example, Figure 15 shows some echoes, exy(t), which correspond to directions of coordinates (2,2), (2,15), (15,2) and (15,15). It is observing that the positions and amplitudes of received echoes are changing in time.

Figure 14. The investigated objects.

Figure 15. A set of recorded echoes, in the case of box like target.

5.3. Ultrasonic Image Generation

Figure 16, [46], introduces the main computation blocks for ultrasonic image generation. Each explored direction is characterized by the coordinates pair (x,y), each from the set of {1, 2,…, n=20}.

The computation of the correct values of x and y and the correct positioning are the tasks of the position control board, which was described already in a previous section.

The image generation method is composed of two structural processing frames, which means processing in hardware and processing in software, with more details in [46]. All processing steps assigned to hardware running are related to the sonar head and consists of orientation, emission of pulse signals, filtering and conversion of the received signals (echoes). Data coming from sonar head are transmitted to the central (personal) computer for intensive processing by MATLAB functions. On each path, left and right, the main considered processing steps are: matched filtering, envelope detection, image construction, and uncertainty compensation.

Images from left and right side are next fused. By fusion, a set of image transforms is considered as registration, enhancement, and filtering. The key element of the image constructor is a maximum value detector and storing, so a pixel of the ultrasonic image has the maximum values of the envelope from a given exploration time (range) interval. Based on the content of the images generated by the first method, object detection and size discrimination algorithms are able to use and develop, e.g. in applications as obstacle avoidance and object recognition.

Some blocks are represented with solid line. They implement complex processing tasks. These are optional, in the sense that these could be replaced with more simple processing functions. Matched filtering can be replaced by cross-correlation and uncertainty processing could miss. Uncertainty processing block consider the effect of the uncertainty of ultrasonic transducers on generated images. If the directivity of transducers is narrow, then this block could be removed.

From the structure of the generation process of images, it could be observed that time is used as a parameter, so the ultrasonic shape (images) of the object is also parameterized by time. Once the image set is finished, all classical algorithms of image processing could be applied, e.g. [47, 48].

Figure 16. Main processing blocks in the ultrasonic image generation process.

The obtained ultrasonic un-processed images are presented in Figure 17, [46]. Because the ultrasonic transducers are located at a small distance from one another, about 5 cm, the left and right images do not differ very much. Thus, a very simple fusion scheme was considered here for the finally generated image, by averaging the left and right images.

Figure 17. Ultrasonic images for the considered cases (second method).

Figure 18, [46], shows an example of the uncertainty processing. The effect of using the block is to introduce a local average. Mainly, a Gaussian response (weighting) surface is considered for each element of the image. The variance of the surface is estimated from the real directivity of the ultrasonic transducer.

CONCLUSION

Preliminary results of airborne ultrasonic image generation and processing were presented. The objective of the work was to investigate the difficulties of ultrasonic image generation with some new sonar heads, having portability, simplicity and biomimetism as specific features.

Figure 18. Ultrasonic images with uncertainty compensation.

Two sonar heads were built, with a small and wide range of working frequency. The present work considers only the small bandwidth (5-8 kHz) and the fixed working frequency (40 kHz). The sonar heads were presented from hardware and software points of view. The structure of the hardware is under portability (among various vehicles) and biomimetism (one transmitter plus two receivers, and six degrees of freedom). Various solutions for narrow beam of directivity were considered. A solution based on parabolic reflectors was presented in this work.

From the imaging point of view, two processes were described: generation and basic processing. Generation process considers a pixel of the image assigned to a pair of signals (transmitted pulse plus received echo). Depending on the amplitude of the received echo, the intensity of the pixel varies from black (zero intensity, corresponding to zero echo) to white (the highest intensity of the received echo). Ultrasonic images have low resolution, 20x20 or 40x40. In both cases, generation time is high and other strategies should be considered in order to obtain ultrasonic images in real time. Image processing process considers here fusion of left and right images.

Ultrasonic images have poor information content, which is explained by the uncertainty generated by the width of the main beam of ultrasonic transducers. Uncertainty must be considered also for generation process, depending also on distance to target. The obtained images are not very close to the images of real objects. This is common to ultrasonic field and – sometimes – an external interpreter could help with the proper use of the image content. Future research steps will consider a wide frequency range, from 40 to 200 kHz and the fusion of ultrasonic images at various exploring frequencies (40, 100, and 200 kHz).

ACKNOWLEDGMENTS

This document contains partial results of ADBIOSONAR research Grant no. 12079/2008 under CNMP (UEFICDI) Romanian Authority.

REFERENCES

• [1] Libelium; 50 Sensor Applications for a Smarter World. Get Inspired. http://www.libelium.com/ top_50_iot_sensor_applications_ranking.
• [2] Kianpisheh, A.; Mustaffa, N.; Limtrairut, P.; Keikhosrokiani, P.; Smart Parking System (SPS) Architecture Using Ultrasonic Detector. International Journal of Software Engineering and its Applications, vol. 6, no. 3, pp.51-58, 2012.
• [3] MIT; City Car, USA, 2012, http://www.media.mit.edu/news/citycar.
• [4] Cybercar. 2006,http://www.cybercars.org/.
• [5] Sun, J; Wu, Z; Pan, G; Context Aware Smart Car: from Model to Prototype, Journal of Zhejiang University, ISSN 1673-565X (Print); ISSN 1862-1775 (Online), vol. 10, no. 7, pp. 1049-1059, 2009.
• [6] Forrest, A.; Mustafa, K.; Autonomous Cars and Society. Department of Social Science and Policy Studies, Worcester Polytechnic Institute, May 1, 2007, http://www.wpi.edu/ Pubs/E-project/Available/E-project-043007-205701/unrestricted/IQPOVP06B1.pdf.
• [7] Fleming, B.; Human/Machine Interface Issues Take Precedence. IEEE Vehicular Technology Magazine, pp. 4-9, March 2009.
• [8] Stevenson, R.;Long Distance Car Radar. IEEE Spectrum Magazine, pp. 50-55, October 2011.
• [9] Paisitkriangkrai, S.; Shen, C.; Zhang, J.; Fast Pedestrian Detection Using a Cascade of Boosted Covariance Features. IEEE Transactions on Circuits and Systems for Video Technology, vol. 18, no. 8,pp. 1140-1151, 2008.
• [10] Jones, W.D.; Building Safer Cars. IEEE Spectrum Magazine, pp. 82-85, January 2002.
• [11] Barshan, B.; Kuc, R.; A Bat-Like Sonar System for Obstacle Localization. IEEE Transactions on Systems, Man, and Cybernetics, vol. 22, no. 4, pp. 636-646, 1992.
• [12] Kampchen, N.; Evolving Pinna Like Surfaces for a Biomimetic Sonar Head. M.Sc. Thesis, University of Edinburgh, 2000.
• [13] Kuc, R.; Biomimetic Sonar Recognizes Objects Using Binaural Information. JASA, vol. 102, no. 2, pp. 689-696, 1997.
• [14] Peremans, H.; CIRCE Project. http://www.ua.ac.be.
• [15] Hallam, J.; ChiROPing Project. http://www.chiroping.org.
• [16] Muller, R.; Kuc, R.; Biosonar-Inspired Technology: Goals, Challenges and Insights. Bioinspiration & Biomimetics, vol. 2, pp. 146-161, 2007.

• [17] Maxx Products International, LLC.,http://www.maxxprod.com.

• [18] Lynxmotion, Inc., Users ManualSSC-32 Ver 2.0.http://www.lynxmotion.com.

• [19] Reinard, L.; Streicher A.; Sutor, A.; Broadband Ultrasonic Transducer. 19th International Conference on Acoustics, Madrid, Spain, 2007.

• [20] Doring, J.; Bovtun V.; Bartusch, J.; Beck, U.; Kreutzbruck, M.; Cellular Polyproplylene Ferroelectret Films Piezolectric Material for Non-Contact Ultrasonic Transducers. 17th World Conference on Nondestructive Testing, Shanghai, China, 2008.

• [21] Bovtun, V.; Doring, J.; Bartusch, J.; Beck, U.; Erhard,A.; EMFIT Ferroelectret Film Transducers for Non-Contact Ultrasonic Testing.ECNDT, 2006.

• [22] Streicher, A.; et al.; Ultrasonic Transducers for a Biomimetic Sonar System. IEEE International Ultrasonics, Ferrolectrics and Frequency Control Joint 50th Anniversary Conference, pp. 1142-1145, 2004.

• [23] Jimenez, A.; et al; EMFI Based Ultrasonic Transducer for Robotic Applications. Sensors and Actuators, A., vol. 148, pp. 342-349, 2008.

• [24] Paajanen, M.; Lekkala, J.; Kirjavainen, K.; ElectroMechanical Film (EMFi) -A New Multipurpose Electret Material. Sensors and Actuators, vol. 84,pp. 95-102, 2000.

• [25] EMFIT Film Specifications. 2009, www.emfit.com.

• [26] Ealo, et al.; A Fabrication Procedure for Airborne Ultrasonic Phased Arrays Based on Cellular Electromechanical Film. IEEE Ultrasonics Symposium, IUS 2008, pp. 891- 894, 2008.

• [27] http://www.st.com/stonline/products/literature/ds/12325/stn83003.htm.

• [28] http://www.st.com/stonline/products/literature/ds/12329/stn93003.htm.

• [29] http://www.national.com/mpf/LM/LM7171.html.

• [30] Carmena, J.M.; Kampchen, N.; Kim, D.; Hallam, J.C.T.; Artificial Ears for a Biomimetic Sonar Head: From Multiple Reflectors to Surfaces. Artificial Life, vol. 7, no. 2, pp. 147-169, 2001.

• [31] Carmena, J.M.; Kim, D.; Hallam, J.C.T.; Designing Artificial Ears for Animat Echolocation: From Animals to Animats. J-A. Meyer et al. (Eds.), MIT Press, vol. 6, pp. 73-80, 2000.

• [32] MA40S5 Product Page. http://search.murata.co.jp.

• [33] Epure S.; Aiordachioaie, D.; On Pinnae Design and Fabrication Technology. 17th IEEE International Symposium for Design and Technology in Electronic Packaging (SIITME-2011), Timisoara, Romania, pp. 133-136, 2011.

• [34] Urena, J.; et al.; Classification of Reflectors with an Ultrasonic Sensor for Mobile Robot Applications. RAS, vol. 29, pp. 269-279, 1999.

• [35] Reijniers, J.; Peremans, H.; Biomimetic Sonar System Performing Spectrum Based Localization. IEEE Trans. on Robotics, vol. 23, no. 6, pp. 1151-1159, 2007.

• [36] Rikoski, R.J.; et al.; On Correlating Sonar Images. Robotics: Science and Systems I, MIT Press, Cambridge, Massachusetts, 2005.

• [37] Sejdic, E.; et al.; Time Frequency Feature Representation Using Energy Concentration: An Overview of Recent Advances. Digital Signal Processing, vol. 19, pp. 153-183, 2009.

• [38] Araujo, E.G.; Grupen, R.A.; Feature Extraction for Autonomous Navigation Using an Active Sonar Head. IEEE International Conference on Robotics and Automation, San Francisco, 2000.

• [39] Gudra T.; et al.; Airborne Ultrasonic Transducers for Ultrasonic Trans. Tomography in Gaseous Media. 20th International Congress on Acoustics (ICA-2010), Sydney, pp. 1- 5, 2010.

• [40] Gomez F.; Althoefer K.; SeneviratneL.D.; Simulation of Ultrasound Imaging Inside Fully Charged Pipes. Automation in Construction, vol. 15, pp. 355-364, 2006.

• [41] Gomez F.; Althoefer K.; SeneviratneL.D.; Simulation of Ultrasound Imaging Inside Fully Charged Pipes. Automation in Construction, vol. 15, pp. 355-364, 2006.

• [42] Baranoski, E.J.; Through-Wall Imaging: Historical Perspective and Future Directions. Journal of the Franklin Institute, vol. 345, pp. 556-569, 2008.

• [43] Borek, S.E.; An Overview of through the Wall Surveillance for Homeland Security. 34th Applied Imagery and Pattern Recognition Workshop, vol. 6, pp. 19-21, 2005.

• [44] Bernus, L.; et al.; Sampling Phased Array: A New Technique for Signal Processing and Ultrasonic Imaging. INDTCM, Publisher: NDT.net, vo. 50, no. 3, pp. 153-157, 2008.

• [45] Jensen, J.A.; Ultrasound Imaging and its Modeling, Imaging of Complex Media with Acoustic and Seismic Waves. Springer Verlag, vol. 84, pp. 135-165, 2002.

• [46] Aiordachioaie, D; Frangu, L.; Some Results on Low Resolution Airborne Ultrasonic Image Generation and Processing. IEEE International Symposium on Signal Processing and Information Technology (ISSPIT-2011), Bilbao, Spain, pp. 434-439, 2011.

• [47] Jahne, B.; Digital Image Processing. 5th Edition, Springer, 2002.

• [48] Pratt, W.K.; Digital Image Processing. John Wiley & Sons, 2001.

Chapter 3

AUTONOMOUS UNDERWATER VEHICLE DYNAMICS

Mustafa Dinc and Chingiz Hajiyev†

Aeronautics and Astronautics Faculty, Istanbul Technical University, Maslak, Istanbul, Turkey

ABSTRACT

The research on underwater systems has gained enormous attention during the last two decades because of applications taking place in many fields. Therefore, the significant number of Unmanned Underwater Vehicles (UUVs) has been developed for solving the wide range of scientific and applied tasks of ocean research and development in the world. Guidance, navigation, and control techniques are key research and development areas for the success of those sophisticated UUV missions.

One of the main objective of this chapter is to provides detailed explanations on the theory behind the main concepts that directly influence the design of the dynamic mathematical model of AUV and then to accomplish dynamic mathematical modeling of an AUV in MATLAB version 7.5 environment under different swimming conditions.

In order to develop high fidelity AUV simulation model and implement control and navigation algorithm for the vehicle, we need to know overall AUV modeling.

Another main focus of this chapter is to realize the parameter identification of hydrodynamic coefficients based on a Least Square Estimation (LSE) algorithm for a nonlinear mathematical modeling of AUV. Parameter Identification is very important to have the estimated values of these coefficients in order to accurately simulate the AUV"s dynamic performance. The estimated hydrodynamic coefficients can be used as inputs not only for a mathematical model to analyze the maneuvering performance but also for the AUV"s motion controller.

Keywords: Autonomous Underwater Vehicle, Dynamic Modeling, Parameter Identification, Least Square Estimation

Mustafa Dinc: Aeronautics and Astronautics Faculty, Istanbul Technical University, Maslak, Istanbul, Turkey. Email: mdinc1972@gmail.com.

^† Chingiz Hajiyev: Aeronautics and Astronautics Faculty, Istanbul Technical University, Maslak, Istanbul, Turkey. E-mail: cingiz@itu.edu.tr.

1.INTRODUCTION

Unmanned Underwater Vehicle (UUV) research began in the 1960s, with the first prototypes emerging in the 1980s. Nevertheless, the research on underwater systems has gained immense interest during the last two decades with applications taking place in multiple fields of marine systems.

Therefore, the significant numbers of UUVs have been developed for solving a wide range of scientific and applied tasks of ocean and seabed research and development in the world [1]. The military, as well as civilian industries can see great potential uses of UUVs in the underwater environment.

UUVs by definition are small submersible vehicles that contain independent propulsion systems and are capable of carrying sensors such as side-scan sonar, video cameras, depth sensor, and other oceanographic measuring devices [1]. UUVs are highly desirable as they can at least limit the level of human life risk and direct physical human involvement in a mission.

Typically, UUVs can be classified into two unmanned underwater systems that are Remotely Operated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs): Each branch has its own pros and cons due to the mission type. ROVs, characterized by direct human assistance, is remotely operated through the presence of a tether cable; the other group is known as the AUVs, which is the topic of this chapter, characterized by their autonomous behavior, having cylindrical geometric shape and absence of a tether cable. This classification is of course not precise given the varying degrees of autonomy in both groups which can differ according to the requirements of the designed mission. In addition, the presence of a tether cable does not necessarily mean that the vehicle cannot perform autonomous tasks. The basic difference between AUVs and ROVs is that AUVs use "intelligence", such as sensing and automatic decision-making. They have predefined plan of operations in its "mind" allowing them to perform tasks autonomously [1]. ROVs are remotely controlled by a human with the help of communication links based on tether cable (such as cupper, fiber optic, etc.).

The AUVs have shown efficiency at performing complex search and inspection missions, and opening a number of new important application areas which include environmental monitoring, surveillance, underwater inspection of harbors and pipelines, geological and biological surveys, mine countermeasures, etc. [1].

Due to their versatility, compact size, independence and covertness, AUVs are a highly valuable asset in these environments. Possible missions for AUVs range from dedicated, rapid environmental assessment, special operations to reconnaissance, surveillance and intelligence. This wide range of possible applications and their partly contradictory nature calls for a large variety of vehicle configurations with regard to speed, range, depth, endurance, launch and recovery, payload and communication systems.

The nonlinear and multivariable dynamics, accompanied with parametric uncertainties and also immeasurable disturbances cause position control of an AUV to become an important issue. Dynamic equations of AUV are strongly nonlinear, multivariable and uncertain. Sea currents are disturbances for the system. Parameters of equations, which called hydraulic coefficients, should be estimated. Some of the parameters, such as hydrodynamic coefficient, do not have accurate value. Computation and stimulation of these parameters is so complex and so time consuming and costly.

These parameters can vary due to environmental conditions. On the other hand, the control of AUV presents a number of unique and formidable challenges. An AUV will be required to move in adverse environments showing good performance and a great deal of autonomy under a variety of load conditions and unknown sea conditions. Moreover, AUV dynamic is highly nonlinear, coupled, time varying and subject to hydrodynamic uncertainties and external disturbances. For trajectory tracking, the problem statement requires the design of a depth and head control system for the vehicle in order to achieve precise trajectory [2].

In order to develop an accurate and robust navigation and control system, we need to know an overall mathematical modeling of AUV, which involves the interdisciplinary study of kinematics, hydrostatics, and hydrodynamics.

Hydrostatic is concerned with the equilibrium of underwater bodies at rest or moving with constant velocity, whereas hydrodynamics is concerned with bodies having accelerated motion [3]. In this chapter, we describe nonlinear dynamic model of an AUV for free swimming conditions.

Otherwise, in many literature, scientists employ parameter specific nonlinear AUV model for only one swimming condition.

Another main focus of this chapter is to realize the parameter identification of hydrodynamic coefficients based on Least Square Estimation (LSE) algorithm for a nonlinear mathematical modeling of AUV. The hydrodynamic coefficients of AUV can be estimated based on the calculations of the first principle of dynamics of AUVs and from statistical LSE of experimental data [4]. The identification of the parameters of the item being tracked can be approached in many different ways. The parameters can be determined once, from a model of the expected input signal. It could also be done adaptively utilizing a recursive method to identify the parameters of the incoming signal.

Another approach that estimates the hydrodynamic coefficients of AUVs is the observer method, in which a model-based estimation algorithm is used. A representative method amongst observer methods is the Kalman filter (KF), which has been widely used to estimate state variables and parameters [4]. Hwang [5] estimated the maneuvering coefficients of a ship and identified the dynamic system of a maneuvering ship using an extended Kalman filter (EKF) technique. Additionally, referencing [6], Meng and Veras, they concentrated on the application and comparison of EKF and iterated EKF for aerodynamic parameter estimation of a fixed wing UAV.

In another reference [7], Chowdhary compared the performance of three recursive parameter estimation algorithms for aerodynamic parameter estimation of two aircrafts derived from real flight data. These algorithms are the EKF, the simplified version of the Unscented Kalman filter (UKF) and the augmented version of the UKF.

On the other hand, the report [8] compares responses obtained by the KF, the least squares estimation, and the linear model for the NPS Phoenix AUV. The LSE provided results similar to those obtained by the KF, but the latter produced a more accurate model.

Referencing [9] Vandersteen, Rolain, Schoukens and Pintelon proposed a robust estimation algorithm for the estimation of static and nonlinear systems which can be described as a nonlinear function corrected with a rational form. The errors-in-variables-based algorithm solves the starting-value problem using an iterative, weighted least-squares procedure, which constructs the rational form such that the set of normal equations becomes best conditioned, and uses a maximum-likelihood estimation step to increase the efficiency of the estimates.

The main focus of this chapter is to provides detailed explanations on the theory behind the main concepts that directly influence the design of the dynamic mathematical model of AUV and then to accomplish dynamic mathematical modeling of an AUV in MATLAB version 7.5 environment under different swimming conditions. Another main focus of this chapter is to realize the parameter identification of hydrodynamic coefficients based on the LSE algorithm for a nonlinear mathematical modeling of the AUV. It is important to have the estimated values of these coefficients in order to accurately simulate the AUV"s dynamic performance. The estimated coefficients can be used as inputs not only for a mathematical model to analyze the maneuvering performance but also for a AUV"s motion controller.

2. DYNAMIC MATHEMATICAL MODELING OF AN AUV

This section provides detailed explanations on the theory behind the main concepts that directly influence the design of the dynamic mathematical model of AUV.

In order to achieve robust and accurate navigation, guidance and control of an AUV, we require overall modeling of AUV, which is fairly complicated, and even when considered as a rigid body, an exact analysis is only possible by including the underlying infinite dimensional dynamics of the surrounding fluid [3]. This can be done using partial differential equations, which are solved by Computational Fluid Dynamics (CFD), but it still involves a formidable computational burden, which is infeasible for most practical applications.

As a result, the conventional approach has been to use finite-dimensional approximations. On the other hand, AUV modeling involves the interdisciplinary study of kinematics, hydrostatics, and hydrodynamics. The study of hydrostatics is concerned with the equilibrium of underwater bodies at rest or moving with constant velocity, whereas hydrodynamics is concerned with bodies having accelerated motion [3].

An increased knowledge of hydrodynamic parameters then leads to a better navigational system design and performance on AUVs.

In this chapter we try to develop nonlinear mathematical modeling of an AUV for free swimming conditions and then as case study, directly use the mathematical model of REMUS AUV that is designed to perform hydrographic analysis in the very shallow water.

REMUS is used for missions such as hydrographic surveys, mine counter-measure operations, harbor security operations, environmental monitoring debris field mapping, search and rescue operations, fishery operations, and scientific sampling and mapping.

2.1. Modeling Assumptions

2.1.1. Environmental Assumptions

The corresponding assumptions are made about the vehicle with respect to its environment [10]:

● The AUV is deeply submerged in a homogeneous and unbounded fluid. In other words, the AUV is located far from the free surface (no surface effects, i.e. no sea wave or vehicle wave-making loads), walls and bottom.

● The AUV does not experience underwater currents.

2.1.2. Dynamics Assumptions

In dynamic modeling of AUV, the following assumptions are used [11]:

• The AUV behaves as a rigid body of a constant mass.
• The earth"s rotation is negligible for acceleration components of the vehicle"s center of mass
• The primary forces that act on the AUV are inertial and gravitational in the center of buoyancy and are derived from hydrostatic, propulsion, thruster, and hydrodynamic lift and drag forces.
• The thruster assumption is that it uses an extremely simple propulsion model, which treats the vehicle propeller as a source of constant thrust and torque.

2.1.3. 6-DOFRigid-Body Equations of Motion

AUVs move in six degrees of freedom (6-DOF) since six independent coordinates are necessary to determine the position and orientation angle of a rigid body dynamics. The first three coordinates and their time derivatives are based off of translational motion along the x, y and z-axes, while the last three coordinates ( ) and their time derivatives are used to describe orientation angle and rotational motion [12]. , ,

Velocity and angular velocity components of the AUV relative to the body axes are denoted by the velocity of surge, sway, heave motion, and angular velocity of roll, pitch, and yaw motion , respectively. X, Y, Z, K, M and N represent the resultant forces and moments with respect to the x, y, and z axis. For AUVs, it is common to use the SNAME notation. In Table 1 below, the six different translational and rotational motion components are defined as: surge, sway, heave, roll, pitch and yaw, respectively [3]. ( , , ) x y z ( , , ) pqr

2.1.4. Coordinate Frames

Typically, three different right-handed and rectangular coordinate frames are used for defining AUV motion, which is a 6 degrees of freedom (6-DOF). First, the body axes have their origin at the center of buoyancy (CB) with x directed toward the bow along the hull centerline axis, y directed to the starboard side, and z toward the keel (Figure 2).

Motion	Forces and	Linear and Angular	Position and Euler
Components	Moments	Velocities	Angles
Surge	X	u	X
Sway	Y	V	Y
Heave	Z	w	Z
Roll	K	P	
Pitch	M	Q	θ
Yaw	N	R	ψ

Table 1. AUV Dynamic Components [1]

Figure 1. 6-DOFnavigational frame [11].

Figure 2. AUV body-fixed and earth-fixed coordinate system.

The axes fixed in the earth are with the plane in the water surface and directed downward into the ocean. Second, the Earth-fixed coordinates frame is also measured to CB of AUV. If roll, pitch and yaw orientation angles of the AUV are zero, the axes will be parallel to the axes, respectively. e z

The third coordinate frame is fixed in the fluid, which can move with a constant velocity relative to the earth-fixed frame. The axes are always parallel to the axes, respectively [3].

A set of axes commonly used with the Earth-fixed axis system is shown in 0, where axis is chosen to point north, axis points east with the orthogonal triad being completed when axis pointing down.

A transformation matrix containing "Euler" angles where is roll, θ is pitch, and ψ is yaw, must be defined. More specifically, the adopted Euler angles move the reference frame to the referred body frame, following a Z-Y-X (or so-called 3-2-1) rotation sequence [13]. As a result, the transformation order from the earth-fixed frame ( ) to the body-fixed frame ( ) is given in the following equation: e b

$$\mathfrak{I}{e} \xrightarrow{u{3}} \mathfrak{I}{1} \xrightarrow{u{2}} \mathfrak{I}{2} \xrightarrow{u{1}} \mathfrak{I}_{b}$$

$$\tag{1}$$

Transformation matrix or direction cosine matrix is defined in the following equation:

Transformation matrix or direction cosine matrix is defined in the following equation: $$T(\phi,\theta,\psi) = \begin{bmatrix} \cos\psi\cos\theta & \sin\psi\cos\theta & -\sin\theta\ \cos\psi\sin\theta\sin\phi - \sin\psi\cos\phi & \sin\psi\sin\theta\sin\phi + \cos\psi\cos\phi & \cos\theta\sin\phi\ \cos\psi\sin\theta\cos\phi + \sin\psi\sin\phi & \sin\psi\sin\theta\cos\phi - \cos\psi\sin\phi & \cos\theta\cos\phi \end{bmatrix}$$ (2)

Transformation from a global velocity vector to the local velocity vector is given by

$$\begin{bmatrix} u \ v \ w \end{bmatrix} = T(\phi, \theta, \psi) \cdot \begin{bmatrix} \dot{X} \ \dot{Y} \ \dot{Z} \end{bmatrix}$$ (3)

The global angular velocity vector can be transformed into the rates of change of the Euler angles as given in Equation (4):

$$\begin{bmatrix} \dot{\phi} \ \dot{\theta} \ \dot{\psi} \end{bmatrix} = \begin{bmatrix} 1 & \sin\phi & \tan\theta & \cos\phi & \tan\theta \ 0 & \cos\phi & -\sin\phi \ 0 & \sin\phi/\cos\theta & \cos\phi/\cos\theta \end{bmatrix} \begin{bmatrix} p \ q \ r \end{bmatrix}$$ (4)

Three dimensional (3D) spatial rotations can be parametrized using both Euler angles and unit quaternions. Unit quaternion provides a convenient mathematical notation for representting orientation angles and rotations of vehicle in three dimensions.

Compared to Euler angles, unit quaternions are simpler to compose and avoid the problem of the gimbal lock. Additionally, compared to rotation matrices, they are more numerically stable and may be more efficient [14].

Nevertheless, Euler angles are preferred in this chapter for simplification reasons.

After a general model structure for AUV is derived, we look further into the modeling aspects in terms of environmental disturbance models, in which waves (wind generated), wind, and sea currents will be considered. In general, these environmental disturbances will both be additive and multiplicative to dynamic EOM [3].

Transformation order from body-fixed frame ( ) to sea current frame ( ) using orientation angle of α and β is given in following equation ^{b w}

$$\mathfrak{I}{b} \xrightarrow{u{2}} \mathfrak{I}{1} \xrightarrow{u{3}} \mathfrak{I}_{w} \tag{5}$$

The transformation matrix from the body-fixed frame ( ) to the sea current frame ( ): b ^w

$$\hat{C}^{(b,w)} = R_2(-\alpha) \times R_3(\beta)$$

$$\hat{C}^{(b,w)} = \begin{bmatrix} \cos\alpha & \cos\beta & -\cos\alpha & \sin\beta & -\sin\alpha \ \sin\beta & & \cos\beta & 0 \ \sin\alpha & \cos\beta & -\sin\alpha & \sin\beta & \cos\alpha \end{bmatrix}$$ (6)

The sea current velocity components are assumed constant relative to axes fixed in the earth. This permits the use of axes fixed in the fluid as an inertial frame of reference. AUV velocity with respect to the sea is expressed as

$$\vec{V}{b/w} = \vec{V}{b/e} - \vec{V}_{w/e} \tag{7}$$

Then, dynamic pressure is calculated as

$$Qd = \frac{1}{2} \rho V{b/w}^2 \tag{8}$$

In this chapter, however, we ignore the sea currents, which normally effects the AUV motion. Additionally, wind and wind generated waves phenomena will not be discussed since the attention is focused on AUVs performing a motion or manipulation task in an underwater environment.

2.1.5. Newtonian and Lagrangian Mechanics

The EOM of AUVs are highly nonlinear, time-varying and coupled due to hydrodynamic added mass, lift, drag, coriolis and centripetal forces, which are acting on the vehicle and generally include uncertainties [3]. Overall 6-DOF nonlinear dynamic EOM can be expressed in the matrix form as

$$M(v)\dot{v} + C(v)v + D(v)v + g(\eta) = \tau$$ (9)

where, M(v) is inertia matrix (including added mass), C(v) is matrix of Coriolis and centripetal terms (including added mass), D(v) is damping matrix, is vector of gravitational forces and moments, is vector of control inputs, and . [ , , , , , ]^T v u v w p q r

The coupled EOM of AUV are derived from two possible modeling approaches; one is a Lagrangian method and the other is a Newtonian-Euler formulation.

Basically, the Lagrangian approach consists of three main steps: first, to formulate a suitable expression for the vehicle's kinetic (T) and potential energy (P), second, to compute the Lagrangian L (L=T-P), and finally to apply the Lagrangian (L) to the Lagrangian formulation [12].

$$\frac{d}{dt} \left( \frac{\partial L}{\partial \dot{\eta}} \right) - \frac{\partial L}{\partial \eta} = J^{-T} \left( \eta \right) \tau \tag{10}$$

Another modeling approach is the Newtonian-Euler formulation, which is based on Newton"s Second Law that relates mass (m), acceleration (a) and force (F). Euler suggested expressing Newton's Second Law in terms of conservation of both linear and angular momentum. The forces (F) and moments (M) refers to the body's center of gravity [3].

In this study, the dynamic behavior of an AUV is described through Newton's laws of linear and angular momentum.

Newton"s Second Law is expressed as:

$$F = \frac{d}{dt} { mV } \tag{11}$$

$$\boldsymbol{M} = \frac{d}{dt}{\boldsymbol{H}}\tag{12}$$

where F represents the sum of all externally applied forces, M represents the sum of all applied torques, and H is the angular momentum.

It is convenient to regard the sums of applied torque (M) and force (F) as consisting of an equilibrium point and a perturbational component. Thus, assuming constant AUV mass

$$\mathbf{F} = \mathbf{F}_0 + \Delta \mathbf{F} = m \frac{d}{dt} { \mathbf{V} }$$ (13)

$$\mathbf{M} = \mathbf{M}_0 + \Delta \mathbf{M} = \frac{d}{dt} { \mathbf{H} } \tag{14}$$

The subscript "⁰ " denotes the equilibrium condition, and " " the component of perturbation. Equilibrium of AUV by definition must be an unaccelerated motion along a straight path; during this motion, the linear velocity vector relative to fixed space is invariant, and the angular velocity is zero. Thus, both and are zero.

Furthermore, since the axis system being used as an inertial reference system is the Earthfixed coordinate system, Equations (15) and (16) can be expressed as

$$\Delta \mathbf{F} = m \frac{d}{dt} { \mathbf{V}_T }_E \tag{15}$$

$$\Delta \mathbf{M} = \frac{d}{dt} { \mathbf{H} }_E \tag{16}$$

The force equation based on rate of change of velocity V relative to Earth"s fixed axis frame is given by

$$\vec{F} = D_e \left( m\vec{V} \right) = Db \left( m\vec{V} \right) + \vec{w}{b/e} \times m\vec{V} \tag{17}$$

where is the angular velocity of the AUV with respect to the Earth fixed coordinate frame. The open form of Equation (17) is given by

$$F \Rightarrow \begin{cases} F_x = X = m(\dot{u} + qw - rv) \ F_y = Y = m(\dot{v} + ru - pw) \ F_z = Z = m(\dot{w} + pv - qu) \end{cases}$$ (18)

After the rearrangement of F, translational accelerations become:

$$\dot{u} = \frac{F_x}{m} - qw + rv$$

$$\dot{v} = \frac{F_y}{m} - ru + pw$$

$$\dot{w} = \frac{F_z}{m} - pv + qu$$ (19)

Similarly, after transforming from the body-fixed frame to the Earth-fixed frame the moment equation and its open form becomes

ment equation and its open form becomes $$\vec{M} = D{e} (\hat{I} \vec{\omega}{b/e}) = D{b} (\hat{I} \vec{\omega}{b/e}) + \vec{\omega}{b/e} \times (\hat{I} \vec{\omega}{b/e})$$

$$\vec{M} = \sum{e} { M{x} = K = I{x} \dot{p} - I{yz} (q^{2} - r^{2}) - I{zx} (\dot{r} + pq) - I{xy} (\dot{q} - rp) - (I{y} - I{z}) qr }$$

$$M{y} = M = I{y} \dot{q} - I{zx} (r^{2} - p^{2}) - I{xy} (\dot{p} + qr) - I{yz} (\dot{r} - pq) - (I{z} - I_{x}) rp }$$

$$M{z} = N = I{z} \dot{r} - I{xy} (p^{2} - q^{2}) - I{yz} (\dot{q} + rp) - I{zx} (\dot{p} - qr) - (I{x} - I_{y}) pq$$ (20)

For a rigid body, angular momentum can be defined as

$$\boldsymbol{H} = \boldsymbol{I}\boldsymbol{\omega} \tag{21}$$

where the inertia matrix is defined as

$$I = \begin{bmatrix} I{xx} & -I{xy} & -I{xz} \ -I{xy} & I{yy} & -I{yz} \ -I{xz} & -I{yz} & I_{zz} \end{bmatrix}$$ (22)

where denotes a moment of inertia, and a product of inertia . In this chapter, we assume that the AUV is symmetrical along the XY and XZ planes, therefore cross inertia parameters become j i

$$I{xy} = I{yz} = I_{xz} = 0$$ and $I_y = I_z$ (23)

As a result of this symmetry, the rotational EOM of AUV in particular, becomes:

$$M{x} = I{x} \dot{p}$$

$$M{y} = I{y} \dot{q} \cdot (I{z} - I{x}) r p$$

$$M{z} = I{z} \dot{r} \cdot (I{x} - I{y}) p q$$ (24)

After rearrangement, Equation (24) is expressed as:

$$\dot{p} = \frac{M_x}{I_x}$$

$$\dot{q} = \frac{1}{I_y} \Big[ M_y + (I_z - I_x) r p \Big]$$

$$\dot{r} = \frac{1}{I_z} \Big[ M_z + (I_x - I_y) p q \Big]$$ (25)

Finally, the derivative of Euler angles is defined as:

$$\dot{\psi} = \frac{q \sin \varphi + r \cos \varphi}{\cos \theta}$$

$$\dot{\theta} = q \cos \varphi - r \sin \varphi$$

$$\dot{\varphi} = p + (q \sin \varphi + r \cos \varphi) \tan \theta$$ (26)

2.1.6. Gravitational Forces

Gravitational forces always exist in the AUV. It can be assumed that gravity acts at the center of gravity (CG) of the vehicle. Since the centers of mass and gravity coincide in an AUV, there is no external momentum produced by gravity on the CG.

Therefore, for the body axis system, gravity contributes only to the external force vector F. Three components of the gravitational force in the body frame depend on the AUV"s attitude relative to an inertia frame. The gravitational force acting upon an AUV is most obviously expressed in terms of the Earth"s axes. With respect to these axes, the gravity vector mg, is directed along the axis. Figure 3 demonstrates the alignment of the gravity vector with respect to the body-fixed axes. In Figure 3, θ represents the pitch angle between the gravity vector and the plane; the angle is positive when the nose of the AUV goes up, Φ

represents the roll angle between axis and the projection of the gravity vector on the plane. Direct solution of the vector mg into , and components produces: X

$$g_x = mg\sin(-\theta) = -mg\sin\theta$$

$$g_y = mg\cos(-\theta)\sin\Phi = mg\cos\theta\sin\Phi$$

$$g_z = mg\cos(-\theta)\cos\Phi = mg\cos\theta\cos\Phi$$ (27)

In general, the Euler angles ( , , ) are not simply the integrals of the angular velocity p, q, and r (Equation (25) ). It is necessary to relate these and their derivatives to the angular velocities p, q, and r. This depends upon whether the gravitational vertical seen from the AUV is fixed or whether it rotates relative to inertial space [16].

The manner in which the angular orientation and velocity of the body axis system with respect to the gravity vector is expressed by depending upon the angular velocity of the body axes about the vector mg. The external forces acting on the AUV can be expressed as:

$$X = F_x + g_x$$

$$Y = F_y + g_y$$

$$Z = F_z + g_z$$ (28)

where are the gravitational terms, and represent the hydrodynamic and thrust forces respectively.

Figure 3. The orientation of gravity vector with respect to the body axis system [15].

Hence, from the Equations (13) and (15), the force equations become:

$$F{x} = ma{x_{CG}} = m(\dot{u} + qw - rv + g\sin\theta)$$

$$F{y} = ma{y_{CG}} = m(\dot{v} + ru - pw - g\cos\theta\sin\Phi)$$

$$F{z} = ma{z_{CG}} = m(\dot{w} + pv - qu - g\cos\theta\cos\Phi)$$ (29)

The gravitational acceleration forces and moments are represented by the weight minus buoyancy (W−B) and weight moment terms respectively [4]. In Figure 4, the Simulink model of gravitational force calculation is shown.

2.1.7. Hydrostatic Forces and Moments

When an AUV is submerged in a fluid under the effect of gravity, two forces act on the vehicle: the gravitational force, which is mentioned in the previous sub-section and the buoyancy, which is called "hydrostatic effect".

The buoyant force acting on the center of buoyancy (CB) is represented in body-fixed frame (Figure 5). It can be recognized that the difference between gravity and buoyancy (W-B) only affects the linear force acting on the vehicle. It is also clear that the restoring linear force is constant in the Earth-fixed frame [11]. On the other hand, the two vectors of the first moment of inertia W and B affect the momentum acting on the vehicle and are constant in the body-fixed frame. A solid body submerged in a fluid will have upward buoyant force acting on it equivalent to the weight of displaced fluid, enabling it to float or at least appear to become lighter. If the buoyancy exceeds the weight, then the object floats; if the weight exceeds the buoyancy, the object sinks. If the buoyancy equals the weight, the body has neutral buoyancy and may remain at its level. Discovery of the principle of buoyancy, which is a result of the hydrostatic pressure in the fluid, is attributed to Archimedes [17].

After applying hydrostatic force coefficients to Equation 29, translational force equations become:

Figure 4. Gravitational force calculation.

Similarly, after applying the hydrostatic moment coefficients to Equation (26), the moment equations become:

Figure 5. Hydrostatic forces and moments [15].

Figure 6. Hydrostatic force and moment calculation.

$$M{x} = I{x}\dot{p} + (z{g}W - z{b}B)\cos\theta\sin\phi$$

$$M{y} = I{y}\dot{q} - (I{z} - I{x})rp + (z{g}W - z{b}B)\sin\theta$$

$$M{z} = I{z}\dot{r} - (I{x} - I{y})pq$$ (31)

In Figure 6, hydrostatic forces and moments are implemented in Simulink.

2.1.8. Hydrodynamic Forces and Moments

In this subsection, the main hydrodynamic effects acting on an AUV moving in a fluid (seawater) will be briefly discussed. This is the key part of AUV modeling. Standard EOM contain only stability derivatives for the specific AUV configuration of interest.

Hence, trajectory simulation or prediction using traditional methods requires a priori knowledge of the hydrodynamic characteristics of the vehicle in the flow regimes, which may occur during the maneuver [15].

All hydrodynamic parameters are defined uniquely for a given AUV shape by formulae based on the results of theory and experiment.

There are primarily three methods of determining the hydrodynamic coefficients in the design process of underwater vehicles: (1) towing tank tests, (2) numerical computations, and (3) field tests. Among these methods, the most reliable results are obtained from the field tests, where the whole designed model is tested in a real sea environment [1]. Towing tank tests are performed with a scaled model and the hydrodynamic forces and moments can accurately be determined. However, experimental testing of designs is a time consuming and costly process (construction of the models, instrumentations, test infrastructure, etc.).

Numerical Computations are mainly based on semi-empirical or CFD methods. Semiempirical or potential theory-based methods are generally utilized in the preliminary design process, where it is important to determine the hydrodynamic characteristics in a short period of time. CFD methods give accurate results and are used in the detailed design process [12].

The theory of fluid dynamics is rather complex and it is difficult to develop a reliable and robust model for most of the hydrodynamic effects. A rigorous analysis for incompressible fluids would need to resort to the Navier-Stokes equations, which are the basic governing equations for a viscous, heat-conducting fluid.

These equations describe how the velocity, pressure, temperature and density of a moving fluid are related. The hydrodynamic prediction method is coupled with a 6-DOF EOM solver to predict vehicle trajectories. The predicted motion characteristics of the AUV are also sensitive to details of the predicted hydrodynamic characteristics of the vehicle.

Small perturbations in the flow field, which cause small; variations in the vehicle forces and moments, accumulate over the length of the hull, and can produce large perturbation; in the calculated trajectory.

In this chapter in order to determine hydrodynamic forces and moments, we used SUBFLO_2, which is an engineering physically based and commercially available software tool. SUBFLO_2 has four major components which are hull separation vortex method, fin horseshoe vortex method, propeller/propulsion models, and 6-DOF equations of motion solver. The hydrodynamic prediction method is coupled with a 6-DOF EOM solver to predict vehicle trajectories. The predicted motion characteristics of the AUV are also sensitive to details of the predicted hydrodynamic characteristics of the vehicle. Small perturbations in the flow field, which cause small; variations in the vehicle forces and moments, accumulate over the length of the hull, and these can produce large perturbation; in the calculated trajectory [23].

In order to determine hydrodynamic coefficients utilizing with SUBFLO_2, first we find three static hydrodynamic force coefficients of CFx, CFy, and CFz, and then three static hydrodynamic moment coefficients of CMSFx, CMSFy, and CMSFz, for the predefined AUV geometry.

These coefficients are calculated due to the parameters of δ^s , δ^r , α, β, Vb/w, which are elevator deflection, rudder deflection, angle of attack, side slip angle, and velocity vector with respect to water, respectively. In Figure 7, the Simulink model of hydrodynamic force and moment calculation is shown.

Figure 7. Hydrodynamic force and moment calculation.

2.1.9. Added Mass and Inertia

When a rigid body is moving in a fluid, the additional inertia of the fluid surrounding the body, which is accelerated by the movement of the body, has to be considered. This effect can be neglected in industrial robotics since the density of the air is much lighter than the density of a moving mechanical system.

However, in underwater applications the density of the water, ρ ≈ 1000 kg/m³ , is comparable with the density of the vehicles. In particular, at zero degrees, the density of the fresh water is 1002.68 kg/m³ ; for sea water with 3.5% of salinity it is ρ = 1028.48 kg/m³ .

Since the fluid surrounding the body is accelerated with the body itself, a force is then necessary to achieve this acceleration (the fluid exerts a reaction force which is equal in magnitude and opposite in direction).

This reaction force is the added mass contribution [16].

The added mass is not a quantity of fluid to add to the system such that it has an increased mass. Different properties hold with respect to the six by six inertia matrix of a rigid body due to the fact that the added mass is a function of the body"s surface geometry.

As an example, the inertia matrix is not necessarily positive definite. The added mass has also an added Coriolis and Centripetal contribution.

In this study, "Added Mass and Inertia" effects of water is not taken into consideration. Added (virtual) mass should be understood as pressure-induced forces and moments due to a forced harmonic motion of the body, which are proportional to the acceleration of the body [12].

Consequently, the added mass forces and acceleration will be 180˚ out of phase to the forced harmonic motion. However, this is not true when AUV is close to surface.

2.1.10. Propeller Effect

The propeller produces the main thrust. Consequently, the reaction of the body to the load torque of the propeller produces a moment with respect to its rotational axis. The vehicle is a nonlinear system: all equations of motion of the system include coupled terms [1]. The main terms of this type are in the longitudinal (X) force and roll moment (K) equations because the thrust forces and moments act in the direction of the x-axis.

A propeller with a rudder can produce a thrust vector within a range of directions and magnitudes in the horizontal plane for low speed maneuvering and dynamic positioning. In our physically based model, we used a simple fixed-trust model due to the AUV velocity.

2.2. Mathematical Model of AUV

Detailed explanations on the theory behind the main concepts that directly influenced the design of the dynamic mathematical model of AUV are given in the previous sub-section. Now we arrive at the combined overall non-linear EOM for the AUV in 6-DOF as follows [18]:

Surge or translational motion along the x-axis:

Surge or translational motion along the x-axis: $$X = X_{\dot{u}}\dot{u} + m\left[-\dot{u} - z_G\dot{q} + yG\dot{r}\right] + X{uu}u|u| + (X{wq} - m)wq + \left(X{qq} + mxg\right)q^2 + \left(X{vr} + m\right)vr + \left(X_{rr} + mx_g\right)r^2 - mzgpr - (W - B)\sin\theta + X{prop}$$ (32)

Sway or translational motion along the y-axis:

Sway or translational motion along the y-axis: $$Y = Y{\dot{v}}\dot{v} + Y{\dot{r}}\dot{r} + m\left[-\dot{v} + z_g\dot{p} - xg\dot{r}\right] + Y{uv}uv + \left(Y{wp} + m\right)wp + \left(Y{ur} - m\right)ur - \left(mzg\right)qr + \left(Y{pq} - mxg\right)pq + Y{vv}v|v| + Y{rr}r|r| + (W - B)\cos\theta\sin\phi + Y{uu\delta_r}u^2\delta_r$$ (33)

Heave or translational motion along the z-axis:

Heave or translational motion along the z-axis:

$$Z = Z_w \dot{w} + Z_q \dot{q} + m \left[ -\dot{w} + x_g \dot{q} - yg \dot{p} \right] + \left( Z{uq} + m \right) uq + \left( Z_{vp} - m \right) vp + \left( mz_g \right) p^2 +$$

$$+ Z_{uw} uw + \left( mzg \right) q^2 + \left( Z{rp} - mxg \right) rp + Z{ww} w |w| + Z{qq} q |q| + (W - B) \cos\theta \cos\phi + Z{uu\delta_c} u^2 \delta_e$$ (34)

Roll or rotation about the x-axis:

Roll or rotation about the x-axis:

$$K = mz_g \dot{u} - myg \dot{r} - (I{xx} - K_p) \dot{p} + (myg) \dot{w} - (I{zz} - I_{yy}) qr - (mz_g) wp +$$

$$+ (mz_g) ur - (z_g W - z_b B) \cos\theta \sin\psi - (y_g W - yb B) \cos\theta \cos\psi + K{pp} p|p| + K_{prop}$$ (35)

Pitch or rotation about the y-axis:

Pitch or rotation about the y-axis:

$$M = -mz_g \dot{u} + \left(mxg + M{\dot{w}}\right) \dot{w} - \left(I{yy} - M{\dot{q}}\right) \dot{q} + \left(M{rp} + I{zz} - I_{xx}\right) rp + \left(mz_g\right) vr - \ - \left(mz_g\right) wq + \left(Muq - mxg\right) uq + M{uw}uw + \left(M_{vp} + mxg\right) vp + M{qq}q|q| + M_{ww}w|w| - \ - \left(z_g W - z_b B\right) \sin\theta - \left(x_g W - xb B\right) \cos\theta \cos\psi + M{uu\delta_e} u^2 \delta_e$$ (36)

Yaw or rotation about the z-axis:

Mustafa Dinc and Chingiz Hajiyev $$N = my_g \dot{u} + \left(N_v - mx_g\right) \dot{v} + \left(Nr - I{zz}\right) \dot{r} + \left(Npq + I{xx} - I{yy}\right) pq + \left(N_{wp} - mxg\right) wp + \ + \left(N{ur} - mxg\right) ur + N{uv} uv + N{rr} r|r| + N{vv} v|v| + \left(x_g W - x_b B\right) \cos\theta \sin\psi + \left(y_g W - yb B\right) \sin\theta + N{uu\delta_r} u^2 \delta_r$$ (37)

where X, Y, Z, K, M, and N represent the resultant forces and moments with respect to the body-fixed coordinates. The parameters of the mathematical model in Equations (32)-(37) with their definitions can be found in Tables 3-5 at the end of this chapter.

2.3. Numerical Integration of the AUV EOM

The nonlinear differential equations defining the AUV accelerations and the kinematic equations give us the vehicle accelerations in two different reference frames. Given the complex and non-linear nature of these equations, we will use numerical integration to solve for the vehicle speed, position, and attitude in time frame [19]. Consider that at each time step, we can express nonlinear differential equation as follows:

$$\dot{x}_n = f\left(x_n, u_n\right) \tag{38}$$

where x is the AUV state vector:

$$x = [u \ v \ w \ p \ q \ r \ x \ y \ z \ \phi \ \theta \ \psi]^{T}$$ (39)

and uⁿ is the input vector:

$$x = [u \ v \ w \ p \ q \ r \ x \ y \ z \ \phi \ \theta \ \psi]^{T}$$

$$\tag{40}$$

There are two common numerical iteration methods to solve the non-linear differential equation: one is Euler"s method and the other is Runge-Kutta method.

2.3.1. Euler's Method First Order

We will first consider Euler's method, a simple numerical approximation that consists of applying the iterative formula:

$$x_{n+1} = x_n + f(x_n, u_n) \cdot \Delta t \tag{41}$$

where is the modeling time step.

Although the least computationally intensive method, Euler's method is unacceptable as it can lead to divergent solutions for large time steps [19].

2.3.2. Runge-Kutta Method

In this subsection, we will introduce one of the most powerful predictor-corrector algorithms- one which is so accurate, that most computer packages designed to find numerical solutions for differential equations will use it by default - the Runge-Kutta fourth order method. The Runge-Kutta method further improves the accuracy of the approximation by averaging the slope at four points [19]. We used this method to solve the nonlinear dynamic model of the AUV. In this method, numerical approximations that consist of applying the iterative formula:

$$k{1} = x{n} + f(x{n}, u{n})$$

$$k{2} = f\left(x + \frac{\Delta t}{2}k{1}, u_{n + \frac{1}{2}}\right)$$

$$k{3} = f\left(x + \frac{\Delta t}{2}k{2}, u_{n + \frac{1}{2}}\right)$$

$$k{4} = f\left(x + \Delta t k{3}, u_{n + 1}\right)$$ (42)

where the interpolated input vector is:

$$u_{n+\frac{1}{2}} = \frac{1}{2} (un + u{n+1}) \tag{43}$$

The combination of these two Equations (42)-(43) yields to:

$$x_{n+1} = x_n + \frac{\Delta t}{6} \left( k_1 + 2k_2 + 2k_3 + k_4 \right) \tag{44}$$

which is simply the x-value of the current point plus a weighted average of four different xjump estimates for the interval, with the estimates based on the slope at the midpoint being weighted twice as heavily as those using the slope at the end-points [10].

For estimations of rigid-body aerodynamic or hydrodynamic models, the second order Runge-Kutta algorithm is usually adequate and hence recommended during the initial iterations of the iterative estimation algorithms, switching over the forth-order Runge-Kutta only during the final iteration [20].

2.4. Dynamic Simulink Model and Control of AUV

The data achieved from the modeling and identification process of the AUV was implemented into a vehicle which was used in a MATLAB Simulink environment.

This makes it possible to simulate the behaviour of the AUV or programs without the need of a real sea environment.

The detailed AUV Simulink models are depicted in Figure 8. In this model, hydrodynamic, hydrostatic, gravitational, and kinematic EOM of AUV and sliding mode control are implemented in the MATLAB version 7.5 Simulink environment.

The sliding mode control algorithm, which is a robust control, is applied to yaw channel of the dynamic model. The control input of course is

$$u = \psi - \Delta \psi - \lambda (r - \Delta \dot{\psi}) \tag{45}$$

The sliding mode control law for course control of AUV is

$$\frac{-5\pi}{18}|u|sign(u)\tag{46}$$

The control strategy used for physically based model of AUV is depicted in Figure 9, which is the Simulink model of sliding mode control of yaw (ψ).

Figure 8. Simulink model of AUV.

Figure 9. Simulink Model of Sliding Mode Control.

2.5. Simulation Results

The simulation of the dynamic model is developed by using the MATLAB version 7.5 Simulink environment. Numerical simulations are made to show the dynamic model of the AUV.

In Figure 10, the simulation results of depth change in z axis, dynamic position maneuver in x and y axis, and yaw angle change are shown. In Figure 16, the simulation result of deflection angle change of rudder under the sliding mode control and total velocity during 50 sec. For this simulation, the AUV has a maneuver speed of 8 m/s and rudder angle applies to 10 deg. From the start, elevator angle applies to 1 deg, and the simulation time is 50 sec.

Figure 10. Simulation Results of X, Y, Z, Roll, Pitch and Yaw Angles (50 sec).

Figure 11. Simulation Results of V_tot, δr and δe (50 sec).

3. PARAMETER IDENTIFICATION WITH LEAST SQUARES ESTIMATION

This section focuses on the parameter identification of hydrodynamic coefficients of AUVs based on the Least Square Estimation (LSE) algorithm for a nonlinear mathematical modeling of AUVs. Hydrodynamic coefficients strongly affect the dynamic performance of an AUV. Therefore, it is important to have the true values of these coefficients in order to accurately simulate the AUV"s dynamic performance. The estimated coefficients can be used as inputs not only for a mathematical model to analyze the maneuvering performance but also for a controller model to design AUVs under development.

However, parameter identification of AUV dynamics is complicated because of its nonlinear identification models and the combination of noisy sensor measurements.

3.1. AUV System Identification

System identification is the determination on the basis of an observation of input and output, of a system within a specified class of systems to which the system under test is equivalent [20].

When the system identification is applied to AUVs, the equations governing the AUV dynamic motion are postulated and an experiment is designed to obtain measurements of input and output variables [21].

The hydrodynamic forces and moments acting on AUV cannot be measured directly.

However, hydrodynamic modeling followed by parameter estimation allows determination of specific hydrodynamic characteristics (such as lift, drag, and side force coefficients, and rolling, pitching, and yawing moment coefficients in terms of stability and control derivatives) from the related measurements such as accelerations, angular rates, etc. [20].

The selected EOM for system identification are sway and yaw

The selected EOM for system identification are sway and yaw $$Y = Y{v}\dot{v} + Y{r}\dot{r} + m\left[-\dot{v} + z{g}\dot{p} - x{g}\dot{r}\right] + Y{uv}uv + (Y{wp} + m)wp + (Y{ur} - m)ur - \left(mz{g}\right)qr + (Y{pq} - mx{g})pq + Y{vv}v|v| + Y{rr}r|r| + (W - B)\cos\theta\sin\phi + Y{uu\delta{r}}u^{2}\delta_{r}$$

$$N = my{g}\dot{u} + (N{v} - mx{g})\dot{v} + (N{r} - I{zz})\dot{r} + (Npq + I{xx} - I{yy})pq + (N{wp} - mx{g})wp + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} - mx{g})ur + (N{ur} -$$

$$-(mz{g})qr + (Y{pq} - mx{g})pq + Y{vv}v|v| + Y{rr}r|r| + (W - B)\cos\theta\sin\phi + Y{uu\delta{r}}u^{2}\delta{r}$$

$$N = my{g}\dot{u} + (N{v} - mx{g})\dot{v} + (N{r} - I{zz})\dot{r} + (Npq + I{xx} - I{yy})pq + (N{wp} - mx{g})wp + (N{ur} - mx{g})ur + +N{uv}uv + N{rr}r|r| + N{vv}v|v| + (x{g}W - x{b}B)\cos\theta\sin\psi + (y{g}W - y{b}B)\sin\theta + N{uu\delta{r}}u^{2}\delta_{r}$$

$$(48)$$

The dynamic Equations (47) and (48) are augmented with output equations that specify the connection of AUV states and controls to measured outputs, along with measurement equations describing the measurement process [21]. In Figure 12, Model Postulation is based on a priori knowledge about the AUV kinematics and hydrodynamics.

The Experiment Design includes selection of an instrumentation system, and specification of an AUV configuration and maneuvers for system identification.

Data Compatibility Analysis, in practice, measured AUV response data that can contain systematic errors, even after careful instrumentation and experimental procedures.

To verify data accuracy, data compatibility analysis can be applied to measured AUV responses. Model Structure Determination in AUV system identification means selecting a specific form for modeling from a class of models, based on measured data. Parameter and State Estimation constitutes a principal part of the AUV system identification procedure. Parameter estimation for linear dynamic systems based on maximum likelihood and the least squares principles can also be formulated in the frequency domain.

Collinearity diagnostics is in almost all practical applications of linear regression and the model terms are correlated to some extent. Diagnostic information can aid in deciding what corrective actions are necessary.

Model Validation is the last step in the identification process [22]. The identified model must demonstrate that its parameters have physically reasonable values and acceptable accuracy, and that model has a good prediction capability on comparable maneuvers [21].

3.2. Parameter Estimation for AUV

The parameters can be estimated from calculating the first principles of dynamics of AUVs, from statistical LSE of experimental data, or a combination of the two. The identification of the parameters of the item being tracked can be approached in many different ways. The parameters can be determined once, from a model of the expected input signal.

It also could be done adaptively utilizing a recursive method to identify the parameters of the incoming signal.

The modern era of system identification is marked by the implementation of the maximum-likelihood method [20]. Therefore, we used the Least Square model for the uncertainties in the parameters and measurements in this chapter. Based on this model, is a vector of unknown constant parameters and v is a random vector of measurement noise. There are two different models that are Bayesian and Fisher [21]. The LSE algorithm always gives the best linear fit when the noise is white and Gaussian.

Figure 12. AUV System Identification Method [21].

Moreover, it is easier to tune the identification process in the latter part with a priori knowledge of the system [21]. A recursive LSE was chosen in this study for the parameter estimation since it is easy to apply the system dynamic and allows for the consideration of modeling and measurement errors.

The parameter estimation process consists of finding values of unknown model parameter in an assumed model structure, based on noisy measurements z. An estimator is a function of the random variable z that produces an estimate of the unknown parameters . Since the estimator computes based on noisy measurements z, is a random variable [21]. In order to estimate the hydrodynamic coefficients of AUVs, the LSE is designed using the observer model. The estimated parameters yields to ˆ

$$\theta = [Y{vv}Y{rr}Y{uv}Y{ur}Y{wp}N{vv}N{rr}N{uv}N{ur}N{wp}]$$ (49)

The output y is given by:

$$y = X\theta \tag{50}$$

where the matrix X is assumed to be known. Then the measurement equation can be expressed as:

$$z = X\theta + v \tag{51}$$

where is the length measurement vector, is the vector of unknown parameters (np=n+1), is the matrix of vectors of ones and regressors, and is the vector of measurement errors.

In general, there are n⁰ measured outputs, and a vector of measurements is taken at each sample i, where i = 1,2,3,…, Nis the number of sampled data points. A single measured output is assumed, son0=1, and z is a vector composed of N scalar measurements [21].

Neglecting interactions from other parts of the system, the sway and yaw measurement subsystems can be modeled as

$$Z_{Ymeas} = Y - mwp + (mz_g)qr + (mxg)pq - Y{uu\delta_r}u^2\delta_r$$ (52)

$$Z{Nmeas} = N - (N{pq} + I{xx} - I{yy})pq + (yg W)\sin\theta + N{uu\delta_r} u^2 \delta_r$$ (53)

For the least squares model, there are no probability statements regarding , but v is assumed to be zero mean and uncorrelated, with a constant variance:

$$E(v) = 0$$ and $E(vv^T) = \sigma^2 I$ (54)

The unknown parameter vector can be estimated via LSE as follows

$$\hat{\theta} = (X^T X)^{-1} X^T z \tag{55}$$

The n^p n^p matrix matrix is always symmetric. If the regressor vectors that make up the columns of X are linearly independent, then is positive definite and the eigenvalues of are positive real numbers, and the associated eigenvectors are mutually orthogonal so the exists [21].

Finally, the covariance matrix of the parameter estimate also known as the covariance matrix of the estimation error is simplified to

$$cov(\theta{est}) = E[(\theta{est} - \theta)(\theta_{est} - \theta)^T] = \sigma^2 (X^T X)^{-1}$$ (56)

3.3. Simulation Results of Parameter Identification

Numerical simulations are made to show the parameter identifications of the AUV. The hydrodynamic coefficients associated with horizontal and vertical motions are estimated by simulating the combined yaw and sway motion of the AUV.

The numerical values of the AUV physical parameters, hydrodynamic force coefficients and hydrodynamic moment coefficients are given in Tables 3-5 respectively.

We will describe how state-of-the art algorithms are used to find a dynamic model that enables accurate simulation of steering dynamics.

In summary, the system identification task to estimate the parameters (56). The parameter identification results are given in Table 2.

Numerical results demonstrate that the LSE estimates the hydrodynamic coefficients very well.

Parameters	Real Value	Est. Value with Int. Nav.	Percent %	Difference
Yvv	-196,26	-192,650	1,839	3,610
Yrr	8,30	8,580	3,373	0,280
Yuv	38,39	38,277	0,295	0,113
Yur	13,41	13,320	0,674	0,90
Ywp	35,50	36,105	1,704	0,605
Nvv	-7,95	-7,733	2,736	0,218
Nrr	-24,13	-24,051	0,328	0,079
Nuv	-21,89	-21,843	0,216	0,047
Nur	1,93	1,922	0,435	0,008
Nwp	-19,30	-19,367	0,347	0,067

Table 2. Parameter Identification Results

The simulation results of the real model and identified model are shown in Figures 13-15 respectively. Figures show that the AUV had a surge velocity changed between 5 and 0 m/s. The depth changed between -15 m and -13.5 m. The elevator was operated at the maximum deflection of 15 deg. for most of the time and similarly the rudder deflection operated at the maximum deflection of -11 deg. The iteration number *N*is chosen to be 5,000 and the sampling time is chosen to be 0.001 sec. As seen from Figures 13, 14 and 15, outputs of real and identified models are very close. On the one hand, this is obviously seen in Figure 16. As a result, Simulation results show that the proposed method works very well. For model verification, the existing in literature methods [10, 20, 22] can be used.

Figure 13. Velocity simulation results of real and identified model.

Figure 14. Position simulation results of real and identified model.

Figure 15. Attitude simulation results of real and identified model.

Figure 16. Attitude simulation results of real and identified model (Zoom in Version).

Parameter	Value	SCI unit	Definition
ρ	1010	g/m3	Fluid density
g	9.81	m/s2	Gravitational acceleration
d	0.191	m	Hull cylindrical radius
l	1.33	m	Overall hull length
Sw	0.7981	2 m	Submerged Area (πdl)
Ap	0.2540	2 m	Body base area (dl)
Af	0.0287	2 m	Body cross area (πd2 /4)
cds	0.004	-	Surface drag coefficient
cdF	0.3	-	Drag coefficient (Af)
cda	0.0166	-	Total drag coefficient (for Sw)

Table 3. (Continued)

Parameter	Value	SCI unit	Definition
lcp	0.2645	M	Distance to center of pressure
tfin	1010	g/m3	Fluid density
hfin	0.0960	m	Rudder height
Ar , As	0.0071	2 m	Vertical/horizontal rudder surface area
cdf	1.558	-	Rudder drag coefficient
W	300	N	AUV weight
B	306	N	Buoyant force
xb	0	m	Buoyancy center about the x-axis
yb	0	m	Buoyancy center about the y-axis
zb	0	m	Buoyancy center about the z-axis
xg	0	m	Gravity center about the x-axis
yg	-0.008	m	Gravity center about the y-axis
zg	0.0196	m	Gravity center about the z-axis
Ixx	0.177	kg m2	Moment of inertia about the x-axis
Iyy	3.45	kg m2	Moment of inertia about the y-axis
Izz	3.45	kg m2	Moment of inertia about the z-axis
R	5.87E-2	m	Diameter of propeller
u0	1.5	m/s	Design velocity

Table 4. AUV Hydrodynamic Force Coefficients [10]

Parameter	Value	SCI unit	Definition
Xuu	-6.68	kg/m	Axial drag
Xu	-0.513	kg	Added mass
Yvv	-196.26	kg/m	Cross flow drag
Yrr	8.30	kg m/rad2	Cross flow drag
Yuv	-38.39	kg/m	Body and rudder buoyancy
Yv	-42.13	kg	Added mass
Yr	-5.16	kg m/rad	Added mass
Yur	13.41	kg/rad	Added mass + Rudder buoyancy
Yuu r	16.99	kg/m/rad	Buoyant force of rudder
Zww	-196.26	kg/m	Cross flow drag
Zqq	-8.30	kg m/rad2	Cross flow drag
Zuw	-38.39	kg/m	Body and rudder buoyancy
Zw	-42.13	kg	Added mass
Zq	5.16	kg m/rad	Added mass
Zuq	-13.41	kg/rad	Added mass + Rudder buoyancy
Zuu s	-16.99	kg/m/rad	Rudder buoyancy

Parameter	Value	SCI unit	Definition
Kpp	-5.03	kg m2 /rad2	Roll drag
Kp	-0.095	kg m2 /rad2	Added mass moment
Mww	7.95	kg	Cross flow drag
Mqq	-24.13	kg m2 /rad2	Cross flow drag
Muw	21.89	kg	+ Rudder buoyancy Added mass + MunkMomenti
Mw	5.16	kg m	Added Mass Inertia
Mq	-7.57	kg m2 /rad	Added Mass Inertia
Muq	-16.56	kg m/rad	Added mass + Rudder buoyancy
Muu s	-13.92	kg/rad	Rudder Buoyancy Moment
Nvv	-7.95	kg	Cross flow drag
Nrr	-24.13	kg m2 /rad	Cross flow drag
Nuv	-21.89	kg	+ Rudder buoyancy Added mass + MunkMomenti
Nv	-5.16	kg m	Added Mass Inertia
Nr	-7.57	kg m2 /rad	Added Mass Inertia
Nuu r	-13.92	kg rad	Rudder Buoyancy Moment

Table 5. AUV Hydrodynamic Moment Coefficients [10]

CONCLUSION

The research on underwater systems has gained immense attention during the last two decades because of applications taking place in many fields. Therefore, the significant number of UUVs has been developed for solving the wide range of scientific and applied tasks of ocean research and development in the world. The military, as well as civilian industry can see great potential uses of AUVs in the underwater environment.

This chapter is concerned with implementing dynamic modeling of an AUV. This chapter has also given an insight and theoretical background about AUV kinematics, hydrodynamics and hydrostatic. Because physically based mathematical modeling of an AUV is an interdisciplinary study of kinematics, hydrostatics, and hydrodynamics.

To achieve an accurate hydrodynamic model for any vehicle that is not a standard shape by purely analytical means is extremely difficult at best.

This chapter also discussed the use of parameter identification methods to predict the hydrodynamic derivatives of the AUV. LSE is used to tackle the problem of parameter identification of an AUV. This classical approach seems to have a better performance in cases where a specific parameter will be identified and the LSE results obtained are satisfactory. Therefore, numerical results demonstrate that the proposed method works very well, in both cases of using simulated data. It has been shown that the characteristic functions for LSE can be obtained in a stable way even in noisy environments.

Numerical simulations are made to show the parameter identifications of the AUV. The hydrodynamic coefficients associated with horizontal and vertical motions are estimated by simulating the combined yaw and sway motion of the AUV.

The simulation results indicate that for AUV hydrodynamic parameter identification purposes, the LSE algorithm remains a feasible tool, which consistently returns quality results and is the least costly in terms of computational demand.

Autonomous guidance, navigation, and control techniques are key research and development areas for success of the specific AUV missions.

Moreover, further work is needed for in precision navigation, sensor development and integration, and improving the reliability and robustness of long term and complex mission completion.

REFERENCES

• [1] Inzartsev, A. V., Underwater Vehicles. In-Tech, Rijeka, 2009.
• [2] Narimani, M., Loueipour, M. and Depth Control of Autonomous Underwater Vehicles. UDT Europe, 2007.
• [3] Fossen, I. T., Guidance and Control of Ocean Vehicles. University of Trondheim, John and Wiley Sons Press, 1994.
• [4] Fauske, K. M., Gustafsson, F., Hegrenaes, O., Estimation of AUV Dynamics for Sensor Fusion. 10th International Conference on Information Fusion, 2007.
• [5] Hwang, W. Y., Application of System Identification to Ship Maneuvering. Ph.D. Dissertation, Massachusetts Institute Technology MIT, Cambridge, US, 1980.
• [6] Meng, L., Li, L., Veres, S. M., Aerodynamic Parameter Estimation of an Unmanned Aerial Vehicle Based on Extended Kalman Filter and Its Higher Order Approach. 978- 1-4244-5848-6, 2010.
• [7] Chowdhary, G., Jategaonkar, R., Aerodynamic Parameter Estimation from Flight Data Applying Extended and Unscented Kalman Filter. Aerospace Science and Technology, vol. 14, pp. 106-117, 2010.
• [8] Marco, D. B., Martins, A., Healey, A. J., Surge Motion Parameter Identification for the NPS Phoenix AUV. Naval Postgraduate School, Center for AUV Research, Monterey, CA, 93943-5000, 2005.
• [9] Vandersteen, G., Rolain, Y., Schoukens, J., Pintelon, R., On the Use of System Identification for Accurate Parametric Modeling of Nonlinear Systems Using Noisy Measurements. IEEE Transactions on Instrumentation and Measurement, vol. 45, no. 2, pp. 605-609, 1996.
• [10] Prestero, T., Verification of a Six-Degree of Freedom Simulation Model for the REMUS Autonomous Underwater Vehicle, M.Sc. Thesis in Ocean and Mechanical Engineering, MIT, US, 2001.

• [11] Healey, A. J., Good, M. R., NPS AUVII Autonomous Underwater Vehicle Testbed: Design and Experimental Verification. Naval Engineers Journal, vol. 104, pp. 191-202, 1992.

• [12] Hemminger, D. L., Vertical Plane Obstacle Avoidance and Control of the REMUS AUV Using Forward Look Sonar. M.Sc. Thesis, Naval Post Graduate Scholl, US, 2005.

• [13] Bekir, E., Introduction to Modern Navigation Systems. World Scientific Publishing Co. Pte. Ltd., 2007.

• [14] Guo, Z., Hao, Y., Sun, F., Gao, W., Design of Seamless Changing Method for AUV Integrated Navigation System. First International Workshop on Education Technology and Computer Science, IEEE Computer Society, pp. 160-165, 2009.

• [15] Healey, A. J., Dynamics of Marine Vehicles (MA-4823). Class Notes, Naval Postgraduate School, Monterey, CA, 1995.

• [16] Antonelli, G., Fossen, T. I., Yoerger, D. R., Underwater Robotics. US, pp. 989-996, 2007.

• [17] http://en.wikipedia.org/wiki/Buoyancy.

• [18] Dinc, M., Hajiyev, Ch., Development of Autonomous Underwater Vehicle Navigation System Based on Kalman Filtering Technique. 18th St. Petersburg International Conference on Integrated Navigation Systems (ICINS-2011), St. Petersburg, Russia, pp. 238-242, 2011.

• [19] Mendel, J. M., Lessons in Estimation Theory for Signal Processing, Communications and Control, Prentice Hall PTR, Englewood Cliffs, NJ, US, 1995.

• [20] Jategaonkar, R. V., Flight Vehicle System Identification: A Time Domain Methodology. Reston, Va.: AIAA, 2006.

• [21] Klein, V., Morelli, E. A., Aircraft System Identification Theory and Practice. Reston, VA.: AIAA, 2006.

• [22] Gadzhiev, Ch. M., The Information Provision of Offshore Platform Supervision and Control. Elm, Baku, Azerbaijan, 1996 (in Russian).

• [23] Yalcin, L., Ilgaz, M., Erdem, B., Hydrodynamic Analysis of DARPASUBOFF Model. SubSeaTech, 2007.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 4

ADVANCES IN ENVIRONMENT SENSING AND PERCEPTION TECHNOLOGIES AND ALGORITHMS FOR AUTONOMOUS GROUND VEHICLES

Constantin Ilas,*¹Irina Mocanu¹and Mariana Ilas²

¹Department of Computer Science, University Politehnica of Bucharest, Bucharest, Romania ²Department of Electronics, Telecommunications and IT, University Politehnica of Bucharest, Bucharest, Romania

ABSTRACT

This chapter discusses the main environment sensing and perception technologies and algorithms used in autonomous ground vehicles (AGVs). For the beginning, a review of main environment sensing technologies is presented, focusing mostly on LIDAR, video camera and radar. The chapter shows that in the area of environment sensing technologies, several main trends exist today, including an increased usage of video cameras, as well as introduction of integrated solutions, consisting of two complementing technologies, such as video camera plus LIDAR or video camera plus radar. The objectives and the generic architecture of the perception subsystem are then presented and the most used perception algorithms employed in AGVs are discussed, insisting on classification algorithms. Also, the chapter discusses the most used feature descriptors employed by these algorithms, i.e. HOG and DSIFT. The chapter includes a review of several implementations, as presented in the literature in the past years. Thus, the main trends are revealed: increasing the complexity and the performance of algorithms, reducing the number of sensors and increasing the usage of video camera. The chapter presents an implementation of the SVM algorithm based on video camera images and compares the performance of this method when used with HOG and DSIFT respectively. Using four different classes, the classifier gives information not only on the global car vs. non-car decision, but also on the orientation and relative distance of the detected vehicle. The overall performance of this classifier, in terms of precision and recall rate is

^* E-mail: constantin.ilas@cs.pub.ro.

presented. It confirms the expectation that video camera can play a more important role in the perception sub-systems of AGVs.

Keywords: Autonomous Ground Vehicle, Driverless Car, Environment Sensing Technologies, Image Classification, Support Vector Machine, Dense Scale-Invariant Feature Transform, Histogram of Gradients, Video Camera, LIDAR

NOMENCLATURES

AGV Autonomous Ground Vehicles

AEB Autonomous Emergency Braking Systems DARPA Defense Advanced Research Projects Agency DSIFT Dense Scale-Invariant Feature Transform ESC/ESP Electronic Stability Control/ Program

FOV Field of View

HOG Histogram of Oriented Gradients IMU Inertial Measurement Unit LIDAR Light Detection and Ranging SIFT Scale-Invariant Feature Transform SURF Speeded Up Robust Features

SVM Support Vector Machine US Ultra-Sonic

1.INTRODUCTION

In recent years, there has been a substantial research activity in the area of autonomous ground vehicles (AGVs). Today, several prototypes developed by universities as well as companies exist [1-4]. Many companies are working on such vehicles or related technologies [5, 6], some of them envisioning first commercially available autonomous cars (with limited functionality) in the medium future [5]. In parallel, semi-automated driver assistance systems have started to be introduced in production cars and much more have been announced for the near future [6]. The fact that AGV is a hot topic is confirmed by the studies related to its impact and implications [6, 7, 8] as well as by the fact that discussions on this topic have grown during the past year (mid-2012 to mid-2013) in the US and so has the consumer receptivity [6]. Washington DC and three US states allow the operation of AGVs on public road, for testing purposes [6].

Despite this evolution, the technical challenges on the development of fully autonomous vehicles are very high and sustained research activity is still needed in order to improve the performance and the robustness of these prototypes. The sensing and the perception technologies play a critical role in the overall performance of AGVs, and consequently they are areas of sustained research activity.

Figure 1. Control system architecture of an autonomous robot - high level diagram.

Looking at the history of the evolution of AGVs, we can see that important milestones were the challenges organized by the Defense Advanced Research Projects Agency -DARPA. DARPA initiated and organized some challenges for driverless vehicles. Thus, the first two, organized in2004 and 2005, were held in a desert area, while the third, from 2007 was held in an urban environment [4, 9, 10, 11, 12, 13]. Many prototypes of autonomous vehicles were designed and developed in order to participate in these challenges. After 2007, research has continued to evolve, key focus areas being perception algorithms as well as sensing devices. The main goal of these recent research activities has been to improve the performance and robustness of the perception and sensing stages.

The AGV evolution is in correlation with the evolution of other types of autonomous robots, for different applications. These include not only ground robots, but also robots designed to move and operate in other environments, such as on/ in water or in air [14]. However, despite the fact that all autonomous robots share a similar operating principle of the control system, their specific challenges can be very different [1, 2, 15]. Thus, in the case of AGVs the navigation speed should reach high values (similar to normal cars), while the environment can be unstructured, i.e. complex and rapidly changing, or even extremely unstructured. Normally, the urban traffic is the highest on the levels of unstructured environment. This is mainly because of existence of multiple intersections, pedestrian crossings, traffic signs, as well as of a variety of other traffic participants. These include vehicles which move in different directions, change their lanes or speed, but also vehicles which are parked or enter the traffic, as well as pedestrians who cross the street (not always on the crossing areas), or approach crossings, etc.

As the history of the DARPA Challenges confirms, autonomous ground vehicles have evolved from first prototypes operating in different types of desert areas (i.e. mediumstructured environments) towards complex urban traffic, which is a very unstructured environment.

Figure 1 [15] shows the high-level diagram of AGV (and autonomous robots in general) architecture, with its main subsystems. Thus, the course information on the environment is captured by sensors and transmitted to the perception subsystem. This extracts the useful information, and outputs data that can then be used by planning and control subsystem. Based on this information, as well as on navigation algorithms, this subsystem generates the references used by the actuators control subsystem. For both perception and planning and control subsystems, several algorithms can be used, thus different levels of precision and complexity being possible.

2. MAIN TYPES OF ELECTRONIC SENSING DEVICES FOR ENVIRONMENT PERCEPTION IN AUTONOMOUS GROUND VEHICLES

By sensing devices for environment perception we understand the devices providing information on the vehicle surroundings (such as road/terrain, obstacles, traffic signs, etc.). The data obtained from these sensors is then processed by perception algorithms in order to extract real-time, accurate information, based on which navigation decisions can be made. Since the algorithms as well as the performance and robustness of perception depend on the types and characteristics of the environment perception sensors, it is useful to review the main types of such sensing devices. As first shown in [15], we will analyze both those used in AGV prototypes as well as those currently employed in mass production cars.

2.1. Type of Electronic Sensing Devices for Environment Perception Used in AGV Prototypes

The main types of electronic sensing devices used in AGV prototype cars for environment perception are:

• LIDAR (Light Detection and Ranging) sensors;
• cameras;
• radars;

Other types of sensors which are less frequently used include ultra-sonic (US) sensors and contact sensors. It is worth mentioning that many AGVs use GPS for position detection. GPS accuracy can be increased if data from IMU (Inertial Measurement Unit) devices is used. Also, IMU data can also be used in some applications to adjust the indications from some of environment perception sensors.

LIDAR, radar and US are active sensors, and are all based on the principle of sending a pulse of wave (of certain nature and frequency), receiving it reflected by the obstacle and measuring the total time of flight.

An important parameter of LIDAR sensors is the range in which they can detect (and determine the distance to) objects. The range is heavily influenced by the reflectivity of the object as well as by weather conditions (such as humidity). Therefore, range is sometimes indicated by either its maximum value, or by a value at which poor reflecting objects (typically objects with 10% reflectivity) can be detected. For example, the maximum range can be 80 m, whereas its value for objects with 10% reflectivity is 30 m. LIDAR sensors can be referred to as long-range (typically with maximum range beyond 50m) or short-range.

By definition, a laser beam is highly concentrated, meaning that in order to scan a region of the 3D environment, either the beam periodically changes its angle, or multiple beams are used. In reality both approaches are used simultaneously. Based on the shape of the environment part they scan, LIDARs can be either 2D (or planar), when they scan in a given plane, or 3D, when they cover a 3D region of the surrounding space. 2DLIDARs spin and scan radially (typically in the horizontal plane), covering a sector of a circle, typical values including 80^o or 270^o . This is also referred to as field of view (FOV). 3DLIDARs cover, for each given position in the horizontal plane, a given sector in the vertical plane. A 3DLIDAR scanning sector is indicated by the horizontal (azimuth) coverage (e.g. 270^o , or 360^o ) as well as the vertical (elevation) coverage (e.g. 26.8^o ). To achieve a 3D coverage, several laser beams are positioned so that together they cover the given part of the vertical plane, whereas the entire unit spins to cover the horizontal one. Because typically a laser beam covers less than one degree, a few dozens of laser beams are normally used (e.g. 32 or 64). For each of these beams, separate emitter and receiver units are needed.

Cameras provide digital image of the covered region of space. They are based on an array of passive light sensors. Each sensor determines the value of an image pixel. The image can be either monochrome (grey-scale) or color and the resolution can be in the range of 640x480 to 1328x1048 pixels. Since the image is 2D, distance to particular objects in image can only be determined using complex processing algorithms.

In automotive applications, there are two particular challenges for a camera: the sensitivity to low intensity of light as well as the capability of simultaneously capturing very bright and dark areas (also referred to as dynamic range). The sensitivity to low intensity light is given by the efficiency of the sensor as well as by the electronic amplification of the signal. The higher this gain, the more sensitive is the camera, but the amplification has to be applied only to useful information (not the noise). For night vision, cameras should capture also infrared (IR) radiations. Typically, cameras spectrum cover also near-IR, which is reflected by objects also at night. It is also possible to use an external IR source (IR LED) to illuminate the objects, thus enhancing the images in the near-IR. Far-IR (thermal) cameras, which measure the IR radiation emitted by objects (thermal radiation) are not commonly used in automotive.

Considering that the intensity of the light reflected by an object can vary from 10-1 lux (night) to 105 lux (sunshine), the dynamic range for very good cameras for automotive should be 120 db (standard cameras offer around 60 db). Radar technology uses electromagnetic radiation in several frequency bands (e.g. 24 GHz, 77 GHz). It has the advantage of being less affected by weather conditions, but, in order to achieve a good angular resolution, the emitted signal has to be restricted to a narrow beam. Also, it is possible to employ several other techniques to improve the resolution (such as direction of arrival method, spatial power spectrum measurement, etc.).

Radars are classified as short - sometimes called also mid-range - (up to 100-150m) or long-range (up to 250 m). Obstacles at short distances (a few meters) can be detected only by short-range radars. Their FOV is typically 45^o (for front radars) and the resolution within a few degrees. Given the range (typically above 100m), and the resolution, radars can detect possible obstacles at a high distance, but the resolution normally allows only preliminary actions to be taken.

An important advantage of radars is that they also can determine the relative speed of the detected obstacle, which is very important for decision making in AGVs as well as for applications such as automatic braking or ACC.

Based on the above considerations, as well as by the results in [16], we summarize the main properties of the main technologies for environment perception in Table 1. Camera is a particular case, as by itself it cannot detect objects, but it depends on image processing algorithms to do so. As it can be seen, none of these technologies is ideal and an attractive idea is to complement two or more to obtain best results.

Criteria	LIDAR	Radar	Camera	US
Very short range (0-1m) detection	Poor	Only for short range radar	Ok	Very good
Short range (1-30 m) detection	Very good	Very good	Good	Poor
Long range (30-100+ m) detection	Medium	Very good	Poor	No
Angle < 10	Very good	Good	Good	Poor
Angular resolution	Very good	Good	Good	Poor
Velocity measurement	No	Yes	No	No
Operation in adverse weather conditions	Poor	Very good	Poor	Good
Operation at night	Very good	Very good	Limited	Very good

Table 1. Comparison of main technologies for environment perception

2.2. Types of Electronic Sensing Devices for Environment Perception Currently Used in Mass-Production Cars

When discussing about the sensors for the future AGV, it is interesting to have a brief review of the types of sensors for environment perception used in current mass-production cars. As mentioned before, some mass-productions cars have started to offer advanced driver assistance features, in some cases even limited versions of some of the functions of an AGV. Consequently, these cars have been equipped with several types of advanced sensors for environment perception. Most common are radars and cameras, which are replacing more mature US sensors [17]. More recently, also (short range) LIDARs have been introduced, and also an integrated camera and short range LIDAR device has been announced [18, 19].

These LIDARs currently operate for short distances (up to around 10 m) and can detect other cars (and their speed) in a couple of sectors (e.g. front, front-left, front-right), thus facilitating the decision of braking vs. steering. In the near future, with the announced of the integrated camera and LIDAR, the reliability of the perception will be improved (by comparing the data provided by each of the two sensors).

Multi-layered LIDARs (combining usually 4 different planes of view) seem particularly useful for automotive, allowing for object detection even when, due to humps or holes on the road, a simple 2DLIDAR beam would not reach the obstacle.

Radar deployment is increasing. A challenge for introducing radars on mid-range production cars is the relatively large size of the device, because of the size and number of the antennas and because of the semiconductor technology which offers limited integration capabilities [20]. However, in recent years there has been a great amount of research activities for finding new, more compact and cheaper technologies [20, 21, 22]. Also, many current cars have inertial sensors (in most cases needed for electronic stability control/program - ESC/ESP) [23].

A general trend is integration of two or more sensors into one single device. Most noteworthy are the devices combining a camera and a radar [24], and even a camera and a LIDAR [19]. This is justified by the fact that the two systems complement each other for better perception and by the need of having fewer and more compact devices.

Generally speaking, the evolution of object detection sensors in mass-production cars (both long range and short range) is stimulated by the growing importance of Autonomous Emergency Braking Systems (AEB). Such systems will be included, starting with 2014 in the Euro NCAP rating scheme [25], therefore they will probably be more widely applied to midrange cars also.

As already happening [15, 26] some of the algorithms designed in AGV prototyping for perception and control, may be adapted in the future to be used in various systems for driver assistance. It is also interesting the trend [15] of decreasing the total number of sensors in AGV prototypes. This is possible because of the evolution of the perception algorithms, through a more optimal design of the architecture of sensing and perception subsystems. Based on [26], we will discuss the objective and the architecture of the perception subsystem in AGVs, as well as review the main algorithms used in this subsystem.

3. OBJECTIVES AND ARCHITECTURE OF PERCEPTION SUBSYSTEM

3.1. Objectives of Perception

The perception subsystem [26] has the goal of determining the parameters required by the motion planning and control subsystems. This is done by processing the information provided by various sensors. The output parameters required include the coordinates of the road, the position and speed of obstacles (on the road), the position of pedestrian crossings, the position of lane markers, etc. It seems natural that the outputs of this subsystem would depend on the type of environment for which the vehicle is designed to navigate [26]. Of course, the more unstructured the environment is, the higher is the number and complexity of its outputs (Table 2).

Thus, in highly unstructured environments, such as urban traffic, the perception subsystem is required to detect diverse types of objects and to classify them, but also to track them and to predict their future position. This results in a perception subsystem with a very complexity. Such a classification is useful to understand the perception subsystems used in different prototypes, as well as those designed for particular applications. A fully operational vehicle, designed to navigate in most of these environments, including urban, (as current manned vehicles do) would clearly need to determine all these parameters.

As the review of literature will reveal (see following paragraphs), many papers discuss implementation and results of only a limited portion of the perception subsystem. This is due to the complexity of the algorithms required for perception as well as to the growing maturity of the area, which allow researcher to reuse existing solutions and to focus on improving particular blocks and algorithms. Also, some AGV prototypes are designed for particular applications, and hence they operate only in specific environments such as off-road, or in simpler conditions, such as vehicle follower. Clearly, in these cases, the perception subsystem has reduced requirements.

3.2. Generic Architecture of Perception Subsystem

The perception subsystem consists in general of three main blocks: detection, classification and tracking. Detection is the first step and it deals with separating different objects from the environment data provided by the sensors. For sensors such as cameras or LIDAR, the data is a matrix containing information for each point in the area covered by that sensor.

The number of points depends on the sensor resolution and is in the range of 1M for each image of a camera and 1M per second in the case of 3DLIDARs [27].

Classification refers to identifying the detected object as belonging to a particular class (e.g. car, pedestrian, etc.) whereas tracking deals with determining the change in the position of a given object from one set of data to the other.

Tracking is very useful for determining the speed of the respective object and for predicting its future movement [28]. Apart from detection results, tracking and classification use also data from each other, to enhance their performance. A diagram of the perception subsystem is presented in Figure 2.

Perception output Type of environment Off-road (variable degree of structure) Highway (medium degree of structure) Country roads (low degree of structure) Urban (highly unstructured) Obstacle position Needed Needed Needed Needed Obstacle speed Sometimes needed Needed Needed Needed Road Sometimes needed Needed Needed Needed Lane markers n/a Needed Needed Needed Traffic signs n/a Needed Needed Needed Pedestrian crossing n/a n/a Needed Needed Obstacle classification Not needed Sometimes needed Sometimes needed Critical Obstacle movement prediction Not needed Sometimes needed Sometimes needed Critical

Table 2. Perception subsystem outputs for different type of environments

In many implementations presented in the literature, several particular aspects exist. For instance, in order to limit the total time required for perception, several tasks are done in parallel, on independent subsystems.

This can be the case of lane markers detection, which may be done with dedicated sensors (planar LIDARS or camera) and also traffic signs can have a dedicated subsystem, based on a separate camera [29, 30]. Similarly, radar is used for detecting fast moving cars, especially outside cities.

Figure 2. Architecture of perception system in AGVs.

4. MAIN ALGORITHMS USED FOR PERCEPTION

In this section we will give some details on the main algorithms used in AGV for object detection, classification and tracking. Data used for complex perception algorithms comes, in most AGVs, from camera or 3GLIDARs. In both cases it has the form of a matrix, each point containing a structure of several data. Because of this, the algorithms used for perception are usually similar (at least as principle). In the description of the algorithms, we will use the term of image to describe the matrix output of either cameras or LIDARs.

4.1. Detection Algorithms

The goal of detection is to separate the objects existing into an image frame. Several approaches exist, including background subtraction and segmentation. Background subtraction is based on the idea of eliminating the image background to find existing objects. This is best used for identifying moving objects, when the background is relatively fixed from one frame to the other. In such case, the difference in the image comes from the moving objects. Segmentation separates the regions of the image which have "similar" points. Consequently, a segmentation algorithm uses a certain quantification of similarity. One of the segmentation algorithms is the mean-shift clustering. It starts with a large number of potential cluster centers, spread across the image. At each step, the new centers are moved to the mean of the data contained inside an ellipsoid constructed based on the old centers. The vector defined by the old and the new cluster center is called mean-shift vector. The algorithm is repeated until all centers remain fixed. During this process, several potential cluster centers may merge.

4.2. Tracking Algorithms

Tracking can be formulated as the correspondence of detected objects represented by points across different frames [28]. Several solutions for this problem exist [28], one of the most widely used being the Kalman filter. A Kalman filter starts from the basic structure of a linear observer, but the gain matrix (Kalman gain) is computed as a solution of the Riccati equation, taking into account the noise and its covariance [28]. Kalman filters are designed for state variables having a Gaussian distribution. For other type of distributions, a particle filter may be used [28].

4.3. Classification Algorithms

4.3.1. SVM

Support vector machine (SVM) is a non-probabilistic binary classifier. Using a set of training data, it determines on which of the two possible classes a certain object belongs. Because it is a binary classifier, it has to be run several times in order to classify objects belonging to more than two possible classes. Basically, it starts from a set of training vectors (xi , yi) of dimension L and based on them, it computes i coefficients (by solving a quadratic problem [31, 32]), and the support vectors S (of size Ns) are then given by the (x, y) elements corresponding to i > 0. With these the w and b value can be computed:

$$w = \sum_{i=1}^{L} \alpha_i \ y_i \cdot x_i \tag{1}$$

$$b = \frac{1}{Ns} \sum{j \in S} \left( yj - \sum{k \in S} \infty_k \ y_k x_k \cdot x_j \right)$$ (2)

And finally, for any new input vector x , the classification is given by:

$$y' = \operatorname{sign}(w \cdot x' + b) \tag{3}$$

As presented, this is a linear problem, but SVM can also be used for non-linear ones, by performing the so-called kernel trick [32].

The linear SVM full algorithm [33]:

Notation: The training set consisting of L classified examples {(x1, y1), …, (xL, yL)}, with yⁱ either 1 (meaning vector xⁱ belongs to the specified class, or -1 (if it doesn"t).

Training (obtaining the SVM):

1. Set up and solve the dual optimization problem (using a quadratic programming solver):

maximize $$\sum_{i=1}^{L} \inftyi - \frac{1}{2} \sum{i,j=1}^{L} \infty_i \left( y_i y_j x_i \cdot x_j \right) \infty_j$$ (4)

subject to: $$\infty_i \ge 0$$ (5)

and $$\sum_{i=1}^{L} \infty_i \ y_i = 0 \tag{6}$$

1. Determine the weights:

$$w = \sum_{i=1}^{L} \alpha_i \ y_i x_i \tag{7}$$

• 3. Determine the set of support vectors j x S , by finding the indices j so that 0 ^j . Note that for any support vector, we have: 1 j j y w x b .
• 4. Determine b coefficient by solving the above relation over the entire set of support vectors (get an average value):

$$y_j \left( w \cdot x_j + b \right) = 1 \tag{8}$$

Classification:

For any new x , the classification output y (1 if it belongs to that class, -1 otherwise) is determined, based on the above w and b, as:

$$y' = \operatorname{sign}(w \cdot x' + b) \tag{9}$$

The value w x b is referred to as the classification score and it gives information on the confidence of the estimation, and can also be used to compute the probability of correct estimation, i.e. the posterior probability.

Figure 3. Illustration of the mapping principle used in SVM for non-linear classification problems.

In some situations, the vectors xⁱ cannot be perfectly separated by solving this linear problem. However, it was shown [32] that it is always possible, for a finite set of points, to find a dimension where separation of the point set can be achieved by a hyper-plane. This means that the vectors xⁱ should be mapped with a mapping function (x) to a space of sufficiently high dimension so that they become separable. In this case, the original space of vectors *xi*is called input space, whereas the space of the mapped vectors is called the feature space. In the feature space the linear SVM can be applied, see Figure 3. Thus, in such cases the above algorithm can be applied, by replacing x with (x). The mapping function (x) depends on the actual problem and, it was shown [32] that an adequate selection of this function can also increase the performance of the SVM. This is because if data is not perfectly separable in one dimension, the SVM will still work, but with a higher error rate. For instance, in the case of a handwriting recognition problem, linear SVM has around 8.5% error rate, whereas, with a particular mapping, the error rate is reduced to around 1% [32]. A typical mapping function used in many applications is the polynomial mapping, which for 2 dimension input vectors is:

$$\Phi: R^2 \to R^3, (x{i1} \ x{i2}) \to (z{i1} \ z{i2} \ z{i3}) \Phi(x{i1}, x{i2}) = (x{i1}^2 \ \sqrt{2}x{i1}x{i2} \ x_{i2}^2)$$ (10)

Interestingly, computation error does not increase massively with mapping, because the only operation needed in the feature space is the vector product. This can be verified easily by replacing x with (x). For instance, if we take the last equation (the classification stage):

replacing $$x$$ with $\Phi(x)$ . For instance, if we take the last equation (the classification stage):

$$y' = \operatorname{sign}(w \cdot x' + b) \Rightarrow y' = \operatorname{sign}(w \cdot \Phi(x') + b) = \operatorname{sign}\left(\sum_{i=1}^{L} \infty_i \ y_i \Phi(x_i) \cdot \Phi(x') + b\right)$$ (11)

This operation of computing the vector product in the feature space is also called computation of the kernel function:

$$K(x_1x_2) = \Phi(x_1) \cdot \Phi(x_2) \tag{12}$$

It can be shown [32, 33] that the kernel function can be computed directly, which is less complex than performing the actual vector product in the feature space. For instance, in case of the above polynomial mapping, the kernel function can be directly computed as:

$$K(x_1x_2) = (x_1 \cdot x_2)^2 \tag{13}$$

instead of:

instead of:

$$K(x_1x_2) = \Phi(x_1) \cdot \Phi(x2) = \begin{pmatrix} x{11}^2 & \sqrt{2}x{11}x{12} & x{12}^2 \end{pmatrix} \cdot \begin{pmatrix} x{21}^2 & \sqrt{2}x{21}x{22} & x_{22}^2 \end{pmatrix}$$ (14)

This property of the kernel function gives the name of kernel trick used when referring to the SVM computations in the feature space. Without getting into details [32, 33], we will note that mapping the vectors into a higher dimension increases the likelihood that data can be separated, but does not guarantee this. Hence, it is not surprising to note that finding the mapping function for a practical problem is not always an easy task [32].

Despite the fact that the output of the SVM classifier is binary, a posterior probability can be calculated, starting from the SVM classification score. There are several techniques for this, such as binning, sigmoid fitting or isotonic regression Bayesian framework [34]. Also there are some approaches extending the SVM formulation to handle multi class classification. Several ways are defined in order to extend SVMs [35, 36, 38, 39]:

• One-Versus-All SVMs (one-versus-rest SVMs), convert a k-class problem into k two-class problems [35, 37, 38]: all the training samples are used for training each SVM. For each SVM, i, i ≤ k, samples are labeled as positive or negative: samples from the in the *i*th class are positive, and the other are negative. After that five-fold cross validation is used for parameter selection. During the test phase, the *i*th SVM separates class i from all other classes.
• Pair-Wise SVMs (one-versus-one SVMs), use a SVM for each pair of classes, thus converting a k-class problem into k*(k-1)/2binary (two-class) problems. Each of these SVM classifiers is trained for its two classes. Thus, we can denote sij the SVM classifier which is trained on data from the *i*th class and *j*th class, with i< j. The classifier parameters are automatically selected by a five-fold cross validation. During the testing phase, the sij SVM divides all data into classes i and j. The twoclass classifiers have to be combined using a voting scheme. Possible voting schemes include:
  • Max-Win Majority Voting (based on the Max Wins strategy described in [40]): for an input x, if the SVM classifier sij categorizes x in the *k*th, where k {i, j} class, then the vote of x for the class k increases by one. Once all classifiers have voted the class that the largest voting score is assigned to x. In [38] it is chosen the class with a smaller index as the prediction result when two classes have identical voting scores.
  • Confidence Score Voting: as described in [41], proposes an absolute decision value distance measure indicating the closeness between a sample and the optimal separating hyper-plane. If the absolute value of confidence score is small, the sample is close to the optimal separating hyper-plane; if the confidence score is large, the sample is far away from the optimal separating hyper-plane. Decision values are computed by:

$$D(x) = \sum_{i=1}^{L} \alpha_i y_i K(x_i, x) + b$$

$$\tag{15}$$

If +1 and *-*1 are labels of the two classes, a confidence score is imposed on the decision value: -1 ≤ d ≤ 1, in order to ensure high generalization ability (if d> 1, d is rounded to 1; if d< -1, d is truncated to -1). In case of SVM classifier sij categorizes x into the *k*th class, k {i,

j}, then the vote of x for the class k increases by |d|. Once all SVM classifiers have voted, sample x is categorized to the class that has the largest voting score.

4.3.2. Boosting Algorithms

Boosting framework is a method of combining the classification given by a set of multiple, simple classifiers (weak-classifiers) into a single, accurate one (strong-classifier). Ada Boost, Gentle Boost and Joint Boost are some versions. The weak classifiers are constructed in several ways. For instance, the weak classifiers can be constructed iteratively, using different weighting factors for the training data, factors which are computed based on the error obtained by the previous classifier.

The Adaboost algorithm (Ada stands for Adaptive) has the following steps [42]:

Notation: The training set consisting of L classified examples {(x1, y1), …, (xL, yL)}, with yⁱ either 1 (meaning vector xⁱ belongs to the specified class, or −1 (if it doesn"t). M is the number of weak classifiers used.

Training:

1. Initialize the observation weights

$$w_i^0 = \frac{1}{L} \tag{16}$$

• 2. For *m=*1, …, M , (and for all training vectors, i.e. i=1, …, L):
• Construct/select a (weak) classifier hm(x) using the wⁱ weights, so that it minimizes:

$$\sum_{y_i \neq h_m(x_i)} w_i^{m-1} \tag{17}$$

• Compute the error of this classifier:

$$err{m} = \frac{\sum{i=1}^{L} w{i} I\left(y{i} \neq h{m}\left(x{i}\right)\right)}{\sum{i=1}^{L} w{i}}$$

$$(18)$$

$$I(y_i \neq h_m(x_i)) = \begin{cases} 1 \Leftrightarrow y_i \neq h_m(x_i) \ 0 & \text{otherwise} \end{cases}$$ (19)

• Compute

$$\alpha_m = 0.5 \log \left[ \frac{\left( 1 - err_m \right)}{err_m} \right] \tag{20}$$

• Set new, increased weights for the points where the (weak) classifier gave a false output:

$$w_i^m = w_i^{m-1} \exp(\alpha_m) I(y_i \neq h_m(x_i)) = w_i^{m-1} \sqrt{\frac{1 - err_m}{err_m}} I(y_i \neq h_m(x_i))$$ (21)

1. The final (strong) classifier is obtained as:

$$h(x) = \operatorname{sign}\left(\sum{m} \infty{m} h_{m}(x)\right)$$ (22)

Classification: apply the final classifier for any new vector x":

$$y' = h(x') = \operatorname{sign}\left(\sum{m} \alpha{m} h_{m}(x')\right)$$ (23)

The number of weak classifiers, M, can be set at the beginning of the training stage, or it can be determined during the iterations, e.g. until the classifier error cannot be further minimized.

4.3.3. Bayes Filter

Considering the conjoint probabilities:

P B A -the probability of B if A is true and P A B - the probability of A if B is true, an important Bayes"s theorem gives the relationship between these two probabilities in the form [36]:

$$P(A|B) = \frac{P(A)P(B|A)}{P(B)}$$ (24)

In classification problems, we can consider that A is our hypothesis (prediction) based on an old set of data, and B is a new set of data. Applying Bayes"s theorem we obtain that: the updated probability of the hypothesis A based on the new data B equals the product of the prior probability P(A) and the probability of data under the hypothesis (or likelihood)P(B|A), divided by the probability of the new data P(B). If P(B)cannot be computed, it can be regarded just as a normalizing constant. In other words, the posterior (probability of our hypothesis) is the product between the prior (probability of our hypothesis) and the likelihood (of the new data to exist, considering our hypothesis), divided by a normalizing term.

In classification problems, we will compute in parallel both the probability of object belonging to class A and, respectively to class B. The approach can thus be extended to cover multiple classes, or, if just a binary classification is needed, class B means that object does not belong to class A. The algorithm starts with an initial probability P(A) and P(B), which can be based for instance on the statistical differences of class A and B sizes. Then, at each step a new feature is read and the two probabilities are updated, considering that the probability of the existence of that feature for objects in class A and B respectively, is known. These probabilities are determined during the training stage.

The algorithm steps are [43]:

Notation: The training set consisting of L classified examples {(*x*1,*y*1), …, (xL, yL)}, with yⁱ either 1 (meaning vector xⁱ belongs to a specified class A, or −1 (if it doesn"t belong, or otherwise said it belongs to class B – of "non A" objects). Consider that the elements of all training vectors xi(each having size M) belong to a finite set D, of size N.

Training:

Compute the probability of each of the n elements d^j of set d to appear in a training vector x if the object belongs to class a or class b respectively.

$$P(Dj|A) = \frac{\sum{j=1..N} y_i = I(x_i = Dj)}{\sum{j=1..N}^{M} I(y_i = 1)}$$ (25)

Compute the overall probability of an object to be in class a or class b based on the training set

$$P(A) = \frac{\sum_{y_i = 1} yi}{L} , P(B) = \frac{\sum{y_i = -1} |y_i|}{L}$$ (26)

Classification:

Start with the initial probability of the object being in class A or B. For each new object, described by vector x":

• 1) For each element of x , x , with i j x D : read from the training results the probabilities that the element appears for class A and respectively class B objects: P D A ^j and P D B ^j .
• 2) Compute the updated probabilities (without dividing by the normalizing term, which is independent on the probabilities updates):

2) Compute the updated probabilities (without dividing by the normalizing term, which is pendent on the probabilities updates): $$P(A){updated} = P(A) \prod{j} P(D{j}|A) , P(B){updated} = P(B) \prod{j} P(D{j}|B)$$ (27)

Optimization: instead of computing the above values (which decrease continuously and may become very small), it is better to compute:

$$\ln P(A){updated} = \ln P(A) + \sum{j} \ln P(D_j | A)$$ (28)

3) Classify:

$$y' = \operatorname{sign}\left(\left|\ln P(A){updated}\right| - \left|\ln P(B){updated}\right|\right)$$ (29)

All these algorithms are machine learning algorithms, needing a set of training data. Training data consists of pairs of object features and associated outputs (classified labels). A common parameter for the feature descriptors used by classification algorithms is the histogram of oriented gradients (HOG) [44]. This is an interesting aspect, because several other approaches for feature descriptors exist, such as SIFT and SURF. The major characteristic of such approaches is the fact that the descriptor is computed only for key (interest) points within the image. These are points which allow constructing descriptors which are invariant to scale and rotation and robust to change of illumination [45].

Figure 4. Orientation assignment principle, as applied for SIFT method.

The main benefits of SIFT method, as indicated by the author [45] are: it is scale invariant (in other words different sizes of the object do not make a difference) and rotation invariant (perspectives from different angles do not make a difference). It is also quite robust to affine distortion, noise and changes in illumination. The image descriptor is a so called local descriptor, computed not for the entire image, but only for selected key points in the image. The algorithm consists of two principal parts, key points identification and descriptor computation. The key points identification process consists on several main steps: determine the interest points (points in the image which are not affected by different image scales), outliner rejection to obtain the key points and orientation assignment (selecting the dominant orientation for each key point).

The descriptor computation involves computing the gradients in a 16x16 region around each key point, dividing the region in 16 (4x4) blocks and for each compute a 8-bin histogram. For this, the orientation of the gradients is expressed relative to the key point orientation. The descriptor around each key point will be a vector with a 16x8=128 size.

The major stages in the algorithm for computing the SIFT descriptor are [45]:

• 1) Scale-space extrema detection identifies locations and scales that can be repeatedly assigned under differing views of the same object.

  • 2) Key point localization to reject points with low contrast or poorly localized.

• 3) Orientation assignment to assign a consistent orientation to each key point based on local image properties. The descriptor can be represented relative to this orientation and achieve invariance to image rotation, see Figure 4, reproduced from [45].

• 4) Key point descriptor to compute a descriptor for the local image region that is highly distinctive yet is as invariant as possible to remaining variation.

Experimental results on SIFT descriptors applied to image classification shows that better classification results are often obtained by computing the SIFT descriptor over dense grids in the image domain opposite to sparse interest points (practically skipping the first stages in the algorithm, which is the selection of the key points).A larger set of local image descriptors computed over a dense grid usually provide more information than corresponding descriptors evaluated at a much sparser set of image points. Based on these results [46, 47] proposed the dense SIFT descriptors. DSIFT is usually accompanied by a clustering stage, where the individual SIFT descriptors are reduced to a smaller vocabulary of visual words, which can then be combined with a bag-of-words model or related methods [48, 49].

By contrast, HOG descriptors are computed over the entire image (on a so-called dense grid) and have no orientation alignment. In general, for image classification problems HOG descriptors are preferred, as their performance is higher. The main idea behind this method, as revealed in [44], is that the look and shape of an object can be sufficiently characterized by the local distribution of intensity gradients, even if gradients are calculated exactly. This method is implemented by dividing the image into small windows called and cells. For each cell it is computed one dimensional local gradient histogram. After that local histograms are combined to create a global representation. To solve the problem of lighting and shadows zonal contrast normalization is applied. This is done by accumulating value of a number of such cells (a block) and normalizing all the cells in the block. These blocks typically overlap. Two main types of geometries are:

• 1. R-HOG: blocks are square grids represented by three parameters: the number of cells per block, the number of pixels per cell and the number of channels per cell histogram.
• 2. C-HOG: blocks are circular. It can be described by four parameters: number of angular and radial bins, radius of the center bin and the expansion factor for the radius of additional radial bins.

The main steps for obtaining a HOG descriptor for an image (region) are (typical values are indicated):

• 1) Divide the image in blocks of size 16x16 pixels, blocks with 50% overlapping
• 2) Each block will consist on 2x2 cells (each of 8x8 pixels)
• 3) Compute gradients, on the x and y axis, for each cell
• 4) Compute gradients orientation, i.e. the angle between the resultant gradient and the x axis
• 5) Construct the histograms (9-bin) by grouping the gradients within corresponding bins, given their orientation (a gradient does not fall within a single bin, but is proportionally distributed between adjacent bins, according to the distance between its orientation and the closest centers of two adjacent bins)
• 6) Weight the gradients (with Gaussian function), in order to down-weight the cells near the image edges

7) Construct the global image (region) descriptor, by concatenating all HOGs (for all cells within the image)

For example, considering the case of an image region of 64x128 pixels (this is typically for person representation), there will be 7x15=105 blocks and 105x4=420 cells, thus the feature descriptor for the image will have 420x9=3780 elements.

In the case of images from a LIDAR, it is possible to construct similar descriptors. However, in literature, the number of blocks per image used for obtaining the HOG is different (e.g. 6 or 10).

5. PERCEPTION ALGORITHMS IN AGV-REVIEW OF EXISTING SOLUTIONS

In this section we will review several implementations of perception algorithms for different AGV prototypes [3, 4, 27, 50-62]. This review is based on description of algorithms and their results, as presented in literature between 2008 and 2012. Consequently, it includes both some of the prototypes participating in the DARPA Urban Challenge (2007) and results of research activities carried on after that date. The latter are either full AGV prototypes, or partial implementations, used for testing and validating new perception algorithms.

In Table 3 we summarize the type of detection, tracking and classification algorithms used in several relevant studies, as well as the reported results (where available).

The performance of a classification algorithm is indicated by two main parameters: precision and recall rate. Precision is defined as the ratio of correctly identified objects per category (NCIO) versus the total number of identified objects in that category (NIO):

$$p = NCIO / NIO (30)$$

whereas the classification recall rate, is the ratio of correctly identified objects per category versus the total number of objects in that category (NO):

$$r = NCIO/NO$$ , with $NO = NCIO + NMO$ (31)

where NMO is the number of missed objects (or "false negative").

For example, if at a certain moment there are 10 different objects in the image (6 cars, 3 persons, 1 other) and the perception algorithm identifies 4 cars (3 correctly, 1 which actually is a different object, wrongly identified or "false positive"), the precision will be: 3/4 and the recall rate 3/6. Clearly, in most applications, both precision and recall are important, because both false positive and false negative classifications can generate big issues. It is somehow intuitive that a classifier with low misses (good recall) may have a higher probability of generating false positives (lower precision). Therefore, when assessing the performance of a classifier, a graphic showing the recall rate vs. the number of false positives (or maybe vs. precision) may be shown. Of course, in practice the classifier is tuned according to the point on the graphic corresponding to the best compromise between recall and precision, depending on the application. Generally, the classifiers described in literature have a precision which is higher than recall (number of false positives smaller than number of false negatives).

Because real-time implementation is a critical condition for perception (and control) algorithms in AGV, we also included in results information on the computational time (where available). Also, where available, we included the AGV operation speed, which depends on several factors, including total computational time.

As it can be seen in the table, only few implementations perform detection, tracking and classification of several classes of objects. In some cases only detection and either tracking or classification is performed, whereas in others a simple classification is done, separating for instance obstacles from road. Also, some implementations focused on achieving real-time functionality of the vehicle (trade-off between precision and computational time), whereas others proposed and tested new algorithms which have to be optimized (and possibly simplified, affecting their performance) in order to become real-time. Despite of these differences, it can be seen that Kalman filter (extended or linear) is a common choice for part of the tracking algorithm (being present in all implementations performing tracking), whereas SVM is similarly common for classification (in 3 out of 4 solutions performing classification of several classes of objects). Out of these, 2 use HOG as main way of generating the feature (descriptor) vectors.

Historically, most AGV prototypes used 3DLIDAR data for classification algorithms, but more recent studies present also results achieved using images from cameras.

Talos [4], a prototype competing in the DARPA Urban Challenge Event (2007) had a perception system consisting of the following blocks: lane detection (based on camera and digital map), fast vehicle detector (based on radar), obstacle detector (based on LIDAR) and hazard detector (based on LIDAR), the latter being used to detect bumps and holes in the road, implemented on several CPUs [4]. The lane detection is achieved using matched filter to determine the lane direction and random sample consensus algorithm (RANSAC) to select the lane centerline. Obstacle detection and tracking combined data from 12 planar and 1 3DLIDARs. First, the ground plane is identified starting from points that are on the same plane, which are then fit into a non-parametric ground model, based on maximum acceptable road slopes, to exclude false candidates (e.g. car roofs). Based on knowing the ground plane, obstacle points are detected considering areas above ground plane which generate a certain minimum numbers of returns from lasers.

A clustering algorithm groups data from neighboring obstacle points, to generate an obstacle. For tracking, obstacles from new data frames are associated with previous ones based on a voting scheme, which considers the region of overlapping between the new and old data frame obstacles. Speed is estimated using a Kalman filter, thus separating between stationary, slow and fast moving obstacles. As it can be seen, this approach is based on relatively simple, but robust algorithms which allows for the perception and control system to be implemented in real-time (max. vehicle speed was 40 km/h).

In [27] it is described a more complex perception subsystem, focusing mainly on classification of several types of objects. The algorithms were tested on Stanford Track Collection database, and operate in real-time, considering just objects on the road and an 8 core computer. Detection is done by segmentation, using the connected components algorithm and clustering using flood fill algorithm. Tracking uses a Kalman filter and data association is achieved based on distance. Classification is based on boosting method, applied twice, for two types of classifiers: segment and holistic (track) classifiers.

Table 3. Summary of implementation details of perception algorithms for AGVs

Implementation	Detection	Tracking	Classification	Results
A. Teichman,	Clustering (no	Tracking of	Boosting + Discrete	Augmented DBF
J. Levinson,	details)	clusters	Bayes Filter (DBF)	best
S. Thrun, [27]		centroids, using	variants (augmented,	97.5% precision
2011		linear Kalman	normalized, naïve)	(overall)
		filter (no	Using 29	Real time (for
LIDAR sensors		details)	dimensional segment	road objects only)
			descriptors
M.	Occupancy grid	Kalman filter	Support Vector	Used in vehicle
Himmelsbach,	Object hypothesis		Machine (SVM)	tracking
A. Muller,	for segmentation		Using 28	Max. error (for
T. Luttel,			dimensional feature	tracked vehicle):
H.J. Wunsche			vector (incl. HOG)	1m
[50]				Low speed,
2008				distance 15-20 m
				from tracked
LIDAR sensors				vehicle
J. Leonard,	Clustering of	Voting scheme	Only for road and	96.7% success on
J. How, S.	chunks by	(based on	obstacle	tracking (all
Teller, et al. [4]	proximity into	overlapping	identification	existing objects)
2008	groups	groups)	Identify candidate	Velocity
		Velocity	points (within a	estimation:
		estimation using	plane)	- good for objects
		Kalman filter	Fit non-parametric	with high speed
			road model to	- noisy for
			exclude false	objects with low
LIDAR sensors			candidates	speed
			Detect obstacles	Real-time. Max
			based on height from	speed 40 km/h.
			the ground plane.
J. Choi, J. Lee,	Camera		Used only for road/
D. Kim,	data(dedicated		obstacle
G. Soprani,	system) for:		Based on z vs. x,y
P. Cerri [51]	- pedestrian		coordinates variation
2012	crossing via IPM transform and		Fit model Risk integration
LIDAR sensors	Sobel filter
	- speed bumps via
	color
N. Suganuma,	Occupancy grid	Extended	-	Robust to
T. Uozumi [3]	maps for detection	Kalman filter		weather
2011	Clustering of	(to estimate		conditions
	dynamic objects	motion)		Real time. AGV
	Lane marking:			speed= 27.5 km/h
	dedicated system
LIDAR sensors	(separate sensor)

Implementation	Detection	Tracking	Classification	Results
D.Z. Wang,	Graph-based		SVM Classifier:	Precision > 95%
I. Posner,	clustering using		separates foreground	Non real time:
P. Newman [53]	Euclidean		data in 3 classes (car,	clustering 3.3 s/
2011	Minimum		pedestrian, bicyclist)	frame
	Spanning Tree		Using 120
	(EMST)		dimensional feature
	Separate		vector
LIDAR sensors	foreground (all
	classes of interest)
	from background
	(others)
A. Ess,	Together with	Hypothesis-and	SVM using HOG as	Recall rate: 60-
K. Schindler,	classification	verify approach	input features; run	70%
B. Leibe,	algorithm (output	Dynamic	independent	Prediction
L. Van Gool	of classifiers	motion using	classifiers for each	precision: 70-
[52]	indicating	extended	class (car,	80% (next 2
2010	probability of an	Kalman filter	pedestrian)	frames),
	image region to be	Trajectory	Ground plane and	60-70% (next 10
	a pedestrian/ car)	generation in	object estimation	frames) – next 1 s
		parallel for all	using Bayesian	Non real time
Video Camera		detected objects	network	(HOG
				computation 6s/
				frame)

Table 3. (Continued)

The segment classifiers operate on segment images, i.e. images of a detected object at a particular moment in time. The holistic (or track) classifiers operate with a set of images of a particular object, tracked during a time interval, in other words a succession in time of segments corresponding to the same object.

The boosting method used is a version of Gentle Boost and Joint Boost, creating both a strong segment classifier, and a strong holistic (track) classifier respectively. Each weak classifier (used to create the two strong classifiers) is based on a particular descriptor of a segment or respectively a track. The descriptors used by track classifiers include maximum and mean speed and acceleration of the object (computed for the entire track duration, relative to the track centroid) while segment classifiers descriptors include HOG and spin images representations. Spin images is a technique used particularly in the case of dense collections of 3D points, such as those generated by 3DLIDARS, to represent and compare a particular objects with a set of rotated images of the reference object [51]. Of course, for efficiency purpose, compressed representations of spin images are used [51].

In the final stage, both strong classifiers are combined into a single, final one, using different versions of discrete Bayes filters. In these versions, the log of odds for the segment and the holistic classifier are taken into account. As the paper shows, the best results are obtained for the augmented discrete Bayes filter (ADBF). The precision is better if classification is done per track (not for each segment, i.e. not for the object at any given moment, but for the tracked object). In this case, the overall precision considering all three classes of objects (car, pedestrian, and bicyclist) is 98.5% for the two combined classifiers and 97.6% if only the segment classifier is used. If on the other hand, the classification is done for each segment, employing just the segment classifier gives a 93.0% precision. In order to use the holistic classifier, holistic data have to be approximated based only the information on that segment. In this case, there is still an important improvement, the overall precision being 97.5%. [27] Consequently, an important conclusion is that holistic data can improve the overall precision, even (especially!) for classification done for each segment.

In [52] a complex system, performing detection, classification as well as tracking is described. The system uses video image processing, with data achieved from a forwardlooking, stereo camera pair. The main focus is object tracking and path prediction, based on the output of pedestrian and car classifiers. The dynamic models used are constant-velocity one for pedestrians and Ackermann model for car motion.

The object classification is performed using SVM and HOG descriptors. Both pedestrian and cars categories are classified. In order to cope with the variance of a car image depending on the viewing angle, seven separate classifiers are used, one for each relevant angle (e.g. back, lateral, front-profile, back-profile – where last three are doubled, as per left/ right side). The output of each classifier is the probability of a true/ false match, rather than a final decision. There is no clear indication on the method used to convert SVM score in a posterior probability. Using these probabilities, as well as the stereo depth, the authors use a Bayesian framework to jointly estimate the objects as well as the ground plane (assuming all objects reside on a common ground plane). This joint estimation of objects and ground plane has the advantage, according to the authors, that improves the results. For example, in a scene crowded with objects, ground plane is difficult to be estimated alone, whereas in scenes with few objects, ground plane estimation helps in objects detection.

Tracking is done considering separate movement models for pedestrians and respectively cars. The state vector for the pedestrian movement model at moment *t*is T t t t t x y v , with x and y the coordinates, the pedestrian orientation angle and v its speed, expressed as a state element in order to estimate it. The movement model for cars is based on the Ackermann model, thus the state vector is T t t t t t t x y v a , where the first four components have the same meaning as in the pedestrian model, and is the car steering angle, whereas a is its acceleration. Both models are nonlinear and thus Kalman filter cannot be directly used, Extended Kalman filter EKF being the algorithm used by the authors.

Classification recall rate is in the order of 60-70%, whereas prediction precision is around 70-80% for short prediction horizon (4 frames) and 60-70% for longer (10 frames). The overall algorithm, implement on a graphical processor (GPU) is not real time, the bottleneck being the HOG computation, but authors indicate that this time can be reduced, if parallel implementation of HOG computation is used.

As it can be seen also from the data summarized in Table 3, typically the implementations increased the complexity of the algorithms over the past years, with the goal of classifying with good precision objects of at least several classes. The increased complexity makes real-time implementation a challenge, even when using multiple processors. In fact, in many of the reviewed papers, the research focused mostly on algorithm precision improvement, while implementation and complexity optimization is left as a future work.

6. EXPERIMENTAL RESULTS

For experimental results we created a database with cars viewed from different angles: rear view (near and far), left and right view. Each of these views is considered a separate class. The database was created staring from Vehicle Image Database [63], which integrates images from Caltech Database [64, 65] and also from TU Graz-02 Database [66, 67]. The database was completed with images from Traffic Signs Dataset [68]. These images are first segmented in order to identify possible component objects. After segmentation, the component objects are validated by the user and new images are created by cutting off each object using a surrounding space. For each of the four classes we have 800 images, and each image contains one car.

We evaluated image classification using DSIFT and HOG feature descriptors using the basic measures for evaluating classification methods: precision, recall as well as the confusion matrix for the test set. Precision and recall rates are computed based on the definitions (30) and (31), in section 5. Since for AGV navigation it is essential not to miss existing cars, we will mostly refer to this parameter.

The confusion matrix displays the number of correct and incorrect predictions made by the model compared with the actual classifications in the test data. The matrix is n-by-n, where n is the number of classes. It contains information about actual and predicted classifications done by the classification system. Of course, ideally this matrix should have all elements outside the main diagonal equal to zero.

Table 4. Optimal number of training images for SVM with DSIFT, for each class

Cars	Number of training images
Left Cars	115 - 120
Right Cars	115 - 120
Middle Cars	90
Far Cars	95

Table 5. Optimal number of training images for SVM with HOG, for each class

Cars	Number of training images
Left Cars	110
Right Cars	95
Middle Cars	90
Far Cars	95

The database is manually divided in four classes: Left Cars, Right Cars, Middle Cars and Far Cars (we use the defined classes in the Vehicle Image Database [63]). In this case we train a SVM for each class (Left, Right, Middle and Far Cars classes). Each SVM was trained with different numbers of training images in order to obtain the best testing accuracy for each class separately. The recall of each SVM based on the number of training images is presented in graph from Figure5 and Figure 6, for DSIFT and HOG methods.

The numbers of training images which leads to best performance for each class - for DSIFT and respectively HOG descriptors are shown in Tables 4 and 5. For the classification rates, we use one-versus-all SVM in order to compute the best number of training images.

The overall recall rates are computed using three types of decisions: one-versus-all SVM, pair-wise SVM with majority voting and pair-wise SVM with confidence score voting.

Figure 5. SVM recall rate for different number of training images for DSIFT method.

Figure 6. SVM recall rate for different number of training images for HOG method.

The recall results for both methods and all four classes are shown in Table 6. It can be observed that both pair-wise SVM methods outperform the one-versus all SVM method. The confidence score voting method performs better than the majority voting method. Based on the overall recall, the class Far class has the lowest performance, for all three types of SVMs. This is of course understandable and is not essential, since it does not impact immediate decisions of the control algorithm.

Table 6. Summary of image classification recall tests using DSIFT and respectively
HOG

| | Recall | | | | | | | |--------|--------------------|-----|--------------------------------|--------|--------------------------------------------------|-----|--| | Cars | One versus all SVM | | Pair-wise SVM
with majority | voting | Pair-wise SVM
with confidence score
voting | | | | | DSIFT | HOG | DSIFT | HOG | DSIFT | HOG | | | Far | 69% | 71% | 72% | 74% | 73% | 75% | | | Left | 83% | 85% | 84% | 86% | 86% | 89% | | | Right | 81% | 84% | 82% | 85% | 85% | 88% | | | Middle | 75% | 79% | 78% | 81% | 80% | 73% | |

Figure 7. Precision/Recall curve for tests on entire image database using DSIFT feature descriptor.

The confusion matrix on the set test of four classes using DSIFT with pair-wise SVM with confidence score voting is presented in Tables 7 and 8, for DSIFT and HOG methods. The tables contain the classification accuracy for the individual categories, too. The most difficult categories are Far Cars and Middle Cars for DSIFT method, respectively Far Cars for HOG method.

Precision and recall curves are given in Figure 7 and Figure 8 for DSIFT and respectively HOG method, for each class from image database, using one-versus-all SVM.

Figure 9 shows precision and recall curve for Left Cars class (the class with the highest overall accuracy) using both methods: DSIFT and HOG. It can be observed that HOG feature description method outperforms DSIFT method in terms of precision and recall curve.

Figure 8. Precision/Recall curve for tests on entire image database using HOG feature descriptor.

performance achieved for this class.

We can see that HOG method performs generally better, on all considered image classes. Figure 10 shows examples of cars correctly categorized by the SVM with HOG descriptor, for each of the four classes. As it can be seen, for Far Cars, not only the cars are smaller, but the quality of the image is also poorer, which also contributes to the lower

| | | Predicted values | | | | | | | |-------------|----------|------------------|-----|-----|--|--|--|--| | | Far Cars | Middle Cars | | | | | | | | Far Cars | 73 | 4 | 3 | 11 | | | | | | Left Cars | 9 | 86 | 9 | 3 | | | | | | Right Cars | 8 | 8 | 85 | 6 | | | | | | Middle Cars | 10 | 2 | 3 | 80 | | | | | | Accuracy | 73% | 86% | 85% | 80% | | | | |

Table 7. Confusion matrix for DSIFT feature descriptor

Table 8. Confusion matrix for HOG feature descriptor

| | Predicted values | | | | | | |-------------|------------------|-----------|-------------|-----|--|--| | | Far Cars | Left Cars | Middle Cars | | | | | Far Cars | 75 | 2 | 4 | 9 | | | | Left Cars | 10 | 89 | 4 | 3 | | | | Right Cars | 7 | 6 | 88 | 5 | | | | Middle Cars | 8 | 3 | 4 | 83 | | | | Accuracy | 75% | 89% | 88% | 83% | | |

Figure 9. Precision/Recall curve for tests on Left Cars using DSIFT and respectively HOG feature descriptor.

Figure10. Example of correct images classified with HOG feature and pair-wise SVM with confidence score voting.

Figure 11. Example of correctly detected cars from images using HOG method for classification (black contour for each correctly detected car).

Figure 12. Example of wrongly detected cars from images (white contour for falsely reported cars).

After classification, car detection is performed on test images, in order to separate the object in that class. A detection result is considered true or false, according to the overlap ratio between the predicted bounding box B^p and the ground truth (the actual) bounding box Bgt.

$$\frac{area\left(B{p} \cap B{gt}\right)}{area\left(B{p} \cup B{gt}\right)} > 50\%$$

Figure 11 shows a set of correct car detection based on the HOG method. For a true detection we used the above equation. We obtained an overall detection rate of over 90%.

Also, in Figure 12 we present a couple of examples of wrongly detected cars. As it can be seen, objects which appear small in the image, or portions of other objects were wrongly classified as cars.

CONCLUSION

The review of existing sensing technologies and perception methods used in AGVs revealed some important trends, including the more emphasis given recently to video camera for vehicle detection and classification. Therefore, we investigated in more detail the object classification based on camera images. For this, we implemented an SVM classifier and we compared two types of feature detectors: DSIFT and HOG. We used four distinct classes of cars, depending on their orientation and relative distance to the camera and present the results for the classification in each of these classes. Obtaining the information on detected cars orientation and distance is helpful for the AGV control. Of course, the classification information can also be aggregated, in a car vs. non-car decision. In general, the HOG feature descriptors lead to a higher performance than DSIFT. The achieved results confirm the expectation that video cameras can play a much higher role in perception sub-systems for future AGVs.

ACKNOWLEDGMENTS

In this chapter we used several existing image databases: mainly the Vehicle Image Database [63], which integrates images from Caltech Database [64, 65] and also from TU Graz-02 Database [66, 67]. The database was completed with images from Traffic Signs Dataset [68].

REFERENCES

• [1] Folsom, T.C.; Energy and Autonomous Urban Land Vehicles. IEEE Technology and Society Magazine, pp. 28-38, 2012.
• [2] Luettel, T.; Himmelshach, M.; Wuensche, H.J.; Autonomous Ground Vehicles Concepts and a Path to the Future. Proceedings of the IEEE, Centennial Issue, 2012.
• [3] Suganuma, N.; Uozumi, T.; Development of an Autonomous Vehicle -System Overview of Test Ride Vehicle in the Tokyo Motor Show 2011. 2012 SICE Annual Conference, Japan, pp. 215-218, 2012.

• [4] Leonard, J.; How, J.; Teller, S.; et al.; A Perception-Driven Autonomous Urban Vehicle. Journal of Field Robotics, Special Issue: Special Issue on the 2007 DARPA Urban Challenge, Part III, vol. 25, no. 10,pp. 727-774, 2008.

• [5] Bigman, D.; Driverless Cars Coming to Showrooms by 2020, Says Nissan CEO Carlos Ghosn. Forbes,www.forbes.com/sites/danbigman/2013/01/14/driverless-cars-comingto-showrooms-by-2020-says-nissan-ceo-carlos-ghosn/.

• [6] KPMG, LLP; Self-Driving Cars: Are We Ready. Report, www.kpmg.com/US/en/Issues AndInsights/ArticlesPublications/Documents/self-driving-cars-are-we-ready.pdf.

• [7] Beiker, S.A.; Legal Aspects of Autonomous Driving. Santa Clara Law Review 1145, vol. 52, no. 4,pp. 1145-1156, 2012.

• [8] KPMG,LLP; CAR (Center for Automotive Research); Self-Driving Cars: The Next Revolution.Report,www.kpmg.com/US/en/IssuesAndInsights/ArticlesPublications/Doc uments/self-driving-cars-next-revolution.pdf.

• [9] Dominguez, R.; Onieva, E.; et al.; LIDAR Based Perception Solution for Autonomous Vehicles.11th International Conference on Intelligent Systems Design and Applications, Spain, pp. 790-795, 2011.

• [10] Fernandez, C.; Gavilan, M.; et al.; Free Space and Speed Humps Detection Using LIDAR and Vision for Urban Autonomous Navigation. Intelligent Vehicles Symposium, Spain, pp. 698-703, 2012.

• [11] Ozguner, U.; Stiller, C.; Redmill, K.; Systems for Safety and Autonomous Behavior in Cars: the DARPA Grand Challenge Experience. Proceedings of IEEE, vol. 95, no. 2, pp. 397-412, 2007.

• [12] Rouff, C.; Hinchey, M.; Experience from the DARPA urban challenge. Springer-Verlag London Ltd., 2012.

• [13] Cremean, L.B.; Foote, T.B.; et al.; Alice: An Information-Rich Autonomous Vehicle for High-Speed Desert Navigation. Journal of Field Robotics, Special Issue: Special Issue on the DARPA Grand Challenge, Part 2, vol. 23, no. 9, pp. 777-810, 2006.

• [14] Guizzo, E.; Deyle, T.; Robotics Trends for 2012. IEEE Robotics and Automation Magazine, pp. 119-123, 2012.

• [15] Ilas, C.; Electronic Sensing Technologies for Autonomous Ground Vehicles: A Review.8th International Symposium on Advanced Topics in Electrical Engineering (ATEE 2013), Romania, pp. 1-6, 2013.

• [16] Jenkins, A.; Remote Sensing Technology for Automotive Safety. Microwave Journal, Dec. 2007, www.microwavejournal.com/ articles/5689-remote-sensing-technology-forautomotive-safety.

• [17] Conner, M.; Automobile Sensors May Usher in Self-Driving Cars.EDN Magazine, May 26, 2011. www.edn.com/design /automotive/4368069 /Automobile-sensors-may-usherin-self-driving-cars.

• [18] Continental, A.G.; SRL 1 and SRL1C Infrared Short Range LIDAR Sensor Datasheet. www.conti-online.com/generator/www/ de/en/continental/industrial_sensors/themes/download/srl1_datasheet_en.pdf.

• [19] Continental, A.G.; SRL-CAM400CMOSCamera and Infrared LIDAR in a Compact Unit. PressRelease, www.conti-online.com/generator/www/com/en/continental/press portal/themes/press_releases/3_automotive_group/chassis_safety/press_releases/pr_201 2_10_17_srl_cam_en.html.

• [20] Stevenson, R.; Long-Distance Car Radar. IEEE Spectrum, Oct. 2011, http://spectrum.ieee.org/green-tech/advanced-cars/longdistance-car-radar.

• [21] Fleming, B.; Recent Advancement in Automotive Radar Systems. IEEE Vehicular Technology Magazine, vol. 7, no. 1, pp. 4-9, 2012.

• [22] Bosch; LRR3- Third Generation Long-Range Radar Sensor Datasheet 2009. www.bosch-automotivetechnology.cz/media/db_application/downloads/pdf/safety_ 1/en_4/lrr3_datenblatt_de_2009.pdf.

• [23] Bosch; Combined Inertial Sensor for Vehicle Dynamics Control SMI650 Datasheet. www.bosch-semiconductors.de/media/pdf_1/einzeldownloads/vehicledynamics_ systems/datenblatt_smi650.pdf.

• [24] Delphi Inc.; Integrated Radar and Camera System. RACam Product Info, http://delphi.com/shared/pdf/ppd/safesec/racam.pdf.

• [25] Euro NCAP; Autonomous Emergency Braking -AEB information. www.euroncap.com/Content-Web-Page/2788ff8c-788d-468e-996b-7405569412/autonomous-emergency-braking---aeb.aspx.

• [26] Ilas, C.; Perception in Autonomous Ground Vehicles A Review. Electronics, Computers and Artificial Intelligence (ECAI) International Conference, Pitesti, Romania, pp. 1-6, 2013.

• [27] Teichman, A.; Levinson, J.; Thrun, S.; Towards 3DObject Recognition via Classification of Arbitrary Object Tracks. IEEE International Conference on Robotics and Automation (ICRA 2011), pp. 4034-4041, 2011.

• [28] Yilmaz, A.; Javed, O.; Shah, M., Object Tracking: A Survey. ACM Computing Surveys, Article 13, vol. 38, no. 4, 2006.

• [29] Moegelmose, A.; Manubhai Trivedi, M.; Moeslund, T.B.; Vision-Based Traffic Sign Detection and Analysis for Intelligent Driver Assistance Systems: Perspectives and Survey. IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 4,pp. 1484-1497, 2012.

• [30] Ruta, A.; Yongmin, L.; Xiaohui, L.; Real-Time Traffic Sign Recognition from Video by Class Specific Discriminative Features. Journal of Pattern Recognition, vol. 43, pp. 416-430, 2010.

• [31] Burges,C.J.C.; A Tutorial on Support Vector Machines for Pattern Recognition. Data Mining and Knowledge Discovery, vol. 2, issue 2, pp. 121-167, 1998.

• [32] Weston, J.; Support Vector Machines (and Statistical Learning Theory) Tutorial. www.cs.columbia.edu/~kathy/cs4701/documents/jason_svm_tutorial.pdf

• [33] Forsyth, D.A.; Ponce, J.; Computer Vision: A Modern Approach. Prentice Hall, 2011.

• [34] Zadrozny, B.; Elkan, C.; Transforming Classifier Scores into Accurate Multiclass Probability Estimates. 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM 2012),pp. 694-699, 2002.

• [35] Abe, S.; Support Vector Machines for Pattern Classification. Springer, New York, 2005.

• [36] Schlkopf, B.; Smola, A.J.; Learning with Kernels: Support Vector Machines, Regularization, Optimization and Beyond. 1stEd., Cambridge, The MIT Press, MA, 2001.

• [37] Vapnik,V.N.; The Nature of Statistical Learning Theory. 2nd Ed., Springer, New York, 2000.

• [38] Hsu, C.W., Lin, C.J., A Comparison of Methods for Multiclass Support Vector Machines. IEEE Transactions on Neural Networks, vol. 13, no. 2, pp. 415-425, 2002.

• [39] Wang, L.; Fu, X.; Data Mining with Computational Intelligence. Springer, Berlin, 2005.

• [40] Friedman, J.H.; Another Approach to Polychotomous Classification. 1996, http://old.cba.ua.edu/~mhardin/poly.pdf.

• [41] Phung, W.; Naghdy, G.; A Multi-Class Image Classification System Using Salient Features and Support Vector Machines. International Conference on Sensors, Sensor Networks and Information Processing (ISSNIP 2007), Melbourne, Australia, pp. 431- 436, 2007.

• [42] Rojas, R.; AdaBoost and the Super Bowl of Classiers A Tutorial Introduction to Adaptive Boosting. Freie University of Berlin, 2009, www.inf.fu-berlin.de/inst/agki/adaboost4.pdf.

• [43] Downey, A.B.; Think Bayes Bayesian Statistics Made Simple. E-book, Green Tea Press, 2012, www.greenteapress.com/thinkbayes/.

• [44] Dalal, N.; Triggs, B.; Histograms of Oriented Gradients for Human Detection. International Conference on Computer Vision & Pattern Recognition (CVPR'05).vol. 1,pp. 886-893, 2005.

• [45] Lowe, D.G.; Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, vol. 60, no. 2,pp. 91-110, 2004.

• [46] Bosch, A.; Zisserman, A.; Munoz, X.; Image Classification Using Random Forests and Ferns. 11th International Conference on Computer Vision (ICCV'07), Rio de Janeiro, Brazil, pp. 1-8, 2007.

• [47] Bosch, A.; Zisserman, A.; Munoz, X.; Scene Classification via pLSA. 9th European Conference on Computer Vision (ECCV'06), pp. 517- 530, 2006.

• [48] Csurka, G.; Dance, C.; Fan, L.; Willamowski, J.; Bray, C.; Visual Categorization with Bags of Keypoints. International Workshop on Statistical Learning in Computer Vision (ECCV'04), Prague, Czech, pp. 1-22, 2004.

• [49] Lazebnik, S.; Schmid, C.; Ponce, J.; Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories. IEEE Conference on Computer Vision & Pattern Recognition (CVPR'06), New York, USA,pp. 2169-2178, 2006.

• [50] Himmelsbach, M.; Muller, A.; Luttel, T.; Wunsche, H.J.; LIDAR-Based 3D Object Perception. 1st International Workshop on Cognition for Technical Systems, 2008.

• [51] Choi, J.; Lee, J.; Kim, D.; Soprani, G.; Cerri, P.; Environment Detection and Mapping Algorithm for Autonomous Driving in Rural or Off-Road Environment. IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 2, pp. 974-982, 2012.

• [52] Ess, A.; Schindler, K.; Leibe, B.; Van Gool, L.; Object Detection and Tracking for Autonomous Navigation in Dynamic Environments. The International Journal of Robotics Research, vol. 29, no. 14,pp. 1707-1725, 2010.

• [53] Wang,D.Z.; Posner, I.; Newman, P.; What Could Move? Finding Cars, Pedestrians and Bicyclists in 3DLaser Data. IEEE International Conference on Robotics and Automation (ICRA'2012),pp. 4038-4044, 2012.

• [54] Chakraborty, N.; Patel, R.S.; Intelligent Agents and Autonomous Cars: A Case Study. International Journal of Engineering Research & Technology (IJERT), vol. 2, no. 1,pp. 1-7, 2013.

• [55] Esteves, B.A.; Megda, P.T.; Becker, M.; Autonomous Embedded Navigation System-Detection, Classification and Trajectory Prediction of Obstacles in Urban Environments Based on Laser Sensor Data and KalmanFilter. ABCM Symposium Series in Mechatronics, vol. 5, pp. 1032-1041, 2012.

• [56] Chen, L.; Li, Q.; Li, M.; Zhang, L.; Mao, Q.; Design of a Multi-Sensor Cooperation Travel Environment Perception System for Autonomous Vehicle. Open Access Sensors, vol. 12, pp. 12386-12404, 2012.

• [57] Dominguez, R.; Onieva, E.; et al.; LIDAR Based Perception Solution for Autonomous Vehicles. 11th International Conference on Intelligent Systems Design and Applications, Spain, pp. 790-795, 2011.

• [58] Fernandez, C.; Gavilan, M; et al.; Free Space and Speed Humps Detection Using LIDAR and Vision for Urban Autonomous Navigation. Intelligent Vehicles Symposium, Spain, pp. 698-703, 2012.

• [59] Urmson, C.; Baker, C.; et al.; Autonomous Driving in Traffic: Boss and the Urban Challenge.Artificial Intelligence (AI) Magazine, vol. 30, no. 2, pp. 17-29, 2012.

• [60] Luber, M.; Arras,K.O.; Plagemann,C.; Burgard, W.; Classifying Dynamic Objects: An Unsupervised Learning Approach. Autonomous Robots, vol. 26. No. 2-3, pp. 141-151, 2009.

• [61] Spinello, L.; Triebel, R.; Siegwar, R.; Multiclass Multimodal Detection and Tracking in Urban Environments. The International Journal of Robotics Research, vol. 29, no. 12,pp. 1498-1515, 2010.

• [62] Johnson, A.; Hebert, M.; Using Spin Images for Efficient Object Recognition in Cluttered 3DScenes. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 21, no. 5, pp. 433-449, 1999.

• [63] Arrospide, J.; Salgado, L.; Nieto, M.; Video Analysis Based Vehicle Detection and Tracking Using an MCMCSampling Framework. EURASIP Journal on Advances in Signal Processing, Article ID 2012, vol. 2, 2012.

• [64] The Caltech Database. Computational Vision at California Institute of Technology, Pasadena, www.vision.caltech.edu/html-files/archive.html.

• [65] Fergus, R; Perona, P.; Zisserman, A; Object Class Recognition by Unsupervised Scale-Invariant Learning. IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Madison, Wisconsin, USA,pp. 264-271, 2003.

• [66] The TU Graz-02 Database. Graz University of Technology.www.emt.tugraz.at /~pinz/data/GRAZ_02.

• [67] Opelt, A.; Pinz, A.; Object Localization with Boosting and Weak Supervision for Generic Object Recognition. Image Analysis Lecture Notes in Computer Science, vol. 3540, pp. 862-871, 2005.

• [68] Larsson, F.; Felsberg, M.; Using Fourier Descriptors and Spatial Models for Traffic Sign Recognition. 17th Scandinavian Conference on Image Analysis (SCIA'2011), LNCS 6688, pp. 238-249,www.cvl.isy.liu.se/research/traffic-signs-dataset.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 5

INTELLIGENT HIGHWAY SURVEILLANCE AND SAFETY SYSTEMS

Valentin Alexandru Stan , Radu Serban Timnea and Razvan Andrei Gheorghiu†

"Politehnica" University of Bucharest, Bucharest, Romania

ABSTRACT

This chapter presents the concept and the implementation solution for the integrated highway traffic surveillance system with complex safety integrated systems (speed radar detectors, vehicle profile automatic identifier, plate recognition system and weight in motion measurement). This concept has already been implemented for one of the busiest routes in Europe and the results are very good, increasing safety and security on highway.

The continuously increase of traffic value on roads and highways requires an improvement of public and vehicles safety. That is possible by increasing the Police force and Highway administrator system, the best solution being to implement electronic systems for traffic surveillance, speed radar, incident detector and informing tables on the highway.

Modern highway surveillance is not just a local circuit television system, in fact it realizes all the functions which would be done by field personnel (generally police officers): traffic surveillance, traffic re-routing, emergency services announcement, driver informing on traffic conditions or restrictions. Moreover, the power computing system can automatically optimize traffic based on real field conditions and generate alarms in case of emergency.

An Intelligent Highway Surveillance System combines Control Centre, Communication and Operating technology to provide the customer with a traffic management, control, and resolution plan. This concept provides real time active traffic management according to the main elements in the field and improves control, flow and safe on the route.

Email: valentin.stan@smart-tc.ro.

Email: radu.timnea@upb.ro.

Email: andrei.gheorghiu@upb.ro.

In order to increase traffic safety, there was an absolute need to develop and implement a system which would ensure the follow facilities:

• Traffic video surveillance which brings real-time images from the field to the operators and also records those images as evidence of road traffic violations;
• Automatic incident detection which generates alarms in the dispatching center (and, as an option, to the mobile teams) in case of major incident on the road (accidents, abnormal traffic, fire, etc.).
• Automatic adaptation of traffic signalization in according with traffic values on areas of highway, so as to optimize cars fluency in maximum safety conditions.
• Traffic statistics resulted by automatically vehicle count, estimation of vehicle weight and average speed on segments of road, generally dangerous areas, accidents, weather conditions and other.
• Speed detection (speed radar) so as to have permanent measurements of vehicle speed in fixed points or mobile equipment.
• Automatic car identification in fixed points, using high definition cameras and plate recognition software will be able to generate alarms in case any vehicles are identified which are recorded in the data-base of the "wanted" (e.g. stolen cars, cars with no technical or expired inspection clearances or others).
• Law enforcement system with alarm and recording facility (driving over speed limits, ignoring traffic restrictions, ignoring signs etc.) is a useful instrument for the Police force.

The technical solution adopted implements a high speed data network, along the route and connects the technical infrastructure (video surveillance cameras, speed sensors and others) to this network. Thus, this infrastructure can be unlimitedly extended, with new cameras or other sensors.

Keywords: Video Surveillance, Highway Surveillance, Road Surveillance, Road Network, Fibre Optics Network, Logistics, Highway Transport, Highway Monitoring Applications, Information and Communication Technologies (ICT), Informing Service, Monitoring Service, Dispatch, Euro-Corridors, Video-Camera, Speed Radar, Weather Sensors, Weather Monitoring for Highway System, Traffic Video Surveillance, Automatic Incident Detection, Traffic Signaling, Traffic Adaptation, Speed Detection, Optical Character Recognizer, Plate Recognizer, Car Identifier, Doom Camera, Long Way Camera, Radar, Speed Detection, CCTV, Wall Screen, Loop Radar, Traffic Info, Emergency Detection, Emergency Team

1.INTRODUCTION

Generally, a system can be defined as a set of components or elements that are intercorrelated, in order to achieve a certain objective. The system"s components can be objects, concepts, processes or even people. Any system has certain terminators that establish the correlation between system components and environment. The environment is composed from all the external elements and these elements cannot be controlled. The system"s components are connected through interfaces, which control the components interactions. An interurban traffic management system is composed from infrastructure elements, interconnected in order to achieve system"s objectives. These components are physical components (cameras, variable message signs, traffic lights, etc.), communication equipment, a traffic management centre, software and centre operators, the politics and procedures established for events analysis etc. These components can be grouped in subsystems (i.e. traveller information subsystem, ramp metering subsystem) that interact between each other and with the environment.

The system"s objectives are related with the system operating zone, in order to satisfy the user needs and to be adapted to the traffic conditions and policies. However, there are some objectives that are common for all these systems, such as:

• Reducing recurrent congestion and their impact on traffic
• Reducing travel time and of other non-recurrent congestions effects
• Increasing travel safety and efficiency on the interurban road network
• Traveller information, in order to provide a more efficient use of the offered facilities
• Assuring an fast and efficient incident response
• Reducing incident impact on traffic
• Decreasing fuel consumption
• Decreasing air pollution, etc.

According to these objectives, it can be inferred the main functions of a dynamic interurban traffic management system:

• Traffic surveillance the main function of any traffic management system; assures traffic surveillance of the network covered by the system; along with traffic monitoring, it also includes other functions: emergency vehicle localization, public transport vehicle localization, meteorological condition monitoring, pavement monitoring etc.
• Prioritization for public transport or emergency vehicles
• Incident management fast incident detection, necessary resources assessment, fast intervention, reduced traffic impact, incident information for other drivers etc.
• Ramp metering controlled highway access, in order to reduce traffic congestion on highways entries or to offer alternative routes, in case of highway capacity overflow.
• Driver information before or during the travel; the data obtained from the traffic surveillance function are centralized and transmitted to the driver, in order to permit an informed decision; transmitted data are: estimated travel time for a certain route, meteorological condition, pavement condition, incident location, congestions, special events and other information regarding some abnormal traffic conditions.

The main advantage of a dynamic traffic system is its capacity to adapt to real traffic conditions, obtained from the real time traffic monitoring, and to inform the drivers regarding these real traffic data. Knowing this information, the drivers can take better decisions for their routes or itineraries, they can make a better estimation for the travel time and they can travel in safer conditions, by avoiding congestion, adjusting travel speed according to meteorological and pavement condition etc.

2. MODERN CONCEPT ABOUT ACTIVE SAFETY ON HIGHWAYS

Modern highway surveillance is not just a local circuit television system, in fact it realizes all the functions which would be done by field personnel (generally police officers): traffic surveillance, traffic re-routing, emergency services announcement, driver informing on traffic conditions or restrictions. Moreover, the power computing system can automatically optimize traffic based on real field conditions and generate alarms in case of emergency [1].

An Intelligent Highway Surveillance System combines Control Centre, Communication and Operating technology to provide the customer with a traffic management, control, and resolution plan. This concept provides real time active traffic management according to the main elements in the field and improves control, flow and safe on the route.

In order to increase traffic safety, there was an absolute need to develop and implement a system which would ensure the follow facilities:

• Traffic video surveillance which brings real-time images from the field to the operators and also records those images as evidence of road traffic violations;
• Automatic incident detection which generates alarms in the dispatching centre (and, as an option, to the mobile teams) in case of major incident on the road (accidents, abnormal traffic, fire, etc.).
• Automatic adaptation of traffic signalization in according with traffic values on areas of highway, so as to optimize cars fluency in maximum safety conditions.
• Traffic statistics resulted by automatically vehicle count, estimation of vehicle weight and average speed on segments of road, generally dangerous areas, accidents, weather conditions and other.
• Speed detection (speed radar) so as to have permanent measurements of vehicle speed in fixed points or mobile equipment.
• Automatic car identification in fixed points, using high definition cameras and plate recognition software will be able to generate alarms in case any vehicles are identified which are recorded in the data-base of the "wanted" (e.g. stolen cars, cars with no technical or expired inspection clearances or others).
• Automatic law breach detection with alarm and recording facility (driving over speed limits, ignoring traffic restrictions, ignoring signs etc.) is a useful instrument for the Police force. By an efficient identification of drivers who do not observe traffic laws, the safety and normal climate on the road will be significantly improved.

3.INTELLIGENT SURVEILLANCE AND SAFETY SYSTEMS FOR ITS

As explained before, it is obvious that active safety on highways can be obtained only by integrating more systems and their components, by means of unitary intelligent system architecture.

A dynamic interurban traffic management system implies to implement policies, strategies and technologies that will lead to an increased level of performance for the road network. A special attention must be granted to highways traffic. The highways represent those road segments that permit travelling at great speeds and with increased safety

conditions, normally with a controlled access, in order not to overpass the transport capacity and to reduce the risk of congestions or traffic jams. In this chapter, we will report mainly to highways and we will treat the secondary roads as alternative routes, for traffic deviation, if case, and as origins and destinations for highway traffic.

A special consideration will be held to traffic surveillance function, as the main component of the integrated system, and by emphasizing its connections with incident management function and with emergency situation function [2].

The traffic management matter is a very complex one. It cannot be solved only locally, or by taking into account only one segment of an interurban road network. In order to obtain an efficient traffic management, we have to take into consideration also the existing connections between different components of the network, the adjacent roads and their connection (entrance and exit) with the highway, the roads that connect different highways and the connection points between interurban and urban networks. Consequently, an efficient interurban traffic management system must take into account the whole transport corridor [3].

Considering that the main entities of such a logical system are: the highway traffic management system, the secondary roads traffic management system and the travellers, the connection and the data exchange between these subsystems is depicted in Figure 1.

Figure 1. Logical components interface for a transport corridor management system.

Figure 2. Control techniques for an integrated interurban traffic management system

Figure 2 contain a schematic representation of some of the main interurban traffic management techniques, which can be implemented in an integrated system:

• Traffic surveillance ant control
• Ramp metering (access control)
• Variable speed limit management
• Lane management
• Incident management
• Emergency intervention management
• Traveller information and route guidance

4. SENSORS AND DETECTING SYSTEMS

In a traffic management system, the detection and surveillance component represent the part of the system that collect data which describe or characterize the highway traffic conditions. These data are used to offer traffic condition information for other components of the system or for other similar systems. The control and surveillance system assure the necessary data for other traffic management functions, such as: traffic flow measurement, environmental condition analysis, traffic management decisions, information dissemination for travellers, system performance monitoring and evaluation, incident detection and verification, emergency intervention management, ramp control management.

The collected data and their accuracy are reliant on the specific application that is used. The detection and surveillance component is not limited at obtaining data regarding traffic, but it can also offer information regarding the meteorological conditions, road pavement condition, visibility, vehicle position data (for emergency vehicle, for example) and others [4].

The main data assured for interurban traffic management are:

• Traffic volume or traffic flow
• Travel speed
• Occupancy
• Emergency vehicle location or public transport vehicle location [5]
• Queue presence and queue length
• Road pavement condition (ice, snow, wet etc.)
• Meteorological conditions (rain, snowfall, blizzard, humidity, fog etc.)

4.1. Detection Types

The detectors and the detection methods can be classified by many parameters (Figure 3). For example, some detector types register mainly the moving vehicle, other the stopped vehicles, and other the vehicles that are moving with at least 3 – 5 km/h.

The magnetic or normal loop detectors can operate in pulse mode or presence mode. The loop produce a short exist pulse, when detects a vehicle passing over it, but no longer then the vehicle presence in the detection area, due to the fact that vehicle delay will produce a longer signal, directly proportional with the vehicle standing in the detection area.

An extended call detector had an extended exit signal, this meaning the holding or the extension of the call generated by a vehicle for a period of time, previously defined by means of an integrated timer.

A delayed call detector doesn"t emit an exit signal when the detection area is occupied, for a period of time set by means of an incorporated timer.

Figure 3. Detection types.

Actually, hybrid detectors are frequently used for extended and/or delayed call.

It can be also mentioned the detectors that extend the capabilities of regular detectors or controllers, through hardware configuration that contain timers and that display controlling circuits.

The software used allow activating or de-activating of some detectors, control the green time and also activate the extension of the signal for a phase, in order to supplement the controller program.

An example is the "green colour extension system", in order to protect the vehicle situated in a dilemma zone, in a semi-actuated signalling junction. Another type of auxiliary logic detection is the "speed analysis system". This system is a hardware component formed by two loops detectors and an auxiliary logic. The two loops are installed on the same lane, at a measured distance. Vehicles that pass over the loops induce two commands. In order to establish if the vehicle speed is higher or lower than a reference speed, it is measured the time between the two commands [5].

4.2. Functional Description of Detectors

Most detectors are used to identify the movement of the vehicles that pass over a certain area of the road. This information is transmitted to a traffic controller, to a traffic counter or to other devices.

If a detector is equipped with elements that contain directional characteristic, it will register only the vehicles that are passing on a certain direction.

A movement detector will detect a vehicle that pass over it only if its speed is higher than 3-5 km/h. These detectors are used for actuated or semi-actuated controllers. For location with very low speed movements or where parking is permitted, and where directional characteristics are not necessary, there are normally used presence detectors.

When the vehicles are forced to stop or to travel with very low speed, in the proximity of a junction with actuated traffic signals, it is necessary to detect presence, in order to assure the detection of the vehicles stopped in the detection area.

The presence detectors that don"t have a directional recognition characteristic cannot make the difference between the vehicles that enter or exit the detection area. However, the direction it is not important, because the controller normally doesn"t memorize the call, once the vehicle exit the detection area. The call will not be memorised by the passing detector either, for the vehicle that exit the detection area. The difference is the operation mode of the controller: locked or unlocked.

For traffic measurements are usually used passing detectors. This type of detector can be used to count the vehicles for one lane, for more than one lane of a specific direction or for all lanes, both directions. If the lane number that register the number vehicles with one detector increase, the registration accuracy decrease, when the detector is simultaneously occupied by more vehicles.

Passing detectors are also used to measure vehicles flow speed and traffic volumes. The data can be measured for each lane or for all the lanes of one direction.

In order to obtain correct data for vehicles speed and traffic volumes, the detectors must be placed at a convenient distance from the junction, so the queues don"t occupy the detection areas. If this situation occur, only traffic volume can be registered.

Table 1. The Advantages and disadvantages of vehicle detection systems

| Technology | Advantages | Disadvantages | | | | |-----------------|--------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------|--|--|--| | | Flexible design, which allow implementation of many application | Need of pavement works for installing | | | | | | Mature and well known technology | Incorrect installing reduce pavement reliability | | | | | Inductive loop | Can provide the basic traffic parameters (volume, presence, size, speed, | Installing and maintenance works impose lane closing | | | | | | direction and distance between vehicles) | Cable coils can be affected by traffic and temperature
efforts | | | | | | It can be used for vehicle classification, by working at higher frequencies. | Usually, it must be installed more than one detector to provide a zone monitoring | | | | | | Not influenced by meteorological conditions (rain, snow, fog) | Detection accuracy can be lower, in case of necessity to identify many vehicle types | | | | | Magnetometer | Less traffic influenced. | Need of pavement works for installing | | | | | (flow meter | Some models permit
data transfer
by radio
waves. | Incorrect installing reduce pavement reliability | | | | | with two gates) | Not influenced by meteorological conditions (rain, snow, fog) | Installing and maintenance works impose lane closing | | | | | | | Some models have reduced detection
area | | | | | Magnetometer | Can be used in places that are inappropriate for inductive loops (bridges) | Need of pavement cutting
or tunnel digging, for installation. | | | | | (magnetometer | Some models can be installed without pavement cutting. | Cannot detect stationary vehicles. | | | | | with induction | Less traffic influenced. | | | | | | coil) | Not influenced by meteorological conditions (rain, snow, fog) | | | | | | Microwave | Not influenced by meteorological conditions | Antenna radiation
lobe and waveform must be adapted to the specific application. | | | | | radar | Direct speed measurement. | Doppler sensors
don"t detect stationary vehicles. | | | | | | Multiple lane monitoring. | | | | | | | Active detectors transmit multiple beams
in order to accurate detect the | | | | | | Infrared | vehicles
position, speed and class. | Active detectors operation is affected by fog and snow. | | | | | radiation | Multizone passive detectors are used for speed measurement. | Passive detectors can have a lower sensibility in fog or rain conditions. | | | | | | Multiple lane monitoring | | | | | | | | Performances may be affected by external conditions, such as temperature
variation or | | | | | | Multiple lane monitoring. | air turbulence. | | | | | Ultrasound | Possibility of detection
for vehicles
which exceed the height
limit. | Some models are provided with temperature
adjustment
modules.
Increased duration of pulse recurrence
can negatively influence the lane occupancy | | | | | | | | | | | | | | measurements,
when speeds are medium or big | | | | | | Passive detection. | Low temperature
affects measurement
precision. | | | | | Acoustic | Negative influenced by precipitation. | Some models are not recommended for zones with low speed or frequent queues. | | | | | | Multiple lane monitoring. | | | | | | | | Sensibility at any external condition variations, vehicle shadows on adjacent lanes, | | | | | | Multiple lane monitoring and multiple zones per lane monitoring. | day-night transitions, vehicle/load contrast, rain drops, fog etc. | | | | | Video image | Zones can easy be added or modified. | At least 15-21 meters high mounting, for optimal detection
of presence and for speed | | | | | processing | Generous range of data available. | measurements. | | | | | | In case of camera interconnection, it can be provided data for extended areas. | Some models can be affected by strong wind. | | | | | | | Cost effective only if multiple zones detection
is possible with each camera. | | | |

Table 2. Vehicle detection systems: data provided

Technology	Data provided					Multiple lanes,
	Vehicle counting	Vehicle presence	Speed	Occupancy	Classification	multiple detection areas	Communicati on bandwidth
Inductive loop	yes	yes	yes1	yes	yes2		small-medium
Magnetometer (flow meter with two gates)	yes	yes	yes1	yes			small
Magnetometer (magnetometer with induction coil)	yes		yes1	yes			small
Microwave radar	yes	yes	yes	yes	yes	yes	medium
Infrared radiation	yes	yes	yes	yes	yes	yes	small-medium
Ultrasound	yes	yes		yes			small
Acoustic	yes	yes	yes	yes		yes	small-medium
Video image processing	yes	yes	yes	yes	yes	yes	small-high
Triple technology detectors	yes	yes	yes	yes	yes	yes	small-medium

¹ Speed can be measured by using two detectors placed at an established distance.

It can be obtained by using some specialised electronic devices.

The magnetic detector is the most limited, because it can work only in pulse mode (the detector produce a short exit pulse, in the moment of detection). This type of detector can be used only for movement detection at junction entries or as numbering equipment.

Magnetometers and inductive loops can be used for presence and passing detection. Magnetometers can produce satisfying results for vehicle counting if the detection zone is correctly defined. The magnetometers detection area is, normally, less than 1 meter, and the detector may "miss" some vehicles.

The inductive loop detector is recommended especially for presence detection. The size of the loop permits adaptation for many applications.

For small detection area, magnetometers and conventional loops are interchangeable. All three types of detectors are interchangeable only for movement detection (pulse mode). Some designs for high speed junctions are using conventional loops or normal exit magnetometers. Other designs are using control unit, which permit the call extension, even after the vehicle has left the detection area (extended call detectors).

4.3. Advantages and Disadvantages

The vehicle detector systems present many alternatives which are used according to the specific location and to the data which must be obtained, due to the fact that the detectors have different characteristics (measurement distance, influence of external factors etc.). The detectors can be classified in terms of installation mode and detection mode as: in/on pavement detectors and roadside detectors. In table 1 we present a comparison between the most used detectors characteristics. Table 2 present a comparison of detectors, according to the data provided [6].

4.4. Application Examples

4.4.1. Presence and Speed Detection, Using Inductive Loops

Inductive magnetic loops are the most used type of detector. The loops generate an electromagnetic field, which is modified by passing vehicle, hence detecting the respective vehicle. Size and form of loops are different, according to specific situation. The main components of inductive loop detectors are one or more coils from an insulated conductor, placed in a road seating, a power cable connected to a junction box placed roadside, an equipment cabinet ant the interior electronic unit, connected to the power supply. Inductive loops are used to detect passing vehicles, stationary vehicles, for counting vehicles or for size determination. Also, vehicle speed can be calculated by using a pair of loops, placed at a determined distance between each other, on the same lane or by using a single loop, using an algorithm that take into account the loop length, medium vehicle length, loop occupation time and the number of counted vehicle. New version of detectors can provide data for vehicle classification, by using higher frequencies.

In order to control the entire circulation direction (all the lanes), it can be used the scheme presented in Figure 4.

Figure 4. Road monitoring system, by using inductive loops.

The inductive loops can be classified in permanent loops (placed in the pavement and used for a long period of time) or temporary ("glued" to the pavement and used for short periods of time).

The loop is functioning like an inductive element connected to an electronic unit. When a vehicle pass or is stopped over the detector, the inductance decrease. Inductance decreasing conducts to an increase of oscillator frequency. The electronic unit observed this increased frequency and send a pulse to the controller, indicating the vehicle presence [6].

The magnetic flow is evenly generated along the loop, except the end portions. The intensity of the generated magnetic flow is:

$$H = \frac{NI}{l} \tag{1}$$

where, H = magnetic flow intensity [A/m], N = coils number and l = resistor length [m]. Because the generated magnetic flow is even, the magnetic flow is:

$$\Phi = BA \tag{2}$$

where, = [Wb], B = magnetic flow density [T] and A = loop section area [m² ]. The magnetic flow is dependent on the magnetic permeability:

$$B = \mu_0 \mu_r H \tag{3}$$

where, ⁰ = 4 π×10-7 [H/m], ^r = relative permeability of the material (1 for air) [H/m]. Loop inductance is:

$$L = \frac{N\Phi}{I} = \frac{NBA}{I} = \frac{N\mu_0\mu_r HA}{I} = \frac{\mu_0\mu_r N^2 A}{l}$$ (4)

where, I = electrical current intensity [A].

The main parameters that can be obtained based on inductive loops data are:

Volume:

$$Q = N/T \tag{5}$$

where, Q = detected vehicles/hour, N = detected vehicle in time interval T, and T = time interval, in hours.

Occupancy:

$$\theta = \frac{100}{T} \sum_{i=1}^{N} (t_i - D) \tag{6}$$

where, θ = occupancy[%], T = time interval [hours], tⁱ = total detector pulse time and D = descendent slope time - ascendant slope time.

Speed:

$$V = \frac{3.6 \times 10^6 d}{5.280(t_1 - t_0)} \tag{7}$$

where, V = vehicle speed [km/h].

In case of using one detector: d = medium vehicle length + detector length [meters], t⁰ = detection start time [ms] and t¹ = detection end time [ms].

In case of using two detectors: d = the distance between the two detectors [meters], t⁰ = detection start time for first detector [ms] and t¹ = detection start time for second detector [ms].

Based on the previous formula, it can be deduced the exact length of the vehicle:

$$L{v} = \left(\frac{1}{2}\right) \left[ \left(t{11} - t{10}\right) + \left(t{21} - t_{20}\right) \right] \left(\frac{5.280V}{3.6 \times 10^{6}}\right)$$ (8)

where, V = previously established speed, ti⁰ = detection start time for detector i[ms] and ti¹ = detection end time for detector i[ms].

Another formula for speed calculation, function of the volume and the occupancy is:

$$V = C\frac{Q}{\theta} \tag{9}$$

where, C = calibration coefficient, experimental determined.

Vehicle density:

$$K = \left(\frac{1}{T}\right) \sum_{i=1}^{N} \left(\frac{1}{V_i}\right) \tag{10}$$

where, K = vehicle density, N = detected vehicles number per time unit T, T = time unit [hours] and Vⁱ = speed of detected vehicle i.

Speed (Figure 5) = $$K * \frac{\text{(Vehicle length + Loop length)}}{\text{(Pulse end time) - (Pulse start time)}}$$ (11)

Speed (Figure 6) = $$K * \frac{D}{\text{(Pulse start time 2nd Detector)} - \text{(Pulse start time 1st Detector)}}$$ (12)

4.4.2. Automatic Plate Recognition Systems with Video Cameras

Video image processing sensors use visible light range and a closed to infrared light range, in order to determine more traffic parameters, according to the implemented configuration. Video image processing sensor permits obtaining traffic data for complex areas, with multiple lanes. However, the necessity of video image digitization and of model recognition routines imposes spending a lot of energy in the field. Such a sensor can provide data regarding speed and traffic volume and it can also be used for incident detection and traffic signal control. Video image sensors represent a very fruitful solution for traffic flow detection, due to the increasing of performances, high flexibility of characteristics and price decreasing.

Figure 5. Speed detection with one inductive loop.

Figure 6. Speed detection with two inductive loops.

A video detection system is composed by a number of video cameras, digitisation and image processing system and software for data processing and data generation. These systems are looking at the images from an area, in order to observe any modification. Complex algorithms are implemented to identify vehicles (cars, trucks, motorcycles, bicycles), irrespective of external condition, shadows, day-night transitions etc. Dependent on algorithms complexity, it can be obtained data that are not available from other detectors, such as cars registration numbers, it can be provided data about complex situations, such as road incidents or it can be monitored the drivers behaviour.

The following categories of video detection systems are frequently used:

• Tripline The user can define only a limited number of detection zones. Vehicles are detected through pixels changes of the image, against the situation when the road is clear. There are two detection systems: one realises a surface analysis, identifying the objects profile; the other realises a grid analysis and identifies in each square vehicle in movement, stationary vehicle or the absence of vehicles. Speed measurement is made by computation, the system providing data about the time needed by a vehicle to transit a known distance.
• Close loop tracking is an extension of tripline system and offers the advantage of extended zone detection. The system track permanently the vehicles situated in the (visual) detection area. In order to validate a vehicle, it is necessary to obtain more detections of the respective vehicle along its route. After validation, vehicle speed can be obtained, besides other data, such as lane changing. The exit data can be sent to variable message signs or to on-board equipment.
• Data association tracking these systems are tracking certain vehicles or vehicle groups, by searching of connected pixels areas. These areas are tracked then on all the frames, to obtain the identified vehicle or vehicles group tracking. The objects are identified by using some markers, based on gradient and morphology. Gradient based markers identify contours, and morphology based markers identify characteristics and dimensions combinations, which are associated with a known vehicle or a known vehicle group. By successively identifying a vehicle though video system, it can be determined its route, travel time, medium speed, driver behaviour and other useful statistics data.

Image formatting and data extraction are performed through powerful operating systems, which allow real time running of processing algorithms. After data digitisation it is performed a first vehicle identification (identification errors may appear in this stage). A second data processing eliminates the erroneous vehicles and maintains only the real vehicles.

Figure 7 presents the event succession that lead to vehicles identification and tracking.

The following paragraphs present vehicle detection modalities and a registration number recognition algorithm (useful for access control, automated taxes collection, ramp control etc.).

Figure 7. Video image processing.

Figure 8. Cases of wrong vehicle detection.

Figure 9. Contour detection.

4.4.2.1. Vehicle Recognition

The first detection method described regards using symmetry and framing scale to identify vehicles. The algorithms are based on the idea that most vehicles are symmetrical and can be framed in rectangles, on the video camera images. In order to determine symmetric elements it is not enough an image grey tones analysis, due to the possibility of anomalies. In Figure 8 it can be observed three such examples: in the first image, the vehicle is not symmetric due to the solar rays reflection on the left side; in the second image it is detected an uniform zone of the lane; in the third image, a background area has symmetrical characteristics [7].

In order to eliminate reflections and uniform areas it can be performed an object contour detection. Next, symmetrical elements search is performed. The first and second problems are eliminated by this method. In the first case, the object contour can be detected, despite the stronger brightness from the left side of the vehicle. In the second case, there are no contours to be detected, for the uniform area.

The next step is to determine the framing rectangles. First, it is searched the lower side of the rectangle.

As we mentioned before, the framing rectangle determination it is not made only for the original picture, but also for the image with object contours identified. In Figure 9, it is depicted the process to determine the lower side of the framing rectangle, and the last step is vehicle identification.

After the lower side identification, the same criteria are applied to determine the higher side of the rectangle. This area is determined by taking into account the camera location and its perspective over the road.

The last step is a backtracking algorithm which eliminates the determined objects which cannot be vehicles, such as background areas. These areas cannot contain vehicles, due to their position in relation to the road.

Another solution uses the Block Matching Algorithm (BMA) – the movement detection algorithm for MPEG (Moving Picture Experts Group) standard. This algorithm has the advantage that it can be used to determine not only the movement, but also the movement direction.

The image is divided in specified dimensions blocks (6x6 pixels, 8x8 pixels etc.) and for each part of it there are analysed the differences between the current image and the previous image. It has to be done estimation for blocks position in the previous image, in order to calculate the movements and speed. Each block defines a search area, with the respective block in the middle. The initial block is translated in the searching area, to determine the correspondence level for each translation. A movement vector for the block it is generated, in the moment of location determination.

Based on the speed or direction differences, a distinction can be made even for close vehicles. The algorithm can produce erroneous vectors, which can see, for example, the same vehicle as two different objects. In order to eliminate this problem, the next step performs a medium filtering of values, using the eight adjacent vectors and taking into account only their non-zero values, to the effect that it will be avoided the situation of transforming in a zero vector a non-zero vector, surrounded by zero vectors.

The objective is to minimise the sum of distances to all neighbours and the distance between two vectors is:

$$|\vec{u} - \vec{v}| = \sqrt{(x_u - x_v)^2 + (y_u - y_v)^2}$$ (13)

This regularisation significantly reduces the noises and produces block sets with more evenly disposed vectors.

After data filtering and obtaining compact blocks which represent vehicles, it will be performed blocks numbering. The numbering is also performed based on the adjacent blocks. The numbering process is an iterative process. Two blocks are considered adjacent if they have adjacent borders with identical vectors and the distance between the two blocks is lower than a specified admissible detection error.

Vehicle tracking is based on current image blocks identification and numbering. The numbering is temporary, being updated according with the determined movement vectors.

Figure 10. Object identification through constituent blocks.

Each identified block from the current image is relocated back to the previous image, with a vector that is equal with the opposite of the current movement vector ( v v ). Next, it is estimated the superimposing between the current label and the previous label. In Figure 10, the left side shows the previous numbering, and the right side shows the current temporary numbering. It is possible that the superimposing is not perfect, being reliant to the object detection accuracy for each image.

After verifying each "3" pixel superimposing over the previously image pixels, it will be decided that the previous label is better covered than the current image. The respective label will be transferred to the current object (in our example, the current object will have the final label: "5"). The procedure is repeated in order to update the numbers for all last image identified objects. In case that the number for a certain object cannot be identified from the previous sequence, this will be classified as a new object.

4.4.2.2. Vehicle Registration Number Recognition

This procedure can be used for automatic vehicle recognition, by identifying the unique registration number.

Some of the methods are: recognition by OCR (Optical Character Recognition) techniques, model based recognition or neuronal network based methods. Here it is explained a method which identify the registration number zone, decompose plate characteristics and identify the existing characters on this zone.

The first step is to extract registration number zone. The video camera image is transformed in a binary image (0-black and 1-white). This decision is taken by comparing each pixel"s luminosity with an established value.

The resulted image is processed in order to decide the registration number area. A lines and columns scanning is performed, and in case that the white pixels number is not included in a specified range, the white pixels are turned black.

Figure 11. Obtaining of binary image.

Next step is registration number extraction from the digitised image, based on the decided area.

Finally, an image containing the registration number plate is obtained.

The following step aims to identify the characters. First an image filtering is carried out, to eliminate noise or unwanted points. Scanning procedure is again applied, to extract the areas that contain characters.

Individual characters are extracted, by applying this mask on the previously identified number.

Before starting the recognition procedure, the characters are normalizes. This operation consists in modifying characters in order to eliminate any white space between the character and the circumscribed cassette.

Finally, o model comparison is made, by determining the best conformity of each character with a model character. It is possible to use more models, which different character sets, in order to correctly identify all the vehicles, irrelevant of the region or country of provenience.

In conclusion, video detection systems have the larger application range, but imply a high calculation power and complex algorithms for image processing.

5. COMMUNICATION TECHNOLOGIES

5.1. Introduction

Data communications are required for data collection and information sharing. In terms of road infrastructure, communications used are stationary. Data from fixed sources such as loop detectors and closed circuit television may be sent to management centres and then distributed through fixed and mobile communications. Vehicles need mobile communications. Data from mobile sources, such as helicopter patrols and outputs of data processors located inside the vehicles, such as GPS coordinates, must be submitted to the management centre through wireless or mobile communications. All existing and future telecommunications infrastructure, stationary and / or mobile can and should be used for ITS functions in order to minimize costs and benefit from the continuing technological progress in the telecommunications industry.

5.2. Wireless Technologies

Wireless communications are used more and more for ITS services. In many countries, drivers of vehicles report often incidents on highways through mobile phones. New personal communication systems are digital. Data communications via digital cellular systems have become feasible due to cellular digital packet and particular techniques of wireless data communication packet switching. The new digital cellular telephone services based on standards like TDMA (Time Division Multiple Access), CDMA (Code Division Multiple Access) as well as European GSM standard have already been launched on the market.

Mobile communications may be used in large areas, in which the transmitter and receiver are located far away from each other (the order of kilometres). Usually in communications over large areas, ITS data is grouped with other messages. In contrast to these are dedicated short range communications (DSRC) for ITS, where the range is so short that the communications are useful for certain dedicated purposes. In general, those who invest in DSRC transmitters (beacons) infrastructure are not part of the telecommunications industry, but are public or private organizations interested in some type of ITS systems using dedicated short-range communications.

ITS services based on dedicated short-range communications are:

• Electronic toll collection;
• Commercial vehicles operation;
• Parking management;
• Signal programming;
• Signalling inside the vehicles;
• Information of the passengers inside the vehicles;
• Route guidance based on the beacons.

5.3. Radio Data Systems (RDS)

Radio Data System represents a form of channel data coding design to operate along with FM commercial radio emissions. The RDS principle is the following: commercial FM radio has some "spaces" between frequencies. This space, which serves as a buffer between radio stations can be used to transmit special audio programs or to transmit digital information. This sub-sender has been used in the past to broadcast background music in restaurants and shops. The efficiency of the system is based on the fact that there are no connections between the sub-sender and the main audio program.

RDS main motivation is to allow commercial radio stations to be identified on the radio screen and to admit self-regulating radio transmissions, that would automatically select the strongest signals corresponding to a particular type of program (such as news, sports, different types of music and traffic information). From the ITS perspective, an important feature of RDS is TMC (Traffic Message Channel), that can provide passengers information about current traffic. Traffic information transmitted to vehicle passengers by RDS digital coded silent messages will get to a display and / or voice synthesizer receiver inside the vehicle. Messages are language independent; they can be decoded in the language chosen by the user. Incoming messages are stored in the memory of the receiver and may be selected by the user as required.

RDS-TMC requires a special decoder. The reason is that RDS-TMC is a sub-sender consisting of ultrasonic signals beyond human hearing ability. Moreover, conventional radios cannot decode these signals without adding the special decoder. Basic implementation allows a radio to indicate whether the receiver outputs traffic information as part of the current program and if the receiver is now getting a traffic message. This information could be used to stop a cassette or CD or turn on the radio during the announcement. Usually RDS radios allow the auditor to use this information as search criteria for identification of radio stations and choosing only the ones that transmit traffic information. While the receiver/decoder provides the user interface to the travel information service, its functionality may vary substantially according to the technical development and market needs.

5.4. VHF-PMR (Professional Mobile Radio) Communications

PMR is a concept which provides complete support for radio networks users. Generally, PMR sets technical parameters for mobile, fixed or portable radio stations. The following two requirements are fundamental in determining the usability of a PMR system:

• Full rural coverage in normal operating conditions;
• Mechanisms that ensure the coverage in exceptional circumstances.

Additionally, there is often the need to provide dedicated frequencies for air-ground or ground-air communications in an organizational network. Dedicated frequencies are required to eliminate the interferences that the air emitter could induce in ground stations, even for larger distances. PMR technology creates special configurations to provide radio coverage even in case of network downtime such as faults of base stations. PMR key user needs can be classified as follows:

• Specialized functionalities for group and dispatched communications, including groups" management in a safe and dynamic mode, emergency calls management, prioritizing of communications and so on;
• Response times, normally required to establish a voice communication, are in the range 0.3 to 1 second, with an average of 0.5 seconds for operation in large areas;
• Continuous radio coverage throughout the service area, including guaranteed availability of coverage even in exceptional circumstances;
• Voice quality without compromise, allowing voice recognition, even under excessive external noise.

Critical users of the radio system (which currently use private radio networks and PMR technologies) have specialized requirements for functionality, response times, radio coverage and quality of service especially during major incidents - with few possibilities for compromise (if any) when the safety of society or human life must be protected.

5.5. Area Radio Network (ARN)

Area radio signals radiate uniformly in all directions, like point-to-point links between microwaves. The dispersion and reflection allow the signal to expand on built surfaces, although these reduce the signal strength. Land barriers may limit the feasibility of this technology.

There are various frequency bands available for ARN applications; typically, two frequencies are assigned to an application. For voice transmissions all receivers are set to the same broadcaster and the users respond only to messages directed to them. ARN networks have been primarily used for fleets of vehicles.

A more sophisticated version of this technology is ESMR (Enhanced Specialized Mobile Radio). This radio uses digital technology to allow more users per channel and provide more services to the users. Services based on this technology may represent a major competition for cellular phone services in the near future.

5.6. TETRA (TErrestrial Trunked Radio) – Professional Mobile Communications for Closed User Groups

TETRA is a global wireless communications standard that presents a lot of similarities with GSM. In both cases are used cellular systems for digital voice and data communications. Unlike public GSM network, TETRA focuses especially on support offered for special services and restricted professional groups such as police, fire brigade, ambulance, transport companies etc. TETRA networks offer professional support, redundancy and priority levels for the users in the group. In addition, modern TETRA networks are deployed on IP platforms, in order to ensure 100% compatibility with the terrestrial networks. Being standardized since 1994, TETRA is now available in all European countries for nongovernmental group of users. In the TETRA network can be defined closed user groups. It can perform communication between two users or between two groups.

One of the specific features of TETRA is the dispatcher function. A dispatcher fulfils a central role for the communications in company or group. The dispatcher can, optionally, take part in all the connections made (listening and communicating). It is noticeable that TETRA can make a connection without network support (base stations) – in direct mode. Direct mode provides point-to-multipoint communications without any instructions from a dispatcher. A TETRA unit can also function as a repeater between a base station and another terminal, situated outside the base station coverage. Relay option allows a station within the network coverage area to be used as a relay station to another station outside the coverage area, using direct mode.

An important requirement is redundancy of trunk connection circuits" arrangements, allowing traffic routing in case of line faults. In an extreme case, of full isolation, each network element is capable of standalone operation. It is necessary to ensure that the transmitting nodes in the network will not affect the overall performance of the network, even from the design phase. Network architecture must be tolerant to failures in order to provide a good redundancy. To achieve this tolerance, all the modules are redundant, even doubled, or provided with backup.

The system software provides automatic supervision of the operation, even "watch-dog" routines, automatic generation of alarms, fault diagnosis and recovery functions to bypass a faulty module or for routing.

TETRA system supports emergency calling on different priority levels. The emergency call shall get access to a higher priority level, in terms of both radio and network resources. In case of too many emergency calls received at a time, for example in case of major incidents, the system will automatically divide its resources according to priority classes. All radio subscribers have the right to initiate emergency calls, but the calls will be segregated into priority classes to allow end users to distinguish real emergencies from a message sent to the dispatcher. All users are able to initiate emergency calls routed directly to the national emergency number 112. The network is designed to have sufficient resources for emergency calls in any situation that would arise and no user or group of users will have a higher priority level than emergency calls. But, if necessary, network resources can be de-allocated to perform emergency transmissions even if it means to disturb other on-going conversations.

5.7. Special Radio Systems and Short Distance Communication Systems

Some radio systems or radio networks support a specific type of application or are even dedicated to a specific target group. Below we shall present special radio systems used in transport and traffic management applications. DSRC, WLAN or Bluetooth systems are developed for proprietary indoor enterprise applications. These systems may also be used in outdoor applications, for transport and traffic management.

5.7.1. DSRC (Dedicated Short Range Communication)

DSRC is a communications standard developed in Europe, USA and Japan to support ITS type systems. DSRC is created by short-range communication devices capable of transferring data at high speed, using air interface, between the moving or stationary vehicle and devices normally stationary, which is often mounted on structures along the road, or are hand-held type.

DSRC is suitable for communications on limited range of approximately 30 m. Applications are in the field of road traffic, for communications between the RSE (Road Side Equipment) and OBE (On-Board Equipment). RSEs are radio beacons placed above or along the road. OBEs are small and represent units installed in vehicles, trailers or containers. DSRC communications supports bidirectional half-duplex mode, for which RSE controls the connection. For simple applications, such as automatic fare collection, OBE can be pseudopassive, if installed behind the windshield of the vehicle. This type of OBE does not generate radio signals independently, the energy used to emit the response being obtained from the RSE signal. The energy received is sufficient to achieve temporary high-speed links. The DSRC standard describes only communications between road side equipment and equipment inside vehicles. It contains no description of the network structure and data protocols used for RSE interconnections.

DSRC radio channel bandwidth is 3 MHz and modulation type is binary AM (Amplitude Modulation). Due to very low range of DSRC and that the antennas used are unidirectional (small bundle), the same frequencies can be reused after short distance (several hundred meters).

ITS architecture requires DSRC in those specific applications that necessitate a close physical interaction between vehicles and roadside infrastructure such as fare collection, electronic authorization of commercial vehicles, safety inspections etc. The following applications have been identified, that may use DSRC as the main mechanism and communication technique:

• Electronic payment of taxes and parking
• Commercial Vehicle Operations
• Authorizations at international borders
• Electronic authorization

• Safety inspection

• Automatic identification of equipment and freight management

• Off-line check

• Operations with emergency and transit vehicles

• Fleet management

• Vehicle use as probes to obtain connection times

• Information on-board the vehicle

• Avoiding collisions at intersections and automated highway systems

• Commercial applications

• Safety warnings at intersections.

5.7.2. WLAN (Wireless Local Area Network)

Wireless networks, such as wireless LAN (WLAN) are used to achieve mobility and flexibility, such as the extension of a wired network. There are currently a number of standards for these systems, including the US IEEE 802.11 and the European HIPERLAN.

It is customary to link the different equipment and in LANs - Local Area Networks. Every aspect (physical connection, the signals used, transmission protocols) for such networks are standardized so that minimum effort is needed to connect different entities.

The structure of a WLAN can be flexibly changed to meet all requirements. The simplest form of a network is the connection of two computers, in which case the communication will be an "ad-hoc" one. For more complex situations the equipment may be connected using a cellular network concept. This implies that the transmission of data between them is managed and done through a base station (Access Point). To ensure coverage of larger areas there is the need of more cells. This type of connection between different base stations can be done via an existing network (usual LAN) or via a wireless network. WLAN standard supports user mobility: participants may have access to different access points by roaming. However, the mobility is limited, the connection being maintained as long as the user moves relatively slowly (10 km/h).

In WLAN systems are applied some techniques to ensure noise immunity: FHSS (Frequency Hopping Spread Spectrum) and DSSS (Direct Sequence Spread Spectrum). In FHSS frequencies are changed regularly and may jump between different channels in the spectrum. The DSSS system uses a single coding system to spread the information using an extremely wide range of frequencies. Both methods give good results in case of electromagnetic disturbances and contribute to anonymity (sender and recipient addresses) and immunity from unauthorized interception.

WLAN systems based on the IEEE 802.11 standard provide a wireless connection for data transfer speed of 54 Mbps. The effective data transmission rate per terminal is relatively modest. This is because it is necessary to resend the messages and the available capacity of radio channel per unit is dependent on the current requests from other cells operating in the WLAN. Typically, in the absence of other units, the effective rate of transfer between the units is 10 Mbps.

In addition to the frequency band of 2.4 GHz there are WLAN systems available for frequencies around 5.8 GHz. In these WLAN systems, based on HIPERLAN standard, are attainable higher transfer speeds.

5.7.3. Wi-Fi (Wireless Fidelity)

"Wi-Fi" is a trademark of the Wi-Fi Alliance, which uses IEEE protocol 802.11b, a special WLAN version, which was described above. Wi-Fi uses access-points that connect to high-speed data lines from the data network and the Internet.

Wi-Fi is designed to transmit data over wireless network portions at very high speeds, easily reaching Mbps. These high-speed Wi-Fi connections are governed by several factors, including: the number of active users that utilize a single access point, the operating distance from the access point, any obstructions blocking the signal. Generally, Wi-Fi signals are transmitted relatively well through the glass and several types of walls, but do not propagate well through metal, concrete or buildings.

The term "hotspot" became popular, referring to a publicly available access point in a location other than home or work. In general, all hotspots and Wi-Fi access points have a limited radius of action, around 100 m or less. New technologies and the use of multiple access points will make possible the extension of the signal"s range.

5.7.4. Bluetooth

Bluetooth is a relatively newly developed technology for data transmission over short distances. It aims at first to find an alternative for the need to interconnect different peripherals of PCs with cables: cable interconnecting keyboard, mouse, printer, etc. Bluetooth can also be used for various applications such as wireless telephony, shopping, electronic parking payment, public transport or access control.

Bluetooth standard (IEEE 803.15) contains the description of all issues related to data sharing, in this case radio interface, transmission protocols and signal characteristics. Each Bluetooth network has one "master" equipment and one to seven "slaves". Each unit has a unique 12-bit address. The "master" controls the access and internal transmissions.

The Bluetooth transmission power is divided into classes. Class 3 has a maximum transmission power of 100 mW (covering about 100 m) and Class 2 transmit power is only 2.5 mW (covering about 1 m). Bluetooth also support limited mobility. In the coverage area of the "master" station, the link is maintained even if the cell is slowly moving. Bluetooth radio interface uses frequency hopping over the whole ISM band (79 channels). A radio channel has a bandwidth of 1 MHz and is used for uplink and downlink.

Bluetooth connection capability is not necessarily symmetrically divided between uplink and downlink. In the case of a symmetrical partition, the data transmission rate for each link is 432.6 kbps. In the case of an asymmetric division, the maximum speed for a link is 721 kbps, and for the other is 57.6 kbps. The exchange of information is done the data packets. Similarly to WLAN, information exchange takes place without errors based on resending the messages received with errors. Information can be protected against interception using encryption keys.

5.7.5. NTCIP (National Transportation Communications for ITS Protocol)

NTCIP was originally designed as an extension of the NEMATS-2 standard, used in traffic control communications. For the functional implementation of ITS, all the components that make up the traffic and transportation control must be able to communicate through a common language, or at least an understandable one. Words that are "spoken" must have a clear and unambiguous meaning for everyone. NTCIP developers started by defining a language for traffic control and expanded it to include Traffic Management Centres (TCM). Later it was diversified, resulting an open set of protocols to meet the various requirements of ITS.

6.INFORMING PUBLIC SYSTEMS

The continuous increase of traffic in major cities led to the mandatory usage of traffic management systems in order to have an organized traffic. There are many levels such systems may be implemented, depending on the traffic volume, starting from the static signalling systems up to management systems based on complex algorithms that consider a series of key factors.

Information obtained by the management traffic systems must be disseminated to the users of the system: the travellers. Beside the static signalisation methods, it is essential to obtain dynamic information of the traffic participants, meaning an information regarding the real traffic condition.

There are many technologies that can be used to realise this goal, the most usual being: variable message signs, highway advisory radio, personal information systems, vehicle information systems, commercial radio broadcasting etc.

Below we present a project focused on Multimodal Real Time Traffic and Travel Information (RTTI) services with the goal to reduce drastically energy consumption in urban areas across the different modes of transport by changing the mobility behaviour (modal shift) of the single traveller. The same project can be used to transmit information also for the travellers that are using the interurban road network.

6.1. The In-Time Project - Introduction

The In-Time (Intelligent and Efficient Travel management for European Cities) project is developing a traffic management system that provides mainly 3 services:

• Business-to-business (B2B) services that will enable European-wide Traffic Information Service Providers (TISPs) to get access to regional traffic and travel data and services of the single pilot cities via a harmonized standardized open interface. This will enable the TISP to provide interoperable and multimodal RTTI services (eservices) to their end-users.
• E-services will influence the on-trip travel behaviour by optimizing journeys taking the energy consumption into account. The community will be the users of mobile devices or navigational devices.
• Web based interoperable and intermodal pre-trip information will be provided by the pilot operators and has the potential to influence the travel behaviour in the trip planning stage by taking environmental aspects into account. The typical users are persons that are planning an urban trip on short term. [8]

The fundamental concept and goal of the In-Time project is to pilot and validate – in terms of transport benefits, user acceptance and market potentials – an innovative panEuropean approach to RTTI service provision based on an open, standardized service oriented infrastructure and B2B services that will facilitate access to urban traffic related data, RTTI service provision and interoperability by TISPs.

The central part of the In-Time concept is the interoperable and multimodal Regional Data/Service Server (RDSS), a service-oriented middleware infrastructure providing a standardized B2B interface and a number of data/services covering individual traffic, public transport, weather, location based services, inter-modal transport planning. Investigated and tested at a prototype level in the EU project eMOTION the In-Time RDSS enables the operation of end-user applications (web based B2C applications and services) as well as the B2B access via the standardized interface and e-service provision by European TISPs. The In-Time RDSS will be set up in all pilot sites to ensure the easy access of real-time multimodal traffic data for external TISPs. This model ensures the easy access to all urban traffic related data within one region resulting in the distribution to the end-users via several information channels and in parallel enhancing user acceptance.

The RDSS can be seen as a service-oriented middleware infrastructure providing a number of data/services, covering:

• individual traffic,
• public transport,
• weather,
• location based services,
• intermodal transport planning,

and enables the operation of the In-Time services.

Figure 12. The In-Time concept.

Figure 13. RDSS concept.

6.2. Description of In-Time Services

Web based interoperable and intermodal pre-trip and on-trip information are very common interfaces from the single operators to their users. "Pre-trip information is already an important contribution to sustainable mobility" (Final Report of the eSafety RTTIWG), but has in many cases the disadvantage that a door-to-door trip planning taking different modes into account, is currently impossible.

The new approach within In-Time will ensure that on the one hand each pilot site will have the complete travel information (public and individual traffic) combined and interpreted at one location - the Interoperable and Multimodal In-Time Regional Data/Service Server (RDSS). This ensures that the user gets accurate and precise door-to-door information on the current travel times within the city, including alternatives. But as shown in above usage example, the setup of the RDSS will give also the possibility for a seamless transition of data/services between the single regions and to result in interoperable cross-border travel information. It is expected that this service will facilitate inter-modal changes, thus attracting travellers to less energy consuming transport modes.

For the implementation of web based interoperable and intermodal pre-trip and on-trip information the costs will be mainly with the operators, that might be public authorities and ppp-models.

Accurate reliable intermodal real-time on-trip Traffic and Travel information as planned in the In-Time e-services has the potential to avoid congestions and bottlenecks, to select alternative routes and/or alternative transport modes. Hereby one key factor is, that the user needs to be used to the device he uses, therefore within In-Time standard products of Wayfinder will be enhanced by the e-service.

Figure 14. In-Time pilot sites.

With the assessment of the operating Traffic Management the environmental impact of new traffic management technologies, that have reduced energy efforts, it is possible not only to improve visibility but also to cut the energy consumed to at least one fifth (in terms of CO2 emissions, comparable to the local impact of vehicle control measures!). The pilot of Bucharest is an opportunity to make direct measures of the savings in terms of overall energy efficiency across the city network, as here historical data from before the implementation of modern traffic management technologies are available.

Beside the energy consumption also the efficiency of the Infrastructure Device Product Life Cycle needs to be taken into consideration. A global approach must consider also energy efficiency of the whole life-cycle of products used for traffic control. Again the example comes from the signal heads, specific LED-based products have gone through the assessment of their Global Warming Potential and demonstrated dramatic improvement if compared with traditional productions. This technology is used in the field in the Bucharest demonstrator and analysis for future "green-certification" rules will be performed.

6.3. IN-TIME Pilot Sites

The In-Time services is piloted and operated in following European cities (Figure 14):

Vienna: 3.5 M inhabitants Brno: 0.4 M inhabitants Munich: 1.3 M inhabitants Bucharest: 2.3 M inhabitants Oslo: 0.5 M inhabitants Florence: 0.6 M inhabitants

6.4. IN-TIME Bucharest Pilot Site

Bucharest is one of the major growing cities within Europe with a potential of approx. 7 million inhabitants until 2015. In Bucharest it was installed and it is functional a complete traffic management system, which consists of:

• 140 intersections controlled with the UTOPIA UTC system;
• 300 vehicles monitored and managed by the MIZAR Fleet Management system FLASH
• about 300 probe vehicles equipped with monitoring equipment
• 40 CCTV cameras used for security monitoring purposes.

The dynamic traffic control is implemented in order to measure the impact of the adaptive traffic control system configured to reduce fuel consumption (i.e. implement the ecoflow concept) with selective bus priority function the data collected used as a basis will include:

• travel time both private cars and public transport vehicles (300) on selected fixed routes within the area where the UTC system is active (these are referred to in the following as fixed route trials);
• measurement of number of stops made at traffic lights and overall time vehicle is stationary (based on probe vehicle data);
• average speed of private vehicles when in motion.

To assess the impact of pre-trip and on-trip traveller information and guidance on traveller route and modal choices, it is planned to collect the following data:

• travel time broken down into component areas (e.g. on board, waiting time at bus stop etc.) for selected OD relations within the area where the Bucharest city systems are available. This will be done where users have free path selection and will be carried out with the Bucharest systems ON and OFF. These trials will be referred to as OD trials. The survey will be designed mainly to assess the differences in total distances travelled and, where possible, vehicle speed. The "Politehnica" University participated in 2007 as subcontractor in the project "Bucharest Transport Masterplan", with the Municipality of Bucharest as beneficiary. Part of the project involved collecting of OD data and also Household Surveys (questionnaires about public travel by private cars and public transport: time of departure, time of arrival, mode of transport, waiting time and so on). This data can be updated and completed with other necessary information.
• Energy consumption estimates: it is intended to apply energy consumption models and algorithms (sing a methodology to be developed during the first phase of the project) to the above data in order to produce estimates of energy consumption in the different ITS areas. In addition, overall fuel consumption and motorisation statistics for Bucharest will be used to calculate:
• the change in energy consumption (volume of fuel consumed per million inhabitants/year for vehicles within the controlled system).

The efficiency of technology choices with regard to energy consumption of the infrastructure and especially the adoption of LED technology for signal heads will be assessed by means of:

• an estimate of the amount of energy needed to operate the traffic control system in the area concerned; Here, in addition, it is also planned to make
• a product life cycle analysis, with information from the manufacturer to permit an estimate of total costs over full lifespan of the LED technology;
• an estimate of CO² emissions is also possible.

The pilot project is an opportunity to make direct measures of the savings in terms of overall energy efficiency across the city network.

As a part of the pilot, it is planned to use data collected from probe vehicles to constantly monitor the operational effectiveness in terms of energy efficiency of the measures described above. The continuous measurement of actual average fuel consumption of vehicles driving through the network will provide evidence of the overall trend and quantify the impact of any new strategies. This approach will guarantee constant improvement in the policies adopted, as for other network parameters are already commonly measured (e.g. travel time).

7. COMMAND AND CONTROL CENTRE

7.1. Introduction

In the actually world when more and more actions are centralized managed by specialized experts, and the information are transferred and stored using electronically devices, it is absolutely necessary to ensure reliable data infrastructure, including central managing points and communication data networks. As a beginning, a special attention in this domain has been ensured by the emergency organizations around the world, those being some of the first entities which have to involve in such cases. For effective activities, they need the most realistic information from the field and well coordination in action, all of this needing a reliable communication infrastructure.

7.2. Concept

The concept of Command and Control Centre for traffic management starts from the idea that the action has to be fully controlled from one central point.

The Command and Control Centre are permanently working systems, typically 24/7, representing the "heart" and the "brain" of the complex integrated system. The operators are responsible of all systems and actions, their job being the coordination and supervising of the automatic process in the field.

Each place in the Command and Control Centre has to ensure well working conditions, both for operators, support teams and, not at last, the equipment.

The main spaces in any Centre are:

• Operators room (Dispatch) probably the most important area and de operative core, this area will be dimensioned so as to ensure enough space for all regular operators, supervisors and visitors. From the design phase, any control room has to ensure enough space for all future extensions. Technically, the space provides a highreliability powerful data network, redundant, capable to cover all voice and data needs. The control room ensures conditions for at least the follow categories of operators:
  • Permanent operators;
  • Supervisors;
  • Experts;
  • Technical referents and specialists;

As an option, in the Dispatch there will be possible to add more operators, permanent or visitors. For this, the infrastructure can be developed so as to ensure enough technical resources.

• Equipment room dedicated to all technical facilities in the Centre, it provides enough space for all servers, wall-display equipment, telecommunication equipment and network. The space must be well ventilated, secured and dotted with inert gas fire protection. Preferable, the space should be designed in according with TIER 4 standards suite.
• Other spaces dedicated spaces, equipped with modern facilities, will be delivered. The role of those spaces is to ensure all facilities for normal working conditions for operators.

7.3. Architecture

The Dispatcher should contain equipment with high processing capacity, redundant to ensure availability, built-in security solutions, easy to manage and high performance, capable of processing data traffic received from video cameras and other sensors place in the field and which need to be monitored.

In intersections where FO is available industrial switches should be used, fan-less and optimised for ensuring connectivity between the surveillance / miniaturization and dispatcher. Industrial switches offer the possibility of using ring topology that can suffer single point of failure (broken cable or removed connector) while continuing to work as if no incident occurred in the network. This solution ensures continuous functioning to the surveillance cameras, traffic lights and other sensors with the command centre even in case of an incident within the communication network.

Figure 15. Command and Control Centre typical design.

The architecture of the system is based on a high reliable core data-network, partitioned in sub-network which covers all the IT&C needs. Those sub-networks are grouped in categories, as follow:

• Operators network;
• Server farm;
• Communications link;
• Wall-display and console.

The network uses a unique time reference, such a GPS time-extractor. This solution is considered the most reliable case, especially because most of the linked operators (public, private or governmental) use the same time reference, making the interface simple and more efficient. More than this, the time generator equipment is standard, easy configurable and well known by most of the technical support personnel.

The server farm is implemented by using a complex infrastructure, made by independent machines or virtualised servers – the final solution can be chosen in according with the budget, software licence policy of specific needs. The servers should provide enough computing power and storage volume in order to ensure the general capacity to manage and integrate all the processes driven from the central point.

Display walls for Control Centres usually consist of a matrix of rear projection modules. Main task of a controller is combining these projection modules to one single display wall, in other words to one giant monitor where application windows can be moved anywhere without any impact on the application itself. Hence applications can benefit from the entire resolution of the giant monitor, e.g. displaying the entire network of an energy utility, and several video windows with information.

Moreover, the controller integrates the display wall seamlessly in a modern digital IT infrastructure via standardized digital networks (LAN).

Basically a controller consists of a high performance PC based digital graphical cards for controlling the individual projection modules. As an option, additional video or RGB input cards ("frame grabbers") allow to hook up analogue signals which are displayed in moveable and sizeable windows on the display wall.

The controller and the projection modules are connected via DVI-D (Digital Video Interface), providing a pure digital configuration from LAN up to the projectors for maximum display quality and reliability.

The connection between the controller and the projection modules has to be ensured on the standardized DVI-D (Digital Video Interface).

Transportation medium is either Copper or optical cable.

All wall management functions have to be accessible via a convenient graphical user interface, running either on any operator"s workstation or on the display wall controller itself (remote mode).

Using wall-display management software, operators have to be able to switch predefined layouts on the display-wall, containing prearranged windows as previously defined or other functions.

Figure 16. Command and control centre IT architecture.

Figure 17. Wall-display architecture.

For remotely operating all functions on the display wall, remote pointer functionality is required, enabling to re-route a local keyboard and mouse of an operator workstation to the display wall.

A pointer finder application assists finding the pointer on large display walls, which otherwise could be troublesome.

Every working station has to be equipped with complete facilities for all specific activities. Thus, all working station have to be technically equivalent and be able to operate full specific jobs, no matter what operator is working on it (traffic operator, CCTV operator, supervisor, expert, maintenance technician etc.).

The main wiring structure should be mounted in the suspended floor, so as to enter simply in each cabinet. All the wires should have fire-safe insulations and halogen-free. The voice / data cables should be separated by the power supply cables in according with general accepted standards and equipment supplier specifications. Also, a special attention should be accorded to radio communication cables, which have to be separated by any other cables.

7.4. Redundancy and Back-up Solution

Through the benefits of implementing such solutions, we can present:

• Reliable telecommunication infrastructure both for voice and data, capable to work in any conditions, including damaged-field (for example in case of an earthquake);
• Command and control infrastructure, implemented in dedicated spaces which ensure well conditions for 24/7 work, with specialized operators and experts. The IT infrastructure will be based on a power-full data centre, equipped at least according with TIER-4 standards. This facility is connected to the reality in the field by using the previews communication infrastructure;
• For major incidents and disasters, there will be a second Command and Control Centre, mobile, capable to be placed in the field and ensure all the coordination activities. More than this, by using a extended communication infrastructure, the Mobile Command and Control Centre (MCCC) will be able to ensure all the operational facilities, including a virtual local radio network. More than this, the MCCC will cover the operational back-up in case of Command and Control Centre malfunction;
• Back-up and recovery solution, placed far away from the main command and control centre, which will ensure all the data-space for real time storage of all information from the main data-centre. Using specific applications, form the Back-up Centre there will be the possibility to remote all the applications so as to manage a recovery procedure in the main data-centre, in case of a damage. The communications between the mail data-centre, the mobile centre and the back-up solution will be ensured using a high-speed data link, implemented on a dedicated infrastructure, with at least one loop.

• The IT&C infrastructure has to be designed so as to be reliable enough for operating with all terminal full active, all communications channel running and all data recorded. For this the newest technological solutions has been adopted:

• Modern data-centre, designed in according with Tier-4 standard, capable to work permanently, in any external conditions. This facility will be organized in dedicated space, with air conditioning system and equipped with monitoring applications, power back-up system (with UPS and generator) and a multi-layer electronic alarm system installed for security reasons;

• Virtualized servers, installed on powerful machines, capable to ensure all processing and delivery data needs: applications servers, unified communications, storage, video streaming and wall display management applications. For compatibility reasons, keeping account that a extensive application like this imply lot of suppliers which have to collaborate, the virtualization solution will ensure 100% compatibility with all modern operating system;

• Redundant data network, capable to ensure flow and jitter parameters for all terminals (data and voice), in fully redundant conditions. The network will be real separated in at least 4 networks, organized in according with the category of operators: external access network, operator"s network, server farm network, and wall display. All those networks will be managed using dedicated tools and it will be interconnected by OSI Layer 3 equipment, security full granted;

CONCLUSION

Modern highway depend on ITS to increase the safety level. Many sensors that are used in cities may also be employed on highways with the purpose of traffic management or even specific applications like speed detection, license plate recognition, vignette check or automatic fare collection. The use of these detectors requires communications with RSU (Road Side Units) and then data transmission to a regional Command Centre. After data processing the communication loop is closed by sending the information back to the field equipment. Depending on the application, this may imply a large bandwidth and hence powerful communication systems. Other systems require short distance communications with a higher level of security. In this case a dedicated short range communications may be used, and some of them was being developed especially for road traffic applications.

All these aspects were covered in this chapter, where the most relevant devices and techniques were detailed. The data collected may also be used to inform the drivers about different aspects related to traffic. This may be done via variable message signs or by transmitting the information inside the vehicle. In this chapter an information system was presented, which wad the purpose of providing updated traffic data with the purpose of generating routes in a navigation system depending on the current road traffic status (free flow, pre-congested, congested, etc.).

All the equipment and techniques presented in this chapter are part of intelligent transport systems for highways, providing a safer journey and more information to the drivers, but also contributing to law enforcement on these high speed routes.

REFERENCES

• [1] Timnea, R.; Minea, M.; Nemţanu, F. C.; The ITS Architecture One of the Most Important Component for Planning and Developing of the Intelligent Transportation Systems and a New Approach of the Information and Communication Systems in Transports Field. International Congress CONAT 2004, Brasov, Romania, p.178, ISBN 973-635-394-X, 20-22 October 2004.
• [2] Federal Highway Administration, Traffic Management Centers The State-of-the-Practice. US Department of Transportation, Washington, DC, US.
• [3] Federal Highway Administration, Innovative Traffic Control Technology and Practice in Europe. US Department of Transportation, Washington, DC, US.
• [4] Banciu, D.; Hrin, R.; Alexandrescu, M.; Anghel, L.; Intelligent Transport Systems, Ed. Tehnica. 2003. Timnea, R.; Minea, M.; Nemţanu, F.C.; Public Transport Management Development Based on ITS Architecture. ITS 05 Prague Conference Proceedings, Prague, Czech, pp. 42-45, 2005.
• [5] Di Stefano, L.; Viarani, E.; Vehicle Detection and Tracking Using the Block Matching Algorithm Federal Highway Administration. US DoT Traffic Detector Handbook, Washington, DC, US, 2003.
• [6] Pline, J. J.; Traffic Control Devices Handbook. Institute of Transportation Engineers, 2001.
• [7] Bates, T.; Accuracy Is Everything: Perfecting the License Plate Reader. Traffic Technology International, The international Review of Advances Traffic Management, 1999.
• [8] UPB CEPETET Group; In-Time Project. Technical Materials, 2011.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 6

CYBER SECURITY EVALUATION OF CRITICAL INFRASTRUCTURES SYSTEMS

Alecsandru Patrascu* and Emil Simion†

Advanced Technologies Institute, Bucharest, Romania

ABSTRACT

Over the last 25 years we have seen a massive change in how we manage and think about information. The Internet is a rather new technology for us, but nevertheless it has become a force that is working its way into all our lives and there is no doubt in this. However it is important to realize the security issues involving it. In the 1980s the information was mostly stored and used in analogue form, as it was stored on papers, binders, shelves and safes. Today almost all information is digital. It is created and stored on individual computers and transmitted over computer networks. From the security point of view, this means that the information sent, often secret or private, can now be intercepted from anywhere over the planet. This has strong consequences because now we no longer have to be physically where the data is.

Cyber security has emerged within the latest years as an issue of vital importance for everyone as governments, businesses and individuals are under constant attack from other governments, cyber criminals and hackers. These attacks steal large amount of money, in form of personal-business data, and compromise sensitive government operations. Industrial software and equipment represents a new attack vector for our modern industry that involves both the risk of unintentional employee errors as well as aimed attacks.

Attack types such as social engineering, port scanning, packet sniffing, password cracking, denial of service attacks, machine bugs, server bugs and exploits are no longer targeted for the desktop and server environments only as they are becoming a real threat for all non-IT industries. This makes work harder even for security specialists and researchers, as these directions are new and have not been yet properly evaluated. Since almost every critical infrastructure like transportations systems, automotive factories, telecommunication, water supply and electricity generation has a computer network for decision-making and control in the background, this is vulnerable to a series of threats. This chapter is focused on these threats with accent on new forms of it, more exactly the

^* E-mail: alecsandru.patrascu@gmail.com, ati@dcti.ro.

E-mail: esimion@fmi.unibuc.ro.

issues about industrial control systems like SCADA, which are configured to control and monitor specific industrial processes, with a case study on Stuxnet. State of the art vulnerabilities like car viruses are also covered. For in-depth understanding of the concept involved in cyber security we have included some attack scenarios explained in detail.

Keywords: Cyber Security, Cryptography, Steganography, Critical Infrastructure, Industrial Control Systems, Car Viruses, Information Warfare, Smart Grid, Malware, SCADA, Transportation Network, Intelligent Transport Systems

1.INTRODUCTION

In our days, Internet Technologies optimize the time needed for taking human and organizational decisions. For example, we are using Internet for electronic communications, electronic commerce, transactions and banking, for accessing different data bases in order to process the information. All these actions, which interact with virtual activities, must be "protected" from electronic fraud. Thus, we need to implement, in the virtual space, security measures similar to real security measure. But there is a difference: behind real life thefts there are humans which interact with the goods and in the virtual space the thefts are made by viruses, worms, malware applications which interact and monitor the actives of the system. If we think that all these weapons (viruses, worms, malware applications etc.) are produced by humans we can conclude that in the protection of the system actives the human intelligence plays a definitive role and the critical decisions must be taken by humans. The problem is how we can order these decisions, which are in large number, in such a way that to be made, corrected and modified by a limited human intervention [1, 2].

In this chapter we are going to talk about a new and controversy topic, which is closely related to the problems that affected until now only traditional computers and networks – cyber security of critical infrastructures and intelligent transport systems.

Traditionally, critical infrastructures represent the field of utilities and facilities that we are using daily. These include among many other: electricity grids, gas, transport and transportation systems. More recent actors are represented by the more and more involvement of computer systems within all other classic actors that were presented earlier. And since we rely more and more on computers for proper system function we open a new window of opportunity to the hackers and all cyber warfare manifestations.

These manifestations had until recently only a single goal: to silently gather data and multiply across all vulnerable devices reachable from an infection point. This was ended in 2010 when a computer worm called Stuxnet was discovered across different critical infrastructures across the globe. We can say for sure that everything that is linked to a traditional computer network can be applied to modern intelligent transport systems, including the use of zero-days attacks, botnets and all other malware.

Along with other specific tools and techniques that are currently used in traditional computer networks we are going to present how can we use the same instruments for this specific field. In this case, the use of scenarios is the main way in which people involved in incident response can be trained. Often these scenarios are presented to the regular people in a masked form, and a good example in this field is represented by an international challenge called "Cyber MITRE", organized by the Federal Bureau of Investigation and the Fordham University. We will present in this case seven basic scenarios that can be used in real life transportation network security breach:

• 1. Identification of encrypted data in a file
• 2. Decrypted a piece of encrypted data;
• 3. Identification of steganography and revealing the hidden data;
• 4. Identification of a suspect communication between two computers and reveal the stolen data;
• 5. Identification of the incorrect usage of cryptographic algorithms usage and finding the private key used for signing.
• 6. Identification of host to host wireless communication
• 7. Interception of host to host communication

The chapter is organized as follows. Section 2 contains a brief introduction in critical infrastructures; we talk about what is a critical infrastructure from the industry point of view, what are the security problems that surround these infrastructures and we provide examples for problems and solutions in this field. In section 3 we talk about the concept of intelligent transport systems and in section 4 we talk about the concept of smart grids, a rather new topic which combines a classic approach of electric grids with the new approach of information and communication technology. Section 5 is dedicated entirely for the newer botnets and malware threats that are intended for desktops and industrial computer networks. During section 6 we talk about cyberattacks on SCADA systems; we present what are these systems, how they work and we provide real examples of attacks, like the Stuxnet malware. In section 7 we present a new and controversy domain in the security of automotive industry - the protection of the transportation networks from malware; we present the main types of these new threats, how they affect our daily lives and as a case study we talk about a subject that is gaining more and more attention to security specialists, the problem of car viruses. In section 8 we briefly talk about some of the most used tools in case of cyber security investigations and we present some practical scenarios used in cyber trainings. Section 9 is dedicated for presenting different views about critical infrastructures in different countries, like United Kingdom, Germany, and USA. Finally section 10 of the chapter contains conclusions and an outline of the directions for future research.

2. CRITICAL INFRASTRUCTURES

People and businesses today are connected using some kind of device or through the cellular network, Wi-Fi or a local area network. Nevertheless, organizations have not yet fully understood that endpoint security is vital. Saying this, we can define a critical infrastructure as a network or system which is vital for an organization, society or economy day-to-day operations. Historically [3], critical infrastructures have been gathered under terms like "utilities" and "facilities" that are commonly associated with:

• electricity, gas, oil products generation, transmission and distribution;
• telecommunication;
• water supply;
• agriculture
• food production and distribution;
• public hospitals and ambulances;
• transportation systems (like railways, airports, etc.);
• banking and other financial services;
• people security services.

The criteria that we can use to determine what asset is a critical infrastructure or not, have been changing over time. According to a report created for the USA Congress in 2002 [4], "originally were considered as critical the infrastructures whose prolonged unavailability could cause significant military and/or economic disruption. Today, the term expanded and covers even national monuments, where an attack can lead to loss of life and affect the attacked nation morale, and chemical industry, where an attack can alter the proper functioning of a factory and threaten the safety of surrounding communities".

It"s easy to see that in order to have means to deal with threats affecting these systems, the governments across the globe are working together for a multi-purpose action [5]. Nevertheless, if we find Interpol as the international agency for all criminal activity, for the cyber space there is no such agency yet, with responsibility closer to its relative. This can be found in international statements, such as the one made by Subhendu Sahu, director of Symantec Asia: "the doctrine of national security has to change. There is a need to be aware and to have the right intelligence. Security is not enough, resilience is required." This means that entire economies must know about, understand and prioritize what is important for their computer networks, especially for those parts of the critical infrastructures.

In order to protect critical infrastructures from different cyberattacks, system administrators can use many technologies and tools existing today. And since the lately emergence of central computation nodes or cloud computing technologies are used more and more, new technologies and tools must be developed if we want to migrate all or part of the computation made inside such infrastructure to them. Current technologies, like secure access control, remote system management and checking for system integrity help administrators and security expert to avoid unauthorized access to information and data that is stored, processed and transported inside such infrastructures. We can say that all these technologies should be gathered under a single cyber security framework for "Critical Infrastructure Protection", as presented in [6]. According to the authors of this document, "such a framework must include:

• 1. determining the business requirements for security;
• 2. performing risk assessments;
• 3. establishing a security policy;
• 4. implementing a cybersecurity solution that includes people, processes, and technologies to mitigate identified security risks;
• 5. continuously monitoring and managing security."

Nevertheless, we must pay attention when implementing and using such frameworks because they have great costs, which must be justified to the owners. In this entire process we must not neglect two important factors.

The first one represents the fact that each used technology has a certain limitation and due the fact that these are designed to work in pair with other technologies they must be used by professional personnel.

The second one is the lack of international or standardized solutions for these problems. When working in such environments we often state the fact that, in order to make security a first-class element, we must be able to develop and impose a long-term plan that contains such standards and all other practical solutions for the problems involved. In this part, a lot of help must be given by the government and the designated authorities for it in deploying a national plan for critical infrastructures protection and allowing agencies in charge with national security to intervene and collaborate for all the matters involving cyber-security.

3.INTELLIGENT TRANSPORT SYSTEMS

Intelligent transport systems, or ITS for short, are systems designed to provide solutions to modern transportation modes. We can see them as an automatic traffic management system that uses modern technology such as sensors and computers to make safer, coordinated and smarter decisions that affect transport systems.

Why we have need for such ITS? The problem appeared together with the exponential increase in number of cars and humans on the globe: traffic congestion. To be more efficient, computing technology was used, both on the field and in dedicated laboratories, for simulations and models. Of course, along with traffic congestion comes a lot more problems like pollution, fuel consumption and large travelling times.

Over time, the concept of ITS has become larger, covering more transportation fields than cars. For example railroads, water and air transport, and all other means of transportation for both critical and non-critical infrastructures, benefits from this technologies.

Since ITS are very complex systems, they are designed to use multiple technologies for different modules management. We can find some basic management for cars like traffic lights, number plate recognition, speed cameras, etc. up to advanced guidance and monitoring systems, like parking guidance, weather monitor, etc.

In order for these systems to communicate with each other, ITS use different communication channels such as wireless communication. Wireless communication is a good alternative to wired communication because it can cover a lot more areas with much less impact over the environments. In this type we can find technologies used for short-range or long-range communication.

In case of short-range communications we find the standard wireless IEEE 802.11 protocols. These come along with all their advantages and disadvantages. As advantages we can mention the fact that they are flexible, as with radio coverage, the nodes can communicate without many restrictions and furthermore, radio waves can pass through walls. Of course, a deployment for such technology does not require so much planning as in case of a wired network. As a downside, a major problem is given by the fact that a lot of devices and interfere with this kind of communication.

In case of long-range communications we find standards like IEEE 802.16 (WiMAX), GSM, 3G or satellite. The problem when using such technologies is the cost for the infrastructure that is going to be used and the response time for the sub-systems.

The development of ITS was marked substantially by the evolving technologies. If ten years ago a medium priced car contained several tens of chips and microcontrollers for controlling all important car functions, like braking, today"s cars contains fewer, but more powerful processing units, with dedicated memory, sensors and operating system.

We can find applications for ITS all over the world, starting from the notification systems existing on highways, up to dynamic semaphore lights and ending with those used for collision avoidance.

4. SMART GRIDS

In the context of critical infrastructure systems we consider a smart grid an evolution from the classic electrical grid. It is "smart" because it uses the latest evolvements in information technology and computer network communications in order to automate its internal work queues. This leads directly to an improved efficiency, lower maintenance costs and lower effort used for electricity distribution to end clients.

The term of "smart grids" is considered new, since it is used broadly since 2005. It was first mentioned officially in an article written by Amin and Wollenberg [7] called "Toward a Smart Grid". Nevertheless we can trace back this concept, in a primitive form, as far back as the early 2000. We can give many definitions to this term, from a functional and/or technological perspective, but all of them have the same central idea: the involvement of computation technologies to create a centralized managed electric grid.

Currently there are a lot of technologies that we can use in the context of smart grids. Technically, we can see the entire picture split in two distinct layers. The first one is composed from the technologies and physical computing resources that are used, and the second one, relative to the classical electric grid, is composed from the business management units, which are responsible with the interaction with the end clients.

The main threats surrounding these infrastructures are represented by attacks to the first layer presented in the previous paragraph. As presented in [8] by the US Council on Foreign Relations, "concerns chiefly center around the communications technology at the heart of the smart grid. Designed to allow real-time contact between utilities and meters in customers" homes and businesses, there is a very real risk that these capabilities could be exploited for criminal or even terrorist actions. One of the key capabilities of this connectivity is the ability to remotely switch off power supplies, enabling utilities to quickly and easily cease or modify supplies to customers who default on payment. This undoubtedly a massive boon for energy providers, but also raises some significant security issues".

But what are those "security issues" mentioned before? To answer this question we must first try and put ourselves in the place of a cyberattacker. The first thing he would try is to paralyze the entire activity of a chosen target. And since all their equipment is running on electric power, we find that attacking a smart grid can be very attractive. We can make a parallel with a classic war, meaning that this kind of attack can be compared with dropping an atomic bomb over a targeted region – nothing will stand in its way.

Looking back into our history we can find clear examples, like the one from the Kosovo war in the late 90s in which electric grids were key targets for fighting the enemies. This is the main reason for which these infrastructures are critical and a great deal of attention and money are spent to secure it. This also leads to a psychological war because people panic once they see they do not have electricity to power their utilities, like cooling or heating devices.

The concepts behind smart grids were always on the first page on the standardization institutes. On the "US infrastructure" website we can find some interesting paragraphs regarding our topic. In their vision [9], "over the next 10 years there will be an opportunity for ill-intentioned foreign governments to have their own agents inserted in key facilities. This is generally outside the system of national vetting and clearances in most countries. Second, it is straightforward to install malware on peoples" PCs. Social engineering, spear phishing can be used easily".

Furthermore, in the same declaration we can see that "there are other ways in which the smart grid revolution could bring unwelcome side effects. In the rush to implement security standards on a national scale a number of organizations are getting involved, including NIST, the Department of Homeland Security and even the Securities Exchange Committee. With new standards come new compliance rules, along with all their associated costs. The desire to circumvent that compliance burden is already having negative impacts.".

5. CYBER THREATS TARGETING CRITICAL INFRASTRUCTURES

Today"s critical infrastructures and all its parts, including intelligent transport systems, suffer the same weaknesses that can be found, for example, in a computer operating system. And since more and more infrastructures migrate to using operating systems and other software programs the same or derived from personal computers software, we consider this a great threat to all infrastructures and its components.

In this following subchapter we will talk about the most important and most often encountered cyber threats. We will present the concept of zero-day attacks, that stands the ground for every great cyberattack, and continue with botnets and malware.

5.1. Zero Day Attacks

A "zero-day" attack or threat is an attack that aims to exploit previously unknown vulnerabilities in software applications. The term comes for the fact that the attack occurs on "day zero" of awareness of the vulnerability and the developers of the affected application had no days available for patching it.

These attacks come in a large number. Malware, viruses and Trojans all represent attacks vectors that target modern software and delivery networks. In this equation the web browsers and the operating systems on top, which they are running, represent the most widely targets because they are widespread on all devices, starting with mobile phones and ending with desktops. Mail delivery networks are also targeted because they can carry to a potential victim an infected e-mail attachment. To cope with these threats organizations like US-CERT [2] and Zero Day Initiative [10] dedicate their work in providing users cyber security.

5.2. Botnets

A botnet is a number of Internet computers that, although their owners are unaware of it, have been mangled and have been set up to send and forward different types of transmissions, including spam or viruses, to the other computers on the Internet. The term comes from the fact that any infected computer becomes a "robot", or "bot" for short, that serves an attacker. Reports from well knows security companies like Symantec and Kaspersky Labs reach a common conclusion: botnets currently are the biggest threat to the Internet.

Computers that are used inside a botnet are those whose owners fail to provide effective firewalls or other safeguards from the Internet. Furthermore, we see that an increasing number of home computers benefit from high speed Internet connections, thus aiding the efforts of the attackers. A bot is a program attached to one of the computer ports that is left open and through this port a remote program can connect to it.

One example is the usage of a botnet to redirect HTTP traffic to another specific computer or website, in a Distributed Denial-Of-Service (DDoS) attack. The remote website will be closed down because it cannot handle all the traffic.

5.3. Malware

Malware represents the software used or created by hackers to alter computer and system operations. The goal is to gather sensitive information or to gain access to private computer systems. Its form varies from a full software program to a script. It is a general term that is used to refer to all forms of hostile and intrusive software, like computer viruses, Internet worms, Trojan horses, spyware, adware, and rootkits.

In the evolution of malware we can establish two big periods: before 2010 and after 2010. Malware before 2010 was mainly targeted to single computers or medium-sized computer networks. Since 2010 we can find the so-called "modern malware", which now has migrated from personal computers to large systems, even critical industrial systems.

We will present on the following paragraphs one of the main treats in this field, responsible with industrial systems malfunction called Flame.

Flame, also known as Flamer or sKyWiper [11] is a computer malware that attacks computer and industrial systems running Microsoft Windows operating system. It was used for regular and critical infrastructure penetration in several countries around the world [12].

This malware is capable of spreading to other systems over a local area network (LAN) or through USB sticks. It can record audio, screenshots, keyboard activity and network traffic. Furthermore it can record instant messaging conversations and can turn mobile infected computers and auxiliary wireless modules, into Bluetooth beacons which attempt to connect and download information from other Bluetooth devices nearby. This data was then sent to one of the several master servers existing in the Internet.

What is interesting on this malware is its built-in capability to "suicide". The creators of Flame implemented a "kill" command that automatically removed itself from the infected computers. This command was caught by the Symantec Security Company using computers set up to watch the malware"s actions. More exactly, when it received that command, Flame located every file existing on the victim PC, deleted it and then overwrote its memory location with random data to prevent a forensic examination.

According to cryptographic experts [13] [14], Flame was the first malware to use a rather obscure cryptographic technique called "prefix collision attack". This allowed it to fake digital credentials that had helped it to spread. The exact method for this kind of attack was demonstrated in 2008, but the creators of Flame implemented their own variant. This determined Marc Stevens to state that "the design of this new variant required world-class cryptanalysis". All these findings give support to claims that Flame must have been built by a nation state rather than cybercriminals due to the large amount of time, effort and resources that have been put into its creation.

6. CYBERATTACKS ON SCADA SYSTEMS

Supervisory Control and Data Acquisition (SCADA) systems, created by Siemens, represent an industrial large scale control system used to automate different industrial processes existing in facilities such as power generation, water supply, etc. One of its main advantages is that this technology can be used in more delicate fields, where precision is vital, such as nuclear fusion facilities. Their main purpose is to monitor and control all the operation inside such facility using a large network of sensors and computers for central decision making. Since SCADA systems are designed to be used as a control system, they must be complex and safe. This is one of the main strength points for this system, as it can manage up to million subsystems input and output channels.

As a brief description, we will present the main components of a SCADA system. The main parts are represented by the signaling hardware for input and output process lines, the network between subsystems, the control devices, the user interface to the subsystems (Human-Machine Interface, or HMI), communication equipment and modules, and, of course, the operating system and all the software for the modules and subsystems. The measurement and control modules of a SCADA system has one so-called Master Terminal Unit (MTU), which is considered the central management node. The chosen operating system for it is Microsoft Windows. This MTU handles a series of Remote Terminal Units (RTU), which are responsible with local data gathering.

In this current form, SCADA systems use open-loop control systems with a series of some closed-loop characteristics built in. Because of this, SCADA systems cannot use feedback from remote modules to check the final results. In other words, in SCADA we cannot use machine-learning software and algorithms.

Because SCADA is a complex system that can be used with success even in critical infrastructures, it represented an attraction to cybercriminals. In the following subchapter we will talk about Stuxnet, a computer worm that targeted Siemens industrial software and equipment running on Microsoft Windows. The analysis that follows represents a compilation of information that has been gathered from different international sources [15, 16] regarding Stuxnet.

Stuxnet is a computer worm discovered in June 2010 and it is the first public menace that targeted Siemens industrial control systems – SCADA [15]. It used advanced techniques for spreading by taking advantage of several Microsoft Windows vulnerabilities in the USB stack. Also, it is the first malware that used malicious STL code (Statement List) and the first to use a PLC (programmable logic controller) rootkit.

Researchers at Symantec have cracked Stuxnet cryptographic system, and they say it is the first worm built not only to spy on industrial systems, but also to reprogram them. Once installed on a PC, Stuxnet uses Siemens default passwords to seek out and try to gain access to systems that run the WinCC and PCS 7 programs, the PLCs that are used to manage largescale industrial systems on factory floors and in military installations and chemical and power plants. The software operates in two stages following infection. First it uploads configuration information about the Siemens system to a command-and-control server. Then the attackers are able to pick a target and actually reprogram the way it works.

Based on information in the code, Stuxnet was targeting a facility that had 33 or more of the frequency converter drives installed, all operating at between 807Hz and 1210Hz. The malware would sit quietly on the system doing reconnaissance for about two weeks, then launch its attack swiftly and quietly, increasing the frequency of the converters to 1,410Hz for 15 minutes, before restoring them to a normal frequency of 1,064Hz. The frequency would remain at this level for 27 days, before Stuxnet would kick in again and drop the frequency down to 2Hz for 50 minutes.

The drives would remain untouched for another 27 days, before Stuxnet would attack again with the same sequence. The extreme range of frequencies suggested Stuxnet was trying to destroy whatever was on the other end of the converters.

7. PROTECTION OF TRANSPORTATION NETWORKS

The area of intelligent transport systems and intelligent cars is becoming more and more a reality surrounding us. And since we rely more on more on technology to make our daily tasks easier, we tend to forget about security involving these kind of systems.

The topic of protection transportation network and attacks that can be made to a car is a new and emerging field of research and more and more scientists are joining every day. This is important because, as we have seen until now, securing critical infrastructures require mixing many of the technologies existing for regular computer or networks.

We can use a brief introduction in this topic from Zhang and Levinson"s document, "Investing for reliability and security in transportation networks" [30]. They consider that "transportation networks support various vital human activities, and their performance affects the efficiency of virtually all economic transactions. Decision makers need to consider the potential trade-off between network efficiency and reliability when they forge policies that could shape future transportation networks."

Checkoway, Mc Coy and Kantor present in their paper entitled "Comprehensive experimental analyses of automotive attack surface"[17], how a modern car is built, what electronics does it have, and most important, how can someone hack into them. According to them, "a modern automobile is controlled by tens of distinct computers physically interconnected with each other via internal (wired) buses and thus exposed to one another. A non-trivial number of these components are also externally accessible via a variety of I/O interfaces."

Current automotive systems can be considered as a network of embedded systems, like Electronic Control Units (ECUs) and sensors, which are connected to each other via different bus systems and thus, exhibit a considerable complexity. Moreover, in the recent years, automotive systems became more and more software intensive systems. This is caused by the fact that more and more functionality of a car is implemented by software functions.

Of course we must not neglect the software modules existing in these embedded systems. In this direction we must agree with Schulze et al. in their work "IT security in automotive software development"[18] saying that "in order to develop software in an efficient way and to overcome the still increasing complexity anyway, techniques and concepts from the field of software engineering (SE) and requirements engineering (RE) have been proposed and applied to the development process. Although the software should contribute, amongst others, to the reliability of the system, one aspect has been neglected so far in the development process: holistic concepts for IT security. Successful attacks on an automotive IT system can have negative implications on the safety of its human users or on the reliability of the system itself. However, today it is common that the mostly generated code is retrofitted at the end of the development process in order to satisfy security concerns. This, in turn, counteracts the effort of reducing complexity by using modern concepts of SE and RE. Furthermore, this process is prone to simply ignore serious vulnerabilities and thus, allows for security leaks remaining in the system".

This is the main reason why we must not concentrate only on hardware, and we must pay more attention to the particular software engineering problems that appear when developing for cars. As Shulze continues, "along with the mentioned fundamental changes within automotive systems, the IT security becomes an important issue. Per definition, IT security means reliability in terms of preserving security aspects of information, namely integrity, availability, authenticity, non-repudiability, confidentiality and privacy. Because of its networked character, automotive systems exhibit vulnerabilities to malicious attacks, which, in turn, can violate one or more aspects of IT security. The access for the execution of an attack on the system can take place in multiple forms from outside or inside the car. Regardless of how access is achieved by the user, the basic attack principle is always the same: it aims at influencing a certain behavior or state of the automotive system. Since this is done by manipulating the respective functionality (e.g., by introducing malicious code, communication or manipulating data), it directly addresses the software responsible for this functionality. Subsequently, it is reasonable to ensure the IT security of the respective software in order to increase the security of the overall system".

7.1. Car Viruses

In the last years there was a lot of debate regarding the concept of car viruses. The reason is because, as we can see in [19, 20], in the past, car viruses were rare because one of the only ways to infect a vehicle was by a mechanic and via the computer or software he used to diagnose problems with the car.

A good definition of this concept can be extracted from [21]. In this article we are introduced to the idea of car viruses and what they require. As the authors stated, "modern cars are full of computers and they seem more the realm of an IT expert than a regular car mechanic. And it certainly is true that modern vehicles have plenty of computers, although they"re not exactly like PCs. "Cars have much simpler processors than a home computer and are designed to do simple, dedicated tasks," says Cameron Camp, a researcher at ESET, a technology security company.

Indeed, most cars today have numerous embedded systems, which are small computers controlling very specific aspects of the car"s functioning, such as air bag deployment, cruise control, anti-lock braking systems and power seating. While these embedded systems share the same architecture as a PC - they utilize hardware, software, memory and a processor they are more akin to a smartphone in sophistication than a laptop. Automotive computers have been more or less immune to hackers and viruses because, unlike PCs, there have been few ways for outside computers or people to connect with vehicle computers.

Not surprisingly, automakers are said to be working on ways to prevent hackers from introducing viruses into cars and otherwise making mischief, although details of their efforts are not readily available. Still, opinions are mixed about how much of a concern this really is for future drivers".

7.2. Protecting the Transportation Network from other Types of Software Attacks. Solutions

Schulze et al. also present in [18] the connection between automotive systems and software. "As already stated, automotive systems more and more rely on software to fulfill certain functionalities. Furthermore, the complexity of such systems steadily increases, while the reliability has to be ensured. Altogether, this is a challenging task to be managed during the software development process of such systems. Hence, different approaches of software development found their way into the systems engineering domain in order to overcome these challenges. For instance, software product lines (SPL) as a special concept of software engineering are used to manage commonalities and variability of automotive software. Another common practice is model-based development (MBD) of software for automotive systems, where functionality is described by models. Afterwards, the code is generated automatically based on these models. Finally, requirements engineering gains more and more importance within automotive systems since a good requirements analysis is inevitable for all other stages of the development process".

Also as can we see in [17],"past work has illuminated specific classes of threats to automotive systems, such as the technical security properties of their internal networks; we believe that it is critical for future work to place specific threats and defenses in the context of the entire automotive platform. Technical capabilities describe the assumptions concerning what the adversary knows about its target vehicles as well as her ability to analyze these systems to develop malicious inputs for various I/O channels. For example, we assume that the adversary has access to an instance of technical capabilities that describe our assumptions concerning about what the adversary knows about its target vehicles as well as his ability to analyze these systems to develop malicious inputs for various I/O channels.".

In the following paragraphs we have presented also some interesting parts from Checkoway"s et al. [17] research regarding the vulnerabilities surrounding a car along with means to physical or wireless access it.

"Modern automobiles provide several physical interfaces that either directly or indirectly access the car"s internal networks. We consider the full physical attack surface here, under the constraint that the adversary may not directly access these physical interfaces herself but must instead work through some intermediary.

A. OBD-II. The most significant automotive interface is the OBD-II port, federally mandated in the U.S., which typically provides direct access to the automobile"s key CAN buses and can provide sufficient access to compromise the full range of automotive systems. While our threat model forbids the adversary from direct access herself, we note that the OBD-II port is commonly accessed by service personnel during routine maintenance for both diagnostics and ECU programming.

In both situations Windows-based computers directly or indirectly control the data to be sent to the automobile. Thus, if an adversary were able to compromise such systems at the dealership she could amplify this access to attack any cars under service.

B. Entertainment: Disc, USB and iPod. The other important classes of physical interfaces are focused on entertainment systems. Virtually all automobiles shipped today provide a CD player able to interpret a wide variety of audio formats like raw "Red Book" audio, MP3, WMA, and so on. Similarly, vehicle manufacturers also provide some kind of external digital multimedia port, typically either a USB port or an iPod/iPhone docking port, for allowing users to control their car"s media system using their personal audio player or phone. Some manufacturers have widened this interface further; BMW and Mini recently announced their support for "iPod Out," a scheme whereby Apple media devices will be able to control the display on the car"s console."

8. TOOLS AND SCENARIOS THAT HELP INVESTIGATING INTELLIGENT TRANSPORT SYSTEMS INCIDENTS

Attacks on intelligent transport systems can be handled using tools and methods common to the field of classic computer networks and regular computing systems. As stated in the first section, it is common in this field to use of scenarios. These scenarios represent the main way in which people involved in incident response can be trained. Often these scenarios are presented to the regular people in a masked form. This means that an expert handling a modern transport system can use in case of incident handling and response the same tools as an expert in computer networks.

In this part we will present some of these tools, along with their capabilities. Also, in order to be more relevant to the reader, we present these tools in action, applied to the "Cyber MITRE" international challenge mentioned before.

8.1. Tools

In general, the tools used in incident response tools are designed as a general application, which can have multiple other uses. We are going to briefly talk about three main representative categories: stand-alone tools, statistical tools and security oriented distributions.

8.1.1. Stand-Alone Tools

Stand-alone tools represent tools that can be used independently over an operating system. Representative to this field is CrypTool. This is a software package dedicated to cryptographic simulation, analysis and cracking which has a user graphic interface.

CrypTool has been developed in cooperation with prestigious universities and thus has become excellent educational software and also a tool for learning cryptology. CrypTool covers both branches of cryptology: cryptography and cryptanalysis. Thus, the product has implement facilities of each field, such as classic cryptography (Caesar and Vigenère ciphers, mono-alphabetic substitution, etc.), symmetric cryptography (IDEA, RC2, AES, etc.), asymmetric cryptography (RSA and elliptic curves, etc.), hash functions (MD2, MD5, SHA-1, etc.), cypher text attacks, plaintext attacks, adaptive attacks, side channel attacks.

8.1.2. Statistical Tools

In order to test the degree of randomness for input or output for such transportation systems, we need a set of different tools, tests and theoretical models – we need statistical tools. Using this kind of tools we can test for true randomness of functions that are part of the software implementation of these systems. Good examples in this direction are:

• "A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications" [22], is a publication of sixteen statistical tests. The authors provide also an implementation for it.
• "The Art of Computer Programming, Seminumerical algorithms, Volume 2", by Donald Knuth [23] contains the theoretical description for some of these tools, that are based on permutations, birthday spacing, serial correlations, etc.
• The Crypt-XS suite developed by the researchers from Queensland University of Technology, Australia. This suite contains the implementation for some tools described by Knuth, along with other custom ones for binary derivative, sequence and linear complexity measurement, etc.
• The DIEHARD suite [24] developed by George Marsaglia adds to the tools mentioned before, tests such as random spheres, overlapping sums, etc.

8.1.3. Security Oriented Distributions

Penetration testing [25] and security auditing are now part of every system administrator"s "other duties as assigned". In this direction, BackTrack Linux (BTL) or its newer version, Kali [26], comes to help. BTL is a custom distribution build on top of Debian Linux, designed for security testing for all skill levels from novice to expert. It is the largest collection of wireless hacking, server exploiting, web application assessing and socialengineering tools available in a single Linux distribution. It contains more than 300 security tools and utilities that are all open source, grouped in major categories like: information gathering, vulnerability assessment, exploitation tools, privilege escalation, maintaining access, reverse engineering, RFID tools, stress testing, forensics, reporting tools, services.

8.2. Practical Scenarios

Of course, to use such tool, the incident response experts must know how to use them, and especially what tool fits better in a context. In order to cross this border different sets of tests have been created. Next we will present five of the most important and most used one, that have also a great impact on today"s intelligent transport systems: finding what kind of encryption is used over a system, identifying what data is leaked from or into our system computer network, identifying and intercepting eventually live transmissions between an infected host from our network and its malicious command and control servers over the Internet, identifying and recognizing fake or mangled signatures keys for public certificate access and finally, in case of using large transport system deployments with many nodes and a lot of intermediary wireless communication, identifying the weakest points that can be used by attacker to gain unauthorized access to the system.

8.2.1. Identification of Encryption Systems

In this scenario the investigator finds a suspect file on the system or computer. This file represents an encrypted data with a classical encryption system. We need to recognize encryption system, decrypt the data and find the password hidden in the encrypted file. The investigator will need this password in solving the second task.

To solve this scenario we will use CrypTool. If we take a look at the encrypted file we see that this file contains only 26 characters A to Z. If we perform a statistics of these letters we see that the characters A-Z appears to be random. Thus, we can think that there is a classic encryption such as substitution (Playfair, Caeser, Vigenère etc.). Using cipher text only attack on a Vigenère cipher we find the encryption password <>. If we take a look at the end of the decrypted file we find the text THEPASSWORDFORTOMORROW ISSTRONGPASSWORDSAREGOOD. Thus, we conclude that the password we are looking for is: STRONGPASSWORDSAREGOOD.

8.2.2. Identification of Hidden Data inside other Files

Over the investigated system network have been intercepted some images. One of these images has a huge size reported to its format. We need to find the data founded in the image and decrypt it if necessary.

We start by analyzing the intercepted image. It is in gif format but has a large size, approximately 13 MB. If we investigate this image with a hex editor like UltraEdit we can see at the end of the file a zip file header (PK). Thus, if we exchange the extension of gif file to zip file and open it with a supported archiver like WinZip we find an encrypted archive. After cracking the password, we find three gif files. Opening each file we can see that one of these one has the content: hollenger.dll. Thus, the malicious file name that the attacker is trying to access is hollenger.dll.

8.2.3. Identifying a Suspect Communication between Two Computers and Reveal Stolen Data

By analyzing our system network we find regular traffic between a node and another host over Internet. The local administrator gives to the investigator a traffic capture between the node and an outside unknown source. The task is to find what information is stolen.

Using the Wireshark application we are opening the target file. This tool is used for network traffic analysis or any other general network troubleshooting. Using Follow TCP stream option we locate within the capture a file found on the source node and dump it into a file. Opening the dumped file with a hex editor like UltraEdit we see the zip file header (PK); thus we change the extension of the dumped file into zip and open it. This archive contains a file and after a visual inspection with UltraEdit we find that it contains the magic header GIF. Thus, if we exchange the extension of the file into gif and open it with an image viewer we can see an image which has the content The Root Password is Pengu1nsRock.

8.2.4. Identifying and Recovering Mangled Signature Keys

In this scenario we are given two ECDSA (Elliptic Curve Digital Signature Algorithm) signatures. After a close inspection we find out that something about them looks strange. Using the known public key and its public parameters we must find a way to recover the private key used to generate the signatures.

Before presenting how an investigator can resolve the scenario, we will present briefly the concept behind ECDSA. Basically, ECDSA is a variant of DSA algorithms that use elliptic curve cryptography in order to be more reliable. The public parameters are the prime number p, an elliptic curve E[Fp], a point G E[Fp] with ord(G)=q, q prime number. The public key VE[Fp] is derived from the signing key 1≤d≤q-1: V=dG. The signature of the hash h is computed using the ephemeral key k mod q is the pair (r,s)=(xkG mod q,(h+dr)k -1 mod q), where xeG is the first component of the point eGE[Fp]. To verify the signature (r,s) of the hash h we need to check if *xv*1G+*v2^V* mod q=r, where *v*1=hs-1 mod q and *v*2=rs -1 mod q. It is essential to have for different signatures (*r*1,*s*1) and (*r*2,*s*2) different ephemeral keys *k*1≠*k*2. If this two keys are equal then the signatures of the two hashes looks like (r,*s*1) and (r,*s*2). Thus, we can derive *s*1-*s*2=k -1 (*h*1-*h*2) mod q and find the ephemeral key k=(*h*1-*h*2)(*s*1-*s*2) -1mod q. Since *s*1=k -1 (h+dr) mod q we derive the private key d=(s1k-h1)r -1 mod q.

The investigator receives three files. The first file contains the hash, in hex codification, of two messages *h*1, h² and their ECDSA signatures (*r*1,*s*1) respectively (*r*2,*s*2) :

h1=DE37B3145DB7359A0ACC13F0A4AFBD67EB496903
r1=ACB2C1F5898E7578A8A861BDF1CA39E7EF41EAC0B6AAA49468DD70E2
s1=BE4FA99C9D261C5F387A3ACE025702F6FB7884DD07CE18CAD48654B8
h2=28469B02BF0D2CFC86FF43CB612EE8FC05A5DBAA
r2=ACB2C1F5898E7578A8A861BDF1CA39E7EF41EAC0B6AAA49468DD70E2
s2=D3540E2B13E51605F5FEB8C87EE8E176E59213F31EA8B8FFDAD077E2

The second file, parameters.der, contains, in der codification, the public parameters of the EC. This file can be interpreted using OpenSSL:

opensslecparam -inform DER -in /cygdrive/e/parameters.der -outformPEM -out /cygdrive/e/parameters.pem

opensslecparam -text -in /cygdrive/e/parameters.pem -noout

Field Type: prime-field

Prime:

00:ff:ff:ff:ff:ff:ff:ff:ff:ff:ff:ff:ff:ff:ff:

ff:ff:ff:ff:ff:ff:ff:ff:ff:fe:ff:ff:e5:6d

A: 0

B: 5 (0x5)

Generator (uncompressed):

04:a1:45:5b:33:4d:f0:99:df:30:fc:28:a1:69:a4:

67:e9:e4:70:75:a9:0f:7e:65:0e:b6:b7:a4:5c:7e:

08:9f:ed:7f:ba:34:42:82:ca:fb:d6:f7:e3:19:f7:

c0:b0:bd:59:e2:ca:4b:db:55:6d:61:a5

Order:

01:00:00:00:00:00:00:00:00:00:00:00:00:00:01:

dc:e8:d2:ec:61:84:ca:f0:a9:71:76:9f:b1:f7

Cofactor: 1 (0x1)

The third file, public.oct, contains the public key:

X^V =85CEEE9C98EFDFDFCF64CB522A773F1435D568173677D1D28FC00643 Y^V =58A105CC1AB1A53D77B278850776E144197F3FA4E27AA676408DFE22

At this point, because the two signatures collide on the first half, we have all the elements to finalize the investigation. The only thing we need to do is to compute the private key using the formula:

$$d=(s_1k-h_1)r^{-1} \bmod q,$$

where k=(*h*1-*h*2)(*s*1-*s*2) -1mod q. We can perform these computations using, for example, MAPLE:

*h*1:=convert("DE37B3145DB7359A0ACC13F0A4AFBD67EB496903",decimal, hex);

*h*2:=convert("28469B02BF0D2CFC86FF43CB612EE8FC05A5DBAA",decimal, hex);

r:=convert("ACB2C1F5898E7578A8A861BDF1CA39E7EF41EAC0B6AAA49468DD7 0E2",decimal,hex);

*s*1:=convert("BE4FA99C9D261C5F387A3ACE025702F6FB7884DD07CE18CAD48654 B8",decimal,hex);

s2:=convert("D3540E2B13E51605F5FEB8C87EE8E176E59213F31EA8B8FFDAD077E 2",decimal,hex);

q:=convert("010000000000000000000000000001DCE8D2EC6184CAF0A971769FB1F 7",decimal,hex);

d:=(-h2*s1+h1*s2)*(r*(s1-s2))^(-1) mod q;
convert(d,hex,decimal);

After the compilation of the program we find the private key:

d=8E88B0433C87D1269173487795C81553AD819A1123AE54854B3C0DA7

8.2.5. Identification of Insecure Wireless Connection Points

In this scenario the investigators are trying to find what wireless access points are unsafe in public use and they are trying to gain access to a private Wi-Fi network that is secured using the WPA (Wi-Fi Protected Access) protocol. WPA is a security protocol and security certification program developed by the Wi-Fi Alliance. It is known as the IEEE 802.11i standard. More exactly a Temporal Key Integrity Protocol (TKIP) is used, that involves using a dynamic 128 bit key for every packet transmitted. The newest version, WPA2 also includes Counter Cipher Mode with Block Chaining Message Authentication Code Protocol (CCMP), a new AES based encryption mode with strong security.

In order to solve this scenario, the investigators will use a series of intercept nodes that will have a single goal: to continuously scan the entire wireless networks available and try to crack their password. For this, all the intercept nodes will run two tools called Reaver and Airmon-ng.

Reaver implements a brute force attack against Wi-Fi Protected Setup (WPS) registrar PINs in order to recover WPA/WPA2 passphrases. The algorithm behind this tool is described in detail in http://sviehb.files.wordpress.com/2011/12/viehboeck_wps.pdf. Currently there is no wide-spread solution for this and it affects even big network vendors like Linksys and Cisco.WPS is a feature that exists on many modern routers and its intention is to provide to the users an easy setup process. The problem is that it is tied to a PIN that is hard-coded into the device.

The first step is to set our node into a monitor mode using the command airmon-ng start wlan0. After this, we need to find the BSSID of the target node that we want to test. The BSSID is a unique series of letters and numbers that identifies a target router. We find it by using the command airodump-ngwlan0. In the list shown inside the terminal we copy the one for our network. We will assume the following made up BSSID: 8D:AE:9D:65:1F:B2.Now, with the BSSID and the monitor interface name in hand we have everything we need to startup Reaver. Inside a terminal we issue the command reaver –imon0 –b 8D:AE:9D:65:1F:B2 –vv After this, Reaver will try a series of PINs on the router in a brute force attack, one after another. After cracking is complete we will have the output of the total time needed to crack the password, the router PIN and its password.

9. PRACTICAL ISSUES AND EXPERIENCES IN DIFFERENT COUNTRIES

As we have seen until now, the field of critical infrastructure protection, including the most important of them - water, gas, and electricity providers, and especially the area of intelligent transport systems, is very important for a country, mainly for those that have strong industries and big population. As we can see in national reports such as those from United Kingdom [27] or Germany [28 - 29] these infrastructure represent important assets.

On one side we have the technological requirements and on the other side we have the public opinion in case of cybercrime, such as banking fraud, identity and content theft. A warning signal was triggered with the discovery of Stuxnet malware as it offered the level of progress made by the cybercriminals. We can say for sure that every country in this world must pay attention to all its infrastructures and "do something" about cyber security.

For a better image we have briefly presented the situation existing in three different countries around the world. They have been chosen because they are representative for their power and importance.

9.1. United Kingdom

In the United Kingdom we can find the National Security Strategy (NSS) and the Strategic Defense and Security Review (SDSR) documents that specify the need for critical infrastructures protection. These documents were released in 2010, and we can see how in the late years this problem has become more and more important. This is also supported by the fact that their annual budget for cyber security is around 650 million pounds.

In charge with applying these documents is the Center for Protection of Critical National Infrastructure (CPCNI) and since their publishing, it has implemented many incident response services, such as energy, economies, health, etc.

9.2. Germany

In Germany we can see a good example of a secure critical infrastructure deployment. They have all their important infrastructures connected through a full mesh network. This means that any failure is covered thoroughly. As examples we can give health, industries, water supplies, IT and food supplies.

This implementation was possible thanks to the mature legal environment existing in this country that threats security as a vital element. For example, we can analyze their power distribution infrastructures; both public and private companies are obligated by the law to have an internal failure and recovery system.

Of course all these implementation are monitored constantly by different enforcers, such as "Energiewirtschaftsgesetz" – the Act on the Supply of Electricity and Gas and the "Bundesnetzagentur" – the Federal Network Agency in the field of smart grids and power distribution. Telecom services also have their own procedures and laws regarding protection from network penetration, unknown call interception and any other form of unauthorized access.

9.3. USA

In the USA, a lot of effort has been put in order to secure their critical infrastructures since the terrorist attacks from 11th September 2001. Since then, an unauthorized access to this kind of infrastructure was never spotted.

In charge with this field is the Department of Homeland Security (DHS). It has the means to coordinate the cyber activity to all these systems and it also can use the help from the other governmental agencies in case something happens.

The entire plan can be found in the USA "Critical Infrastructure Protection Plan" which enforces each agency with a specific task. For them, these infrastructures represent a top priority.

CONCLUSION

As we can see from this chapter the problem of critical infrastructures represents a current and menacing threat, which involves strong knowledge of computer communications techniques, secure programming techniques, algorithm and software implementation analysis, cryptography, steganography, probability and finally applied mathematics. The process of high level of assurance of cyber security must take into account all the above-specified domains.

Fast advances in cybercrime technology and techniques have resulted since the beginning of 2012 in an unprecedented rise in data breaches. We think that planning to ensure that our critical infrastructures and intelligent transport systems are trustworthy and secure we need to consider the fundamental changes that are occurring in the cyberspace and try to adapt to them. In our opinion, looking forward into the future of more than 3 billion Internet users existing today we can see four big directions for resolving the cyber security issues: online users security education, cryptography, online data obfuscation and cloud services transparency and security.

REFERENCES

• [1] Hough, P.; Understanding Global Security. Routledge, 2009.
• [2] www.us-cert.gov/.
• [3] www.expresscomputeronline.com/index.php/sections/events/796-critical-infrastructureprotecting-the-heartbeat-of-the-nation.
• [4] Critical Infrastructures: What Makes an Infrastructure Critical?. Report for Congress, 2003, http://fpc.state.gov/documents/organization/13839.pdf.
• [5] Kramer, F.D.; Starr, S.H.; Wentz, L.; Cyberpower and National Security. Potomac Books Inc., 2009.
• [6] Cybersecurity for Critical Infrastructure Protection. United States General Accounting Office, www.gao.gov/new.items/d04321.pdf.
• [7] Massoud Amin, S.; Wollenberg, B.F.; Toward a Smart Grid. IEEE P&E Magazine, 2005.
• [8] U.S. Infrastructure: Smart Grid, Renewing America. Council on Foreign Relations, 16.
• [9] www.americainfra.com/article/Smart-grid-security/.
• [10] www.zerodayinitiative.com/.
• [11] www.crysys.hu/skywiper/skywiper.pdf.
• [12] www.certcc.ir/index.php?name=news&file=article&sid=1894&newlang=eng.
• [13] www.securelist.com/en/blog/208193522/The_Flame_Questions_and_Answers.
• [14] www.bbc.co.uk/news/technology-18365844.

• [15] www.codeproject.com/Articles/246545/Stuxnet-Malware-Analysis-Paper.

• [16] Zhang, L.; Levinson, D.; Investing for Reliability and Security in Transportation Networks. 2008.

• [17] Checkoway, S.; McCoy, D.; Kantor, B.; Comprehensive Experimental Analyses of Automotive Attack Surface, Autosec, 2011.

• [18] Schulze, S.; Pukall, M.; Hoppe, T.; *ITSecurity in Automotive Software Development.*GI Softwaretechnik-Trends, 2009.

• [19] Clayton, M; Scientists Hack into Car Computers - Control Brakes Engine, The Christian Science Monitor, www.csmonitor.com/USA/2010/0813/Scientists-hack-intocars-computers-control-brakes-engine.

• [20] Goretsky, A.; Researcher at ESET:A Technology Security Company. Personal correspondence, 2011.

• [21] http://auto.howstuffworks.com/car-computer-virus.htm.

• [22] A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications. 800-22, NIST Special Publication, 2010.

• [23] Knuth, D; *The Art of Computer Programming, Seminumerical Algorithms.*3rd Ed., vol. 2, Addison Wesley, Reading, Massachusetts, 1998.

• [24] Marsaglia, G.; DIEHARD Statistical Tests. http://stst.fsu.edu/geo/diehard.html.

• [25] Engebretson, P.; The Basics of Hacking and Penetration Testing. Syngress, 2012.

• [26] www.backtrack-linux.org/.

• [27] www.chathamhouse.org/sites/default/files/public/Research/International% 20Security/r0911cyber.pdf.

• [28] www.bmi.bund.de/SharedDocs/Downloads/EN/Broschueren/cip_stategy.pdf?__ blob=publicationFile.

• [29] ww.bsi.bund.de/SharedDocs/Downloads/EN/BSI/Kritis/paper_studie_en_pdf. pdf?__blob=publicationFile.

• [30] www.pcworld.idg.com.au/article/360645/siemens_stuxnet_worm_hit_industrial_ systems/.

• [31] Hills, R.; Senior Education Program Manager at Universal Technical Institute. Personal Correspondence,2011.

• [32] Mills, E.; *Hacking a Car (Q&A).*CNET, http://news.cnet.com/8301-27080_3- 20005047-245.html.

• [33] Mollien, C.; Information and Communication Technology Strategist at Basic Blue. Personal Correspondence, September 14, 2011.

• [34] Petraglia, D.; Director of Forensic and Information Security Services at Chartstone Consulting. Personal Correspondence, 2011.

• [35] Shaer, M.; Disgruntled Hacker Remotely Disables 100 Cars. The Christian Science Monitor, www.csmonitor.com/Innovation/Horizons/2010/0318/Disgruntled-hackerremotely-disables-100-cars.

• [36] www.cfr.org/united-states/us-infrastructure-smart-grid-security/p26842.

• [37] http://wwwiti.cs.uni-magdeburg.de/~sanschul/papers/GI-WS_regensburg2009.pdf.

• [38] www.autosec.org/pubs/cars-usenixsec2011.pdf.

• [39] http://nastfenews.org/index.php?option=com_content&view=article&id=107:makingautomotive-software-safe&catid=27:features&Itemid=49.

• [40] www.isticom.it/documenti/news/pub_003_eng.pdf.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 7

SUSTAINABLE INFORMATION SYSTEMS MANAGEMENT FOR SPATIAL GOVERNANCE

Angela Ionita*¹ , Marcel Foca†² and Maria Visan‡²

¹Research Institute for Artificial Intelligence, "MihaiDrăgănescu" of Romanian Academy, Romania 2 Intergraph Computer Services S.R.L.

ABSTRACT

This chapter has the intention to discuss the complexity of sustainable information systems management in the framework of spatial governance. There is a diversity of motivations to introduce sustainability into Information Systems based on Information Technology and Communication. The discussion about sustainability and Information Systems based on Information Technology and Communication innovation appears in different contexts. Based on the current literature, this article is a synthesis and proposes a reorientation of the way the concept of sustainability is dealt with, positioning spatial governance based on human behavior and the processing of spatial knowledge at the centre of the concept. The sustainability approach presented in this chapter is then translated into requirements needed for a Sustainable Information Systems Management for Spatial Governance focusing on (Cooperative) Intelligent Transportation Systems, Spatial Data Infrastructure and emphasizing Spatial Governance.

Keywords: Sustainability, Spatial Governance, Intelligent Transportation Systems, Cooperative Intelligent Transportation Systems, Spatial Data Infrastructure

^* E-mail: aionita@racai.ro.

^† E-mail: office@ingr.ro.

E-mail: office@ingr.ro.

1.INTRODUCTION

"Moving forward with sustainability is no more a subject matter of intellectual discussion. Instead it has become a harsh reality in evolution of human civilization. However, scope and definition of sustainability has constantly expanded manifold to include essential utilities and requirements of day to day life. Inherent consumerism of growth/development has constantly evolved luxuries into commodities causing incremental consumption per capita" [1]. Sustainability is in many cases, but not correctly, associated with the environmental aspects only, while corporate social responsibility is more often perceived as primarily social-oriented. Additional confusion is created by the various social and environmental-focused initiatives related to climate change or fair trade. Current literature [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24) acknowledges that sustainability is a broad, complex concept [25] involving environmental as well as social issues, and which requires continuous learning in order to be understood and tackled. However all of them are about vision for the future.

Rio+20, a joint project of the entire United Nations System, is engaging all constituencies and programs of the global governance body to encourage governments to adopt measures for implementing sustainable practices developed over the past 20 years [26]. The global economy and economic progress paradigm as in the late 20th century was completely revolutionized with the development of the personal computer and rise of the Internet. Analytics and metric tools enabled by Information Technology and Communications (IT&C) now allows for better and more efficient reporting for all sorts of fields - including sustainable development. As Daniel Kent mentioned [26]: "by investing further in information and environmental technologies, it is possible for some countries to become exporters of clean technologies and thus turn the environment into a new engine for competitiveness and growth, rather than a cost as currently perceived. Embedding environmental perspectives into social and economic policy requires a more holistic and cross-sectoral way of working, strengthening local to global interactions and improving scientific understanding. In this respect, greater transparency about assumptions, shared access to data and better communication of scientific findings will become critical to gathering and retaining the support of society and policy-makers".

The discussion about sustainability and Information Systems (IS) based on Information Technology and Communications (IT&C) innovation appears in different contexts. For instance, models and tools have been developed to assess corporate sustainability [27] and sustainability of Management Information Systems [28]). The notion of sustainability of IS based on IT&C innovation (IS IT&C) stems from the broader notion of Sustainable Development (SD) and is applied in a specific domain. Also, contributions concerning sustainability and IS IT&C originate from joining domains, sustainability and systems [29, 30].

2. SPATIAL GOVERNANCE AS AN ASPECT OF GOVERNANCE

'Governance' relates to human organization of various kinds: government and nongovernment, commercial and non-commercial, private and public, formal and informal. It is thus a broader term than 'government' and deals with a variety of more general human social interactions and their social, economic and ecological outcomes. In generally sense, spatial governance (SG) is an aspect of governance. Spatial governance involves modern ideas about governance; but it focuses more particularly on a wide variety of situations where knowledge and mutual understandings about both time and place are important to successful social, economic and ecological outcomes (http://spatialgovernance.com/).The Spatial Data Infrastructures (SDI) could be considered as kernel of Spatial Governance. An analysis of SDI in Governance shows that the idea of SG has a wide variety of practical applications starting with the built environment, ways of creating enabling scenarios for making timely and information-driven decisions to solve existing or identified spatial problems. These aspects are important to the ownership, occupation, government and administration of territory in all its forms; as well as the activities that occur in a particular territory.

2.1. Multiple Facets of Spatial Data Infrastructures

There are many definitions of spatial data infrastructure (SDI) [31, 32, 33, 34) and many critical discussions about existing definitions and the complexity of the SDI concept, perspectives [35, 36, 37, 38, 39].

In this chapter, we accepted a more comprehensive and easy to understand definition of the International Society for Photogrammetry and Remote Sensing (ISPRS) Commission IV - Geodatabases and Digital Mapping, 2008-2012: "SDI is generally defined as the technologies, policies, and people necessary to promote sharing of geospatial data throughout all levels of government, the private and non-profit sectors, and the academic community. Goal of SDI is to reduce duplication of effort among agencies, improve quality and reduce costs related to geographic information, to make geographic data more accessible to the public, to increase the benefits of using available data, and to establish key partnerships with states, counties, cities, tribal nations, academia and the private sector to increase data availability".[40]

2.2. Spatial Information and Sustainable Development

Hadley [41] notes the often-quoted statistic that 80% of data used by government is spatial and that the proportion used for planning could well be even higher. Information on resources, (e.g. ecosystems and environment, human resources, linguistic resources, cultural and economic aspects etc.) becomes imperative for decision-making in SG. The implementation of Sustainable Development strategies should be based on scientific policy making, which demands great deal of real-time information. It is one essential step to obtain real-time information and to use IT&C in order to build IS IT&C for Sustainable Development. The nature of the built environment requires an IS capable of handling both the descriptive characteristics (attributes) and the spatial component of this environment. This important feature is what an IS for management spatial information offers. Such system implemented by IT&C specialists for the acquisition and verification, storage, updating, management and exchange, manipulation, retrieval and representation, analysis of geospatial data [42, 43] and using the expertise of specialized people, stressed the relevance of spatial information in decision making process when defined it [44, 45, 46]. Considered as superset of Geographic Information Systems (GIS) formed by the intersection of GIS and other techniques, the Spatial Decision Support Systems (SDSS) which contain specific decision models, and using the general spatial capabilities of the GIS [47], becomes a more relevant domain to the generation of more conducive decision making environment [48, 49, 50] for the problems of SG.

3. TRANSPORTATION SYSTEMS AND SUSTAINABLE DEVELOPMENT

"Every problem has an "environment," to which it is inextricably united. If you stop x from growing (or declining), you will make other things grow (or decline), and these changes you have created may very well be as serious, and as disastrous, as the growth of x." (Churchman"s classic "environmental fallacy [51]).

Each country recognizes that transportation is an important tool to help meet overall sustainability objectives. The development of new technologies and its introduction in everyday life has been a regular scenario in several areas including also transportation. The introduction and development of new technology for the transportation field has been occurring with different rhythms, being expected that the future lead unconditionally to the presence of such systems in all transportation areas. Intelligent Transportation Systems (ITS), whose introduction has been bringing several benefits to the transportation field and, more specifically to transports users, are the most visible consequence of this evolution. Attributes of sustainable transport follow from the expanded definition of sustainable development sustainable transport is safe, high quality, and accessible to all; ecologically sound; economical; and a positive contributor to regional development.

According to European Roadmap – Infrastructure for Green Vehicles (2012) "a sustainable transportation system, or to say sustainable mobility, is one that:

• Allows the basic access and development needs of individuals, companies and society to be met safely and in a manner consistent with human and ecosystem health, and promotes equity within and between successive generations.
• Is affordable, operates fairly and efficiently, offers a choice of transport mode, and supports a competitive economy, as well as balanced regional development.
• Limits emissions and waste within the planet"s ability to absorb them, uses renewable resources at or below their rates of generation, and uses non-renewable resources at or below the rates of development of renewable substitutes, while minimizing the impact of the use of land and the generation of noise. In the context of sustainable mobility, with its individual mobility, public and goods transport, the pertinent infrastructure technology carry an important role to overcome the problems of environmental impact, air quality and sustainable mobility." [52].

Specific goals for sustainable transport may include improved service quality and quality of access to goods and services, safety, improved air quality, noise reduction, improved water quality, protection of habitat and open space, historic preservation, reduced carbon emissions, increased social equity, economic development, and a satisfying quality of life, as well as local goals consistent with the overall objective.

One of the main ideas of this chapter is that "to be successful, sustainable transportation policy must avoid the common transportation pitfall of ignoring the larger systems in which transport is embedded" [53]. As mentioned also in [53] a fundamental characteristic of the transportation sector that continually frustrates policy development is "the porous nature of the transport system itself. Transportation is not a closed, self-contained system; rather, it is tightly intertwined with other systems".

3.1. Intelligent Transportation Systems

A first report, made by ITS America and published in 1994, tried to make a comparison between the Intelligent Transportation Systems (ITS) developed in Japan, USA and Europe. In [54] considered to be a very important base to build a common idea of ITS, is considered as opened way to the development of a second work, published in 1997 [55]. In this more recent report an international comparison between ITS and the related activities was made as an effort to collect information and gather consensus and in [54] suggested that categories for ITS technology need to be identified. Besides the classification in each country may not exactly be the same, they presented a table (Table 1) that indicates ITS technologies utilized for the three mentioned regions: Europe, Japan and America. Beside this observed division in the ITS subsystems, in order to accomplish for their work the authors choose to focus on the seven major categories of ITS commonly present in the three studied areas: ATMS, ATIS, AVCSS, CVO, APTS, EPS &ETC, and EMS.

Table 1. Major ITS technology categories in the US, Europe and Japan. Based in data collected from ([56], ITS Japan and ITS America [54])

Designation		Region
		USA	Europe	Japan
ATMS	Advanced Traffic Management Systems			
ATIS	Advanced Traveller Information Systems			
AVCSS	Advanced Vehicle Control and Safety Systems			
CVO	Commercial Vehicle Operations			
APTS	Advanced Public Transportation Systems			
EPS&ETC	Electronic Payment System and Electronic Toll Collection			
EMS	Emergency Management Systems			
Electronic Navigation	System			
	Road and Safety Management System			
	Information Management			
	Maintenance and Establishment Management	

Now talking about at least two dominant concepts in this research area: Intelligent Transportation Systems (ITS) and Cooperative Intelligent Transport Systems and Services (C-ITS)[56].

According to European Telecommunications Standards Institute (ETSI): "Intelligent Transport Systems (ITS) add information and communications technology to transport

infrastructures and vehicles in an effort to improve their safety, reliability, efficiency and quality. ITS services are also designed to optimize transportation times and fuel consumption thus providing greener and safer transportation. However, the deployment of Intelligent Transport Systems and provision of corresponding services are not limited to road transport sector only, but includes other domains such as railways, aviation and maritime as well." [57]

Intelligent Transportation Systems (ITS) are a vision of a future that integrates existing transportation infrastructure with communication networks and spatial data infrastructures in an effort to reduce congestion and travel time. In doing so on a mass scale, the larger effect of ITS is to limit the release of carbon emissions into the atmosphere, cut back on fuel consumption and improve road safety.

In [58] has been identified five trends as most relevant for transportation needs: urbanization; demographic developments; individualization; informatization; globalization. And, as mentioned in [59], applying IT&C, ITS delivers five classes of benefits by: increasing safety; improving operational performance, particularly by reducing congestion; enhancing mobility and convenience; delivering environmental benefits and boosting productivity and expanding economic and employment growth.

3.2. Cooperative Intelligent Transport Systems and Services

As mentioned in [56], modern vehicles equipped with driver assistance systems can "feel" (by sensors), "see" (by cameras) and – in future – "speak" (by communication systems). The new technology of C-ITS enables communication between vehicles and traffic infrastructure (Figure 1). It is based on the principle that cooperative parties (ITS stations, i.e. in vehicles, road side units) exchange information among each other in terms of standardized message sets. Deployment of C-ITS in Europe based on ITS G5 technology beginning in 2015. C-ITS applications are manifold and may not all be implemented at the same time. Instead, a phase model will be applied.

The ITS world is vast and includes a wide variety of sub-taxonomies and definitions that are currently utilized to describe intelligent systems. A considerable variety of equipment"s are exclusively dedicated to the road sector and much of them are directly related with single vehicles. A specific cluster of systems have received a great deal of attention from psychologists and human factor professionals that study road safety [60]. This cluster is characterized by on-board systems that can, in a direct or indirect manner, influence the driving task and the driving activity. Two main categories are usually distinguished according to their impact on driving: In-Vehicle Information Systems (IVIS) and Advanced Driver Assistance Systems (ADAS). This classification has been used in some European scientific projects that aimed to study the implications of these new technological systems, like for example the CONSENSUS [61], AIDE [62], HASTE [63], COST Action 352 [64] and HUMANIST [65].).

The In-Vehicle Information Systems "provide drivers with information or communication while the vehicle is in motion. The delivered information may not be related to driving (e.g. conversing by phone, consulting emails, or listening to the radio) or related to the trip management (e.g. traffic or weather information, navigation map, or route guidance)." The Advanced Driver Assistance Systems "are designed to support the driving task on vehicle manoeuvring by informing, warning and actively assisting drivers on the basis of vehicle surroundings analysis and with the requirement of immediate driver actions (e.g. intelligent speed adaptation, lateral and longitudinal control, blind spot warning, or night vision)" [66].

Figure 1. Mind map of the most important facts on cooperative Intelligent Transport Systems and Services (C-ITS) (Source: ITS Niedersachsen cited in [56])

3.2.1. Interaction with Navigation/Guidance Systems

In the last decades, with the introduction of new technologies, several systems have been developed with the main aim of present to drivers" useful information for their navigation. Once navigation can be described as: "to achieve movement through a space" [67] these systems can help drivers to travel throughout unfamiliar places and avoid making deviations from the wanted course [68, 69].

To accomplish this task, systems can transmit guidance instructions in strategic decision points [70] so drivers can make a more efficient use of the road network (Jeffrey as cited in [69]). In a broader way, the advantages obtained by their use include preventing drivers to get lost, and if traffic information is available, prevent them to travel through congested roads [71]. As a consequence to the presentation of other route options, these systems give the possibility of more equal distributions of the traffic through the transportation network, helping to solve quite a lot of congestion problems in several cities around the world [72]. Additionally, they can also contribute to a decrease in the consumption of fuel and to a decline in the release of pollutant gases ([73, 72, Jeffrey as cited in [69]). The use of these equipments can also augment the confidence and security of road users, once they can receive useful instructions to avoid or outline dangerous situations [73, 74, 71, 69].

In order to make the navigation possible, based on local SDI, these systems provide updated information about the place where the vehicle is at, offering also guidance instructions to select the most adequate roads. The information supplied can have different features, and be transmitted in different modes in order to guide the driver to the destination. Some of these systems are also capable of transmitting on-line information about traffic jam, to calculate the time needed to complete the course, or notify about incidents or other kinds of malfunctions on the road [75].

Some years ago, navigation systems could be characterized as being an electronic help that offered the driver several kinds of information upon request [76]. With the development of equipments that displayed a map which could give current location of the vehicle as well as the destination position; or a display that used symbols or voice instructions and also the increased variety of systems available in the market, these two groups of systems received different denominations: Navigation Systems and Route Guidance Systems.

Navigation Systems are considered as passive systems due to the fact that they do not provide route guidance information, offering only "geomatic information" [77]. They usually display a digital map on a screen that might highlight a route, the current position of a vehicle and the destination point [78]. With this information, drivers are expected to actively make their own navigation decisions at each intersection in order to get to the destination [79].

On the other hand, Route Guidance Systems are able to calculate the best route to follow, based on specific criteria. The suggestion of a course is given through route guidance information, being these messages considered as important and useful to guide the driver [78, 77, 79].

On the other hand, Route Guidance Systems are able to calculate the best route to follow, based on specific criteria. The suggestion of a course is given through specific instructions, also named as route guidance information, being these messages considered as important and useful to guide the driver [78, 77, 79].

Besides this systems" classification based on the type of the information transmitted, they can also be classified in a different way [60]: those fitted as standard to the vehicle and the nomadic devices. Devices fitted as standard to the vehicle are originally installed in the car, meaning that when the driver buys the car the navigation systems is already initialized. There are also some options that allow drivers to include a system afterwards; however theses aftermarket navigation devices are professionally installed and can only be used inside the vehicle.

Contrarily, nomadic systems are generally introduced inside vehicles by their own drivers and one of the characteristics that distinguish this type of system is the fact that they can be used for besides the driving context [80, 81]. They can be activated in different situations and with different aims because these equipments can include also mobile phone functions, access to the internet, and electronic agenda, among several other utilities. Concerning the manmachine interaction in a road environment, nomadic systems are similar to systems fitted as standard to the vehicle as they allow the transmission of audio and written messages that can help drivers guide themselves through the road network [80]. However, the way they are positioned and attached inside the vehicle can be considerably different. Unlike the fitted as standard systems, nomadic devices are placed according to drivers" wish. This aspect can foresee two problems: the first is related with the proper positioning of the system and consequently with the efficiency on the acquisition of messages; the second is a safety concern as in case of accident the system can be loose and hurt the driver [80]. Problems related to the single, isolated automotive vehicle and its subsystems are challenging enough (see the grand challenge on Advanced Driver Assistance Systems), but the research community is also exploring the "big picture" of intelligent road transportation - the system, or system of systems, consisting of many vehicles and their drivers interacting on roads. Two related topics are included in this vision: Vehicle-to-infrastructure (V2I) interaction and Vehicle-to-vehicle (V2V) interaction.

3.2.2. Vehicle to Infrastructure Communication

In Vehicle-to-infrastructure (V2I) concept, the infrastructure plays a coordination role by gathering global or local information on traffic and road conditions and then suggesting or imposing certain behaviors on a group of vehicles. One example is ramp metering, already widely used, which requires limited sensors and actuators (measurements of traffic density on a highway and traffic lights on ramps).

In a more sophisticated scenario, the velocities and accelerations of vehicles and intervehicle distances would be suggested by the infrastructure on the basis of traffic conditions, with the goal of optimizing overall emissions, fuel consumption, and traffic velocities. Suggestions to vehicles could be broadcast to drivers via road displays or directly to vehicles via wireless connections. Looking further ahead, in some cases suggestions could be integrated into the vehicle controls and implemented semi automatically (always taking onto account the restrictions on automatic vehicle driving imposed by the Vienna Convention on Road Traffic [82]).

V2I promise revolutionary improvements in transportation - greater energy efficiency, less road construction, reduced collisions, and safety of vehicle occupants as well as pedestrians and bicyclists. Control is a key contributing discipline for both topics. Some experts [83] predict that the first V2I systems may be developed and deployed in the 2015- 2020 time frame.

3.2.3. Vehicle-to-Vehicle

In the Vehicle-to-Vehicle (V2V) concept, when two or more vehicles or roadside stations are in radio communication range, they connect automatically and establish an ad hoc network enabling the sharing of position, speed, and direction data. Every vehicle is also a router and allows sending messages over multihub to more distant vehicles and roadside stations. The routing algorithm is based on the position of the vehicles and is able to handle fast changes of the network topology. Control technology comes into play at local and higher layers of the architecture. Uncertainties, delays, partial measurements, safety and performance objectives, and other aspects must be considered, and the system must be capable of making automatic or semiautomatic decisions, providing warnings/information and potentially effecting actions.

A lot of infrastructure technologies and processes are already known, some already developed, interesting and promising ideas appear and research initiatives have been undertaken. The smart electricity grid, charging devices in car parks, filling stations for hydrogen, gas and renewable fuels, logistics and services, information systems for users, traffic management are only some infrastructure needs for new research, development and implementation.

As mentioned in [84] there are currently many initiatives across the globe in V2V and V2I communications. Though performed by different groups with different objectives, the overall goal is the same: to provide drivers with real-time information beyond what is available through the senses alone. This allows for better decision making, which enhances safety and driving efficiency. Between these initiatives has been enumerated: FRICTI@N [85],GST[86], PReVENT [87], VSC[88], Centrico [89], COMeSafety [90], CVIS[91], eCall [92], COOPERS [93], NoW [94], Staufreies Hessen 2015[88], INFONEBBIA "Safety in the Fog"[95] etc.

Within SmartCities and Communities [96] has been mentioned some initiatives such as listed:

• in [97] projects addressing the problems of ITS, such as: in Germany [98], on 6 August 2012 has been proposed a Pilot Project regarding the improvement of road safety and efficiency and a Pilot Project concerning the enhancement of application of C-ITS; in The Netherlands, on 20 September 2012 has been proposed a Pilot Project towards a Smart Mobility Roadmap and on 24 September 2012, Schneider Electric proposed a Pilot Project addressing V2I on-board information and a Pilot Project for City Congestion Management;
• in [99] Schneider Electric proposed on 24 September 2012 a Pilot Project concerning the data exchange between existing systems (multiagency interaction); EIGSI proposed on 1 October 2012 a Pilot Project addressing visibility and accessibility of medium sized cities and in 6 October 2012 launch an idea regarding the integration city/logistics in global transport system; GMV proposed on 21 December 2012 a Pilot Project addressing efficient and sustainable mobility.
• In [100] it is presented an initiative undertaken by the U.S. Department of Transportation (DOT) to provide an infrastructure where vehicles can identify threats and hazards on the roadway and communicate this information over wireless network to alert and warn drivers, named Vehicle Infrastructure Integration (VII). This initiative, recently renamed IntelliDrive, has significantly impact DOT operations and should be monitored on a continual basis.

In summary, for ensuring of a sustainable development of the society, what intelligent transportation systems do is empowers actors in the transportation system – from commuters, to highway and transit network operators – with actionable information (or, intelligence) to make better-informed decisions, whether it"s choosing which route to take; when to travel; whether to mode-shift (take mass transit instead of driving); how to optimize traffic signals; where to build new roadways; what the true cost of roadways are and how best to price their use; or how to hold providers of transportation services accountable for results. "The big opportunity at hand is to bring information to bear on transportation networks, transforming them into truly intelligent transportation systems." [59]

4. SUSTAINABLE INFORMATION SYSTEMS MANAGEMENT

The increasing dissemination of IS IT&C into all areas of business and personal life has drawn attention to its economic, environmental and social effects. The purpose of sustainable IS IT&C management is to contribute to the vision, the overall strategy and the strategic objectives of the society as whole, which are not only economic, but also ecological and social [101]. These objectives do not only apply to information as a resource but also all other resources identified by the resource-based view. A sustainable IS IT&C management covers not only the economical, but also the ecological years [101] and social perspective, extending traditional concepts of IS management. The social dimension of sustainable IS management includes topics of IT&C utilization by individuals and organizations, but also comprises the challenges of generating, preserving and safeguarding knowledge, information and data. Sustainability is sometimes compared with the internet revolution, given its wide-ranging impact on businesses.

According to [102], a Sustainable Information System (SIS) is an Information System which: adapts to its environment, involves relevant stakeholders, and supports the knowledge lifecycle, i.e. knowledge creation, knowledge evaluation and knowledge integration/ application.

The new generation of SIS will not be a single system but "multiples connected infrastructures based on multiple technological platforms and on open access and will require new governance models" [103].

Starting from the above considerations, in our opinion, the development of Sustainable Information Systems Management for Spatial Governance (SISMSG) requires the integration of various IT&C tools, models, formats of data and methodologies. The selection of IT&C tools and the merging of methodologies should result in specifications of domain-specific software architecture, suitable for development a particular application domain. The SISMSG development requires taking into consideration of cultural and political factors in addition to technological components, comprehensive studies by multidisciplinary terms of researchers with diverse cultural background and domain expertise.

4.1. Requirements for SISMSG

"Today, we live in a world that is dramatically transformed with technological advances. The immediate opportunity is to integrate the data residing in disparate places and create value out of it. The convergence of various stakeholders including users, technology providers and the industry will enable us to meet the needs of the future" (KK Singh in [1]).

In the generally sense, the IS IT&C consists of stakeholders, process and equipment (hardware and software) to collect, analyze and proliferate information across relevant decision makers (http://www.managementstudyguide.com/measuring-market-demand.htm).

Starting from SIS"s accepted definition [103] and from the article elaborated by Jakobsson and Tsoulos [104], we try to exemplify the evaluation of SISMSG by using an architecture that seems suitable to support a multi-layered approach, underling the benefit to users, benefit to society and the gain (Figure 2). Based on our experience, we would like to discuss some relevant characteristics of a SISMSG to constitute a base of evaluation.

All of our considerations start from the fact that must to be a demand for any SISMSG which comes from the society, from the users and/or stakeholders in order to provide a harmonized content and of good quality as support for decision making in spatial governance.

Demand. As an economic principle that describes a consumer"s desire and willingness to pay a price for a specific good or service, the demand for spatial information systems has been growing sharply in recent years, especially with the proliferation of SDI at different levels and with entering SG in political and economical areas.

Figure 2. Evaluation of SISMSG.

The resources, which are playing an important role in this context, are primarily the infrastructure(s), the internal and external relationships, and the market/demand responsiveness. For this reason the model of sustainable IS IT&C management has the ability to illustrate complex relationships and effects between stakeholders, objectives and measures. High demand is also a prerequisite for building a stable and significant user base, which definitely would help to sustain a system over time.

4.2. Resources for SISMSG

In order to satisfy the existing and future demand for SISMSG and having Spatial Governance as goal, the resources are represented by spatial data, spatial knowledge, geospatial information and human resources (HR) represented by specialized people and/or simple citizens.

Spatial Data also known as geospatial data (GD) or geographic information (GI) it is the data that identifies the geographic location of features and boundaries on Earth, such as natural or constructed features, oceans, and more. Geospatial Data is usually stored as coordinates and topology, and is data that can be mapped. Geospatial Data is often accessed, manipulated or analyzed through GIS.

Spatial Knowledge. Creating *geospatial knowledg*e (GK), as mentioned in [105], is an effortful process the analyst undertakes; it is a cognitive intellectual endeavour that arrives at a conclusion through geospatial reasoning. The notion of "geospatial reasoning" is the core of geospatial knowledge creation. Geospatial reasoning is the cognitive process of integrating spatial evidence and geographic problem representations in the process of making judgment about space and spatial relationships. Evidence is geospatial information helpful in forming a conclusion or judgment. Problem representation is what we know about the structure in the system and its properties. "There is a need for greater collaboration and association between industry and government to develop a global platform for sharing geospatial knowledge" [1] and GIS is used as a knowledge management platform to create, disseminate, renew and manage explicit (codified) and tacit (experiential) knowledge toward organisational sustenance and survival. Also is used to collaboratively integrate multiple forms of GK to develop meaningful solutions to complex environmental and natural resources management issues, challenges and opportunities.

Geospatial Information. "Geospatial information is becoming more and more central to many capabilities … The nature of geospatial is a combination of three different technologies – sensors that capture data, software that turns data into information and the wireless communication that enables that information to be used everywhere in relatively seamless manner" (Steven Berklund in [1]).

The terms Geospatial Data (GD), Geospatial Knowledge (GK) and Geospatial Information are difficult to define, but cannot be used interchangeably. Geospatial information (GI) provides the basic framework for space visualization. It is information produced by multiple sources to common interoperable data standards. It may be presented in the form of printed maps, charts, and publications; in digital simulation and modeling databases; in photographic form; or in the form of digitized maps and charts or attributed centerline data. Geospatial services include tools that enable users to access and manipulate data, and also include instruction, training, laboratory support, and guidance for the use. It should be noted that the definitions and usage of the terms GD, GK and GI are not used consistently or unambiguously further exacerbating the situation. Geospatial Data (GD) can (usually) be applied to the output of a collector or collection system before it is processed, i.e., data that was sensed. GI is geospatial data that has been processed or had value added to it by some human or machine process. Geospatial Knowledge (GK) is a structuring of Geospatial Information (GI), usually with some interpretation or analysis.

4.3. The Characteristics of a SISMSG

In order to evaluate the really gain, in our opinion, for the characteristics of a SISMSG it is necessary to pay attention at least to two points of view: of the users and of the society.

a) From the point of view of users, we have to pay attention to: longevity, accessibility, responsiveness, and robustness.

Longevity. Given the amount and the complexity of investments in the collection, storage, handling and visualization of the resources of SISMSGS mentioned above, one of the important characteristic of a SISMSG is that it is to be able to survive over time. Digital longevity is an area of growing concern [106]. It is widely accepted that *IT&*C is revolutionizing the concepts of documents and records – in general sense - in an upheaval at least as great as the introduction of printing, if not of writing and/or designing itself. The current generation of digital records therefore has unique historical significance; yet the digital documents and records are far more fragile than paper. In fact, the record of the entire present period of history is in danger. The content and historical value of many governmental, organizational, legal, financial, and technical records, scientific databases, geospatial databases and personal documents may be irretrievably lost to future generations if we do not take steps to preserve them. The longevity of the GI is one of the most complex problem if it is considered only the discouraging characteristic of being software-dependent. Preserving digital documents and records containing GI, GK, and GS is almost analogous to preserving ancient written texts [107].

Accessibility. One of the most prominent features of the Web is the easy and ubiquitous access. Services channeled through standard browsers reach the most distant corners of our globe. The illusion that standards provide an answer is only an illusion and the promise of "migration" seemly to be false and time and costs consuming. Although defining ultimate standards for digital documents and for the resources of SISMSG may be an admirable goal, it is premature [107]. The field seemly to be too new to have developed an accepted, formal understanding of the ways that humans manipulate GI and GK. It would be presumptuous to imagine that we are ready to enumerate the most important kinds of digital applications containing GI and GK, let alone to propose that we are ready to circumscribe their capabilities by standardizing them. Any attempt to force users to settle for artificial limitations imposed by such standards would be futile, since the momentum of the information technology revolution derives directly from the attraction of new capabilities. It may become feasible to define long-term standards for digital documents containing spatial information sometime in the future, when information science rests on a more secure, formal foundation, but standards do not yet offer a perfect solution to this problem.

Responsiveness. Specialists and/or simple citizens need feedback and the feeling of being heard and participating in the processes of spatial governance. Most systems will be strengthened by processes allowing users to contribute to the development but as mentioned above, we cannot imagine till now the complexity of the applications in the area of SG starting from the GIS, SDSS including Artificial Intelligence techniques and sensors.

Robustness. Robust systems are fault tolerant in the sense that sub systems may fail without bringing the system down. This might be implemented by exploiting redundancy or by providing backup systems/methods. Robustness also implies that it should be easy to replace modules of the system without redesigning the whole system. This aspect is very important in Spatial Governance having in mind the diversity of processes and relation between them.

b) Considering the society"s point of view, we have to discuss about adaptivity, scalability, simplicity, stability and environmentally friendly.

Adaptivity is the challenge that faces all researchers in this field is the development of adaptive systems for unpredictable and unforgiving applications domain [108]. A system should have an underlying structure that allows for easy adapt of technological innovations and improvements.

Scalability. In this context the scalability does not refer to the scalability from geospatial domain but it refers to the meaning from IT&C domain. Accelerated interest in and rapidly increasing number of request against a web server may bring a system to halt, and in several cases start a domino effect which eventually will crash the system fundamentally. A system has to gracefully handle an increasing work load, and also be prepared for a rapid growth of the content.

Simplicity. "To facilitate easy adoption of these technologies by the user community, the industry needs to make the applications cheaper and simpler for them" [1]. This is a common denominator for long-time survivors in the world of IT&C systems and in geospatial world. A sustainable system should be well planned with easy-to-use-and-understand design and clear.

Stability. A sustainable system should be run by an organization, or one or more key responsible if needed, with regards to maintenance and development. Commercial companies, public and/or private, may administer sustainable systems, but would most likely drop a project if becomes financially uninteresting.

Environmentally Friendly. Environmentally friendly is a term used to refer to goods and services, laws, guidelines and policies claimed to inflict minimal or no harm on the environment.

c) But the really gain is acquired by applying process management and it is expected to reach interoperability and affordability in order to use the harmonized content and of good quality, preserving the previously investments.

Process Management in *Spatial Governance*. The generic process management mechanisms needed to provide SG are those pertaining to: authority; planning; monitoring[109]. Interaction is necessary between the generic mechanisms. Authority can involve rights and responsibilities in respect of planning - and planning can have implications for the kind of authority necessary to implement plans. Planning has implications for monitoring to indicate whether planning is successful - and monitoring provides the experience on which further planning can take place. Monitoring has implications for whether an authority is performing satisfactorily - and an authority can instigate changes in monitoring regimes as a test of its own performance.

Spatial Planning Processes. Practices of spatial planning (SP) in Europe have borne witness to considerable challenges in recent decades as a consequence of changing policy priorities, heightened expectations and increased socio-economic and demographic disparities [110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122]. However, recent decades have, also witnessed the emergence of new strategic approaches to SP at local, regional, national and transnational scales, which have sought to broaden the scope and enhance the governance capacity of SP in practice [113, 114]. As presented in [115] the purpose of strategic planning is to inform decision-making in practice, rather than to directly influence material outcomes (i.e. the object of project plans). From this perspective, strategic SP is viewed as a continuous process, which provides a frame of reference for subsequent decisions and actors [115]. The concept of strategic SP places emphasis on the development of coordinated or integrated perspectives that transcend traditional sectoral policy divisions through a specific focus on the spatial impacts of sectoral policies [114, 116, 117]. Challenging an instrumentalist view of planning, a number of authors have stressed the role of values in strategic SP. For Healey [113] the concept of strategic SP places particular emphasis on the development of approaches to policy-making where objectives and values relating to future socio spatial development within a territory or functional space are shaped, framed and negotiated as part of the strategy-making process. In [114] similarly defined strategic Spatial Planning as a: "public-sector-led sociospatial process through which a vision, actions and means for implementation are produced that shape and frame what a place is and may become". The concept of a re-emergence or revival of strategic SP in Europe since the 1990s is more specifically associated with the development of SP policy and perspectives at the European scale. In particular, the publication of a European Spatial Development Perspective [123] and the adoption of the principle of territorial cohesion policy have served to highlight the significance of territorially differentiated perspectives on processes of social and economic development in the context of the European Union ([118, 119, 120, 121, cited in [122]). Planning involves a range of information-intensive activities aimed to improve performance by design or 'on purpose' for improving performance taking advantage of future possibilities - or avoiding - future possibilities. Design involves working out how to achieve some purpose within constraints of what is deemed possible in: a technical sense (something that can be done); an economic sense (something that is affordable) and a political sense (something that is socially permissible and acceptable) [109]. This issue could be expanded in SDI in Spatial Governance: "Today, every local action has a global reaction. So, every global challenge influences local living. Geospatial technology helps understand and manage our resources. I believe geospatial industry has the true potential to be a game changer in the current world order" (Sanjay Kumar in [1]).

Interoperability. "We were working in silos and this is particularly prevalent in intelligence communities. Increasingly now, that is not the way things work. We need to fuse across traditional boundaries, inculcate the culture of sharing and enable the databases to become interoperable. It is also about the interoperability of the mind" (Marchal Stuart Peach in [1]).

As mentioned in the literature [124, 125, 126, 127, 128, 129, 130, 131] information system interoperability occurs at multiple levels within and across platforms, providing the capacity of interchanging data, services, and applications among different organisations as well as users [130, 131]. Several interoperability levels have been proposed to achieve system interoperability, including technological, syntactic and semantic levels (ISO 2002, 126, 127]. In a GIS context, interoperability has emerged as a research issue on computation implementations of GD exchange since a decade ago when eight interoperability levels were identified: these being the information community and institution, enterprise, application tools, middleware, data store, distributed computing environment and network [132]. In a SDI context, the Open Geospatial Consortium (OGC) and International Organisation for Standardisation (ISO) TC211 have played a major role in improving GD and System Interoperability through the specification of object models and XML schemas to store and transfer data, Open Service Interfaces, and data and services metadata standards. The formalisation of interoperability models became essential for developing abstractions of critical measures for the connectivity of different interoperability levels.

"In order to work more efficiently and in a unified manner, there is an increased need to have more integrated and active SDIs and standardization of data to facilitate interoperability" [1].

In our opinion, in connection with SISMSG, it is necessary to pay attention also to the aspect closely related to interoperability named in the literature persistence, the "interoperability with the future", i.e. the independent systems able to exchange meaningful information and initiate actions from each other are separated by time.

Affordability. This aspect is in connection with identification of bottlenecks, missing links and quality of service in infrastructure networks. This poses two types of challenges:

• How to speed up a process in order to achieve a sustainable development with legal instruments which are its key;
• How can effectively improve the implementation management mechanisms and how much could rely on cooperation with the partners.

Harmonized Content. "Opportunities are tremendous and it is time for geospatial revolution, but we all face challenges, we need to work together on standardisation and harmonisation on sharing data. That is how we can create a geospatial revolution" (Melanie Schultz van Haegen-Maas Geesteranus in [1]).

A resource of SISMSG (such as GD, GK and GI) can be part of more than one domain (document, records, scientific domain etc.), and can be identified by different systems, so it is necessary to guarantee interoperability between different identification systems as well as implementations based on the same namespace [125]. OGC in parallel with World Wide Web Consortium (W3C) and ISO develop standards and specifications to support the interoperability between repositories with geospatial content. The scope of these standards and specifications is to enable an application developer to use any geospatial content (resources and services) available on "the net" within a single environment and a single workflow [133]. However, standardization may not solve the problem of interoperability by itself, because [134] the construction and maintenance of a single and integrated model is a hard task and the requirement to communicate with geospatial sources that do not complying the adopted standard will be always present; existing geospatial sources have their own models which may not always be mapped to the common model without information loss; the standards/specifications are subject to continuous change, but systems will not all simultaneously change to conform.

Quality can be described from different viewpoints [135, 136]. Based on [104] quality can be defined as fitness for use, including both quality of design, conformance to the design (production oriented quality), customer satisfaction and the fulfilment of the needs of society or environment. Most of the quality descriptions of the resources of SISMSG have been developed to serve the production-oriented approach. ISO 19113 and ISO 19114 standards follow this approach based on the data quality concepts developed already in the 1980"s [137, 138, 139, 140, 141], on earlier standards like the Spatial Data Transfer Standard (SDTS) (FGDC, 1991) and CEN pre-standard on quality in 1998. In the context of SISMSG, the concept of spatial quality (SQ) could be developed, in our opinion, by means of three components: use value, perception value and future value. These three terms have become the point of reference in discussions on SQ [142]. A high use value occurs when space can be used in a safe manner for several purposes (to live and to work, but also to recreate and to move through). The idea is that these different functions do not hinder and possibly reinforce each other. Perception value plays an important role in people"s living environment. Cultural awareness and diversity, the presence of characteristic properties (identity) and of history and beauty, are all elements that can be attributed to perception value. Perception value also involves spatial diversity and variation. Future value includes characteristics such as sustainability, bio-diversity, robustness and flexibility, both concerning suitability for new use forms and admissibility for new cultural and economic meanings [142]. The future value are in direct relation with persistence.

Preserving the Previously Investments. Despite the technological advances, the problem of archiving remains: there is, as yet, no satisfactory solution for to the long term preservation of the vast amounts of the resources for SISMSG (spatial data, spatial knowledge, geospatial information) already created. It is hard at this point to see how century long preservation of digital media can be achieved, together with the means to search and retrieve useful information.

Regarding Human Resources, significant research work that considers how C-ITS impacts on the citizens as drivers [60] and passengers has been carried out over a number of years in research projects and by in-house work from vehicle manufacturers and road transport testing laboratories. Much of this work has been undertaken in European research projects. A detailed analysis has been made of existing ITS projects (National and European) and also of the work of Consortiums/Organizations/Institutes working in the ITS related area, and the high-level observations on what work are contained in ([57] annex B). In the same time, in order to improve the infrastructure "… there is a need to build capacity for enhanced integration of geospatial technologies into the mainstream education by means of partnerships between the industry, government and academia" [1].

ITS has been attracting attention from many professionals in several countries all over the world. Groups of discussion have been created in order to contribute to the effective development of ITS and also to take an active role in sharing its future standards since the eighties [54].

CONCLUSION

Starting from the fact that it is recognized that "ICT reinforces the convergence of all three pillars of development -- social, economic and environmental, and that ICT by definition embraces the fundamental principles of sustainability" [26] and from the fact that United Nations Istanbul Plan of Action for Least Developed Countries (2011-2020) [143] and based on the current literature, this chapter proposes a reorientation of the way the concept of sustainability is dealt with, Spatial Governance based on human behavior and the processing of Geospatial Knowledge at the centre of the concept. The sustainability approach presented in this chapter is then translated into requirements needed for a Sustainable Information Systems Management for Spatial Governance focusing on (Cooperative) Intelligent Transportation Systems, Spatial Data Infrastructure and emphasizing Spatial Governance.

To conclude this chapter, the discussion of the characteristics of Sustainable Information Systems Management for Spatial Governance has been used to briefly revisit the complexity of involved in the concept of sustainability, sustainability of IS IT&C innovation, Spatial Data Infrastructures and his multiple facets in connection with Spatial Governance and the resources. Positioning Spatial Governance based on human behavior and the processing of Geospatial Knowledge at the centre of the concept, on the one site, and inspired from the paper elaborated on the way quality should be taken into account in the development of innovation, Spatial Data Infrastructures – the Spatial Governance"s kernel - of [60] based on four quality management viewpoints, i.e. a production-centred, planning-centred, customercentred and a system-centred perspective, to exemplify has been used an architecture that seems suitable to support a multi-layered approach, underling the benefit to users, benefit to society and the gain.

Based on the experience of the authors and on the literature, has been highlighted that such Sustainable Information Systems Management for Spatial Governance:

• will come from the holistic and multilevel vision and possibility of for real time actions of the various systems constituting the mobility space because such approach is the starting point for integration and interoperability between systems;
• will be not just about space and spatial relations but also about place, culture, identity, spanning physical and virtual space;
• will emphasise the analysis of networks and flows and relationships between places (including cultural influences) in addition to the traditional spatial analysis;
• will be easy to use with different levels of functionality for different types of users;
• the competition for limited resources implies reduced redundancies in the system and a large vulnerability to random failures.

In order to highlight the complexity, the authors suggested some characteristics of a Sustainable Information Systems Management for Spatial Governance that should be applicable, in general, for any infrastructure based on IT&C innovation but also to any domain"s application including (cooperative) intelligent transportation systems and defining spatial governance.

It is necessary to recognize that the implementation of the Sustainable Information Systems Management for Spatial Governance require changes in behaviour and adaptation to new modes of governance, from the design and implementation of new governance to individual awareness and acceptance of new mobility modes. The consequences of the implementation of such Sustainable Information Systems Management for Spatial Governance will be the empowerment of individuals, organizations, governments and society.

REFERENCES

• [1] Geospatial World Forum Report. RAJ Convention Center, Amsterdam, The Netherlands, 23-27 April 2012, http://anax1em.pressmart.com/GeospatialMedia Proceedings/In dex.aspx.
• [2] Glavic, P.; Lukman, R.; Review of Sustainability Terms and their Definitions. Journal of Cleaner Production, vol. 15, no. 18, pp. 1875-1885, 2007.
• [3] Cato, S.M.; Green Economics. ISBN 9781844075713, Earthscan, London, UK, pp. 36- 37, 2009.
• [4] Adams, W.M.; The Future of Sustainability: Re-Thinking Environment and Development in the Twenty-First Century. Report of the IUCN Renowned Thinkers Meeting, 29-31 January 2006, http://cmsdata.iucn.org/downloads/iucn_future_of_ sustanability.pdf.
• [5] Dictionary.com. http://dictionary.reference.com/browse/sustain.
• [6] Onions, C.T.; The Shorter Oxford English Dictionary. Clarendon Press, Oxford, p. 2095, 1964.

• [7] United Nations General Assembly; Report of the World Commission on Environment and Development: Our Common Future. Transmitted to the General Assembly as an Annex to Document A/42/427 - Development and International Co-operation: Environment; Our Common Future, Chapter 2: Towards Sustainable Development; Paragraph 1, March 20, 1987, www.un-documents.net/ocf-02.htm.

• [8] United Nations General Assembly; World Summit Outcome. Resolution A/60/1, Adopted by the General Assembly, 15 September 2005.

• [9] Forestry Commission of Great Britain; Sustainability, www.forestry. gov.uk/forestry /edik-59fmzf.

• [10] Manning, S.; Boons, F.; Von Hagen, O.; Reinecke, J.; National Contexts Matter: The Co-Evolution of Sustainability Standards in Global Value Chains, Ecological Economics, Forthcoming, 2011, http://papers.ssrn.com/sol3/papers.cfm?abstract_id= 1752655.

• [11] Reinecke, J.; Manning, S.; Von Hagen, O.; The Emergence of a Standards Market: Multiplicity of Sustainability Standards in the Global Coffee Industry. Organization Studies, Forthcoming, 2012, http://papers.ssrn.com/sol3/papers.cfm?abstract_id= 1970343.

• [12] SAI Platform; Sustainability Indicators. Sustainable Agricultural Initiative. 2010, www.saiplatform.org/activities/alias/sustainability-indicators.

• [13] Alvarez, G.; Sustainable Agriculture and Value networks, Lausanne, Switzerland: Latitude. 2012, www.intracen.org/WorkArea/DownloadAsset.aspx?id=51770.

• [14] What is Sustainable Development?. International Institute for Sustainable Development, 2009, www.iisd.org/sd/.

• [15] Sustainable Development: Introduction. EurActiv, 2004, www.euractiv.com/ sustainability/sustainable-development-introduction /article-117539.

• [16] Kates, R.; Parris, T.; Leiserowitz, A.; What Is Sustainable Development?. Environment, vol. 473, pp. 8-21, 2005, www.euractiv.com/sustainability/sustainable-developmentintroduction/article-117539.

• [17] Holling, C.S.; Theories for Sustainable Futures. Conservation Ecology, vol.42, p. 7, 2000, www.ecologyandsociety.org/vol4/iss2/art7.

• [18] Redclift, M.; Sustainable Development 1987-2005: An Oxymoron Comes of Age. Sustainable Development, vol. 134, pp. 212-227, 2005.

• [19] Porritt, J.; Capitalism as If the World Mattered. ISBN 9781844071937, Earthscan, London, UK, p. 46, 2006.

• [20] IUCN/UNEP/WWF, Caring for the Earth: A Strategy for Sustainable Living. Gland, Switzerland, 1991, http://coombs.anu.edu.au/ ~vern/caring/caring.html.

• [21] Markus, J.; Milne, M.K.; Kearins, K.; Walton, S.; Creating Adventures in Wonderland: The Journey Metaphor and Environmental Sustainability. Organization, vol.136, pp. 801-839, 2006, http://org.sagepub.com/content/13/6/801.abstract, retrieved on 2012-04- 08.

• [22] The Earth Charter. The Earth Charter Initiative, 2000, www.earthcharterinaction.org/ content/pages/Read-the-Charter.html.

• [23] Costanza, R.; Patten, B.C.; Defining and Predicting Sustainability. Ecological Economics, vol. 15, no. 3, pp. 193-196, 1995.

• [24] Pezzoli, K.; Sustainable Development: A Transdisciplinary Overview of the Literature. International Congress of Collegiate Schools of Planning and the European Schools of Planning, Toronto, July 1996.

• [25] Faber, N.R.; Jorna, R.J.; Van Engelen, J.M.; The Sustainability of Sustainability: A Study into the Conceptual Foundations of the Notion of Sustainability. Journal of Environmental Assessment Policy and Management, vol. 71, 2005.

• [26] Kent, D.; Should Information and Communications Technology Experts Have a Seat at the Table During the Rio+20 Summit?. 2012, www.huffingtonpost.com/daniel-kent/rio 20_b_1516056.html.

• [27] Zwetsloot, G.I.; Van Marrewijk, M.N.; From Quality to Sustainability. Journal of Business Ethics, vol. 55, pp. 79-82, 2004.

• [28] Caldelli, A.; Parmigiani, M.L.; Management Information System A Tool for Corporate Sustainability. Journal of Business Ethics, ISSN: 0167-4544, vol. 55, issue 2, pp. 159-171, Dec. (I) 2004.

• [29] Clayton, A.M.H.; Radcliffe, N.J.; Sustainability: A System Approach. Earthscan Publications Ltd, ISBN: 1 85383319 3, ISBN: 1 85383314 2, 1997.

• [30] Edwards, A.R.; The Sustainability Revolution: Portrait of a Paradigm Shift. New Society Publishers, ISBN: 0-86571-531-9, 2006.

  • [31]TWHOMB. The White House, Office of Management and Budget, Circular no. A-16 Revised, August 19, 2002, http://www.whitehouse.gov/omb/circulars_a016_rev/.

• [32] Masser, I.; GIS Worlds: Creating Spatial Data Infrastructures. Redlands, ESRI, CA, USA, 2005.

• [33] Nebert, D.; Developing Spatial Data Infrastructures. The SDI Cookbook, Version 2.0, 2004, www.gsdi.org/docs 2004/Cookbook/cookbookV2.0.pdf.

• [34] Kuhn, W.; Introduction to Spatial Data Infrastructures. Presentation, March 14, 2005, www.docstoc.com/docs/2697206/Introduction-to-Spatial--Data-Infra structures.

• [35] Hendriks, H.J.; Dessers, E.; Van Hootegem, G.; Reconsidering the Definition of a Spatial Data Infrastructure. International Journal of Geographical Information Science, doi:10.1080/136588 16.2011.639301, 2012, www.tandfonline.com/doi/pdf/10.1080/ 136588 16.2011.639301.

• [36] Masser, I.; All Shapes and Sizes: The First Generation of National Spatial Data Infrastructures. International Journal of Geographical Information Science, vol. 13, pp. 67-84, 1999.

• [37] European Commission, Establishing an Infrastructure for Spatial Information in the European Community (INSPIRE). Directive 2007/2/EC of the European Parliament and of the Council, 14 March 2007.

• [38] Giff, G.A.; Crompvoets, J.; Performance Indicators: A Tool to Support Spatial Data Infrastructure Assessment. Computers Environment and Urban Systems, vol. 32, no. 5, pp. 365-376, 2008.

• [39] Salvemini, M.; Van Den Broucke, D.; Results of the INSPIRE State of Play 2010: First Lessons Learnt from the Application of INPIRE Directive in EUROPE. Third Preparatory Meeting of the Proposed United Nations Committee of Experts on Global Geospatial Information Management, New York, 4-6 April 2011.

• [40] www.commission4.isprs.org/wg1/.

• [41] Hadley, C.; Technological Innovations in Land Survey and Large-Scale Mapping: New Techniques for Revision of Large-Scale Map Data: The Great Britain Experience and Lessons Learned. UNCHS Habitat, Land Survey and Large-Scale Mapping in Sub-Saharan Africa, p. 118, 2001.

• [42] Benhardsen, T.; Geographic Information Systems. VIAK IT, Norway, pp. 9-17, 1992.

• [43] Grimshaw, D.J.; Bringing Geographical Information into Business. Longman Group Ltd, pp. 22-40, 1994.

• [44] Densham, P.J.; Spatial Decision Support Systems. Maguire, D.J.; Goodchild, M.S.; Rhind, D.W.; (Eds.), Geographical Information Systems: Principles and Applications. Longman, London, pp. 403-412, 1991.

• [45] Jankowski, P.; Nyerges, T.; Smith, A.; Moore, T.J.; Horvath, E.; Spatial Group Choice: A Spatial Decision Support Tool for Collaborative Decision Making International Journal of Geographical Information Systems, vol. 11,no. 6, pp. 577-602, 1977.

• [46] Keenan, P.B.; Geographic Information Systems. Encyclopedia of Information Systems, Ed. by Bidgoli, H.; pp. 421-432, Elsevier Science USA, San Diego, CA, USA, 2003.

• [47] Keenan, P.B.; Spatial Decision Support Systems: A Coming of Age. Control and Cybernetics, vol. 351, pp. 9-27, 2006.

• [48] Pick, J.B.; Geographic Information Systems. Business, Idea Group Publishing, Hershey, PA, USA, 2005.

• [49] Huerta, E.; Navarrete, C.; Ryan, T.; GIS and Decision-Making in Business: A Literature Review. Pick, J.B.; Ed., Geographic Information Systems in Business. Idea Group Publishing, Hersey, PA, USA, pp. 151-174, 2005.

• [50] Jarupathirun, S.; Zahedi, F.; GIS as Spatial Decision Support Systems. Pick, J.B.; Ed., Geographic Information Systems in Business. Idea Group Publishing, Hersey, PA, USA, pp. 151-174, 2005.

• [51] Churchman, C.W.; The Systems Approach and its Enemies. ISBN: 9780465083428, Basic Books Inc., New York, USA, 1979.

• [52] European Roadmap Infrastructure for Green Vehicles. 2012, www.ertrac.org/pictures/ downloadmanager/6/54/infrastructure-for-green-vehicles_final-october-2012_65.pdf.

• [53] Goldman, T.; Gorham, R.; Sustainable Urban Transport: Four Innovative Directions. Technology in Society, vol. 28, pp. 261-273, 2006.

• [54] Wu, Y.C.J.; Lee, P.L.; The Use of Patent Analysis in Assessing ITS Innovations: US, Europe and Japan. Transportation Research, Part A, vol. 41, pp. 568-586, 2007.

• [55] HIDO; A Bible for Those Involved in ITS Around the World: ITS Review. Ministry of Land, Infrastructure and Transport, Highway Industry Development Organization, Japan, 1997, Retrieved February 2008, www.hido.or.jp/ITSHP_e/Rev/review2- 4/review4/1-3e.html.

• [56] C-ITS;Key to Innovation Integrated Solution: Cooperative Intelligent Transport Systems and Services (C-ITS). April 2013, www.eu-smartcities.eu.

• [57] ETSI; Human Factors (HF). Intelligent Transport Systems (ITS):ICT in Cars. ETSITR 102 762 V1.1.1,DTR/HF-00117, 2010, www.etsi.org.

• [58] Van Voorst Tot Voorst, M.P.; Future of Super Intelligent Transport Systems. EFP Brief, no. 202, 2011, www.foresight-platform.eu/wp-content/uploads/2011/11/EFP-Brief-No.-202_Future-of-Superintelligent- Transport-Systems.pdf.

• [59] Ezell, S.; Explaining International IT Application Leadership: Intelligent Transportation Systems. ITIF, 2010, www.itif.org/files/2010-1-27-ITS_Leadership.pdf.

• [60] De Oliveira Pereira, M.S.; In-Vehicle Information Systems Related Multiple Task Performance and Driver Behavior: Comparison between Different Age Groups. Dissertation Submitted Public Advocacy for the Degree of Doctor in Business Human Kinetics, Specialty Ergonomics, 2009, http://www.repository. utl.pt/bitstream/ 10400.5/1232/1/In-Vehicle%20Information%20System-Related%20Multiple%20Task %20Performance%20and%20Driving%20Behaviour-%20Comparison% 20of%20Two. pdf.

• [61] http://ec.europa.eu/research/social-sciences/projects/360_en.html.

• [62] http://www.aide-eu.org/.

• [63] http://ec.europa.eu/transport/road_safety/projects/doc/haste.pdf.

• [64] http://www.ifsttar.fr/en/europe-international/colonne-1/projets-europeens/actions-cost/.

• [65] http://www.noehumanist.org/news.php.

• [66] Brusque, C.; The Influence of In-Vehicle Information Systems on Driver Behaviour and Road Safety: Synthesis of Existence Knowledge Cost 352. Collections of INRETS. Synthesis, no. 54, 2007.

• [67] Lee, J.; Forlizzi, J.; Hudson, S.; Studying the Effectiveness of MOVE: E-Contextually Optimized In-Vehicle Navigation System. In-Vehicle Interfaces, Portland, Oregon, USA, 2005.

  • [68]Brooks, A.; Lenneman, J.; George-Maletta, K.; Hunter, D.R.; Green, P.; Preliminary Examinations of the Time to Read Electronic Maps: the Effects of Text and Graphic Characteristics (UMTRI-98-36). University of Michigan Transportation Research Institute, 1999, http://www.umich.edu/~driving/publications/UMTRI-98-36.pdf

• [69] Burnett, G.; Usable Vehicle Navigation Systems: Are We There Yet?. 2000, www.cs.nott.ac.uk/~geb/pubs.html.

• [70] Ross, T.; Burnett, G.; Evaluating the Human-Machine Interface to Vehicle Navigation Systems as an Example of Ubiquitous Computing. International Journal of Human-Computer Studies, vol. 55, pp. 661-674, 2001.

• [71] Brooks, A.; Nowakowski, C.; Green, P.; Turn-by-Turn Displays versus Electronic Maps: An on-the-Road Comparison of Driver Glance Behaviour (UMTRI-98-37). The University of Michigan, 1998.

• [72] Jackson, P.J.; In Search for Better Route Guidance Instructions. Ergonomics, vol. 41, no. 7, pp. 1000-1013, 1998.

• [73] Adler, J.L.; Blue, V.J.; Toward the Design of Intelligent Traveler Information Systems. Transportation Research, vol. 6, issue 3, pp. 157-172, June 1998.

• [74] Pauzie, A.; Aging Population and Ergonomics Technological Innovation Communication in Driving. Research in Transport Security, vol. 81, pp. 203-212, 2003.

• [75] Pauzie, A.; Ergonomics and Safety of Mobile Systems in the Vehicle: A Reality Pauzie, A. (Ed.); Ergonomics of Communicating Systems in the Vehicle Use and Safety. pp. 97- 102, INRETS. LESCOT, 2001.

• [76] Schraagen, J.M.C.; Information Presentation in In-Car Navigation Systems. Parkes, A.M.; Franzen, S. (Eds.); Driving Future Vehicles. pp. 171-185, Taylor & Francis, 1993.

• [77] Zaidel, D.M.; Noy, Y.I.; Automatic versus Interactive Vehicle Navigation Aids. Noy, Y.I. (Ed.); Ergonomics and Safety of Intelligent Driver Interfaces. pp. 287-307, Lawrence Erlbaum Associates, 1997.

• [78] Daimon, T.; Kawashima, H.; Akamatsu, M.; Drivers' Cognitive Process and Route Guidance. Noy, Y.I. (Ed.); Ergonomics and Safety of Intelligent Driver Interfaces. pp. 273-286, Lawrence Erlbaum Associates, 1997.

• [79] Pauzie, A.; *Ergonomic Guidelines for the Driver System Interface: Help to the Road Safety Design Process.*4th Asia Pacific Conference on Human Computer Interaction, Singapore, Amsterdam, Elsevier Science Ltd, New York, 2000.

• [80] Pauzie, A.; *Ergonomics and Safety of Mobile Systems in the Vehicle. Ergonomic Inspection Guidance Function.*vol.0501,INRETS. LESCOT, 2005.

• [81] Thompson, S.; Nomadic Devices: Distraction Issues and Specific Results. HUMANIST (Ed.), Impact of IVIS on Driver Workload and Distraction: Review of Assessment Methods and Recent Findings. pp. Annex 5, HUMANIST, 2004.

• [82] Vienna Convention on Road Traffic. www.unece.org/fileadmin/DAM/trans/conven tn/crt1968e.pdf.

• [83] Samad, T.; Annaswamy, A.M.; The Impact of Control Technology. Glielmo, L.; University of Sannio, Italy,2011, http://ieeecss.org/sites/ieeecss.org/files/documents/ IoCT-Part4-13VehicleToVehicle-HR.pdf.

• [84] Motsinger, C.; Hubing, T.; A Review of Vehicle-to-Vehicle and Vehicle-to-Infrastructure Initiatives. Technical Report,CVEL-07-003,2007, www.cvel.clemson. edu/pdf/CVEL-07-003.pdf.

• [85] http://friction.vtt.fi/.

• [86] http://www.ertico.com/gst-website.

• [87] http://ec.europa.eu/information_society/activities/esafety/doc/rtd_projects/fact_sheets/p r event.pdf.

• [88] Kellum, C.C.; Deegener, M.; An Analysis of the Differences Between the Evolving United States and European Dedicated Short Range Communications (DSRC) Systems. Detroit, Society of Automotive Engineers, MI, 2007.

• [89] www.centrico.ten‐t.com/documents/briefing%20notes/03_InnovativeProjects_I.pdf.

• [90] http://ec.europa.eu/information_society/activities/esafety/doc/rtd_projects/fact_sheets/c all_4/ comesafety.pdf.

• [91] http://ec.europa.eu/information_society/activities/esafety/doc/rtd_projects/fact_sheets/c all_4/cvis.pdf.

• [92] http://www.heero-pilot.eu/view/en/ecall.html

• [93] http://ec.europa.eu/information_society/activities/esafety/doc/rtd_projects/fact_sheets/c all_4/coopers.pdf

• [94] Gerlach, M.; Trusted Network on Wheels. October 2005, www.ercim.org/publication/ Ercim_News/enw63/gerlach.html.

• [95] Alampi, C.; Euromobility. 2007, www.euromobilityforum.compresentazioni\_11\_04\_ 07/ EMF_Seminar_11‐04‐07_Alampi‐ANAS.pdf

• [96] www.eu-smartcities.eu.

• [97] Proskawetz, K.O.; Klug, S.; Beckert, B.; Key to Innovation Integrated Solution, Cooperative Intelligent Transport Systems and Services (C-ITS). 2013,www.eusmartcities.eu.

• [98] www.its-nds.de.

• [99] Klug, S.;Breuil, D.;De Wilde, H.; Key Innovation Integrated Solution. Smart Organization of Traffic Flow and Logistics,2013, www.eu-smartci ties.eu.

• [100] Ammana, A.; Overview of IntelliDrive/Vehicle Infrastructure Integration (VII). Report Edited by Virginia Tech Transportation INSTITUTE,2009, http://filebox.vt.edu/users/ aaman na/web%20page/VII-IntelliDrive%20Report-edited.pdf.

• [101] Schmidt, N.H.; Kolbe, L.M.; Erek, K.; Zarnekow, R.; Sustainable Information Systems Management. Business & Information Systems Engineering, vol. 5, pp. 400-402, 2009.

• [102] Maruster, L.; Faber, N.R.; Peters, K.; Sustainable Information Systems: A Knowledge Perspective.CAiSE'08 Forum, pp. 57-60, 2008.

• [103] Craglia, M.; De Bie, K.; Jackson, D.; Pesaresi, M.; Remetey-Fulopp, G.; Wang, C.; Annoni, A.;Bian, L.; Campbell, F.; Ehlers, M.; Van Genderen, J.; Goodchild, M.; Guo,

• H.; Lewis, A.; Simpson, R.; Skidmore, A.; Woodgate, P.; Digital Earth 2020: Towards the Vision for the Next Decade. International Journal of Digital Earth, vol. 5, issue 1, pp. 4-21, 2012.

• [104] Jakobsson, A.; Tsoulos, L.; The Role of Quality in Spatial Data Infrastructures. 23rd International Cartographic Conference, Moscow, Russia, 2007.

• [105] Bacastow, T.S.; Bellafiore, D.J.; Redefining Geospatial Intelligence. American Intelligence Journal, The Magazine for Intelligence Professionals, pp. 38-40, Fall 2009.

• [106] Besser, H.; Digital Longevity: Handbook for Digital Projects, A Management Tool for Preservation and Access. Sitts, M. (Ed.); Northeast Document ConservationCenter,Andover MA,pp. 155-166,2000, www.gseis.ucla.edu/~howard/ Papers/sfs-longevity.html.

• [107] Rothenberg, J.; Ensuring the Longevity of Digital Information. An Expanded Version of the Article, Ensuring the Longevity of Digital Documents. Scientific American, vol. 272, no. 1, pp. 42-47,1999,www.clir.org/programs/otheractiv/ensuring.pdf.

• [108] Patel, N.V.; Adaptive Evolutionary Information Systems. ISBN: 1-59140-034-1, 9- 781591-400349, IDEA GROUP Publishing, 2003.

• [109] Bradbury, R.H.; Sustainable Development as a Subversive Issue*.*UNESCO Flagship, Nature & Resources, October 1998.

• [110] Albrechts, L.; Strategic Spatial Planning Re-Examined. Environment and Planning B, Planning and Design, vol. 31, no. 5, pp. 743-758,2004.

• [111] Nadin, V.; The Emergence of the Spatial Planning Approach. Planning Practice and Research, England, vol.22,no. 1, pp. 43-62, 2007.

• [112] Schmidt, S.; Land Use Planning Tools and Institutional Change in Germany: Recent Developments in Local and Regional Planning. European Planning Studies, vol. 17, no. 12, pp. 1907-1921, 2009.

• [113] Healey, P.; Collaborative Planning: Shaping Places in Fragmented Societies. Palgrave Macmillan, Hampshire, 1997.

• [114] Albrechts, L.; Healey, P.; Kunzmann, K.R.; Strategic Spatial Planning and Regional Governance in Europe. Journal of the American Planning Association, vol. 69,no. 2, p. 113, 2003.

• [115] Faludi, A.; The Performance of Spatial Planning. Planning Practice and Research, vol. 15, no.4, pp. 299-318, 2000.

• [116] Salet, W.; Thornley, A.; Kreukels, A.; Institutional and Spatial Coordination in European Metropolitan Regions. Salet, W.; Thornley, A.; Kreukels, A (Eds.), Metropolitan Governance and Spatial Planning: Comparative Case Studies of European City-Regions. Spon Press, London and New York, pp. 3-19, 2003.

• [117] [Adams, N.; Alden, J.; Harris, N.; Introduction: Regional Development and Spatial Planning in an Enlarged European Union. Adams, N.; Alden, J.; Harris, N. (Eds.), Regional Development and Spatial Planning in an Enlarged European Union. Ashgate, Aldershot, pp. 3-16, 2006.

• [118] Commission of the European Communities CEC; Green Paper on Territorial Cohesion: Turning Territorial Diversity into Strength. 2008, http://ec.europa.eu/regional_ policy/consultation/terco/index_en.htm.

• [119] Faludi, A.; Making Sense of the Territorial Agenda of the European Union. European Journal of Spatial Development, Refereed Articles, vol. 25,2007, www.nordregio. se/EJSD/refereed25.pdf.

• [120] Faludi, A.; Territorial Cohesion and the European Model of Society. Lincoln Institute of Land Policy, Cambridge, MA, 2007.

• [121] Faludi, A.; European Spatial Planning: Past, Present and Future. Town Planning Review, vol. 81, no. 1, pp. 1-22, 2010.

• [122] Allin, S.; Walsh, C.; Strategic spatial Planning in European City-Regions: Parallel Processes or Divergent Trajectories. NIRSA Working Papers Series, September 2010, www. nuim.ie/nirsa/research/documents/WP60_Allin_Walsh.pdf.

• [123] European Spatial Development Perspective.1999, http://ec.europa.eu/regional_policy/ sources/docoffic/officeal/reports/pdf/sum_en.pdf.

• [124] Janowicz, K.; Schade, S.; Broring, A.; Kebler, C.; Maue, P.; Stasch, C.; Semantic Enablement for Spatial Data Infrastructures. Transactions in GIS, 2010, vol. 142, pp. 111-129, 2010.

• [125] Paskin, N.; Identifier Interoperability- A Report on Two Recent ISO Activities. D-Lib Magazine, ISSN 1082-9873, vol. 12, no. 4, April 2006, www.dlib.org/dlib/april06/pas kin/04paskin.html.

• [126] Tolk, A.; Beyond Technical Interoperability Introducing a Reference Model for Measures of Merit for Coalition Interoperability. Command and Control Research and Technology Symposium (CCRTS), Washington DC, 2003.

• [127] Turnitsa, C.; Tolk, A.; Battle Management Language: A Triangle with Five Sides. Paper 06SSIW-016, Spring Simulation Interoperability Workshop, Simulation Interoperability Standards Organization, IEEE CS Press, April 2006.

• [128] Groot, R.; McLaughlin, J.; Introduction: Geospatial Data Infrastructure - Concepts, Cases and Good Practice. Groot, R.; McLaughlin, J. (Eds.), Oxford University Press, Oxford, 2000.

• [129] Bernard, L.; Kanellopoulos, I.; Annoni, A.; Smits, P.; The European Geoportal One Step Towards the Establishment of a European Spatial Data Infrastructure. Computers, Environment and Urban Systems, vol. 29, pp. 15-31, 2005.

• [130] Arms, W.Y.; Hillmann, D.; Lagoze, C.; Kraft, D.; Marisa, R.; Saylor, J.; et al.; A Spectrum of Interoperability. D-Lib Magazine, vol. 81,2002, www.dlib.org/dlib/ january02/ arms/01arms.html.

• [131] Gordon, D.; What Is Interoperability and Why Does It Matter?.2003, http://resource.mapinfo.com/static/files/document/1074288321775/mapinfomag_summ er2003.pdf.

• [132] Goodchild, M.; Egenhofer, M.; Fegeas, R.; Interoperating GISs: Report of a Specialist Meeting held under the Auspices of the VARENIUS Project.1997, www.ncgia. ucsb.edu/Publications/Varenius_Reports/Interop.pdf.

• [133] McKee, L.; Buehler, R.; The Open GIS Guide. Open GIS Consortium Inc.,1998, www.OpenGIS.org/techno/specs.htm.

• [134] Stoimenov, L.;Derdevic-Kajan, S.; Framework for Semantic GIS Interoperability. University of Facta, Ser. Math. Inform., vol. 17, pp. 107-125, 2002.

• [135] Garvin, D.; Managing Quality: The Strategic Competitive Edge. The Free Press, New York, p. 319, 1988.

• [136] Lillrank, P.; Laatuajattelu Quality Thinking. 1998, Otava, Cited in Jackobson and Tsoulos, Keuruu, 2007.

• [137] Chrisman, N.R.;A Theory of Cartographic Error and its Measurement in Digital Data Bases. Proceedings of Auto Carto, vol. 5, pp. 159-168, 1982.

• [138] Goodchild, M.; Gopal, S.; The Accuracy of Spatial Databases. Taylor & Francis Ltd, p. 290, 1989.

• [139] Buttenfield, B.P.; Research Initiative 3: Multiple Representations. Closing Report, National Centre for Geographic Information and Analysis, Buffalo, 1993, http://www. ncgia.ucsb.edu/Publications/Closing_Reports/CR-3.pdf

• [140] Guptill, S.C.; Morrison, J.L.; Elements of Spatial Data Quality. International Cartographic Association, Elsevier, Oxford, 1995.

• [141] Goodchild, M.; Jeansoulin, R.; Data Quality in Geographic Information, From Error to Uncertainty. Hermes, Paris, 1998.

• [142] Assink, M.; Groenendijk, N.; Spatial Quality, Location Theory and Spatial Planning. Regional Studies Association Annual Conference, Understanding and Shaping Regions, Spatial, Social and Economic Futures Leuven, Belgium, April 6-8, 2009, http://doc.utwente.nl/71952/1/Assink09spatial.pdf.

• [143] www.unohrlls.org/UserFiles/File/IPoA.pdf.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 8

STANDARDS OF COMMUNICATIONS IN THE INTELLIGENT TRANSPORT SYSTEMS (ITS)

Maaruf Ali*

Department of Computer Science & Software Engineering, College of Computer Science and Engineering, University of Hail, Hail, Saudi Arabia

ABSTRACT

This chapter covers the latest standards in communication in intelligent transportation systems, especially related to inter-vehicular communication. An overview of the IEEE 802.11 protocol standard relevant to ITS communication is discussed. Along with the latest standardizations in the Dedicated Short Range Communication, DSRC, specification and the IEEE 1609 Wireless Access in Vehicular Environments (WAVE) family of standards. Fully distributed technology as well as infrastructure based and centralised cellular networks such as 3G and 4G are also briefly covered.

Keywords: IEEE 802.11p, Dedicated Short Range Communication, DSRC, IEEE 1609, Wireless Access in Vehicular Environments, WAVE, Inter-Vehicular Communications, IVC, Telematics, Continuous Air Interface Long and Medium range, CALM, Traffic Message Channel, TMC, Transport Protocol Experts Group, TPEG, Multimedia Broadcast Multicast Service, MBMS, VeINS, Vehicles in Network Simulations

1.INTRODUCTION

This chapter is a high level tutorial overview based partly on the presentation by Dressler and Casetti [1] on inter-vehicular communications (IVC) presented at IEEE ICC 2012. This chapter will concentrate purely on those aspects of the various standards and protocols related to ITS communication, covering briefly some applications then the DSRC/WAVE standard.

E-mail: m.maaruf@uoh.edu.sa.

236 Maaruf Ali

The protocols and standards are explained with relevance to vehicular communication including the use of centralised and distributed systems.

The need for inter vehicular communication grew out of mainly safety considerations along with the need to maintain and increase road safety between the drivers with the ever increasing traffic density. Initial communication was mainly of the broadcast type from a central station disseminating traffic information. The need for a feedback channel between vehicles was realized from early on and then consequently the need for inter-vehicular communication for safety applications such as maintaining safe braking distances. The main constraints of a vehicular network are the need to develop a system within the constraints of a real-time system, keeping network latency to safe reaction time limits.

Communication between vehicles may take place using the existing standards whether via cellular means such as 3G or 4G or by means of satellite communications. This chapter will address specifically those radio communication technologies pertinent to inter-vehicular communication.

2. THE TRANSPORTATION SYSTEMS

For a better understanding in devising suitable communication systems suited to the application, the differing types of transportation systems need to be defined.

Figure 1. The Intelligent Transportation System (ITS) network (original diagram inspired by [2]).

Intelligent Transportation Systems deals with the long distance transmission of digitised multimedia or hypermedia realtime and non-realtime information, that is, telematics between vehicles and vehicles to other network entities, stationary and non-stationary, terrestrial and non-terrestrial. Thus the flow of information also includes to those encompassing the railway, airway and maritime domains. Their fundamental basis of access and disseminating this information is currently based on radio transmission using specialised protocols and currently established international standards.

Figure 1, shows the rather complex inter-relationships between the various and rather diverse vehicular networks ranging from satellites, cars, ships, planes, trains to cellular networks and the LAN and MAN computer networks. The arrows between the various network entities have been omitted so as to make the diagram less cluttered. It can clearly be seen that communication between the network entities will require the use of different protocols, frequencies and power of transmission. The ITS network is clearly not static and is constantly evolving in terms of both technology and diversity of uses, beyond its adoption by merely automobiles.

3. CURRENT AUTOMOTIVE RELATED ITS PROJECTS

These are the three major projects for ITS currently under investigation. These are the Dedicated Short-Range Communications (DSRC) system, the Telematics using the 5 GHz and 63 GHz radio bands and the Continuous Air interface Long and Medium range (CALM) initiative. These three projects also come under the umbrella of the European Commission"s eSafety drive specially related to road safety.

The DSRC is designed for short range communication between the vehicle and the road side curb, specifically at designated places such as near parking meters or toll gates. Using such a system will also mean having to accept electronic forms of payment too for such things as parking charges or toll fees.

Specific wireless systems for ITS are being tested and trialled to offer telematics using the 5 GHz and 63 GHz frequencies. Using such frequencies by their nature means the systems are designed to be line-of-sight operation only using relatively low power. However, they do offer higher bandwidth and hence higher data rates to carry complex and detailed information. To ensure international operability, these frequencies and channels do off course have to be standardised and agreed up, specially ratified by the International Telecommunications Union.

The CALM system is designed to use both existing networks and future networks to provide long and medium range continuous services between the vehicles and to their nonvehicles. They can use the existing cellular networks such as 3G, 4GLTE, 4G, WiFi and or WiMax. They are also designed to operate on the newer standardised ITS frequencies of 5 GHz and 63 GHz. Because of the higher frequencies and greater bandwidth, both safety and entertainment applications for the passenger and driver can be easily accommodated. The type of data may include 3D video and immersive multi-surround sound.

It should be noted that the railways have adopted the use of GSM for signalling, known as GSM-R, whose frequencies are mainly in the 400 MHz band. Thus ITS systems for 238 Maaruf Ali

railways should also be able to access the GSM-R network. The main difference between standard GSM and GSM-R is the use of different frequency bands.

The ITS system not only encompasses land based systems but also the maritime and aeronautical systems extending to communication with satellites. Typical services include providing cellular and WiFi onboard connectivity onboard aeroplanes to those serving the air traffic controllers for critical real-time navigation purposes. Maritime usage also concerns primarily navigation and safety use both for sea and for the inland waterways of many countries. For both the maritime and aeronautical specifications, these are governed by the European Telecommunications Standards Institute, ETSI.

3.1. Infrastructure-Assisted Data Exchange

This is using the existing technologies to exchange data between vehicles and between vehicles and other network entities, whether terrestrial, maritime or space based. These technologies are currently based on 2G, 3G and 4G networks. These technologies are used mainly by navigation units in vehicles. But there are off course limitations with such technologies, such as radio coverage outages when transiting through a tunnel or loss of satellite signal when driving under heavy foliage. Thus other technologies are being experimented with and trialled such as Ad Hoc Multihop Broadcasting to counteract the problems of loss of cellular coverage including the problems of deep fading.

Willke et al., [3] partitions into two the major types of data that need to be exchanged in inter-vehicle communications, these being non-safety and safety messages. Safety messages are identified as those that alert the driver to danger and require immediate action. These are best served by short-range radio broadcast with its low network latency. The non-safety data encompasses such things as entertainment data as one example. Such non-critical and nontime sensitive information can make use of the existing cellular infra-structure as they are network delay tolerant.

Historically, it all started with TMC, traffic message channel and TMC Pro. This was a centralized management of traffic information using FM radio broadcasting. The major drawback was the limitation of providing an uplink channel from the vehicle to the central station, there was not any. However, emergency service providers like the Police force could update information to the central station. The other major drawback was the limited data channel rate. Some of these problems were addressed by the implementation of TPEG (Transport Protocol Experts Group), founded in 1997.

Experiments have already been carried out using infra-structure based network utilising the 3G/3.5G network. However, MBMS (multimedia broadcast multicast service) of the 4G network system is ideal to be utilised for inter-vehicular communication. MBMS is currently not being deployed as no application has been found that requires it, however, this may change with more pressure from the inter-vehicular communications pressure group. Using MBMS can be good for non-safety applications as the end-to-end delay being 120 ms is considered too slow for safety critical applications.

Many problems need to be addressed in considering the type of technology to be adopted for ITS. If wireless communication is used, then interference is a big factor, as it leads to errors in the data rate due to packet collisions. These have to be minimized. Utilising Ad Hoc networks has its own particular topological concerns such as the length of the multi-hop forwarding link, the number of unidirectional links and co-existence with multi-radio and multi-network systems. The very nature of the ITS is both mobility and fluidity of the network topology with often disappearing links as the cars go into tunnels, or cars following each other take different directions and routes. This calls for re-routing by the routing protocols. The speed of the vehicles is also very variable and this leads to a constantly changing and thus a dynamic topology. Privacy and security is also an increasing matter of concern and the permanent presence of an internet connection is also not guaranteed.

3.2. DSRC/WAVE (Dedicated Short Range Communication/Wireless Access in Vehicular Environments)

The IEEE 1609 specification is the latest current standard in ITS communication that covers vehicular networks. As shown in Figure 2, all the layers are designed for vehicular communication including accessing the network and security features. Short distance communication is dealt with by the new IEEE 802.11p[3] standard, this being similar to the WiFi IEEE 802.11a standard.

IEEE 1609.1: Core Systems IEEE 1609.3: Network Services IEEE 1609.2: SecurityIEEE 1609.4: Channel Management

Figure 2. The IEEE 1609/WAVE protocol stack (adapted from [4]).

The physical layer is almost similar to the IEEE 802.11a standard and uses orthogonal frequency division multiplexing (OFDM) using 16 level quadrature amplitude modulation (16QAM) for its modulation. This leads to reduced inter-symbol interference due to multipath and Doppler shift effects, which is commonly faced by fast moving vehicular traffic. The timing parameters have also been doubled and the channel bandwidth reduced to 10 MHz instead of 20 MHz. This consequently leads to a reduced throughput ranging from 3 to 27 Mbit/s instead of 6 to 54 Mbit/s. The system is designed to operate for close inter-vehicular traffic and thus the effective radio wave operating distance is set in the standard to be effective up to a range of 1000 metres whilst the vehicle is travelling at a maximum speed of 200 km/h (125 mph). So the timings had to be changed and the protocol made more robust because of the speed of travel of the vehicles. Thus to make the protocol more robust the 240 Maaruf Ali

timings had been doubled. Interference considerations were also an important factor, thus 5.9 GHz has been reserved for the USA and Europe.

The IEEE 802.11p MAC layer has features with extensions to the IEEE 802.11a standard. The MAC layer has randomized MAC addresses, quality of service (QOS) built-in with priorities, features that are also present in IEEE 802.11e. The MAC layer also supports multichannel and multi-radio access with a new ad hoc mode. The protocol allows for both non-BSS and BSS mode of operation.

QoS is an important factor so that safety critical messages can be transported and relayed as a priority whilst entertainment type of data can be relegated to have a lower class of service. These differing QoS and access control mechanisms have been borrowed from the IEEE 802.11e protocol.

The WAVE channel spacing for both the US and Europe are shown in Figure 3. The frequency range and channels are strictly regulated but there will be no licensing cost to the user. Both the US and the European channel allocation allow for four service channels to be used by applications and one control channel. However, the US having an extended frequency range has two further channels, channel 172 for "Critical Safety of Life Use" and channel 184 for "Hi-Power Public Safety" applications. With both the US and European systems the channel are of 10 MHz bandwidth. Even though they may have the same channel number, they do not all carry the same service of operation.

ETSIES 202 663 V1.10 (2010-01): Intelligent Transport Systems (ITS) is an European Profile standard from the physical and medium access control layer of Intelligent Transport System operating in the 5 GHz frequency band.

1 control and 4 service channels to be used by application

In Europe, ETSI has reserved 5 channels of 10 MHz bandwidth each

Figure 3. The WAVE channel spacing (adapted from [5]).

The control channel relays management and safety information. Two-way communication takes place on the service channel. Slot management is critical and the synchronization of timing is achieved using the timing signal from the atomic clock from the GPS satellite. The standard specifies a 100 ms sync interval including spending 50 ms on the control channel (CCH). Single radios are used, so they have to switch to the CCH at specific known times. To allow for jitter, the timeslots all start with a guard interval.

Car-2-Car communication is not defined in DSRC/WAVE but a frame format has been provided to carry such messages. First applications include an application that make the drive aware of his/her current surroundings which include the time, speed, position and heading of other vehicles. This is known as "Cooperative Awareness Messaging" or CAM.

To summarize, there are two methods of relaying traffic information systems. Those which belong in the domain of a centralized system, such as using FM Radio, DAB (digital audio broadcasting), 3G/3.5G and those which belong in the decentralized domain, such as using broadcasting, Ad Hoc routing and peer-to-peer networking.

Historically an early traffic information system was known as "SOTIS" or "Self Organizing Traffic Information System". This was a fully distributed system with no communication infrastructure. It was a basic and simple system with no congestion control. It worked on exchanging a local knowledge base using beacons. The beacons would aggregate all the sensed and received data with the beacon transmitting every five seconds. This system, however, suffered from many shortcomings such as limitation of channel capacity and beacon interval. To gain market penetration of any system, it has to be realized that real networks are heterogeneous and cross platform interoperability is a must for future adoptability and market penetration for the system to be adopted by the public. Thus any system that is to be designed must be able to operate with the existing legacy systems.

Further details regarding beaconing approaches may be found in [6,7,8] and peer-to-peer in [9,10].

Peer-to-peer traffic information system may also be implemented using a hybrid 3G/4G network [5]. The idea behind this was to use all the cars as a massively distributed system. To ease the data processing, the data for the geographical map can be tiled with each vehicle carrying a portion of the map.

Communication between vehicles will also be dependent on traffic density, speed of movement, terrain, obstructions in the propagation and permitted frequencies of operation. There will also be the sensitive issue of different cultural patterns in driving behaviour between nations, age groups, socio-economic background of the driver and the gender. Such factors may affect how communications between these vehicles are in fact modelled. It may be appropriate to use different models for different road types and nations, for example for fast moving autobahn traffic versus those suited to slow moving urban traffic with dense populated traffic volume interacting with the data services of the conurbation. Behaviour and swarm studies are also applicable in modelling group driver behaviour.

Some vehicle networking simulations that can be explored by the independent researcher include such tools as: VeINS – Vehicles in Network Simulation. This simulator is based on tools such as OMNeT++ for network simulation and SUMO for road traffic micro-simulation. Further information on VeINS can be found in [12, 13].

Security and Privacy is also a major research topic in its own right and is intimately linked with the various inter-vehicular communication protocols. Tied in with this also probably the most important issue and these are human computer interaction and usability issues. The protocols must be able to incorporate all these features for its adoption by humans from diverse backgrounds.

242 Maaruf Ali

4. SECURITY AND PRIVACY IN INTER-VEHICULAR COMMUNICATION (IVC)

Security provisioning is important not only for security but to actually protect the data in the hazardous and noisy electromagnetic environment that exist in the fast moving and contentious vehicular and road traffic environment. The key factors that need to be considered in a practical working IVC network are such things as the identity of the user or IVC network entity and its management and privacy protection. As mentioned above, it is not always a question of protection from ominous users or situations but simply the protection of data in a very noisy environment. Messages exchanged between IVC network components also require authentication to prevent spoofing or "man-in-the-middle" attacks from occurring. The data being transported between and through the IVC network need to be protected and also be consistent. This off course requires the use of standardized protocol with international recognition. Thus the IEEE 1609/WAVE protocol is one very important and likely contender for global adoption in the IVC network. The European equivalent IVC protocol is the ETSI Working Group 5. Any IVC network will off course require both the determination and secure exchange of location services.

The IVC network must be a tolerant network such that it can prevent attacks from happening in the first place, as prevention is always better than the cure. This also requires the detection of attacks and recovering from them ideally with the user data or signalling information intact. The IEEE 1609/WAVE Protocol does have a security layer in the protocol stack, this being detailed in the IEEE 1609.2 Security specifications. Protection from undetectable attacks, that is, those that are unknown to the user are tackled by focussing on prevention; whereas attacks that are aggressive in nature are best dealt with by focussing on detection. The issues and experiences from wireless mobile communications and WiFi usage can be applied to the wireless IVC network. Extensive knowledge can be applied from the use of MANETs (Mobile Ad Hoc Networks) and VANETs (Vehicular Ad Hoc Networks) as well. The concepts of encryption using keys, digital signatures, data integrity mechanisms such as message authentication have all been adopted to ensure security in the IVC network.

Figure 4 shows how a smart vehicle integrates various modules necessary for the proper functioning of an IVC network. Some of the features shown are actually present in most modern manufactured cars these days, such as radar and GSM mobile communication. The presence of a hardware security module (HSM) clearly shows the importance attached to network security and integrity. Cryptographic data will be processed by the HSM. Each module requires a high end processor in its own right for the realtime processing of the vast amounts of data that need to be processed. Radar both in the front and rear are necessary for collision avoidance for example. It should also be noted that there is an event data recorder, much like the black box of an aircraft. This can keep a log of the journey taken and provide vital data during an breakdown or accident. The information from the vehicular black box can also be useful for the insurance companies where claims are disputed. The vast amounts of realtime data generated need to be communicated between vehicles for realtime traffic management and accident avoidance.

Figure 4. The smart vehicle (adapted from [14]).

5. KEY APPLICATIONS OF IVC

A vitally important application of IVC is that of emergency vehicles switching traffic lights to force a green so that it can get to an emergency site as quickly as possible or to a hospital. This can actually be taken further in that traffic lights further across the route of the ambulance for example can be controlled to regulate traffic and relieve congestion in certain hotspots along the trajectory of the emergency vehicle.

Dynamic pricing [14] is also one possible revenue generating stream likely to be adopted by the various international highway agencies. "The driver (possibly assisted by a navigation system) decides on a route; the payment of any tolls automatically occurs when entering the toll road or bridge" [14]. This would require the seamless communication between the vehicle and the toll gate charging communication system.

A stationary vehicle also needs to be aware of its surrounding traffic. This can be important if the car is hit by another car. Another example may be that the car can track and monitor another car that has been involved in a criminal incident. The speed, data and time may be used in a court case or help in the identification of the fleeing vehicle [14].

Verifiable multilateration is also possible with IVC networked vehicles [14]. If there are four such vehicles then the target vehicle can be positioned in 3d space. If three IVC networked vehicles are present, then the target vehicle can be positioned in 2d space. The positioning accuracy can be further augmented with GPS and other satellite navigational data.

What most drivers will feel happy with is the concept of "cooperative driving". This is when a vehicle has a special "electronic token" that allows it to safely navigate through a busy blind crossing or approach a busy congestion safely. This is made possible because of 244 Maaruf Ali

the sharing and exchanging of vehicular positioning information and access rights and privileges with each other using a standardized IVC protocol.

6. DISCUSSION

The need for a robust, secure and internationally agreed standard for inter-vehicular communication is now reaching the stage of absolute necessity. This is specially driven by the increasing traffic density in urban cities and the need to manage traffic flows safely and efficiently.

Routing of traffic needs to be managed at peak times to avoid congestions and also to minimize potential accidents at junctions. Safety is of paramount importance as well as the smooth and fast flow of traffic through cities.

If vehicles are considered as network entities then the reliable inter-vehicular communication becomes a necessity. This dictates the adoption of an internationally agreed standard or standards. The current scenario is of two emerging standards, one based on the IEEE 1609/WAVE protocol driven by the North American IEEE institution and the other standard is the European ETSI Working Group 6. Both standards share similarities and have security as one of their utmost concern. It is envisioned that future ITS communication will have to adopt a multi-standard approach, just like cellular communication.

All the applications rely on the secure and robust transport of data between vehicles both moving and importantly also sedentary ones. Communications between other types of vehicles such as aircrafts ships and trains are also incorporated into the standards. Communication is not just limited to terrestrial network entities but also with space bound vehicles such as satellites orbiting at various altitudes.

Applications which cater for the emergency services have a special place in the protocol standards, an example being to allow vehicles such as ambulances safe and fast passage through congested traffic including the control of traffic ahead of the emergency services vehicle. Monitoring of vehicles by each vehicle can also be used for intelligence gathering and crime prevention.

Accurate estimation of a vehicle"s position is essential for the safe monitoring of the trajectories of vehicles to avoid collisions and thus avoid multiple fatalities. This is a special application area that is being researched intensively. The position of a vehicle can be determined by multilateration in three dimensions using four other vehicles equipped with ITS and IVC systems. Again the position of a vehicle can be used for security and crime prevention and detection. GPS and other satellite navigational systems can be augmented with terrestrial and inertial systems for added safety derived from redundancy.

Most applications are safety and traffic flow management based. Revenue generating applications are also starting to appear with the concept of toll collecting based on routing information provided or gathered from the vehicle.

CONCLUSION

The ITS system is a complex, ever evolving network that not only consists of vehicles but other types of mobile and non-mobile entities. This chapter has presented the latest current projects and specifications being applied and adopted for ITS. An overview of the DSRC and CALM systems were given. Technologies covered included the use of the current cellular networks such as the GSM-R network adopted for the railway network and those of the 3G and 4G networks for ITS use. An overview of the IEEE 1609/WAVE specification was also discussed including the coverage of the lower layers of the protocol stack. For an efficient and functioning ITS to operate, interoperability with existing communications protocols and emerging protocols must be ensured. The IEEE 1609/WAVE specification appears to be the successor and the most likely communication protocol for inter-vehicular communication to be internationally adopted as a ratified standard.

REFERENCES

• [1] Dressler, F.; Casetti, C.; Standards, Protocol Design, and Integrated Security Metrics. IEEE Tutorial Lecture on Inter-Vehicular Communication (ICC-2012), http://host. comsoc.org/freetutorial/wiley5/wiley5.html [Accessed 23rd Nov., 2013].
• [2] http://www.etsi.org/technologies-clusters [Accessed 23rd Nov., 2013].
• [3] Jiang, D.; Delgrossi, L.;IEEE 802.11p: Towards an International Standard for Wireless Access in Vehicular Environments. IEEE Vehicular Technology Conference (VTC Spring 2008), doi:10.1109/VETECS.2008.458, pp. 2036-2040, 11-14 May 2008.
• [4] Willke, T.L.; Tientrakool, P.; Maxemchuk, N.F.; A Survey of Inter-Vehicle Communication Protocols and their Applications. IEEE Communications Surveys and Tutorials, doi: 10.1109/SURV.2009.090202, vol. 11, no. 2, pp. 3-20, Second Quarter 2009.
• [5] Uzcategui, R.; Acosta-Marum, G.; Wave: A Tutorial. IEEE Communications Magazine, doi: 10.1109/MCOM.2009.4939288, vol. 47, no. 5, pp. 126-133, May 2009.
• [6] Korkmaz, G.;E kici, E.; Ozguner, F.; An Efficient Fully Ad-Hoc Multi-Hop Broadcast Protocol for Inter-Vehicular Communication Systems. IEEE International Conference on Communications (ICC'06), doi: 10.1109/ICC.2006.254764, vol. 1, pp.423-428, June 2006.
• [7] Sommer, C.; Tonguz, O.K.; Dressler, F.; Traffic Information Systems: Efficient Message Dissemination via Adaptive Beaconing. IEEE Communications Magazine, doi: 10.1109/MCOM.2011.5762815, vol.49, no.5, pp.173-179, May 2011.
• [8] Kargl, F.; Schoch, E.; Wiedersheim, B.; Leinmuller, T.; Secure and Efficient Beaconing for Vehicular Networks.5thACM International Workshop on Vehicular Ad-Hoc Networks (VANET 2008), ACM, San Francisco, USA, pp.82-83, 2008.
• [9] Lee, K.C.; Uichin L; Gerla, M.; TO-GO: Topology-Assist Geo-Opportunistic Routing in Urban Vehicular Grids. 6th International Conference on Wireless On-Demand Network Systems and Services (WONS 2009), doi: 10.1109/WONS.2009.4801842, pp.11-18, 2-4 February 2009.

246 Maaruf Ali

• [10] Rybicki, J.; Scheuermann, B.; Koegel, M.; Mauve, M.; PeerTIS: A Peer-to-Peer Traffic Information System. In Shorey, R.; Weimerskirch, A.; Jiang, D.; Mauve, M.;Ed., Vehicular Ad-Hoc Networks. ACM, pp. 23-32, 2009.
• [11] http://veins.car2x.org/ [Accessed 23rd Nov., 2013].
• [12] Sommer, C.; German, R.; Dressler, F.; Bidirectionally Coupled Network and Road Traffic Simulation for Improved IVC Analysis. IEEE Trans. on Mobile Computing, vol. 10, no. 1, pp. 3-15, Jan. 2011.
• [13] Sommer, C.; Dietrich, I.; Dressler, F.; Realistic Simulation of Network Protocols in VANET Scenarios. 26th IEEE Conf. on Computer Communications (INFOCOM 2007), IEEE Workshop on Mobile Networking for Vehicular Environments (MOVE 2007), Poster Session, Anchorage, AK, pp. 139-143, May 2007.
• [14] Hubaux, J.P.; Capkun, S.; Luo, J.; TheSecurity and Privacy of Smart Vehicles. IEEE Security and Privacy, doi: 10.1109/MSP.2004.26,vol. 2, no.3, pp. 49-55, May-June 2004. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.127.1164&rep=rep1& type=pdf.

PART II. SMART TECHNOLOGIES APPLIED IN THE ELECTRICAL AND HYBRID VEHICLES

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 9

POWER ELECTRONICS AND DRIVES USED IN AUTOMOTIVE APPLICATIONS

Ersan Kabalci*

Electrical and Electronics Engineering Department, Faculty of Engineering and Architecture, Nevsehir University, Turkey

ABSTRACT

This chapter deals with power electronics and drives used in electric vehicles (EV). The power electronic devices are classified in four categories as rectifier, inverter, DC-DC converter, and cycloconverter according to AC and DC conversions together and vice versa. The converters are mostly utilized after the rectifier block to ensure supplying the battery charger with a stable DC voltage, at the output of battery or fuel cell, and before DC motor drives. The propulsion systems of electric, hybrid electric and plug-in hybrid electric vehicles are based on electrical machines. An electric machine used in an EV is chosen according to its construction structure, efficiency, maintenance requirements, electromagnetic interference (EMI), reliability, and speed range parameters. The rapid improvements seen in electrical machines and power electronics devices also contribute the development of EVs. There are a few electric motors among others are known as workhorse in power-train applications. The induction machines (IM), permanent magnet synchronous machines (PMSM), and brushless DC machines (BLDC) are the several electric machine types used in EV power-train applications with specific controllers. The machine controllers are power electronics systems that are combined with sensors and observers in order to ensure the wide speed range in constant torque situations and extended speed range for constant power requirements. Although the power electronics devices and motor drives cover a wide research area, this chapter is dedicated to specific applications used in EV technology. The related topics are arranged from plug to propulsion system for an EV.

Keywords: Power Electronics, Converter, Inverter, Cycloconverter, Electric Vehicle, Electric Machines, PMSM, BLDC, ACIM

E-mail: kabalci@nevsehir.edu.tr.

250 Ersan Kabalci

NOMENCLATURES

ARCP Auxiliary Resonant Commutated Pole

BLDC Brushless DC Machine

CCM Continuous Conduction Mode

CHB-MLI Cascaded H-Bridge Multilevel Inverter

CSI Current Source Inverter

DCM Discontinuous Conduction Mode DC-MLI Diode Clamped Multilevel Inverter

EMF Electromagnetic Force EMI Electromagnetic Interference

EV Electric Vehicle

FACTS Flexible AC Transmission Systems FC-MLI Flying Capacitor Multilevel Inverter

FCV Fuel Cell Electric Vehicle FOC Field Oriented Control HEV Hybrid Electric Vehicle ICE Internal Combustion Engine

IM Induction Machine MLI Multilevel Inverter

PEEM Power Electronics and Electrical Machines

PEV Plug-In Electric Vehicle

PHEV Plug-In Hybrid Electric Vehicle

PI Proportional-Integral

PID Proportional-Integral-Derivative PMBL Permanent Magnet Brushless

PMSM Permanent Magnet Synchronous Machine

UPS Uninterruptible Power Supplies VSI Voltage Source Inverter

1.INTRODUCTION

The plug-in electric vehicles (PEV) are known as a subcategory of electric vehicles that also include pure electric vehicles (EVs) and plug-in hybrid vehicles (PHEVs). Furthermore, the PEVs are defined as electric vehicle conversions of hybrid electric vehicles (HEVs) and conventional internal combustion engine (ICE) vehicles. Although a HEV has battery configuration that is consistently recharged by using the power obtained from the ICE and regenerative braking. Since HEVs cannot be recharged from any off-vehicle electric energy source, they are not categorized in plug-in electric vehicles. A PEV can be defined as an electric motor vehicle that obtains the required energy for propulsion over the electric grid and/or the charged batteries located on board [1-3].

The PEVs require a detailed power electronics and electrical machines (PEEM) control to operate efficiently as other advanced electric drive vehicles such as HEVs, PHEVs, fuel cell electric vehicles (FCVs), and EVs. The power electronics devices convert the grid electricity

to the required ratings and various current waveforms as direct or alternating. The electric drive system in EVs is based on direct current (DC) and alternative current (ac) energy conversion stages. The complete drive system is defined with functions of converter, battery, inverter, electric motor, and wheels. The inverter converts the DC line voltage to AC and supplies the electric motor to generate propulsion in wheels [3,4]. On the other hand, the power electronics and electrical machine stages of PHEVs are more complex according to HEVs. This vehicle configuration uses two electrical machines; one as a generator connected to the engine and another as a motor to drive the wheels. Each of these machines is connected to an inverter that runs at a higher voltage levels comparing to the battery.

The boost converter increases the DC voltage that will be applied to inverter. Main components of a hybrid and an electric vehicle include motors, motor controllers, DC-DC converters, and inverters that condition the electrical energy between the power supply (a fuel cell or battery) and electric motor to provide power to various components, sensors, control systems, and other interface electronics. The electric motor drive converts the stiff DC battery voltage to DC (for DC motor) or AC (for AC motor) voltage levels with controllable amplitude and frequency. Besides performing the propulsion of PEVs with induction motors, the DC motors are used in order to manage the auxiliary systems such as air conditioning (A/C) compressor and pumps [5-7].

The input command of the motor drive is converted to torque command that sets the operating point parameters of the electric motor. The torque command, in conjunction with the feedback signals from sensors, performs by adjusting the turn-on and turn-off transitions of the power switches through the drive system. The motor drive then shifts the power at the desired voltage and frequency to the motor, which in turn delivers the desired torque and speed for propulsion [7,8].

This section covers power electronic converters and inverters, DC and AC electric motors that are most widely utilized in vehicle technology and motor drives specifically for PEVs. The PEEM subsections focus on actual and practical electric vehicle technologies instead of pure power electronics and introduces the application areas of widely known power electronics in PEVs.

2. POWER ELECTRONICS USED IN PEVS

Power electronics is a wide research area owing to its circuit topologies converting the power from AC to DC or vice versa. In addition to current waveform conversions, AC-AC cycloconverters or DC-DC converters are used to generate the required various frequency, current, and voltage levels. The power electronics circuits and components used in a PEV are depicted in Figure 1 where all the components are introduced in the following subsections. Even though the electric motor in the figure is shown as actuating the rear wheels, the front wheel propulsion is widely used referring to actual automotive technology.

The main topics of power electronics required for a PEV design cover:

Rectifiers: The rectifiers are used in ac-DC conversion stages of PEVs where the grid electricity is firstly met with this device. The grid voltage obtained from plug is 110V/60Hz or 220V/50Hz single-phase AC input in household application of PEV. Rectifiers are analysed in ideal circuit topology and uncontrolled or controlled power 252 Ersan Kabalci

• switch topologies. Since the automotive applications are mostly based on uncontrolled rectifiers, controlled rectifiers will not be analysed in this section.
• Power Converters: Power converters are classified according to their input and output types namely DC-DC converters and AC-AC cycloconverters. The DC-DC converters are designed to increase or decrease the input voltage stage to desired value at the output. Therefore, a device known as buck or step-up converter decreases the output value according to a high input voltage while boost or step-up converter generates an increased output voltage. The cycloconverters are used to meet high voltage and high power requirements of large motors where frequency and voltage levels are adjusted by converter. The AC-AC converters are not used in electric vehicles and the frequency or AC voltage requirements are provided by inverters that may be named as DC-AC converters. The cascaded inverter topologies are introduced in brief besides widely used full bridge inverters in PEVs. Control techniques of inverters are also focused in a following subsection.
• Motors and Motor Controllers: Motors or electric drives used in electric vehicles are classified in two main categories as brushed or brushless. Induction motors, brushless DC motors (BLDC), and permanent magnet brushless (PMBL) motors that are widely used PEV electric drives are analysed in terms of physical structures and controllers in a separate subsection.
• Controller and sensors: Almost all the power electronics in PEV require control signals as met in industrial applications. The required control signals are generated by microcontrollers and commutate power electronic devices with switches to provide fixed speed or torque. To obtain the instant control action, interface electronic devices are used besides current and voltage sensors located in the related parts.

Figure 1. Power electronics and electrical components used in a PEV.

The power electronic as being a discipline of electronics itself, the devices mentioned above are mostly known by anyone who is familiar to this research area. Therefore, the following topics are dedicated to specific power electronics used in PEVs instead of analysing entire variations of devices.

3. RECTIFIERS

The rectifier is a circuit that converts AC to DC owing to diodes in uncontrolled structure or thyristors in controlled structure. A general overview is given in Figure2 where the input utility voltage is shown with Vin, while f and q stand for line frequency (50Hz/60Hz) and phase number, respectively.

Figure 2. General rectifier overview.

The half-wave rectifiers are used in low power or industrial applications to isolate load from power line and are widely preferred because of its low cost installation. The power output and therefore, the rectification efficiency is quite low besides its low transformer utilization ratio and high ripple factor requiring a robust filtering[9-11]. The full-wave or bridge rectifiers generate DC output voltage during each half cycle of a period on the contrary of half-wave rectifiers [12, 13].

Figure 3. Single phase full-wave rectifier.

The full-wave rectifiers can be accomplished with two different circuit topologies that are known as center-tapped and bridge rectifier. The bridge rectifier depicted in Figure3 requires 254 Ersan Kabalci

four diodes while it is enough to use two diodes in the center-tapped rectifier due to transformer structure [9]. In the full-wave rectifier with center-tapped transformer, each diode acts as a half-wave rectifier using the tapped half of line transformer and output generates a full-wave waveform to supply the load [11]. The average value of the load voltage (VL) is defined as Vdα in Figure 3 and calculated as given below;

$$V{d\alpha} = \frac{1}{T} \int{0}^{T} v_L(t)dt \tag{1}$$

If this generic equation is rearranged for full-wave bridge rectifier;

$$V{d\alpha} = \frac{1}{\pi} \int{0}^{\pi} V_{m} \sin \omega t \quad d\omega t \tag{2}$$

is obtained. Therefore,

$$V_{d\alpha} = \frac{2V_m}{\pi} = 0.636 \cdot V_m \tag{3}$$

The root mean square (rms) value of load voltage,

$$V{L{-}rms} = \frac{V{m}}{\sqrt{2}} = 0.707 \times V{m} \tag{4}$$

Figure 4. Single-phase full-wave rectifier simulations with RL loads.

The rms value of a full-wave rectified voltage should be equal to that of the original AC voltage as proved in Equation (5) [11-13]. The supply and load waveforms of rectifier with RL load are seen in Figure4. Each diode pairs (D1D3-D2D4) are switched on during positive and negative half cycles of source voltage (Vd) provided by transformer. Therefore, the output current flow is continuously obtained in sinusoidal waveform and there is not any phase shift occurred with output voltage. The load current waveform seen in Figure4 is obtained with an R-L load where a 2mH inductance is selected to observe effect of the inductance on current waveform.

The load voltage is seen in sinusoidal waveform while I^d is distorted by odd harmonics and a phase shift is occurred between V^d and Id. There a distortion is also seen in source current because of source inductance. The bridge diode currents are seen in square waveform in high inductive loads that is a usual case seen in industrial applications [9,11,13].

4. DC-DC CONVERTERS

The DC-DC converters in PEVs are required to keep the output voltages stable against varying source voltages supplied by fuel cells or batteries. Furthermore, the variable and/or stable torque and speed requirements under several conditions also involve DC-DC converters. All the electronic equipments in PEVs require a fixed DC input voltage. This requirement is accomplished by decreasing or increasing the input voltage of the converter and performing circuit is called with its operating principle where a buck converter (stepdown) adjusts the output voltage by decreasing the source voltage or a boost converter (stepup) that increases the output voltage to a fixed value. There is a third type of converter namely buck-boost is widely used in PEVs that performs specific features of both buck and boost converters together. These converters are constituted with a high frequency switching devices and reactive components to ensure operating in the continuous conduction mode [14,15].

The converters are mostly utilized after the rectifier block to ensure supplying the battery charger with a stable DC voltage, at the output of battery or fuel cell, and before DC motor drives that are fed by auxiliary batteries as illustrated in Figure1 [16-17]. The following subsections deals with the most widely used converter topologies such as buck converter, boost converter and buck-boost converter with design criteria for PEVs.

4.1. Buck (Step-down) Converter

The buck converter, which is also known as step-down converter, is shown in Figure5. It involves input source (Vs), controlled switching device (S), a diode (D), low-pass filter with L and C, and load resistance. Inductor and capacitor ensure to generate output voltage at a limited ripple ratio and filters the output according to requirements of the load. The filtering operation affects the circuit analysis mode depending to inductor value. The inductor current never drops to zero if the inductor value is high enough and this analysis mode is called continuous conduction mode (CCM). In case of inadequate inductor values, inductor current drops to zero for a while in each half-cycle where this mode is known as discontinuous 256 Ersan Kabalci

conduction mode (DCM). There are separate analysis required for CCM and DCM and the relation between these modes is described with boundary analysis method.

Figure 5. Circuit diagram of buck converter.

The relationship among the input voltage, output voltage, and the switch duty ratio D can be derived from the inductor voltage V^L waveform. The DT in Figure6 expresses duty ratio or the interval where the switch is on while the left interval shows off state. Inductor and output currents are seen in the first curve of Figure6 [9-11].

Figure 6. Waveforms of buck converter.

The voltage equation of the circuit is calculated by using maximum and minimum inductor currents,

$$I{\text{max}} - I{\text{min}} = \left(\frac{V_S - V_C}{L}\right) DT \tag{5}$$

$$I{\min} - I{\max} = \left(-\frac{V_C}{L}\right)(1 - D)T \tag{6}$$

$$V_C = DV_S \tag{7}$$

The minimum inductor value providing the converter to operate in CCM is determined as,

$$L = \left(\frac{TR}{2}\right) (1 - D) \tag{8}$$

4.2. Boost (Step-up) Converter

A boost converter, which the circuit diagram is shown in Figure7, is a power converter with an output DC voltage greater than its input DC voltage. It is a class of switching-mode power supply containing at least two semiconductor switches (a diode and a controlled switch) and at least one energy storage element (capacitor and/or inductor). Filters made of capacitors are normally added to the output of the converter to reduce output voltage ripple and the inductor connected in series with the input DC source in order to reduce the current ripple. The diode is reverse biased when the switch is on and therefore conduction does not occur in this state. The load is supplied in the off state of power switch. The value of output capacitor should be high enough to obtain a stable output voltage in steady-state analysis [18,19].

The maximum and minimum values of inductor currents are seen in Figure 8 where currents of switch, capacitor, and load voltage are also given respectively.

The boost converter is also analysed for CCM and DCM, and the relation between these modes is described with boundary analysis method. The inductor current (iL) is differentially increased during on state,

Figure 7. Circuit diagram of boost converter.

258 Ersan Kabalci

Figure 8. Waveforms of boost converter.

The capacitor voltage (VC) should be greater than V^S in order to decrease the inductor current. The inductor current in the off state of switch is calculated as follows,

$$\frac{di_L}{dt} = \frac{V_S - V_C}{L} \tag{10}$$

Since the increment and decrement of i^L should be equal in on and off states, the voltage relation of converter is defined as,

$$\frac{di_L}{dt} = \frac{V_S - V_C}{L} \tag{11}$$

4.3. Buck-Boost Converter

The buck-boost converter namely step-up/down is a popular non-isolated, inverting power stage topology as shown in Figure 9. The buck-boost converter is selected because of the output voltage is inverted from the input voltage, and the output voltage can be either higher or lower than the input voltage. The diode voltage is reversed when the switch is turned to on state and the diode current (iD) is obtained at zero value in this case. Furthermore, the source voltage is transferred to inductor and i^L increases as a function of time in conduction mode. The connection between the source and inductor is leaved by switching off the S and i^D gets equal to i^L in this state. The mentioned on and off states are used in steady state analysis as given in the following equation for each operation mode [18,19]. The output voltage equation is obtained by equalling the variations of inductor currents as given in Equation(12).

$$V_C = \frac{D}{1 - D} V_S \tag{12}$$

The operation mode is determined according to value of D where higher values than 0.5 provides to operate in boost mode and lower values to operate in buck mode. The output voltage is negative as seen in Figure 10.

Figure 9. Circuit diagram of buck-boost converter.

Figure 10. Waveforms of buck-boost converter.

5.INVERTERS

The main objective of the power inverters is to generate an AC output waveform by using the DC input. Several industrial applications such as uninterruptible power supplies (UPS), static VAR compensators, active filters, flexible AC transmission systems (FACTS), and voltage compensator require AC waveforms. The magnitude, frequency, and phase should be controllable for sinusoidal AC outputs. In electric vehicles, inverters are used to provide the 260 Ersan Kabalci

required voltage and/or current for AC motors, and consequently the inverter control algorithms perform motor control operations in terms of torque and speed. Inverters are considered as voltage source inverters (VSI) and current source inverters (CSI) according to the type of output generation where VSIs are widely used in many industrial applications and novel topologies are developed to meet the power requirements.

The efficiency parameters of an inverter such as switching losses and harmonic reduction are principally depended on the modulation strategies used to control the inverter [19-21]. There is various control techniques that can be classified into two groups such as feed-back and feed-forward control have been proposed to reduce the switching losses and harmonics in occurred in DC link inverters [22-24]. A complete AC motor drive and control system is shown in Figure 11, where a microcontroller dynamically controls the three-phase full bridge inverter according to feedback data inherited from sensors. The switches of any phase leg in the inverter (S¹ and S4, S³ and S6, or S⁵ and S2) cannot be switched on simultaneously to prevent a short circuit across the DC link voltage supply as considered in the single-phase VSIs. Similarly, the switches of any leg in the inverter cannot be switched off simultaneously in order to avoid undefined switching states and undefined AC output line voltages in the VSI. The phase outputs are mutually phase shifted by 120º angles. The line voltage rates of inverter are determined according to modulation indexes (mi) which define the operating area as in linear modulation ( 1) mⁱ or as over-modulation ( 1) mⁱ range. The line voltages are

limited to ( 3 2 V d ) of DC line in linear modulation range and are limited to 4 3 ( )( ) 2 Vd in over modulation range [25-27].

Figure 11. A complete AC motor drive system with full-bridge inverter for EVs.

The Sinusoidal PWM (SPWM) modulator acts as an amplifier in the linear modulation range with the gain parameter (G) seen as in Equation (13).

$$G = \frac{0.5m_i V_d}{V_p} = \frac{0.5V_d}{V_t} \tag{13}$$

The closed loop control operation is performed to ensure torque or speed stabilisation of AC motor. In recent years, multilevel inverters have gained much attention in the application areas of medium voltage and high power owing to their various advantages such as lower common mode voltage, lower voltage stress on power switches, lower dv/dt ratio to eliminate higher harmonic contents in output voltage and current. The multilevel inverters (MLIs) also have the advantages of lower harmonic ratios of line-to-line voltages fed to load are reduced owing to its switching frequencies when compared to two-level inverter topologies at the same power ratings. The most common MLI topologies classified into three types are diode clamped MLI (DC-MLI), flying capacitor MLI (FC-MLI), and cascaded H-bridge MLI (CHB-MLI). Since MLIs are mostly used in large hybrid electric vehicles instead of EV or PEVs, these topologies are not analyzed in this section. Further information about MLIs can be found in [25].

6. ELECTRICAL MACHINES AND CONTROLLERS

The propulsion systems of electric, hybrid electric and plug-in hybrid electric vehicles are based on electrical machines. The rapid improvements seen in electrical machines and power electronics devices also contribute the development of EVs. There are a few electric motors among others are known as workhorse in power-train applications. Induction machines (IM), permanent magnet synchronous machines (PMSM), and brushless DC machines (BLDC) are the several electric machine types used in EV power-train applications with specific controllers. Due to the developments met in power electronics, various machine control techniques are implemented to increase the efficiency [28-30].

An electric machine used in an EV is chosen according to its construction structure, efficiency, maintenance requirements, electromagnetic interference (EMI), reliability, and speed range parameters. The machine controllers are power electronics systems that are combined with sensors and observers in order to ensure the wide speed range in constant torque situations and extended speed range for constant power requirements [3]. Fixing the gear ratio allows to easily adjust the speed where a high speed electric machine can be used due to its smaller size and weights less in contrast to low speed machines. This section deals with electrical equivalents and controllers of IMs, PMSMs, and BLDCs for EVs. Furthermore, there are various textbooks and application notes introducing electrical machines in detail can be found in literature.

262 Ersan Kabalci

6.1. AC Induction Motor (ACIM) and Controllers

The induction machine, which has two separate operation mode as generator or motor, is used in motor mode for power-train applications. An equivalent circuit can be used to model the induction motor for electric vehicles. The efficiency of the motor can be easily calculated, if the parameters of the equivalent circuit model seen in Figure 12 are known. Besides the starting and maximum torque characteristics of an induction motor, some other steady-state performance characteristics include the variation of current, speed, and losses as the loadtorque requirements.

The rotor of an induction motor can be wounded or squirrel-cage type. A wound rotor is built with a poly-phase winding similar to the stator where the rotor terminals are externally available on the motor. On the other hand, a squirrel-cage rotor has a winding that consists of conductor bars embedded on the rotor iron, which are short-circuited at each end by conducting end rings. The wound rotor is the most common type of ACIM in sizes ranging from fractional horsepower on up. The difference between synchronous speed and the rotor speed is commonly referred to the slip of the rotor [29,31]. The rational slip factor (s) is defined as

$$s = \frac{n_s - n_r}{n_s} \% \tag{14}$$

where the synchronous speed is nsand the rotor speed is depicted with n^r both in terms of revolutions per minute (rpm). The synchronous speed is calculated with Equation (15).

$$n_s = \frac{120 \cdot f}{P} \tag{15}$$

where f is the line frequency in hertz and P is the number of poles in the machine. The equivalent circuit shown in Figure 12 contains the electrical and mechanical power conversion properties of an induction motor in a schematic diagram.

The stator equivalent circuit that is seen on the left hand side up to Rm//X^m magnetizing circuit of Figure 12 is combined with stator resistance (Rs) and stator inductance (Xls). The core losses of stator and rotor are depicted in the magnetizing circuit. The power flow diagram of a polyphase induction motor is shown in Figure 13 by considering input power (Pi), air gap power (Pg), mechanical converted power (Pm), and mechanical output power (*P*0). The m parameter is the number of phase applied to motor while T is torque, and ω^m is the angular velocity.

An induction motor can be operated by vector controls with appropriate feedback signals that are obtained from encoders. This method performs the speed control by a PWM synthesized sinusoidal waveform.

The vector controlled motors that are typically designed include a high efficiency winding, an efficient lamination design, and high-temperature insulation materials. The tighter speed regulation approaching to 0.01% of fixed speed can be attained by vector control according to scalar control methods. A vector controller allows obtaining controllable speed ranges from zero to five times of base speed. The constant horsepower range is also about 3.5 times of the base speed. The smooth stopping and highly efficient operation is obtained by using a vector control. Standard vector controls with dynamic brake resistor are used in many applications.

Figure 12. Modified equivalent circuit model of an induction motor.

Figure 13. Power flow diagram of an induction motor.

The advantages of this approach include low initial cost, reliability, the capability of providing full rated torque from rated speed down to zero, precise speed and torque control, constant output power above the rated speed, and such programmable features as controlling acceleration/deceleration time and tuning. The vector control of an ACIM can be applied with high-performance adjustable-speed applications of EVs. The block diagram of a vector control system is shown in Figure 14. The speed controller is based on a PI regulator that controls the motor slip ratio where the value computed by the regulator is added to the motor speed in order to generate the demand frequency. The latter frequency is also used to generate the demanded voltage in order to maintain the constant V/f ratio of the motor.

The space vector PWM modulator contains three-phase generator, low-pass bus filter, α-β transformation, α-β vector sector, ramp generator, and switching time calculator [32,33]. Three-phase generator provides the modulating signals in 120° degree phase shift and the α-β transformer generates three-phase to two-phase conversion in order to be used in sector preparation. The low-pass bus filter is used to remove fast transients from the DC bus voltage measurement. The feed-forward method is used to compute the voltage vector applied to the motor [34,35].

Another widely used control technique of the induction motor is known as field-oriented control (FOC). The FOC of an induction motor achieves torque (T) and rotor flux (ψr) dynamics in a decoupled way by orthogonal projection of the stator current into torque (T) and flux (ψr) generating components. The orientation angle that allows a complete decoupling

264 Ersan Kabalci

between those two variables is the rotor flux angle, θ. This technique is performed by two basic methods that are namely direct and indirect vector control. The required instant value is obtained by direct measurements using flux sensors or flux estimators in direct field orientation, whereas indirect field orientation is based on the inverse flux model dynamics and there are three possible implementation based on the stator, rotor or air gap flux orientation. The indirect control technique of rotor flux is the most widely used due to its simplicity. Although the FOC methods are attractive, they suffer from one major disadvantage that is related to sensitivity.

Figure 14. Vector control block diagram of an AC induction motor.

The FOC methods are sensitive to parameter variations such as rotor time constant and incorrect flux measurement or estimation at low speeds. The field oriented control block diagram of an ACIM is shown in Figure 15 [36-38]. Since all variables of Figure 15 are expressed in terms of rotating coordinate frame, the electromagnetic torque can be controlled via the imaginary component of the stator current isq [39, 40].

The rotor flux can be commanded by its real part isd according to a first order dynamic with a time constant τ^r , as shown in the following equations [37];

Figure 15. Field oriented control block diagram of an AC induction motor.

$$\psi_r = \frac{L_m}{\taur \cdot s + 1} \cdot i{sd} \tag{16}$$

$$T = \frac{3}{2} \frac{L_m}{L_r} p \cdot \psir \cdot i{sq} \tag{17}$$

Equation(16) expresses the flux estimation while torque estimation is given in Equation(17). The θ calculation is used to find the phase angle of the rotor flux-rotating field [41-43].

6.2. Brushless DC Motor and Controllers

Two separate brushless motor types are known as brushless DC motor (BLDC) and permanent magnet synchronous motor (PMSM). The brushless motors are widely used as inwheel motors due to their high efficiency in electric vehicles. The BLDC term means an AC motor with combination of semiconductor inverter control and rotor position sensor. This combination provides a stable and linear moment-speed characteristic such as classical DC motors. The major advantages of BLDC are compactness, easy control and cooling, low maintenance and low noise, and high reliability besides its efficiency. The BLDC commutation is performed by using simple hall-effect sensors that are compactly located on the motor and provides appropriate current waveforms. The permanent-magnet DC machines are widely found in a wide variety of low-power applications. The field winding is replaced by a permanent magnet, resulting in simpler construction. The permanent magnets provide a number of benefits in these applications. However, a high-resolution encoder is required for the PMSM control mode to generate sinusoidal currents comparing to BLDC [11, 12].

The permanent magnets provide the DC excitation of the field winding in a synchronous. By replacing, the electrical excitation with permanent magnet ensures to eliminate the copper losses. The BLDC machines can be categorized by the position of rotor permanent magnet in which the magnets are mounted on the rotor. The magnets can either be surface mounted or interior mounted. There are several textbooks and publications define the physical structure of BLDC in terms of rotor and stator [28, 31, 40].

Therefore, this part will focus on back electromagnetic force (EMF) properties of BLDC and PMSM and control techniques of brushless motors. The back EMF of BLDC motors are classified in two ways such as trapezoidal-shaped and sinusoidal-shaped. The basic torque and voltage equations of BLDC closely resemble those of brushed DC motors. The PMSM is sinusoidal excited while the BLDC is trapezoidal excited machine motor.

The construction differences between are that while the stator windings of the trapezoidal PM machines are concentrated into a narrow-phase pole, the windings of a sinusoidal machine are typically distributed over multiple slots to approximate a sinusoidal distribution. These differences in construction are reflected in their corresponding motion characteristics. This implies the consequence that the first type of PMSM provides sinusoidal backelectromotive force (back-EMF) generation, and the second type provides trapezoidal back-EMF. The Back EMF is directly proportional to the motor speed while the torque production is almost directly proportional to the phase current. The idealized back EMF and current 266 Ersan Kabalci

waveforms are seen in Figure16 where the BLDC is operating during half cycles of each three-phase. The back EMF waveforms indicate the switching orders of three-phase inverter in torque or speed controller that are shown in Figure 17 and Figure 18, respectively.

Figure 16. Idealized back EMFs (EA, EB, EC) and current signals (IA, IB, IC) of a three-phase BLDC motor.

Figure 17. Block diagram of the torque control of a three-phase BLDC motor.

Figure 17 depicts a torque control scheme for a BLDC motor drive. The desired current I* is derived from the commanded torque T* through a torque controller. The current controller and commutation sequencer acquires the desired current I* from limiter, the position information from the position sensors, and the required current feedback through current transducers, and then generates gating signals. The gating signals are applied to the three-phase inverter to generate the desired phase current to the BLDC machine.

The three-phase inverter is widely used in full-bridge topology. Most of the highperformance traction control systems require torque control to sustain required performance. The speed control of a BLDC shown in Figure 18 is based on torque controller with additional equipments such as speed sensor and speed controller block. The speed control is used in cruise control of traction application where stabilizing the speed is required. The speed sensors acquire the actual rpm (ωr) value of the BLDC to compare the desired reference speed (ωr*). The speed controller block is constituted with classical proportional-integral (PI), proportional-integral-derivative (PID) controllers or advanced software controllers such as artificial intelligence or neural network controllers.

The remainder blocks in the controller perform as in torque controller. The switching signals are generated according to inherited values and are applied to inverter in order to generate three-phase line voltage of BLDC [40, 44, 45].

Figure 18. Block diagram of the speed control of a three-phase BLDC motor.

7. THE LATEST POWER ELECTRONICS IN EVS

The circuit diagram seen in Figure 19 is known as resonant inverter application used for AC motor driving operations. The additional switches located between phase legs and neutral points constitutes the auxiliary resonant commutated pole (ARCP) that provides to operate the inverter in soft switching mode to decrease the switching losses and to increase the efficiency. The auxiliary circuit helps enable the load to be swung to the opposite rail to assure zero turnon voltage. If the auxiliary circuit is not included, then there is a load current constraint to insure zero turn-on voltage.

268 Ersan Kabalci

Figure 19. A zero-voltage-transition inverter for AC motors.

Figure 20. Bidirectional DC-DC converter configurations; (a) cascaded buck-boost, (b) half-bridge,(c) split-π.

Emerging power electronics in electric vehicles also cover the DC-DC converter topologies. There are several topologies developed to increase the conversion efficiency and decrease the switching losses. Besides the unidirectional DC-DC converters introduced in 4th section, the novel applications seen in electric vehicles are based on bidirectional converter topologies. The bi-directional DC-DC converter plays an important role in the four-quadrant EV propulsion system where it covers motoring and generator modes. While the EV drive operates in motoring mode, it consumes supply energy from the batteries. On the other hand, when the EV driver switched into braking mode the batteries are charged by electric motor owing to bidirectional structure of converter. Therefore, the bi-directional DC-DC converter will be the key device in EVs. Bidirectional converters can operate in both motoring and regenerative braking modes, and it is therefore dedicated to four-quadrant operation of EV propulsion systems. The most widely used bidirectional converters are shown in Figure 20 where Figure 20(a) illustrates the cascaded buck-boost while half-bridge and split-π configurations are seen in Figure 20(b) and Figure 20(c), respectively [46, 47].

Although the cascaded buck-boost converter topology decreases the electrical and thermal stresses on switching devices, it requires twice of the active devices according to regular buck-boost topology. The required switch number can be decreased by using halfbridge topology seen in Figure20.b where the passive components are at the same number.

The split-π configuration is relatively a new topology among other DC-DC converters, and has numerous application areas as regenerative braking in electric vehicles. The passive components that are used in a split-π converter are quite smaller in terms of sizes and yields efficiency over 97%. Furthermore, the switching losses, distortions, and ripples on the output current waveform are reduced owing to the topological configuration [46, 47].

8. DISCUSSION

The power electronics and drives have the key role on traction control of electric vehicle technologies. The latest trends involve the most efficient power flow that is realized from battery through the entire vehicle. In order to achieve this, there are numerous studies performed on innovative issues such as highly efficient converters, bidirectional power control, dynamic feedback, and regenerative braking. On the other hand, motor drives are also extensively studied owing to emerging magnetic and electrical materials. This chapter covers the most widely used power electronics and motor drives referring to classical circuit topologies and presenting the novel configurations. It is seen that the energy efficient converters such as resonant or hybrid ones sustain their popularity due to reliable switching specifications at higher frequencies. Another important issue discussed for converter circuits is bidirectional configuration that allows to operate the traction motor as a generator during the braking mode. In this mode, the generated energy is used to charge the batteries if there is a convenient converter configuration used. Consequently, more efficient and high power electric vehicles depend on high performance power electronics.

270 Ersan Kabalci

CONCLUSION

This chapter is dedicated to specific applications used in EV technology although the power electronics devices and motor drives cover a wide research area. The related topics are arranged from plug to propulsion system for a PEV as shown in Figure1. The rectifier circuits are used to convert AC grid voltage to DC voltage, and supplies the converted voltage to DC/DC converters as introduced in fourth subsection. There are several novel converter topologies are also announced in the literature to decrease costs and increase the efficiency besides the introduced main topologies. A converter ensures the charge control of battery system and allows getting the batteries charged in several modes by controlling rectifier output and regenerative braking feedbacks. Furthermore, DC-DC converters also manage the DC motor controls. Inverters cover a wide research area in medium and high power application due to proposed topologies such as full bridge, multilevel and resonant structures. Multilevel inverters generate output voltage in staircase levels that increases the number of voltage steps and decreases the electromagnetic interference (EMI) rates. Low cost topologies such as full-bridge or resonant types are widely used in EVs since the inverters are used in motor drive operations. Three-phase full-bridge inverter that is analyzed in 5th subsection is the main energy conversion part of motor control applications shown in Figures 14, 15, 17, and 18. Resonant inverters, namely soft switching inverters, tackle switching power losses by managing commutating intervals during zero voltage or zero current transitions.

All the power electronics devices introduced up to 6th section require a microcontroller or microprocessor to generate switching signals considering the reference values obtained from voltage or current sensors. The most widely used electric motors in EVs are surveyed in 6th section with drive methods and systems. Induction machines (IM), permanent magnet synchronous machines (PMSM), and brushless DC machines (BLDC) are considered as several electric machine types used in EV powertrain applications with specific controllers. The machine control techniques and controllers are implemented to increase the efficiency of the machine due to developments met in power electronics. Speed or torque control systems can be summarized as a synthesis of electric motor, observers (position and/or speed sensors), and inverter. The microcontroller unit (MCU) manages the closed loop control by acquiring the reference values form sensors, determining the required arrangements, and generating the required switching signals for inverter. The efficiency of closed loop is mostly based on the algorithm that runs on MCU besides self-performance of each device. The estimating algorithms, which allows to sensorless control are developed to remove the sensor requirement in speed and torque control. Other widely known algorithms such as Park, Clarke and Inverse Park, Inverse Clark provide 3-phase/2-phase conversion and generate d-q reference values in current control. BLDC and PMSM motors are frequently used in propulsion systems due to their higher rpm rates comparing to IMs. The BLDCs with 230V supply can generate up to 20000 rpm and 250 HP power. The internal sensor and tachometer options are evaluated as an important advantage. Furthermore, hub or in-wheel BLDCs provide independent control for each wheels as a 4x4like application.

REFERENCES

• [1] Sandalow, D.B.; Plug-In Electric Vehicles: What Role for Washington?. 1st Ed., The Brookings Institution, ISBN 978-0-8157-0305-1, 2009.
• [2] Pistoia,G.; Electric and Hybrid Vehicles Power Sources, Models, Sustainability, Infrastructure and The Market. Elsevier B.V., ISBN 978-0-444-53565-8, 2010.
• [3] Mi, C.; Masrur, M.A.; Gao, D.W.; Hybrid Electric Vehicles: Principles and Applications with Practical Perspectives. John Wiley & Sons Ltd, ISBN 978-0-470- 74773-5, 2011.
• [4] Mapelli, F.L.; Tarsitano, D.; Mauri, M.; Plug-In Hybrid Electric Vehicle: Modeling, Prototype Realization, and Inverter Losses Reduction Analysis. IEEE Transactions on Industrial Electronics, vol. 57, no. 2, pp. 598-607, 2010.
• [5] Emadi, A.; Lee, Y.J.; Rajashekara, K.; Power Electronics and Motor Drives in Electric, Hybrid Electric, and Plug-In Hybrid Electric Vehicles. IEEE Transactions on Industrial Electronics, vol. 55, no. 6, pp. 2237-2245, 2008.
• [6] Lee, Y.J.; Khaligh, A.; Emadi, A.; Advanced Integrated Bidirectional AC/DC and DC/DC Converter for Plug-In Hybrid Electric Vehicles. IEEE Transactions on Vehicular Technology, vol. 58, no. 8, pp. 3970-3980, 2009.
• [7] Gao, Y.; Ehsani, M.; Design and Control Methodology of Plug-in Hybrid Electric Vehicles. IEEE Transactions on Industrial Electronics, vol. 57, no. 2, pp. 633-640, 2010.
• [8] Chau, K.T.; Chan, C.C.; Liu, C.; Overview of Permanent-Magnet Brushless Drives for Electric and Hybrid Electric Vehicles. IEEE Transactions on Industrial Electronics, vol. 55, no. 6, pp. 2246-2257, 2008.
• [9] Bose, B.K.; Modern Power Electronics and AC Drives. Prentice Hall Inc., ISBN 978- 0130167439, 2002.
• [10] Luo, F.L.; Yee, H.; Rashid, M.; Digital Power Electronics and Applications. Elsevier Publishing, ISBN 978-0120887576, 2005.
• [11] Rashid, M.; Power Electronics Handbook. Academic Press, ISBN978-0125816502, 2001.
• [12] Emadi, A.; Handbook of Automotive Power Electronics and Motor Drives. Taylor&Francis CRC Press, ISBN 978-0824723613, 2005.
• [13] Mohan, N.; Undeland, T.M.; Robbins, W.P.; Power Electronics-Converters. Application and Design, John Wiley & Sons Inc., ISBN 978-0471142089, 1995.
• [14] Soylu, S.; Electric Vehicles - Modelling and Simulations. Intech, ISBN 978- 9533074771, 2011.
• [15] Larminie, J.; Lowry, J.; Electric Vehicle Technology Explained. John Wiley & Sons Inc., ISBN 0-470-85163-5, 2003.
• [16] Kabalci, E.; Irmak, E.;Colak, I.; Design of an AC-DC-AC Converter for Wind Turbines. International Journal of Energy Research, Wiley Interscience, vol. 36, no. 2, pp. 169-175, 2011.

• [17] Kabalci, E.; Kabalci, Y.; Develi, I.; Modelling and Analysis of a Power Line Communication System with QPSK Modem for Renewable Smart Grids. International Journal of Electrical Power and Energy Systems, vol. 34, no. 1, pp. 19-28, 2012.

• [18] Colak, I.; Irmak, E.; Yesilbudak, M.; Kabalci, E.; Design of a Parameter Determination System for Non-Isolated Converter Topologies. IEEE International Aegean Conference on Electric Machines and Power Electronics & Electromotion Joint Conference (ACEMP 2010),Istanbul, Turkey,pp.241-244, Sept. 8-10, 2011.

• [19] Babaei, E.; Hosseini, S.H.; Gharehpetian, G.; Reduction of THD and Low Order Harmonics with Symmetrical Output Current for Single-Phase AC/AC Matrix Converters. International Journal of Electrical Power and Energy Systems, vol. 32, pp. 225-235, 2010.

• [20] Ghaemi, A.H.; Abyaneh, H.A.; Mazlumi, K.; Harmonic Indices Assessment by Wavelet Transform. International Journal of Electrical Power and Energy Systems, vol. 33, pp. 1399-1409, 2011.

• [21] Ramasamy, M.; Thangavel, S.; Photovoltaic Based Dynamic Voltage Restorer with Power Saver Capability Using PI Controller. International Journal of Electrical Power and Energy Systems, vol. 36, pp. 51-59, 2012.

• [22] Colak, I.; Kabalci, E.; Increasing the Efficiency of an Inverter Used in Renewable Energy Sources. *2nd International Conference on Nuclear and Renewable Energy Resources (*NURER 2010), Ankara, Turkey,pp. 853-859, July 4-7, 2010.

• [23] Colak, I.; Kabalci, E.; Bayindir, R.; Sagiroglu, S.; The Design Analysis of a 5-Level Cascaded Voltage Source Inverter with Low THD. IEEE Int. Conf. (POWERENG), Lisbon, Portugal,pp. 575-580, 18-20 March 2009.

• [24] Colak, I.; Kabalci, E.; Bayindir, R.; Bal, G.; Modelling of a Three Phase SPWM Multilevel VSI with Low THD Using Matlab/Simulink. EPE 2009, Barcelona, Spain, pp. 1-10, 8-10 September 2009.

• [25] Colak, I.; Kabalci, E.; Bayindir, R.; Review on Multilevel Voltage Source Inverter Topologies and Control Schemes, Energy Conversion and Management. vol. 52, pp. 1114-1128, 2011.

• [26] Colak, I.; Kabalci, E.; Developing a Novel Sinusoidal Pulse Width Modulation (SPWM) Technique to Eliminate Side Band Harmonics. International Journal of Electrical Power and Energy Systems, doi:10.1016/j.ijepes.2012.08.024.

• [27] Colak, I.; Kabalci, E.; Practical Implementation of a DSP Controlled Multilevel Inverter with Low THD for Motor Drive Applications. Journal of Power Sources, vol. 196, pp. 7585-7593, 2011.

• [28] Veltman, A.; Pulle, D.W.J.; Doncker, R.W.; Fundamentals of Electrical Drives. Springer, ISBN 978-1-4020-5503-4, 2007.

• [29] Vukosavic, S.N.; Digital Control of Electrical Drives. Springer, ISBN 978-0-387- 48598-0, 2007.

• [30] Leitman, S.; Brant, B.; Build Your Own Electric Vehicle. McGraw-Hill, ISBN 0-07- 154373-2, 2009.

• [31] Fitzgerald, A.E.; Kingsley, C.; Umans, S.; Electric Machinery. 6th Ed., McGraw-Hill Higher Education, ISBN 978-0073660097, 2002.

• [32] Zheng, C.; Zhang, B.; Qui, D.; Digital Natural Sampling SPWM Based on Inverse Operator Method. IEEE Power Electronics Spec. Conf., Orlando, USA, 2007.

• [33] Kanchan, R.S.; Gopakumar, K.; Kennel, R.; Synchronised Carrier-Based SVPWM Signal Generation Scheme for the Entire Modulation Range Extending up to Six-Step Mode Using the Sampled Amplitudes of Reference Phase Voltages. IEEE Elec. Power App., vol. 1, pp. 407-415, 2007.

• [34] Beig, A.R.; Narayanan, G.; Ranganathan, V.T.; Modified SVPWM Algorithm for Three Level VSI with Synchronized and Symmetrical Waveforms. IEEE Transactions on Industrial Electronics, vol. 54, pp.486-493, 2007.

• [35] Jacobina, C.B.; Lima, A.M.N.; Silva, E.R.C.; Alves, R.N.C.; Seixas, P.F.; Digital Scalar PWM: A Simple Approach to Introduce Non-Sinusoidal Modulating Waveforms. IEEE Transactions on Power Electronics, vol. 16, pp. 351-359, 2001.

• [36] Cirrincione, M.; Accetta, A.; Pucci, M.; Vitale, G.; MRAS Speed Observer for High-Performance Linear Induction Motor Drives Based on Linear Neural Networks. IEEE Transactions on Power Electronics, vol.28, no.1, pp.123-134, Jan. 2013.

• [37] Rodriguez, J.; Kennel, R.M.; Espinoza, J.R.; Trincado, M.; Silva, C.A.; Rojas, C.A.; High-Performance Control Strategies for Electrical Drives: An Experimental Assessment. IEEE Transactions on Industrial Electronics, vol.59, no.2, pp.812-820, Feb. 2012.

• [38] Patel, H.M.; Shah, P.T.; Patel, H.V.; Field Oriented Control of Induction Motor Using Matlab Simulink. *Journal of Information, Knowledge and Research in Electrical Engineering, v*ol. 1, no.2, pp. 56-60, 2011.

• [39] Biswas, D.; Mukherjee, K.; Kar, N.C.; A Novel Approach towards Electrical Loss Minimization in Vector Controlled Induction Machine Drive for EV/HEV. IEEE Transportation on Electrification Conference and Expo (ITEC), pp.1-5, 18-20 June 2012.

• [40] Ehsani, M.; Gao, Y.; Emadi, A.; Modern Electric, Hybrid Electric, and Fuel Cell Vehicles: Fundamentals, Theory, and Design. 2nd Edition, CRC Press, ISBN 978- 1420053982, 2009.

• [41] Yi, W.; Kaiqi, Z.; Field-Oriented Vector Control of Induction Motor for Electric Vehicles. 31st Annual Conference of IEEE Industrial Electronics Society, pp. 1610- 1614, 2005.

• [42] Rehman, H.U.; Xu, L.; Alternative Energy Vehicles Drive System: Control, Flux and Torque Estimation, and Efficiency Optimization. IEEE Transactions on Vehicular Technology, vol.60, no.8, pp.3625-3634, Oct. 2011.

• [43] Mauri, M.; Mapelli, F.L.; Tarsitano, D.; A Reduced Losses Field Oriented Control for Plug-in Hybrid Electrical Vehicle. XIX International Conference on Electrical Machines (ICEM), pp.1-6, 6-8 Sept. 2010.

• [44] Texas Instruments Europe; DSP Solutions for BLDC Motors. 1997.

• [45] Balazovic, P.; Designer Reference Manual: Sensorless PMSM Control for an H-axis Washing Machine Drive. Freescale Czech Systems Laboratories, Roznov pod Radhostem, Czech.

• [46] Tie, S.F.; Tan, C.W.; A Review of Energy Sources and Energy Management System in Electric Vehicles. Renewable and Sustainable Energy Reviews, vol. 20, pp. 82-102, 2013.

• [47] Zhang, X.; Wang, Z.; Cheng, M.; Shichuan, D.; Wang, W.; The Random PWM Based Bi-Directional Buck-Boost Cascade Converter for Electric Vehicles. International Conference on Electrical Machines and Systems (ICEMS), pp.1-5, 20-23 Aug. 2011.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 10

POWER CONTROL OF PLUG-IN ELECTRIC VEHICLES IN SMART GRIDS

Amin Hajizadeh1, , Amir Hossein Shahirinia2,† and David C. Yu2,‡

¹Department of Electrical Engineering, Shahrood University of Technology, Shahrood, Iran ²Department of Electrical Engineering and Computer Applied Science, University of Wisconsin-Milwaukee, Milwaukee, US

ABSTRACT

In this chapter, Plug-in Electric Vehicle (PEV) is considered with dual energy storage sources. In order to respond to a transient power demand, usually battery energy storage device is combined with super capacitor. Battery-super capacitor system is used to create an energy storage system with the high energy density attributes of batteries and the high power density of super capacitor. In order to simulate the PEV model, dynamic state space models of bidirectional DC-DC converter and grid connected voltage source converter with energy storage models are considered to connect the PEV to the main grid. In order to stabilize the DC-link power and distribute the power between dual energy storage sources in PEV during normal and disturbance conditions on the grid voltage, a Lyapunov based Neuro-Fuzzy control strategy has been developed. This controller determines the super capacitor power that should be generated according to the amount of available energy in DC-link. Moreover, a robust sliding mode control strategy for voltage source converter based on positive and negative symmetrical components is presented to investigate the voltage disturbance ride-through and voltage control capability.

Keywords: Plug-in Electric Vehicle, Intelligent Control, Smart Grids, Neuro-Fuzzy, Voltage Disturbances

E-mail: aminhajizadeh@shahroodut.ac.ir.

E-mail: shahiri2@uwm.edu.

E-mail: yu@uwm.edu.

1.INTRODUCTION

Gridable Electric Vehicles (GEV), such as plug-in hybrid electric vehicles and batteryelectric vehicles, have recently become more attractive options over combustion engine-based vehicles, due to the high fuel price and environmental concerns. They have the ability to discharge their batteries and send electricity back to the power grid when they are parked and connected to an electrical outlet. The electricity flowing from electric vehicles to the power grid is then called "vehicle-to-grid" power or V2G [1-3].

The success of widespread electric vehicle implantation depends on the reliability of both the electric grid and the charging equipment. The cost of the batteries make the energy price of battery supplied power expensive, which implies that not only a robust and reliable charger is needed, but also with high power efficiency and minimum impact on the power quality of smart grid distributions systems. The electric vehicle storages can act as controllable loads or generation units to buffer the variable power produced by the wind turbines. V2G systems can act as a provider of the important power system ancillary services like regulation (up and down) and spinning reserves [3]. The EV owners can benefit by the revenues earned from the electricity market by participating in the power balancing act.

The EV storages act as distributed sources with smart metering, intelligent control and communication interfaces with which it is connected to an aggregator [2].

The power flow in V2G can be both unidirectional and bidirectional, though in unidirectional V2G only services such as frequency regulation and delivering power, are provided to the grid. Through V2G, EV owners can potentially generate revenue while charging their cars and at the same time mitigate the negative impacts on the grid from charging. With V2G, an EV can participate in most energy markets, from bulk energy, to spinning reserves and frequency regulation [4-6].

Plug-in Electric Vehicles (PEV) are connected to smart grid via power electronic converters to support the grid.

So it is important that the control strategies are designed to keep the system stable under any disturbance and parameter variations in the electrical distribution system.

The grid-connected power electronic converters are highly sensitive to grid disturbances and it is important to emphasize the necessity to reduce the effects of voltage disturbances on their operation. In this chapter, PEV is considered with dual energy storage sources.

In order to respond to a transient power demand, usually battery energy storage device is combined with supercapacitor. Battery-super capacitor system is used to create an energy storage system with the high energy density attributes of batteries and the high power density of super capacitor.

For simulation studies, dynamic models of bidirectional DC-DC converter and grid connected voltage source converter are considered to connect the hybrid energy storage sources to the main grid. In order to stabilize the DC-link power and distribute the power between dual energy storage sources in PEV during normal and disturbance conditions on the grid voltage, a Lyapunov based neuro-fuzzy control strategy has been developed. This controller determines the super capacitor power that should be generated according to the amount of available energy in DC-link. Moreover, a robust sliding mode control strategy for voltage source converter based on positive and negative symmetrical components is presented to investigate the voltage disturbance ride-through and voltage control capability.

2. ARCHITECTURE OF THE PLUG-IN ELECTRIC VEHICLE

The modeling of Plug-in Electric Vehicle (PEV) is an important issue that needs to be carefully addressed. In Figure 1, overall structure of PEV is shown. Many articles deal with static models that are built up from maps and static relationships between parameters in the model.

These models allow for fast simulation, but they cannot show the oscillations and other dynamic phenomena when switching occurs between different modes of operation. However, a good model should consider both accuracy and simulation time. The electric components of electric vehicle used in this chapter comprise a battery bank, DC/DC and DC/AC converters. The details about modeling of battery energy storage and power electronic converters have been described in [7, 8].

3. POWER MANAGEMENT ON V2G - STATE OF THE ART

EV battery"s energy can be utilized as distributed energy storage and used for reactive power support as well as voltage sag reduction at a particular node. A schematic diagram of a hybrid energy storage including battery and supercapacitor and a bidirectional converter coupled with the distribution node via line reactance X is shown in Figure 2. In this figure, the hybrid energy storage system represents a lumped parameter for the energy available from the charging station in that particular area. With the bidirectional converter, this dual energy storage can behave both as a source or a sink, i.e., the vehicles at the charging station can collectively discharge or charge respectively. The bidirectional converter, interfaced with the EV energy storage, is synchronized with the grid system at point a. Power injection by the battery is the power at connection point b, which is at the utility side.

The grid-connected power electronic converters are highly sensitive to grid disturbances and it makes to emphasize the necessity to reduce the effects of voltage disturbances on their operations [9]. In spite of the growing number of V2G units, their contribution of power delivered to the utility grid remains small, as compared to the power injected by the large centralized power plants.

Figure 1. Overall structure of PEV.

Figure 2. Diagram of charging and discharging modes of V2G.

However, to support the grid in case of disturbances, it will become necessary to keep the V2G units connected to the grid.

In the wide range of power quality disturbances, the interest focuses on voltage sags, which can severely affect the performance of the voltage source converter (VSC). A voltage sag is a drop in voltage with duration between one half-cycle and one minute [9], which is in most cases caused by a short-circuit fault.

Therefore, the operation of V2G under voltage sag has not received much attention in the past. However, as the power generated by V2G increases, this behavior stresses the utility grid and could cause power unbalance, which may turn into instability.

So, the interaction between V2G and the grid during the voltage sag is very important and it must be considered for designing the proper control strategy.

Hence in this chapter a control strategy is proposed for V2G including energy storages under voltage sag conditions. During the voltage sag in distribution systems, a decrease in voltage magnitude ate the grid-connected converter is occurs.

In this case, the current controllers to avoid overloading the converter limit the power that V2G can supply to the grid during voltage sag. For V2G, the power limitation can be a problem resulting the slow dynamics of battery energy storages. According to this limitation, in order to respond the transient of power demand, usually a supercapacitor is combined with battery energy storage. The comparison between battery and supercapacitor energy storages shows that the use of supercapacitor is better than battery for power quality problem studies and there are some limitations to use the battery energy storage.

Due to the low power density of the battery than supercapacitor, it cannot release its charge or discharge fast enough during the voltage sag [7].

Additionally, the main drawback of the batteries is a slow charging time, limited by a charging current, in contrast, the supercapacitor can be charged in a short time, depending on the availability of a high charging power from the main source.

Although the batteries are considered to be the main energy storage device for EV application, their cycle and calendar life should still be improved.

For these reasons, the use of the supercapacitor as an auxiliary source is expected that the very fast power response and high specific power can complement the slower power output of the main source.

So it is important to study the operation and the behavior of whole components of V2G system under voltage sag not only the response of power electronic converters.

Hence, in this chapter a robust power control strategy has been presented for V2Gsystem during voltage sag in smart grid.

4. POWER MANAGEMENT OF GEV BASED ON DC-LINK POWER AND VOLTAGE CONTROL

In this section, the control strategy of V2G system has been presented. The overall control structure is illustrated in Figure 3, which includes the power flow controller and local controllers for power conditioning units and the hybrid energy storages system.

The term, "power flow control", refers to the design of the higher-level control algorithm that determines the proper power level to be generated, and its split between the two power sources [10, 11].

In fact, during voltage sag conditions, the power flow control strategy must be design to stabilize the DC-link power and regulate the DC-link voltage consequently. During the voltage sag, a decrease in voltage amplitude occurs at the converter terminal. To keep the power supplied to the grid constant, the current should increase. It will be limited by the current controller however, to avoid overloading of the converter. This will thus limit the power that the V2G can supply to the grid during voltage sag, resulting the DC-link voltage will increase.

Figure 3. Power Control structure of V2G with hybrid energy storages and power electronic converters.

To avoid a too high DC-link voltage, the power balance between inverter power and V2G power must be satisfied. According to three operation modes for V2G, the DC-link voltage is given by three different state equations:

Discharging mode:

$$C{dc}v{dc}\frac{dv{dc}}{dt} = P{V2G} - P_{inv}$$

$$P{V2G}(k) = P{batt}(k) + P_{sup}(k)$$ (1)

Charging mode:

$$C{dc}v{dc}\frac{dv{dc}}{dt} = P{inv} - P_{V2G}$$

$$P{V2G}(k) = P{batt}(k) + P_{sup}(k)$$ (2)

Disconnected Mode:

$$C{dc}v{dc}\frac{dv{dc}}{dt} = P{V2G} - P_{load}$$

$$P{V2G}(k) = P{batt}(k) + P_{sup}(k)$$ (3)

In this mode, the V2G is not connected to grid and it could be used for supplying electrical motor to propel the vehicle. According to the above mentioned equations, *PV*2G, Pinv, Pload, Pbatt, Psup, Cdc and Vdc are V2G power supplied to the DC-link, inverter power (grid power), electrical load power, battery power, supercapacitor power, DC-link capacitor and voltage respectively. During the voltage sag, the reference power of inverter is determined by the maximum power that can be supplied to the grid. So, the V2G power must be decreased to avoid a too high DC-link voltage. Hence, the power management in hybrid battery/ supercapacitor energy storage system is very important task and its effect on the operation of the system during voltage sag. So it should be robust under any variations in voltage and power. Moreover, the distribution of power between power sources must be done optimally.

4.1. Lyapunov Based Neuro-Fuzzy Control Strategy

In this section, a power flow control structure has been developed for hybrid energy sources during voltage sag. It is based on the Neuro-fuzzy algorithm and distributes the power between the supercapacitor and battery energy storage according to the input parameters [10]. The configuration of the proposed control strategy is shown in Figure 4.

In this control structure, the battery power is determined according to the error of DClink power (power difference between inverter power and V2G power) P(k), the error of DC-link voltage (voltage difference between desired voltage and bus voltage) Vdc(k) and the error of battery power Pbatt(k) (battery power difference between Pbatt(k-1) and Pbatt(k)).

Figure 4. Proposed neuro-fuzzy power flow control strategy.

These inputs are calculated as follow:

$$\Delta P(k) = P{V2G}(k) - P{inv}(k)$$

$$\Delta V{dc}(k) = V{dc-ref} - V_{DC}(k)$$

$$\Delta P{batt}(k) = P{batt}(k) - P_{batt}(k-1)$$ (4)

In fact, the P(k) and Vdc(k) and its derivative are very important during the decision process to stabilize the dc bus voltage during the occurring of voltage sag. Also, Pbatt(k) is very important to battery operation and it makes to minimize the changes of the battery power. The output of neuro-fuzzy controller is the battery system duty ratio (Kbatt). It is a representation of mount power of battery that must be delivered and varies between 0 and 1.

Then according to this parameter, the amount of battery power must be delivered is determined. The following equations show the battery.

$$P{batt}(k) = K{batt}(k) \times P_{batt-nom}$$

$$P{sup}(k) = P{V2G}(k) - P_{batt}(k)$$ (5)

where, Pbatt-nom is the nominal power of battery.

During the voltage sags, the power flow control strategy must be designed to manage the power between the battery and supercapacitor energy sources and the utility grid. Moreover, it is important to consider the dynamic limitation of the battery, its power must change smoothly so that the battery operating point deviation is minimized and its life time is increased. In proposed neuro-fuzzy control strategy, for each input, four fuzzy subsets have been used. These are ZE (zero), L (low), M (medium) and H (high). For each of these fuzzy sets, Gaussian membership function has been used.

As each of the three inputs has seven four sets, there are altogether 64 control rules in the neuro-fuzzy logic controller.

The neuro-fuzzy algorithm developed by Wang [12] uses the hybrid model developed by Takagi-Sugeno in [13]. In this type of model, the condition part uses linguistic variables and the conclusion part is represented by a numerical value which is considered as a function of system condition expressed with the variables*x*1, *x*2, …., xm. These models are suitable for neural-based-learning techniques as gradient methods to extract the rules and generate models with a reduced number of rules.

$$\omega^{(l)} = g(x_1, x_2, ...., x_n) \tag{6}$$

The neuro-fuzzy algorithm uses membership functions of Gaussian type. With Gaussian fuzzy sets, the algorithm is capable of utilizing all information contained in the training set to calculate each rule conclusion, which is different when using triangular partitions. Figure 5 illustrates the neuro-fuzzy scheme for an example with two input variables (*x*1, *x*2) and one output variable (y). In the first stage of the neuro-fuzzy scheme, the two inputs are codified into linguistic values by the set of Gaussian membership functions attributed to each variable. The second stage calculates each rule R its respective activation degree. Last, the inference mechanism weights each rule conclusion ω , initialized by the cluster-based algorithm, using the activation degree computed in the second stage.

The error signal between the model inferred value Y and the respective measured value (or teaching value) y', is used by the gradient descent method to adjust each rule conclusion. The algorithm modifies the values of ω to minimize an objective function E usually expressed by the mean quadratic error (16). In this equation, the value y'(k) is the desired output value related with the condition vector x'(k) = (*x'*1, *x'*2,…,x'm). The element Y(x'(k)) is the inferred response to the same condition vector x'(k) and computed by Equation (7).

$$E = \frac{1}{2} [Y(x'(k)) - y'(k)]^{2}$$

$$Y(x'(k)) = \frac{\sum{l=1}^{c} \left( \prod{j=1}^{m} \mu{A{j-1}^{(l)}}(x'{j}(k)) \right) \omega^{(l)}(k)}{\sum{l=1}^{c} \left( \prod{j=1}^{m} \mu{A{j-1}^{(l)}}(x'{j}(k)) \right)}$$

$$\mu{A}(x'{j}(k)) = \exp\left{ -\frac{(x'{j}(k) - m{j})^{2}}{(\delta_{j})^{2}} \right}$$ (8)

where, the subscript j indicates the *j*th input x^j .

Figure 5. The neuro-fuzzy structure.

By using the back propagation (BP) learning algorithm, the weighting vector of the FNN is adjusted such that the error defined (7) is less than a desired threshold value after a given number of training cycles. The well-known BP algorithm may be written briefly as:

$$\omega^{(l)}(t+1) = \omega^{(l)}(t) - \alpha \frac{\partial E}{\partial \omega^{(l)}}$$ (9)

The symbol α is the learning rate parameter, and t indicates the number of learning iterations executed by the algorithm.

The inference function (8) depends on ω only through its numerator. The expression composing the numerator is now denoted by a and is shown in (10).

$$a = \sum{l=1}^{c} \left( \prod{j=1}^{m} \mu{A{j=1}^{(l)}}(x_j'(k)) \right) \omega^{(l)}(k)$$ (10)

The denominator of function (8) is dependent on a term d (l) , defined in (11), and denoted by b in (12).

$$d^{(l)} = \prod{j=1}^{m} \mu{A_{j=1}^{(l)}}(x_j'(k))$$ (11)

$$b = \sum_{l=1}^{c} (d^{(l)})$$ (12)

To calculate the adjustment of each conclusion value ω , it is necessary to compute the variation of the objective function E, ∂E, in relation to the variation that occurred in ω in the anterior instant, ∂ω(l) . Therefore, using the chain rule to calculate ∂E/∂ω(l) results in Equation (13).

$$\frac{\partial E}{\partial \omega^{(l)}} = \frac{\partial E}{\partial Y} \frac{\partial Y}{\partial a} \frac{\partial a}{\partial \omega^{(l)}}$$ (13)

The use of chain rule looks for the term contained in E that is directly dependent on the value to be adjusted, i.e., the conclusion value ω . Therefore, we can verify by chain Equation (13) that it starts with E dependent of Y value. The Y value depends on term and, at last, the expression a is a function of ω . After some computation, the adjustment to be made in ω (l) can be interpreted as being proportional to the error between the neuro-fuzzy model response and the supervising value, but weighted by the contribution of rule (l), denoted by d , to the final neuro-fuzzy inference.

$$\omega^{(l)}(t+1) = \omega^{(l)}(t) - \alpha \frac{(Y(x'(k)) - y'(k))d^{(l)}}{\sum_{l=1}^{c} (d^{(l)})}$$ (14)

Next, a convergence theorem has been developed to guarantee the stability of learning algorithm used for the above-mentioned FNN. A Lyapunov energy function is defined as follows:

$$V_k = J_k = \frac{1}{2}E_k^2 \tag{15}$$

From Equation (15), we can get

$$\Delta V = V_{k+1} - Vk = \frac{1}{2} (E{k+1}^2 - E_k^2) \tag{16}$$

The error difference, Ek, can be defined as

$$\Delta Ek = E{k+1} - E_k = \frac{\partial E_k}{\partial \omega} \Delta \omega \tag{17}$$

where

$$\Delta \omega = \omega_{k+1} - \omega_k = -\alpha E_k \frac{\partial E_k}{\partial \omega} \tag{18}$$

Using Equation (16), we can get

$$\Delta V = \frac{1}{2} (E_{k+1} - Ek)(E{k+1} + E_k) = \frac{1}{2} (\Delta E_k)(2E_k + \Delta E_k)$$ (19)

Substituting Equation (18) into Equation (19), we have

Substituting Equation (18) into Equation (19), we have $$\Delta V = \frac{1}{2} \frac{\partial E_k}{\partial \omega} \alpha E_k \frac{\partial E_k}{\partial \omega} (-2E_k + \frac{\partial E_k}{\partial \omega} \alpha E_k \frac{\partial E_k}{\partial \omega}) = \frac{1}{2} (E_k \frac{\partial E_k}{\partial \omega})^2 \left[ (\frac{\partial E_k}{\partial \omega})^2 \alpha^2 - 2\alpha \right]$$ (20)

If V<0, the convergence of the algorithm described in Equation (20) can be guaranteed. Therefore, we have

$$\left(\frac{\partial E_k}{\partial \omega}\right)^2 \alpha^2 - 2\alpha < 0 \tag{21}$$

From Equation (21), we can obtain

$$0 < \alpha < \frac{2}{\left(\frac{\partial E_k}{\partial \omega}\right)^2} \tag{22}$$

4.2. Robust Sliding Mode Control Strategy for Voltage Source Converter Based on Positive and Negative Symmetrical Components

Control of grid connected voltage source converter (VSC) is an important problem during voltage disturbances like voltage sag. A voltage sag is a reduction in the RMS voltage in the range from 0.1 to 0.9 pu of the nominal voltage for duration greater than half cycle and less than one minute. The drawback of using VSC is its sensitivity to voltage disturbances. For a VSC, a sudden decrease in grid voltage normally causes an increase in current, as the control attempts at maintaining the power to the DC-link constant.

Moreover, most faults are unbalanced and result in unbalanced voltage sag, which produce undesirable power oscillations of low order frequencies in current harmonics and poor DC-link voltage regulation. Ultimately, this can also lead to tripping of the converter due to DC overvoltage.

The VSC is connected is connected to the grid via a filter inductor. The dq-components of currents and voltages are then used along with the reference current signals. The VSC controller is required to have main functions: Current control and DC-link voltage regulation.

A comparison between different types of current controllers for shunt connected VSC, has proved that Dual Vector Current Controller (DVCC), is capable of providing sinusoidal grid currents and regulated DC-link voltage during unbalanced faults. The current controller used in this chapter consists of two sliding model controllers that control the positive and negative sequence current separately and are implemented in two different rotating coordinate systems [8].

A simplified scheme for the DVCC is shown in Figure 6. The VSC is connected to the grid via filter inductors. The three phase grid currents and voltages are sampled and transformed into its positive and negative sequence components. The positive and negative sequence of dq-components are then used along the with the reference current signals in the DVCC to produce the reference voltage signals for the PWM regulator.

According to the proposed control strategy in [8], the purpose of the current controller is to synthesize a voltage correction vector so that the current error vector can be kept to a minimum value. The error current-vector is defined as:

$$\Delta i(k) = i_{ref}(k) - i(k) \tag{23}$$

where, iref(k) is the specified current vector command for the positive and negative sequence of dq-components and i(k) is the measured current vector feedback. In order to generate proper current references, consider the complex apparent power from the grid:

Amin Hajizadeh, Amir Hossein Shahirinia and David C. Yu $$Sg = (e{dqp} e^{j\omega t} + e{dqn} e^{-j\omega t}).(i{dqp} e^{j\omega t} + i_{dqn} e^{-j\omega t})^* =$$

$$= (P + P{2c} \cos(2\omega t) + P{2s} \sin(2\omega t)) + j(Q + Q{2c} \cos(2\omega t) + Q{2s} \sin(2\omega t))$$ (24)

In order to work with an invertible matrix (4×4), oscillating reactive power (*Q*2c, *Q*2s) can not be included in the current reference calculation. Therefore, oscillating reactive power is not controlled and will flow through the system.

By expanding Equation (24), the following expression in matrix form can be written:

$$\begin{bmatrix} P{g} \ Q{g} \ P{s2g} \ P{c2g} \end{bmatrix} = \begin{bmatrix} e{dp} & e{qp} & e{dn} & e{qn} \ e{qp} & -e{dp} & e{qn} - e{dn} \ e{qn} & -e{dn} & -e{qp} & e{dn} \ e{dn} & e{qn} & e{dp} & e{qp} \end{bmatrix} \begin{bmatrix} i{dp} \ i{qp} \ i{dn} \ i{qp} \end{bmatrix}$$

$$i_{dn}$$

$$i_{qp}$$

$$i_{dn}$$

$$i_{qp}$$

$$i_{dn}$$

$$i_{qp}$$

$$i_{dn}$$

$$i_{qp}$$

$$i_{dn}$$

$$i_{qp}$$

where, P^g and Q^g are the constant active and reactive power, respectively, while the subscripts *Ps2^g* and *Pc2^g* represent the second harmonic sine and cosine component of the active power. These are the oscillating powers due to the unbalance in the grid voltages.

The reference currents can be calculated as follow:

Figure 6. Block diagram of dual vector current controller.

$$\Delta P = R(i{dp}^2 + i{qp}^2 + i{dn}^2 + i{qn}^2) \tag{27}$$

$$\Delta P{c2} = 2R(i{dp}i{dn} + i{qp}i{qn}) + 2\omega L(i{dp}i{qn} - i{qp}i_{dn})$$ (28)

$$\Delta P{s2} = 2R(i{dp}i{qn} - i{qp}i{dn}) + 2\omega L(-i{dp}i{dn} - i{qp}i_{qn})$$ (29)

This algorithm calculates current references (26) by setting active and reactive power references (P * , Q * ), and by forcing the oscillating active power demanded by the filter to be delivered from the grid * 2 2 ( ; P P c c * 2 2 ) P P s s . Then, no oscillating active power flows between the DC-link and the filter.

5. SIMULATION RESULTS

In order to show the effectiveness of proposed control strategy, the simulation model of the proposed hybrid V2G system has been built in Matlab/Simulink environment. The parameters of the hybrid battery/ supercapacitor energy storage system in this study are given in [8, 10]. The supercapacitor operates as a buffer of energy to meet load demand that cannot be met by the battery energy source, particularly during transient or disturbances periods. In this case study, the output power of the battery is limited to 50 kW and the supercapacitor bank has is capable of sustaining the extra load of 20 kW during the voltage disturbances.

Also it is supposed that the initial state of charge for supercapacitor is 0.7. For the investigated system, simulations have been done with unbalanced voltage sag (type C) with the magnitude of between 0.3 and 0.9 pu Maximum grid current and peak- to-peak DC voltage ripple during the sag are shown for sag type C in Figures 7 and 8. In order to ride through sag with minimum retained voltage of 30%, the converter switches should be able to carry a maximum current of 3.65 pu.

The maximum DC voltage ripple is ±0.25% around the nominal value for sag magnitudes below 30%. Moreover, for the evaluation of proposed control strategy, simulations have been run with unbalanced voltage sag (type C) with magnitude 40%. The sag starts at 0.3sec and ends at 0.4sec. The DC-link voltage is shown in Figure 9. The grid currents increase to above 2 pu and DC voltage shows a variation during the transients at the beginning and end of the sag. However, the ripple during the transient is not bigger than 10% peak-to-peak and is quite quickly damped to almost zero. As shown in Figure 10, the generated power of the hybrid energy storage systems decreases during voltage disturbance to increase the voltage sag ridethrough capability. In fact, during the voltage sag, the current controllers limit the output current to avoid the overloading of the converter. As a result the average and instantaneous active power that converter supplies to the grid deceases (Figure 10), resulting in an increase in DC-link voltage (Figure 9). The grid converter now controls the power (with its set point determined by the maximum current). In these conditions, to stabilize the DC-link voltage, the neuro-fuzzy controller manages the power flow between battery and supercapacitor, resulting the reference power of the energy storage bank changes for decreasing the input power to dc bus during voltage disturbances and neuro-fuzzy control strategy switches the energy storage bank to the charging mode and energy storage"s state of charge (Figure 11) goes high and remain constant on a new value. Because after occurring voltage sag, the energy storage dose not discharge and the output power of supercapacitor is equal to zero.

From Figures 12 and 13, it is evident that the battery and supercapacitor together share this load requirement. Because of the slow variations of battery, the output power of battery energy source remains constant during the voltage sag (Figure 13).

The output voltage of battery is shown in Figure 14. The battery voltage has not been affected during the voltage sag. According to the simulation results, it is clear that the parameters of battery system has not changed during the occurring the voltage disturbances in distribution systems.

On the other hand, during occurring voltage sag, the supercapacitor makes to improve the performance of V2G. In this condition, the proposed power flow control strategy finds suitable operating point for supercapacitor and battery energy sources.

Figure 7. Maximum grid current in pu for voltage sag type C with magnitude between 0.3 and 0.9.

Figure 8. DC-link voltage ripples in pu for voltage sag type C with magnitude between 0.3 and 0.9.

Figure 9. DC-link voltage for 40% sag type C (pu).

Figure 10. Generated average and instantaneous active power by hybrid distributed generation system during voltage sag.

CONCLUSION

In this chapter, PEV is considered with dual energy storage sources. In order to respond to a transient power demand, usually battery energy storage device is combined with super capacitor. Battery-super capacitor system is used to create an energy storage system with the high energy density attributes of batteries and the high power density of super capacitor.

For simulation studies, dynamic models of bidirectional DC-DC converter and grid connected voltage source converter are considered to connect the hybrid energy storage sources to the main grid.

Figure 11. Energy storage"s state of charge during voltage sag.

Figure 12. The supercapacitor power during voltage sag (pu).

Figure 13. The battery power during voltage sag (pu).

In order to stabilize the DC-link power and distribute the power between dual energy storage sources in PEV during normal and disturbance conditions on the grid voltage, a Lyapunov based neuro-fuzzy control strategy has been developed. This controller determines the super capacitor power that should be generated according to the amount of available energy in DC-link.

Figure 14. Output voltage of battery during voltage sag.

Moreover, a robust sliding mode control strategy for voltage source converter based on positive and negative symmetrical components is presented to investigate the voltage disturbance ride-through and voltage control capability.

REFERENCES

• [1] Han, S., Sezaki, K., Development of an Optimal Vehicle-to Grid Aggregator for Frequency Regulation. IEEE Transaction on Smart Grid, vol. 1, no. 1, pp. 65-72, 2010.
• [2] Guille, C., Gross, G., A Conceptual Framework for the Vehicle-to-Grid (V2G) Implementation. Energy Policy, vol. 37, no. 11, pp. 4379-4390, 2009.
• [3] Han, S., Sezaki, K., Design of an Optimal Aggregator for Vehicle to Grid Regulation Service. IEEE Innovative Smart Grid Technologies (ISGT 2010), Gaithersburg, Maryland, US, pp. 1-8, 2010.
• [4] Nyns, K. C., Haesen, E., Driesen, J., The Impact of Charging Plug-In Hybrid Electric Vehicles on a Residential Distribution Grid. IEEE Transaction on Power System, vol. 25, no. 1, pp. 371-380, 2010.
• [5] Kisacikoglu, M. C., Ozpineci, B., Tolbert, L. M., Examination of a PHEV Bidirectional Charger System for V2G Reactive Power Compensation. Twenty-Fifth Annual IEEE Applied Power Electronics Conference and Exposition (APEC 2010), Palm Springs, California, US, pp. 458-465, 2010.
• [6] Lee, Y. J., Khaligh, A., Emadi, A., Advanced Integrated Bidirectional AC/DC and DC/ DC Converter for Plug-In Hybrid Electric Vehicles. IEEE Transaction on Vehicular Technology, vol. 58, no. 8, pp. 3970-3980, 2009.
• [7] Barton, J., Infield, D., Energy Storage and its Use with Intermittent Renewable Energy. IEEE Transaction on Energy Conversion, vol. 19, no. 2, pp. 441-448, 2004.
• [8] Hajizadeh, A., Golkar, M. A., Feliachi, A., Voltage Control and Active Power Management of Hybrid Fuel-Cell/Energy-Storage Power Conversion System under Unbalanced Voltage Sag Conditions. IEEE Transactions on Energy Conversion, vol. 25, no. 4, pp. 1195-1208, 2010.
• [9] Bollen, M. H., Understanding Power Quality Problems: Voltage Sags and Interruptions. IEEE Press, 1999.

• [10] Hajizadeh, A., Golkar, M. A., Intelligent Power Management Strategy of hybrid Distributed Generation System. International Journal of Electrical Power and Energy Systems, vol. 29, no. 10, pp. 783-795, 2007.

• [11] Hajizadeh, A., Golkar, M. A., Fuzzy Neural Control of a Hybrid Fuel Cell/Battery Distributed Power Generation System. IET Renewable Power Generation, vol. 3, no. 4, pp. 402-414, 2010.

• [12] Wang, L. X., Back-Propagation of Fuzzy Systems as Nonlinear Dynamic System Identifier. IEEE International Conference on Fuzzy Systems, San Diego, California, US, pp. 1409-1418, 1992.

• [13] ChingHung, L., ChingCheng, T., Identification and Control of Dynamic Systems Using Recurrent Fuzzy Neural Networks. IEEE Transactions on Fuzzy Systems, vol. 8, no. 4, pp. 349-366, 2000.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 11

OPTIMAL SHORT-TERM SCHEDULING OF AGGREGATORS FOR PLUG-IN ELECTRIC VEHICLES

*Mohammad Moradi-Dalvand**¹ *and Behnam Mohammadi-Ivatloo†*²

¹Department of Electrical Engineering, Power and Water University of Technology, Tehran, Iran Faculty of Electrical and Computer Engineering, University of Tabriz, Tabriz, Iran

ABSTRACT

Plug-in hybrid electric vehicles (PHEVs) are fast growing options for the next generation of transportation systems. With the increasing number of PHEVs in power systems, the system load will increase accordingly. If PHEVs are treated as conventional loads (uncontrolled loads), the large amount of charging power will take place at peak hours when drivers head back home in the evening. Since most drivers are expected to charge their vehicles in the morning and have no preferences about the exact time of charging, PHEVs are more flexible loads than the traditional ones. It is predicted that there will be one million PHEVs on U.S. roads within 5 to 10 years. A new organization, which is called a PHEV aggregator, is required for utilizing the flexibilities and charging control of PHEVs. PHEV aggregator is a new participant of electricity markets which is responsible for providing power and controlling the charging pattern of PHEVs under its territory. The aggregator should buy power from the wholesale electricity market. Thus, it needs to forecast power prices for the next day. Moreover, the aggregator should predict PHEVs' consumption, charging duration, and arrival and departure time. This chapter deals with the short-term scheduling of aggregators in day-ahead electricity markets considering price uncertainty. Market prices, as the most important uncertainty, will be modeled using a stochastic programming technique. Proper scenario generation and reduction methods are utilized for generating efficient scenarios. The objective of PHEV aggregator is to maximize its profit by charging PHEVs at low-price hours. Constraints of electricity market and characteristics of PHEVs are incorporated into the proposed method. An illustrative example is provided to describe the proposed method.

^* E-mail: moradi.dalvand@gmail.com.

E-mail: mohammadi@ieee.org.

Keywords: Plug-in electric vehicle (PHEV), Energy market, PHEV aggregator, Stochastic programming

1.INTRODUCTION

1.1. Background and Review of the Related Works

Plug-in hybrid electric vehicles (PHEVs) are a new paradigm in transportation. Some reasons of introducing this paradigm can be expressed as follows:

• 1 Increasing emissions due to burning fossil fuels and related problems such as global warming,
• 2 Restriction on fossil fuel resources and need for using other sources of energy for vehicles, and
• 3 Advances in storage and battery technologies.

Introducing PHEVs to power system will bring some advantages and disadvantages. Using the storage capability of PHEVs may have some advantages for power systems such as frequency regulation and load-shifting. However, since vehicles usually connect to the grid in the evening when the system load is high, charging PHEVs may introduce new challenges to the power system such as a need for more generation capacity and enforcing transmission and distribution systems. A good idea which has been proposed for handling the disadvantages of the connection of PHEVs to the power system is smart charging that can be helpful because the main characteristic of PHEVs, i.e. having storage capability and no need for immediate charging generally. Based on this idea, PHEVs can be charged at midnight hours when the system demand is very low and also grid integration of the renewable generation is facilitated. A new entity which should be introduced for implementing smart charging and relating PHEVs to the electricity market is the PHEV aggregator. PHEV aggregation can result in lower charging cost for PHEV owners through smart charging at low-price hours. Moreover, PHEV aggregator may bring some benefits to the power system via shifting demand to offpeak hours and therefore decreasing the power demand at peak hours. Two main patterns of PHEV charging include unidirectional and bidirectional ones. In the unidirectional charging, PHEV aggregator only purchases energy from grid to charge batteries of PHEVs and can also participate in the ancillary service market using smart charging which controls charging rate of PHEVs. In the bidirectional charging, PHEV aggregator can sell power to the grid in addition to buying electricity from it. Unidirectional charging is the first step in the integration of PHEVs into the electricity grid and markets [1, 2], while bidirectional charging requires additional hardware that is not currently installed on the produced PHEVs [3]. It should be mentioned that using bidirectional charging increases battery cycles and degradation [4]. Hence, in this chapter, the unidirectional charging is considered. Although PHEV aggregator may provide ancillary services for the system [5, 6] and stability [7], this chapter is focused on risk constrained optimal participation in the energy market. The driving cost of PHEVs is less than that of conventional vehicles (CVs) and hybrid electric vehicles (HEVs). Annual cost of a CV is \$1,196, while it is \$549 for PHEV when no control strategy is imposed on PHEV charging [8]. If the charging profile of a PHEV is controlled, this cost is reduced to \$393 [8].

In this chapter, the optimal bidding of a PHEV aggregator to the electricity market which is inspired by the Iran electricity market will be discussed and formulated.

1.2. PHEV Configurations

Series Configuration

The required traction force of the vehicle is solely provided by the electric motor in the series PHEV. Fig, 1 shows this type of configuration. As shown in this figure, the required power for electric motor is provided by either battery pack or a generator driven by an internal combustion engines (ICE). If the battery state of charge (SOC) drops below a certain level, the ICE would drive the electric generator and the battery would be charged by the produced power of the generator. The series configuration requires two electric machines and two converters. The electric motor should be rated for maximum power. There is no mechanical linkage between ICE and mechanical transmission which lets engine to operate in the most efficient region [9].

Figure 1. Series configuration of PHEVs (adapted from [9]).

Parallel Configuration

Fig. 2 shows the parallel configuration of PHEVs, in which the traction force is provided by either electric motor or ICE. The eclectic motor and ICE are connected to mechanical transmission system using a mechanical coupling. The motor can also be operated as a generator. Unlike the series configuration, parallel configuration uses only one electric machine and one converter. The engine and electric motor of PHEV can be rated at reduced levels for short distance trips [9].

Figure 2. Parallel configuration of PHEVs (adapted from [9]).

Series-Parallel Configuration

Series-parallel configuration takes the advantages of both series and parallel configurations. Fig 3. represents the series-parallel configuration of the PHEV. As can be observed in this figure, this PHEV can operate using the electric motor alone or along with ICE. Operating point of the ICE affects the efficiency of the ICE and also power transfer [10]. Similar to the series configuration, it is possible to charge batteries by either grid power or electric generator.

It is worth mentioning that there are other configurations of PHEV that cannot be classified into the aforementioned series/parallel configurations, but are categorized as complex configuration [11].

Figure 3. Series-parallel configuration of PHEVs (adapted from [9]).

1.3. Chapter Organization

The remainder of this chapter is organized as follows: an introduction to stochastic programming in addition to the description of scenario generation/reduction methods is presented in Section 2. Risk controlling approaches in stochastic programming are also discussed in this section. The proposed problem formulation is presented in Section 3. A comprehensive illustrative example with different case studies is provided in Section 4. Finally, the concluding remarks are provided in Section 5.

2. STOCHASTIC PROGRAMMING

2.1. Stochastic Versus Deterministic Programming

Different solutions exist in the real-world decision making problems. Finding the best solution is called programming or optimization. For each problem, there exists some input data, which are called parameters, and some decisions which should be made in the optimization procedure and are called variables. In some cases, value of the parameters is definite. These problems are called deterministic programming problems. However, in most of the real-world problems, all or at least some parameters are uncertain; this type of problems is called stochastic programming. For this type of programming, decision should be made in an uncertain framework. In most problems, despite the exact values of uncertain parameters, their probability distribution functions are known.

Since the solving procedure of deterministic programming problems is usually more straightforward than that of stochastic problems, it is desired to cast stochastic programming problems into the deterministic ones. The main key for this conversion is the probability functions of the uncertain parameters. Generally, such a conversion is done using two approaches. The first way is by replacing uncertain parameters with their expected values. The second one is generating some probable instances of uncertain parameters using the information of their probability functions which are called scenarios and replacing them with uncertain parameters.

2.2. Risk Control in Stochastic Programming

Objective function of the stochastic programming problem can be a random value. Characteristics of this random value should be utilized to obtain a single and deterministic value. The expected value (mean) is the most widely used index for forming the objective function.

The expected value of a random variable does not provide any information about its deviation and therefore can lead to a great risk. Some decision makers are concerned about the associated risk of their decisions. For these decision makers (called risk-averse), risk measurement should be defined and incorporated into the optimization problem as a term of the objective function or constraints in addressing their concerns. Different risk management approaches have been proposed for scheduling conventional market participants (generation companies, large consumers, and retailers) in the literature [12, 13]. A review of the application of stochastic programming methods in deregulated energy markets was provided in [14]. Value at risk (VaR) is a measure that estimates the financial loss of a decision-maker resulting from price fluctuations with a certain confidence level [15]. Conditional value at risk (CVaR) is an extension of VaR which is defined as the difference between the expected profit and the average of potential profit values which are less than VaR [16]. Compared with VaR, CVaR is a more conservative index, because it measures the average loss exceeding the VaR value for a given confidence level [13].

2.3. Scenario Generation and Reduction

Figure 4. Flowchart of the scenario generation approach using MCS.

Each uncertain parameter is a random variable and its probability function is usually known. Some random variables which are varying over time are called stochastic process. A single realization of a stochastic process is called a scenario [17]. In scenario generation approach, a probability distribution function based on the previous experiences is considered for each stochastic variable. Generally, normal distribution is used for modeling the market price [18]. For normal distribution, mean and variance of the stochastic variable should be calculated. It should be mentioned that every distribution can be utilized to model the market price and the proposed solution procedure is not restricted to the normal distribution. In this study, Monte Carlo simulation (MCS) was used to generate scenarios [19, 20]. Fig. 4 shows the flowchart of MCS for scenario generation. Market price sampling will be described in Section 4.

Figure 5. Flowchart of the implemented scenario reduction approach.

Generally, so many samples are generated for extracting the characteristics of the stochastic process. Despite the necessity of generating many scenarios, a large number of scenarios can lead to computational intractability. For solving this problem, some scenario reduction methods have been developed, the main goal of which is decreasing cardinality, while keeping the maximum possible information of the deleted scenarios [17]. In this chapter, backward method was used for scenario reduction [21, 22], the main idea of which is eliminating the scenario with the lowest information and repeating this action until the stopping criterion is satisfied. Two common stopping criteria are reduction percentage and number of preserved scenarios. Reduction percentage is defined in terms of relative distance between reduced and original scenarios. Algorithm of backward method is presented in Fig. 5. Original scenarios and their probabilities are input data. The eliminated set at the beginning of the algorithm is empty. Distance of two scenarios is calculated and denoted by D(.,.) . For each scenario, the nearest scenario and its associated distance are determined. The nearest scenario is denoted by index r . The probability distance for each scenario, i.e., ( ) PD s ^r , is calculated as its probability, i.e., ^s , times distance with its nearest scenario, i.e., ( ) D r s . The scenario which has the minimum probability distance is chosen for elimination, i.e., PD^d . Probability of the deleted scenario is added to the probability of the nearest scenario.

2.4. Two-Stage Stochastic Programming

Stochastic programming problems happen in a horizon. Stage in the problem horizon is defined as a point of time, at which some decisions should have been made when uncertainties partially or totally vanish [17]. Based on this definition, stochastic programming problems can be divided into two-stage and multi-stage stochastic programming problems. Since the studied problem in this chapter was a two-stage stochastic programming and the multi-stage stochastic programming problems can be considered as the extension of the twostage one, only two-stage stochastic programming was introduced.

In two-stage stochastic programming, uncertainties vanish in one step. Therefore, these problems consist of two stages: before and after scenario realization. Thus, two types of decisions can be distinguished: first-stage or here-and-now decisions and second stages or wait-and-see decisions. The first stage decisions should be made before the realization of uncertain parameters. Since the scenarios are not realized at this stage, these decisions do not depend on the value of scenarios. After making the here-and-now decisions, decision-makers should wait until the uncertainties are vanished and make some decisions afterward. The waitand-see decisions depend on the scenarios.

Scenarios can be visualized using a scenario tree. A typical scenario tree with three scenarios is shown in Fig. 6. Each scenario tree is composed of nodes and branches. Nodes represent the state of the problem and branches show transition between stages where scenarios are realized. The first node of the problem is called "root" and the nodes of the last stage are called "leaves". Each way between the root node and leaves is a scenario.

A typical two-stage stochastic programming problem can be formulated as follows [17]:

$$\max_{x} c^{T} x + E(Q(s))$$ subject to $$Ax = b$$

$$x \in X$$ (1)

where Q x s ( , ) is a stochastic part of the objective function obtained using the following formulation.

$$Q(x,s) = \begin{cases} \max_{y(s)} d(s)^{\top} y(s) \ \text{subject to} \end{cases}$$

$$T(s)x + W(s)y(s) = h(s)$$

$$y(s) \in Y }, \forall s \in S$$ (2)

where x and y s( ) are the first and second stage decision variables and A , b , d s( ) , T s( ) , W s( ) , and hs( ) are parameters, and E (.) represents the expected value of its argument. This problem can be expressed as deterministic problem considering some assumptions using the following formulation [17]:

$$\max{x,y(s)} c^{\top} x + \sum{s \in S} \pi(s) d(s)^{\top} y(s)$$ subject to $$Ax = b$$

$$T(s)x + W(s)y(s) = h(s), \forall s \in S$$

$$x \in X, y(s) \in Y, \forall s \in S$$ (3)

CVaR can be used as a risk controlling measure and incorporated into two-stage stochastic programming. Mathematical formulation of the CVaR is as follows [17]:

CVaR( $$\alpha, x$$ ) = max $\left{ \eta - \frac{1}{1-\alpha} E_s \left{ \max \left{ \eta - f(x, s), 0 \right} \right} \right}, \forall \alpha \in (0, 1)$ (4)

CVaR can also be used as a term in the objective function and/or a constraint for risk limitation. Finally, the formulation of the extended model can be stated as follows:

$$\max_{x,y(s),\eta,scs} \mathbf{c}^\top x + \sum{s \in S} \pi(s)d(s)^\top y(s) + \beta \left( \eta - \frac{1}{1-\alpha} \sum_{s \in S} \pi(s)sc_s \right)$$ subject to $$Ax = b$$

$$\eta - \left( \mathbf{c}^\top x + q(s)^\top y(s) \right) \le sc_s, \forall s \in S$$

$$sc_s \ge 0, \forall s \in S$$

$$T(s)x + W(s)y(s) = h(s), \forall s \in S$$

$$x \in X, y(s) \in Y, \forall s \in S$$

$$(5)$$

where is an auxiliary variable, s sc equals max{ ( ) ( ) ,0} c x q s y s , and is a positive weighting parameter of risk measure.

Figure 6. Simple illustration of two-stage scenario tree.

3. PROBLEM FORMULATION

Main objective of PHEV aggregator is maximizing its total profit. Revenue of the PHEV aggregator comes from selling power to PHEVs and cost of aggregator is related to purchasing electricity from the grid. It is assumed that PHEV charging tariffs are predetermined based on the regulatory constraints and market gaming with other aggregators. If an aggregator sells electricity at a more expensive cost than others, PHEV owners will migrate to other aggregators. Determining energy tariff is a long- or mid-term decision which is considered as the input data for the short-term scheduling problem. Therefore, PHEV aggregator should charge PHEVs at the low-electricity-price hours such as off-peak hours to maximize its profit.

For increasing total profit and avoiding potential risks, uncertainties should be handled by the aggregator in the decision making procedure. The main uncertainty in this current framework is energy price. Economic effect of the price forecasting errors of four forecasting methods on the short-term scheduling of thermal and hydro producers was presented in [23]. Considering this uncertainty, the optimal short-term operation scheduling of the aggregator would become a stochastic programming problem.

3.1. Market Structure

One side of electricity market was considered in this chapter, in which only electricity producers were competing. The generators participate in the day-ahead market through offering value‒price pairs and consumption side of the hypothesized market, such as retailers and PHEV aggregators, only submit their demands to the system operator. If demand side players deviate from their stated demand, they incur imbalance costs. This market model is inspired by Iran electricity market.

3.2. Decision Framework

In this model, the PHEV aggregator should decide about its consumption for all dayahead hours. Thus, the aggregator should estimate the required power for charging PHEVs. This decision is made before the realization of energy prices. Afterward, the cost associated with the consumed energy is determined. The scenario set in this problem consists of electricity prices for each hour of the decision horizon and their associated probabilities. After scenario generation and reduction, the proposal of the aggregator to the market should be optimized. Since the power required for charging PHEVs ( E ,^t ) should be determined before the realization of energy price, it is a first stage or here-and-now decision and does not depend on the scenario realization. Other state variables which depend on E ,^t such as state of charge ( SOC ) and scheduled demands ( Ft ) are first stage decisions. Electricity cost is determined after the realization of scenarios as the second stage variables.

3.3. Mathematical Formulation

The problem objective is to maximize the aggregator benefit which is formulated in (6).

$$\max{F{i},E{v{f}}} \sum{t=1}^{T} \sum{s=1}^{N{s}} \pi{s}(\rho{t} - \lambda{s,t}) F_{t}$$ (6)

where, ^s is the probability of the occurrence of the s th scenario, t is the price of selling energy to PHEV owners at the t th hour (\$/KWh), and st, is the s th realized electricity price at the t th hour (\$/KWh). In this formulation, it was assumed that the aggregator sells energy to PHEV owners with constant and predetermined tariffs. Based on this assumption, the benefit term is fixed and therefore the objective function can be rewritten as (7):

$$\max_{Ft, E{xJ}} \sum{t=1}^T \sum_{s=1}^{N_s} \pis \left( -\lambda{s,t} \right) F_t \tag{7}$$

Formulation of the risk-averse two-stage stochastic programming problem using CVaR risk measure results in the following object function:

$$\max_{Ft, E{r_s}, sc, \eta} \sum{t=1}^{T} \sum_{s=1}^{N_s} \pis \left( -\lambda{s,t} \right) Ft + \beta \left( \eta - \frac{1}{1 - \alpha} \sum{s \in S} \pi_s s c_s \right)$$

$$(8)$$

Also, the two following constraints should be added to the model to limit the risk.

$$\eta - (\sum{t=1}^{T} - \lambda{s,t} F_t) \le sc_s, \forall s \in S$$

$$\tag{9}$$

$$sc_s \ge 0, \forall s \in S$$ (10)

Arrival, departure, and battery usage of PHEVs are considered known according to [1, 26]. The sold electricity to PHEV owner should be provided by the purchased energies from the grid. This relation is shown in the following equation:

$$\sum{v} E{v,t} = F_t \qquad \forall t \tag{11}$$

SOC of the battery of each PHEV is updated using power which is drawn from grid, the power used for driving ( G ,^t ), and its SOC at the previous hour using the following equation:

$$SOC{v,t} = SOC{v,t-1} + (\varsigmav E{v,t} - G_{v,t}) / C_v \qquad \forall t, \forall v$$

$$(12)$$

where, C is the battery capacity of the th vehicle and is its charging efficiency. SOC and rate of charge bounds are forced using (13) and (14), respectively.

,

t

$$\underline{SOC}{v} \leq SOC{v,t} \leq \overline{SOC}_{v} \qquad \forall t, \forall v$$

$$\underline{\underline{E}}{v} \le \underline{E}{v,t} \le \overline{\underline{E}}_{v} \qquad \forall t, \forall v \tag{14}$$

(13)

where SOC is minimum SOC and SOC is the capacity limit of battery of the v th vehicle. Moreover, minimum and maximum charging rates are denoted by E and E , respectively. Constraints (15) imposed on PHEV can be charged only at plugged times.

$$E{v,t} = 0 { \forall t, \forall v | G{v,t} \neq 0 } (15)$$

PHEVs should be charged to a sufficient level before departure time, representing by the following constraints:

$$SOC{v,t} \ge SOC{v,t}^{dep} \qquad { \forall t, \forall v \mid I_{v,t}^{dep} = 1 }$$

$$\tag{16}$$

where dep I is a flag that indicates the departure of vehicle at hour t and the required level of charging is denoted by dep ^t SOC .

4. ILLUSTRATIVE EXAMPLE

In this section, a simple case study is used to demonstrate different bidding strategies of a typical PHEV aggregator. Characteristic data PHEVs were adopted from [27]. For the sake of simplicity, a few numbers of PHEVs were considered, i.e. five PHEVs in the six hour scheduling period. Each PHEV battery had maximum and minimum allowable capacities of 25 KWh and 3 KWh. Charging efficiency of PHEVs was assumed to be 100% for the sake of simplicity. The maximum charging rate of PHEVs was limited to 50% of their capacities and therefore approximately 2 h was required for full charging. Minimum charging rate was considered zero. The optimization problem was solved using CPLEX solver [29] under general algebraic modeling system (GAMS) framework [28].

Means and variances of the electricity price are shown in Table 1.. Fig. 7 shows the typical discretization of the normal probability distribution function. The continuous normal probability distribution function is discretized at seven levels. Each error level has the width of one forecast error standard deviation ( ) [30]. Fig. 8 shows the flowchart of the generating electricity price scenarios.

Hour	e  (\$/KWh) t	e  (\$/KWh) t
1	50	10
2	47.5	9.74
3	45	9.48
4	42.5	9.22
5	40	8.94
6	42.5	9.22

Table 1. Expected and standard deviations of electricity price (\$/KWh)

Departure and arrival times of PHEVs along with the their initial SOC at the beginning of the scheduling period, arrival SOC, and the required SOC at the end of the scheduling period are shown in Table 2..

The Monte Carlo method which is described in Fig. 4 and Fig. 8 was used to generate 100 scenarios for energy prices. These scenarios were reduced using backward method. The SCENRED tool under GAMS environment was used for scenario reduction. Ten scenarios were retained after scenario reduction, as shown in Table 3..

Figure 7. Discretization of normal probability density function.

Figure 8. Flowchart of the generating electricity price scenarios.

PHEV number	Initial SOC	Departure Time	Arrival Time	Arrival SOC (%)	Required SOC (%)	final
1	50.00%	13:00	14:00	60%	100%
2	50.00%	13:00	15:00	20%	100%
3	15.00%	14:00	16:00	20%	100%
4	15.00%	15:00	17:00	20%	60%
5	15.00%	16:00	17:00	60%	100%

Table 2. Characteristics of PHEVs

Table 3. Reduced scenarios for power prices over the planning horizon

Scenario	Hour 1	Hour 2	Hour 3	Hour 4	Hour 5
1	40	37.76	63.96	42.5	48.94
2	50	47.5	45	51.72	40
3	50	47.5	35.52	42.5	31.06
4	60	47.5	54.48	42.5	48.94
5	40	47.5	54.48	42.5	40
6	60	47.5	54.48	42.5	40
7	50	57.24	45	42.5	48.94
8	50	37.76	45	42.5	48.94
9	50	57.24	35.52	33.28	48.94
10	40	47.5	35.52	42.5	31.06

Different cases were investigated. At first, deterministic scheduling was done. Then, stochastic scheduling was performed without considering risk and finally CVaR as a risk measure was added to the constraints. In this step, risk-neutral and risk-averse scheduling were compared. Moreover, the effect of the degree of risk aversion was studied.

4.1. Deterministic Scheduling

In deterministic scheduling, uncertain parameters, i.e. energy price, are replaced with their expected values. SOC of PHEV in the planning horizon is shown in Fig. 9. Expected cost of supplying PHEVs in the scheduling horizon was \$6.489.

Table 4. Cost of each scenario in the case of stochastic programming

Scenario	Cost (\$)	Scenario	Cost (\$)
1	7.147	6	7.648
2	7.686	7	7.242
3	6.983	8	7.081
4	8.334	9	6.971
5	7.48	10	6.138

4.2. Stochastic Scheduling

The reduced ten scenarios were used for the realization of energy prices. SOC of PHEV is shown in Fig. 10. In comparison with the results of deterministic scheduling, the schedule was different for the 4th hour of the first and second PHEVs and the 5th hour of the second PHEV. Schedules were the same for other PHEVs at other hours. The expected cost of supplying PHEVs was \$6.863, which seemed to be larger than the expected cost of deterministic scheduling of PHEVs. Costs of all the scenarios are shown in Table 4..

Figure 9. SOC of PHEVs during planning hours for the case of deterministic programming.

Figure 10. SOC of PHEVs during planning hours for the case of stochastic programming.

In order to evaluate the effect of stochastic programming on the final cost of aggregator, it was assumed that scenarios #1 and #8 were realized at real operation hours, which could also be called after the fact energy prices. The real operation cost for the stochastic and deterministic cases are compared in Table 5.. It can be observed that the real cost of supply was less for both of the after the fact values when stochastic scheduling was employed.

4.3. Risk-Constrained Scheduling

The scheduling was done for -CVaR with different values for . -CVaR with = 0.75 was used and results are shown in Table 6.. SC equaled zero for all scenarios, expect scenario #8. It can be seen that the expected cost decreased and the objective value increased with increasing .

Table 5. Comparing real cost of aggregator for the realizations of scenarios 1 and 8 for deterministic and stochastic programming

Scenario #1 as after the fact values
Deterministic	7.2432
Stochastic	7.1466
Scenario #8 as after the fact values
Deterministic	7.1778
Stochastic	7.0812

Table 6. Results of risk-constrained stochastic scheduling

	Objective Value	Cost	SC(8)	
0	-6.864	6.864	0.368	-1.487
1	-8.288	6.877	0.368	-1.102
2	-9.699	6.877	0.368	-1.102
3	-11.111	6.877	0.305	-1.102
5	-13.932	6.921	0.305	-1.146
10	-20.944	6.921	0.368	-1.146

CONCLUSION

In this chapter, optimal short-term operation scheduling of a typical PHEV aggregator was studied. A typical aggregator faced different uncertainties for day-ahead scheduling and an important one was energy price. Stochastic programming can be used as a powerful method for robust scheduling of PHEV charging. The objective of the aggregator in the proposed method was to maximize its total profit, while supplying the required energy of PHEVs. For a typical aggregator, the uncertainty of electricity price forecasting was considered and the short-term operation of aggregator was formulated as a stochastic optimization problem. Different stochastic and deterministic results were presented and compared for the aggregator, demonstrating that stochastic programming can result in more benefit (through lower cost) than deterministic programming for two after the fact energy prices. Moreover, for risk-averse aggregator, different schedules with naturally more costs were obtained.

REFERENCES

• [1] Sortomme, E.; El-Sharkawi, M.; Optimal charging strategies for unidirectional vehicleto-grid. IEEE Transactions on Smart Grid, 2 (1) (2011) 131-138.
• [2] Sortomme, E.; El-Sharkawi, M.; Optimal combined bidding of vehicle-to-grid ancillary services. IEEE Transactions on Smart Grid, 3 (1) (2012) 70-79.
• [3] Peterson, S. B.; Whitacre, J.; Apt, J.; The economics of using plug-in hybrid electric vehicle battery packs for grid storage. Journal of Power Sources, 195 (8) (2010) 2377- 2384.
• [4] Peterson, S. B.; Apt, J.; Whitacre, J.; Lithium-ion battery cell degradation resulting from realistic vehicle and vehicle-to-grid utilization. Journal of Power Sources, 195 (8) (2010) 2385- 2392.
• [5] Guille, C.; Gross, G.; A conceptual framework for the vehicle-to-grid (v2g) implementation. Energy Policy, 37 (11) (2009) 4379-4390.
• [6] Tomic, J.; Kempton, W.; Using fleets of electric-drive vehicles for grid support. Journal of Power Sources, 168 (2) (2007) 459-468.
• [7] Kempton, W.; Tomic, J.; Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable energy. Journal of Power Sources, 144 (1) (2005) 280-294.
• [8] Sioshansi, R.; Miller, J.; Plug-in hybrid electric vehicles can be clean and economical in dirty power systems. Energy Policy, 39 (10) (2011) 6151-6161.
• [9] Zhang, X.; Mi, C.; Vehicle Power Management Modeling, Control and Optimization. Springer-Verlag, 2011.
• [10] Park, J.; Oh, J.; Park, Y.; Lee, K.; Optimal power distribution strategy for series-parallel hybrid electric vehicles. The 1st International Forum on Strategic Technology, Gyeonggi, South Korea, (2006) 37-42.
• [11] Zhang, P.; Yin, C.; Zhang, Y.; Wu, Z.; Optimal energy management for a complex hybrid electric vehicle: Tolerating power-loss of motor. Journal of Shanghai Jiaotong University (Science), 14 (4) (2009) 476–481.
• [12] Kazempour, S.; Moghaddam, M. P.; Haghifam, M. R.; Yousefi, G. R.; Risk-constrained dynamic self-scheduling of a pumped-storage plant in the energy and ancillary service markets. Energy Conversion and Management, 50 (5) (2009) 1368–1375.
• [13] Mohammadi-Ivatloo, B.; Zareipour, H.; Amjady, N.; Ehsan, M.; Application of information-gap decision theory to risk-constrained self-scheduling of gencos. IEEE Transactions on Power Systems, 28 (2) (2013) 1093-1102.
• [14] Most, D.; Keles, D.; A survey of stochastic modelling approaches for liberalised electricity markets. European Journal of Operational Research, 207 (2) (2010) 543–556.

• [15] Garcia-Gonzalez, J.; Moraga, R.; Mateo, A.; Risk-constrained strategic bidding of a hydro producer under price uncertainty. 2007 IEEE Power Engineering Society General Meeting, Madrid, Spain, (2007) 1–4.

• [16] Garcia-Gonzalez, J.; Parrilla, E.; Mateo, A.; Risk-averse profit-based optimal scheduling of a hydro-chain in the day-ahead electricity market. European Journal of Operational Research, 181 (3) (2007) 1354–1369.

• [17] Conejo, A. J.; Carrion, M.; Morales, J. M.; Decision Making Under Uncertainty in Electricity Markets. Springer, 2010.

• [18] Li, T.; Shahidehpour, M.; Li, Z.; Risk-constrained bidding strategy with stochastic unit commitment. IEEE Transactions on Power Systems, 22 (1)(2007) 449–458.

• [19] Kall, P.; Wallace, S. W.; Stochastic Programming. John Wiley & Sons, 1994.

• [20] Birge, J. R.; Louveaux, F.; Introduction to Stochastic Programming. Springer, 2011.

• [21] Growe-Kuska, N.; Heitsch, H.; Romisch, W.; Scenario reduction and scenario tree construction for power management problems. 2003 IEEE Bologna Power Tech Conference, Berlin, Germany (2003).

• [22] http://www.gams.com/dd/docs/solvers/scenred.pdf, [accessed on September, 2012].

• [23] Mohammadi-Ivatloo, B.; Zareipour, H.; Ehsan, M.; Amjady, N.; Economic impact of price forecasting inaccuracies on self-scheduling of generation companies. Electric Power System Research, 81 (2) (2011) 617-624.

• [24] U. D. of Energy, Benefits of demand response in electricity markets and recommendations for achieving them, A report to the United States congress, February, 2006.

• [25] Chang-Chien, L.R.; An, L. N.; Lin, T.W. ; Lee, W.J.; Incorporating demand response with spinning reserve to realize an adaptive frequency restoration plan for system contingencies. IEEE Transactions on Smart Grid, 3 (3) (2012) 1145-1153.

• [26] Sortomme, E.; El-Sharkawi, M.; Optimal scheduling of vehicle-to-grid energy and ancillary services. IEEE Transactions on Smart Grid, 3 (1) (2012) 351–59.

• [27] Pantos, M.; Stochastic optimal charging of electric-drive vehicles with renewable energy, Energy. 36 (11) (2011) 6567-6576.

• [28] Brooke, A.; Kendrick, D.; Meeraus, A.; Raman, R.; GAMS: A user"s guide. GAMS Development Corporation, 1998.

• [29] Brooke, A.; Kendrick, D.; Meeraus, A.; aman, R.; GAMS/Cplex 7.0 User Notes. GAMS Development Corporation, 2000.

• [30] Amjady, N.; Aghaei, J.; Shayanfar, H.; Stochastic multiobjective market clearing of joint energy and reserves auctions ensuring power system security. IEEE Transactions on Power Systems, 24 (4) (2009) 1841-1854.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 12

ENERGY EFFICIENCY OF THE HYBRID POWER SOURCE USED IN THE PLUG-IN FUEL CELL VEHICLES

Nicu Bizon1,2, , Marian Raducu¹ , Mihai Oproescu¹ and Luminita Mirela Constantinescu¹

¹University of Pitesti, Pitesti, Romania ²University Politehnica of Bucharest, Bucharest, Romania

ABSTRACT

In this chapter a mathematical analysis of the energy efficiency for the series and parallel Multiport Power Converter (MPC) topologies used in the Plug-in Fuel Cell Vehicles (PFCVs) is presented. Besides this study, that is performed and validated in the second part, the first part presents a brief overview over the electric vehicles with a range extender, which also include the electrical motors and inverter systems used, the main parameters of the DC input sources for electrical vehicle, and the usual strategies for the energy management unit. The aim of the analytical study shown in the second part is to provide some general strategies for the energy management unit operating under a wide range of PFCV regimes. The goal is to choose the optimal parameters of the MPC control for an efficient use of the PFCV during each particular drive cycle. In relation with the FC system of PFCV, the Energy Storage System (ESS) can be operated in the following regimes: (1) charge-sustaining, (2) charge-depleting, and (3) charge-increasing.

Considering the imposed window for the ESS state-of-charge, the MPC can be connected to the renewable plug-in charging stations to exchange power with a smart grid, when it is necessary for both. The energy management unit that communicates with the smart grids will establish the moments to match the PFCV power demand with the grid supply availability, stabilizing it. The MPC energy efficiency of the PFCVs is studied when the ESS is charged (discharged) from (to) the smart grid. The comparative results were shown for both PFCV architectures through the analytical calculation performed and the appropriate Matlab/Simulink® simulations presented.

Corresponding author: Nicu Bizon. University of Pitesti, Pitesti, Romania. E-mail: nicubizon@yahoo.com, nicu. bizon@upit.ro.

Keywords: Plug-in Vehicles, Fuel Cell Hybrid Power Source, Energy Storage System, Stateof-Charge, Multiport Power Converter, Energy Efficiency

NOMENCLATURES

RES Renewable Energy Sources

PV Photovoltaic panels WT Wind Turbines HT Hydraulic Turbines

PEMFC Proton Exchange Membrane Fuel Cell

ESD Energy Storage Devices HPS Hybrid Power Source HEV Hybrid Electric Vehicles FCV Fuel Cell Vehicles

EGS Energy Generation Systems DG Distributed Generation

UPS Uninterruptible Power Supplies EMU Energy Management Unit ESS Energy Storage System

MPPT Maximum Power Point Tracking MEP Maximum Efficiency Point

SOC State-of-Charge EV Electric Vehicle

ICE Internal Combustion Engine PCS Plug-in Charging Stations PHEV Plug-in Hybrid Electric Vehicles PFCV Plug-in Fuel Cell Vehicles

FEV Full Electric Vehicle

EREV Electric Vehicle with a Range Extender

PI Power Interface IM Induction Motor

SRM Switched Reluctance Motor

PMSM Permanent Magnet Synchronous Motor

BLDC Brushless DC motor PMBLDC PM Brushless DC ZSI Z-Source Inverter DSP Digital Signal Processing

CVS Controlled Voltage Source CCS Controlled Current Source

UC Ultracapacitor AGM Absorbed Glass Mat

VRSLAB Valve Regulated Sealed Lead-Acid Battery

CS Charge-Sustaining CD Charge-Depleting

CI Charge-Increasing V2G Vehicle-to-Grid MPCs Series MPC MPCp Parallel MPC

1.INTRODUCTION

In the last decade an increased interest in flexible and effective systems of power management, which are able to harvest energy from various sustainable sources of energy, such as the Renewable Energy Sources (RES) (photovoltaic panels - PV, wind turbines - WT, hydraulic turbines - HT etc.) or hydrogen-based energy sources (Proton Exchange Membrane Fuel Cell - PEMFC) appeared.

Using at least one energy source from those mentioned above and Energy Storage Devices (ESD: batteries, ultracapacitors, etc.) combined in a hybrid stack (passive, semiactive or active controlled mode). These merged technologies are used as Hybrid Power Source (HPS) in Hybrid Electric Vehicles (HEVs) and Fuel Cell Vehicles (FCVs) [1], in renewable Energy Generation Systems (EGS) [2], in distributed generation (DG) in smart grids [3], in uninterruptible power supplies (UPS) [4], and so on. Usually, the voltage-current characteristics of the energy sources and storage devices are nonlinear and their voltage levels are different from those of the loads. Consequently, a Multi-Port Converter (MPC) must be used to manage the power flow between the inputs and the outputs (energy sources, storage devices and loads) based on an Energy Management Unit (EMU) [5].

The concept of MPC (or Multi-port Power Electronic Interface) is commonly adopted to process the renewable power from multiple sources and loads [6-8], having the following main features: (1) maximum energy harvesting from renewable sources, (2) optimal management of energy from multiple sources, (3) optimal Energy Storage System (ESS) management, and (4) adaptive energy management system for the best performance. The MPC represents a particular case of energy hub concept that is considered as a unit where multiple energy carriers can be converted, conditioned, and stored, representing an interface between different energy sources and loads [9-11]. For example, the MPC interfacing the HEV DC buses must meet the following requirements:

• › Electrical isolation between the low voltage (LV) and the high voltage (HV) buses;
• › Bidirectional power flow for all input and output ports;
• › Continuous supplying of the loads;
• › Short response time to load changes;
• › High energy efficiency and power density
• › Safe start-up without complex circuits.

The MPC creates an interface between the loads, the renewable sources, and the storage devices to efficiently provide and recover the power. Consequently, the Maximum Power Point Tracking (MPPT) control guarantees the optimal energy harvested from the energy sources that have a power characteristic with a maximum at their MPP. Generally, it is better to operate the FC system at the Maximum Efficiency Point (MEP), but MPP can also be used because this is easier to track in comparison with the MEP that is more difficult to be estimated [12]. The generic FCV architecture shown in Figure 1 is based on the flexible MPC topology, generalizing the most used MPC topologies in automotive applications such as the series and parallel MPC topologies. These topologies request the use of the ESS on the LV and HV bus, respectively [13, 14].

In [14, 15] it is shown that a hybrid MPP topology is more efficient than both series and parallel MPC topologies when the ESS is operated to have final State-Of-Charge (SOC) different from the initial SOC. This case may be of interest in operating the PFCV. In this chapter will be shown for basic MPC topologies that each could work more efficiently under certain conditions that will be specified by analytical calculation performed. Consequently, the basic MPC topology helps to boost the energy efficiency performance as well as the flexibility in operating the PFCV to increase the driving range and fuel economy. Also, the mass of the MPC could be reduced by using the modularization and integration techniques for MPCs [6], but this is off the topic of this chapter. Finally, it can be noted that the MPC concept represents both integrated PIs and their appropriate control, and also the fact that they operate efficiently under the energy management strategies implemented in the EMU.

A brief comparison of the energy management approaches applied for the PEMFC systems in the literature is made in [16]. The energy management of multiple sources and loads is a complex and difficult task that needs to be solved by the specialists using different strategies based on (1) intelligent concepts [17, 18], (2) local and global optimization approaches [19, 20], (3) frequency decoupling techniques [21, 22], (4) linear and nonlinear controllers used in multiple control loops [23] etc.

The ECU design took recently a more important and detailed place in literature, especially based on the fuzzy logic and on the nonlinear optimization [16, 24].

Figure 1. Plug-in Fuel Cell Vehicle - generic MPC topology [14].

This design issue is mainly related to the ECU features of easy adaptation to more complex MPC topologies, computational efficiency, driving rules integration, capacity to modeling uncertainties, etc. [25].

Real-time optimization based on equivalent consumption minimization strategy may be a good choice for the online implementation of the ECU [20]. Finally, all the proposed ECU strategies compare the MPC energy efficiency in term of equivalent hydrogen cost.

In the first part of this chapter (Introduction and the next four sections) the state of art for EVs and HEVs was briefly presented, especially for the FCVs with or without the plug-in feature integrated.

So, these sections shortly present the common electrical motors and inverter systems usually used in automotive applications, the main features of the FC stack and ESS, and the appropriate EMU strategies.

Consequently, a MPC must be used to manage the power flow between the inputs and the outputs (energy sources, storage devices and loads) based on an optimum EMU strategy. Section 6 presents the main MPC topologies proposed in the literature in order to understand the energy efficiency analysis of the MPC performed in the last sections.

The features of the MPCs, which are scalable to meet different power requirements, are presented in relation with the DC input sources in Section 7.

The analysis of the two MPC topologies is presented in the last five sections in term of energy efficiency related to the ESS power flow and the MPC parameters" values.

This part is organized as follows. Sections 8 and 9 present the series and parallel MPC topologies operating in the three operating modes of the ESS. The next two sections (Sections 10 and 11) will compare the energy efficiency for the series and parallel MPC topologies in each ESS operating mode over another.

This analysis is performed to increase the energy efficiency of the whole MPC for series and parallel HPS architectures, which means the use of a LV or HV ESS type, respectively. The analysis is made in term of energy efficiency related to the ESS power flow and MPC parameters. Finally, some conclusions are given related to the PFCV capacity to stabilize the grid by charging/discharging the ESS under EMU control.

2. PLUG-IN ELECTRIC VEHICLES WITH A RANGE EXTENDER

The Electric Vehicle (EV) is a technology that promises to drastically reduce emissions associated with the road transport. In the last decade the technology has been supported by different manufacturers and specialists as a key element in reducing CO² emissions (as well as emissions of pollutants and noise) of cars and light commercial vehicles.

But at the same time, EV technology is still far from being projected as necessary, emphasizing too many uncertainties regarding the issues to be addressed, such as [26]:

• › Well-to-wheel impacts on emissions;
• › Interaction with the DG system;
• › Cost of large scale introduction;
• › The batteries technology (their capacity related to vehicle and road range, fast charging, durability, availability and environmental impacts of used materials).

In addition, advantages such as high energy-efficiency and zero environmental pollution will impose the battery-powered EV technology to the internal combustion engine (ICE) vehicles conventional, even if the EV performance is far less competitive than ICE vehicles, due to the much lower energy density of the batteries than that of gasoline.

Consequently, to overcome their disadvantages, the HEV technology was promoted by the main companies that design and produce cars in order to exploit the advantages of the both ICE and EV technologies. HEV combines an ICE with an on-board rechargeable ESS to achieve better performances regarding the fuel economy and road range. The HEVs models are currently produced by Toyota (Lexus), Honda, Chevrolet, Ford, Mercedes Benz etc.

The plug-in concept means the interaction of the vehicle with the DG system through an internal or external charger. Plug-in vehicles can be classified into different categories such as EV, HEV, Plug-in Hybrid Electric Vehicles (PHEV), and plug-in fuel cell vehicles (PFCV) [27]. EVs classification is given in [28], as below:

• Full Electric Vehicles (FEVs) that have an electric motor and no ICE or Fuel Cell (FC) system.

PHEVs have in addition a charger;

• Electric Vehicles with a Range Extender (EREVs) that have one or more electric motors and an ICE or a FCS, which is used to extend the road range.

A FCV is an EV with a FC system operating as range extender. The car producers announced their FCVs based on ongoing research into the development of the hydrogen technology, even if hydrogen production, storage and distribution are in early stage of implementation. Thus, the FCVs will have a long way before entering the market.

In this chapter will be shown that many design options for the PFCV-MPC topologies are available. The FC system powers the DC bus via a DC-DC power converter (Power Interface 1 - PI1) integrated in the PFCV-MPC topology (see Figure 1), where the inverter system and PI2 are of bidirectional type based on different power converter"s topologies [29, 30]: bibuck, bi-boost, or hybrid integrated topologies.

Note that the power flow during the regenerative braking process is from the load to the ESS via the inverter diodes and PI2 that operate in buck mode [30]. Considering the higher cost of the bidirectional type against the unidirectional type, the bidirectional power flow is split in two unidirectional power flows.

The load is powered via the unidirectional inverter and boost converter - PI2 and the ESS is charged during the brake stage via the unidirectional PI3 and PI4 (of DC-DC and AC-DC type) [29]. PI 4 operates equivalent to the series connection of PI5 and PI2 during the regenerative braking stage.

Thus, the MPC shown in Figure 1 is a generic topology to study MPC energy efficiency, considering the FC system and the ESS as energy sources, and the AC electrical machine(s) as output(s).

3. ELECTRICAL MOTORS AND INVERTER SYSTEMS USED IN THE HYBRID ELECTRICAL VEHICLE

Considering the efficiency, cost, size and weight, lifespan and maintainability as performance indicators, the electric traction motors used for HEVs are of following type: the Induction Motor (IM), the Switched Reluctance Motor (SRM), the Permanent Magnet (PM) Synchronous Motor (SM) motor (PMSM), the Brushless DC motor (BLDC), and the new motors such as PM Brushless DC (PMBLDC) [1, 13].

The electric motor operates as a generator to recharge the ESS during the braking stage until to the emergency stopping phase or if the ESS is near to full charge. In a single foot pedal both regenerative braking and mechanical braking modes appear. In the first part of the foot pedal the regenerative braking power flow charge the ESS.

However, the braking requirements for all cars will have to be enhanced in the next decade to meet the new safety stop requirements.

The PM motors have higher efficiency in comparison with the IM, due to the elimination of magnetizing current and copper loss in the rotor. Note that high performance torque control could be obtained using PMBLDC motors, even if these require complex control to operate in both motoring and generating modes [31]. Also note that the IMs and SRMs have the highest reliability, besides other new motors designed for the HEV applications [32].

The electric motor power varies from 10 kW to 100 kW. So inverter system must be designed accordingly. The basic inverter topologies are shown in Figure 2.

The first (a) is a traditional PWM inverter that operates in rectified mode during the regenerative braking. The second (b) is a DC-AC PI (also named inverter) plus a bidirectional DC-DC PI. The third (c) is a Z-Source Inverter (ZSI), with the unidirectional variant shown in plot (d).

Figure 2. (Continued)

Figure 2. Inverter topologies [14], (a): Traditional PWM inverter, (b): Inverter plus a bidirectional DC-DC PI, (c): Bidirectional Z-source inverter, (d): Unidirectional Z-source inverter.

The bidirectional PIs (Figure 1) will operate in boost and buck mode during the motoring and regenerative braking regime. The voltage on ESS bus usually has a large variation and must be stabilized to the input level required by the inverter [1, 13]. The inverter system will be designed to handle this ESS voltage range at full power load. Note that double of the rated current must be considered at 50% of rated ESS voltage, so the cost of the inverter system will increase. Consequently, the bidirectional PI2 must to be used in order to minimize the inverter stress [33]. The ZSI topology shown in Figure 2c is a PI topology that has interesting features such as buck-boost characteristics and single stage conversion, which can overcome the above mentioned problems [34]. Note that any desired AC output voltage can be obtained, even one greater than the input DC voltage, making the ZSI fed adjustable speed drive systems competitive for the FC automotive applications [35, 36].

A bidirectional ZSI with nine-switches can replace the conventional HEV inverter [37], using the resonant-phase leg inverters instead of hard-switching inverter and new control techniques to increase the energy efficiency and reduce the ripple on the DC bus [38, 39]. Note that the inverter energy efficiency is usually from 0.8 to 0.95.

To better use the PIs must be modular and scalable [6], obtaining a higher production volume and thus a lower cost. Besides this design, the system package, standardization and interoperability among PIs and system are necessary to be considered in MPC design. Such of MPCs will allow designers to use a hub-energy-block approach based on specific functions defined. Thus, the following MPC features should be provided for the design [13]:

• › Each MPC should have at least one DC-port and one AC-port;
• › All MPC ports must be bidirectional and should be able to operate in a buck or boost mode;
• › MPC device"s rating is limited by manufacturer datasheets, so paralleling of MPCs is necessary for high power applications;
• › The wiring will be based on the required load power;

› Filters, transformers, and other external circuit designs will be dependent on the implemented operational power.

The overall cost of power system used in vehicle or Plug-in Charging Stations (PCS) will be decreased based on this design concept.

4. THE DC INPUT SOURCES OF THE FC VEHICLE

The DC input sources for the FC vehicle shown in Figure 1 are the FC system and ESS.

4.1. Fuel Cell System

In comparison with the ICE, the FC system has the advantage of high energy efficiency of the FC system, but its slow response requests the mandatory use of ESS. The ESS employs high-power and high-energy density devices, such as hybrid stack of UCs and batteries. Hybridization of the FC system with an ESS is an effective technology to overcome the disadvantages of the FC-alone-powered vehicles.

The PEMFC stack represents one of the most used solutions as main energy source in automotive applications because of its small size, the ease of construction, a fast start-up and low operating temperature. Unfortunately, its relatively short life cycle is still an impediment to FCV commercialization [13, 28]. The performances as energy efficiency [40, 41] and PEMFC life cycle [42] are dependent to MPC topology and control. For example, the magnitudes of low frequency harmonics of the PEMFC current ripple affect both performance indicators mentioned above. Consequently, the PI2 must have a control loop for ripple mitigation, which will force the ESS to supply the sharp load power profile [23]. The PEMFC stack (PFC) and ESS stack (PESS) assure the power flows on the DC bus via the PI1 (*Pout*1) and PI2 (*Pout*2), respectively. The power balance on the DC bus is PDC=*Pout*1+*Pout*2, where the load power flow, PDC, is the input DC power required by the inverter system. The EMU control the power flows of Controlled Voltage Source (CVS) and Controlled Current Source (CCS) in order to operate the FC system near to MPP and ESS inside of the SOC window [1, 43]. The MPP current (IMPP) can be tracked based on current-mode control using the FC current as control variable.

4.2. Energy Storage System

The on-board batteries of the ESS allow the energy to be stored during regenerative braking and provide peak power to the motor controller during vehicle acceleration. To achieve a higher energy balance on the ESS bus of the FCV, the ESS batteries stack is charged by a fraction (kFC) of the FC power (PFC), too. Also, taking into account the plug-in facility of the PFCV, all of these requirements mean the necessity of a high charge acceptance capability for batteries used.

There is a wide variety of batteries" technologies from standard flooded and enhanced flooded to next Absorbed Glass Mat (AGM) and lithium-ion polymer batteries, which are all claimed to be "best choices" for automotive and DG applications.

Because there are other chapters dedicated to the ESS, in this section only a brief review of main batteries" technologies is shown (Table 1) [44]:

• Nickel Cadmium (NiCd) mature technology, but relatively low in energy density. The NiCd is used where long life, high discharge rate and lower costs are important. It is better to note that NiCd contains toxic metals, so is environmentally unfriendly.
• Nickel-Metal Hydride (NiMH) has a higher energy density compared to the NiCd at the expense of reduced cycle life. NiMH contains no toxic metals.
• Lead Acid mature technology and most economical for high power applications where weight is of little concern.
• Lithium Ion (Li-ion) fastest growing battery technology. Li-ion is used where highenergy density and lightweight is of main importance. The technology is still fragile and a protection circuit is required to assure safety.
• Lithium Ion Polymer (Li-ion polymer) takes the features of the Li-ion technology in ultra-slim geometry and advantaged packaging. These batteries may also power the next generation vehicles, but the cost of an electric car using this type of batteries is prohibitive now.

Typically, the current Li-ion batteries use a carbon-based anode and a metal oxide cathode. The research on the next Li-ion batteries will continue the development of electrode and electrolyte materials and chemistries in order to increase the life cycle and energy density of the battery while reducing size and weight. The most promising technologies appear to involve silicon, sulfur and air (oxygen) (LiS and Li air technologies in development shown in Figure 3) and another important development is research into nanotechnologies.

The AGM batteries are constructed differently than the traditional flooded or gel battery. In AGM batteries (also called starved electrolyte), there is a thin ultra-fine fiberglass mat sandwiched between the plates that are saturated with battery acid to about 95% of what they can hold. AGM sealed battery technology was developed and introduced in 1985 for military aircraft where energy density, safety, and reliability were the main design parameters.

An AGM battery is named as VRSLAB (Valve Regulated Sealed Lead-Acid Battery), and sometimes is referred to as VRLA (Valve Regulated Lead-Acid). AGM batteries are not the best choice for all applications - they are rather expensive compared to flooded batteries.

However, their safety and design features make them the battery of choice for many applications (for example, applications where freezing temperatures could occur in some situations), including automotive applications.

As it is known, the most current HEVs use NiMH batteries [44]. However, the most likely alternative battery technologies for use in PHEVs, EREVs and FCEVs are Li-ion and ACM batteries that have the advantage of higher energy densities and charge-discharge efficiency (Figure 3).

Table 1 compares the characteristics of the six most commonly used rechargeable battery systems in terms of energy density, cycle life, static and dynamic requirements and cost.

The efficiency of a battery can be calculated as the amount of power discharged by the battery divided by the amount of power delivered to the battery.

This takes into account the loss of energy to heat, which warms up the battery. The charge-discharge efficiencies of various batteries are summarized in Table 1, too. The internal resistance of a battery represents an overall charge/discharge efficiency indicator.

A low value of it means a high dynamic capacity for example to deliver high currents without significant drops in voltage, but also to obtain a short time in fast charging regime.

The new semi-solid flow batteries described in [45] and nanotechnologies in development will overcome the limitation of energy density, providing a possible 10-fold improvement in this performance over the present liquid flow-batteries, and a lower-cost manufacturing than the conventional lithium-ion batteries 44.

The SOC of the ESS should be kept inversely proportional to the vehicle"s speed [1, 13]:

• At low speeds it is probable that vehicle will accelerate in the near time, so the ESS should have a high SOC to power that acceleration.
• At high speeds, it is probable that vehicle will be brake soon, so the ESS should have a low SOC to have capacity to accept as much of the regenerative braking energy as possible.

Consequently, the durability and lifetime of the battery under SOC deep cycling are key measures of the battery performance. Automotive batteries require a cycle life of at least 5,000 cycles of charge – discharge and a calendar lifetime of over ten years [44].

To overcome this situation, the hybridization of ESS by using UCs stack seems to be now the best way to obtain high performance for ESS.

Several ESS topologies (passive, semi active and active battery–UC hybrid) have been proposed and tested for usage in PFCV [13, 26]. To boost the UC stack voltage to ESS voltage a bidirectional PI is required based on the UC semi-active ESS topology. The operating window centered on SOC middle (around 0.7 SOC) assures a high efficiency and flexibility of the ESS operation in all regimes [13] in comparison with the passive ESS topology, which is still the most used because it is cheap and easy to implement [26]. The active or semi-active ESS topologies use a large SOC window, reducing the incremental cost [16, 46] and increasing the use of the overall energy [47].

Figure 3. Battery technology roadmap.

	NiCd	NiMH	Lead Acid	Li-ion	Li-ion	AGM
					polymer	VRSLAB
Gravimetric Energy Density (Wh/kg)	45-80	60-120	30-50	100-200	150-250	>100
Internal Resistance [mΩ] (includes peripheral circuits)	100 to 200 12V pack	200 to 300 6V pack	<100 12V pack	150 to 250 7.2V pack	200 to 300 7.2V pack	<50 12V pack
Cycle Life [cycles] (to 80% of initial capacity)	1500	500	300	1000	500	3000
Fast Charge Time	1h typical	2-4h	8-16h	2-4h	2-4h	<1h
Overcharge Tolerance	moderate	low	high	very low	low	Very high
Charge/Discharge Efficiency [%]	70% - 90%	66%	50% - 92%	80% - 90%	> 90%	98%
Self-discharge / Month (room temperature)	20%	30%	5%	10%5	~10%	1-3%
Cell Voltage(nominal)	1.25V	1.25V	2V	3.6V	3.6V	2V
Load Current - peak / rated value [%]	2000	500	500	>200	>200	>2000
Operating Temperature	-40 to	-20 to	-20 to	-20 to	0 to	-40 to
(discharge only)	60°C	60°C	60°C	60°C	60°C	70°C
Maintenance Requirement	60 days	60 to 90 days	3 to 6 months	not req.	not req.	not req.
Typical Battery Cost (US\$, estimative only)	\$100 (7.2V)	\$60 (7.2V)	\$25 (12V)	\$100 (7.2V)	\$100 (7.2V)	\$100 (12V)
Cost / Life Cycle (\$/cycles)	\$0.08	\$0.12	\$0.1	\$0.1	\$0.29	\$0.03- 0.04
Commercial use since	1950	1990	1970	1991	1999	1985

Table 1. Characteristics of commonly used rechargeable batteries

However, until the PFCV will be plugged to PCS, EMU must decide how best will use the FC stack to force the ESS to operate in (1) Charge-Sustaining (CS), (2) Charge-Depleting (CD), or (3) Charge-Increasing (CI) mode. Thus, the battery energy must be sufficient to operate the vehicle on anticipated use-restricted zone or distance to the next PCS based on the stored electrical energy during initial CD operation. The ECU will switch to CS operation if the next PCS is far away or to CI mode if the next PCS has communicated its energy demand availability. The best EMU energy strategy for different load profiles [48] based on the communication features implemented [49] will be presented briefly in next section.

5. ENERGY MANAGEMENT UNIT

From the power flows balance in hybrid MPCs point of view, the energy is balanced by the DC bus voltage regulation, usually using an ESS [15]. This chapter is focuses on this kind of MPC topology powered by a FC system and equipped with an ESS that usually has two energy storage devices: UC and battery.

The UCs stack, as a high dynamic and high power density device, operates for supplying power to regulate the DC bus voltage on load transitions. The batteries stack operates as a high energy density device to satisfy the power flows balance on the DC bus and to keep the UC stack charged, which could be connected to the ESS bus via a bidirectional PI to obtain high performance in operating the ESS in CS mode [50].

Because the FC system has the slow response, the EMU will control the FC system to operate at the average load power to keep the ESS stack in CS mode [1, 13]. Besides this power control loop, there are other two voltage controls" loops are shown in Figure 1: the DC bus voltage regulated by the ESS and the ESS voltage regulated by the FC system. If the ESS active topology will be used, then the UC voltage will be regulated by the battery stack [51]. The FC system and hybrid ESS can assure the best performance for FCVs propulsion [52, 53]. The FC system and ESS can be downsized to fulfill the average and transient power demand based on the load following strategy [52]. The dynamic performances of a FCV operating with and without the assistance of the ESS are investigated in [53, 54].

As it was mentioned above, the MPC topology and EMU features implemented set the (1) cost, (2) efficiency, and (3) reliability performances of the PFCV. The trade-off between the complexity of the EMU and the system performance may provide different results for different types of PFCVs that operate in the specified regimes under a given driving cycle [13, 54]:

• CS Mode is the operating mode in which the ESS SOC may fluctuate, but on average this is maintained at a certain level while driving.
• CD Mode operating mode in which the average of ESS SOC decreases while driving.
• CI Mode operating mode in which the average of ESS SOC increases while driving.

The CI mode is a specific operating mode only for FCV in which the ESS is charged (PESS< 0) while driving. Note that PESS represents the average power exchanged by the ESS during a drive cycle:

• If PESS> 0, then this power represents the net power discharged by the ESS, which operates in CD mode. The PFCV must be connected to the PCS as soon as the SOC level reaches the low value imposed.
• If PESS< 0, then this power represents the net power from the FC system and regenerative load that charged the ESS, which operates in CI mode. The PFCV must be connected to the PCS as soon as the SOC level reaches the high value imposed.

In this way the PFCV could operate as grid stabilizers if EMU will match the power demand to supply availability of the DG system via PCSs. It is obvious that connection to the DG system allows more opportunities for all plug-in vehicles, including their autonomy. The plug-in vehicles autonomy means that these can not only charge their ESS, but they can also discharge the available ESS energy into the DG grid. In these operating regimes the plug-in vehicles can support the DG grid and this concept is named vehicle-to-grid (V2G). However, plug-in vehicles should not be confused with the V2G concept, in which the electric drive vehicles can provide value to electric utilities through peak saving and auxiliary services. V2G defines system architecture in which plug-in vehicles, such as EVs, PHEVs and PFCVs, communicate with the DG system to sell demand response services by either delivering electricity into the grid or by controlling their charging rate. Such operating regimes for PFCV will be considered in the energy efficiency analysis performed in this chapter.

The EMU controls the PIs to operate the ESS in CD, CI or CS mode in order to increase the energy efficiency of the whole MPC. The main design factors for EMU are the SOC regulation, ESS life, power dynamic of load, power capacity of the FC system near to its MPP, and available or required power to or from the grid.

In next sections it will be performed the analysis of the energy efficiency of the whole MPC based on the SOC window width, the energy efficiency of each PI, the k^L regenerative braking factor, the kFC fraction of FC power used to charge ESS, and the road profile. To implement the PCSs into DG system it is necessary to upgrade the level of [55]:

• Wide Area Monitoring Systems.
• Two way distribution grids.
• Two-way high speed communications.
• Renewable Energy Sources integration.

Furthermore, plug-in vehicles can be used as backup power to a home via the bidirectional charger, which usually is a bidirectional two-stage topology based on high frequency transformer for isolation and/or obtaining of a HV conversion ratio.

The EMU receives the measured state variables from the driver (signals from the accelerator pedal and brake pedal) and feedback from the drive train (for example, the speed signal) and other components (Figure 1) to set the levels for power flows based on the power balance.

6. MULTI-PORT CONVERTER TOPOLOGIES

In this section the MPC topologies usually used for the PFCV that exchange power with the PCS grid connected or autonomous renewable PCS (Figure 4) will be analyzed. The key issue related to the grid integration will be the matching of the PFCV power demand to the grid supply availability. It is obvious that a grid connection opens more opportunities for all plug-in vehicles. The PFCV with a bidirectional charger can not only charge the ESS from the EP system, but it can also discharge the available ESS energy into the grid. The PFCV based on the EMU strategies supports the grid using two-way communication between the EP system and the vehicle, enables a better management of the energy sources, and allows the vehicle owners to earn money by selling power back to the grid. In this way plug-in vehicles could become more like "cash cows" that produce income from existing equipment and less like vehicles that merely consume energy [15]. The EREVs having as Range Extenders a FC or/and ICE (Figure 5) are in production phase, but the research in reducing the losses is under work. The data from Figures 4 and 5 give the values of the losses under an urban (highway) cycle [56].

In last the decade a lot of work was devoted to reducing the MPC efficiency [56], besides the work to obtain low cost and compact structure for the MPC topologies proposed in the literature [57, 58].

Figure 4. PFCV architecture.

Figure 5. Electric Vehicles with a Range Extenders (FC and ICE).

The MPC are designed for various applications, such as achieving three bus voltages of 14 V/42 V/HV bus (voltage on HV bus been of around 500 V) in EV and HEV [1, 13], interfacing the PV panel and a battery to a regulated 28-V bus in a satellite platform power systems [59], PV energy harvesting [60], FC energy harvesting [61], and HPS [62].

From the topology point of view, the MPC are based on buck-boost, half and full bridges and interleaved topologies 63. The basic bidirectional switching cells used in the MPC are shown in Figure 6.

The MPC must meet the isolation requirement and the bidirectional capabilities needed in vehicle applications. So, from the isolation requirement point of view the MPC topologies can be classified in three main classes:

• › Non-isolated (Figure 7);
• › Partially isolated (Figure 8);
• › Fully isolated (Figure 9);

The main research directions for the MPC are:

• › Reducing the number of active switches and passive components;
• › Finding a simple transformer windings connection that assures bidirectional power transfer between each pair of ports;
• › Simple control of the bidirectional power flows;
• › Designing of compact interleaved topologies for high-power applications without the use of bulky inductors or transformers to mitigate the current ripple;
• › Reducing switching loss and stresses based on zero-voltage and zero-current switching command techniques;
• › Reducing conduction loss due to circulation current;
• › Obtaining a higher power density based on a symmetrical topology that is favorable to achieving magnetic integration and modularizing;
• › Reducing the DC offset of the transformers caused by unbalance on the voltages or the switching times.

7. PFCV WITH SERIES AND PARALLEL MPC TOPOLOGIES

The purpose of this chapter is to demonstrate that the EMU strategies could improve the energy efficiency of the whole MPC for the both series and parallel MPC topologies, named as MPCs and MPCp, respectively.

Figure 6. Basic bidirectional switching cells used in the MPC.

The FC system powers the DC bus (HV type) and the ESS bus (LV type), which is used in the parallel and series MPC topologies, respectively, via a DC-DC power converter that will boost the FC voltage to the HV level needed on the DC bus [1, 89].

(a) Adapted from [64]. (b) Adapted from [65]. (c) Adapted from [66]. (d) Adapted from [67]. (e) Adapted from [68]. (f) Adapted from [69].

Figure 7. Non-isolated MPC topologies.

Figure 8. (Continued)

(a) Adapted from [70]. (b) Adapted from [71]. (c) Adapted from [72]. (d) Adapted from [73]. (e) Adapted from [74]. (f) Adapted from [75].

Figure 8. Partially isolated MPC topologies.

Figure 9. (Continued)

(a) Adapted from [76]. (b) Adapted from [77]. (c) Adapted from [78]. (d) Adapted from [79]. (e) Adapted from [80]. (f) Adapted from [81]. (g) Adapted from [82]. (h) Adapted from [83]. (i) Adapted from [84]. (j) Adapted from [85]. (k) Adapted from [86]. (l) Adapted from [87, 88].

Figure 9. Fully isolated MPC topologies.

The FC system usually operates in a range between the MAE and MPP for both high efficiency and reliability. So, a PI is necessary and this must be appropriately controlled [20, 90]. The ESS exchanges power with the HV DC bus via a DC-DC bidirectional power converter, named PI2 in the both MPC topologies. Thus, the PI2 is necessary in the parallel and series MPC topology to assure the power flow balance on the DC bus (if it is properly controlled) and to boost the LV of the EES bus to the HV DC bus, respectively. The inverter system for the traction motor is usually of bidirectional type and the used topologies are of classical or of Z-source inverter type in the both MPC topologies [1, 34]. The competitive types of electric traction motors include the induction motor (IM), the switched reluctance motor (SRM), the permanent magnet (PM) Synchronous Motor (SM) motor (PMSM), and the Brushless DC motor (BLDC) [1, 13].

These motors have many advantages (such as efficiency, cost, size and weight, life span and maintainability) if compared with other motor types.

The EMU controls the power flow in different operating modes (CS, CD, and CI) of the ESS, monitoring its SOC. The energy efficiency depends by the charge/discharge ratio of the ESS power (SOC level), energy efficiency coefficient of each PIs, regenerative braking factor, and the drive cycle. For example, the EMU must to select the CD and CI mode when the PFCV runs up and down a hill, but all CD, CI or CS modes could be used in efficient manner until the PFCV will be plugged into a PCS. The most strategies propose to use the stored electrical energy during the initial CD operation up to exhausting it, and then will switch to CS mode and later to CI mode if a PCS on route has announced their availability to sell power back to the grid. The best energy strategy will be shown in next sections for main MPC topologies.

The on-board ESS allows energy storage from FC system and regenerative load and supply the motor during the vehicle acceleration, which means a high charge/discharge capability for the used batteries.

A wide variety of battery technologies have claimed to be "the best choice" for automotive applications, starting from standard flooded and enhanced flooded to next AGM and Li-ion polymer batteries [91-93]. Hybridization of the ESS for pulsed load is made using UCs and batteries in different ESS topologies (passive, semi active and active) [94].

This section summarized few EMU strategies to assure the grid stabilization and cover the distance over which the vehicle will likely operate before a connection to the near PCS.

The next sections will evaluate the energy efficiency of the whole MPC for different ESS operating modes. The obtained results are a generalization of those shown in [50], where only the CS mode is considered.

8. PARALLEL MPC TOPOLOGY

The parallel MPC (MPCp) topology is shown in Figure 10 without a plug-in feature. The whole MPC has as inputs the FC system and the ESS bus, and as output the AC electrical machine. The average power of the FC system and ESS on a load cycle is PFC and PESS, respectively. The PFC is always positive, but the sign of the PESS depends on the SOC regulation performed by the EMU.

The MPCp topology shown in Figure 10 contains the following PIs:

• PI1 unidirectional DC-DC power converter, having the energy efficiency coefficient denoted by 1.
• PI2 bidirectional DC-DC power converter, having in direct and reverse mode the 2and2^R coefficients, respectively.
• PI5 bidirectional inverter system, having in direct and reverse mode the ⁵ and 5^R coefficients, respectively.

The inverter system powers the AC electrical machine with P^L from the DC bus via PI5. During regenerative braking process the reverse power flow, kLPL, charges the ESS via the PI5 and PI2 connected in series. The net power delivery to the AC electrical machine is *(*1-kL)PL, where k^L is the regenerative braking factor.

8.1. Equivalent Parallel MPC Topology

In terms of the power flow, an equivalent MPC topology of the MPCp topology is shown in Figure 11.

The power flow will circulate in the motor regime from both FC and ES system to the AC motor via the series connection of the PI1 and PI5, and the PI2 and PI5, respectively.

Figure 10. Parallel MPC topology [14].

Figure 11. Flowchart of power flow of the parallel MPC topology [15].

The FC system powers the load with PFC(L) , and the rest of the needed power on the DC bus is obtained from the ESS that is discharged by the average power, PESS(D) , via the PI2. The power flow balance on DC bus in the CD mode is:

$$\eta2 P{ESS(D)} + \eta1 P{FC(L)} = P_L / \eta_5$$ (1)

where the used notations are mentioned above.

The reverse power flow during regenerative braking process, kLPL, will circulate from the AC electrical machine to the ESS via the PI5 and PI2 connected in series (that is formally represented by PI4 in Figure 11). The FC system also powers with PFC(ESS) the ESS via PI1 and PI2 connected in series (that is formally represented by PI3 in Figure 11).

The power flow balance on the ESS bus related to the CI mode of the ESS (which is charged in average with PESS(C)) is:

$$P_{ESS(C)} = \eta3 P{FC(ESS)} + \eta_4 k_L P_L = \eta1 \eta{2R} P{FC(ESS)} + \eta{2R} \eta_{5R} k_L P_L$$ (2)

where the used notations are mentioned above.

The equivalent MPCp topology proposed in Figure 11 contains additional PIs:

• PI3 unidirectional DC-DC power converter, having the energy efficiency coefficient denoted by 3*=*12R.
• PI4 unidirectional AC-DC power converter, having 4*=*5*R*2R. The plug-in feature of PI 4 is not shown in Figure 11.

This equivalent representation of the MPCp is more clear and convenient for an analysis of energy efficiency performed in this section.

Also, it can be noted that this topology is obviously better than the MPCp topology because the power is less processed and easier to be controlled. The efficiency analysis will be performed in the next section considering as parameters the kESS and kCD ratios, where kESS(p)=PESS/P^L and kCD=PESS(C) /PESS(D) . As it was noted above, PESS(C) and PESS(C) represent the average power exchanged by the ESS on CI and CD mode, respectively.

8.2. Energy Efficiency of Parallel MPC Topology

Note that the energy efficiency coefficients are defined as average value because their values vary depending on the operating conditions.

The energy efficiency for the MPCp topology is given by the relationship:

$$\etap = \eta{MPCp} = (1 - k_L) PL / (P{FC} + P_{ESS})$$ (3)

It is obvious that:

$$P{FC}=P{FC(L)}+P_{FC(ESS)}$$

$$\tag{4}$$

$$P{ESS} = P{ESS(D)} - P_{ESS(C)} \tag{5}$$

Relationships (6) result by simple mathematical manipulations of (5) using the definitions for the kESS(p) and kCD ratios:

$$P{ESS(C)} = k{ESS(p)} k_{CD} PL / (1 - k{CD})$$ (6a)

$$P{ESS(D)} = k{ESS(p)} PL / (1 - k{CD})$$ (6b)

If relationships (6) will be used on (1) and (2), then relationship (4) becomes:

$$P_{FC} = P_L \cdot \left( 1/\eta5 - \eta{5R} kL - k{ESS(p)} \eta_{2m} \right) / \eta_1 \tag{7}$$

where

$$\eta_{2m} = (\eta2 \eta{2R} - k{CD}) / [\eta{2R} (1 - k{CD})], k{CD} \neq 1$$ (8)

It is obvious that:

$$k{ESS(p)} \stackrel{\leq}{=} 0 \Leftrightarrow P{ESS(D)} \stackrel{\leq}{>} P{ESS(C)} \Leftrightarrow k{CD} \stackrel{\geq}{=} 1$$ (9)

where the symbol used for comparison can be <, =, or >.

The case kCD=1 (kESS(p)*=*0) will be briefly discussed in section 11 of this chapter because it has been widely analyzed in [50]. Considering *kCD*1, the energy efficiency for the MPCp topology is finally given by relation:

$$\eta{p} = (1 - k{L})\eta{1} / [1/\eta{5} - \eta{5R}k{L} - k{ESS(p)}(\eta{2m} - \eta_{1})]$$ (10)

9. SERIES MPC TOPOLOGY

The series MPC (MPCs) topology is shown in Figure 12. The MPCs topology contains the same PIs, marked as it was mentioned in the MPCp topology. If the ESS bus is of HV type (not of LV type), then PI2 will not be used, and their energy efficiency coefficients will be considered unitary in all relations written in this section.

Here, for the simplicity of the analytical comparison performed, PI2 is considered having the same values of the energy efficiency coefficients (as PI2 from the MPCp topology), even if the design parameters are different. The same notations as above, PFC and PESS, will be used for the average power on a load cycle of the FC system and ESS, respectively.

Also, the net power delivery to the AC electrical machine is (1-kL)PL.

Figure 12. Series MPC topology [15].

9.1. Equivalent Series MPC Topology

In terms of the power flow, an equivalent MPC topology to the MPCs topology is shown in Figure 13. The inverter system powers the AC electrical machine with the P^L power from the the ESS bus via the PI2 (2) and PI5 (5) connected in series. Also, during regenerative braking process the reverse power flow, kLPL, charges the ESS via the PI5 (5R) and PI2 (2R) connected in series.

So, PI3 will represent the series connection of the PI2 and PI5, and PI4 will represent the series connection of PI2 and PI5 that operate in direct and reverse mode, and have the 3=^{2 5} and 4*=*2*R5^R* coefficient, respectively. The ESS bus is powered by the ESS and the FC, directly and via PI1, respectively. Using the ratio of the ESS power exchanged on the ESS bus, kESS(s)=PESS/PL, the power flow balance on the ESS bus is:

$$\eta1 P{FC} + P{ESS} + \eta{2R} \eta_{5R} k_L P_L = P_L / \eta_2 \eta_5$$ (11)

The same relationship is obtained if the power flow balance on the ESS bus is written in the CI and CD mode of the ESS, respectively:

$$P_{ESS(C)} = \eta1 P{FC(ESS)} + \eta{2R} \eta{5R} k_L P_L \tag{12a}$$

$$P_{ESS(D)} + \eta1 P{FC(L)} = P_L / \eta_2 \eta_5 \tag{12b}$$

where the notations for the split power flows are the same.

The split power flows were not shown in the diagram for the clarity of the equivalent MPCs topology. Considering (4) and (5), the same relationship as in (11) will be obtained by subtracting (12a) from (12b).

9.2. Energy Efficiency of Series MPC Topology

In this case, the energy efficiency for the MPCs topology is given by (13):

$$\etas = \eta{MPCs} = (1 - k_L)PL/(P{FC} + P_{ESS})$$ (13)

Figure 13. Flowchart of power flow of the series MPC topology [15].

Using the kESS(s) ratio, by simple mathematical manipulations of (11), results:

$$\eta{s} = (1 - k{L})\eta{1} / [1/(\eta{2}\eta{5}) - \eta{2R}\eta{5R}k{L} - k{ESS(s)}(1 - \eta{1})]$$ (14)

The case kESS(s) = 0 has been widely analyzed in [50], therefore will be briefly discussed in section 11 of this chapter.

10. ENERGY EFFICIENCY OF MPCP VS. MPCS

If both MPC topologies (MPCp and MPCs) exchange the same ESS power (positive or negative), i.e. kESS(p)=kESS(s)=kESS, it is also important to know which ones operates more efficiently:

$$\eta{p} \stackrel{<}{=} \eta{s} \tag{15}$$

10.1. The Results of Analytical Calculations

Relationship (15) is equivalent with (16) by simple mathematical manipulations of its, using (10) and (14):

$$k{ESS} \cdot (1 - \eta{2m}) = \left[ 1 - \eta_2 + k_L \eta2 (1 - \eta{2R}) \eta5 \eta{5R} \right] / (\eta_2 \eta_5)$$ (16)

Using (8) will result (17):

$$1 - \eta{2m} = [\eta{2R}(1 - \eta2) + k{CD}(1 - \eta{2R})] / [\eta{2R}(1 - k{CD})], k{CD} \neq 1$$ (17)

So, (16) will become equivalent with (18):

$$\frac{k{ESS}}{1 - k{CD}} \stackrel{\leq}{=} \frac{\eta_{2R}}{\eta_2 \eta_5} \cdot \frac{1 - \eta_2 + k_L \eta2 (1 - \eta{2R}) \eta5 \eta{5R}}{\eta_{2R} (1 - \eta2) + k{CD} (1 - \eta{2R})}, k{CD} \neq 1$$ (18)

For an easy interpretation of the kESS values on the CI and CD mode, the following notation is made related to the CS mode:

$$\frac{C{\eta CD}}{1 - k{CD}} \stackrel{\Delta}{=} \frac{\eta_{2R}}{\eta_2 \eta_5} \cdot \frac{1 - \eta_2 + k_L \eta2 (1 - \eta{2R}) \eta5 \eta{5R}}{\eta_{2R} (1 - \eta2) + k{CD} (1 - \eta{2R})} > 0, k{CD} \neq 1$$ (19)

Finally, relationship (15) will become equivalent with (20):

$$\frac{k{ESS}}{1 - k{CD}} \stackrel{<}{>} \frac{C{\eta CD}}{1 - k{CD}}, k_{CD} \neq 1$$ (20)

Analyzing (20) and (9), some final conclusions related to the MPC energy efficiency on both CD and CI operating modes are shown in (21a) and (21b), respectively, considering kESS(p)=kESS(s)=kESS:

$$\begin{split} k{ESS} &> 0 \; (\Leftrightarrow k{CD} < 1) \Rightarrow k{ESS} \stackrel{<}{=} C{\eta CD} \; \Leftrightarrow \ \Leftrightarrow \eta_p \stackrel{>}{=} \etas \Rightarrow \begin{cases} 0 < k{ESS} < C_{\eta CD} \; \Leftrightarrow \eta_p > \etas \ k{ESS} > C_{\eta CD} > 0 \Leftrightarrow \eta_p < \eta_s \end{cases} \end{aligned} \tag{21a}$$

$$\begin{aligned} k{ESS} &< 0 \ (\Leftrightarrow k{CD} > 1) \Rightarrow k{ESS} \stackrel{>}{<} C{\eta CD} \Leftrightarrow \ \Leftrightarrow \eta_p \stackrel{>}{<} \etaz \Rightarrow \begin{cases} 0 > k{ESS} > C_{\eta CD} \Leftrightarrow \eta_p > \etaz \ k{ESS} < C_{\eta CD} < 0 \Leftrightarrow \eta_p < \eta_z \end{cases} \end{aligned} \tag{21b}$$

Relationships (21) can be rewritten in a concise equivalent form as:

$$(|k{ESS}| < |C{\eta CD}| \Leftrightarrow \eta_p > \etas) \lor (|k{ESS}| > |C_{\eta CD}| \Leftrightarrow \eta_p < \eta_s)$$ (22)

where the symbol is used for the logical OR operator.

Of course, if kESS(p)=kESS(s)=CCD, then both MPC topologies considered are equivalent in terms of energy efficiency, so CCD is a border value. If kESS(p) kESS(s) , then p/^s ratio of energy efficiency coefficients is computed in (23):

$$\frac{\eta_p}{\eta_s} = \frac{1/(\eta_2 \etas) - \eta{2R} \eta_{5R} kL - k{ESS(s)} (1 - \eta_1)}{1/\etas - \eta{5R} kL - k{ESS(p)} (\eta_{2m} - \eta_1)}$$ (23)

Obviously, if considering kESS(p)=kESS(s)=kESS in (23), the final conclusion (22) can be obtained, too. The analytical results are validated by the simulation performed. Some relevant results will be shown in next section.

10.2. Simulation Results

If it is not specifically mentioned, the energy efficiency coefficients of PIs are set, in all simulations performed, to 1=0.9, 2=0.9, 2R=0.9, 5=0.9, and 5R=0.9. The values used for the charge-discharge parameter, kCD, are 1/1.25 and 1.25 for the CD and CI operating mode, respectively.

Figure 14. ^p /^s (kESS, kL) surface [15].

The energy efficiency ratio for the MPCs topologies is shown in Figure 14 based on (23), considering kESS(p)=kESS(s)=kESS. Using kCD=1/1.25 and kCD=1.25 in regions where kESS>0 and kESS<0, respectively.

It can be noted that: (1) the unitary plane delimits the border lines where the MPCs and MPCp topologies have the same energy efficiency; (2) if the ECU manages a low level of the ESS power exchanged on a load cycle, then the MPCp is more energy efficiency than the MPCs topology; (3) if the ESS accepts a large SOC window, then the ECU can manage a higher value of the ESS power exchanged on a load cycle and thus the MPCs is more energy efficiency than the MPCp topology; (4) the border lines (values of the CCD coefficient) are less dependent by the regenerative braking factor [15].

The dependence of the CCD coefficient to k^L parameter can be obtained based on (19) computed for different values of the kCD parameter. The regions where p>^s can be extended for values of the kCD parameter that are far from unity value. A kCD value greater or less than unity means that the ESS will be charged (PESS(C)>PESS(D)) or discharged (PESS(C)<PESS(D)), respectively, but at higher values for the split powers, PESS(C) and PESS(D) . For a PFCV under a load cycle it is obvious that these high values for average power can produce unacceptable peak values for the ESS. Consequently, in this case the ESS hybridization must be more carefully designed. Note that each of the MPC topologies (MPCp or MPCs) can operate efficiently in both operating modes (CD or CI) (Figure 14).

Also, the CCD coefficient is dependent on energy efficiency coefficients. It was shown that CCD coefficient is less sensitive to energy the efficiency coefficients of PI5 (⁵ and 5R) [15]. This aspect is explained trough topological considerations: PI5 has the same position (between the DC bus and load bus) into both MPC topologies. As it was explained above, the bidirectional PI5 can exchange power between the load port and the DC bus.

This dependence related to an energy efficiency coefficient () can be obtained by computing the CCD/ coefficient sensitivity using (19). Also, relationship (19) clearly shown that the CCD coefficient is independent to the PI1 energy efficiency coefficient. This is explained by the same position of PI1 on both MPC topologies and the fact that the FC system powers with the same power (PFC) the load and ESS.

Figure 15. *kESSs*vs *kESSp*for 1=2=2R=5=5R=0.9 and different k^L values [15].

It is obvious that both ^p and *s*coefficients increase if each of energy efficiency coefficients increases. The exact variation of the p/^s ratio is given by (23). A mathematical analysis might reveal all dependencies [15], but this is not the main goal of this chapter. Only a brief validation of the analytical results based on topological analysis is given below.

The load port is powered in both topologies with the same power flow from FC and ESS via PI1, PI2, and PI5, but only the position of the PI2 is different on the MPC topologies.

If the PI2 is an ideal converter (2*=*2R=1), then MPCp and MPCs topologies are identical, so p/s=1 must be obtained. Therefore, if 2*=*2R=1 is considered in (8), then 2*m*=1, and thus from (23) p/s=1 is obtained (q.e.d.).

It is obvious from (22) that the energy efficiencies of the MPCp and MPCs topologies have different values, ^p and ^s , respectively, for a value of common kESS parameter, which is different by the CCD value. On the other hand, if different kESS values are used in the operation of the MPCp and MPCs topologies (kESSp and kESSs, respectively), then both topologies could have the same efficiency on the conditions given by the relationship (24):

$$\eta{p} = \eta{s} \Rightarrow k{ESSs} = \frac{\eta{2m} - \eta{1}}{1 - \eta{1}} k{ESSp} + \frac{1 - \eta{2}}{\eta{2} \eta{5} (1 - \eta{1})} + \frac{\eta{2} (1 - \eta{2R})}{1 - \eta{1}} k_{L}$$ (24)

Relationship (24) is shown in Figure 15 for different k^L values and1*=2=*2*R=5=* 5R=0.9. The same efficiency could be obtained for both MPC topologies if the ESS operates in the CD or CI mode at different kESS values. If the both MPC topologies operate in the CS mode, then the energy efficiency analysis will be performed in the next section based on (10), (14) and (22), considering kESS(p)=kESS(s)= 0.

11. ENERGY EFFICIENCY OF MPCP AND MPCS TOPOLOGIES IN CS MODE

The average value of charge exchanged by the ESS under a drive (load) cycle is zero in a CS mode, so:

$$k{ESS} = 0 \Rightarrow P{ESS0(D)} = P_{ESS0(C)} \tag{25}$$

The energy efficient coefficients given by (10) and (14) will become as below:

$$\eta{p0} = \eta{p} \Big|{k{\text{EXE}(p)} = 0} = \frac{(1 - k{L})\eta{1}}{1/\eta{5} - \eta{5R}k_{L}}$$ (10a)

$$\eta_{z0} = \etaz \Big|{k_{\text{ESS}(z)} = 0} = \frac{(1 - k_L)\eta_1}{1/(\eta_2 \eta5) - \eta{2R} \eta_{5R} k_L}$$ (14a)

where k^L is the regenerative braking factor.

The kL-ESS⁰ coefficient is defined as a measure of load dynamic that requests an appropriate ESS dynamic:

$$k{L-ESS0} \stackrel{\Delta}{=} P{ESS0(D)} / P_L \tag{26}$$

If both MPC topologies (MPCp and MPCs) operate in a CS mode (i.e. kESS(p)=kESS(s)=0), it is also important to know which ones operates more efficiently (e.g. Figure 16). Relationship (26) is equivalent with (27) by simple mathematical manipulations of its, considering (10a) and (14a):

$$k{L-ESSO} \le \frac{\eta{2R} \left[ 1 - \eta_2 + k_L \eta2 \eta{5R} (1 - \eta_{2R}) \right]}{\eta_2 \eta_5 (1 - \eta2 \eta{2R})} \stackrel{\Delta}{=} C_{\eta_0} > 0$$ (27)

where the C⁰ coefficient defined in (27) is a positive value, which depends on the energy efficiency coefficients of the PI2 and PI5, and the regenerative braking factor, kL.

Figure 16. ^p /^s (kL, kL⁰ ) surface for 1=2=2R=0.93 and 5=5R=1 [15].

The energy coefficients of PI5 can be set to 1 (5*=*5R=1) for an easy comparison of these particular results obtained in the CS mode with same results that have been presented in 50.

An example of a load cycle profile applied on the DC bus is shown in Figure 17, too [94]. The average (AV) and effective (RMS) value of the load power are dependent on the duty cycle, DL, the ratio of positive and negative amplitudes, CL=PL(max)/PL(min), and the regenerative braking factor, kL, as in (28):

$$P_{L(AV)} = P_L k_L(D_L CL - 1), P{L(RMS)} = P_L k_L \sqrt{D_L C_L^2 + 1 - D_L}$$ (28)

The results presented in Figure 17 show that the MPCp topology may or may not be more efficient than the MPCs topology, and this depends on the values set for the D^L and kL-ESS0 parameters. Further details about MPCp and MPCs topologies operating on CS mode can be found in [50].

CONCLUSION

In this chapter, besides a brief overview over the EV and HEV issues, a systemic analysis of the MPC energy efficiency was presented for the series and parallel MPC topologies used in PFCVs. The analysis has been focused on the FC/ESS HPS that supply the electrical motors via a MPC and via inverter systems. The power flow flowchart of the parallel MPC topology shown in [14] is similar to the MPC topology proposed in [15], to overcome the limitations of both series and parallel hybrid MPC topologies for uncertain load profiles. The main properties of the MPC topologies were underlined in this chapter.

The efficiency of MPC's topologies is evaluated and compared in CS, CD and CI operating modes using different control's factors.

Figure 17. ^s (DL) - line width set to 2, and ^p (DL) - line width set to 1, for 1=2=2R=0.93, 5=5R=1, CL=9 and different kL⁰ [15].

Therefore, the ESS exchanges the power under the EMU control and the PFCV must be connected to the PCS when it is the case. The EMU sets these moments and notifies the driver in due time. It was shown that each MPC topology can operate efficiently under certain conditions, which have been specified.

The impact of the PFCVs on the electricity sector depends on the (1) magnitude of the market penetration, (2) timing of ESS charging/discharging, (3) energy management, (4) structure of the power sector, and (5) availability of the FC systems for automotive vehicle. PFCVs can contribute by an efficient use of the renewable intermittent potential and can potentially solve imbalances between electricity supply and demand.

Besides these conclusions, a number of questions also arise for further exploration: (1) What type of energy management strategy is more suitable for the PFCV EMU in relation with upcoming smart grids? (2) What are the optimum ESS requirements for various PFCV configurations? (3) Can be quantified the impact of plug-in feature on battery life? (4) Can be quantified the PFCV impact on renewable intermittent potential? (5) Can be evaluated the PFCV consumption characteristics under a given EMU strategy? The CI mode will be used in relation to the upcoming smart grids to sell the power back to the grid in order to stabilize the grid. Overall, the potential to increase the MPC efficiency is shown for all CS, CD and CI operating modes of the ESS. The use of multi-operating modes makes the PFCV attractive, but also makes the estimation of its in-use fuel economy a challenging design action. If it is not necessary to operate the ESS in the CD and CI modes, then the CS mode will be used, but designing of appropriate EMU strategy remains a challenging action in order to approximate real-world operation.

ACKNOWLEDGMENTS

Some figures, tables and text are reproduced from [14, 15] here with kind permission from Elsevier Limited, UK [November 18, 2013].

REFERENCES

• [1] Mehrdad, E., Yimin, G., Ali, E., Modern Electric: Hybrid Electric, and Fuel Cell Vehicles. 2nd Ed., CRC Press, New York, 2010.
• [2] Bizon, N., Advances in Energy Research: Distributed Generation Systems Integrating Renewable Energy Resources. Nova Science Publishers Inc., US, 2012.
• [3] Bizon, N., Tabatabaei, N. M., Advances in Energy Research: Energy and Power Engineering. Nova Science Publishers Inc., US, 2013.
• [4] Zhan, Y., Guo, Y., Zhu, J., Wang, H., Intelligent Uninterruptible Power Supply System with Back-up Fuel Cell/Battery Hybrid Power Source. Journal of Power Sources, vol. 179, no. 2, pp. 745-753, 2008.

• [5] Metaxiotis, K., Intelligent Information Systems and Knowledge Management for Energy: Applications for Decision Support, Usage and Environmental Protection. IGI Global, US, Chapter 2 (Bizon, N., Intelligent Control of the Energy Generation System), pp. 40-96, 2009.

• [6] Jiang, W., Fahimi, B., Multiport Power Electronic Interface: Concept, Modeling, and Design. IEEE Trans. on Power Electron, vol. 26, no. 7, pp. 1890-1900, 2011.

• [7] McDonough, M., Shamsi, P., Fahimi, B., Application of Multi-Port Power Electronic Interface for Contactless Transfer of Energy in Automotive Applications. Vehicle Power Propulsion Conference, pp. 936-940, 2009.

• [8] Geidl, M., Andersson, G., Optimal Power Flow of Multiple Energy Carriers. IEEE Trans. on Power Electronics, vol. 22, no. 1, pp. 145-155, 2009.

• [9] Geidl, M., Koeppel, G., FavrePerrod, P., Kloeckl, B., Andersson, G., Frohlich, K., Energy Hubs for the Future. IEEE Power Energy Mag., vol. 5, no. 1, pp. 24-30, 2007.

• [10] Krause, T., Andersson, G., Frohlich, K., Vaccaro, A., Multiple-Energy Carriers: Modeling of Production, Delivery, and Consumption. Proceedings of the IEEE, vol. 99, no. 1, pp. 15-27, 2011.

• [11] Del Real, A. J., Arce, A., Bordons, C., Optimization Strategy for Element Sizing in Hybrid Power Systems. Journal of Power Sources, vol. 193, no. 1, pp. 315-321, 2009.

• [12] Loo, K. H., Zhu, G. R., Lai, Y. M., Tse, C. K., Development of a Maximum-Power-Point Tracking Algorithm for Direct Methanol Fuel Cell and its Realization in a Fuel Cell/Supercapacitor Hybrid Energy System. 8 th International Conferenceon Power Electronics, pp. 1753-1760, 2011.

• [13] Chan, C. C., Chau, K. T., Modern Electric Vehicle Technology. Oxford University Press, 2001.

• [14] Bizon, N., Energy Efficiency of Multiport Power Converters Used in Plug-In/V2G Fuel Cell Vehicles. Applied Energy, vol. 96, pp. 431-443, 2012.

• [15] Bizon, N., Energy Efficiency for the Multiport Power Converters Architectures of Series and Parallel Hybrid Power Source Type Used in Plug-In/V2G fuel Cell Vehicles. Applied Energy, vol. 102, pp. 726-734, 2013.

• [16] Erdinc, O., Uzunoglu, M., Recent Trends in PEM Fuel Cell-Powered Hybrid Systems: Investigation of Application Areas, Design Architectures and Energy Management Approaches. Renewable and Sustainable Energy Reviews, vol. 14, pp. 2874-2884, 2010.

• [17] Gao, D., Jin, Z., Lu, Q., Energy Management Strategy Based on Fuzzy Logic for a Fuel Cell Hybrid Bus. Journal Power Sources, vol. 185, no. 1, pp. 311-317, 2008.

• [18] Erdinc, O., Vural, B., Uzunoglu, M., A Wavelet-Fuzzy Logic Based Energy Management Strategy for a Fuel Cell/Battery/Ultracapacitor Hybrid Vehicular Power System. Journal of Power Sources, vol. 194, no. 1, pp. 369-380, 2009.

• [19] Rodatz, P., Paganelli, G., Sciarretta, A., Guzzella, L., Optimal Power Management of an Experimental Fuel Cell/Supercapacitor-Powered Hybrid Vehicle. Control Eng. Pract., vol. 13, no. 1, pp. 41-53, 2005.

• [20] Feroldi, D., Serra, M., Riera, J., Energy Management Strategies Based on Efficiency Map for Fuel Cell Hybrid Vehicles. Journal of Power Sources, vol. 190, no. 2, pp. 387- 401, 2009.

• [21] Erdinc, O., Vural, B., Uzunoglu, M., Ates, Y., Modeling and Analysis of an FC/UC Hybrid Vehicular Power System Using a Wavelet-Fuzzy Logic Based Load Sharing and Control Algorithm. International Journal of Hydrogen Energy, vol. 34, no. 12, pp. 5223-5233, 2009.

• [22] Liu, G., Zhanga, J., Sun, Y., High Frequency Decoupling Strategy for the PEM Fuel Cell Hybrid System. International Journal of Hydrogen Energy, vol. 33, no. 21, pp. 6253-6261, 2008.

• [23] Thounthong, P., Rael, S., Davat, B., Energy Management of Fuel Cell/Battery/ Supercapacitor Hybrid Power Source for Vehicle Applications. Journal of Power Sources, vol. 193, no. 1, pp. 376-385, 2009.

• [24] Wang, Y., Chen, K. S., Mishler, J., Cho, S. C., Adroher, X. C., A Review of Polymer Electrolyte Membrane Fuel Cells: Technology, Applications, and Needs on Fundamental Research. Appl. Energy, vol. 88, no. 4, pp. 981-1007, 2011.

• [25] Doucette, R. T., McCulloch, M. D., Modeling the Prospects of Plug-In Hybrid Electric Vehicles to Reduce CO² Emissions. Applied Energy, vol. 88, no. 7, pp. 2315-2323, 2011.

• [26] Hacker, F., Harthan, R., Matthes, F., Zimmer, W., Environmental Impacts and Impact on the Electricity Market of a Large Scale Introduction of Electric Cars in Europe - Critical Review of Literature. ETC/ACC Technical Paper, vol. 4, pp. 56-90, 2009.

• [27] www.pluginamerica.org/vehicles.

• [28] http://evworld.com.

• [29] Bizon, N., A New Topology of Fuel Cell Hybrid Power Source for Efficient Operation and High Reliability. Journal of Power Sources, vol. 196, no. 6, pp. 3260-3270, 2011.

• [30] Qian, Z., Al-Atrash, H., Modeling and Control of Three-Port DC/DC Converter Interface for Satellite Applications. IEEE Trans. on Power Electronics, vol. 25, no. 3, pp. 637-649, 2010.

• [31] Singh, B., Singh, S., State of the Art on Permanent Magnet Brushless DC Motor Drives. Journal of Power Electronics, vol. 9, no. 1, pp. 1-17, 2009.

• [32] Zeraoulia, M., Benbouzid, M. E. H., Diallo, D., Electric Motor Drive Selection Issues for HEV Propulsion Systems: A Comparative Study. IEEE Trans. on Vehicular Technology, vol. 55, no. 6, pp. 1756-1764, 2006.

• [33] Shen, M., Peng, F. Z., Converter Systems for Hybrid Electric Vehicles. Electrical Machines System Conference, pp. 2004-2010, 2007.

• [34] Peng, F. Z., Yuvan, X., Fang, X., Qian, Z., Z-Source Inverter for Motor Drives. IEEE Trans. on Power Electronics, vol. 20, no. 4, pp. 857-63, 2005.

• [35] Thangaprakashy, S., Krishnan, A., Current Mode Integrated Control Technique for Z-Source Inverter Fed Induction Motor Drives. Journal of Power Electronics, vol. 10, no. 3, pp. 285-292, 2010.

• [36] Haiping, X., Peng, F. Z., Chen, L., Wen, X., Analysis and Design of Bidirectional Z-Source Inverter for Electrical Vehicles. APEC Conference, pp. 1252-1257, 2008.

• [37] Dehghan, S. M., Mohamadian, M., Yazdian, A., Hybrid Electric Vehicle Based on Bidirectional Z-Source Nine-Switch Inverter. IEEE Trans. on Vehicular Technoloy, vol. 59, no. 6, pp. 2641-2653, 2010.

• [38] Yamanaka, M., Koizum, H., A Bidirectional Z-Source Inverter for Electric Vehicles. Power Electronic Drive System Conference, pp. 574-578, 2009.

• [39] Ding, X., Qian, Z., Yang, S., Peng, F. Z., A New Feedforward Compensation to Reject DC-Link Voltage Ripple in Bidirectional Z-Source Inverter ASD System. APEC Conference, pp. 1809-1813, 2008.

• [40] Fontes, G., Turpin, C., Astier, S., Meynard, T. A., Interactions between Fuel Cells and Power Converters: Influence of Current Harmonics on a Fuel Cell Stack. IEEE Trans. on Power Electronics, vol. 22, no. 2, pp. 670-678, 2007.

• [41] Thounthong, P., Davat, B., Rael, S., Sethakul, P., An Overview of Power Converters for a Clean Energy Conversion Technology. IEEE Industrial Electronics Magaz., vol. 3, no. 1, pp. 32-46, 2009.

• [42] Wang, Y., Chen, K. S., Mishler, J., Cho, S. C., Adroher, X. C., A Review of Polymer Electrolyte Membrane Fuel Cells: Technology, Applications, and Needs on Fundamental Research. Applied Energy, vol. 88, no. 4, pp. 981-1007.

• [43] Bizon, N., Nonlinear Control of Fuel Cell Hybrid Power Sources: Part I Voltage Control. Applied Energy, vol. 88, no. 7, pp. 2559-2574, 2011.

• [44] Duleep, G., Van Essen, H., Kampman, B., Grunig, M., Impacts of Electric Vehicles Assessment of Electric Vehicle and Battery Technology. Delft, CE Delft, vol. 2, pp. 9- 29, 2011.

• [45] Duduta, M., Ho, B., Wood, V. C., Limthongkul, P., Brunini, V. E., Carter, W. C., Chiang, Y. M., Semi-Solid Lithium Rechargeable Flow Battery. Advanced Energy Materials, vol. XX, pp. 1-6, 2011.

• [46] Gruning, M., Witte, M., Boteler, B., Kantamaneni, R., Gabel, E., Bennink, D., et al.; Impacts of Electric Vehicles - Assessment of the Future Electricity Sector. Delft, CE Delft, vol. 3, pp. 35-70, 2011.

• [47] Tang, Y., Yuan, W., Pan, M., Li, Z., Chen, G., Li, Y., Experimental Investigation of Dynamic Performance and Transient Responses of a KW-Class PEM Fuel Cell Stack under Various Load Changes. Applied Energy, vol. 87, no. 4, pp. 1410-1417, 2010.

• [48] Bayindir, K. C., Gozukucuk, M. A., Teke, A., A Comprehensive Overview of Hybrid Electric Vehicle: Powertrain Configurations, Powertrain Control Techniques and Electronic Control Units. Energy Conversion and Management, vol. 52, no. 2, pp. 1305-1313, 2011.

• [49] Wagner, G. O., Real-Time Diagnosis and Maintenance for Automotive System. Journal of Electronics and Computer Science, vol. 2, pp. 51-56, 2011, http://fecc_bulletin.upit. ro/PublicSite/Main.aspx?cid=14.

• [50] Ramos Paja, C. A., Romero, A., Giral, R., Calvente, J., Martinez Salamero, L., Mathematical Analysis of Hybrid Topologies Efficiency for PEM Fuel Cell Power Systems Design. Electrical Power and Energy Systems, vol. 32, pp. 1049-1061, 2010.

• [51] Kuperman, A., Aharon, I., Battery Ultracapacitor Hybrids for Pulsed Current Loads: A Review. Renewable and Sustainable Energy Reviews, vol. 15, pp. 981-992, 2011.

• [52] Tang, Y., Juan, W., Pan, M., Wan, Z., Experimental Investigation on the Dynamic Performance of a Hybrid PEM Fuel Cell/Battery System for Lightweight Electric Vehicle Application. Applied Energy, vol. 88, no. 1, pp. 68-76, 2011.

• [53] Barelli, L., Bidini, G., Ottaviano, A., Dynamic Analysis of PEMFC-Based CHP Systems for Domestic Application. Applied Energy, vol. 91, no. 1, pp. 13-28, 2012.

• [54] Thounthong, P., Rael, S., Davat, B., Control Strategy of Fuel Cell/Supercapacitors Hybrid Power Sources for Electric Vehicle. Journal of Power Sources, vol. 158, no. 1, pp. 806-814, 2006.

• [55] Markel, T., Kuss, M., Denholm, P., Communication and Control of Electric Drive Vehicles Supporting Renewables. Vehicle Power Propulsion Conference, pp. 27-44, 2009, www.nrel.gov/vehiclesandfuels/vsa/pdfs/46224.pdf.

• [56] Zhao, C., Isolated Three-Port Bidirectional DC-DC Converter. Thesis, 2010, www.pes. ee.ethz.ch/uploads/tx_ethpublications/Zhao_no_CV.pdf.

• [57] Liu, Y., Chen, Y. M., A Systematic Approach to Synthesizing Multi-Input DC-DC Converters. IEEE Trans. on Power Electronics, vol. 24, no. 2, pp. 116-127, 2009.

• [58] Kwasinski, A., Identification of Feasible Topologies for Multiple-Input DC-DC Converters. IEEE Trans. on Power Electronics, vol. 24, no. 3, pp. 856-861, 2009.

• [59] Qian, Z., Abdel-Rahman, O., Al-Trash, H., Batarseh, I., Modeling and Control of Three-Port DC/DC Converter Interface for Satellite Applications. IEEE Trans. on Power Electronics, vol. 25, no. 3, pp. 637-649, 2010.

• [60] Bizon, N., Energy Harvesting from thePV Hybrid Power Source. Energy, vol. 52, pp. 297-307, 2013.

• [61] Bizon, N., FC Energy Harvesting Using the MPP Tracking Based on Advanced Extremum Seeking Control. International Journal of Hydrogen Energy, vol. 38, no. 14, pp. 1952-1966, 2013.

• [62] Jiang, W., Fahimi, B., Multi-Port Power Electric Interface for Renewable Energy Sources. IEEE Applied Power Electronics Conference, pp. 347-352, 2009.

• [63] Tao, H., Kotsopoulos, A., Duarte, J. L., Hendrix, M. A. M., Family of Multiport Bidirectional DC-DC Converters. IEEE Proceeding on Electric Power Applications, vol. 153, no. 3, pp. 451-455, 2006.

• [64] Caricchi, F., Crescimbini, F., Napoli, A. D., Honorati, O., Santini, E., Testing of a New DC/DC Converter Topology for Integrated Wind-Photovoltaic Generating Systems. Fifth European Conference on Power Electronics and Applications, vol. 8, pp. 83-88, 1993.

• [65] Napoli, A. D., Crescimbini, F., Solero, L., Caricchi, F., Capponi, F. G., Multiple-Input DC-DC Power Converter for Power-Flow Management in Hybrid Vehicles. Industry Applications Conference (37th IAS Annual Meeting), vol. 3, pp. 1578-1585, 2002.

• [66] Chen, Y. M. M., Liu, Y. C. C., Lin, S. H., Double-Input PWM DC/DC Converter for High/Low Voltage Sources. IEEE Transactions on Industrial Electronics, vol. 53, issue 5, pp. 1538-1545.

• [67] Marchesoni, M., Vacca, C., New DC-DC Converter for Energy Storage System Interfacing in Fuel Cell Hybrid Electric Vehicles. IEEE Transactions on Power Electronics, vol. 22, issue 1, pp. 301-308, 2007.

• [68] Rodriguez, F. D., Imes, W. G., Analysis and Modeling of a Two-Input DC/DC Converter with Two Controlled Variables and Four Switched Networks. IECEC Energy Conversion Engineering Conference, vol. 1, pp. 322-327, 1996.

• [69] Dobbs, B., Chapman, G., Lyle, P., A Multiple-Input DC-DC Converter Topology. IEEE Power Electronics Letters, vol. 1, issue 1, pp. 6-9, 2003.

• [70] Tao, H., Kotsopoulos, A., Duarte, J. L., Hendrix, M. A. M., Multi-Input Bidirectional DC-DC Converter Combining DC-Link and Magnetic-Coupling for Fuel Cell Systems. Industry Applications Conference, vol. 3, pp. 2021-2028, 2005.

• [71] Su, G. J., Peng, F. Z., A Low Cost, Triple-Voltage Bus DC-DC Converter for Automotive Applications. IEEE Appl. Power Electron. Conference, vol. 2, pp. 1015-1021, 2005.

• [72] Su, G. J., Cunningham, J. P., Tang, L., A Reduced-Part, Triple-Voltage DC-DC Converter for Electric Vehicle Power Management. IEEE Power Electronics Specialists Conference, pp. 1989-1994, 2007.

• [73] Al-Atrash, H., Tian, F., Batarseh, I., Tri-Modal Half-Bridge Converter Topology for Three-Port Interface. IEEE Trans. on Power Electronics, vol. 22, no. 1, pp. 341-345, 2007.

• [74] Al-Atrash, H., Pepper, M., Batarseh, I., A Zero-Voltage Switching Three-Port Isolated Full-Bridge Converter. IEEE International Telecommunication Energy Conference, pp. 1-8, 2006.

• [75] Al-Atrash, H., Batarseh, I., Boost-Integrated Phase-Shift Full-Bridge Converter for Three-Port Interface. IEEE Power Electronics Specialists Conference, pp. 2313-2321, 2007.

• [76] Matsuo, H., Lin, W., Kurokawa, F., Shigemizu, T., Watanabe, N., Characteristics of the Multiple-Input DC-DC Converter. IEEE Trans. on Industrial Electronics, vol. 51, no. 3, pp. 625-631, 2004.

• [77] Chen, Y. M., Liu, Y. C., Wu, F. Y., Multi-Input DC/DC Converter Based on the Multiwinding Transformer for Renewable Energy Applications. IEEE Trans. of Industry Applications, vol. 38, no. 4, pp. 1096-1104, 2002.

• [78] Krishnaswami, H., Mohan, N., Constant Switching Frequency Series Resonant Three-Port Bidirectional DC-DC Converter. IEEE Power Electronics Specialists Conference (PESC"08), pp. 1640-1645, 2008.

• [79] Liu, D., Li, H., A ZVS Bidirectional DC-DC Converter for Multiple Energy Storage Elements. IEEE Trans. on Power Electronics, vol. 21, no. 5, pp. 1513-1517, 2006.

• [80] Tao, H., Duarte, J., Hendrix, M., Three-Port Triple-Half-Bridge Bidirectional Converter with Zero-Voltage Switching. IEEE Trans. on Power Electronics, vol. 23, no. 2, pp. 782-792, 2008.

• [81] Zhao, C., Round, S. D., Kolar, J. W., An Isolated Three-Port Bidirectional DC-DC Converter with Decoupled Power Flow Management, IEEE Trans. on Power Electronics, vol. 23, no. 5, pp. 2443-2453, 2008.

• [82] Tao, H., Kotsopoulos, A. J., Duarte, J. L.; Hendrix, M. A. M., A Soft Switched Three-Port Bidirectional Converter for Fuel Cell and Supercapacitor Applications, In: Proc. IEEE Power Electronics Specialists Conference (PESC"05), Recife, Brazil, pp. 2487- 2493, 2005.

• [83] Inoue, S., Akagi, H., A Bidirectional DC-DC Converter for an Energy Storage System with Galvanic Isolation, IEEE Trans. on Power Electronics, vol. 22, no. 6, pp. 2299- 2306, 2007.

• [84] Duarte, J. L., Hendrix, M., Simoes, M. G., Three-Port Bidirectional Converter for Hybrid Fuel Cell Systems. IEEE Trans. on Power Electronics, vol. 22, no. 2, pp. 480- 487, 2007.

• [85] Zhang, Z., Thomsen, O. C., Andersen, M. A. E., Nielsen, H. R., Dual-Input Isolated Full-Bridge Boost DC-DC Converter Based on the Distributed Transformers. IET Power Electronics. vol. 5, issue 7, pp. 1074-1083, 2012.

• [86] De Doncker, R. W., Divan, D. M., Kheraluwala, M. H., A Three Phase Soft-Switched High-Power Density DC/DC Converter for High-Power Applications. IEEE Trans. on Industry Applications, vol. 27, no. 1, pp. 63-73, 1991.

• [87] Chen, Y. M., Liu, Y. C., Wu, F. Y., Multi-Input DC/DC Converter Based on the Multiwinding Transformer for Renewable Energy Applications. IEEE Trans. on Industry Applications, vol. 38, no. 4, pp. 1096-1104, 2002.

• [88] Tao, H., Duarte, J. L., Hendrix, M. A. M., High-Power Three-Port Three-Phase Bidirectional DC-DC Converter. Industry Applications Conference (42nd IAS Annual Meeting), pp. 2022-2029, 2007.

• [89] Hwang, J. J., Chen, Y. J., Kuo, J. K., The Study on the Power Management System in a Fuel Cell Hybrid Vehicle. International Journal on Hydrogen Energy, vol. 37, no. 5, pp. 4476-4489, 2012.

• [90] Saha, S. S., Efficient Soft-Switched Boost Converter for Fuel Cell Applications. International Journal on Hydrogen Energy, vol. 36, pp. 1710-1719, 2011.

• [91] Markel, T., Simpson, A., Energy Storage Systems Considerations for Grid-Charged Hybrid Electric Vehicles. CP-540-38538, www.ctts.nrel.gov/.

• [92] Park, C. R., Next Generation Lithium Ion Batteries for Electrical Vehicles. Vukovar, In-Tech, 2010.

• [93] Caux, S., Hankache, W., Fadel, M., Hissel, D., On-Line Fuzzy Energy Management for Hybrid Fuel Cell Systems. International Journal on Hydrogen Energy, vol. 35, no. 5, pp. 2134-2143, 2010.

• [94] Kuperman, A., Aharon, I., Battery Ultracapacitor Hybrids for Pulsed Current Loads: A Review. Renew Sustainable Energy Review, vol. 15, pp. 981-992, 2011.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 13

DYNAMIC RESPONSE OF ELECTRIC MACHINES FOR ELECTRIC VEHICLES/HYBRID ELECTRIC VEHICLES (EV/HEV)

Robert Beloiu , Mariana Iorgulescu† and Corina Savulescu University of Pitesti, Pitesti, Romania

ABSTRACT

This chapter presents some of the most common traction solutions used machine for electric vehicle/hybrid electric vehicle (EV/HEV) adopted by different car manufacturers. It also covers the dynamic response of the most used electric machines applied in EV/HEV. We describe three types of electric machines: brushed direct current machine (DC), brushless machine (BLDC) and induction machine (IM). The brushed DC machine presents the important disadvantage related to its maintenance cost. The need for brushes change after a certain period of time makes it very costly. Brushless machines do not have this maintenance cost related problem. . For the induction machine the cost of production and maintenance is the smallest among the three analyzed machines. Consequently, this type of electric machine is more fitted to be used than brushed or brushless DC machines. .

This chapter describes the behavior of three types of electric machines used nowadays in EV/HEV commercial vehicles. To compare these types of machines we developed a testing regime based on the experience of a driver facing all functional situations met. Thus, we took into consideration the rapid and slow start-stop regime, speed change regime and the backward movement of the vehicle. We have studied the behavior of the three types of machines using these testing regimes. In spite of their advantages and disadvantages of one versus the others, these types of electric machines are all used nowadays in the prototypes and commercial EV/HEV.

Keywords: Electric Machine, Electric Drive, Electric Vehicle, Hybrid Electric Vehicle, DC Machine, Induction Machine, Brushless Machine, Dynamic Regime

E-mail: robert.beloiu@upit.ro.

E-mail: mariana.iorgulescu@upit.ro.

E-mail: corina.savulescu@upit.ro.

1.INTRODUCTION

Nowadays, there is a great interest in the field of EV/HEV. Many car companies develop solutions for EV/HEV both commercial as prototypes still under investigation. These solutions include different types of electric machines (DC, brushless, induction, synchronous, etc.). However, these machines present advantages and disadvantages and there has not been adopted a standardized unique solution adopted yet.

These solutions are available today on EV/HEV market [1,2]:

• DC machine: PSA Peugeot-Citroen/Berlingo
• Induction machine: GM Silverado ISG, Continental ISAD, Delphi-Automotive ISG, Valeo ISA, Renault Kangoo, Chevrolet Silverado, BMX X5
• Brushless permanent magnet machine: Honda Insight and Civic mild hybrids, FCX-V3 FCEV; Mannesman-Sachs
• Synchronous machine: Toyota Prius, Estima, Ford Hybrid Escape, Nissan Tino, Honda Insight

The electric drive is an essential part in the EV/HEV structure. In order to ensure the desired capabilities of a vehicle, the traction sub-ensemble has to fulfill certain characteristics given by the electric drive system which includes the electric machine.

It is known that electric machines are manufactured in a variety of configurations and that they are able to drive all kind of industrial applications. Theoretically, any kind of electric machine could be used to drive EV/HEV, yet the car companies focused on some specific electric machines that proved to provide advantages over the others.

The electric machine for EV/HEV needs to have various specific characteristics related to the different functional regimes: acceleration, braking, very fast change in the functional regime, capability of developing very low and also very high speeds, frequent starts and stops, good proportion between mass and volume, robustness, functionality and low maintenance cost, etc. [3]. The mechanical characteristics for electric machines used for EV/HEV are presented in Figure 1.

Figure 1. Typical Performance Characteristics of Electric Machine for Traction [4].

In this chapter we present some of the most common electric machines for EV/HEV adopted by different car manufacturers. We also describe the dynamic response of the best know electric machine used in EV/HEV, and we end by comparing the studied electric machines.

2. ELECTRIC MACHINES USED IN EV/HEV

2.1. The DC Machine

There are four constructive types of DC machines, depending on the excitation field: series, independent, shunt and compound. The DC permanent magnet machine falls in the category of independent field excitation machines. The constructive type depends on the way the winding field is connected to the winding armature shown in Figure 2.

Figure 2. DC machines configurations.

The use of DC machines in electric traction is not novel. The most used DC electrical machines in traction applications are the series excitation winding, due to their mechanical characteristics that allows a smooth start with a large resistive torque. Even though the behavior of the DC machine is well known and well covered in the literature, there is still interest in developing new functional regimes. The ideal traction machine should have both the characteristics of series field connection (for starting regime) and of the shunt connection (for regenerative braking and speed control). Such behavior should improve the EV/HEV functionality and the battery use [5].

2.1.1. Mechanical Characteristics

The steady state equivalent circuit of the DC machine is indicated in Figure 3, and it is widely covered in the technical literature.

Figure 3. Equivalent circuit for DC machines.

The basic equations, using the standard notations that describe the steady state functional regime of a DC machine, are as follows:

$$\begin{cases} U_A = R_A \cdot I_A + E; & E = K \cdot \Phi \cdot \Omega \ T = K \cdot \Phi \cdot I_A \end{cases} \tag{1}$$

where - armature resistance, - armature current, - developed torque, - field excitation, - machine constant, - rotor angular speed. R^{A A} I T K

The relation between the rotor angular speed and the developed torque is derived from the Equation(1), represents the mathematical form of the mechanical characteristic: T

$$\Omega = \frac{U_A}{K \cdot \Phi} - \frac{R_A}{\left(K \cdot \Phi\right)^2} \cdot T \tag{2}$$

The Equation (2) refers to all types of DC machines, and the graphical representation is indicated in Figure 4.

Figure 4. DC machines" mechanical characteristics.

Before development of power electronic devices, the DC series machine was the only fitted for electric traction applications, due to its mechanical characteristic, such as: development of high rated torque values for low values of speed. Even though this special characteristic is very important for electric traction vehicles, the use of DC series machines has its limitations, especially because the speed rises as the torques values decrease.

Nowadays, when power electronics and numerical control developed very much, manufacturers overcame the past limitation of using DC shunt machine for EV/HEV applications. Due to the modern drive systems, they combined the ease in controlling the DC independent/shunt connection machines and the advantages of DC series connection machines.

Analyzing the Equation (2) for DC shunt machines, there are several functional regions that could be defined as indicated in Figure 5:

• Nominal functionality
• Below nominal functionality
• Above nominal functionality

The "nominal mechanical characteristic" is obtained when the machine functions at the nominal parameters. The point in Figure 5 represents the Static Operating Point (SOP) when the voltage and torque are at nominal values.

The "below mechanical characteristic" is obtained by varying the armature voltage. The point in Figure 5 represents the SOP when applied voltage is below the nominal value. The "below mechanical characteristic" is characterized by lower speed comparing to the nominal value. The mechanical characteristics are parallel to the nominal. When the machine works at low and very low speed, the ventilation is lower and the temperature of the machine rises. Thus when working at this parameters, cooling of the machine should be taken into consideration.

The "above mechanical characteristic" is obtained by varying the field voltage. The point in Figure 5 represents the SOP when applied voltage is above the nominal value. The "above mechanical characteristic" is characterized by higher speed comparing to the nominal value. The mechanical characteristics are not parallel to the nominal. The load capacity of the machine is lower than the nominal and below nominal mechanical characteristics. Thus, when working at these parameters, the load capacity has to be taken into consideration.

Figure 5. Mechanical characteristics for shunt DC machines.

2.1.2. Chopper Control of DC Machines

The chopper control is not the only controlling method by which the speed of a DC machine can be varied. A different method of speed control is by adding supplementary resistors. Because of energetic consideration, this method is not appropriate for EV/HEV, and therefore it is not covered in this chapter.

The variation of speed for the machine can be controlled by variation of the field voltage variation. This method is not covered in this chapter.

2.1.2.1. Chopper Control for Single-Quadrant DC Functionality

The control of the armature voltage applied to the DC machine can be obtained by the use of chopper control techniques.

The generic chopper schematic consists in a power electronic switch connected in series with the DC machine, like in Figure 6.

The power source has constant value V, which in case of an EV/HEV is provided by the main battery. The voltage is applied to the DC machine through the power electronic switch PES. The free regime diode (FRD) ensures power dissipation from the machine while PES is open.

In the duty period of time , the PES is closed and the source voltage is applied to the machine. In this period of time, the current rises from the minimum value to the maximum value . In the freewheeling period of time , the PES is open and no voltage is applied to the machine. In this time interval the current falls from the maximum value to the minimum value . The FRD ensures the electric current while the PES is open, thus improving the current waveform. 0 t T ^A_ min I ^A_ max I T t T ^A_ min I

2.1.2.2. Chopper Control for Multiple-Quadrant DC Functionality

The problem of the indicated schematic in Figure 6 is that the DC machine functions only in one functionality quadrant. For EV/HEV the DC machine has to be able to operate in all four functionality quadrants.

In order to obtain this, a generic schematic for the chopper is used in Figure 7. All PES are closed and opened alternatively PES1 + PES3 and PES2 +PES4 in order to ensure full functionality of the DC machine for all four quadrants. FRDs also work in groups of two depending of the working quadrant of the DC machine.

Figure 6. Single-quadrant DC machine chopper drive.

Figure 7. Four quadrant DC machine chopper drive.

2.1.2.3. Driving System of Chopper Controllers for DC Machines

Either for single-quadrant or multiple-quadrant functionality of DC machines, the driver is a closed loop system that controls both the current and speed.

The schematic of closed loop control of shunt DC machine is indicated in Figure 8.

In Figure 8, the is the current transducer and the is the speed transducer. The R^w and RIa are the speed and current regulators placed in the speed and respectively current closed control loops. The measured current and speed are compared with the prescribed values and the result is applied to the respective regulator. The output signal of the current regulator, is the G command signal for the PES. Thus the voltage level applied to the machine armature is obtained.

Figure 8. Schematic for a drive system for chopper controllers for DC machine.

2.2. Induction Machine

There are two types of induction machines: squirrel cage (short circuit rotor) and wounded rotor.

Because of their high maintenance cost due to brushes, the wound-rotor induction machines are not as convenient as the squirrel cage induction machines for EV/HEV applications.

The squirrel cage induction machines have a simple construction and thus their fabrication and maintenance cost is lower comparing with other types of electric machines.

One disadvantage of the induction machines is the difficulty of controlling in closed loop systems. Sometimes the controller is more expensive than the machine, making these machines difficult to integrate in variable speed applications. However, these difficulties were overcome by the evolution of power devices and industrial computing.

2.2.1. Mechanical Characteristics

In order to determine the functional characteristics of the induction machine, its equivalent schematic is used in the specialized literature. This is indicated in Figure 9.

In Figure 9, the , , , and are the stator quantities of voltage, current, resistance U^{s s} I R^s L^s

and inductance. The , , and are the rotor quantities of current, resistance and inductance reported to the stator. The rotor slip is s, which indicates the delay between the angular speed of the stator field and the rotor angular speed defined as: r I Rr s Lr

$$s = \frac{\omega_1 - \omega_2}{\omega_1} \tag{3}$$

where is the synchronous speed (the speed of the inductor field), is the rotational rotor speed and p the number of pole pairs. 1 2

Analyzing the equivalent schematic, the Kloss relation indicates the relation between the torque and slip speed, is approximated by:

$$\frac{T}{T{cr}} = \frac{2}{\frac{s}{s{cr}} + \frac{s_{cr}}{s}} \tag{4}$$

Figure 9. The equivalent schematic of induction machine.

In the Equation (4), and represents the critical torque and respectively the critical slip speed, while and represents the torque and slip speed at certain functional conditions. Tcr s

From Equation (4) we obtain the mechanical characteristic for the induction machine, displayed in Figure 10.

Figure 10. Mechanical characteristic of induction machines for U, f constant control, (a) , (b) T f s f T

As mentioned above, Figure 10 presents the mechanical characteristic of an induction machine when the supply voltage and frequency are constant. It can be observed that the starting torque of the induction machine has low values and the speed variation is very small with the load variation. When the induction machine is operated, the stable functional domain is when . This domain is very narrow and therefore the speed can vary in a very limited domain. For the EV/HEV applications, this behavior is not acceptable. cr s s

2.2.2. Control of Induction Machines for Constant U/f

In the domain of constant U/f the induction machine behaves similar with DC machines. The mechanical characteristics of its control are indicated in Figure 11.

The "nominal mechanical characteristic" is obtained when the machine functions at nominal parameters. The point in Figure 11 represents the Static Operating Point (SOP) when all physical quantities applied to the machine are at nominal values.

Figure 11. Mechanical characteristics for U/f constant control.

The "below mechanical characteristic" is obtained by varying the supply voltage and frequency so that U/f is maintained constant. The point in Figure 11 represents the static operating point (SOP) when the applied voltage and frequency are below the nominal values. The "below mechanical characteristic" is characterized by lower speed compared to nominal value. The below mechanical characteristics are parallel to the nominal. When the machine is working at low and very low speed, the ventilation decreases and consequently its temperature rises. Therefore when working at these parameters, machine cooling has to be taken into consideration.

The "above mechanical characteristic" is obtained by varying the frequency of the supply voltage. The value of the voltage is maintained constant at the nominal value. The point in Figure 11 represents the static operating point (SOP) when the applied frequency is above the nominal value. The "above mechanical characteristic" is characterized by higher speed than the nominal. The mechanical characteristics are not parallel to the nominal. The load capacity of the machine is lower than the nominal and below nominal mechanical characteristics. Therefore, when working at these parameters, capacity load has to be taken into consideration.

2.2.3. Driving System of U/f Controllers for Induction Machines

The U/f controller consists of 6 PESs, displayed in Figure 12. These PESs ensure that the source voltage is applied to the stator windings of the induction machine IM. The PESs are commanded with signals provided by three comparators. The command signal is obtained by comparing the 3ph sinus voltage provided by with the triangular signal provided by . *U*sin Utri

Figure 12. Schematic for a U/f constant controller drive system of s for induction machines.

The PES1…6 are power electronic switches that could be of different types, depending on the power of the machine, and consequently on the absorbed current. The groups of PES1,3,5 and PES2,4,6 are closed alternatively depending on the output signal of COMP1-3. When the sinusoidal voltage the output signal . When the output signal COMP1<0 (Figure 13). COMP¹ 0

Figure 13. (Continued)

Figure 13. The schematic of command circuit of U/f constant controllers for induction machines

Depending on the polarity of signal either PES1 or PES2 conducts providing maximum or minimum values of the voltage supply applied at machine"s phases, obtaining the alternating current values in its windings.

2.3. Brushless Machine

Regarding construction principles, brushless (BLDC) machines are similar to the permanent magnet synchronous machines. The similarity between brushless machines and synchronous machine lies in the fact that both magnetic fields produced by the stator and the rotor rotate at the same frequency.

In a brushless machine, the stator windings" configuration is similar to that of poly-phase AC machines, especially to the induction machines. However, the winding configuration is distributed differently.

The structure of the rotor of a brushless machine contains one or more permanent magnets, and can vary from 2 to 8 pole pairs.

The brushless machines are manufactured in 2-phase or 3-phase configuration, although the 3-phase is wider spread in industrial applications.

Figure 14. Mechanical characteristic of BLDC machines.

Compared to the brushed DC machines, the BLDC present some important advantages, such as:

• Easier cooling due to windings placement in the stator
• Lower maintenance cost due to the lack of brushes
• Higher efficiency, as the Joule loses are lower than the corresponding DC brushed permanent magnet machine due to the windings placement in the stator

2.3.1. Mechanical Characteristics

In the technical [7] and scientific [8] literature, it is considered that the mechanical characteristic of BLDC is very similar with the brushed DC machines.

Analyzing the mechanical characteristic of BLDC machine, it can be stated that its behavior is similar with the brushed DC machine. At starting the machine develops a high peak torque, which leads to a high peak current.

2.3.2. Control of Brushless Machines

The brushless driving system contains six PES controlled by a Gate Command System (GCS). The GCS block receives signals from the machine and the imposed speed *. Based on the input signals, the GCS forms the command signals for the PES1-6 that drives the machine. Both the electronic system GCS and PES1-6 function as an electronic commuter which drives the machine. In traction applications, the speed control for brushless machine raises a special interest [7].

Figure 16 displays some waveforms that appear in a driving schematic of brushless machines. Figure 16(a) indicates the currents waveforms. Depending on the inertia of the machine and the resistant torque at the shaft, the current of the machine at starting has a high peak value.

The CGS block forms the command signals for the PES1÷6 indicated in Figure 16(c). The inputs of CGS are the machine currents, feedback signals and the output of a PI speed regulator.

Figure 15. Driving system schematic for a brushless machine.

Figure 16. Waveforms in the driving circuit of brushless machine.

3. DYNAMIC REGIME OF ELECTRICAL MACHINES USED IN EV/HEV

In EV/HEV the functional regime of the electric traction machine could be of two types:

• transient or dynamic
• stable

The dynamic regime for EV/HEV has some particularities compared to the industrial applications. In the normal use of a vehicle, various situations can occur consisting in: acceleration, braking, very fast change in the functional regime, capability of developing very low and also very high speeds, frequent starts and stops etc [9].

For the purpose of comparison between different types of electric machines used nowadays for EV/HEV, a speed profile was created to be applied to each type of machine (Figure 17).

Figure 17. Testing speed profile.

The following notations are used in Figure 17:

• 0 periods of time when the vehicle is stopped
• 1, 2 rapid variation of speed
• 3, 4, 5, 8 stable speed regime when the vehicle is moving forward
• 6 stable speed regime when the vehicle is moving backward
• 7, 9 slow speed variation

The speed variation situations presented above are intended to simulate several functional regimes most likely to appear during the use of a real vehicle. The 1 and 2 could be speed regimes likely to appear at the traffic light crossroads. Situations 3, 4, 5, 8 could be various levels of speed cruise, while 6 is moving backward. Situations 7, 9 are likely occur while starting/stopping the vehicle. The transition 0-3 simulates very rapid starting, while 3-4-5 simulates rapid braking and accelerating situations. The 5-0 simulates a dangerous situation when a very rapid stopping of the vehicle is needed. The situation simulated by 3-4-5 could very easy be a cruise situation, when a rapid increase of speed is needed for overtaking other vehicle. All these transitory regimes are very common in daily driving, especially when using vehicles in crowded urban areas. For the experimental data, we applied the same resistant torque at the shaft to the analyzed machines: DC, induction and brushless.

Nominal data of the machines are indicated in Table 1.

DC machine Induction machine BLDC machine UAn = 220 V c.c. I^A = 2 A, Ifield = 0,2 A nⁿ = 1.500 rpm Pⁿ = 0.373 kW Uⁿ = 230/400 V Iⁿ = 3,5/2,1 A fⁿ = 50 Hz Pⁿ = 1 CV = 0.735 kW nⁿ = 1500 rpm M⁰ = 2.6 Nm *M*max = 13 Nm I⁰ = 4.2 A *I*max = 21 A nⁿ = 4000 rpm

Table 1. Nominal data of tested machines

3.1. DC Machine

The DC machine functions in a closed loop control speed schematic. The schematic contains also a subordinate current control loop. In order to accurately function, the driver should be set before use and the most appropriate configuration for current and speed regulators should be chosen. The procedure of setting up the DC machine"s driver is presented in previous works of the authors [10,11]. Briefly, this procedure consists in adjusting the values for the current and speed regulators" parameters to obtain the most appropriate behavior in the transient regime.

Figure 18. (Continued)

Figure 18. Set-up of DC machine driver control loops.

Figure 18 represents the excitation speed signal and the acquired output speed signal for various configurations of the driving schematic. During the parameter adjustment, different improved output signals are obtained. The best configuration of the driver is the one that provides the quickest answer from the machine. The best response is a compilation of the time delay, the override output signal and the delay of the steady state.

Figure 19. Dynamic behavior of DC machine.

Once the best considered configuration is chosen, the system is tested using the userdefined speed testing profile as indicated in Figure 19. It is noteworthy that the output speed of the machine follows the testing speed profile.

3.2. Induction Machine

The induction machines function in an open loop U/f converter. This means that there is no speed feedback to the controller. For this reason, there is no need for adjustment of internal regulators dynamic parameters.

Figure 20. Dynamic behavior of induction machine.

The dynamic behavior of the system containing the induction machine is indicated in Figure 20. The output speed signal follows the testing speed input signal. The shape of the input and output speed signals is similar, but there are substantial errors between the two signals. The main reason for these errors is that the system functions in open loop configuration. Thus the speed error is not corrected in any way through a feedback.

3.3. Brushless Machine

The brushless machine, likewise the DC machine, functions in a closed speed control loop. The brushless machine"s driver parameters should be set-up before any experiment is conducted. The procedure of setting up the brushless machine"s driver is covered in previous works of the authors [12]. Just as for the DC machine, the procedure consists in establishing the parameters for the current and speed regulators" parameters.

Figure 21 presents the set-up procedure for the brushless system. Equally as the DC brushed machine, the BLDC machine drive includes a regulator parameter configuration in order to achieve the best behavior. As in any closed loop systems, monitored parameters are: the delay of the dynamic response, the override signal and the delay to reach the steady state.

Figure 21. Set-up of brushless machine driver control loops.

Figure 22. Dynamic behavior of BLDC machine.

The effect of the feedback loop of the driver for BLDC machine is displayed in Figure 22. The shape of the output speed signal is similar to the input speed excitation signal. There are slight errors between the input and the output speed signals.

4. COMPARATIVE BEHAVIOR OF ELECTRIC MACHINES FOR EV/HEV

For the comparison of the dynamic regime for the three electric machines, an excitation speed signal was applied to its" drivers (Figure 23). The dynamic response of these machines was recorded (Figure 24)

Figure 23. Testing speed profile.

Figure 24. Comparative dynamic behavior of DC, induction and BLDC machines.

Firstly the speed responses of DC and BLDC machines are very similar, which is consistent to other results from technical [8] and scientific literature [13] . Secondly, for the forward move of the EV/HEV equipped with these kinds of machines, the error between the input and the output speed signals is small. Therefore in all imposed functional regimes, the output signal follows the input signal with an acceptable error.

Third, in the slow deceleration regime (9 in Figure 24) a smoother behavior for the BLDC machine compared to the brushed DC machine can be noticed. In addition, at a sudden change of functional regimes, the DC machine has bigger override value of measured speed than the BLDC machine. When the slow acceleration is applied (7 Figure 24) the override of the DC machine is again bigger than the one for the BLDC machine. Nevertheless, the response is attenuated rapidly and the brushed DC machine has a very similar behavior with the BLDC machine.

Forth, a different situation can be observed for the induction machine. The experiments revealed a very large error of the induction machine"s output speed compared to the input speed-testing signal. At every regime change there is a notable delay, in both senses of rotation. The profile of output speed for the induction machine follows the profile of the testing signal, but there is an important difference in amplitudes.

One of the reasons for such behavior is the fact that in this circumstance, the induction machine"s controller is an open loop U/f driver. Therefore the output speed is never compared with the input signal and speed regulators never process the error.

CONCLUSION

From the above mentioned analysis the following conclusions could be drawn:

• The most accurate dynamic behavior for regime changes is offered by the BLDC machine as indicated by our tests
• The dynamic behavior of brushed DC machine is similar with the BLDC machine
• The dynamic behavior of the induction machine controlled by a U/f driver in an open loop configuration presents very large errors

Each analyzed solution for traction for EV/HEV in this chapter has its advantages and disadvantages.

The brushed DC machine fabrication technology and control systems arrived at their maturity, it is envisaged its costs will decrease [14]. Nevertheless the brushed DC machine presents the important disadvantage of high maintenance cost.

The brushless machines do not present the same problem as the brushed DC machines related to maintenance cost, as the brushes are eliminated. Since the fabrication process of magnets improves and the price for control and power electronics decreases, the BLDC machines become a very suitable solution for replacing the brushed DC machines.

Comparing the brushless machines with induction machines, it is outlined in the literature that the stators for a three-phase induction machine and for a BLDC machine are practically identical [14]. However, these machines differ in the rotor construction and the electronic controller.

The cost of production and maintenance for induction machines is the smallest among the three types analyzed. For this reason they are more fitted to be used than the brushed or the brushless DC machines in EV/HEV, but they are more difficult to control.

In spite of presented advantages and disadvantages, all three categories of machines are used in prototypes and already commercial EV/HEV. There is still more room for further studies to develop the best electric machine configuration to be implemented in EV/HEV.

NOMENCLATURES

BLDC	Brushless DC
COMP	Comparator
DC	Direct Current
FRD	Free Regime Diode
GCS	Gate Command System
PES	Power Electronic Switch
SOP	Static Operating Point

REFERENCES

• [1] Miller, J. M.; Propulsion Systems for Hybrid Vehicles. The Institution of Engineering and Technology, London, UK, 2008.
• [2] Zeraoulia, M.; Benbouzid, M. H.; Diallo, D.; Electric Motor Drive Selection Issues for HEV Propulsion Systems: A Comparative Study. IEEE Transactions on Vehicular Technology, vol. 55, no. 6, pp. 1756-1764, 2006.
• [3] Xue, X. D.; Cheng, K. W. E.; Cheung, N. C.; Selection of Electric Motor Drives for Electric Vehicles. Australasian Universities Power Engineering Conference (AUPEC'08), Sydney, Australia, paper 170, pp. 1-6, 2008.
• [4] Dong, G. G.; et al.; Studies of Electric Motor for Light-weight Electric Vehicle. Malaysian Universities Transportation Research Forum and Conferences (MUTRFC2010), Putrajaya, Malaysia, pp. 135-148, 2010.
• [5] Khan, R. U.; Khan, M. A.; Arjumand, H.; Analysis of Innovative Applications of Single DC Motor in Series and Separately Excited Mode for Hybrid Electric Solar Car. International Journal of Engineering Science and Technology, vol. 2, no. 3, pp. 312- 316, 2010.
• [6] www.powersimtech.com.
• [7] Mehrdad, E.; Yimin, G.; Ali, E.; Modern Electric, Hybrid Electric, and Fuel Cell Vehicles. CRC Press, 2010.
• [8] Jung, G. H.; Yeom, J. H.; Kim, M. G.; Determination of Torque-Speed-Current Characteristics of a Brushless DC Motor by Utilizing Back-EMF of Non-Energized Phase. Journal of Magnetism and Magnetic Materials, vol. 310, pp. 2790-2792, 2007.

• [9] Xue, X. D.; Cheng, K. W. E.; Cheung, N. C.; Selection of Electric Machine Drives for Electric Vehicles. Australasian Universities Power Engineering Conference (AUPEC'08), Sydney, Australia, pp. 1-6, 2008.

• [10] Beloiu, R.; Iorgulescu, M.; Dumitru, O.; Combei, P.; Stancescu, A.; Electric Drive System with DC Motor with Closed Control Speed Loop. Electronics, Computers and Artificial Intelligence (ECAI 2005), Pitesti, Romania, pp. 47-50, 2005.

• [11] Beloiu, R.; Iorgulescu, M.; Dumitru, O.; Combei, P.; Stancescu, A.; Electric Drive System with DC Motor with Closed Current Loop and Open Speed Loop. Conference on Electronics, Computers and Artificial Intelligence (ECAI 2005), Pitesti, Romania, pp. 50-53, 2005.

• [12] Beloiu, R; Iorgulescu, M.; Set up of Brushless Control Drivers in Hybrid Electric Vehicle. Conference on Electronics, Computers and Artificial Intelligence (ECAI 2011), Pitesti, Romania, pp. 109-113, 2011.

• [13] G. Ellis; Advances in Brushless DC Motor Technology, Control, and Manufacture, PCIM-Europe, 1996, www.kollmorgen.com/uploadedfiles/Files/Website/Common/ Images/pcim1996brush_vs_brushless.pdf, (accessed January 2014).

• [14] W. Rippel; Induction versus DC Brushless Motors, http://www.teslamotors.com/blog/ induction-versus-dc-brushless-motors, (accessed January 2014).

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 14

USE OF THE EXTERNAL MAGNETIC FIELD FOR SUPERVISION AND DIAGNOSIS OF ELECTRICAL TRACTION MOTORS

Remus Pusca 1,2,* Raphael Romary1,2,† and Andrian Ceban1,2,‡ University of Lille Nord de France, Lille, France University of Artois, LSEE, Bethune, France

ABSTRACT

The work presents a supervision method for detection of AC electrical machine faults, generally used in electrical traction, exploiting the information given by the external magnetic field in the vicinity of the machine. This magnetic field is an image of the air-gap flux density and consequently, it contains information concerning the failure presence. In this work, two methods based on the analysis of the magnetic field outside the machine are presented. They are reliable, original, inexpensive and simple to implement in electrical automotive vehicles. They present the advantage of being noninvasive and just get rid from the main drawback presented by other diagnosis methods based on the comparison with a healthy state assumed to be known. Both methods need simple coil sensors placed in specific positions near the motors for flux density measurements. These sensors induce an electromotive force whose derivative effect amplifies the low and medium frequency harmonic components. These methods are developed in the case of AC electrical rotating machines (induction and synchronous machines) often used in electrical traction. The analysis is focused on the low frequency components of the magnetic field which is not influenced by the presence of electronic converters usually used in traction applications. The theoretical analysis and the experimental validation are also presented.

^* E-mail: remus.pusca@univ-artois.fr.

^† E-mail: raphael.romary@univ-artois.fr.

E-mail: andrian.ceban@univ-artois.fr.

Keywords: AC Motors, Electrical Traction Application, Fault Diagnosis, Low Frequency, Magnetic Field Spectral Analysis

1.INTRODUCTION

The work presented in this chapter is a part of the current concerns in the development of electric traction systems where EV (electric vehicle) and HEV (hybrid electric vehicle) are used. Although electric traction is not a new idea, its application to electrical automotive vehicle has be previously abandoned, consequence of an insufficient level of technological developments. Today, the interest in electrical traction grows in order to respond to current problems like the development of a non-polluting vehicle with sufficient reliability and autonomy. In this context, the automotive vehicle should not be a modification of the conventional one but it must be a "new vehicle".

Taking that point into consideration the work presented in this chapter shows a noninvasive method for supervision and diagnosis of electrical motors used in automotive vehicles. The presence of new traction control systems in automotive vehicles allows faster control of drive torque and improved dynamic performance [1]. However, these systems require new methods of monitoring and diagnosis to detect the faults because strong vibrations in embedded systems reduce the reliability of all components including electrical machines (motors and generators). To improve the reliability of the "new vehicle", the proposed method, presented in this chapter, can insure a preventive diagnosis based on the analysis of the magnetic flux leakage measured outside of the electrical machines by noninvasive sensors.

Considering the electrical motor as part of EV and HEV, this chapter presents the sensors used, their location in the vicinity of motor and the analysis of measured signals allowing an easy use of the method for supervision and diagnosis of AC motors. Generally, electrical motor diagnosis methods require a good knowledge of the machine and its behaviour in the presence of internal faults. To diagnose electrical machines used in classical applications, different methods can be distinguished [2]:

• diagnosis techniques that rely on knowledge methods and which usually use artificial intelligence methods. These methods are enriched by feedback and do not require the mathematical model of a faulty machine.
• methods of identification and estimation of the physical parameters of the machine. In this case, monitoring is established by comparing the measurement results with those obtained by analytical modelling or by calculation of a state vector.
• methods based on signal modelling. These methods analyze the time evolution or the spectral content of measured signals in order to detect a fault of the electrical machine.

Today, many works are dedicated to motor fault analysis using different methods such as the current signature analysis [3, 4, 5], exploitation of the slotting harmonics [6, 7], artificial intelligence techniques [8], vibrations and noise measurement [9, 10, 11] or flux analysis [12, 13, 14]. All these techniques have been validated beyond the research laboratories in order to find their place in industry [15], and it could be interesting to check if these methods can be implemented in electric traction motors in order to diagnose incipient faults. In electric traction applications, the speed variation is often insured by static converters which feed the motors with adapted waveforms of the currents. These signals are generally not perfectly sinusoidal and are source for electromagnetic disturbances. Moreover, the interpretation of the results obtained with these methods requires an advanced level of expertise which hinders a real democratization of the techniques among users of electrical machines. Indeed, only systems where a breakdown of the machine can have disastrous consequences (e.g. the electricity generating plant) are equipped with a supervision system.

Generally, all diagnosis methods require the knowledge of the healthy state of the machine regardless of the exploited variable. The fault detection is then based on the comparison of a given signature with that of the healthy state assumed to be known, considering an indicator sensitive to a particular fault. The difficulty with this approach is that, often, the sensitive indicators exist even for the healthy machine (case of a spectrum lines). Then, it is their change, often their increase, that is likely to provide information on the presence of a fault. On the other hand, the load of the machine can be considered for diagnosis as a disturbing factor, because it induces several healthy states. Additional difficulty lies in the fact that the supposed healthy state is almost never known before the failure appears, because the operator of the machine has not taken the precaution of recording the parameters that characterize the healthy machine.

Recently, methods based on the analysis of the external magnetic field have appeared, they have the advantage of being non-invasive and easy to implement [13, 16, 17]. These techniques exploit information in the stray field around the machine in order to respond to industrial problems concerning the diagnosis [18], but also on monitoring the aging of insulation or the monitoring of eco efficiency [19, 20]. The various applications differ in the analyzed frequency range.

The diagnosis techniques exploiting the low and medium frequency ranges, up to the slotting effect frequencies of the electric machine, give the possibility to detect stator faults (inter-turn short circuit) and rotor faults (broken bars, inter-turn short circuit) by analyzing the external magnetic field of induction and synchronous machines. These techniques are based on an analytical model which considers the decomposition of the external magnetic field into a radial and axial field. The both fields do not have the same properties, because each has a different origin. This model is able to predict the spectral components of the stray field affected by different faults.

The work presented in this chapter mainly considers the effects of a fault on the magnetic stray field. It is oriented toward the diagnosis of rotating electrical machines including electric traction motors since. The proposed diagnosis methods are not influenced by the static converters and use the load operating condition for detection of the faults. Nowadays, rotating electrical machines are presented in all areas of modern life including the electrical vehicle and one of the most important objectives of the maintenance systems is the detection of incipient faults in order to anticipate a shutdown of the system. In this work, two methods based on the analysis of the magnetic field outside the machine are presented.. These methods are developed in the case of AC electrical rotating machines (induction and synchronous machines). The analysis focuses on the magnetic stray field in the vicinity of the machine, especially its radial an axial distribution which presents different sensitivity according to various faults.

To apply these methods in electrical traction, two coil sensors are used to measure the stray field in the motor vicinity (Figure 1). These sensors induce an electromotive force (emf) which derivative effect amplifies the low and medium frequency components. They are simple to carry out, non-invasive and low cost. Moreover, they can be set up outside the motor without needing much space in electrical system of hybrid vehicles. The diagnosis procedure is able to indicate the failure presence and can be implemented within a simple microcontroller. For diagnosis, each fault gives specific signature that are identified and analyzed.

Figure 1. Positions of sensors for on-line supervision of HEV traction motor.

Two diagnosis methods are developed in this work. The first one deals with the inter-turn short-circuit faults in the stator winding of an AC machine. It uses two external flux coil sensors placed in specific positions that measure the external magnetic field. The signal delivered by the both sensors is exploited for diagnosis. The analysis concerns the magnitude of specific harmonics of the electromotive force delivered by the coil and the method does not require any knowledge of a presumed healthy former state of the machine. Its originality is the use of the load to improve the diagnosis. Here, the load is not longer a disturbing factor but it is used for the fault discrimination. Actually a comparison between machine operating conditions (no-load and load) enables to detect a stator fault.

The second method is based on the variations of the axial flux density in the presence of rotor faults. Any changes of magnetic circuit conditions, such as a broken bar or the appearance of an eccentric air-gap due to bearing wear, will be reflected with a corresponding change in the axial flux. The analysis is focused on the low frequency components of the magnetic field, which are not influenced by the presence of electronic converters usually used in industrial and traction applications. Moreover, due to their low frequency, these components are less attenuated during their penetration through the stator magnetic core or the machine external frame. Both methods need simple coil sensors for flux density measurement.

In the end of the chapter, the experimental implementation of the sensors and its results are presented. The faults have been artificially created in the rotor and in the stator of different AC machines, especially designed for the tests.

2. MAGNETIC FIELD

From a diagnosis point of view, the advantage of methods based on the measurement of the external magnetic field is that they are non-invasive. These methods are simple to implement and they need a sensor placed near the electrical machine, able to measure the external magnetic field. Considering the position of the sensor, the measured stray magnetic field does not originate from the same source, and does not result from the same physical phenomenon [21].

2.1. Measurement of External Magnetic Field

In a rotating electrical machine, the leakage flux is created by the different elements of the machine. A distribution of these field lines is shown in Figure 2 [22].

From this distribution, one can decompose the external field in two parts called axial field and radial field. These two fields are combined to generate the stray magnetic field: the axial field is in a plane that includes the axis of the machine and the radial field is in a plane perpendicular to the axis of the machine. The stray field may give information about the presence of a fault, and sometime its location. Figure 3 shows the different positions for the measurement of the stray field using a coil sensor.

Figure 2. Distribution of leakage flux lines in electrical machine.

Assuming the field lines shown in Figure 3, the sensor will be sensitive to different parts of the field, according to its position:

• in position 1 (Position 1), the coil measures the axial field. In this case the amplitude of the measured flux is maximum at the motor end-shield (near the end-windings), and minimum at the middle of the machine length; this minimum is practically null.
• in position 2 (Position 2), the coil measures the normal component of the radial field. Taking the coil sensor away from the motor end-shields increases the sensitivity to the air gap flux density.
• in position 3 (Position 3), the coil is used to measure the external field called "pure radial" sensitive to the tangential component. In theory, field lines generated by the end-windings cannot cross the section of the coil sensor in this position.

Figure 3. Measure of leakage field in near vicinity of the machine.

2.2. Measurement Sensors

Several types of sensors can be used to measure the external field (magnetoresistive sensor, Hall effect sensor, coil sensor ...).

In the diagnosis method presented in this chapter, a coil sensor is used. The principle of the coil sensor is to measure the flux linkage in a coil located in the vicinity of the electrical machine [23]. The sensor provides an emf equal to the derivative of the magnetic field with respect to the time. Therefore, the measurement is only possible for variable fields. With this type of sensor, the measurement is simple and direct because it does not require associated electronics, unless it is necessary to amplify the induced emf signal [21]. The performance of the measurement depends on the size of the sensor. The main advantage of these sensors is their simple design which allows users to build it in accordance to their needs [24]. The sensor used for experimental tests and its frequency response are represented in Figure 4.

This small sensor will be used for all experimental measurements presented in this work. It has been built for small and low power machines. The coil diameter is about 3.2 cm and it has 1200 turns. Its frequency response shows a resonance at 93 kHz, so, the sensor is suitable to be used for measurement of the field in a frequency range lower than this frequency, what actually corresponds to the frequency range exploited by the proposed diagnosis methods. This type of sensor is used for flux measurements because it is simple, non-invasive, low cost, and it induces an emf with a natural derivative effect that amplifies the low and medium frequencies [25].

Figure 4. Coil sensor: (a) photo; (b) frequency answer.

3. STATOR FAULTS IN AC ELECTRICAL ROTATING MACHINES

The most of stator faults are caused by the degradation of winding insulation. They manifest themselves as a short-circuit between turns, by a short circuit between two phases or by a short-circuit between a phase and the ground. In this section the analytical model is presented for the short-circuit between turns of the same phase. This is a common defect that can occur either at the coil ends or in the notches, causing a decrease of effective winding turns. On the other hand, this fault leads to increase the stator current in the faulty phase and to a slight variation of the amplitude in the others. In the case of induction machines, the inter-turn short-circuit fault also amplifies the current in the rotor circuit [26, 27]. Thermal stress induced by short-circuit can cause the spread of the fault to other coils and can lead to the breakdown of the machine.

The analytical approach developed makes it possible to identify the properties of the air gap flux density waves generated by the fault and proposes a diagnosis method based on those flux density waves.

3.1. Analytical Model

For the analytical study of stator faults in AC motors, a simplified geometry of the machine has been considered (Figure 5) to study the external magnetic field.

Figure 5. Transverse section and considered parameters of simplified machine.

The point M" outside the machine has polar coordinates x, α^s . The stator is composed of a magnetic yoke and an external frame; the gap which can possibly exist between the stator yoke and the external frame is neglected. To simplify, the considered machine presents smooth air gap where the flux density is imposed. Figure 5 also gives the position of the sensor, placed at the level of the point M. Here the ts Rint and ts Rint are internal and external radii of the stator.

For short-circuit model let us consider p 2 pole pair, three-phase stator winding machine given in Figure 6 [25].

Figure 6. Three phase stator windings p 2

Each coil is made up of s m elementary coils of / s s n m turns, where s n is the machine per phase, per pole pair turn number. The coils of the same phase are series connected. As a result the per phase turn number is s pn .

Considering the phase q of the machine, the proposed model supposes that an elementary coil with y turns among / s s n m is short-circuited. In this model, it is considered that the currents are unchanged when y is small compared to s pn . So the model which implements the inter-turn short circuit can preserve the healthy structure and the faulty stator winding are composed by the healthy winding and y short-circuited turns. A s qf i fictitious current circulates in this coil (denoted "additional coil") which is composed of inter-turn short circuit induced current s sc i and the current in opposite direction s q i

$$i{qf}^{s} = i{sc}^{s} - i_{q}^{s} \tag{1}$$

A damaged elementary coil is shown in Figure 7. Here, the model preserves the initial structure and simplifies the fault analysis by separating the fault effects and those created by the healthy coils. For this case, the resulting air-gap flux density b can be expressed by:

$$b^* = b + b_{qsc}^s \tag{2}$$

where K H K H b b , , , , ˆ cos ^s K H K H K H b b K t H (K is the frequency rank and H

the polarity of K H, b ) and s qsc b is obtained as consequence of the damaged coil.

Figure 7. Elementary coil with y short-circuited turns

Taking in consideration the y short-circuited turns, s qsc b is obtained by connecting in series the s qel mmf generated by two healthy coils of the same phase and the s qsc mmf generated by the additional coil and shifted of as presented in Figure 8(a). The mmf generated by the faulty turns is shown in Figure 8 (b).

In this approach, the additional coil generates a unidirectional wave s qsc compared with s qel which is a rotational one. Its polarity H is equal to 1 and creates a dissymmetry in the stator mmf. The rotational mmf waves associated to the unidirectional one evolve per pair in opposite directions. Therefore, s qsc can be written in the stator reference frame s d tied to the stator phase 1 axis:

Figure 8. The mmf components generated by elementary damaged coil created by: (a) two healthy elementary sections; (b) y short-circuited turns

$$\varepsilon{qsc}^{s} = I{f}^{s} \sum{h} A{h}^{s} \cos\left(\omega t - h\alpha^{s} - \varphi_{h}\right)$$ (3)

where s I is the rms value of s qf i , is the angular frequency, h is a not null relative integer, which takes among others, all the values of s h expressed by: 6 1 ^s h k ( k is an integer which varies between to ). The winding coefficient tied to the rank s h is taken in consideration by function s h A . The h is the phase lag which takes into account the place where the short-circuit occurs and the position of the elementary coil concerned by the shortcircuit.

The flux density s qsc b generated by the fault is obtained by multiplying s qsc with air-gap permeance expressed by:

neance $$\lambda$$ expressed by:

$$\lambda = \sum{k^s = -\infty}^{+\infty} \sum{k^r = -\infty}^{+\infty} \Lambda_{k^s k^r} \cos \left[ \left( k^s N^s + k^r N^r \right) p \alpha^s - p k^r N^r \theta \right]$$ (4)

where s r k k is a coefficient which depends on the ranks s k and r k and the air gap geometry, and s N is the stator per pole pair tooth numbers.

3.2. Model for Salient-Pole Synchronous Machine

For a healthy salient-pole synchronous machine, K and H are defined as:

$$K = 1 + 2k^{r}$$

$$H = p\left(h^{sr} + k^{s}N^{s} + 2k^{r}\right)$$ (5)

where sr h includes all the values taken by s h and r h sr s r h h h and 6 1; 1,3,5, , s r h k h .

If a fault exists, the flux density s qsc b can be expressed as:

$$b{qsc}^{s} = \sum{K{sc}, H{sc}} \hat{b}{sc{K{sc}, H{sc}}} \cos\left(K{sc}\omega t - H{sc}\alpha^{s} - \varphi{sc{K{sc}, H{sc}}}\right)$$

$$(6)$$

with

$$K_{sc} = 1 + 2k'^{r}$$

$$H_{sc} = h + pk'^{s}N^{s} + 2pk'^{r}$$ (7)

s k and r k are equivalent to s k and r k . Consequently they vary from to .

After attenuation, it is possible to measure, near the machine, the resulting flux density which traverses the frame and which generates in the signal given by the sensor, the harmonics at K and Ksc angular frequencies. The analysis of K and Ksc values expressed by (5) and (7), allows the following remarks:

• in failure presence, no new frequencies are brought by Ksc . Only the variation of the existing lines amplitudes will be visible.
• The spectral lines with higher sensitivity in presence of the fault correspond to harmonics obtained for 1 r k , ( and 3 ). One of these low frequencies is confused with the fundamental [12].
• the spectral components at angular frequencies 3 , 5 , 7 , which can exist for a healthy machine(supply harmonic, saturation) are superposed to other sensitive lines obtained for 2, 3, r r k k

Considering all those remarks, it is difficult to achieve the analysis of the fault by measuring the harmonics amplitude variation. So the following model takes into consideration the dissymmetry generated by the fault and exploits the properties generated by this dissymmetry.

3.3. Model for Induction Machine

Under healthy conditions the quantities K and H in the case of induction machine can be written in the following form:

$$K = 1 + k^{r} N^{r} (1 - s)$$

$$H = p \left( h^{s} + k^{s} N^{s} + k^{r} N^{r} \right)$$ (8)

The sensitive lines to the fault are dependent by the parameters Ksc and Hsc which can be expressed by:

$$K{sc} = 1 + k'''N''(1-s)$$ , $H{sc} = h + p(k'^s N^s + k'''N'')$ (9)

A numerical application for induction machine is presented in Appendix. The results obtained by compilation show that each harmonic of the external field, at given r k is essentially created by components with polarities the lowest as possible. The analysis will focus on the diagnosis harmonic at 850 Hz, because it is generated by a component with the lowest polarity both for the healthy machine and for the faulty machine 2, 1 H H cc . As the numerical application (see Appendix) shows this harmonic has a high amplitude value and can be more easily measured outside of the motor frame.

3.4. Principle of the Tests

For the new diagnosis method, which exploits the change of external magnetic field in presence of the faults, the measurements are carried out in two diametrically opposed positions: 1 0 ^s (Position 1) and 2 0 ^s (Position 2) in Figure 9.

The new principle consists in measuring at the same time and in both positions of the machine, the variation of specific harmonic components that appear in presence of the faults. Practically, taking into consideration the attenuation of the air-gap flux density in the stator core, the method will be concentrated on the harmonics defined by the lowest values of the parameters , , , , , s r s r s r h h k k k k . The objective is to obtain a value of H smallest as possible.

Figure 9. Sensors position for measurement of external magnetic field.

In this context, the spectral line with magnitude at angular frequency K 1 has to be considered. However, for this component, 1,1 1, ˆ ˆ b b ^H (with H different from 1) and generates an important effect because it corresponds to the fundamental of the air-gap flux density. So for the components obtained for H 1, the sensitivity of 1 ˆ b is lower. 1,1 ˆ b

Therefore, in the analysis of salient-pole synchronous machine, the study will focus on the spectral lines corresponding to K 3 , ( harmonic at 150 Hz). In the case of a four pole machine p 2 , the harmonics are obtained considering the flux density components given by:

• for healthy case: 1, 0 r s h k , 1, 1 r sr k h in (5), which leads to K 3 and H p 3, 6
• for faulty case: 3, 0, 1 s r h k k in (7), which gives K K sc 3 , Hsc 1 .

The flux density component x b measured in the machine vicinity is obtained taking into consideration a term relative to the faulty case (of amplitude 3 ˆ x sc b ) and a component with 3 angular frequency relative to the healthy case. In Position 1, the flux density component is given by:

n by:

$$b_3^{*x(1)} = \hat{b}_3^x \cos(3\omega t - 6\beta_0 + \varphi3^x) + \hat{b}{sc3}^x \cos(3\omega t + \beta0 + \varphi{sc3}^x)$$ (10)

In Position 2, it becomes:

In Position 2, it becomes: $$b_3^{*x(2)} = \hat{b}_3^x \cos(3\omega t - 6\beta_0 + \varphi3^x) - \hat{b}{sc3}^x \cos(3\omega t + \beta0 + \varphi{sc3}^x)$$ (11)

The difference between the flux density measured by the sensors in the position 1 and position 2 is the sign of the component ( 3 ˆ x sc b ) corresponding to the short –circuit case. In the healthy case this component is null ( 3 ˆ 0 x sc b ) and theoretically the both components expressed by (10) and (11) have the same amplitude. This one changes in faulty case because the component 3 ˆ x b sc is added to the healthy component in Pos. 1 and subtracted in Position 2. Practically the amplitudes of the components measured in Position 1 and Position 2 are different even in the healthy machine and that can be the consequence of:

• a non symmetrical alignment of the sensors in Position 1 and Position 2
• a different attenuation of the signal for the both sensors as consequence of the building techniques which are not perfect leading to a geometrical dissymmetry.

In order to solve this difficulty a test in load operating condition has been achieved. The application of this test can be refined as follows:

• for healthy machine, practically the air-gap flux density does not change in the load case ( when the stator current increases) if the supply voltage is maintained constant. So, in the load case, the elements which influence the amplitudes 3 ˆ x b and (2) 3 ˆ x b act in order to influence the evolution of the signals measured in Position 1 and Position 2 in the same direction.
• for faulty machine, the test in load condition gives a variation of the component (corresponding to angular frequency) measured in Position 1 in opposite direction that those measured in Position 2, according to (10) and (11). Considering this different evolution of the sensitive harmonic in the both positions, the originality of the method is to use it as an indicator of the fault. It can be noticed here the interest in using the load tests to determine the lines variation and so, to eliminate any initial knowledge of a healthy state.

In the case of induction machine with N^r =32, the sensitive harmonic which can be easily highlighted is to one at 850 Hz. This line corresponds to a flux density component having a low pole pair number

• for healthy case: 1, 1, 7 r s s k k h lead to K H 17, 2 in (8)
• for faulty case: 1, 1, 15 r s s k k h lead to K H sc sc 17, 1 in (9).

Expression similar to (10) and (11) can be obtained, considering K K sc 17 (instead of 3) and H 2 (instead of 6).

4. ROTOR FAULTS IN AC ELECTRICAL ROTATING MACHINES

This section briefly presents the main phenomena characterizing the rotor faults. It will focus on the influence of a broken bar on the change of the external magnetic field of an induction machine.

An analytical approach requires some simplifying assumptions but it will allow (one) to give the specific harmonics of the fault, including their frequency and polarity. The analytical study will highlight the need to consider the axial field as part of stray flux for which a simplified model will be proposed. A specific signature will be presented, visible in the spectrum of the external magnetic field.

4.1. Types of the Rotor Faults

4.1.1. Bearings Faults

The majority of faults in electrical machines are relative to bearing defects that have mechanical (lubrication, fatigue, etc.) or electrical ones (circulation of leakage currents induced by the converters [28]) causes. Bearing faults lead to more general mechanical effects such as noise and vibration increase or by the appearance of movements of the rotor about the longitudinal axis of the machine [29]. This type of faults induces also variations (oscillations) of the load torque. The ultimate point of the bearing defect is the rotor blocking.

4.1.2. Eccentricity

The eccentricity of an electrical machine is a phenomenon that evolves over time and exists from its manufacture. It passes through various stages of building and assembly tolerances, which induces an off-centering of the rotor relatively to the stator. During operation of the machine, two main causes increase rotor eccentricity. The first is inherent to the mechanical environment in which the machine operates and can impose a radial force on the shaft of the machine. This can lead to bearing wear and amplification of the eccentricity. The second phenomenon that may amplify the eccentricity is itself inherent to the operation of the machine. Actually, the rotor off-centering creates an unbalance in the distribution of radial forces between the stator and the rotor [30, 31] that tends to increase the eccentricity. The ultimate point of eccentricity is the friction between stator and rotor, that leads to a fast destruction of the machine.

4.1.3. Broken Bar

Inter-turn short circuit fault may also occur in a wound rotor. For squirrel cage induction machines, broken rotor bars and the breaking of short-circuit rings are the most common rotor failures (Figure 10).

The broken bar or the breaking of the short-circuit ring may be due, for example, to mechanical overload (frequent starts, etc.), to excessive local heating or to a manufacturing defect (air bubble or bad welds). This generates oscillations of currents and electromagnetic torque even more apparent when the inertia of the system has a high value (constant velocity) [32, 33]. If the system inertia is lower, oscillations appear on the angular speed and the amplitude of the stator currents. The occurrence of a broken bar does not cause an immediate breakage of the machine, because the current crossing the broken bar flows over the adjacent bars. Therefore, these bars are overloaded and the thermal and electromechanical constraints can lead to their breaking and so on until the breakage of a sufficient large number of bars that causes the shut-down of the machine.

Figure 10. Fault of squirrel rotor cage: (a) broken bar; (b) breaking of the short-circuiting ring.

4.2. Rotor Broken Bar

4.2.1. Influence on the Speed Variation

The presentation of the phenomenon will consider only the flux density component of p pole pairs. Under healthy operating conditions, the balanced three-phase current system creates a fundamental clockwise rotating magnetic field in the air-gap. This field induces a current in the rotor bars with a frequency proportional to the rotor slip s . The rotor bars then generate a clockwise field rotating at s s in a rotor referential (Figure 11(a)). For a symmetrical motor (healthy machine), there is not any anticlockwise field in the air-gap [8, 34].

Under rotor faults, in addition to the clockwise field an anticlockwise field appears, created by the unbalance of rotor bar currents and which rotates at an angular velocity ^s s shown in Figure 11(b). It is possible to associate with this anticlockwise field a negativesequence current system with amplitude rn I , which is added to the positive-sequence current system with amplitude rp I .

Figure 11. Operation schema for: (a) healthy motor; (b) faulty motor.

Figure 11(b) shows the phenomenon due to the appearance of rotor failure. A flux density component rotating at fault appears in the stator referential:

$$\Omega_{fault} = \Omega_r - s\Omega_s = \Omega_s \left( 1 - 2s \right) \tag{12}$$

which leads to the well-known frequency in literature fault f , appearing in the radial external magnetic field and in the line current:

$$f_{fault} = (1 - 2s)f \tag{13}$$

The interaction of the rotor currents with the clockwise stator field produces the electromagnetic torque. In case of broken rotor bar fault, the total torque can be expressed as follows:

Use of the External Magnetic Field for Supervision ... $$395$$

$$C{e_tot} = C{e_srp} + C_{e_srm} = \pi \frac{RLe_m}{\mu_0} k_s I_s \left[ kr I{rp} \sin(\varphi_{Bs} + \varphi{B_{rp}}) + k_r Im \sin(2s\omega t + \varphi{Bs} - \varphi{B_m}) \right]$$ (14)

where Ce srp _ represents the mean torque due to the interaction of the stator and the rotor fields in the clockwise direction, Ce srn _ is the pulsating torque of pulsation 2s due to the interaction of the stator clockwise field with the anticlockwise rotor field; Bs , B^s and Brn are respectively the phases relative to the stator phase one current of the stator flux density, the rotor flux densities in the clockwise direction and in the anticlockwise direction; s k is a coefficient that depends on the stator winding arrangement and r k is a coefficient that depends on the rotor cage constitution.

The torque oscillation at frequency 2sf , also presented in [8], generates a speed oscillation at this frequency whose amplitude depends on the rotor asymmetry, the motor load level [34, 35], and the total moment of inertia J . The speed is calculated with the relation:

$$J\frac{d\Omegar}{dt} = C{e_tot} - Cr = C{e_srn}$$ (15)

where Cr is the load torque:

$$\Omegar = \frac{1}{J} \int C{e_srn} dt + \Omega_0 \tag{16}$$

By calculating the integral (11) one obtains:

$$\Omega_r = \Omega_0 + \Delta\Omega_r \cdot \cos(2s\omega t + \varphi) \tag{17}$$

where 0 1 2 m r s rn rn RLe k Ik I sJ is the magnitude of the speed oscillation, ⁰ is the

mean value of r which changes, depending on the number of broken bars at given load torque, and . B B s rn

It is well known that the speed oscillation also generates an additional air-gap flux density component at 1 2 s f .

4.2.2. Influence on the Axial Flux

For the modelling of the axial field, it is necessary to consider the field lines existing at the machine extremities. The distribution of these ones strongly depends on the geometry and the mechanical components that exist at the machine extremities. As mentioned in [32], the total axial flux AT outside the induction machine with an eccentricity fault can be written in the following form:

$$\Phi_{AT} = \hat{\Phi}\cos(\omega t - p\beta) \tag{18}$$

with ^{0 r} t , indicates the position of the minimum air-gap thickness. Under broken rotor bar fault, a speed variation appears and will influence the external axial flux. In this case (18) can be written as follows:

$$\Phi_{AT} = \hat{\Phi}\cos\left[\omega t - p(\beta_0 + \Omega_r t)\right]$$ (19)

Developing the argument of the cosine, using (17) one obtains:

$$\Phi_{AT} = \hat{\Phi}\cos\left[s\omega t - \Delta\Omega_r pt\cos(2s\omega t + \varphi) - p\beta_0\right]$$ (20)

A frequency modulation appears in AT . The time variation of (20) for ˆ 1 Wb , 1 rad/s ^r and s 2.5% is presented in Figure 12(a). A direct analysis of the signal is difficult; hence the change in the frequency domain is performed by Fast Fourier Transform (FFT). The spectrum obtained is given in Figure 12(b).

Figure 12. Simulation of axial flux variation: (a) axial flux signal; (b) FFT for the axial flux.

In the spectrum it can be observed the appearance of the harmonic 3sf caused by the speed variation and sf due to dynamic eccentricity on the axial flux. One can notice that a small speed variation can produce an important component at frequency 3sf . With the considered speed variation that corresponds approximately to 2 broken bars; the component 3sf is higher than the initial component sf .

5. EXPERIMENTAL TESTS

5.1. Achievements of the Tests

The application of the proposed method for stator fault detection is presented considering a salient-pole synchronous machine and an induction machine. For rotor faults an experimental bench using two identically squirrel-cage induction motors (one healthy and another with a broken bar) is used. The principle of measurement is shown in Figure 13.

Figure 13. Measurement and treatment of emf signal.

5.2. Stator Tests

5.2.1. Inter-Turn Short Circuit in Synchronous Machines

The application of the proposed method is presented considering a test carried out on a salient-pole synchronous machine characterized by: 7.5 kW, 50 Hz, p 2 , 230/400V, N 5 =18, grid connected. This machine has been rewound so that the terminals of the different stator elementary sections are extracted from the winding and are brought back to a connector block which is fixed above the machine as indicated in Figure 14. So it is possible to produce short-circuits between elementary coils.

The tests consist in measuring and analyse the emf delivered by the sensors in order to validate the suggested procedure when an elementary coil is short-circuited (what corresponds to 16.6% of the whole winding of a phase). In order to avoid total damage of the winding, the short-circuit current is limited by an external resistance R. Verification has also been done in order to make sure that the currents in the wires going up from the machine towards the connector block do not disturb the measurements. The Position 1 and Position 2 correspond respectively to 0 and ⁰ , ⁰ is chosen in the axis of the tested phase. The interest is focused on the harmonic at 150 Hz. Figures 15 a and b show the variations of the harmonic at 150 Hz relative to the healthy machine (motor) when the state (marked with arrow) varies from no-load to load.

Figure 14. Modified salient-pole synchronous machine and the sensors" positions.

Figure 15. Harmonic components at 150 Hz for a healthy salient-pole synchronous machine (motor) at no-load and load operations.

It can be observed that the magnitudes vary in the same direction (decrease) for the two positions. Under faulty conditions at full load (Figure 16) it can be observed that the magnitudes in both positions vary in the opposite direction. This indicates the failure presence and confirms the proposed method.

Other tests are carried out on the machine which operates as generator. The corresponding results are shown in Figures 17 and 18. One can notice the same phenomenon: under healthy case the magnitudes varies in the same direction, but in the faulty case they vary in opposite direction.

Figure 16. Harmonic components at 150 Hz for faulty salient-pole synchronous machine (motor) at noload and load operations.

Figure 17. Harmonic components at 150 Hz for a healthy salient-pole synchronous machine (generator) at no-load and load operations.

Figure 18. Harmonic components at 150 Hz for faulty salient-pole synchronous machine (generator) at no-load and load operations.

Figure 19. Harmonic components at 150 Hz for a healthy permanent magnet synchronous motor at noload and load operations.

Figure 20. Harmonic components at 150 Hz for a healthy permanent magnet synchronous motor at noload and load operations.

The next test has been performed on a four pole, 4.4 kW, 50 Hz, 3×380 V permanent magnet synchronous motor. The results obtained under healthy and faulty conditions are given in Figures 19 and 20. It can be observed that under healthy case the harmonic magnitudes at 150 Hz vary in the same direction, but when the inter-turn short circuit exists in the stator winding these harmonic magnitudes vary in different direction.

5.2.2. Inter-Turn Short Circuit in Induction Machines

In order to generalize the proposed method, the tests have been also performed on a p 2, 50 Hz, 11 kW, 380 / 660V, 24 ^s N and 16 ^r N squirrel cage induction motor (Figure 21).

This machine, connected to the grid, has been rewound so that the terminals of the different stator elementary sections are extracted from the winding and are brought back to a connector block which is fixed above the machine. So it is possible to produce short-circuits between elementary coils. Position 1 and Position 2 correspond respectively to 0 and

⁰ , ⁰ is chosen "arbitrarily". One will be interested to the spectral line at 850Hz 1 r k . The results obtained by considering the harmonic of rank K K sc 17 ( 850Hz at no-load and load), are presented in Figures 22 and 23.

Figure 21. Modified squirrel cage induction motor and sensors" positions

The phenomena are similar than those obtained in the case of synchronous machines:

• under healthy condition the lines relative to K K sc 17 evolve in the same way with the load (Figure 22),
• under faulty condition - these lines evolve in opposite way with the load (Figure 23).

Figure 22. Harmonic components relative to K K sc 17 for healthy induction motor at no-load and load operations.

Figures 23. Harmonic components relative to K K sc 17 for faulty induction motor at no-load and load operations.

This method is thus applicable to salient-pole synchronous machines, permanent magnet synchronous motors and induction motors often used in electrical traction applications and that confirm the universal quality of the proposed diagnosis method.

Figure 24. Frequency spectrum of the external magnetic field (Position 1) for healthy squirrel cage induction motor (s = 5%).

Figure 25. Frequency spectrum of the external magnetic field (Position 1) squirrel cage induction motor with one broken rotor bar (s = 5%).

5.3. Broken Rotor Bar in Induction Machine

The tests are performed on a 4 poles, 4 kW / 50 Hz, 3×380 V squirrel cage induction motor under healthy and faulty (broken rotor bar) conditions. The results obtained at full load are shown in Figures 24 and 25.

If the rotor has a broken bar, the frequency component 3sf is more sensitive than other components and it denotes the speed variation at frequency 2sf , caused by the fault. Figure 25 shows that the amplitude of this harmonic is higher in comparison with the component sf , which is an indicator of eccentricity.

Also, one can observe in the case of healthy machine, the presence of frequency components sf and 3sf with low amplitude. It is due the dissymmetry that exists even for a healthy machine, inherent to the building tolerances.

CONCLUSION

The chapter presents a simple method to insure the supervision of AC machines used in electrical traction. This supervision can detect stator and rotor faults using original signatures presented in the chapter. Under stator winding faults the signature concerns the harmonic at 150 Hz for synchronous machines and a slotting harmonic for induction machines. This diagnosis method does not require the knowledge of the healthy signature. It is based on comparison between the states at no-load and load operations.

The indicators of rotor faults are the low frequency components of the flux spectrum at sf and 3sf frequencies. The diagnosis method consists in checking the magnitude of the component 3sf generated by the speed variation that can be compared to this of the component sf .

The advantages of the proposed non-invasive methods compared to other existing methods are their simplicity to implement, to interpret and a low cost set-up. The processed emf signal, delivered by a simple coil sensor, is able to indicate the presence of the fault. Moreover, the low frequency analysis is not disturbed by other sources of magnetic field presented in vicinity. Finally, the experimental tests realized on the different AC machines confirm the validation of used models and proposed methods for stator and rotor fault detection. The simulation and experimental analyses confirm the models and approaches used.

APPENDIX

For a numerical application it is considered an induction machine with p 2 , 24 ^s N , ^{16 r N} , 0.0825 ts Rint , 0.121 ts ^Rext and a short circuit in the phase 1. Here 12.5% of the phase coils are short-circuited and the short-circuit current is 3 s s sc i i . The values that can take the coefficients that contribute to the description of predominant components |are given by ; ; r k ; ; s k ; ; r k 6 1; ^s h k 1,3,5, , r h .

If one consider the first rotor slotting harmonic obtained for 1 r k , for a healthy induction machine with a slip null and considering 1 r k , it comes 16 r r k N and K 15, f 750 Hz .

In order to have s s r r k N k N the lowest possible (here equal to -6), we must choose 1 s k . Also, we find that the dominant mode is obtained for 5 s h , H 6 (in absolute terms and because p 2 ). For 1, ^r k we obtain 16 r r k N and K 17, f 850 Hz .

As s s r r k N k N must be the lowest possible (here equal to -2), we must have 1 s k . Also, we find that the dominant mode is obtained for 7 s h , H 2 (in absolute terms).

Table 1. Components generated by the healthy IM

r k	s k	s h	K	H	K (Hz)	ts  H	ˆ ˆ x x K H b / b , 11
-1	1	-5	-15	6	-750	5.09e-05	2.99e-05
-1	1	-11	-15	-6	-750	5.09e-05	8.16e-06
1	-1	7	17	-2	850	8.16e-04	2.62e-04
1	-1	13	17	10	850	3.97e-06	5.32e-07
-2	1	7	-31	-2	-1550	8.16e-04	-2.49e-04
-2	2	-17	-31	-2	-1550	8.16e-04	5.66e-05
2	-1	-5	33	6	1650	5.09e-05	-2.85e-05
2	-1	-11	33	-6	1650	5.09e-05	-7.76e-06
2	-2	13	33	-6	1650	5.09e-05	3.81e-06
2	-2	19	33	6	1650	5.09e-05	4.04e-06

Table 2. Components generated by the faulty coils

r k	s k	h	Kcc	Hcc	Kcc (Hz)	ts  H cc	ˆ ˆ x x b / b 11 ccK H cc cc ,
-1	1	-10	-15	6	-750	5.09e-05	1.82e-05
-1	1	-13	-15	3	-750	3.76e-04	-7.31e-05
-1	1	-14	-15	2	-750	8.16e-04	-2.07e-04
-1	1	-15	-15	1	-750	0.0023	-3.78e-04
-1	1	-18	-15	-2	-750	8.16e-04	1.60e-04
-1	1	-19	-15	-3	-750	3.76e-04	4.94e-05
-1	1	-21	-15	-5	-750	9.72e-05	-1.15e-05
-1	1	-22	-15	-6	-750	5.09e-05	-8.09e-06
1	-1	10	17	-6	850	5.09e-05	1.82e-05
1	-1	11	17	-5	850	9.72e-05	2.23e-05
1	-1	14	17	-2	850	8.16e-04	-2.07e-04
1	-1	15	17	-1	850	0.0023	-3.78e-04
1	-1	17	17	1	850	0.0023	3.33e-04
1	-1	18	17	2	850	8.16e-04	1.60e-04
1	-1	21	17	5	850	9.72e-05	-1.15e-05
1	-1	22	17	6	850	5.09e-05	-8.09e-06
-2	1	17	-31	1	-1550	0.0023	-3.16e-04
-2	1	21	-31	5	-1550	9.72e-05	1.09e-05
-2	1	22	-31	6	-1550	5.09e-05	7.70e-06
-2	2	-26	-31	6	-1550	5.09e-05	3.78e-06
2	-1	-10	33	6	1650	5.09e-05	-1.73e-05
2	-1	-13	33	3	1650	3.76e-04	6.96e-05
2	-1	-14	33	2	1650	8.16e-04	1.97e-04
2	-1	-15	33	1	1650	0.0023	3.60e-04
2	-1	-18	33	-2	1650	8.16e-04	-1.52e-04
2	-2	26	33	-6	1650	5.09e-05	3.78e-06

The other components defined for values , , , , , s r s r s r k k k k h h of the lowest possible for the healthy machine and for the faulty coils are given respectively in Table 1 and Table 2 with the amplitude in s ext x R given relatively to the fundamental.

REFERENCES

• [1] Pusca, R.; Contribution to the Study of the Driving of a Mobile Platform to Test Electrical Power Train Components - Design of a Simulator and Real Time System. Doctoral Thesis, Science for Engineering, University of Franche-Compte, Belfort, France, 2002.
• [2] Bellini, A.; Filippetti, F; Tassoni, C; Capolino, G.A.; Advances in Diagnostic Techniques for Induction Machines. IEEE Trans. on Industrial Electronics, vol. 55, no. 12, pp. 4109-4125, 2008.
• [3] Thomson, W.T.; Fenger, M.; Current Signature Analysis to Detect Induction Motor Faults. IEEE Industry Applications Magazine, vol. 7, no. 4, pp. 26-34, 2001.
• [4] Kliman, G.B.; Koegl, R.A.; Stein, J.; Endicott, R.D.; Madden, M.W.; Noninvasive Detection of Broken Rotor Bars in Operating Induction Motors. IEEE Trans. on Energy Conversion, vol. 3, no. 4, pp. 873-879, 1988.
• [5] Kliman, G.B.; Stein, J.; Methods of Motor Current Signature Analysis. Electrical Power Component System, vol. 20, no. 5, pp. 463-474, 1992.
• [6] Khezzar, A.; Kaikaa, M.Y.; El-Kamel Oumaamar, M.; Boucherma, M.; Razik, H.; On the Use of Slot Harmonics as a Potential Indicator of Rotor Bar Breakage in the Induction Machine. IEEE Trans. on Industrial Electronics, vol. 56, no. 11, pp. 4592- 4605, 2009.
• [7] Romary, R.; Corton, R.; Thailly, D.; Brudny, J.F.; Induction Machine Fault Diagnosis Using an External Radial Flux Sensor. EPJ. Applied Physics, vol. 32, no. 2, pp. 125- 132, 2005.
• [8] Filippetti, F.; Franceschini, G.; Tassoni, C.; Vas, P.; AI Techniques in Induction Machines Diagnosis Including the Speed Ripple Effect. IEEE Trans. on Industry Applications, vol. 34, no. 1, pp. 98-108, 1998.
• [9] Arthur N.; Penman, J.; Induction Machine Condition Monitoring with Higher Order Spectra. IEEE Trans. on Industrial Electronics, vol. 47, no. 5, pp. 1031-1041, 2000.
• [10] Menacer, A.; Nat-Said, M.S.; Benakcha, A.H.; Drid S.; Stator Current Analysis of Incipient Fault Into Induction Machine Rotor Bars. Journal of Electrical Engineering, vol. 55, pp. 122-130, 2004.
• [11] Li, W.D.; Mechefske, C.K.; Detection of Induction Motor Faults: A Comparison of Stator Current, Vibration and Acoustic Methods. Journal of Vibration and Control, vol. 12, pp. 165-188, 2006.
• [12] Thailly, D.; Romary, R.; Lecointe, J.Ph.; Brudny, J.F.; Suau, P.; Synchronous Machine Diagnosis Using an External Magnetic Flux Sensor*.17th International Conference on Electrical Machines (ICEM 2006),*Chania, Crete, Greece, pp. 462-468, 2006.

• [13] Yazidi, A.; Henao, H.; Capolino, G.A; Artioli, M.; Filippetti, F.; Casadei, D.; Flux Signature Analysis: an Alternative Method for the Fault Diagnosis of Induction Machines. IEEE PowerTech Conference, St. Petersburg, Russia, pp. 1-6, 2005.

• [14] Ceban, A.; Pusca, R.; Romary, R.; Eccentricity and Broken Rotor Bars Faults Effects on the External Axial Field. International Conference on Electrical Machines (ICEM 2010), Rome, Italy, pp. 1-6, 2010.

• [15] Tavner, P.; Ran, L.; Penman, J.; Sedding, H.; Condition Monitoring of Rotating Electrical Machines. 2nd Ed., Stevenage, IET, UK, 2008.

• [16] Henao, H.; Demian, C.; Capolino, G.A.; A Frequency-Domain Detection of Stator Winding Faults in Induction Machines Using an External Flux Sensor. IEEE Transactions on Industry Applications, vol. 39, no. 10, pp. 1272-1279, 2003.

• [17] Kokko, V.; Condition Monitoring of Squirrel-Cage Motors by Axial Magnetic Flux Measurements. Doctoral Thesis, Department of Electrical Engineering, University of Oulu, Oulu, Finland, 2003.

• [18] Negrea, M.D.; Electromagnetic Flux Monitoring for Detecting Faults in Electrical Machines. Doctoral Thesis, Electrical and Communications Engineering, Helsinki University of Technology, Helsinki, Finland, 2006.

• [19] Werynski, P.; Dielectric Aging and in Situ Monitoring of Electrical Machines. Doctoral Thesis, Laboratory of Electrotechnical Systems and Environment (RSAA), Artois University, Bethune, France, 2006.

• [20] Zidat, F.; Designing a Monitoring Unit Performance Eco-Efficiency of Rotating Electrical Machines with Alternating Current. Doctoral Thesis, Laboratory of Electrotechnical Systems and Environment (RSAA), Artois University, Bethune, France, 2012.

• [21] Thailly, D.; Field Study of Dispersion of Alternating Current Machines: Application to Diagnosis. Doctoral Thesis, Laboratory of Electrotechnical Systems and Environment (RSAA), Artois University, Bethune, France, 2007.

• [22] Thomson, W.T.; A Review of Online Condition Monitoring Techniques for Three-Phase Squirrel Cage Induction Motors - Past, Present and Future. IEEE International Symposium on Diagnostics for Electrical Machines, Power Electronics and Drives (SDEMPED'99), Gijon, Spain, pp. 3-17, 1999.

• [23] Tumanski, S.; Induction Coil Sensors A Review. Measurement Science and Technology, vol. 18, pp. 31-46, 2007.

• [24] Ceban, A.; Pusca, R.; Romary, R.; Supervision of Electric Traction Motors by External Magnetic Field Analysis. International Conference on Electronics, Computers and Artificial Intelligence (ECAI 2011), Pitesti, Romania, pp. 1-10, 2011.

• [25] Pusca, R.; Romary, R.; Ceban, A.; Brudny, J.F.; An Online Universal Diagnosis Procedure Using Two ExternalFluxSensorsAppliedtoACElectrical Rotating Machines. Sensors, vol. 10, pp. 10448-10466, 2011.

• [26] Bonnett A.H.; SoukupG.C.; Cause and Analysis of Stator and Rotor Failures in Three-Phase Squirrel Cage Induction Motors. IEEE Transactions on Industry Applications, vol. 28, pp. 921-937, 1992.

• [27] Nandi, S.; Toliyat, H.A.; Xiaodong L.; Condition Monitoring and Fault Diagnosis of Electrical Motors: A Review. IEEE Transactions on Energy Conversion, vol. 20, pp. 719-729, 2005.

• [28] Busse, D.F.; Erdman J.M.; Kerkman, R.J.; Schlegel, D.W.; Skibinski, G.L.; The Effects of PWM Voltage Source Inverters on the Mechanical Performance of Rolling Bearings. IEEE Transactions on Industry Applications, vol. 33, pp. 567-576, 1997.

• [29] Vaseghi, B.; Contribution to the Study of Electrical Machines in the Presence of between Failure Spires. Doctoral Thesis, INPL-GREEN University of Henri Poincare - Nancy I, Nancy, France, 2009.

• [30] DorrellD.G.; Smith A.C.; Calculation of UMP in Induction Motors with Series or Parallel Winding Connections. IEEE Transactions on Energy Conversion, vol. 9, pp. 304-310, 1994.

• [31] Dorrell, D.G.; Thomson, W.T.; Roach, S.; Analysis of Air Gap Flux, Current, and Vibration Signals as a Function of the Combination of Static and Dynamic Air Gap Eccentricity in 3-Phase Induction Motors. IEEE Transactions on Industry Applications, vol. 33, pp. 24-34, 1997.

• [32] Ceban, A.; Pusca, R.; Romary, R.; Study of Rotor Faults in Induction Motors Using External Magnetic Field Analysis. IEEE Transactions on Industrial Electronics, vol. 593, pp. 2082-2092, 2001.

• [33] Bossio, G.R.; De Angelo, C.H.; BossioJ.M.; Pezzani, C.M.; Garcia, G.O.; Separating Broken Rotor Bars and Load Oscillations on IM Fault Diagnosis Through the Instantaneous Active and Reactive Currents. IEEE Transactions on Industrial Electronics, vol. 56, pp. 4571-4580, 2009.

• [34] Didier, G.; Razik, H.; Rezzoug A.; An induction Motor Model Including the First Space Harmonics for Broken Rotor Bar Diagnosis. European Transaction of Electrical Power, vol. 15, pp. 229-243, 2005.

• [35] Bellini, A.; Filippetti, F.; Franceschini, G.; Tassoni, C.; Quantitative Evaluation of Induction Motor Broken Bars by Means of Electrical Signature Analysis. IEEE Trans. on Industry Applications, vol. 37, no. 5, pp. 1248-1255, 2011.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1

Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 15

BATTERIES FOR ELECTRICAL VEHICLES: A REVIEW

Bogdan Enache*¹ , Emilian Lefter †¹ and Costin Cepisca‡²

1 Faculty of Electronics, Communications and Computers, University of Pitesti, Pitesti, Romania Faculty of Electrical Engineering, University Politehnica Bucharest, Bucharest, Romania

ABSTRACT

Batteries represent one of the key elements of electric and hybrid vehicles and they constitute an important factor in the development of the automotive industry in an environment-friendly direction.

Nowadays, there is a wide range of batteries available on the market: lead-acid, nickel, lithium, sodium, metal-air, etc., but none of these technologies managed to overcome the gasoline in the terms of specific energy, specific power, density of energy and power density.

The new developments in the field of nanotechnology and fuel cells aim at offering solutions for the creation of batteries able to provide autonomy comparable to that offered by gasoline, but a lot of the present results are only applicable to experimental research.

Keywords: Batteries, Electric Vehicles, Hybrid Vehicles

^†* E-mail: b0gdan.enache@yahoo.com.

^† E-mail: emilianlefter@yahoo.com.

E-mail: costin@wing.ro.

1.INTRODUCTION

By the end of the 19th century, the electric vehicles (EV) seemed to have a bright future in front of them. They represented the preferred choice of the society"s elite due to the fact that they were silent and they didn"t require the starter"s manual actuation at start-up. This can be also noticed from the analysis of the USA sales statistics [1]: in 1899, 1575 electric vehicles were sold as compared to only 936 internal combustion engine vehicles (ICE). The performances of the two types of automobiles were comparable. Thus, in 1899 the Belgian Jenatzy reached a record speed of 100 km/h with his electric automobile named «Jamais Contente».

Within a few years, with all the improvements brought to it, the electric vehicle begins to lose its place on the market. In 1910 the number of the sold electric automobiles increased significantly, to almost 4000, but the sales of the internal combustion engine automobiles exceeded to 120000 pieces, namely 30 times higher number [1]. This was due, especially, to certain problems related to the vehicle"s battery: a long recharging period, a reduced reliability, a limited autonomy, a high cost price, etc.

One hundred years later, even though batteries technology made a considerable leap forward, a great deal of the electric vehicles issues remained unsolved. But new elements started to bring our attention back to them: the repeated petrol crises and the environment pollution. The noxious emissions contributed to the deterioration of the environment quality, to the development of the greenhouse effect, to climatic modifications, all leading to the negative change of people"s life conditions. Suzuki Foundation shows in its studies that diesel engines are responsible for numerous diseases such as: asthma, pneumonia, other respiratory diseases, cancer, heart diseases [2].

In this context of environmental care completed with the diminution of petrol reserves, a series of organizations appeared, which have their sole purpose to reduce the automobiles impact on the environment and on people"s health. This is carried out by accelerating the development process of batteries, for the electric and hybrid vehicles. Among these organizations, the best known are: USABC and PNGV.

USABC is a consortium made up of the Department of Energy (DOE) of the United States, Daimler-Chrysler, Ford, General Motors and the Electric Energy Research Institute of the USA (EPRI). USABC determined the required performances, so that the batteries used within EV will be compatible with ICE engines from the performance point of view. Among these we can mention:

• High density energy, obtained by an single charge in order to provide increased autonomy and high currents during the acceleration period;
• An increased life span with maintenance free;
• Nonpolluting materials in batteries structure.

The other mentioned organism, PNGV, was a research cooperative program in the USA, initiated in 1993 and completed in 2003 and was composed by federal agencies, national laboratories, universities and motor vehicles manufacturers. The program was completed with the elaboration of three diesel-electric car prototypes: General Motors - Precept, Ford -

Prodigy, Chrysler -ESX3, which complied with the imposed requirements of providing a consumption of 3.78l/130 km/ (80 miles/gallon).

Both USABC and PNGV made up a series of tests, grouped in 2 manuals, which are used for the batteries performances evaluation.

2. EXISTING ELECTRIC VEHICLES

Under the influence of the new 21st century technologies, the electric vehicles started to appear again on the automobile market. At present, two categories of electric vehicles can be found on the streets in the entire world: pure electric and hybrid electric.

2.1. Electric Vehicles

The Electric Vehicles (EV) use an electric engine and a battery as its main power supply for the engine. These cars are also called zero-emission vehicles (ZEV).

Before use, these vehicles require the charging of the battery from an external source of energy. This charging must be performed as fast as possible and it must provide high autonomy. The corroboration of the two conditions leads to a series of difficulties because the ensuring high autonomy implies a great energy density which cannot be obtained in a short charging time. This is why the batteries meant for the ZEVs are permanently searching for the right balance between the energy density and the charging time.

Transposing the problem numerically, the quantity of energy required for a medium-type ZEV is of 32 kWh and for a 5 minutes charge the battery requires a power source which can supply 384 kW, a power too big to be practically. Due to this reason, the medium charging time, for this type of vehicle, is between 6-8 hours. A comparison between the different existing technologies, in relation to a 250kg battery, is presented in Table 1.

Technology	Pb-acid	NiCd	NiMH	Li-Ion
Vehicle mass [kg]	1200	1200	1200	1200
Battery mass [kg]	250	250	250	250
Energy density [Wh/kg]	33	45	70	120
Supplied energy [kWh]	6.4	8.8	13.0	23.4
Power density [W/kg]	75	120	170	370
Battery supplied power [kW]	15	24	33	72
Autonomy for 120 Wh/ton/km [Km]	53	73	114	195

Table 1. Performances for different batteries technologies [1]

2.2. Hybrid Electric Vehicles

A hybrid electric vehicle uses an internal combustion engine and an electric engine, and needs two different power sources, one for each engine. The goal of the hybrid vehicle is to combine the autonomy of the internal combustion engine with the pollution-free and economy of the electric one in a single vehicle [3].

The electric engine and implicitly the battery are used especially at start-up, but they assist the internal combustion engine whenever it requires additional power. This operating regime is characterized by high currents, and so the battery used must have a high power density.

There are several hybrid cars architectures: Micro-Hybrids - "Stop & Start", Soft-Hybrids -"Stop & Go", Mild Hybrids, Full Hybrids - Power Assist, Plug-in Hybrids -PHEV.

2.2.1. Micro-Hybrids - "Stop and Start"

The fundamental characteristic of these vehicles is that they maintain the structure of the internal combustion engines unchanged. They usually append to the existing engine a starteralternator system that is powered from a battery-inverter device. The starter-alternator system is only used at start-up.

At these vehicles the recovery of energy at braking doesn"t take place. During idling the internal combustion engine is stopped and only the starter system in powered up. This operating mode produces a fuel economy of 15% [1].

The micro-hybrid vehicles are equipped with lead-acid batteries, usually called start-stop batteries, which are maintained in 100% SOC.

2.2.2. Soft-Hybrids - "Stop and Go"

These vehicles borrow a lot from the micro-hybrids vehicles characteristics but, as compared to these, they use regenerative braking. Within the braking process the electric car, goes into a generator regime and the energy produced by it is transferred to the battery or to certain ultra-cap.

The start-up of the soft-hybrids vehicles is performed in an electric mode and afterwards the thermal engine is started, thus producing a maximum fuel economy of 25% [4].

The battery mounted on these vehicles is lead-acid or Ni-Cd because they have to endure high current peaks. The operating mode of the battery is between 100% and 80% of SOC.

2.2.3. Mild Hybrids

The mild hybrids vehicles depart from the classic vehicles with internal combustion engines and they propose a new topology characterized by the incorporation of the electric engine in the drive.

The start-up for this type of vehicle is performed in the electric mode and this mode is maintained until the speed of 20 km/h is reached, after which the thermal engine interferes. While driving, the electric engine brings its contribution only during the acceleration periods, thus reducing the fuel consumption.

The battery mounted on these vehicles is Ni-MH or Li-Ion because they have to endure high operating currents and profound discharges.

2.2.4. Full Hybrids - Power Assist

These vehicles use simultaneously the electric engine and the internal combustion engine, although they have the possibility, for a certain number of kilometers (approximately 30), to operate only in electric mode. Due to this reason, the main characteristic of the batteries is high energy density and the preferred choice of technology is Ni-MH.

The battery that equips these vehicles is either in a discharge process, or in a charging one and it usually operates at 50% SOC in order to be able to endure the high currents at discharge and the energy recovery at braking.

2.2.5. Plug-in Hybrids -PHEV

These vehicles combine the characteristics of the electric vehicles - their battery can be charged from an external source and hybrid ones– the propulsion system has a thermal engine and an electric one.

The vehicle can function in the electric mode until the battery reaches a certain discharge level (around 30% SOC), which usually is translated into 50 km autonomy, afterwards the thermal engine interferes. During the electric engine functioning period, in case a higher speed is required, the thermal engine supplies the additional power.

The PHEV vehicles battery is charges from an external source when the vehicle is not used, and from this reason, both the power density and the energy density are important battery parameters.

The parameters that a battery should provide, in order to be used on one of the existing electric vehicles, are synthesized in Table 2.

Nr. Crt.	Type	Power [kW]	Energy [kWh]	Voltage [V]
1	Micro Hybrids	2.5-5	0.5	12-36
2	Soft Hybrids	3-6	0.75	12-48
3	Mild Hybrids	15-20	1	120-160
4	Full Hybrids	30-50	2-3	200-350
5	Plug-in Hybrids	30-100	5-15	200-350
6	Electric Vehicle	35-75	25-40	200-350

Table 2. Electric vehicles requirements [1]

3. BATTERY TECHNOLOGY USED FOR ELECTRIC VEHICLES

The batteries used in electric vehicles are devices that transform chemical energy into electric energy, at discharge, respectively the electric energy into chemical energy at charge.

A battery is made up of several interconnected cells, where a cell is an independent unit that possesses all the chemical properties of the battery.

Usually, a cell is made up of: electrodes (anode, cathode), electrolyte, separators and recipient. The functioning of a battery relies on the oxidation and reduction reactions that take place at the anode and cathode, and on the flow of the electrons released by oxidation into the battery exterior circuit [5].

If at the anode the active substance is A, the oxidation reaction will be:

$$A \to A^{z+} + ze_{-} \tag{1}$$

Namely, each molecule of substance A sends in the electrolyte the positive ions z A (cation radical), and z electrons remain on the electrode e , z being the cation radical valence.

At cathode, the active substance C accepts the electrons directed through the exterior circuit (the reaction is called reduction), becoming a negative ion (anion):

$$C + ze_{-} \to C^{z^{-}} \tag{2}$$

The anions z C sent in the electrolyte are combined with the cautions z A brought in this area due to the diffusion forces, producing the element AC:

$$A^{z+} + C^{z-} \to AC \tag{3}$$

As it can be seen, the process of electric energy generating can be resumed as follows: pairs of charges with contrary names are generated at the anode: electrons and ions. The ions are sent in the electrolyte, which is a sort of ionic diode, reaching the cathode. The electrons can circulate only through the exterior circuit, if it is connected. Through the exterior circuit the electrons reach the cathode where the described phenomena take place.

3.1. Lead-Acid Batteries

The lead-acid batteries are still the most used batteries in the automotive industry. They are used especially in the vehicles with internal combustion engines having the role of electric source during idle and cranking.

The power generation is based on the use of sponge lead as active material at the anode and on the use of lead dioxide (PbO2) as active material at the cathode. The electrodes are immersed in an electrolyte made up of a diluted solution of sulphuric acid.

At discharge, the sulphuric acid combines with the sponge lead and with the lead dioxide thus forming the lead sulphate (PbSO4) and water, at charge the process is being reversed. As the discharge process takes place, the electrolyte becomes more and more diluted. In exchange, during charge the concentration of sulphuric acid in the electrolyte increases.

The total reaction is:

$$Pb + PbO_2 + 2H_2SO_4 \leftrightarrow 2PbSO_4 + 2H_2O$$ (4)

One of the most important characteristics of the lead-acid batteries is the very low internal resistance, fact that makes the voltage loss (internal voltage drop) very small.

There are two important categories of lead-acid batteries: with maintenance and without maintenance.

The lead acid batteries with maintenance have a liquid electrolyte and allow gas escape at the end of charge, when water electrolysis takes place; because of this it requires a refill with water in order to maintain the electrolyte concentration.

The lead-acid batteries with maintenance free (VRLA- Valve Regulated Lead-Acid), allow water recombination at the end of the discharging process. The oxygen is released at the cathode, passes through the electrolyte and it"s recombined at the anode, so that the refill with water is no longer required.

In order to be used in electric vehicles, the lead-acid batteries require certain modifications:

• The lead plates (forming the electrodes) should be thicker;
• The quantity of active material deposited on these plates should be higher ;
• The cathode is covered with one or several layers of fiberglass in order to prevent the active material from getting loose;
• The distance between the electrodes should be higher in order to facilitate the interaction between the active material and the electrolyte.

At present, the lead-acid batteries are used in micro-hybrids vehicles. Citroen launched in 2005 the model C3 equipped with a 55Ah battery which managed to obtain a fuel economy of approximately 10% in urban mode [3].

Parameter	Value
Specific energy [Wh/kg]	20-35
Energy density [Wh/l]	54-95
Specific power [W/kg]	250
Nominal voltage [V]	2
Internal impedance [Ω]	0.022
Lifespan [no. of cycles]	800
Self-discharging [%/day]	2
Temperature [°C]	10-55
Charging time [h]	8

Table 3. Lead-acid battery characteristics [4]

3.2. Nickel Batteries

The nickel batteries were the first serious rival for the lead batteries offering, first of all, a higher life span and also a better power and energy density.

There are four major technologies based on nickel: Ni-Fe, Ni-Zn, Ni-Cd and Ni-MH, out of which, at present, only the last two are still in use.

3.2.1. Ni-Cd Batteries

Ni-Cd batteries started to be produced after thirty years from the entry in the production of lead batteries, occupying, initially, areas non-covered by the lead batteries: railroads, safety lighting, aviation etc. [6].

The Ni-Cd batteries use cadmium at anode, nickel oxy-hydroxide (NiOOH) at cathode and an electrolyte made up of potassium hydroxide (KOH).

The total reaction is:

$$2NiOOH + 2H_2O + Cd \leftrightarrow 2Ni(OH)_2 + Cd(OH)_2$$ (5)

There are two great categories of Ni-Cd batteries: with and without maintenance.

For the batteries with maintenance free, the gases resulting after the charge process are recombined in order to form water, element leading to the electrolyte concentration diminution.

For the batteries with maintenance, this process does not take place and the resulting gases are eliminated through a valve, fact that leads, with time, to the necessity of adding water in order to dilute the electrolyte.

The Ni-Cd batteries are equipped with a mechanism of protection against overcharge, thus having from construction an excess of cadmium hydroxide (Cd(OH)2) at anode. From here, it results that the cathode will be always the first to be charged. If the charge continues, the following reaction takes place:

$$4OH^{-} \rightarrow 2H{2}O + O{2} + 4e_{-}$$ (6)

The resulting oxygen diffuses to the anode where the following reaction takes place:

$$O_2 + Cd + 2H_2O \rightarrow 2Cd(OH)_2 \tag{7}$$

As the charge process continues, at this electrode, the following reaction occurs:

$$2\text{Cd}(\text{OH})2 + 4e- \rightarrow 2\text{Cd} + 4\text{OH}^-$$ (8)

From the analysis of the mechanisms described by these reactions we can observe that the rate of cadmium hydroxide production is the same with the cadmium conversion rate, from where it results that we have a stable system that does not consume materials from the battery.

Parameter Value Specific energy [Wh/kg] 40-55 Energy density [Wh/l] 70-90 Specific power [W/kg] 125 Nominal voltage [V] 1.2 Internal impedance [Ω] 0.06 Lifespan [no. of cycles] 2000 Self-discharging [%/day] 0.5 Temperature [°C] -40–80 Charging time [h] 1

Table 4. Ni-Cd battery characteristics [4]

At present, the Ni-Cd batteries are used in the electric vehicles, especially in Europe: Peugeot 106, Citroen AX, Renault Clio, Renault Kangoo, but also in the USA - Ford Th!nk.

Considering the required voltage for powering these vehicles, between 8and 206V modules are used. Each module is made up of 5 cells connected in series and provided with a liquid cooling system, necessary in order to dissipate the heat resulting in the charge/discharge processes.

3.2.2. Ni-MH Batteries

The study of these batteries started in the 80s, but the technology became commercial available only in 1991, mainly due to the R&D programs. They are the result of the crisis generated by the increase of Cd price in the 80s, the Ni-Cd battery being the only power source for the portable electronic devices at that time. The R&D programs aimed at the same performance of the batteries based on Ni, but without cadmium. The environment issues related to Cd were revealed in the marketing period for this new type of battery – Ni-MH battery.

These batteries are similar in construction to the Ni-Cd ones, but the anode uses hydrogen instead of cadmium, which is absorbed in a metal alloy named metal hydride, MH. When the battery is discharged, the metal hydride oxidizes and forms an alloy, and at the cathode the nickel oxy-hydroxide is reduced and forms the nickel hydroxide.

The total reaction that occurs is the following:

$$MH + NiOOH \leftrightarrow M + Ni(OH)_2 \tag{9}$$

The interesting fact in that Ni-MH batteries is that the anode behaves exactly as a fuel cell. Thus, the hydrogen is attached to a metal hydride (an alloy specific to each manufacturer) and released when required. The functioning of the Ni-MH batteries is conditioned by keeping the hydrogen to a certain constant pressure (approximately 0.5 bars). Also, air penetration within the cell should be prevented because it may react with the metal hydride and replace the hydrogen.

Even though they have a quite high energy density, these batteries equipped specially prototypes, but in 1996 they were the power source for the EV1 an electric vehicle manufactured by General Motors, that provided a 250 km autonomy.

Parameter	Value
Specific energy [Wh/kg]	65
Energy density [Wh/l]	150
Specific power [W/kg]	200
Nominal voltage [V]	1.2
Internal impedance [Ω]	0.06
Lifespan [no. of cycles]	1000
Self-discharging [%/day]	5
Temperature [°C]	-40–80
Charging time [h]	1

Table 5. Ni-MH battery characteristics [4]

The development of hybrid vehicles made the Ni-MH batteries the preferred choice for several Full Hybrids vehicles, such as: Toyota Prius, Honda Civic, Honda Accord or Ford Escape. The batteries for the hybrid Toyota Prius, were manufactured in 1997 by Matsushita and contained 240D-type registered units. For the Prius II variant for Europe and USA they use also Ni-MH batteries with a reported density of 1100W/Kg.

The first generation of Ni-MH batteries used cylindrical cells and an air-cooling system. The new generations, in exchange, use rectangular cells loaded in metal cases. This implied a diminution of the volume occupied by batteries with 26%, at the level of the entire module, as well as a power increase with 55% (in relation to the volume). For a good thermal management, the batteries are equipped with channels allowing the air coming from the fans to reach all the cells.

First Generation cylindrical First Generation rectangular Second Generation rectangular New Model Nominal voltage [V] 7.2 7.2 7.2 7.2 Rated capacity [Ah] 6 6.5 6.5 6.5 Power [W] 872 1050 1352 1352 Mass [g] 1090 1050 1040 1197 Dimensions l[mm] H[mm] L[mm] 19.6 106 285 19.6 106 285 19.6 106 271.5 18.4 96 279

Table 6. Evolution of the Ni-MH batteries [1]

3.2.3. Batteries Based on Sodium

In the 80s, following a new development direction, batteries using liquid sodium as active material for the anode, started to be produced. Their main characteristic is that they operate at high temperature. These batteries did not meet a rapid development because, in the first place, they addressed only to one segment of the market, namely to the automotive industry.

There are two types of batteries based on sodium: with sulphur and with nickel chloride (Zebra), out of which only the Zebra batteries imposed on the market.

Parameter	Value
Specific energy [Wh/kg]	100
Energy density [Wh/l]	150
Specific power [W/kg]	150
Nominal voltage [V]	2
Lifespan [no. of cycles]	>1000
Self-discharging [%/day]	13
Temperature [°C]	300-350
Charging time [h]	8

Table 7. Zebra battery characteristics [4]

Zebra use the nickel chloride (NaCl2) as active material for the cathode and liquid sodium for the anode. The electrolyte is made up of two parts: a ceramic electrolyte surrounding the liquid sodium and a secondary electrolyte made up of sodium chloride and aluminum for the cathode. The Cl ions are the mobile ions from the electrolyte (they provide the ionic current). The electric energy obtained at discharge results from the combining of the sodium with the nickel chloride, nickel and sodium chloride resulting as product.

The total reaction is the following:

$$2Na + NiCl_2 \leftrightarrow Ni + 2NaCl$$

In order to provide the high functioning temperature, Zebra batteries use two stainless steel shells which include a 2-3 cm insulating layer. Nevertheless, except for short periods of time of 2-3 hours, the batteries must be connected to a power source; otherwise they consume approximately 100 W/h. It can be calculated that within 24 hours they consume nearly 13% of the stored energy.

At present, Zebra batteries are mounted on the electric vehicle Renault Twingo within the promotion program for electric vehicles in Mendrisio -Switzerland.

3.2.4. Batteries Based on Lithium

Lithium is the lighter of all metals used within batteries and presents interesting electrochemical characteristics. It allows a high thermodynamic voltage, fact that leads to very important energy and power densities. Li is a highly reactive substance. In 1970 the researchers from the Oxford University discovered the solution for intercalating Li in a cobalt or nickel crystalline network, thus creating the first batteries based on Li.

At present there are two major technologies based on lithium: lithium polymer (Li-P) and lithium-ion (Li-ion).

3.2.5. Li-P Batteries

Li-P batteries use lithium at the anode and an oxide of a transition metal at the cathode. This oxide presents a structure in which the lithium ions can be introduced or from which they can be removed according to the process taking place: charge or discharge.

The electrolyte, for these batteries, is a solid polymer (SPE) obtained by the dissolution of a lithium salt (Li(CFSO2)2N, LiTFSI), in a polyethylene-polypropyleneoxide (PEO-PPO) [7].

The total reaction is the following:

$$xLi + M_yO_z \leftrightarrow Li_xM_yO_z \tag{11}$$

At discharge, the lithium ions formed at anode migrate through SPE and they are introduced in the cathode structure, at charge the process is being reversed.

Parameter	Value
Specific energy [Wh/kg]	155
Specific power [W/kg]	150
Nominal voltage [V]	3
Lifespan [no. of cycles]	>1000
Self-discharging [%/day]	0.5
Temperature [°C]	60-80
Charging time [h]	8

Table 8. Li-P battery characteristics [4]

In the 2000s, these batteries were mounted on the Ford Th!nk vehicles within the programs sponsored by USABC. The battery structure was made up of 8 cells connected in series (one module) and 8 modules connected in parallel in order to create a 24V and 80Ah battery.

Starting with 2005 these batteries were withdrawn from the program, mainly because of the security issues related to oxygen generation within the discharge process, the situation presenting a risk of explosion. Now they are used especially in the field of telecommunications.

3.2.6. Li-Ion Batteries

The term Li-ion hides, in fact, an entire family of batteries. There is no universal structure, each manufacturer having a proper "recipe". Nevertheless, there are a few similarities related to the anode composition - usually made up of carbon or graphite and to the electrolyte containing, in most of the cases, a lithium salt (LiPF6) in an organic solvent.

For the cathode, the most known structures are the oxides of Co, Mn, Ni: LiCoO2, LiMn2O4, LiNiO2, or more complex structures: LiNiMnCoO2, LiNiCoAlO2, LiFePO4, etc.

The LiCoO2cells were among the first cells modules that used a lamellar structure for the cathode (Figure 1) in order to allow the Li ions to pass from the anode to the cathode at discharge, respectively from the cathode to the anode at charge.

Figure 1. LiCoO² cathode structure [8].

These cells are characterized by a high energy density, but also by a reduced power density, fact that makes them adequate for the automobiles requiring high autonomy and reduced currents. These were used by Nissan for the automobiles in Japan, but due to the overload issues they were withdrawn and now are used only for portable devices.

The LiMn2O4cells present a crystalline (spinel) structure of the cathode (Figure 2) which allows a high electron transfer rate fact that is translated in a high power density.

These batteries were tested on electric and hybrid vehicles prototypes in Japan, but, due to their short lifespan they were withdrawn and are now used only for electric tools.

The LiNiMnCoO2cells can have either a high power density or high energy density, but never both at the same time. Their "secret" resides in finding the right balance between Niproviding a high energy density and Mn- which, due to its crystalline structure, provides an increased flow of electrons and, implicitly, a high power density.

The LiNiCoAlO2cells have the advantage of both high power density and high energy density, fact that renders them ideal for the automobile industry. At present, Toyota tests this type of cells on the Vitz model, but problems related to safety may lead to their withdrawal from the market.

Figure 2. LiMn2O⁴ spinel structure [8].

The LiFePO⁴ cells came more recently on the market, but they drew attention of the manufacturers in the automobile industry by the fact that they are the safest batteries based on Li. This quality, corroborated with a high power density and with a short recharging time, makes them ideal for the hybrid vehicles, being already in Toyota"s attention. The only impediments come from the relatively low voltage, of only 3.3V supplied, and from the capacity loss at low temperatures.

The basic characteristics of Li-ion batteries are presented in Figure 3.

Figure 3. Li-Ion cells characteristics [8].

3.3. Metal-Air Batteries

The metal-air batteries followed a completely different development process as compared to rest of the batteries, combining the technology of the fuel cells with the technology of secondary batteries [9].

The defining characteristic for these batteries is their open structure, due to the necessity of absorbing oxygen from the atmosphere, within the discharge process [10, 11].

There are several types of metal – air batteries using: Ca, Al, Fe, Cd, Zn, Li etc., and the most known being those with Al, Zn and Li.

In the context where charging a battery consists in the reversal of the electric current direction, the metal – air batteries proposed a new method, by the complete change of the electrodes and electrolyte, because they are consumed during the discharge process. The main advantage of this type of charge is the short time in which it can be performed, usually between 8 and 10 minutes. The materials resulted after the charge can be afterwards recycled in order to recover the main components.

3.3.1. Al-Air Batteries

The chemical reaction occurring within these batteries involves the combining of Al with the oxygen in the air and with water in order to form the aluminum hydroxide (Al(OH)3), energy being released during the process. The aluminum represents the negative electrode and it is, usually, made up of a 1 cm thick plate. The electrolyte is an alkaline solution of potassium hydroxide.

The total reaction is irreversible and is:

$$4Al + 3O_2 + 6H_2O \rightarrow 4 Al(OH)_3$$ (12)

Parameter	Value
Specific energy [Wh/kg]	225
Energy density [Wh/l]	195
Specific power [W/kg]	10
Nominal voltage [V]	1.4
Lifespan [no. of cycles]	>1000
Temperature [°C]	10-55
Charging time [h]	10

Table 9. Al-air battery characteristics [1,4]

The battery is charged by the replacement of the anode and of the electrolyte, now containing the aluminum hydroxide. The electrolyte and the anode are processed in a recycling factory that recovers Al.

Due to the much reduced power density, this type of battery is not practical for being mounted on electric or hybrid vehicles.

3.3.2. Zn-Air Batteries

The Zn – air batteries are similar to the Al – air batteries, but they have an approximately 10 times higher power density. A complete drawing describing the composition and the operating process for these batteries is presented in Figure 4.

As we can see a Zn-air battery is made up of three parts: an anode composed of Zn, an air cathode divided into two layers, a gas diffusion layer and a catalytic active layer and a separator. This design was chosen to prevent the accumulation of reaction products in the air electrode by improving the oxygen transport into the inner regions [12, 27].

Figure 4. Zn-air operating process [12].

The battery operation implies the diffusion of the oxygen from the atmospheric air, through the carbon electrode, to the active catalytic layer from the cathode, where it"s transformed into hydroxyl with the help of the electrons generated from the anode after the oxidation reaction of the metallic Zn [12].

The total reaction is the following:

$$2Zn + O_2 \rightarrow 2ZnO \tag{13}$$

Parameter Value Specific energy [Wh/kg] 230 Energy density [Wh/l] 270 Specific power [W/kg] 105 Nominal voltage [V] 1.2 Lifespan [no. of cycles] >2000 Temperature [°C] 10-55 Charging time [h] 10

Table 10. Zn-air battery characteristics [1,4]

The Zn-air batteries represent the most mature technology in the field of metal-air batteries, combining in the most efficient way the low cost price, the operating safety and the specific energy density [13].

3.3.3. Li-Air Batteries

Encouraged by the results obtained with the Zn-air batteries, the researchers focused on lithium, which is the lightest material and which has the highest theoretical energy density. In this context, appeared the Li-air batteries, batteries with an energy density of approximately 12 kWh/kg (Figure 5), being the closest to the gasoline energy density of 13 kWh/kg.

Figure 5. Energy density for several battery technologies [12].

There are 4 types of Li-air batteries (Figure 6), all using Li (metal) and air oxygen as active materials at the anode and at the cathode. Differencing them is the electrolyte, which can be:

• With Li salts dissolved in nonaqueous solvents;
• With Li salts dissolved in aqueous solvents;
• With Li salts dissolved in hybrid (nonaqueous/aqueous) solvents;
• With solid electrolyte.

Figure 6. Li-air cell configuration [12].

The use of metallic lithium could make it possible to have high theoretical specific energy density value with very simple mechanism. During the discharge reactions, lithium ions dissolve into the electrolyte, from the Li metal anode, and oxygen reduction was occurred on air electrode. When oxygen reduction reaction is occurred, the processes are so complex depending on the type of electrolyte. Li2O² is the principal product of the reaction, but in some case, Li2O² is converted to Li2CO³ resulting a low cycle stability [14].

A key element for a long life for a Li-air battery is protecting the lithium metal anode from air moisture [15]. This is done by using an O² selective membrane, in the cathode, which allows O² into the cell while stopping or slowing the ingress of water vapor [16, 17].

At present, the Li - air batteries are not available on the market, but they present the highest potential for solving the automotive industry problem related to batteries energy density.

4. NEW TRENDS IN THE DEVELOPMENT OF THE BATTERIES FOR THE ELECTRIC DRIVE

Of all the existing batteries, the batteries based on lithium raised the highest interest and they have priority in the development as energy sources, due to the high energy density, increased lifespan, increased tolerance to mechanical shocks etc. [18].

In order to improve the lithium batteries performances new materials were tested. The conclusion was that in order to obtain high efficiency materials with a high lithium interconnection in their structure must be used [19, 20, 21]. This analysis conducted to the introduction of materials obtained by nanotechnology as electrodes for the lithium batteries.

A first step in this sense was made by the development of batteries using carbon nanotubes (CNT) instead of the graphite anode– Figure 7. Carbon nanotubes have extraordinary thermal and mechanical stability and high electronic conductivity, which makes them an ideal candidate for nanoscale matrix material for the anode [22]. The direct implication is that, now the entire mass of the electrode becomes active material. Due to this, the surface area, between the electrodes and the electrolyte, increases which leads to high electrical conductivity and high capacity [23]. This technology applied to the LiCoO2, LiNiCoAlO² and LiFePO⁴ batteries lead to an increase with 80% of the specific energy and with 60% of the energy density [24].

Figure 7. Carbon nano-tubes structure [13].

Encouraged by these results, other materials obtained by nanotechnology were also tested as materials for the anode. Among these, the silicon was the most promising, due to its specific capacity. Thus, the batteries using silicon nano-fibers at the anode presented a significant capacity increase but also a performance increase to almost 90% [25].

Nano-tin powders supported by multiwall carbon nanotubes (MWCNT) were also tested as materials for the anode. One of these is a Sn-MWCNT anode with 11-20% higher volumetric capacity than a graphite anode and excellent reversible capacity stability. This anode retains 71% of its initial reversible capacity up to the 280th cycle. In addition, the anode has a significant advantage over the graphite anode in terms of battery safety [22].

Attempts were made also for the increasing of the lifespan, and in this direction, the nano-materials using sulphur demonstrated a reduction of only 20% of the battery capacity after 3000 cycles.

The research concerning the improvement of the Li batteries did not stop only to the anode, it also involved the cathode. The main characteristics for the cathode materials should be superior electrochemical properties without any rare materials. A way of obtaining these materials is by "blending" two or more distinct lithium intercalation compounds. The motivation for blending these compounds is to achieve a more balanced performance compared to what is possible with any individual compound [25]. Among these, iron nitrodisulphate combined with lithium bromide lead to the obtaining of batteries with125 mAh/g capacity which that can be easily extended up to 200mAh/g, in the conditions where at present, the highest performance present by a Li batteryis150 mAh/g capacity [26].

CONCLUSION

After more than 100years of continuous development, the electric vehicle reached a decisive threshold, especially due to the development of hybrid vehicles. The preoccupation for the environment and the diminution of the fossil fuel reserves will contribute to the acceleration of their further development [1].

Due to the fact that there are several types of electric vehicles, each one with specific requirements, several types of batteries will have to be developed. Figure 8 shows a chart representing the evolution of batteries technologies for electric vehicles and it is the starting point for predicting their new trends.

The lead-acid batteries are the best represented on the batteries market, and for a few years they will continue to be, because of their superior cost to stored energy ratio. Nevertheless, due to the specific reduced energy, it is hard to believe that they will one day be mounted on the highly autonomous electric vehicles.

The Ni-Cd batteries are the only ones presenting a downward trend due to the fact that their active element - Cd is a polluting material. Even though they have certain remarkable performances, such as increased lifespan and excellent resistance to mechanic shock, they will be replaced on the automotive market, but will find their way in the electric tools industry.

The Ni-MH batteries dominate for now hybrid vehicles market segment. The clear advantages related to the high energy and power density, as well as the fast recharging capacity will help them to maintain themselves on this position on the medium term, but on a long term they might be surpassed by the lithium batteries.

Figure 8. Evolution of the batteries technologies 1996-2012 [8].

The lithium batteries, propelled by new technologies and also by the existing financing programs, will become a serious competitor for the Ni-MH batteries, and, on the long term, they could even become the most used power source in the automotive industry [3]. Among these, the LiFePO4batteries, promoted by a much reduced price and by the highest operating safety, will impose themselves in front of the other types of Li batteries.

For electric vehicles, the only considerable progress is made by the metal-air batteries and, especially, by the Li ones. Due to the high energy densities and the short charging time, these batteries can prove themselves to be the most important power source for this type of vehicles.

REFERENCES

• [1] Braussely, M; Pistoia, G; Industrial Applications of Batteries: From Cars to Aerospace and Energy Storage. Elsevier, 2007.
• [2] www.davidsuzuki.org/publications/downloads/2007/DSF-Prescription-for-Canada-Full-Research.pdf.
• [3] Dhameja, S; Electric Vehicle Battery Systems. Newnes, 2002.
• [4] Larminie, J; Lawry, J; Electric Vehicle Technology Explained. John Wiley & Sons, 2003.
• [5] Tomescu, A; Tomescu, I.B.L; Tomescu, F.M.G; Direct Energy Conversion. Matrix Rom, Bucharest, Romania, 2008.

• [6] Lefter, E; Power Supply Vehicles. Ed. Mediamira, Cluj-Napoca, 2006.

• [7] Ehsani, M; Gao, J; Emandi, A; Modern Electric, Hybrid Electric, and Fuel Cell Vehicles, Fundamentals, Theory and Design. CRC Press, 2005.

• [8] http://batteryuniversity.com/learn/article/types_of_lithium_ion.

• [9] Martin, J.J; Neburchilov, V; Wang, H; Qu, W; Air Cathodes for Metal-Air Batteries and Fuel Cells. IEEE Electrical Power & Energy Conference (EPEC 2009), Montreal, Canada, 22-23 Oct. 2009.

• [10] Wang, J; Li, Y; Sun, X; Challenges and Opportunities of Nanostructured Materials for Aprotic Rechargeable Lithium-Air Batteries. NanoEnergy Journal, vol. 2, issue 4, pp. 443-467, July 2013.

• [11] Zhang, J.G; Wang, D; Xu, W; Xiao, J; Williford, R.E; Ambient Operation of Li/Air Batteries. Journal of Power Sources, vol. 195, pp. 4332-4337, 2010.

• [12] SooLee, J; Kin, S.T; Cao, R; Choi, N.S; Liu, M; TaeLee, K; Cho, J; Metal-Air Batteries with High Energy Density: Li-Air vs Zn-Air. Advanced Energy Materials, no. 1, pp. 34-50, 2011.

• [13] Cao, R; Lee, J.S; Liu, M; Cho, J; Recent Progress in Non-Precious Catalysts for Metal-Air Batteries. Advanced Energy Materials, no. 10, pp. 1-14, 2012.

• [14] Jang, I.C; Hidaka, Y; Ishihara, T; Li Metal Utilization in Lithium Air Rechargeable Batteries. Journal of Power Sources, vol. 244, pp. 606-609, 2013.

• [15] Shimonishi, Y; Zhang, T; Johnson, P; Imanishi, N; Hirano, A; Takeda, Y; Yamamoto, O; Sammes, N; A Study on Lithium/Air Secondary Batteries: Stability of NASICON-Type Glass Ceramics in Acid Solutions. Journal of Power Sources, vol. 195, pp. 6187- 6191, 2010.

• [16] Crowther, O; Keeny, D; Moureau, D.M; Meyer, B; Salomon, M; Hendrickson, M; Electrolyte Optimization for the Primary Lithium Metal Air Battery Using an Oxygen Selective Membrane. Journal of Power Sources, vol. 202, pp. 347-351, 2012.

• [17] Chervin, C.N; Long, J.W; Brandell, N.L; Wallace, J.M; Kucko, N.W; Rolison, D.R; Redesigning Air Cathodes for Metal-Air Batteries Using MnOx-Functionalized Carbon Nanofoam Architectures. Journal of Power Sources, vol. 207, pp. 191-198, 2012.

• [18] Tao, H; Feng, Z; Liu, H; Kan, X; Chen, P; Reality and Future of Rechargeable Lithium Batteries. The Open Material Science Journal, vol. 5, Suppl. 1: M2, pp. 204-214, 2011.

• [19] Phan, A; High Performance Lithium-ion Battery Anodes Using silicon Nanowire. COSMOS Cluster 8, 2012.

• [20] Xiong, Z; Yun, Y.S; Jin, H.J; Applications of Carbon Nanotubes for Lithium Ion Battery Anodes. Materials, vol. 6, no. 3, pp. 1138-1158, 2013.

• [21] Bresser, D; Oschmann, B; Tahir, M.N; Tremel, W; Zentel, R; Passerini, S; Stabilizing Nanostructured Lithium Insertion Materials via Organic Hybridization: A Step Forward Towards High-Power Batteries. Journal of Power Sources, vol. 248, pp. 852-860, 2014.

• [22] Menkin, S; Barkay, Z; Golodnitsky, D; Peled, E; Nanotin Alloys Supported by Multiwall Carbon Nanotubes as High Capacity and Safer Anode Materials for EV Lithium Batteries. Journal of Power Sources, vol. 245, pp. 345-351, 2014.

• [23] Kang, C; Lahiri, I; Baskaran, R; Kim, W.G; Sun, Y.K; Choi, W; 3-Dimensional Carbon Nanotube for Li-Ion Battery Anode. Journal of Power Sources, vol. 219, pp. 364-370, 2012.

• [24] Landi, B.J; Cress, C.D; Raffaelle, R.P; High Energy Density Lithium Ion Batteries with Carbon Nanotube Anodes. Journal of Material Research, vol. 25, no. 8, pp. 1636-1644, 2010.

• [25] Chikkannanavar, S.B; Bernardi, D.M; Liu, L; A Review of Blended Cathode Materials for Use in Li-Ion Batteries. Journal of Power Sources, vol. 248, pp. 91-100, 2014.

• [26] Liu J; Yu, X; Hu, E; Nam, K.W; Yang, X.Q; Khalifah, P.G; Divalent Iron Nitridophosphates: A New Class of Cathode Materials for Li-Ion Batteries. Chemistry of Materials, vol. 25, no. 20, pp. 3929-3931, 2013.

• [27] Williford, R.E; Zhang, J.G; Air Electrode Design for Sustained High Power Operation of Li/Air Batteries. Journal of Power Sources, vol. 194, pp. 1164-1170, 2009.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 16

COMPUTATIONAL STUDY OF Li-ION BATTERIES FOR AUTOMOTIVE APPLICATIONS

Inigo Aramendia*¹ , Unai Fernandez Gamiz¹ , Jose Antonio Ramos†² , Javier Sancho¹ and Ekaitz Zulueta‡³

¹Nuclear Engineering and Fluid Mechanics Department, University of the Basque Country, Vitoria Gasteiz, Araba, Spain Electrical Engineering Department, University of the Basque Country, Vitoria Gasteiz, Araba, Spain 3 Systems Engineering and Automatics Department, University of the Basque Country, Vitoria Gasteiz, Araba, Spain

ABSTRACT

The battery is one of the most important elements of an electric vehicle. The implementation of these vehicles is determined by the performance to be derived from their batteries, as described in [1]. Vehicle manufacturers must ensure that the life expectancy for the battery of eight to ten years is viable before the beginning of series production, because it is undesirable to inform to the buyer of a car the need to replace the battery, associated with high costs, after only a few years.

Previous experimental work, e.g. [2], shows that the application of lithium-ion batteries in the automotive world is growing significantly as a result of increased commitment by electric and hybrid vehicles. Lithium batteries [3] are considered the most attractive of all the storage solutions available today and they are expected to be used in the next generation of hybrid and electric vehicles.

The main goal of this chapter is to analyze a battery submodule design consisting of lithium ion cells that allows an optimum cooling battery, and assess the heat transfer and temperature of the cells along the testing time. To that end, computational fluid dynamic simulations (CFD) have been carried out by CFD. This is a very versatile and useful tool which allows simulate many industrial cases as models and prototypes.

^* E-mail: email: inigo.aramendia@gmail.com.

^† E-mail: josean.ramos@ehu.es.

E-mail: ekaitz.zulueta@ehu.es.

Li-ion batteries performance and life are significantly affected by operating temperatures of the battery. Thermal management is one of the main issues to be considered in the design of the battery, [4]. The control of the maximum temperature and of the maximum temperature difference in a cell is critical parameter in cell durability and in element safety [8].

The battery model investigated in this work is based on cold plates, as showed in [9]. Two versions will be displayed, differentiated by varying the geometry of the cold plates; in one, they will be solid and in the other one the cold plate is created by cold extrusion. In both models will be shown the geometry and mesh of the elements. The physical properties and initial conditions will be analyzed, as well as the convergence of waste products, air speed, cells temperature and heat transfer.

Keywords: Li-Ion Batteries, CFD, Refrigeration, Automotive

NOMENCLATURES

VOC Open Circuit Voltage V Nominal Voltage

Vmin, Vmax Maximum and Minimum Voltage per cell in volt

C Stored Electrical Charge in A-hr

n Recommended and Actual Number of Charge Cycles

SOC State of Charge DOD Depth of Discharge SOH State of Health

Specific Energy, Energy Density, Specific Potential in Wh/kg, Wh/L,

W/kg

Maximum Charge and Discharge Currents in Ampere

Internal Resistance in mOhm Thermal Operating Range in ºC Charge-discharge efficiency in %

1.INTRODUCTION

1.1. History

The first lithium battery prototypes were developed in the 1970s by M. Stanley Whittingham, while he was working for Exxon, using titanium disulphide for the negative electrode (cathode) and lithium metal for the positive electrode (anode). Nevertheless, the first workable lithium batteries started to emerge from Bell Labs, where graphite anodes were developed as a more efficient alternative to lithium metal.

By 1983, Michael Thackeray and his group had already identified manganese spinel as a material for the cathode, as it has a high conductivity, relatively low cost and an internal structure that offered good stability for use in next-generation storage batteries. Initially, the drawback of this approach was that the spinel degraded after the first cycle of use, but this has been addressed in subsequent designs.

Two years later, in 1985, Akira Yoshino proposed a prototype with a carbonaceous material in which lithium ions could be inserted for one electrode and lithium cobalt oxide (LiCoO2), which is stable in air, for the other. This design markedly improved the safety of the batteries, as it avoided the use of lithium metal. Moreover, the use of LiCoO2 meant that production could readily be scaled-up to an industrial level.

In 1989, Arumugam Manthiram and John Goodenough at the University of Texas at Austin took these early lithium batteries another step forward by incorporating so-called polyanion cathodes, which were able to produce higher voltages than the oxides given the inductive effect of the polyanion and their higher capacitance.

The first commercial battery, however, was launched by Sony and Asahi Kasei in 1991 with new features (lithium cobalt). The design made these storage batteries truly competitive.

Subsequently, in 1996, John Goodenough, AkshayaPadhi and coworkers discovered the electrochemical advantages of using lithium iron phosphate (LiFePO4) and phospho-olivines (lithium metal phosphates with the same structure as mineral olivine) as cathode materials. These new materials further improved the safety of the batteries and were important in the development of this technology, being used above all in mobile telephones and digital watches.

In the last decade, the performance of Li-ion batteries has continued to grow: Yet-Ming Chiang and his group at MIT demonstrating, in 2002, a considerable improvement by increasing the conductivity of components by doping with aluminium, niobium and zirconium; and the same group further improving the performance, in 2004, using iron phosphate particles of less than 100 nm in diameter.

By 2011, Li-ion batteries accounted for 66% of the rechargeable battery market in Japan.

1.2. Background Knowledge

A battery is a set of electrochemical cells, generally connected in series to increase the final voltage, that is able to store energy by means of a chemical reaction and later release the chemical energy as electrical energy. All electrochemical cells are formed by anodes, cathodes, electrolytes and an external circuit connecting the electrodes.

Nevertheless, not all batteries are alike. Some can only be used once: they cannot be recharged as their internal structure does not enable the energy extracted from the battery to be restored. This class of batteries would have to be disposed of after use and are known as "primary cells". There are different types:

• "Dry cells", first those with Zn and C electrodes and mostly a ZnCl² electrolyte and later others, known as alkaline batteries, based on Zn/MnO2.
• Li-ion cells [1], the most widely used being those based on Li and MnO2. These are relatively cheap to manufacture, but more expensive than dry cells. They are thought to account for 80% of the batteries in general (rechargeable and non-rechargeable).

On the other hand, when the chemical process can be reversed by an electrical current without producing an immediate degradation of the system, this is called a "secondary cell" and the battery is rechargeable. Today, rechargeable batteries are a practical and effective source of energy.

Until recently, batteries in cars have been an element that only provided the electrical energy for them to work. For that, they have used small batteries kept charged by the running of the car. Now, however, electric cars aim to go a step further; the goal is for their movement to be powered by battery systems, abandoning the petrol- and diesel-powered motors we have known to date. This type of batteries for electric cars can be divided into three types:

• Lead-acid batteries. The cheapest and most common type, these batteries have been used to start combustion engines for decades. They have a good specific power (W/kg), behave well at a wide range of temperatures, retain charge well over time and are relatively easy to recycle. On the other hand, they have only a rather low energy density (40 Wh/kg).
• Ni-metal batteries. Their specific power is satisfactory, they have a long lifecycle and do not present environmental problems, but they have a relatively short shelf-life (losing around 30% of their charge in a month when unused). Their production cost is somewhat high given that rare earth metals are used in the anode. At around 60 Wh/kg, they provide a higher energy density than lead-acid batteries, but their performance is not sufficiently good to allow a true expansion in electric car use.
• Lithium-ion batteries. These offer better performance than the other two types in terms of voltage, energy density, specific power, usable charge, recharge efficiency and lifecycle. They also have a slower self-discharge rate. They are, however, less robust to voltage variations and are more expensive.

The following lithium-ion batteries are the most widely used subtypes.

• Lithium cobalt (LiCoO2): with an energy density of 170-185 Wh/kg, these provide only 500 charge cycles. If damaged, there may be exothermic reactions and these can even result in fire. They tend not to be used in electric vehicles.
• Lithium iron phosphate (LiFePO4): these offer an energy density of 90-125 Wh/kg. They also offer up to 2000 charge cycles, and with this are the cheapest, together with the cobalt batteries. Further, with a high thermal and chemical stability they are safer. Indeed, they present a relatively low-risk option for powering vehicles.
• Lithium manganese (LiMn2O4): these provide an energy density of 90-115 Wh/kg. This energy range is lower than that of the cobalt batteries, but they are more stable and operate at higher voltages. What is more, manganese is less polluting.
• Lithium nickel cobalt manganese (LiNixCoyMnzO2): with an energy density of 155- 190 Wh/kg, these have a good cost-performance ratio. They are starting to be used on a massive scale. They offer some 1500 charge cycles at among the highest voltages.
• Lithium titanium (Li4Ti15O12): these have the longest lifecycle providing some 12000 charge cycles, but offer the lowest range of energy density, 65-100 Wh/kg, and are very expensive.

In a short time, lithium-ion batteries have become the main source of energy for electronic equipment, due to the fact that they can store more energy per unit volume or weight than lead-acid batteries and nickel-metal hydride or nickel-cadmium batteries (both nickel-based designs being developments from alkaline batteries). Lithium-ion batteries are usually stable for around 2200-2500 deep charge cycles, which corresponds to around 10 years of use. Moreover, a battery in fact still has 80% of its maximum charge capacity at this stage. A wide range of sizes and designs of this type of battery is already manufactured for various different applications, including information technology devices, and portable tools and machines, as well as vehicles, see [2].

1.2.1. Charge and Discharge Processes

*Charging*: During this process, energy is supplied in the form of an electric current, causing Li+ ions to flow from the anode into the electrolyte and thence to the cathode where the host material is reduced and host ions (Li+) are inserted. At the same time, the positive electrode is oxidised.

The electrolyte allows the flow of ions, but not of electrons. During charging, electrons also flow from the anode to the cathode, but through the external circuit. In this way, the cathode becomes more negatively charged and the anode more positively charged, increasing the potential difference between them, and hence the voltage of the cell.

*Discharging*: The process is exactly the reverse of what occurs during charging; the battery now supplies an electric current. This process forces lithium ions to leave the cathode (extraction), oxidising this electrode, while the ions are inserted into the anode, leading to its reduction. As the discharging process progresses, the electric potential (E) at each electrode falls, decreasing the voltage across the battery.

The mobility of Li⁺ ions between the electrodes is possible due to the fact that both are in contact with an electrolyte formed by a lithium salt dissolved in a non-aqueous solvent. LiClO and LiPF6 are among the most commonly used salts, while ethylene carbonate (EC), propylene carbonate (PC), dimethoxyethane (DME), diethyl carbonate (DEC) and dimethyl carbonate (DMC) are the most common solvents.

1.2.2. Important Technical Considerations

From a technical point of view, there are five key factors to consider: safety, durability, performance, cost, energy density and power.

• Good safety is undeniably essential for the success of a technology, and we tend to associate Li-ion batteries with a risk of explosion in certain circumstances [3], this being particularly severe in the case of large batteries, as are required for the automotive application under study. A small number of accidents could result in the technology being rejected by potential users. We should, however, keep this in perspective given that, through it tends not to be perceived by society, a full 60-litre tank full of petrol also presents certain hazards.
• Durability is not a trivial issue for current batteries. The manufacturers aspire to batteries working for 10 years, this lifetime being considered to be around half that of the vehicles themselves. On the other hand, cells progressively lose their capacity with charge cycles, and this makes it necessary to design batteries with excess power at the start, increasing their weight, volume and cost.
• Performance, another key characteristic, depends on many other factors such as, for example, the temperature of the environment. With current batteries, we cannot expect the same level of performance in a hot summer in Seville and on a cold day in Ávila (both in Spain) or say Moscow. They could be optimised to improve performance in hot or cold conditions, but many improvements need to be made to

obtain the same output in the full range of conditions that a given vehicle might be expected to operate.

• The energy density and power must also be considerably improved, as the values of these parameters offered by currently available batteries are far from that of petrol (little over 1%), and this limits the autonomy of electric vehicles to a few hundred kilometres at best.
• Lastly, all the aforementioned parameters should be optimised [4] without excessively increasing the cost of the battery and, in turn, of the vehicle. For the year 2020, the goal set by the battery sector itself is some €175 per kWh, which is approximately equivalent to the cost of the small batteries used in consumer electronics. The batteries required are, however, much more complex to construct and the current cost is still over four times this target. Clearly, large-scale production of the batteries will allow the final price to be considerably lower than at present, but it is nevertheless difficult to believe that this goal will be achieved.

1.3. Computational Fluid Dynamics (CFD) Tools

Currently, one of the most popular approaches to development in the industrial sector is the simulation of processes. Design remains important in this context but it is now seen that this has to go hand-in-hand with simulation. Thanks to simulation programs it is possible to explore the details of multiple thermal processes that it was previously impossible to visualise and analyse. By running simulations, time and money can be saved with respect to investigating the real processes, which tend to be more expensive, lengthy and, in some cases, more complex to interpret.

With the help of CFD tools it is possible to simulate and analyse any industrial process to obtain a desired model. On the basis of such a CFD computer based model, a prototype can then be validated and results obtained to later be applied to the real system. The usefulness of CFD analysis lies in its ability to predict fluid behaviour in great detail in the domain studied, support design and development of prototypes, and obtain solutions rapidly avoiding expensive experiments, allowing visualisation and animation of processes in terms of characteristics of the fluids involved. On the other hand, the CFD approach requires highly trained users with extensive experience, and substantial hardware and software resources that require significant investments, in some cases, the computational cost being high. In particular, the equations of fluid mechanics do not have analytical solutions and hence need to be solved numerical methods. For this, CFD requires discretisation of space. The working approach will be:

• 1 Description of the problem
• 2 Base model/geometry
• 3 Discretisation/Mesh generation
• 4 Physical definition of the problem
• 5 Iterative calculations
• 6 Post-processing/Analysis

Given the increases in the price of petrol, its limited supply and the growing concern for the global environment, the automation sector is committed to electrical and hybrid vehicles. The device for the storage of electrical energy continues to be the component that is proving the greatest headache for the manufacturers. This challenge requires the development of new types of batteries which are more efficient and offer higher energy densities. Among the solutions available today, lithium batteries are considered the most attractive candidates, given their strengths over the conventional nickel-metal hydride batteries:

• Better performance and energy densities
• More compact size
• High charge/discharge efficiency
• More charge cycles

2. STATE OF THE ART: PAST AND PRESENT OF THE LITHIUM-ION BATTERY COOLING TECHNOLOGY

Below it summarize the main candidate solutions, in terms of types of refrigeration and arrangements of the cells.

2.1. Lithium-Ion Battery with Passive Cooling

This patent [5] is focused on high energy density batteries and, more particularly, on a package of Li-ion batteries with a plate of metal, such as aluminum, to dissipate the heat of the cells in the battery to the central chamber (Figure 1).

Figure 1. Passive cooling with an aluminum plate [5].

The battery consists of 28 cells, with rapid recharge times and long lifecycles and is relatively environmentally friendly.

2.2. Modular Battery System with Cooling System

This is a modular system with at least two battery modules (14 15) (Figure 3) in which in each module is a cooling element [6]. Figure 2 shows the distribution of batteries (in two parallel rows and 3 in series in each row). A cooling bracket, a light alloy extruded section, extends over the two rows of cells (Label 8 in Figure 3).

Figure 2. Distribution of the cells in a module [6].

Figure 3. System consisting of 6 modules [6].

If adjacent modules are connected in series a problem arises: the length of the ductwork for the refrigerant near the main connection (the inlet) is shorter than for those at the back. This translates into a greater flow of refrigerant to the front modules and, therefore, better cooling.

To avoid this, the coolant return path could be on the opposite side of the connection and then take it to the main common connection on the other side of the storage device through a dedicated channel. However, this requires additional space and increases costs.

In principle, the coolant could be gaseous or liquid. In this patent, a liquid coolant was proposed.

2.3. Battery Cooling Device

In this design, a first cooling medium is injected (air). It would enter by (112) and exit by (114) (Figure 4). In addition, a fan can be installed to increase the volume of flow [7].

The second cooling medium can be water. It would be stored in the tank (136), supplied by the cooling pipe, which can be a spiral to increase the refrigeration efficiency (332) through the water pump (334).

Figure 4. Cooling device with air and liquid refrigerant (water) [7].

3. APPLICATION TO A REAL WORLD EXAMPLE

3.1. Technical Data

The base data/requirements considered are the key specifications equivalent to a battery for a bus, tram or similar vehicle.

Nominal Voltage: 380 V Nominal Capacity: 100 Ah Maximum Power: 100 kW

These base data, corresponding to the full battery pack, will be divided into modules and, in turn, into submodules. The analysis will focus on the specific case of cooling a submodule, for which is knows (from theoretical calculations) the number of cells to be placed in series and parallel, as well as their specifications.

The following proposals, studied in more detail with simulations, are composed of Li-ion cells LiFeBATT40138F1 (Figure 5). These cells have the characteristics in Figure 6.

Figure 5. LiFeBATT cell.

Figure 6. Physical characteristics and operation conditions of LiFeBATT cells.

Each submodule is able to deliver 14 Ah at 48 V.

Each submodule is able to deliver 14 Ah at 48 V.

Number of submodules = $$\frac{\text{Total Capacity}}{\text{Submodule Capacity}} = \frac{100 \text{ Ah}}{14 \text{ Ah}} = 7.14 \approx 8 \text{ submodules}$$ (1)

Actual total capacity = Submodule Capacity × No. of submodules = 14.8 = 112 Ah (2)

(2)

Hence, each module is able to deliver 112 Ah at 48 V.

Number of submodules = $$\frac{\text{Total Voltage}}{\text{Module Voltage}} = \frac{380}{48} = 7.916 \approx 8 \text{ modules}$$ (3)

Actual total voltage = Module Voltage × No. of modules = $48.8 = 384 \text{ V}$ (4)

Actual total voltage = Module Voltage $$\times$$ No. of modules = $48.8 = 384$ V (4)

Then, is calculated the number of cells to be placed in series and in parallel in each module and submodule to meet requirements.

Submodule:

Submodule:

Cells in series = $$\frac{\text{Total Voltage Submodule}}{\text{Nominal Voltage}} = \frac{48 \text{ V}}{3.2 \text{ V}} = 15 \text{ cells}$$ (5)

Nominal Voltage 3.2 V

Cells in parallel = $$\frac{\text{Total Capacity Submodule}}{\text{Nominal Capacity}} = \frac{10 \text{ Ah}}{14 \text{ Ah}} = 0.714 \approx 1 \text{ cell}$$ (6)

Total Submodule Cells 15.1 15 cells (7)

Module:

Cells in series = $$\frac{\text{Total Voltage Module}}{\text{Nominal Voltage}} = \frac{48 \text{ V}}{3.2 \text{ V}} = 15 \text{ cells}$$ (8)

Nominal Voltage 3.2 V

Cells in parallel = $$\frac{\text{Total Capacity Module}}{\text{Nominal Capacity}} = \frac{100 \text{ Ah}}{14 \text{ Ah}} = 7.14 \approx 8 \text{ cells}$$ (9)

Total Module Cells 15 .8 120 cells (10)

Total cell Pack LiFeBATT Total Module Cel ls No. of modules 120 .8 960 cell

(11)

To calculate the maximum current flowing through each cell is used the equation: P V I . (12)

First, the maximum power is divided by the total number of cells:

$$\frac{\text{Maximum Power}}{\text{Number total of cells}} = \frac{100.000 \text{ W}}{960 \text{ cells}} = 104.16 \frac{\text{W}}{\text{cell}}$$ (13)

Hence

$$I = \frac{P}{V} = \frac{104.16 \text{ W/cell}}{3.2 \text{ V}} = 32.55 \text{ A}$$ (14)

Now the efficiency of each cell is estimated, with three possible cases, high, moderate and low efficiency; from this, the heat generated in each cell (losses) is calculated to determine how much heat needs to be dissipated from the system (heat losses).

$$Q = I^2 R = 32.55^2 \times 0.006 = 6.36 \text{ W}$$ (15)

$$Q_{total} = (I^2 R) \times \text{No. of cells} = 6.36 \times 960 = 6103.52 \text{W}$$ (16)

If is focused on a submodule, then:

If is focused on a submodule, then:

$$Q{sub \, \text{mod} \, ule} = (VI) \, \eta{low} = (3.2 \times 32.55) \times (1 - 0.8) = 20.83 \, \text{W}$$ (17)

First design: Forced convection cooling system

The first design studied is based on forced convection, as described in [8]. Is considered a submodule of 15 cells, with 5 cells in series in 3 parallel rows. All this is placed inside an aluminium housing, which has a series of inputs and outputs to ensure that the airflow is as uniform as possible.

Figure 7. (a) In-line and,(b) staggered configurations.

$$T = 298 \text{ K } (25^{\circ}\text{C}) \qquad P = 1.013 \text{ bar (atmospheric pressure)}$$

$$V = 1 \text{m/s} \qquad b/a = 1.5$$

$$c/a = 2 \qquad T_{w1} = 303 \text{K}$$

$$T_{w2} = 323 \text{K}$$

$$Cell \text{ surface area} = \pi dh = \pi \times 0.0406 \text{m}^2 \times 0.1388 \text{m} = 0.0177 \text{ m}^2$$

$$(18)$$

Cell surface area = $$\pi dh$$ = $\pi \times 0.0406 \text{m}^2 \times 0.1388 \text{m} = 0.0177 \text{ m}^2$ (18)

Cell surface area = $$\pi dh$$ = $\pi \times 0.0406 \text{m}^2 \times 0.1388 \text{m} = 0.0177 \text{ m}^2$ (18)
Total heat flux = $0.4369 \frac{\text{kW}}{\text{m}^2} \times 0.0177 \text{ m}^2 \times \frac{1000 \text{ W}}{1 \text{ kW}} \times 15 \text{ cells} = \boxed{115.99 \text{ W}}$ (19)

Figure 8. Geometry of the linear system of cells with convection cooling.

The simulation was conducted with two different cell arrangements: in-line and staggered. First, these were simulated with the "QTPro² " program. By entering the parameters describing each configuration, you obtain the average heat flow.

The geometry of the model is first drawn with a CAD program and then imported to the CFD solver, which generates the base mesh. For this analysis, is used a polyhedral mesh with an average size of 0.002 m, obtaining 836,000 elements for the aligned cells and 975,000 for the staggered configuration.

As for boundary conditions, is considered an air inlet velocity of 5.5 m/s. The physical conditions (physical values) applied in the simulation are as Table 1.

	Physics 1 (Cells)	Physics 2 (Air)
Space	Three dimensional	Three dimensional
Motion	Stationary	Stationary
Time	Steady	Steady
Material	Solid	Gas
Equation of state	Constant density	Constant density
Flow	()	Segregated Flow
Viscous Regime	()	Turbulent
Reynolds Average Turbulence	()	K-Epsilon Turbulence
Optional physical model	Segregated Solid Energy	Segregated Fluid Temperature

Table 1. Physical Conditions of the model

Figure 9. Plot of temperatures.

The first result that is considered from the simulation is a map of the temperature across the system. The air flows past all the cells to the exit. It can be observed that the cooling of the cells is poorer the further they are from the inlet. This is due to the air being warmed as it flows through the system by the heat generated by the cells.

In the in-line configuration, the temperature of the cells is around 333 K close to the inlet, increasing to a maximum value of 356 K in the central cell of the last row. With the same air velocity (5.5 m/s) we obtain the temperature patterns are illustrated in Figures 9 and 10.

Another output that from the simulation is the air velocity; this should be explored in to determine whether there are problems such as the generation of eddies and vortices. These effects cause the phenomenon of reverse flow.

The staggered configuration can provide significantly better results. On the other hand, this system would require higher air flow velocities.

Figure 10. Detailed view for two cells of the velocity vectors field of the air to illustrate reverse flow.

3.2. Definition and Concept of Cold Plate Cooling

This consists of a plate of aluminum or another metal that contains an inner channel through which a liquid coolant is circulated to absorb the heat transferred to the plate by transistors and other elements mounted on it, in this case battery cells. The cooling method is based on direct contact between the plate and the object to be cooled, direct contact allowing the heat to be efficiently conducted away (Figure 11).

In fact, this system is a type of heat exchanger that uses metal plates (aluminum) to transfer heat between two fluids. Its main advantage over a conventional heat exchanger is that fluids are exposed to greater surface because they are distributed across the plate, as showed in [9]. This facilitates the transfer of heat and increases the speed of the temperature change. Generally, the pipe is formed into a serpentine shape in order that the liquid follows a longer path and thereby the cooling is greater.

Figure 11. Cold plates [10].

3.3. System of Solid Aluminum Cold Plates

The cells are placed between cold plates (Figure 12), here made of solid aluminium, and coolant pipes run through these plates, as shown in Figure 13. We illustrate the results for the inline configuration. In this case, the geometry shown is only half of the system, the other half being symmetrical. In this way, the simulation times are reduced significantly.

Figure 12. Geometry of the solid aluminum cold plate.

Figure 13. Geometry and meshing system with solid aluminum cold plates.

| | | Table 2. Physical properties of the solid aluminum cold plates model | |--|--|----------------------------------------------------------------------| |--|--|----------------------------------------------------------------------|

	Physics 1 (Cells)	Physics 2 (Cold plates)	Physics 3 (Pipes)
Space	Three dimensional	Three dimensional	Three dimensional
Motion	Stationary	Stationary	Stationary
Time	Implicit Unsteady	Implicit Unsteady	Implicit Unsteady
Material	Solid (Aluminum)	Solid (Lithium)	Liquid (Water)
Equation of state	Constant density	Constant density	Constant density
Flow	()	()	Segregated Flow
Viscous Regime	()	()	Turbulent
Reynolds Average Turbulence	()	()	K-Epsilon Turbulence
Optional Physical Model	Segregated Solid Energy		Segregated Solid EnergySegregated Fluid Temperature

Table 3. Material characteristics

	Density (kg/m3 )	Specific Heat (J/kg*K)	Thermal conductivity (W/m*K)
Aluminum	2702	903	237
Lithium	2267	1016	1

Figure 14. Temperature fields.

Figure 15. The temperature of the cooling water: along the length of the pipe and a detailed.

Figure 14 shows the cell distribution in the computational domain. The mesh has 566,646 elements and the physical properties considered in the model are summarized in Table 2. The characteristics of aluminum cold plates and lithium cells are in Table 3. The final solution of the simulation is obtained with 6200 iterations.

It can be clearly seen how the temperature rises as the water progresses through the system. Even so, most of the water increases only slightly in temperature (Figures 14 and 15). The highest values correspond to the small fraction that is in contact with the aluminum plate. Figure 16 shows the temperatures at the centre of each of the 8 cells.

Figure 16. Model chart with the internal temperatures of the 10 cells.

4. SYSTEM OF EXTRUDED ALUMINUM COLD PLATES

The second option that will be discussed is based on aluminum extrusions, the option of producing solid sections being very costly in materials and hence expensive. In the extrusion process, material is pushed or drawn through a die of the required cross-section. This method has several advantages, such as being able to create complex sections and obtain an excellent surface finish.

In Figures 17 and 18, can be seen both the new geometry and the meshing of the complete model. This total system has 757,477 elements.

Figure 17. Geometry of the extruded aluminum cold plate.

Figure 18. Geometry and meshing system with extruded aluminum cold plates.

Figure 19. Temperature fields.

The boundary conditions and physical properties considered are equivalent to those specified above for the system of solid cold plates (Figure 19). As for initial conditions, a temperature of 298 K (25°C) corresponding to room temperature, is specified for the entire system, while the inlet velocity of the cooling water is set at 0.33 m/s. It must also set the "Energy Source" property for cells. This is the amount of heat generated by the cells divided by the volume of the cell (W/m³ ).

The heat generated, obtained by theoretical calculations, is 20.83 W as Equation (20).

The heat generated, obtained by theoretical calculations, is 20.83 W as Equation (20).

Energy Source = $$\frac{\text{Heat Generated (W)}}{\text{Volume of the cell (m}^3)} = \frac{20.83 \text{ W}}{\frac{\pi \times (0.0406 \text{ m})^2}{4} \times 0.1388 \text{ m}} = 115919.86 \frac{\text{W}}{\text{m}^3}$$ (20)

After calculating the full solution with 6200 iterations, the temperature field illustrated in Figure 20 was obtained. It can be seen how refrigeration works at the surface of the cells, the temperature decreasing as it moves away from the inside (Table 4).

| Cell-surface temperatures | | | | | Cell-internal temperatures | | | | |----------------------------|-----------|---------|----------------------------|--------|----------------------------|---------|----------|--| | Entire cells
Half cells | | | Entire cells
Half cells | | | | | | | Cell 2 | 305.873 K | Cell 7 | 305.475 K | Cell 2 | 313.83 K | Cell 7 | 313.39 K | | | Cell 4 | 305.644 K | Cell 9 | 306.324 K | Cell 4 | 313.57 K | Cell 9 | 314.14 K | | | Cell 5 | 305.811 K | Cell 6 | 306.607 K | Cell 5 | 313.75 K | Cell 6 | 314.50 K | | | Cell 1 | 306.103 K | Cell 10 | 306.858 K | Cell 1 | 314.02 K | Cell 10 | 314.74 K | | | Cell 3 | 306.646 K | Cell 8 | 306.528 K | Cell 3 | 314.64 K | Cell 8 | 314.43 K | |

Table 4. Cell Temperatures

CONCLUSION

There is not a notable difference in temperatures between the systems with solid cold plates and with cold plates formed by extrusion. In both cases the internal temperatures of the cells do not exceed the operating range specified by the manufacturer, and hence both systems could be considered valid approaches to cooling batteries for electric vehicles.

All things being equal, however, it could be argued that the system using cold plate formed by extrusion is the best candidate, since it represents a considerable saving both in materials, aluminium in this case, and in costs.

REFERENCES

• [1] Vayrynen, A.; Salminen, J.; LithiumIon Battery Production. The Journal of Chemical Thermodynamics, vol. 46, pp. 80-85, March 2012.

• [2] Chacko, S.; Chung, Y.M.; Thermal Modeling of Li-Ion Polymer Battery for Electric Vehicle Drive Cycles. Journal of Power Sources, vol. 213, pp. 296-303, 1 September 2012.

• [3] Wang, Q.; Ping, P.; Zhao, X.; Chu, G.; Sun, J.; Chen, Ch.; Thermal Runaway Caused Fire And Explosion of Lithium Ion Battery. Journal of Power Sources, vol. 208,pp. 210-214, 2012.

• [4] Santee, S.; Xiao, A.; Yang, L.; Gnanaraj, J.; Lucht; B.L.; Effect of Combinations of Additives on the Performance of Lithium Ion Batteries. Journal of Power Sources, vol. 194, issue 2, pp. 1053-1060, 1 December 2009.

• [5] Polk, R.C.; Villarreal, R.; Sunderlin, T.A.; Lithium Ion Battery Pack Having Passive Cooling. April 1,2011,www.epo.org, page bookmark WO2010141854 (A1).

• [6] Maier, G.;Michelitsch, M.; Modular Battery System with Cooling System. April 1, 2011,www.epo.org, page bookmark US2011033742 (A1).

• [7] Song, T.W.; Park, T.S.; Oh, D.J.; Jeong, J.Y.; Cooling Device of Battery Pack. April 1,2011,www.epo.org, page bookmark US2010136406 (A1).

• [8] Mahamud, R.; Park, Ch.;Reciprocating Air Flowfor Li-Ion Battery Thermal Management to Improve Temperature Uniformity. Journal of Power Sources, vol. 196, issue 13, pp. 5685-5696, 1 July 2011.

• [9] Zhang, Y.P.; Yu, X.L.; Feng, Q.K.; Zhang, R.T.; Thermal Performance Study of Integrated Cold Plate with Power Module. Applied Thermal Engineering, vol. 29, issues 17-18, pp. 3568-3573, December 2009.

• [10] www.aavid.com/product-group/liquidcoldplates.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 17

Ni-Cd BATTERIES FOR AUTOMOTIVE APPLICATIONS: THE ac ELECTRICAL DATA OF THE Ni-Cd BATTERY IN THE HAVRILIAK-NEGAMI DIELECTRIC REPRESENTATION

Mohammad A. Alim¹* , Sudip Bhattacharjee² , Sanjida Khanam³ , Nasiha Muna⁴and Sabiha Runa⁵

¹Department of Electrical Engineering, Alabama A&M University, Huntsville, Alabama, US ²Department of Mechanical and Civil Engineering, Alabama A&M University, Huntsville, Alabama, US 3 ITT Technical Institute, Madison, Alabama, US ⁴Department of Mathematics, University of Alabama in Huntsville, Huntsville, Alabama, US ⁵Department of Chemical and Materials Engineering, College of Engineering, University of Alabama in Huntsville, Huntsville, Alabama, US

ABSTRACT

The non-Debye dielectric relaxation via Havriliak-Negami (H-N) empirical relation has been observed for the ac small-signal electrical data of the Ni-Cd battery in the impedance plane. Various analytical aspects of the same data are demonstrated using the Davidson-Cole (D-C) representation. It is clear that the continuous relaxation process corresponding to high frequencies yields skewness beside the breadth of the H-N distribution function. Possible meaning of the depression parameters α and β is extracted from the breadth and skewness or deviation of the semicircular relaxation loci which is normally observed for either Debye or Cole-Cole dielectric relaxation. Both α and β are overall responsible for the skewness of the loci in the high frequency segment. In reality α and β cannot be separated and viewed as individual parameters causing breadth and

E-mail: mohammad.alim@aamu.edu.

skewness in conjunction with the presence of the apparent depression angle for the skewed relaxation loci.

Keywords: Complex Planes, Relaxation Time, Equivalent Circuit, Impedance Spectroscopy, Immittance, Debye and Non-Debye Relaxations, Cole-Cole Relaxation, Davidson-Cole Relaxation, Havriliak-Negami Relaxation, Ni-Cd Battery

1.INTRODUCTION

Batteries operate by converting chemical energy into electrical energy through electrochemical discharge reactions. Interest exists in compact size solid-state thin-film batteries that serve energy-handling issues for various applications associated with cell phones, camcorders, lap-tops, toys, etc. popular electronic systems. Basically a battery is a device that stores electrical energy which is consumed. It is, in general, used as a single unit building block cell possessing termination and insulation. In reality a battery refers to a series combination of individual cells assembled in a pack and provided with some means for connecting to the system it serves. The attributes of the ideal battery includes: (a) high energy density, (b) long life, (c) rechargeable capability, (d) safe handling, (e) flexibility in applications, etc. There are two types of batteries available for use: (1) primary type discharged after single usage and discarded, and (2) secondary type for repeated use which is rechargeable by reversing the chemical reaction within it. The primary batteries are not of concern in this context as the rechargeable batteries are drawing consistently increasing interest for handling large energy density [1-4].

There are several types of secondary batteries drawn attention depending on the nature of the applications. The Hg-Zn and Hg-Cd batteries got popularity for the miniaturized feature. The Li-based batteries are used to power the microelectronic devices. There are two basic rechargeable type batteries used in the commercial arena. The first type is very commonly used as the automotive starting battery which is maintenance free during the service life. The second type is sealed and does not allow venting of gas to the atmosphere. This battery requires that the gasses generated when charging the cell be recombined as part of the process. This recombinant technology is employed in all sealed Ni-Cd and in some sealed lead cell types. Other types of rechargeable batteries require partial sealed structure having electrolyte immobilization scheme to reduce the possibility of acid leakage. These batteries are open to the atmosphere and release gases during charging and discharging [2-4].

The Ni-Cd batteries showing outstanding features of long durability and high power performance beside very quick discharging and charging, and low cost have been in use in the cordless electronic products and expanded the market for several decades as it reduces the environmental effects significantly [3]. Three fundamental components of Ni-Cd battery are the nickel species as the positive electrode active material, the cadmium species as the negative electrode active material, and the aqueous alkali solution as electrolyte. In order to ensure the sealing by means of the oxygen absorption by the negative electrode, the capacity of the negative electrode is designed to be larger than that of the positive electrode so that the negative electrode does not reach a state of full charge. Another advantage of Ni-Cd batteries is the wide variety of the product line up which permits a diversity of applications [4].

The dielectric behavior of the Ni-Cd battery is of interest concerning characterization and determining functional capabilities as it handles high energy density. It is used for the acquisition of the ac small-signal electrical data in the form of immittance (impedance or admittance). These data are analyzed in the complex plane formalisms, and then distinguished between the Havriliak-Negami and the previously [5] analyzed Davidson-Cole representations.

2. NI-CD BATTERY: STATUS AND PROSPECT

The continuous rechargeable batteries are used in the spacecraft [6] hybrid electric vehicle (HEV) [7] for the capability of handling high power capacity and high energy density. The major parameters adjudged for the rechargeable batteries are: (1) kilowatt hour, (2) massto-volume ratio, and (3) operating (environmental) temperature. Based on functional capabilities Ni-Cd batteries indicate better performance, operating temperature range, and durability [8]. The Li-ion batteries though handle large energy density but charging cycle life is less than the Ni-Cd batteries. Also Li-ion batteries have a chance to get accidental ignition or violent reaction while evolving hydrogen gas. Otherwise Li-ion batteries, in general, are safe to use in electronics. Other types of rechargeable batteries are not considered in analyzing the dielectric data in the present investigation because those batteries possess less advantage over Li-ion and Ni-Cd batteries when energy density, mass-to-volume ratio, and safety aspects are considered.

The applications of the Ni-Cd batteries for high power requirements satisfy temperature extremes, severe deep cycling, and mechanical (pressure or thrust or stress) abuse. Other continuous rechargeable batteries exhibit high heat dissipation during charging, and very limited tolerance to over-discharge. When the design capacity is low for the negative electrode so that it reaches fully charged state, then a sharp pressure rise occurs in the Ni-Cd battery. Since the reaction of oxygen with hydrogen evolve during over-charging of negative electrode is slow, hydrogen evolution itself causes a serious pressure rise inside the battery [4].

The Ni-Cd rechargeable battery uses nickel-oxide-hydroxide and metallic cadmium as electrodes with alkaline electrolyte (such as potassium hydroxide) material [4]. The Ni-Cd batteries suffer *memory effec*t when discharged and recharged at the same state of charge (SOC) several times. This means when a battery is repeatedly recharged before it has discharged more than half of its power, it will "forget" its original power capacity. Thus, the battery is capable of remembering charging cycle at the recharging point. While in use, the Ni-Cd battery suffers abrupt drop in voltage as if the battery is discharged. Thus, memory can be attributed to changes in the cadmium electrode reflecting charging involving conversion of cadmium hydroxide to metallic cadmium which is microcrystalline (polycrystalline). This is the nucleation process to get enlarged Cd polycrystals in the Cd electrode. Scientific literature concerning this aspect is not documented but interchangeably words such as whiskers or dendrites or crystals, etc. are used. This is undesirable for the Ni-Cd battery since it makes Cd harder to dissociate during the high current discharge, and leads to high internal resistance attributing to the Cd polycrystals having variation in grain-size.

The aforementioned difficulties identified require extensive research for the Ni-Cd batteries in order to bring complete satisfaction form the users. Thus, improvement of the Ni-Cd battery becomes obvious with the materials development [9]. For continuous rechargeable applications the Ni-Cd batteries have potential due to longer time usage in the HEV and longrange aircraft. This is primarily due to the safety and recycling charging issues despite some minor problems with the memory effect.

3. THIN-FILM STACK OR LAYER BATTERIES

Battery power can be either increased or optimized by introducing parallel structure of the stacks or layers as opposed to a single stack or layer. Multiple stacks or layers reflect parallel capacitance-like structure as found in most junction devices. Stacks or layers cause reduction in battery impedance by introducing more equivalent cell area which aids in voltage regulation for the high current applications. Possible length or mass is likely to increase slightly due to the thin-film stacks or layers. Thick-film stack or layer can be used as stack or layer in the form of pellets due to the easiness of making pellets. Any kind of reduction in materials for making batteries is a reflection of saving in the eventual manufacturing. This is the overall fundamental concept of thin-film based stack or layer batteries.

The thin-film based Ni-Cd structure can be evaluated to compare the performance with respect to the Li-ion technologies. This structure reflects Metal Foil Substrate as Cathode Current Collector surrounded by a bi-metallic oxide material as Cathode having insulation between them. The metallic electrode surrounds this structure as Anode. This basic structure can be repeated sequentially to form multi-layers having Cathode Current Collector terminal and Metal Anode Encapsulation with terminal for use [1, 10, 11]. Using the fundamental concept of the thin-film technology, the Ni-Cd structure is expected to encounter severe competition with the alternate technologies such as NiMH (nickel-metal-hydride) based battery having very less memory effect. Electrode based on sintered or fiber material in the Ni-Cd thin-film structure can be used in the NiMH technology.

Due to the memory effect, there is a great prospect for the NiMH technology [12] in lieu of the Ni-Cd battery. This needs to be examined employing the acquisition of the dielectric data and subsequent analysis [13] via the powerful novel tool/technique [13] known as the Impedance or Dielectric Spectroscopy (IS/DS). The analysis of the dielectric data employing complex plane formalisms may indicate direction for optimized performance of the NiMH technology.

The comparison of the dielectric data for the Ni-Cd battery with the NiMH battery may indicate overall performance. These data may play an important role in minimizing the memory effect in the Ni-Cd battery or aid in complete removal of the memory effect in the NiMH battery. This is not done as current concentration remained in understanding the dielectric response via the data analysis and possible interpretation for the underlying operative mechanisms. Various aspects of the relaxation mechanisms are explained utilizing relevant complex functions and subsequent derivations from the analyses.

4. BACKGROUND AND ANALYTICAL APPROACH OF THE DIELECTRIC DATA

The relaxation behavior of the ac small-signal electrical data of the Ni-Cd battery is well understood in the complex plane formalism where the imaginary part of the immittance function is plotted against the real part [13]. The ideal or Debye relaxation process from the measured ac small-signal electrical data is very rarely observed [14-16] in the engineering material systems and devices. The immittance response (I * ) for the Debye behavior [16] given by

$$I^{*}(\omega) = I{\infty} + (I{dc} - I_{\infty}) / (1 + j\omega\tau)$$ (1)

is represented as the ideal response in the form of semicircular loci in the complex plane formalism where the center lies on the real axis (x-axis). The non-ideal or non-Debye behavior is usually observed via depression parameter and/or skewed nature of the semicircular loci. In the Cole-Cole(C-C) or ideal-like non-Debye representation [16] the semicircular response is depressed without skewing the loci or the relaxation curve. The other non-Debye type responses occasionally observed via skewed behavior [17-20] of the semicircular loci (curve) where the depression parameter is not accounted alone for that non-Debye response. Overall, non-Debye dielectric response is often described as simplistic C-C relaxation [16] as:

$$I{CC}^{*}(\omega) = I{CC\infty} + \frac{(I{CCs} - I{CC\infty})}{1 + (j\omega\tau_{CC})^{\alpha}}$$

$$\tag{2}$$

while the complex non-Debye dielectric response is often described as the Davidson-Cole 17, 18relaxation in the form

$$I{DC}^{*}(\omega) = I{DC\infty} + \frac{(I{DCs} - I{DC\infty})}{(1 + j\omega\tau_{DC})^{\beta}}$$

$$\tag{3}$$

and Havriliak-Negami 20, 21 relaxation in the complex plane formalism as

$$I{HN}^{*}(\omega) = I{HN\infty} + \frac{(I{HNs} - I{HN\infty})}{\left[1 + (j\omega\tau_{HN})^{\alpha}\right]^{\beta}}$$

$$\tag{4}$$

where the measurement frequency increases in the anti-clock wise direction along the relaxation curve. The subscript CC or DC or HN represents Cole-Cole or Davidson-Cole or Havriliak-Negami relaxation, respectively. The subscript s denotes dc (or static condition) measurement corresponding to f → 0 Hz, subscript ∞ indicates high frequency measurement corresponding to f → ∞, and τ is the relaxation time for respective situation.

The depression parameter for the simplistic non-Debye (i.e., C-C) behavior is derived from the semicircular relaxation [13, 16]. This is not the case for the relaxations in D-C and in H-N loci. The curve fitting procedure [13] of the data points in the complex plane formalism for the semicircular relaxation provides extraction of the associated parameters.

In an earlier work [5], the D-C depression parameter was found to be complex quantity where real and imaginary parts of the D-C depression parameter were illustrated for the Ni-Cd battery. In this work, extraction of the depression parameters is demonstrated for the H-N relaxation using real quantity as well as complex quantity for the same ac small-signal electrical data. The degree of precision depends on the user and the meaning of the extracted parameters in the complex domain is not ascertained. However, the interpretation of the extracted parameters is likely to remain the same as does exist for conventional analysis.

4.1. Dielectric Relaxation: Ideal (Debye)and Cole-Cole (Non-Ideal) Responses

The ideal or Debye relaxation is straightforward from the admittance expression [18, 19]

$$Y^* = G_p + j\omega C_p \tag{5}$$

where the resistance Rp[(= 1/Gp) or the conductance G^p (= 1/Rp)] is in parallel with the capacitance, Cp. The parallel Rp-Cp(or *R*1-*C*1) combination constitutes the Debye relaxation time τ (= RpC^p = *R*1*C*1). The semicircular relaxation function F * (ω) is given by

$$F^*(\omega) = \frac{1}{1 + j\omega\tau} \tag{6}$$

where

$$\tau = \frac{1}{\omega{peak}} = \frac{1}{2\pi f{peak}} \tag{7}$$

satisfying

$$\omega{peak} \, \tau = \omega{peak} \tau_{peak} = 1 \tag{8}$$

when represented in the complex plane reveals zero depression angle. The ideal semicircular relaxation containing an intercept on the left-side of the semicircle is presumed to be zero for F * (ω) in Equation (6). The existence of a finite intercept on the left-side of the semicircle may be referred to as the left-intercept. In representing a semicircular behavior the ordinate (*y*axis) and abscissa (x-axis) must have the same plotting grid scale so that the relaxation in the complex plane can clearly be delineated. This means that each unit grid or graphical segment in length on the ordinate must be equal to the abscissa.

Figure 1. Generic representation of the relaxation curves complex plane for the relaxation function F ( ) given in Equation (11): (1) Debye formalism with depression angle θ = 0^o or α = 1 and β = 1, (2) Non-Debye Cole-Cole formalism with depression angle θ = 9^o or α = 0.9 and β = 1, (3) Davidson-Cole formalism with α = 1 and β = 0.5, (4) Havriliak-Negami formalism with α = 0.85 and β = 0.45, (5) Havriliak-Negami formalism with α = 0.45 and β = 0.85.

A semicircle whose center lies on the x-axis is defined as Debye or Debye-like relaxation, whereas for a non-Debye relaxation the center of the semicircle lies below the x-axis giving a finite depression angle (θ) measured from the point at the left-intercept to the center of the semicircle below the x-axis. This is displayed in Figure 1. The depression angle gives rise to the amount of the depression demonstrated originally by Cole-Cole [17] as the ratio of θ and /2 (= 90^o ) [13]. Thus, the Cole-Cole type depression parameter h is given by

$$h = \frac{\theta}{\pi/2} = \frac{2\theta}{\pi} \tag{9}$$

The C-C depression parameter h ranges between 0 and 1 depending on the idealized Debye and/or extreme simplistic non-Debye response of a semicircular relaxation. In Equations (2) and (4) the parameter α is related to the Cole-Cole type depression parameter h showing dependence to each other as:

$$1 - h = \alpha \tag{10}$$

The value of α ranges between 0 and 1. The Equation (2) takes Debye form as given by Equation (1) when *h*→0. This implies that when α = 1 Equation (2) leads to Debye form as given in Equation (1). The same Equation (2) leads to a point on the x-axis when *h*→1. This implies that when α = 0 Equation (2) leads to extreme non-Debye response as a point on the x-axis. From Equation (10) it is observed that as *h*0 then *α*1, and likewise as *h*1 then *α*0. The ideal Debye or Debye-like behavior can be achieved as *θ*0 o implying *α*1, and an extreme simplistic non-Debye response can be visualized with *θ*90^o implying *α*0. That is why α can be termed as the "depression parameter" in the present context.

4.2. Analysis of the Havriliak-Negami Representation

The H-N relaxation function can be ascribed from Equation (4) representing the skewed relaxation curve (arc) given by

$$F^*(\omega) = 1/[1 + (j\omega\tau)^{\alpha}]^{\beta} \tag{11}$$

$$F^{*}(\omega) = 1/\left[1 + \left(\omega\tau\right)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right) + j\left(\omega\tau\right)^{\alpha}\sin\left(\frac{\alpha\pi}{2}\right)\right]^{\beta}$$ (12)

Both depression and skewed nature of the H-N relaxation curve are represented by the parameters α and β. In order to separate real and imaginary parts of Equation (11) F*(ω) is expressed in the form:

$$F^*(\omega) = \frac{1}{\left(A + jB\right)^{\beta}} \tag{13}$$

with

$$A = 1 + \left(\omega \tau\right)^{\alpha} \cos\left(\frac{\alpha \pi}{2}\right) \tag{14}$$

$$B = (\omega \tau)^{\alpha} \sin\left(\frac{\alpha \pi}{2}\right) \tag{15}$$

Then

$$F^*(\omega) = \frac{1}{\left(A + jB\right)^{\beta}} = \frac{1}{\left(X \exp\left(j\gamma\right)\right)^{\beta}}$$ (16)

$$F^{*}(\omega) = X^{-\beta} \exp(-j\beta\gamma) = X^{-\beta} \left[\cos(\beta\gamma) - j\sin(\beta\gamma)\right]$$ (17)

where

$$X^{2} = A^{2} + B^{2} = \left[1 + \left(\omega\tau\right)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right)\right]^{2} + \left(\omega\tau\right)^{2\alpha}\sin^{2}\left(\frac{\alpha\pi}{2}\right)$$ (18)

$$X^{2} = 1 + (\omega \tau)^{2\alpha} + 2(\omega \tau)^{\alpha} \cos\left(\frac{\alpha \pi}{2}\right)$$ (19)

and

$$\tan(\gamma) = \frac{B}{A} = \frac{(\omega \tau)^{\alpha} \sin\left(\frac{\alpha \pi}{2}\right)}{1 + (\omega \tau)^{\alpha} \cos\left(\frac{\alpha \pi}{2}\right)}$$ (20)

Thus, real and imaginary parts become:

$$F' = X^{-\beta} \cos(\beta \gamma) \tag{21}$$

and

$$F'' = X^{-\beta} \sin(\beta \gamma) \tag{22}$$

$$\frac{F''}{F'} = \tan(\beta \gamma) \tag{23}$$

The H-N response can be achieved either in the impedance plot or in the complex capacitance plot where the measurement frequency moves anti-clockwise for each measurement point on the curve representing the total data. The presence of the intercept on the left side of the relaxation arc has a finite magnitude while the relaxation portion can be expressed as in empirical Equation (11). Like the D-C representation the skewed nature of the generic H-N curve is shown in Figure 1. The time constant is not necessarily corresponding to the conventional ωτ = 1 at the peak of the skewed arc. Thus, ωτ = 1 is likely to be dubious while deriving the simplified equivalent R-C(resistance-capacitance) circuit for the relaxation function F * (ω) shown in Figure 2.

Differentiating imaginary part *F*˝(ω) in Equation (22) with respect to (ωτ) α

Differentiating imaginary part $$F''(\omega)$$ in Equation (22) with respect to $(\omega \tau)^{\alpha}$ : $$\frac{dF''}{d(\omega \tau)^{\alpha}} = -\beta X^{-(\beta+1)} \sin(\beta \gamma) \frac{dX}{d(\omega \tau)^{\alpha}} + \beta X^{-\beta} \cos(\beta \gamma) \frac{d\gamma}{d(\omega \tau)^{\alpha}}$$ (24)

where

$$\frac{dX}{d(\omega\tau)^{\alpha}} = \frac{1}{2} \left[ 1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha} \cos\left(\frac{\alpha\pi}{2}\right) \right]^{-\frac{1}{2}} 2 \left[ (\omega\tau)^{\alpha} + \cos\left(\frac{\alpha\pi}{2}\right) \right]$$ (25)

Figure 2. Simplified equivalent circuit for the semicircular relaxation in the impedance plane presuming either only Debye or non-Debye (Cole-Cole) response.

$$\frac{dX}{d(\omega\tau)^{\alpha}} = \left[1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha} \cos\left(\frac{\alpha\pi}{2}\right)\right]^{-\frac{1}{2}} \left[(\omega\tau)^{\alpha} + \cos\left(\frac{\alpha\pi}{2}\right)\right]$$ (26)

with

$$\gamma = \tan^{-1} \left( \left[ (\omega \tau)^{\alpha} \sin \left( \frac{\alpha \pi}{2} \right) \right] / \left[ 1 + (\omega \tau)^{\alpha} \cos \left( \frac{\alpha \pi}{2} \right) \right] \right) = \tan^{-1} \varphi$$ (27)

Differentiating γ in Equation (27) with respect to (ωτ) α

$$\frac{d\gamma}{d(\omega\tau)^{\alpha}} = \frac{1}{1+\varphi^2} \cdot \frac{d\varphi}{d(\omega\tau)^{\alpha}}$$ (28)

where

2 2 2 2 2 2 ( ) sin 1 ( ) 2( ) cos 2 2 1 1 1 ( ) cos 1 ( ) cos 2 2 (29)

and

$$\frac{d\varphi}{d(\omega\tau)^{\alpha}} = \frac{d}{d(\omega\tau)^{\alpha}} \left[ \frac{(\omega\tau)^{\alpha} \sin\left(\frac{\alpha\pi}{2}\right)}{1 + (\omega\tau)^{\alpha} \cos\left(\frac{\alpha\pi}{2}\right)} \right] = \frac{\sin\left(\frac{\alpha\pi}{2}\right)}{\left[1 + (\omega\tau)^{\alpha} \cos\left(\frac{\alpha\pi}{2}\right)\right]^{2}} \tag{30}$$

Rearranging Equation (28) it is obtained:

$$\frac{d\gamma}{d(\omega\tau)^{\alpha}} = \frac{\sin\left(\frac{\alpha\pi}{2}\right)}{1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right)}$$ (31)

In order to obtain the maximum of *F*˝(ω), setting Equation (24) as:

$$\frac{dF''}{d(\omega\tau)^{\alpha}} = 0 \tag{32}$$

Thus,

Ni-Cd Batteries for Automotive Applications 463 $$\frac{dF''}{d(\omega\tau)^{\alpha}} = -\beta X^{-(\beta+1)} \sin(\beta\gamma) \frac{dX}{d(\omega\tau)^{\alpha}} + X^{-\beta} \beta \cos(\beta\gamma) \frac{d\gamma}{d(\omega\tau)^{\alpha}} = 0$$ (33)

$$\frac{dF''}{d(\omega\tau)^{\alpha}} = -X^{-1}\sin(\beta\gamma)\frac{dX}{d(\omega\tau)^{\alpha}} + \cos(\beta\gamma)\frac{d\gamma}{d(\omega\tau)^{\alpha}} = 0$$ (34)

$$d(\omega\tau)^{\alpha} = -\left[1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right)\right]^{-\frac{1}{2}}\sin(\beta\gamma)\left[1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right)\right]^{-\frac{1}{2}} \times \left[(\omega\tau)^{\alpha} + \cos\left(\frac{\alpha\pi}{2}\right)\right] + \frac{\cos(\beta\gamma)\sin\left(\frac{\alpha\pi}{2}\right)}{1 + (\omega\tau)^{2\alpha} + 2(\omega\tau)^{\alpha}\cos\left(\frac{\alpha\pi}{2}\right)} = 0$$ (35)

Switching the quantities of both sides of the equal sign Equation (35) becomes:

Switching the quantities of both sides of the equal sign Equation (35) becomes: $$\left[1 + (\omega \tau)^{2\alpha} + 2(\omega \tau)^{\alpha} \cos\left(\frac{\alpha \pi}{2}\right)\right]^{-1} \sin(\beta \gamma) \left[(\omega \tau)^{\alpha} + \cos\left(\frac{\alpha \pi}{2}\right)\right] = \frac{\cos(\beta \gamma) \sin\left(\frac{\alpha \pi}{2}\right)}{1 + (\omega \tau)^{2\alpha} + 2(\omega \tau)^{\alpha} \cos\left(\frac{\alpha \pi}{2}\right)}$$ (36)

Rearranging Equation (36):

$$\sin(\beta\gamma)\left[(\omega\tau)^{\alpha} + \cos\left(\frac{\alpha\pi}{2}\right)\right] = \sin\left(\frac{\alpha\pi}{2}\right)\cos(\beta\gamma) \tag{37}$$

Thus, at the maximum value of F˝, βγ = (βγ)peak, ωτ = (ωτ)peak, and then Equation (37) becomes:

$$\tan(\beta \gamma){peak} = \frac{\sin\left(\frac{\alpha \pi}{2}\right)}{(\omega \tau){peak}^{\alpha} + \cos\left(\frac{\alpha \pi}{2}\right)}$$ (38)

Equation (38) represents the condition when *F˝ = F˝*max. When α → 1, then Equation (38) leads to:

$$\left(\omega\tau\right){peak} \tan\left(\beta\gamma\right){peak} = 1 \tag{39}$$

And Equation (20) yields

$$\tan(\gamma{peak}) = (\omega \tau){peak} \tag{40}$$

Then Equation (39) becomes D-C response [5, 20, 21]:

$$\tan(\gamma{peak})\tan(\beta\gamma){peak} = 1 \tag{41}$$

Expression of β appears as real quantity for α being real using Equations (38):

$$\beta = \left(\frac{1}{\gamma{peak}}\right) \arctan \left[\frac{\sin\left(\frac{\alpha\pi}{2}\right)}{(\omega\tau){peak}^{\alpha} + \cos\left(\frac{\alpha\pi}{2}\right)}\right]$$ (42)

where, γ and απ are in radians; and (ωτ)peak corresponds to the maximum value of *F*˝(ω).

4.3. Geometrical Interpretation of H-N Relaxation at the Limiting Case

Figure 3(a) displays a generic representation of the H-N relaxation formalism. This feature has a similarity with the D-C feature [17, 18] with α = 1. Equation (23) gives the ratio between the imaginary and real part of the H-N relaxation function yielding tan (βγ). From Equation (20) it is seen that tan (γ) increases as ω increases anti-clockwise, i.e., the angle βγ increases. Further examination of the imaginary and real parts of the H-N relaxation function and imposing limiting case, Equation (23) gives [19, 20]:

$$\lim{\omega \to \infty} \frac{F''}{F'} = \lim{\omega \to \infty} \tan(\beta \gamma) = \tan(\beta \gamma_{\text{max}})$$ (43)

$$\tan \gamma = \lim{\omega \to \infty} \left( \frac{\sin\left(\frac{\alpha\pi}{2}\right)}{\left(\omega\tau\right)^{-\alpha} + \cos\left(\frac{\alpha\pi}{2}\right)} \right) = \tan\left(\frac{\alpha\pi}{2}\right) = \tan\left(\gamma{\text{max}}\right) \tag{44}$$

$$\gamma{\text{max}} = \gamma{\text{max}} = \frac{\alpha \pi}{2} \tag{45}$$

Combining Equations (43) and (45) it is obtained:

$$\lim{\omega \to \infty} \frac{F''}{F'} = \tan(\beta \gamma) = \tan(\beta \gamma{\max}) = \tan\left(\frac{\alpha \beta \pi}{2}\right)$$ (46)

where

$$\beta \gamma{\text{max}} = (\beta \gamma){\text{max}} = \frac{\alpha \beta \pi}{2} \tag{47}$$

The product *α*×β has a proven connection with the Kohlrausch-Williams-Watts (KWW) function illustrated by Alvarez et al. [21]. Equation (44) indicates the ratio between imaginary and real part of the H-N relaxation function attains its maximum value when *ω*→∞. This situation is shown in Figure 3(a) where the bisector of the angle *βγ*max intersects the relaxation curve at the point ωτ= 1. The angle created by this bisector with the x-axis is given by:

$$\frac{\beta \gamma_{\text{max}}}{2} = \frac{\alpha \beta \pi}{4} \tag{48}$$

Now, when ωτ = 1, Equation (20) yields:

$$\tan \gamma \big|_{\omega r = 1} = \frac{\sin\left(\frac{\alpha \pi}{2}\right)}{1 + \cos\left(\frac{\alpha \pi}{2}\right)} = \tan\left(\frac{\alpha \pi}{4}\right) \tag{49}$$

Therefore,

$$\gamma\big|{\omega\tau=1} = \frac{\alpha\pi}{4} = \frac{\gamma{\text{max}}}{2} \tag{50}$$

and

$$\tan\left(\frac{\beta\gamma{\max}}{2}\right) = \tan\left(\frac{\beta\alpha\pi}{4}\right) = \tan\left(\beta\gamma\big|{\omega\tau=1}\right) = \frac{F''}{F'}\bigg|_{\omega\tau=1}$$ (51)

Equation (51) proves that the bisector of the angle *βγ*max intersects the H-N relaxation curve at a point where ωτ = 1. This important information can be used to determine the value of τ.

Gerischer impedance [22] representation in the impedance plane lies for α = 1 and β = 0.5 which is a special case. This impedance gives ωτ = 1 in the low-frequency domain whereas the peak of the loci gives ωτ = 3 ≈ 1.732 and the corresponding coordinate gives *F'*peak = F'ωτ=1.732= 3 / 8 ≈ 0.6124 and *F"*peak = F"ωτ=1.732 = 1/ 8 ≈ 0.3536. Since the Ni-Cd battery data indicated β = 0.6 with α < 1, the D-C relaxation formalism is not emphasized like Reference 8. Figure 3(b) refers to the D-C formalism while Figures 3(c) and 3(d) represent the H-N formalism. Thus, α = 0.999 and β = 0.6 as well as α = 0.995 and β = 0.6 indicate H-N representation of the experimental data.

Figure 3(a). Havriliak-Negami representation of the relaxation function F ( ) per Equation (6) where measurement frequency runs from right to the left.

Figure 3(b). Havriliak-Negami representation of the relaxation function F ( ) with α = 1, β = 0.6 that corresponds to *F'*peak = F'ωτ=1.497 = 0.5843 and *F"*peak = F"ωτ=1.497 = 0.3905 along with F'ωτ=1 = 0.7237, F"ωτ=1 = 0.3888.

Figure 3(c). Havriliak-Negami representation of the relaxation function () F with α = 0.999, β = 0.6 that corresponds to *F'*peak = F'ωτ=1.498 = 0.5842 and *F"*peak = F"ωτ=1.498 = 0.3899 along with F'ωτ=1 = 0.7236, *F"*ωτ=1 = 0.3682.

Figure 3(d). Havriliak-Negami representation of the relaxation function () F with α = 0.995, β = 0.6 that corresponds to *F'*peak = F'ωτ=1.501 = 0.5841 and *F"*peak = F"ωτ=1.501 = 0.3877 along with F'ωτ=1 = 0.7229, F"ωτ=1 = 0.3662.

4.4. Extraction of the Relaxation Time τ and the H-N Depression Parameters α and β

The H-N impedance representation given by

$$Z^{*}(\omega) = Z{HF} + \frac{Z{LF} - Z_{HF}}{[1 + (j\omega\tau)^{\alpha}]^{\beta}}$$ (52)

can be ascribed as D-C representation for α = 1 with ZHF = high frequency (ω → ∞) impedance, and ZLF = low frequency (ω → 0 or dc) impedance. The generic representation of both H-N and D-C responses shown in Figure 1 depicts the real world situation for the Ni-Cd battery impedance data shown in Figure 4 as the H-N response. The same data approximate both D-C and H-N representations within the close range of the value of α. The iteration process of the fitting of each relaxation illustrated in Figure 4 was previously employed for the D-C response [5]. For β = 0.6 three separate values of α provide near identical fitting. Thus, two H-N representations of Figure 4 cannot be easily distinguished from the D-C representation for the same data. The error in each fitting was brought to as minimum as possible. Table 1 shows the observed and calculated (predicted) values of the real and imaginary part of the impedance for both D-C and H-N responses. The errors are indicated therein.

The values of ZLF, ZHF, α, and β are obtained by trial and error for the measured data of the Ni-Cd battery previously represented as the D-C dielectric behavior [5, 19]. Figures 5 and 6 represent the ratio of the imaginary to the real impedance of the relaxation domain versus angular frequency for H-N response and D-C response, respectively. For the H-N case, when α = 0.995 and β = 0.6, then *β*γmax = *αβ*π/2 = 0.9378 and tan(*βγ*max) = 1.3628. For the D-C case, when α = 1 and β = 0.6, then *βγ*max = *αβ*π/2 = 0.9425 and tan(*βγ*max) = 1.3764. Realistically the horizontal tangent on the curve in Figure 5 and/or 6 yields tan(*βγ*max) = 1.2449. This finite difference between the theoretical and the experimental values is attributed to the difference between the tangent of the line containing experimental data points and the tangent of the fitted curve, respectively.

Figure 4. Havriliak-Negami representation of the Ni-Cd battery via impedance plot per reference 9 for (a): α = 1 and β = 0.6; (b): α = 0.999 and β = 0.6; and (c): α = 0.995 and β = 0.6.

In order to determine the value of relaxation time τ for the H-N response as well as for the D-C response, Equations (48) and (51) are used. According to Equation (51), a horizontal line is drawn for tan(*βγ*max/2) which intersects the experimental curve at a point from which a perpendicular line is drawn on the abscissa to extract the relaxation time corresponding to τ = 1/ω. This gives τ = 0.2025 sec for the H-N case (Figure 5) and 0.2009 sec for the D-C case (Figure 6). The inset in Figures 5 and 6 shows enlarged intersecting region of the curve.

By setting various values of the product *ω*×τ in the relaxation function F * (ω) in Equation (11) it is possible to obtain the impedance function given in Equation (52). Either of these two functions can be used in plotting the product *ω*×τ on the abscissa versus imaginary part of the relaxation or impedance function on the ordinate for specific values of α and β. Thus, the imaginary part of Equation (52) can be plotted using Equation (11) for Debye, C-C, D-C, and H-N responses. These responses are shown in Figure 7 wherein the values of α and β prompted (ωτ)peak. It is clear that (ωτ)peak = 1 only for ideal Debye and ideal non-Debye (i.e., Cole-Cole) responses. Further (ωτ)peak ≠ 1 for non-ideal non-Debye D-C and H-N responses.

For the real world situation of Figure 4 represented in the F" format in Figure 7 that gives (ωτ)peak ≠ 1 (i.e., (ωτ)peak> 1) is shown in Figure 8. The imaginary part F" as a function of the product *ω*×τ is displayed in Figure 7 for Debye response with α = 1 and β = 1 yielding (ωτ)peak = 1; C-C response with α = 0.5 and β = 1 yielding (ωτ)peak = 1.0; D-C response with α = 1 and β = 0.6 yielding (ωτ)peak = 1.192; and H-N response with α = 0.5 and β = 0.6 yielding (ωτ)peak = 1.273. Figure 8 displays the product *ω*×τ versus Z (ω) for both D-C and H-N responses. For D-C response with α = 1 and β = 0.6 yields (ωτ)peak = 1.4970; for H-N response with α = 0.999 and β = 0.6 yields (ωτ)peak = 1.4980 and with α = 0.995 and β = 0.6 (ωτ)peak = 1.5010. The values of (ωτ)peak are within a close range presumably because of close proximity of α or β for D-C or H-N response. It is interesting to demonstrate the same Ni-Cd battery data for both D-C and H-N dielectric formalism.

To see the effect of change in the peak value of the product ω × τ on the parameter α and β, different values of α and β are plotted against (ωτ)peak. Figure 9 shows the variation of β with (ωτ)peak for various values of α indicating increase in β with decrease in α. Figure 10 shows the similar variation of α with (ωτ)peak for various values of β indicating increase in α with decrease in β. In general, both α and β decrease as (ωτ)peak increases. Also when the values of α and β approach zero (i.e., highly non-Debye situation), then (ωτ)peak becomes extremely large β. Thus, the extreme non-Debye situation gives (ωτ)peak>> 1. The values of (ωτ)peak for various values of α and β are shown in Tables 2 and 3.

4.5. Test of Depression Parameter β When α Is Real

In order to verify β being a real quantity, it is necessary to test its nature. Thus, considering α as a constant real parameter for β being complex quantity given by

$$\beta = \beta_1 + j\beta_2 \tag{53}$$

and using it in Equation (13):

	f	Measured Data		D-CData (× 10-4 Ω)				H-NData (× 10-4 Ω)
Observation		(× 10-4 Ω)		Calculated				Calculated
Number	(Hz)			2 R		Error		2 R		Error
				0.998545	0.994314			0.997304	0.976946
		Z′	Z″	Z′	Z″	Z′	Z″	Z′	Z″	Z′	Z″
1	50.110	14.320	0.240	14.38200	0.32400	0.062	0.084	14.40586	0.48162	0.086	0.242
2	39.800	14.420	0.380	14.44000	0.38580	0.020	0.006	14.46283	0.54686	0.043	0.167
3	31.610	14.500	0.490	14.50500	0.45830	0.005	0.032	14.52813	0.62064	0.028	0.131
4	25.110	14.590	0.610	14.57900	0.54290	0.011	0.067	14.60304	0.70386	0.013	0.094
5	19.950	14.710	0.720	14.66300	0.64130	0.047	0.079	14.68913	0.79750	0.021	0.078
6	15.850	14.820	0.840	14.76000	0.75530	0.060	0.085	14.78842	0.90265	0.032	0.063
7	12.590	14.940	0.950	14.87300	0.88670	0.067	0.063	14.90334	1.02037	0.037	0.070
8	10.000	15.070	1.070	15.00400	1.03700	0.066	0.033	15.03670	1.15135	0.033	0.081
9	7.9430	15.220	1.210	15.15800	1.20710	0.062	0.003	15.19195	1.29599	0.028	0.086
10	6.3090	15.390	1.370	15.34200	1.39680	0.048	0.027	15.37332	1.45410	0.017	0.084
11	5.0110	15.570	1.540	15.56100	1.60410	0.009	0.064	15.58587	1.62443	0.016	0.084
12	3.9800	15.790	1.740	15.82400	1.82350	0.034	0.083	15.83540	1.80402	0.045	0.064
13	3.1610	16.070	1.950	16.13900	2.04480	0.069	0.095	16.12824	1.98746	0.058	0.037
14	2.5110	16.390	2.160	16.51200	2.25180	0.122	0.092	16.47003	2.16557	0.080	0.006
15	1.9950	16.800	2.350	16.93900	2.42260	0.139	0.073	16.86462	2.32491	0.065	0.025
16	1.5850	17.270	2.500	17.40900	2.53390	0.139	0.034	17.31151	2.44755	0.042	0.052
17	1.2590	17.790	2.570	17.89700	2.56750	0.107	0.002	17.80198	2.51249	0.012	0.058
18	1.0000	18.350	2.530	18.37000	2.51860	0.020	0.011	18.31539	2.50046	0.035	0.030
19	0.7943	18.800	2.400	18.79600	2.39940	0.004	0.001	18.82081	2.40163	0.021	0.002
20	0.6309	19.220	2.200	19.15600	2.23310	0.064	0.033	19.28393	2.22184	0.064	0.022
21	0.5011	19.52	2.060	19.44400	2.04590	0.076	0.014	19.67722	1.98273	0.157	0.077
22	0.3980	19.730	1.890	19.66500	1.85860	0.065	0.031	19.98790	1.71435	0.258	0.176
23	0.3161	19.920	1.740	19.83200	1.68490	0.088	0.055	20.21853	1.44506	0.299	0.295

Table 1. The AC electrical data for the Ni-Cd battery

$$F = \frac{1}{\left(A + jB\right)^{(\beta_1 + j\beta_2)}}$$

$$F = X^{-\beta_1} \exp(\beta_2 \gamma) \left[ \left{ \cos(\beta_2 \ln X) \cos(\beta_1 \gamma) - \sin(\beta_2 \ln X) \sin(\beta_1 \gamma) \right} \right]$$ (54)

$$(A + jB)^{(\beta_1 + j\beta_2)}$$

$$F = X^{-\beta_1} \exp(\beta_2 \gamma) \begin{bmatrix} {\cos(\beta_2 \ln X)\cos(\beta_1 \gamma) - \sin(\beta_2 \ln X)\sin(\beta_1 \gamma)} \ -j{\sin(\beta_1 \gamma)\cos(\beta_2 \ln X) + \cos(\beta_1 \gamma)\sin(\beta_2 \ln X)} \end{bmatrix}$$ (55)

$$F = X^{-\beta_1} \exp(\beta_2 \gamma) \left[ \cos(\beta_1 \gamma + \beta_2 \ln X) - j \sin(\beta_1 \gamma + \beta_2 \ln X) \right]$$ (56)

Therefore, Equation (23) becomes:

$$\frac{F''}{F'} = \tan(\beta_1 \gamma + \beta_2 \ln X) \tag{57}$$

Figure 5. H-N response for tan (βγ) versus angular frequency ω of the Ni-Cd battery data with α = 0.995 and β = 0.6. Maximum theoretical value of tan (*βγ*max) obtained as 1.3628, while experimental value shows 1.2449. Inset shows ω = 4.9390 rad/sec corresponding to τ = 0.2025 sec.

Figure 6. D-C response for tan (βγ) versus angular frequency ω of the Ni-Cd battery data with α = 1.0 and β = 0.6. Maximum theoretical value of tan (*βγ*max) obtained as 1.3764, while experimental value shows 1.2449. Inset shows ω = 4.9766 rad/sec corresponding to τ = 0.2009 sec.

Figure 7. The product *ω*×τ is displayed as a function of the imaginary part of the relaxation function F * (ω) for (1) Debye response with α = 1 and β = 1 yielding (ωτ)peak = 1; (2) Non-Debye (i.e., Cole-Cole) response with α = 0.5 and β = 1 yielding (ωτ)peak = 1; (3) Davidson-Cole response with α = 1 and β = 0.6 yielding (ωτ)peak = 1.497; and (4) Havriliak-Negami response with α = 0.5 and β = 0.6 yielding (ωτ)peak = 2.637, Vertical dotted line indicates ωτ = 1.

Figure 8. The product *ω*×τ is displayed as a function of the imaginary part of Z * (ω) for curve 1: Davidson-Cole response with α = 1.0 and β = 0.6 yielding (ωτ)peak = 1.4970; curve 2: Havriliak-Negami response with α = 0.999 and β = 0.6 yielding (ωτ)peak = 1.4980; and curve 3: Havriliak-Negami response with α = 0.995 and β = 0.6 yielding (ωτ)peak = 1.5010.

From Equations (45) and (46), it is seen that *γ*max (when *ω*→ ∞) is a finite quantity. Similarly taking logarithm of Equation (19) and imposing the limiting case:

$$\lim{\omega \to \infty} \ln X = \frac{1}{2} \lim{\omega \to \infty} \ln \left[ 1 + (\omega \tau)^{2\alpha} + 2(\omega \tau)^{\alpha} \cos \left( \frac{\alpha \pi}{2} \right) \right] \to \infty$$ (58)

Figure 9. The parameter β versus (ωτ)peak for various values of α.

Figure 10. The parameter α versus (ωτ)peak for various values of β.

Taking limit on both sides of Equation (57) and using Equation (58), it is seen that no definite finite limit exists. However, in reality the observed data indicate finite ratio *F*˝/*F*΄ when *ω*→ ∞ as depicted in Figure 1 and Figure 3. This gives

$$\beta_2 = 0 \tag{59}$$

which in turn yields β as a real quantity. In the recent D-C analysis [5] it has been presented that the parameter β is the complex quantity for α = 1. However, β is a real quantity for the H-

N analysis. If the same analytical approach applied to the D-C analysis [5], β will turn to real quantity. The difference for the parameter β between this analysis and reference 4 is due to the fact that when the linear least square fitting equations are used, the independent and dependent variables are real against the complex quantities considered earlier.

Table 2. Values of peak ( ) for various values of for fixed

		( )  peak			( )  peak			( )  peak
0.1	0.1	9.660E+09	0.2	0.1	9.354E+04	0.3	0.1	1.98E+03
	0.2	9.509E+06		0.2	2.958E+03		0.2	196.783
	0.3	1.656E+05		0.3	393.550		0.3	51.754
	0.4	9.375E+03		0.4	94.267		0.4	20.110
	0.5	1009.5		0.5	31.133		0.5	9.672
	0.6	163.750		0.6	12.598		0.6	5.322
	0.7	35.145		0.7	5.864		0.7	3.213
	0.8	9.271		0.8	3.02		0.8	2.075
	0.9	2.862		0.9	1.686		0.9	1.412
0.4	0.1	275.991	0.5	0.1	84.345	0.6	0.1	37.733
	0.2	50.042		0.2	21.727		0.2	12.375
	0.3	18.552		0.3	9.932		0.3	6.496
	0.4	9.204		0.4	5.717		0.4	4.135
	0.5	5.353		0.5	3.732		0.5	2.92
	0.6	3.441		0.6	2.637		0.6	2.201
	0.7	2.369		0.7	1.967		0.7	1.734
	0.8	1.715		0.8	1.527		0.8	1.411
	0.9	1.290		0.9	1.221		0.9	1.176
0.7	0.1	21.019	0.8	0.1	13.412	0.9	0.1	9.363
	0.2	8.177		0.2	5.951		0.2	4.611
	0.3	4.762		0.3	3.749		0.3	3.093
	0.4	3.263		0.4	2.718		0.4	2.346
	0.5	2.440		0.5	2.124		0.5	1.900
	0.6	1.927		0.6	1.740		0.6	1.603
	0.7	1.580		0.7	1.471		0.7	1.389
	0.8	1.331		0.8	1.273		0.8	1.228
	0.9	1.145		0.9	1.121		0.9	1.102
							0.999	1.001

4.6. Test of Depression Parameter α When β Is Real

In order to verify α being a real quantity, it is necessary to test its nature. Thus, considering β as a constant real parameter for α being complex quantity given by

$$\alpha = \alpha_1 + j\alpha_2 \tag{60}$$

		( )  peak			( )  peak			( )  peak
0.1	0.1	9.66E+09	0.2	0.1	9.51E+06	0.3	0.1	1.66E+05
	0.2	9.35E+04		0.2	2.96E+03		0.2	393.55
	0.3	1.98E+03		0.3	196.783		0.3	51.754
	0.4	275.991		0.4	50.042		0.4	18.552
	0.5	84.264		0.5	21.752		0.5	9.932
	0.6	37.733		0.6	12.375		0.6	6.496
	0.7	21.019		0.7	8.177		0.7	4.762
	0.8	13.412		0.8	5.951		0.8	3.749
	0.9	9.363		0.9	4.611		0.9	3.093
0.4	0.1	9.38E+03	0.5	0.1	1009.5	0.6	0.1	163.75
	0.2	94.267		0.2	31.133		0.2	12.598
	0.3	20.11		0.3	9.672		0.3	5.322
	0.4	9.204		0.4	5.353		0.4	3.441
	0.5	5.717		0.5	3.732		0.5	2.637
	0.6	4.135		0.6	2.92		0.6	2.201
	0.7	3.263		0.7	2.44		0.7	1.927
	0.8	2.718		0.8	2.124		0.8	1.74
	0.9	2.346		0.9	1.9		0.9	1.603
0.7	0.1	35.145	0.8	0.1	9.271	0.9	0.1	2.862
	0.2	5.864		0.2	3.02		0.2	1.686
	0.3	3.213		0.3	2.075		0.3	1.412
	0.4	2.369		0.4	1.715		0.4	1.29
	0.5	1.967		0.5	1.527		0.5	1.221
	0.6	1.734		0.6	1.411		0.6	1.176
	0.7	1.58		0.7	1.331		0.7	1.145
	0.8	1.471		0.8	1.273		0.8	1.121

Table 3. Values of ( ) peak for various values of for fixed

and using it in Equation (13):

$$F = 1/\left[1 + \left(j\omega\tau\right)^{\alpha_1 + j\alpha_2}\right]^{\beta} \tag{61}$$

0.9 1.389 0.9 1.228 0.9 1.102

0.999 1.086

$$F = 1/\left[1 + (\omega\tau)^{\alpha_1} \exp\left(-\frac{\alpha_2\pi}{2}\right) \cos\phi + j(\omega\tau)^{\alpha_1} \exp\left(-\frac{\alpha_2\pi}{2}\right) \sin\phi\right]^{\beta}$$ (62)

$$F = 1/\left(C + jD\right)^{\beta} \tag{63}$$

where,

$$\varphi = \frac{\alpha_1 \pi}{2} + \alpha_2 \ln \omega \tau \tag{64}$$

$$C + jD = X \exp(j\psi) \tag{65}$$

$$\tan \psi = \frac{D}{C} = \frac{\left(\omega \tau\right)^{\alpha_1} \exp\left(-\frac{\alpha_2 \pi}{2}\right) \sin \varphi}{1 + \left(\omega \tau\right)^{\alpha_1} \exp\left(-\frac{\alpha_2 \pi}{2}\right) \cos \varphi}$$ (66)

$$\tan \psi = \frac{\sin \varphi}{\left(\omega \tau\right)^{-\alpha_1} \exp\left(\frac{\alpha_2 \pi}{2}\right) + \cos \varphi} \tag{67}$$

$$\tan \psi = \frac{\sin\left(\frac{\alpha_1 \pi}{2} + \alpha_2 \ln \omega \tau\right)}{\left(\omega \tau\right)^{-\alpha_1} \exp\left(\frac{\alpha_2 \pi}{2}\right) + \cos\left(\frac{\alpha_1 \pi}{2} + \alpha_2 \ln \omega \tau\right)}$$ (68)

and

$$X^2 = C^2 + D^2 (69)$$

Therefore,

$$F = \frac{1}{\left(Xe^{j\psi}\right)^{\beta}} = X^{-\beta}e^{-j\beta\psi} = X^{-\beta}\left(\cos\beta\psi - j\sin\beta\psi\right)$$ (70)

Now Equation (23) becomes:

$$\frac{F''}{F'} = \tan \beta \psi \tag{71}$$

Taking the limiting case for Equation (67) as well as using Equations(64) and (68):

$$\lim{\omega \to \infty} \tan \psi = \lim{\phi \to \infty} \tan \varphi \tag{72}$$

which does not exist, however, from the observed data and Equation (46):

$$\lim_{\omega \to \infty} \frac{F''}{F'} = \tan\left(\frac{\alpha\beta\pi}{2}\right) = \text{finite value}$$ (73)

Therefore, using Equations (68), (71), and (73) it is seen as:

$$\alpha_2 = 0 \tag{74}$$

in order to have a finite limit as ω → ∞.

4.7. Effect of α and β on the H-N Distribution Function

The H-N distribution function is given by [19, 20, 23, 24]:

$$g_{HN}(x) = \frac{1}{\pi} \frac{x^{\beta} \sin(\beta \theta)}{\left[x^{2\alpha} + 2x^{\alpha} \cos(\alpha \pi) + 1\right]^{\beta/2}}$$ (75)

where

$$\theta = \tan^{-1} \left[ \frac{\sin(\alpha \pi)}{x^{\alpha} + \cos(\alpha \pi)} \right]$$ (76)

and

$$x = \tau / \tau_{HN}$$ for $0 < \tau < \infty$ , $0 < x < \infty$ (77)

τ is the continuous relaxation time within a limiting range and τHN is the fitted relaxation time obtained from the same data. So far the relaxation time τHN has been referred to as τ. When *α*→1then *gHN*→gDC, where gDC is the D-C distribution function. When gHN is written as a function of ln(τ), following properties is identified:

$$g{HN}(\ln \tau) > 0$$ , for $-\infty < \ln \tau < \infty$ , and $\int{-\infty}^{\infty} g(\ln \tau) d\ln \tau = 1$ (78)

Figure 11 displays the effect of β on the H-N distribution function for a fixed value of α = 0.5. The inset shows the same response for the fixed value of α = 0.995 obtained from the H-N fitting of the NiCd battery data (Table 1). The curves indicate shift of the peak of the H-N distribution function as β increases with respect to x = ln(τ/τHN) = 0 line. As β increases, curves become symmetrical about the x = 0 line for the H-N distribution function and gradually tend to the C-C distribution function.

Figure 11. Effect of β on the H-N distribution function for α = 0.5, and τ = *τ*HN.

The spread of the H-N distribution function (standard deviation) also reduces as it approaches the C-C distribution function. When β = 1, the H-N distribution function coincides exactly with the C-C distribution function with the peak at τ = τCC. Similar behavior is observed at any other value of α.

Figure 12 displays the effect of the change in α on the H-N distribution function for a fixed value of β = 0.6 which was obtained from the H-N fitting of the Ni-Cd battery data. The figure indicates that the increase in α shifts the peak of the H-N distribution function toward the x = ln(τ/τHN) = 0 line. Each curve becomes more skewed in the negative domain, i.e. larger portion of the curve lies left to the x = ln(τ/τHN) = 0 line. This indicates the H-N distribution function tends to the D-C distribution function. The spread of the H-N distribution function (standard deviation) also reduces as it approaches the D-C distribution function. When α = 1, the H-N distribution function coincides exactly with the D-C distribution function which lies to the left of the x = ln(τ/τHN) = 0 line. Similar behavior is also observed at any other value of β.

In Figure 11 the breadth of the distribution function becomes narrower as *β*→1 (i.e., C-C) which indicates decrease in skewness reflecting near homogeneous distribution of the R-C equivalent circuit values. In Figure 12 the breadth of the distribution function becomes narrower as *α*→1 (i.e., D-C) indicating asymmetrical distribution of the R-C equivalent circuit values. The value of the resistance is very much limited by the chord of the relaxation loci indicating singular contribution to the relaxation time whereas capacitance is extracted from the y-axis attributing to the skewness. The capacitance is presumed to be nonlinear causing drift in the *ω*×τ = 1 from the peak of the relaxation arc.

Figure 13 shows the nature of the skewed behavior of the H-N distribution function with limiting case of C-C for β = 1. Here the skewness is quantified as the difference between the expected (mean) value of τ and the parameter τHN. The expected value of τ from the distribution differs significantly from the fitted parameter τHN and the difference is dependent on the parameters α and β. The expected value is always less than the fitted value, which indicates the skewed nature of the HN relaxation function. With increase in β (i.e., H-N→C-C) for fixed α = 0.5, the difference reduces and the expected value of ln(τ) coincides with the fitted relaxation time constant at the C-C limit, i.e., E[ln(τ)]→ln(τCC). Similar behavior is also observed in the case of H-N→D-C situation via the effect of α (for fixed β = 0.6). For the limiting C-C case the difference between the expected value of ln(τ) and ln(*τ*HN) is zero. Increase in α reduces the difference between the expected value and the fitted parameter value of τHN. However, unlike the C-C limit, the difference does not vanish at the D-C limit condition since the distribution is highly unsymmetrical at the D-C condition.

Figure 12. Effect of α on HN distribution function for β = 0.6.

Another way to determine the skewness of the H-N distribution function and the effect of α and β on it is to calculate the skewness statistic of the distribution function. The skewness of the distribution function is defined as [25]:

$$\gamma_1 = \mu_3 / \mu_2^{3/2} \tag{79}$$

whereμ² and μ³ are the second and third moment about the mean of the distribution, respectively. The skewness of *g*HN given by γ¹ is plotted in Figure 14 as a function of α and β. In reality α is not the skewness but amalgamated with β = 0.6 in Figure 14. The horizontal line shows β = 1. This figure indicates that the skewness of the H-N distribution function reduces when β increases for fixed α, and increases when α increases for fixed β. This is expected since the increase in β indicates H-N→C-C condition, and the distribution becomes more symmetrical. Thus, making it more gradual skew positively, i.e., the skewness vanishes when the H-N distribution becomes C-C distribution. On the other hand, the increase in α indicates H-N→D-C condition and the distribution becomes more non-symmetrical, thus making it increasingly negatively skewed distribution.

Figure 13. Effect of α and β on difference between expected value of ln(τ), ln(τHN) for α = 0.5 and β = 0.6.

4.8. Meaning of the Depression Parameters α and β

The straightforward meaning of the depression parameter α implies the presence of the depression angle of the relaxation arc in the absence of β as found in the C-C behavior. This is not the case for the ac electrical data of the Ni-Cd battery displayed in the D-C configuration [4]. The presence of the finite depression angle represents non-ideal conduction processes wherein a considerable degree of non-uniformity is associated with the capacitance term than the resistance term. It also indicates the breadth of the distribution of the relaxation times. The depression parameter α indicates broadness of the frequency dependence of the imaginary part of the impedance (Z" ≡ F"). This is related to the average relaxation time attributing to the reactance. Often the average relaxation time referring to the singular lumped R-C circuit (showing one relaxation time) is termed as the distribution of a set of relaxation times [26, 27]. For all purposes α lies between 0 and 1 in order to satisfy Debye and C-C responses. The straightforward meaning of the depression parameter β in the D-C behavior implies the breadth and skewness [28] of the relaxation arc for the ac electrical data of the Ni-Cd battery. The breadth and skewness of the relaxation arc is not necessarily overlapping with the behavior of the depression angle for the relaxation process resulting skewness in the relaxation arc. For the D-C behavior using the Ni-Cd data the relaxation arc is found to be skewed when α = 1 and β = 0.6. The equivalent circuit model of the D–C behavior may be viewed considering both *ω*×τ = 1 corresponding to the maximum of the y-axis parameter and ωτ ≠ 1 corresponding to the maximum of the relaxation function F * (ω). Another way to view equivalent circuit is to refer to Figure 3 where *ω*×τ = 1 does not correspond to the maximum of the y-axis parameter but takes the value on the low-frequency side.

Considering the role of both α and β causing the H-N behavior it is understood that these two depression parameters are not isolated from each other. The frequency domain product α × β for the H-N parameters is related to the time domain KWW (Kohlrausch-Williams-Watts) parameter [21, 24]. Further interpretation of combined α and β may be extended to both diffusion and activation controlled dielectric behavior of the device under test.

The parameters α and β are not optimized as far as H-N representation is concerned. The analytical approaches presented herein reflect that the Ni-Cd battery system is surely too complex to comprehend the underlying conduction processes. Based on the present analysis it is understood that the internal configuration of the Ni-Cd battery is responsible for the evolution of the parameters α and β. The solution of the problems associated with the internal construction and overall content of the Ni-Cd batteries may not further simplify the parameters α and β. The previous analyses [5] indicate that there is a chance to view the present H-N representation in the D-C format. This dual complexity requires clarity from the internal construction viewpoint of the Ni-Cd battery. Another study [27] also indicates similar situation for the Ni-Cd batteries though curve fitting procedure for the identical complex plane analysis is not conducted. Nevertheless, it is clear that the Ni-Cd battery is too complex for the impedance spectroscopy to simplify the conduction processes. It is possible to use the information of the parameters α and β to achieve better performance characteristics. Perhaps in this way impedance spectroscopic approaches can provide improved functions of the Ni-Cd batteries.

Figure 14. Effect of α and β on the skewness of the H-N distribution function for α = 0.5 and β = 0.6.

CONCLUSION

The analysis of the ac small-signal electrical data in the H-N dielectric representation in the impedance plane allowed understanding of the depression parameters α and β for the Ni-Cd batteries. These two parameters are real quantities and cannot be separated to display independent role for each of them for the H-N formalism. The breadth and skewness are associated with these two parameters primarily at high frequencies when the data are displayed in the impedance plane. The product (*ω*×τ)peak corresponding to the maximum value of the imaginary part of the complex function F * (ω) is not equal to unity for the H-N representation. This product depends on the parameters α and β and the extreme non-Debye situation which is characterized by (ωτ)peak> 1. The geometrical relationships among various parameters of the H-N relaxation function can be used to determine the relaxation parameter *τ*HN. Based on the behavior of the two depression parameters, it can be concluded that the optimized values to be determined in future may yield useful battery performance concerning the reduction of the memory effect.

ACKNOWLEDGMENT

The authors gratefully acknowledge Mr. Stephen R. Bissell for his initial contribution in the D-C analysis using AC electrical data of the Ni-Cd battery presented in Reference 5. The authors appreciate interest and intellectual support of Mrs. Jayati Bhattacharjee.

REFERENCES

• [1] Greiner, J.; Microfabricated Thin-Film Batteries: Technology and Potential Applications. Thesis, Massachusetts Institute of Technology, 2006.
• [2] Sequeira, C.A.C.; Hooper, A; Solid State Batteries. NATO Science, Series E, Applied Sciences, No. 101, Springer, 1985.
• [3] Rydh, C.J.; Karlstrom M.; Life Cycle Inventory of Recycling Portable Nickel-Cadmium Batteries. Resources, Conservation and Recycling, vol. 34, pp. 289-309, 2002.
• [4] Morioka, Y.; Narukawa, S.; Itoh, T.; State-of-the-Art of Alkaline Rechargeable Batteries. Journal of Power Sources, vol. 100, pp. 107-116, 2001.
• [5] Alim, M.A.; Bissell, S.R.; Mobasher, A.A.; Analysis of the AC Electrical Data in the Davidson-Cole Dielectric Representation. Physica B, vol. 403, pp. 3040-3053, 2008.
• [6] NASA Practice No. PD-EC-1103, Ni-Cd Conventional Spacecraft Battery Handling and Storage. October 1995.
• [7] Dhar, S.K.; Ovshinsky, S.R.; Gifford, P.R.; Corrigan, D.A.; Fetcenko, M.A.; Venkatesan, S; Nickel/Metal Hydride Technology for Consumer and Electric Vehicle Batteries - A Review and Up-Date. Journal of Power Sources, vol. 65, pp. l-7, 1997.
• [8] Kiehne, H.A.; Battery Technology Handbook. 2ndEdition, CRC Press, 2003.
• [9] Dell, R.M.; Batteries: Fifty Years of Materials Development. Solid State Ionics, vol. 134, pp. 139-158, 2000.
• [10] N.J. Dudney, N.J.; Solid-state Thin-Film Rechargeable Batteries. Materials Science and Engineering B, vol. 116, pp. 245-249, 2005.
• [11] Bates, J.B.; Dudney, N.J.; Neudecker, B.; Ueda, A; Evans, C.D.; Thin-Film Lithium and Lithium-Ion Batteries. Solid State Ionics, vol. 135, pp. 33-45, 2000.
• [12] Yang, C.C.; Polymer Ni-MH Battery Based on PEO-PVA-KOH Polymer Electrolyte. Journal of Power Sources, vol. 109, pp. 22-31, 2002.
• [13] Alim, M.A.; Electrical Characterization of Engineering Materials. Active and Passive Electronic Components, vol. 19, no. 3, pp. 139-169, 1996.

• [14] Von Hippel, A.R.; Dielectrics and Waves. John Wiley & Sons, 1954.

• [15] Smyth, C.P.; Dielectric Behavior and Structure. McGraw-Hill Book Company, 1955.

• [16] Cole, K.S.; Cole, R.H.; Dispersion and Absorption in Dielectrics: Alternating Current. Journal of Chemical Physics, vol. 9, pp. 341-351, 1941.

• [17] Davidson, D.W.; Cole, R.H.; Dielectric Relaxation in Glycerine. Journal of Chemical Physics, vol. 18, no. 10, pp.1417-1417, 1950.

• [18] Davidson, D.W.; Cole R.H.; Dielectric Relaxation in Glycerol, Propylene Glycol, and n-Propanol. Journal of Chemical Physics, vol. 19, no. 12, pp.1484-1490, 1951.

• [19] Havriliak, S.; Negami, S.; A Complex Plane Analysis of Dispersions in some Polymer Systems. Journal of Polymer Science, Part C, vol. 14, pp.99-117, 1966.

• [20] Havriliak, S.; Negami, S.; A Complex Plane Representation of Dielectric and Mechanical Relaxation Processes in some Polymers. Polymer, vol. 8, pp.161-210, 1967.

• [21] Alvarez, F.; Algeria, A.; Colmenero, J.; Relationship between the Time-Domain Kohlrausch-Williams-Watts and Frequency-Domain Havriliak-Negami Relaxation Functions. Physical Review B, vol. 44, no. 14, pp. 7306-7312, 1991.

• [22] Sluyters Rehbach, M.; Impedances of Electrochemical Systems: Terminology, Nomenclature and Representation - Part I: Cells with Metal Electrodes and Liquid Solutions. Pure and Applied Chemistry, vol. 66, no. 9, pp.1831-1891, 1994.

• [23] Zorn, R.; Applicability of Distribution Function for the Havriliak-Negami Spectral Function. Journal of Polymer Science, Part B: Polymer Physics, vol. 27, pp.1043- 1044, 1999.

• [24] Alvarez, F.; Alegria, A.;Colmenero, J.; A New Method for Obtaining Distributions of Relaxation Times from Frequency Relaxation Spectra. Journal of Chemical Physics, vol. 103. No. 2, pp.798-806, 1995.

• [25] Zorn, R.; Logarithmic Moments of Relaxation Time Distributions. Journal of Chemical Physics, vol. 116, no. 8, pp. 3204-3209, 2002.

• [26] Alim, M.A.; Admittance-Frequency Response in Zinc Oxide Varistor Ceramics. Journal of the American Ceramic Society, vol. 72, no. 1, pp.28-32, 1989.

• [27] Alim, M.A.; Seitz, M.A.; Singular Nature of Preferential Conducting Paths at High Electric Fields in ZnO Based Varistors. Journal of the American Ceramic Society, vol. 71, pp.C246-C249, 1988.

• [28] Havriliak, S.; Havriliak, S.J.; A Rigorous Comparison of Theoretical Autocorrelation Functions with Dielectric Relaxation Parameters for the Alpha Process of Polymers. Journal of Polymer Science, Part B: Polymer Physics, vol. 35, pp.1887-1897, 1997.

• [29] Hammouche,A.; Karden, E.; De Doncker, R.W.; Monitoring State-of-Charge of Ni-MH and Ni-Cd Batteries Using Impedance Spectroscopy, Journal of Power Sources, vol. 127, pp. 105-111, 2004.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 18

ADVANCED DRIVE ARCHITECTURES AND STATE OF THE ART IN-WHEEL MOTORS

Blaz Stefe and Gorazd Lampic

Elaphe Propulsion Technologies Ltd., Slovenia

ABSTRACT

The following chapter describes different concepts of drive architectures for conventional 4-wheel vehicles, most typical for passenger cars and delivery vans. Besides the conventional drive architectures, a state of the art direct drive in-wheel propulsion system and its key advantages are presented. Conventional drive architectures include an internal combustion engine and a central motor electric drive, while the concept of the direct drive in-wheel motor, boasting a high torque to mass ratio, zero maintenance requirements, compactness and efficiency, is researched in more detail. This chapter reviews the major mechanical design possibilities that are currently known in the field of in-wheel electric motors, including direct drive motors and the concepts with gears or a reduction gear. The drive cycle and extreme drive conditions, such as driving uphill and driving at top speed, are simulated for several vehicle segments from B to E. In-wheel motor requirements, such as peak torque and RPMs, are presented. As it is known, inwheel motors increase the total unsprung mass of the vehicle. For its consideration, a mathematical model is described and the effect on the drivability is both tested and theoretically analyzed and presented by charts. Being one of the crucial topics when designing in-wheel motors for different kinds of vehicles, a table reviewing a notable amount of different mechanical constructions is presented in this chapter.

Keywords: Direct Drive, In-Wheel Motor Propulsion System, Electric Vehicle, Central Motor, Mechanical Design, Drive Architecture, Unsprung Mass

Email: blaz@elaphe-ev.com.

1.INTRODUCTION

Energy supply for the internal combustion engines, low efficiency, high maintenance costs, a requirement for its own compartment (usually under the hood of the car), and environmental concerns all call for a radical change of the propulsion concept of the vehicle. Centrally mounted electric motors have especially turned out to solve environmental issues and to reduce maintenance costs. However, according to the research and market demand, the most promising propulsion concept, providing ultimate solutions for the customer and urban society, seems to be a highly adjustable direct drive in-wheel motor propulsion system. Numerous electro-magnetic and mechanical designs and concepts of in-wheel motors have already been produced by several companies and research institutions. Here we present the findings and concrete results.

The drive architecture of an internal combustion engine (ICE) powered vehicle is limited by the constraints related to the design and performance of their engines. Gasoline, diesel, LPG and other internal combustion engines provide the required performance only at high RPMs, while their performance is poor at low power and their mechanical design fit only as a central motor. There, it takes a notable percentage of the vehicle"s total space and keeps the center of gravity high. This kind of drive architecture requires mechanical transmissions and gears (gear-box, reduction gear, homokinetic joints, differentials, axles and kardan joint couplings), both contributing to energy loss, weight and complexity 1. Despite the undisputable reliability and also respectful durability, there are needs and possibilities for new solutions.

Figure 1. The drive architecture of an internal combustion engine (ICE) powered vehicle [2].

As a natural evolution, the central internal combustion engines are being gradually replaced by electric motors. The improvements of the first generation of vehicles are done by even more complex hybrid solutions, gaining benefits from both technologies. The drawbacks are a complex production, increased total weight of the vehicle, required maintenance of both propulsion systems and of course a higher price of the vehicle (which is in some jurisdictions kept competitive by the environmental law policy). The second generation of electric vehicles runs purely on batteries and central electric motors, as shown in Figure 2. However, the advantages of this system are still related only to the advantages of the electric motor technology in relation to the internal combustion engine. No other benefits of this transition are exploited as all the beforehand mentioned mechanical parts remain in place and contribute to the higher weight of the car, energy loss and in general decrease the vehicle"s efficiency. When taking advantage of the electric motor, only the gearbox is optional and its necessity depends on the type and characteristics of the electric motor. However, mechanical parts that waste energy are not omitted. Therefore, the efficiency of this propulsion system (providing around 70 kW of power) with a central motor can range between 60 to 65%. The mass of the motor with an inverter normally turns out to exceed 100 kg and, including all the shafts, gearing and the differential, exceeds the value of 150 kg. On the other hand, a 2WD propulsion system with a low torque, high speed in-wheel electric motors with a reduction gear (or planetary gear-box) and inverters weighs between 90 kg and 120 kg (saving on the differential and axles) and can reach efficiencies in the range of 65 to 70%. But the state of the art direct drive in-wheel in a permanent magnet synchronous motor (PMSM) [8], providing the power around 70 kW for a 2WD propulsion system, can reach efficiencies up to 88% (including the loss of the inverter) and can, in the case of Elaphe construction, weigh below 60 kg.

Figure 2. Vehicle propulsion system powered by a central electric motor [2].

2. HISTORY

Currently the most advanced drive architecture, using the real benefits of electric motors, is the so-called direct drive propulsion architecture with implemented in-wheel motors. Motors deliver the torque exactly where it is needed; at the wheel. This solution has already been proposed and implemented for the testing purposes by innovators in the late 19th century (Figure 3), but numerous technological, mechanical, economic and political constraints have, for more than 100 years, kept such a motor from the market and every-day use. In the last decades, technology has made a giant step forward in the fields of production, electromagnetic design and energy storage. The new era of mobility seems to be closer than ever before. Using the technology of chemical energy storage on the one hand and a development of fuel cells on the other, the development of advanced rechargeable batteries with high energy/mass ratio has also pushed forward the development of in-wheel motors as a fusion of advanced electro-magnetic topology and mechanical and production engineering, as shown in Figure 3.

Figure 3. A car powered by in-wheel motors built by Dr. Ferdinand Porsche in 1900-1901.

3. DIRECT DRIVE IN-WHEEL PROPULSION

A direct drive in-wheel propulsion system brings several advantages in the following fields:

• uncompromising performance, satisfying and exceeding a modern driver"s demands;
• clean propulsion with zero on spot emissions (in most configurations, except the inseries or parallel hybrid with the ICE), or total zero emissions when vehicles run on electricity from renewable resources;
• unbeatable efficiency;
• zero maintenance;
• space saving technology.

3.1. Uncompromising Performance

Uncompromising performance comes from the very high specific torque state of the art in-wheel electric motors and is usually provided from a standstill. In most configurations with 2 in-wheel motors, a torque applied on the road already exceeds the maximum torque applied on the road by a standard ICE engine in the 1st gear. For the vast majority of vehicles, the torque, applied on the road at the speeds where 2nd gear was engaged with a normal ICE engine, is significantly in favor of the in-wheel motors. The in-wheel motor can supply very high torque to the road at low RPM and full torque at a very broad RPM range, while the ICE engine can only do that at higher RPM. The total torque reserve, which can be applied to the road at any speed, is highly in favor of the in-wheel electric motor, as is evident from the comparison of the charts in Figure 4.

As technology and economies are moving towards the oil-independent transport, it can be expected that in-wheel motors will soon be implemented particularly into the vehicles with rechargeable batteries or fuel cells instead of in-series hybrids with the ICE [7]. Clean energy can in many cases be gained entirely from renewable resources, such as the wind, water or solar energy. Even in cases where the infrastructure for the CO² neutral energy resources does not exist, emissions from the power plants (like gasses and particles) can be much better controlled and cleaned than those exhausted from many independent vehicles. Moreover, they can be emitted into the environment in uninhabited areas instead of in the city centers.

(a) Torque and power characteristics of the ICE

(b) Torque characteristics of the in-wheel direct drive electric motor

Figure 4.

The comparison of the torque and power characteristics of the ICE and in-wheel direct drive electric motor: an ICE motor (Figure 4(a)) provides a very low torque at low RPM and requires a transmission, while an in-wheel motor (Figure 4(b)) provides a full torque already at the lowest RPM and does not require a transmission due to the high torque and low specific speeds (the diagram is for an Elaphe in-wheel motor).

3.2. Unbeatable Efficiency

Unbeatable efficiency of the direct drive in-wheel motors comes from the already mentioned absence of numerous mechanical parts, transmissions and gears, a low mass of the active part of the motors and a state of the art electromagnetic topology of the motors.

3.3. Zero Maintenance

It is well known that a conventional ICE, besides the fuel, also requires regular maintenance for a reliable operation during the years. Regular maintenance includes regular checks and changing the motor oil, air, oil and fuel filters, timing belt, seals, and so on). Apart from the bearing, there is no friction between the moving parts as the drag force of the electric motor is literally generated in the air gap. No lubricating liquids and no filters are needed, therefore such an electric motor is a maintenance free machine. In certain configurations, seals and bearings might be a subject to change after a certain amount of miles are driven. But even in conventional cars, the bearing must be changed after a longer period of continuous use. What is also believed to become a revolutionary factor in the car industry in the next 5 to 10 years is a completely new concept of a car"s chassis, body and passenger compartment. That is particularly due to the absence of the ICE under the hood of the car and because of the absence of the kardan joint couplings in case of the RWD vehicles, differentials, exhaust collector and exhaust pipes with catalytic converter and other emission gases treatment devices, such as diesel particle filter as shown in Figure 5. Moreover, the gearbox that usually takes a large space in the engine compartment or is placed in the passenger compartment in case of sporty or top-end vehicles is absent and the space can be used for the passenger"s comfort.

Figure 5. Major components of a conventional ICE vehicle that are not required in the in-wheel motor vehicle. This way, saved space can be used for passengers, luggage, or simply saved, which can result in more compact urban vehicles.

A new dimension of using a so far empty space inside the wheel of the vehicle, allows constructors to design a completely flat floor of the car and use it for the passengers comfort (Figure 6). A lot of space, previously occupied by the motor parts, can now be used as a luggage compartment and the designers of the car are much less limited and can bring their creativity to an entirely new level. City cars can therefore be smaller in all dimensions and can with their agility and compactness contribute to a better traffic flow, less traffic jams and can significantly help the utilization of the parking areas in the city.

Figure 6. Electric vehicle propulsion system powered by an in-wheel motor: the propulsion elements do not take up a lot of space, which can therefore be used for the passenger"s comfort [2, 3].

4. HIGH TORQUE AND LOW RPMS REQUIREMENTS FOR THE DIRECT DRIVE IN-WHEEL MOTORS

The simplest way to design the drive system is without any gears. Such motors need to produce up to ten times the torque required by geared motors, but the high speed can be reached by relatively low RPMs. Beside the peak torque and RPM shift, the approach of analyzing the efficiency of the motor is also different in geared and gearless propulsions. Let us examine this on the example of the urban car [6].

The first step of the analysis includes the vehicle data, driving demands and any dynamic driving cycle representing the expected typical use of the vehicle.

| Table 1. Technical specifications
of the compact city vehicle representing the | | | | | |-----------------------------------------------------------------------------------|--|--|--|--| | L7equadricycle/microcar segment | | | | | | | | | | |

Parameter	Value
Frontal area	2.2 m²
Total mass (with 2 people)	650 kg
Air drag coefficient	0.35
Friction coefficient	0.015
Number of in-wheel electric motors	4
Outer radius of a wheel	0.25 m

Figure 7. The selected complex urban driving cycle lasts for 1369 seconds and covers 12 kilometers driven with an average speed of 31.5 km/h.

Figure 8. Probability density function of working conditions for an in-wheel direct drive motor: the dark color represents the area where the electric motor runs most of the time. White areas are less occupied. The working point of the stand still time is not shown, since it does not contribute to power consumption in no hotel load case.

The example is made for the super compact city vehicle, also called the heavy quadricycle, representing the L7e market segment with the following parameters (Table 1). The main calculation has been performed using the Urban Dynamometer Driving Schedule (Figure 7).

Using the vehicle and the driving cycle data, we can calculate a diagram showing torque and wheel rotation speed demand (Figure 7). The torque and frequency demand diagram shows what part of the driving time an electric motor spends at different torques T and revolutions per second ν (Figure 8). By calculating such a diagram, we can determine the torque and the frequency range that have to be achievable with an electric motor [6].

The Equation (1) shows Ptime, which is the probability density function showing the probability to find the motor during the driving cycle (Figure 7) at the working point of given rotations ν and torque T. In the equation, t is the duration of the driving cycle.

$$t = \int_{v0}^{v{\text{max}}} \int_{T0}^{v{\text{max}}} P_{time}(v, T) dv dT$$ (1)

In Figure 8, we can see for how long an electric motor works at different conditions (torque and rev/s). From these data we can deduce that, for most of the time, the electric motor is in regime at 15 Nm of torque and at mechanical frequency of 7 rev/s.

Figure 9. The function of energy distribution showing the relative parts of released energy at all torque and mechanical frequency related working points during driving cycle.

The peak torque and the peak mechanical frequency needed can also be calculated. The peak mechanical frequency depends only on the top speed in the driving cycle and on the wheel diameter. It is found to be 16.3 rev/s or 978 RPM, which is far less than in electric motors used in combinations with gears.

The peak torque has been determined from three different conditions:

• Firstly, the torque and frequency demand diagram in a standard driving cycle shows a peak torque of 69.7 Nm. However, this torque may not be sufficient for some special driving conditions where higher force or acceleration are required.
• Secondly, the peak torque needed can be calculated from acceleration requirements (0 to 60 km/h in 6 seconds). It is found to be 121 Nm.
• The third way of the peak torque determination is the hill climbing ability. In order to climb a 35% slope, the peak torque needed is 134 Nm, which is the highest number and is therefore the required value of peak torque.

In order to develop an appropriate electric motor for a direct drive application, its efficiency has to be optimized according to the working conditions in the predicted driving cycle. By multiplying the probability of density function with torque and angular speed at each working point of the diagram (2), a new diagram is obtained (Figure 9). It shows the amount of energy released at every working point (torque and mechanical frequency) during driving cycle.

The best overall efficiency is obtained when the electric motor and power electronic efficiencies are optimized according to the working points where most of the energy is released. Such diagrams are useful for the optimization of direct drive electric motors and power electronics for all efficiency critical applications (Figure 9).

Total energy required in the driving cycle E is defined in Equation (2) and is only an extension of the Equation (1).

$$t = \int_{v0}^{v{\text{max}}} P_{time}(v, T) dv dT$$ (2)

Absolute values of energy are used, so in case of energy recuperation at regenerative breaking, the energy shown is positive.

Approximately 40% of released energy is in the middle area of the diagram, 21% at high speeds, 9% at low speeds and 16% at regenerative braking area. This calculation shows that the efficiency of the propulsion system has to be high in a wide area of torques and RPMs, which differs from the systems with gears where narrower high energy efficiency area is acceptable.

5.IN-WHEEL MOTOR CHARACTERISTICS FOR DIFFERENT SEGMENTS OF PASSENGER VEHICLES

The characteristics of the motor, calculated in section 4, apply to a propulsion system of a compact electric city car - A segment according to the EU classification. This kind of passenger vehicles started to gain popularity around 2010 but the conventional cars are still well demanded by the customers and a long transition period is expected before the market will show a notable demand for the super compact electric vehicles. Until then, hybrids, converted ICE and also some electric cars in the B, C, D and E market segments are expected to appear on the market. We had to make a research of the characteristics of cars in the mentioned segments in order to be able to develop appropriate propulsion systems for these vehicles. In every car segment we have chosen one of the most common representatives and calculated the required performance of the propulsion system for such a vehicle.

The representative vehicles are: Renault Clio – B segment, Kia Ceed – C segment, Volkswagen Golf – C segment, Ford Mondeo – D segment and Audi A6 – E segment.

The performance (torque and power) of a direct drive propulsion system that meets the hill climbing, top speed and acceleration requirements is presented in Tables 2-5.

Figure 10. Complex extra urban driving cycle on which the calculations for the segments B, C and D have been performed.

Table 2. Torque, RPS and power requirements for different segments of the vehicles to drive up a 30% slope at 30 km/h

Vehicle	Torque driving up a 30% slope at 30km/h [Nm]	RPS at 30km/h [rev/s]	Power driving up a 30% slope at 30km/h [kW]
B segment*	898	4.40	24.8
C segment*	1266	4.18	33.3
C segment*	1063	4.18	28.0
D segment*	1441	4.20	38.0
E segment*	1709	3.92	42.1

Table 3. RPS, power and torque requirements for different segments of the vehicles to reach top speed

Vehicle	Top speed [km/h]	RPS at top speed [rev/s]	Torque at top speed [Nm]	Power at top speed [kW]
B segment*	140	20.53	269	34.8
C segment*	150	20.92	408	53.6
C segment*	150	23.7	364	54.2
D segment*	160	22.38	350	49.3
E segment*	160	20.92	355	46.8

The results obtained from these calculations and presented in Tables 2-5 can help developers of in-wheel or any other type of direct drive motor with defining the motor requirements as far as peak torques and RPS are concerned. As the data are relevant for the vehicle, the total torque values have to be divided by 2 or 4 to obtain required "per motor" specifications for 2- or 4-wheel drive systems. From the data above, one can see that a direct drive motor in class B segment car with a 4-wheel drive needs approximately 300 Nm of peak torque and has to reach up to 1200 RPM. On the other hand, the direct drive motor for a class E segment car with a 2-wheel drive needs 1200 Nm and also has to work at approximately 1200 RPM. In addition, they show the differences between car segments and the most frequent working conditions in the complex urban and complex extra-urban driving cycle.

Table 4. Results of the calculations of the most frequent working point and peak torque in the complex urban driving cycle (Figure 7) for the direct drive in-wheel motor in different car segments

| Vehicle | Complex urban
driving | Complex urban
driving cycle most frequent
working point | | | |------------|--------------------------|---------------------------------------------------------------|-------------|--| | | cycle peak torque [Nm] | Torque [Nm] | RPS [rev/s] | | | B segment* | 533.3 | 91.4 | 13.2 | | | C segment* | 627.3 | 99.4 | 15.9 | | | C segment* | 729.6 | 106 | 12.4 | | | D segment* | 848.4 | 120 | 12.5 | | | E segment* | 1003 | 142 | 11.7 | |

Table 5. Results of the calculations of the most frequent working point and peak torque in the complex extra-urban driving cycle (Table 4) for the direct drive in-wheel motor in different car segments

	Complex extra-urban	Complex extra-urban	driving cycle most frequent
Vehicle	driving cycle peak	working point
	torque [Nm]	Torque [Nm]	RPS [rev/s]
B segment*	1212	88.1	14.6
C segment*	1709	112.0	13.9
C segment*	1687	119	13.9
D segment*	1669	138	16.1
E segment*	2337	100.6	13.0

Physical dimensions and the weight of the vehicle are taken from the gasoline version of the mentioned representative vehicles. Top speed and some other parameters are taken in such a way that they best represent a possible electric version of the vehicle in the same segment.

6. MECHANICAL DESIGN POSSIBILITIES OF AN IN-WHEEL MOTOR

There are several design possibilities for in-wheel motor and other required elements near the wheel. Among these elements, the brake, bearing, motor position sensor and seals are the most important. In the case of front drive, the A-arms for the steering mechanism have to be integrated as well and the attachment to the suspension has to be considered in all cases.

Let us examine all of them systematically and list their advantages and drawbacks. In-Wheel motors

• In-wheel motors with gears (reduction gear or a gear-box)

  • Michelin [5]
  • Toyota Fine X

• Eliica (Keio University, Japan) (Figure 11)

• Direct drive in-wheel motors

• Elaphe Propulsion Technologies

• SIM-LEI [15]

• Protean Electric

• Siemens VDO [16]

• Schaeffler [17]

• TM4 [18]

• Evans Electric Limited [19]

• e-Traction [20]

• Fraunhofer hub motor

Generally, direct drive in-wheel motors are divided into motors with:

• Outer runner
• Inner runner

6.1. In-Wheel Motors with Gears

In-wheel motors are generally divided into motors with gears and direct drive motors. Some producers have reported a use of a reduction gear [1] or even a whole gearing system, such as Eliica research car from the Keio University in Japan [1]. Implementation of a gearing system allows the usage of conventional high speed electric motors with a low specific torque and in the applications with a gear-box a planetary gear-box system is usually implemented as it provides a high gear ratio (3 to 35). Furthermore, from the point of view of the mechanical structure, other types of gear boxes are even less convenient. But it must be taken into account that a gearing system means an additional weight of the propulsion system, approximately 0.1 kg/Nm [1].

It is typical for motors with any gearing system and a conventional low torque high speed motor [1, 5] to be axially longer. This means they are not protected by the rim of the vehicle, so damages can occur when parking on the pavements or simply by the stones and obstacles on the road. Moreover, such motors present notable constraints when mounted into the steering wheels as it is technically very inconvenient to steer axially long and overall complex assembly and mechanical construction. A conventional low torque and high speed motors connected to the gear-box or reduction gear usually have a lower diameter than a wheel itself and therefore take space right in the center of the wheel, where a brake is placed in the conventional cars. Therefore a standard brake usually has to be replaced by a newly developed braking system and installed outside the center of the wheel. This not only increases the cost of such braking topology and often reduces the braking ability of the vehicle, but also causes steering torques on the steering system, which can result in a dangerous situation when all the braking wheels do not have equal grip to the driving surface. Regarding the energy efficiency, the coupling of the motor and gearbox itself implicate additional energy dissipation and therefore reduce the power and range of such vehicles.

Figure 11. Axially long low-torque high speed motors of the Eliica research car from the Keio University in Japan with a gear-box [1].

As already mentioned in section 4, the direct drive motors typically boast high torque already at low RPM and therefore present a notable potential for a high performance passenger car. Moreover, they are usually axially short so that they can be entirely implemented into the rim of the car. This protects them from environment (stones, splashing water) and makes it safer to park them on the pavement. As such motors are designed in a way to have active parts of the motors distributed only on a large circumference and are therefore "hollow" at the lower diameters, there is a space left in the center of the vehicle for the implementation of a conventional mechanical brake. Due to a less complex construction and absence of the gearing, these kinds of motors are also lighter, which means lower unsprung mass and consequentially higher durability, driving comfort and car turning agility. As the motor can be entirely integrated into the rim of the car, there are no more constraints with the maximum steering angle of the car. Wheels can theoretically steer even at the angle of 90° and can therefore benefit from this uncompromising maneuverability and can park sideways.

Increased unsprung mass of the wheels is a concern engineers come across when developing in-wheel electric motors. Higher mass of the wheel also means higher inertia of the wheels, which results in higher acceleration forces from the road on the wheels. A lot of skepticism has been expressed whether these factors can have a negative impact on the vehicle"s stability, driving comfort, safety, passenger"s comfort in the cabin and the durability of the bearing and mechanical components of the motor [11]. The distribution of the mass of the vehicle is also notably changed as the mass of the engine is moved from the center of the vehicle towards the outer edges. It increases the moment of inertia of the vehicle around the vertical axis through the center of gravity of the vehicle so the handling performance of the vehicle is decreased. However, tests showed that this can only influence the performance of racing cars and sports cars and cannot be noticed in an everyday ride with a passenger car. Schalkwyk and Kamper [6] have presented a simple mathematical model of the quarter of the vehicle"s suspension for the analysis of the mentioned parameters on the vehicle"s stability and comfort, as shown in Figure 12.

Figure 12. A mathematical model of the quarter of vehicle"s suspension.

In the Equations (3) and (4), M^v and M^s are the sprung and unsprung masses, K^t and K^s are the spring coefficients (of the tire and of the damper), while damping coefficients are expressed by B^t and B^s . The dynamic equations of the system are as follows:

$$M_V \ddot{z} = -K_s (z - y) - B_S (\dot{z} - \dot{y}) - M_V g \tag{3}$$

$$M_V \ddot{z} = -K_S (z - \dot{y}) - B_S (z - \dot{y}) - M_V g$$

$$M_S \ddot{y} = K_S (z - \dot{y}) + B_S (\dot{z} - \dot{y}) - K_T (y - \dot{x}) - B_T (\dot{y} - \dot{x}) - M_S g$$

$$\tag{4}$$

Interestingly, Schalkwyk, Kamper [6] and Anderson and Harty [4] have shown a slight impact of the in-wheel motors on the vehicle"s stability, maneuverability and driving comfort. The wheel hop (a phenomena where a wheel loses contact with the ground) almost never occurs and the unsprung mass does not decrease the driving comfort of passengers. It does, however, imply higher vibration forces to the motor hub, which can decrease the endurance of certain components of the vehicle, such as the bearing. Furthermore, the components of the motor must be constructed in a way that makes them resistant to such working conditions [9- 10].

Anderson and Harty [4] have presented a Cooper rating scale describing some particular properties and performances of the vehicle with integrated 30 kg in-wheel motors in each wheel.

As can be noticed from the figures below, only a few performances of the car can be notably changed in a car powered by the in-wheel motors. Considering the ride, these two parameters are pitch control and a great impact feeling. As already mentioned, these two factors are only notable at higher speeds and sharp cornering so they do not really affect passenger cars (Figures 13-14) [11-17].

Considering the handling properties, tracking stability and yaw stability can be influenced, yet again it has been discovered that the most worrying cornering grip level remains unaffected.

Figure 13. Ride behavior of the vehicle with 30 kg unsprung mass at each end [4].

Figure 14. Handling properties of the vehicle with 30 kg unsprung mass at each end [4].

6.2. Direct Drive In-Wheel Motors

As far as the mechanical design of direct drive in-wheel motors is concerned, they are divided in two main groups: motors with an outer runner rotor and motors with an inner runner rotor. During almost 30 years of research and prototyping, we have produced and tested the following main types of the in-wheel motors.

6.2.1. Outer Runner Motors

The most typical and at the same time the most convenient topology of the direct drive in-wheel motor is the one where the standard car brake can remain in the hub of the wheel. In such cases an expensive development and homologation of a new braking system is avoided and a braking performance remains unchanged. Drawbacks of such a topology can be heat transfer from the brake to the motor, which demands a higher performance of the cooling system and can, in extreme cases, also harm the magnets. Moreover, such motors demand high diameter seals, which are usually expensive to make, complex to install and produce high energy dissipation and high braking torque. As the seals have a large diameter, the lowermost part of the seal is much closer to the road than a small diameter seal, which means a higher probability of the ingress of water when sealing is not perfect. But on the other hand, these types of seals allow such a design that the condensed water can exit either at standstill or at high speed. Two seals are necessary for this solution as the motor has to be opened on the outer side of the wheel to ensure the proper cooling conditions for the brake. The described solution can be implemented around different types of the brakes [18-25].

A motor with a disc brake and a caliper between the motor and the attachment plate (Figure 15) boasts the advantage of good cooling capabilities for the disc brake and brings no stator lower radial dimension boundary constraints, as the caliper"s upper boundary can obviously be on a larger radius. Moreover, the arrangement of this kind of motor is very convenient, but maintenance of brake discs and changes of brake pads can be complicated in some implementations. A seal can be on a smaller diameter than in the previous solution, which means lower friction force and low energy dissipation. The rotor plate can be closed on the outer axial side of the wheel so only one seal can be used and even this one on a small diameter, consequently causing little energy loss. Such a topology of the drive is axially wider than the previously described one.

Figure 15. The disc brake mounted between the motor and the attachment plate/mounting point.

The already mentioned energy loss of the seal is also presented by the Equation (5), where T^B is braking torque, kfr is a coefficient of friction, Fpl is a preload force and R is the radius on which the seal is mounted.

$$T_B = kf F{pl} R \tag{5}$$

It is important to note that the preload force is proportional to the circumference of the seal as presented in Equation (6):

$$F_{pl} \propto 2\pi R$$ (6)

Based on Equations (5) and (6) we can conclude that the braking torque T^B is proportional to the R , clearly supporting engineers" aims to mount the seal on the lowest radius possible and with a lowest preload possible, while still ensuring the appropriate sealing properties.

The solution with the brake on the inner side of the attachment plate (Figure 16) shares the majority of properties with the solution where the brake is between the attachment plate and the motor, but is axially even wider, so that the steering constraints are much more limiting.

Figure 16. The disc brake mounted on the inner side of the attachment plate.

An unconventional solution for refined products is a motor with a brake disc attached to the rotor and a caliper attached to the strut of the stator. One large seal is required for such an application but the disc and the caliper are unconventional, which requires a development and homologation of a new concept of the brake. Unfortunately, the disc (and in some topologies the caliper) is thermally coupled with the rotor. This can be a limiting factor for the performance of the motor as too much heat from the disc could dissipate to the magnets.

The brake is mechanically outside and uncoupled with the motor and the rim. The main advantages of such a construction of the motor are two seals on a very small diameter producing the lowest amount of friction of all the solutions presented so far. Such a motor, called Elaphe*3, requires a two-sided mounting. It is therefore not usable in modern cars but has shown a great reliability as a scooter motor.

The solution of the brake between the motor and the rim is an absolute winner from the point of view of the produced total friction, as it requires only one seal on a small diameter but faces issues with a caliper arm design.

6.2.2. Inner Runner Motors

Generally speaking, motors with an inner runner can be slightly more difficult to produce as the winding faces towards the center of the motor so that there is less space for the tooling and machine manipulation. Moreover, the magnets are attached to the outer side of the rotor, which means the centrifugal force tries to tear them off. This way, in case of glue (or other attachment means) failure, the magnet is thrown into the airgap, where it is ground by the friction between the stator and rotor and the motor loses its power and gets jammed. In case water cooling is required for a particular type of motor, the motor with inner runner is also less favorable. As we are speaking about in-wheel motors we have to be aware of the outer diameter limit. It is defined by the rim size and, as the cooling system of the stator is on the opposite side of the winding, it means the winding, and consequently the air gap, have to be on a lower diameter. With this, some torque potential of the motor is lost. It is true that the total cooling area can be larger in such cases, but it is usually not a critical factor.

There is a strong analogy of the properties, advantages and drawbacks between the designs of motors with inner and motors with outer runners. As there is a clear logic behind it, different types of inner runner motors are only graphically presented in Figure 17, while their properties are presented in Table 6.

Figure 17. The brake inside the motor – the inner runner design.

7. SOME TYPES OF THE IN-WHEEL MOTORS

Different types of vehicles have their own specific requirements when talking about the in-wheel motor integration, especially when the prototype motors are produced for the conversion of the stock vehicle and when motors have to be developed for a newly designed electric vehicle [14]. The boundary conditions for an in-wheel motor integration are usually the following:

• Inner diameter of the rim;
• Outer diameter of the brake (usually the limiting factor is a drum brake or a disc caliper);
• Attachment points of the motor
  • o Specific original mounting points on the car or specific knuckles
  • o Specific bearings and their integration;
• The motor has to be adjusted to the braking system of the vehicle, as newly designed brakes are very unfavorable in the automotive industry, also because of the homologation requirements and costs;
• Heat dissipation from the brake should not be limited after the integration of an inwheel motor. Heat should also not be transmitted to the in-wheel motor where it could limit its performance or even cause damage. When there is no other design option, the heat from the brake can only be dissipated to the motor"s cooling system instead on the active parts, were magnets are the component most at risk.

For research purposes we have designed, produced and tested several prototypes of inwheel electric motors with inner or outer runners in order to recognize specific challenges and advantages of every single design and its compatibility with certain designs of a braking system.

A closer look at Table 6 and Figures 15-17 shows us that, generally, a motor with an inner runner provides much more conventional options for seal integration and its full performance. It is also more compact, stiff and provides better options for the cooling system integration but its cooling system cannot absorb the heat dissipated by the brake.

Table 6. Reviewing some types of the in-wheel Elaphe motors

EM		Example	Key advantages
Solution Outer runner	Brake inside the motor	Elaphe*Chebela1 Elaphe*Smart1 Elaphe*Smart2 Protean Electric in-wheel Siemens VDO e-corner Schaeffler E-wheel drive e-Traction The Wheel Fraunhofer hub motor	Compact design, access to the brake without removing the motor from the car
	Brake between the attachment plate and the motor	Elaphe*Chebela2	Easy rim mounting/removal, small seal diameter and consequentially low friction, only one seal, simple brake pads change
	Brake on the other axial side of the attachment plate		Robustness, brake thermally decoupled from the motor, easy brake pads change, only one seal
	Brake (disc brake) attached to the rotor, caliper attached to strut or stator	Elaphe*Sports Car	Low axial space requirement, low weight, only one seal
	Brake outside the motor and the rim	Elaphe*3 (scooter motor)	More simple realization of environmental protection for the motor
	Brake between the motor and the rim		Only one seal used, not much additional weight, caliper from the inner side
Slim Bearings		Elaphe*1 and Elaphe*2
Inner runner	Brake inside the motor		Compact design, access to the brake without removing the motor from the car
	Brake disc attached to the rotor		Only one seal
	Brake on the other side of the attachment plate	Elaphe*Hycat3 ATV motor Elaphe*Hycat1 Elaphe*Fido motor	Robustness, thermally decoupled from the motor, easy removal of the brake, only one seal
	Brake outside the motor and the rim Brake between the motor and the rim		Only one seal, thermally decoupled from the motor

In reality, it turns out the brake is usually much more accessible in the motor with an inner runner and the adjustability for different types of unconventional vehicles (e.g., with tracks) is higher. The cooling abilities for the magnets turn out to be worse than in the case of an outer runner for the following reasons: the total outer surface is smaller, circulation of the air is at a lower rate than at the outer side of the motor and in cases when the brake is integrated into the wheel itself, the probability for heat ingress to the magnets is higher than in the case of an outer runner [16, 18, 19, 24, 25].

Crucial factors when talking about the mounting specifics of an electric motor are the type and position of the position sensor (usually hall sensors, incremental or absolute encoder). Its type depends also on the type and configuration of the inverter.

In both cases (inner or outer runner) a braking system which is thermally decoupled from the motor is preferred. The easiest way is when the brake is mounted on the inner axial side of the motor, though space saving advantages of the direct drive in-wheel motor are the most obvious when the brake is integrated directly inside the wheel.

CONCLUSION

Mobility trends in western societies (with eastern following quickly) are obviously moving towards technologies of cleaner personal transport and concepts of small, compact and agile vehicles with a high level of space utilization. The concept of the direct drive inwheel motor is obviously the brightest star on the horizon for the mobility of the coming decades, not only because of the technological advantages presented in this section, the high efficiency, high power/weight and torque/weight ratios and a very competitive production cost, but also because of its unbeatable convenience for the user and society. The research and development trends are therefore expected to notably increase in the field of unsprung mass of the vehicle"s components as well as in the field of great space utilization from the point of view of the active parts of an electric motor. In the majority of vehicles, all active components in a motor have to be implemented on a relatively thin radial space between the brake and the rim, which suggests a notable increase in usage of the high number of pole pairs in permanent magnet motors. Moreover, from the electrical point of view, the present lack of standard inverters for high currents in a relatively specific frequency range is expected to accelerate and deepen the research in this field, in order to provide the demanded supply of controlling units for electric motors in different segments and topologies of the electric propulsion systems.

REFERENCES

• [1] Gerling, D.; Dajaku, G.; Lange, B.; Electric Traction for Automobiles – Comparison of Different Wheel-Hub Drives. University of Federal Defense Munich; FEAAM GmbH, Munich & Neubiberg, Germany.
• [2] Pavasovic, U.; Internal Combustion Engine. http://picture_Internal_combustion_engine.
• [3] Lampic, G.; Slivnik T.; Detela, A.; Holistic Approach in Developing Propulsion System for Urban Electric Vehicles. IEEE Vehicle Power and Propulsion Conference, Chicago, USA, 2005.
• [4] Anderson, M.; Harty, D.; Unspring Mass with In-Wheel Motors – Myths and Realities. 10th International Symposium on Advanced Vehicle Control, Lough borough, UK, 2010.

• [5] Michelin Active Wheel. Paris Motor Show, Paris, France, 2008.

• [6] Schalkwyk, D.V.; Kamper M.; Effect of Hub Motor Mass on Stability and Comfort of Electric Vehicles. University of Stellenbosch, Stellenbosch, South Africa.

• [7] Kim, D.; Shin, K.; Kim, Y.; Cheon, J.; Integrated Design of In-Wheel Motor System on Rear Wheels for Small Electric Vehicle. World Electric Vehicle Journal, vol. 4, pp. 597- 602, 2010.

• [8] Wang, P.; Guang Ning, W.; Gao, B.; Luo, Y.; Cao, K.; Study of Partial Discharge Characteristics at Repetitive Square Voltages Based on UHF Method. China Technological Sciences, vol. 56, no. 1, pp. 262-270, 2013.

• [9] Luo, Y.; Tan, D.; Study on the Dynamics of the In-Wheel Motor Systems. IEEE Transactions on Vehicular Technology, vol. 61, no. 8, pp. 3510-3518, 2012.

• [10] Muller, S.; Uchanski, M.; Hedrick, K.; Estimation of the Maximum Tire-Road Friction Coefficient. Journal of Dynamic Systems, Measurement, and Control, vol. 125, no. 4, pp. 607-617, 2013.

• [11] Besselink, I.; Nijmeijer, H.; Influence of In-Wheel Motors on the Ride Comfort of Electric Vehicles. Eindhoven University of Technology, Department of Mechanical Engineering, Dynamics &Control, Eindhoven, The Netherlands, 2010.

• [12] Janulevicius, A.; Pupinis, G.; Damanauskas V.; Effect of Tires' Pressure on the Kinematic Mismatch of a Four-Wheel-Drive Tractor. Mechanika, vol. 19, no. 1, pp. 73- 80, 2003.

• [13] Hartani, K.; Bourahla, M.; Miloud, Y.; Sekour, M.; Electronic Differential with Direct Torque Fuzzy Control for Vehicle Propulsion System. Turk Journal of Electrical Engineering and Computer Science, vol. 17, no. 1, pp. 21-38, 2009.

• [14] Gaing, Z.L.; Chiang, J.A.; Robust Design of a Direct-Drive PM Hub Motor by Fuzzy-Inference Taguchi Method. International Journal of Applied Electromagnetic sand Mechanics, vol. 36, pp. 295-307, 2011.

• [15] Yirka, B.; SIM-Drive Corp Announces New 'In-Wheel' Electric Car. http://phys.org/news/2011-04-sim-drive-corp-in-wheel-electric-car.html.

• [16] Siemens VDO; Technology for the Environment – Environmentally Friendly Transportation. www.siemens.com/innovation/en/publikationen/publications_pof/pof_ spring_2007/technology_for_the_environment/green_transportation.htm.

• [17] Schaeffler; The Drive for the City Cars of Tomorrow – Schaeffler Displays Wheel Hub Drive E-Wheel Drive. www.schaeffler.de/content.schaeffler.de/en/press/pressreleases/press-details.jsp?id=23019393.

• [18] TM4; Electric Corner Module.www.tm4.com/en/electric_corner_module.aspx.

• [19] Evans Electric; The Automotive Mega Trend of the 21st Century –The Electrification of the Automobile. www.evans-electric.com.au/.

• [20] E-Traction; TheMotion.www.evans-electric.com.au/.

• [21] Whitehead A.; Plug-in Hybrids Made Easy - Selectable EV Drive Using In-Wheel Motors. Protean Electric Ltd, Farnham, UK, 2011.

• [22] Rahman, K.M.; Application of Direct-Drive Wheel Motor for Fuel Cell Electric and Hybrid Electric Vehicle Propulsion System. IEEE Transactions on Industry Applications, vol. 42, no. 5, pp. 1185-1192, 2006.

• [23] Busse, M.; Development of Components for Electric Vehicles. Fraunhofer Institute for Manufacturing and Advanced Materials (IFAM), 2010.

• [24] Kock, A.; Mertens, A.; Fault Tolerant Wheel Hub Drive with Integrated Converter for Electric Vehicle Applications. IEEE Vehicle Power and Propulsion Conference, Seoul, Korea, pp. 19-23, 2012.

• [25] Juris, P.; Brune, A.; Ponick, B.; A Coupled Thermal-Electromagnetic Energy Consumption Calculation for an Electric Vehicle with Wheel Hub Drive Considering Different Driving Cycles. IEEE Vehicle Power and Propulsion Conference, Seoul, Korea, pp. 28-31, 2012.

In: Autonomous Vehicles ISBN: 978-1-63321-324-1 Editors: N.Bizon, L. Dascalescu and N.M. Tabatabaei © 2014 Nova Science Publishers, Inc.

Chapter 19

THE FUTURE OF ELECTRICAL VEHICLES

Wim J. C. Melis

University of Greenwich at Medway, Kent, UK

ABSTRACT

Electrical Vehicles (EV) entered the transportation market much earlier than their combustion engine driven counterparts. After a long period of no development, currently they have found their way back into use, but then mainly as small goods vehicles on large estates where they then often carry a label in the sense of "Zero Carbon Vehicle". Due to the increasing social pressure for environmentally friendly transportation, their zero emissions comes to their benefit also for mainstream goods and person transport. However, there are still a few problems to be resolved to smoothen the path for a better market penetration. Even though the electrical motor has a much higher efficiency, in comparison to fossil fuel powered vehicles, overall, electrical vehicles suffer from a limited range due to the lower energy density for electrical storage. Current hybrid vehicles provide a solution to this problem by using an internal combustion engine used as a range extender. The environmental impact of this range extending motor as well as the actual electricity production itself both have an environmental impact unless green electricity is being used as an energy source and used to charge the plug-in hybrid. The environmental impact depends on the source-fuel being used, but in essence highlights the importance of finding more environmentally friendly energy sources. Additionally, it is essential to improve the overall efficiency of our transport vehicles and any energy conversions that take place during transport, by for example recovering the kinetic energy of the vehicle when it is being decelerated. As much as there has been a move towards electrical vehicles, there has been a move towards more Information and Communication Technology (ICT) within vehicles for a variety of reasons, moving towards electrical vehicles the integration of ICT with the actual vehicle is likely to become easier and therefore more efficient allowing for vehicles to become fully autonomous.

Keywords: Fuel efficiency, environmental impact, hybrid vehicles, bio-fuels, alternative fuels, energy storage for transport

E-mail: wim.jc.melis@gre.ac.uk.

1.INTRODUCTION

Even though electrical vehicles have been around since the early 20th Century, they did not quite make it to the mass market for a variety of reasons. Looking at the current situation, there are a number of reasons why they are more likely to break through now than they ever could have in the past. One of them is the environmental concern in relation to fossil fuel powered vehicles, which are generally considered to be none environmentally friendly. In the past, vehicles with lower consumption have been presented, but they never quite made it to the market. Within the search towards more environmentally friendly vehicles, hybrid and electrical vehicles have been considered to form the solution. This is partly related to the efficiency of an electrical vehicle being rated at 80-90% [1] in comparison to the efficiency of fossil fuel cars, which is only 25-30% [2, 3]. One however needs to look at the full picture before drawing any premature conclusions about the environmental friendliness of the electrical vehicle, since a lot depends on how the actual electricity is generated.

One way to improve the integration of electrical vehicles has been the introduction of hybrid vehicles, and these are gradually being developed towards plug-in hybrid vehicles. This should allow for the vehicle to be charged by green power, while still managing a long range due to the presence of another engine that provides for added acceleration and range. Being able to charge the vehicle with green energy does not only make it more environmentally friendly, but also helps with the need for grid storage when renewable sources are being used for energy generation. Hybrid vehicles as such, benefit from the advantages of two worlds when it comes to the acceleration, however, they do not manage to fully recover energy during deceleration. Hence, fully regenerative breaking is being presented, its efficiency investigated and advantages discussed [4].

Current hybrid vehicles have, besides an electrical motor, another energy source that helps during acceleration. This other energy source is generally a combustion engine, or alternatively a Fuel Cell can be used when the energy is stored as Hydrogen rather than using batteries. The use of batteries as well as fossil fuels however still has an environmental impact. Therefore, the chapter looks at alternative fuels as well as energy storage mechanisms.

After a short look into the past and present developments of Electrical Vehicles, the first section of this chapter will look into the potential future of the electrical vehicle, and why now it is more likely than ever that electrical vehicles will become mainstream. The next section then looks at current vehicles and their efficiency, comparing EVs with Combustion Engine (CE) vehicles, how to improve their efficiency and also their respective impacts on the environment. The final section then introduces some alternative fuels and storage mechanisms that could be used to overcome some of the problems that EVs currently face.

2. PAST, PRESENT, FUTURE

Electrical vehicles have been around since the early 20th Century. Their history can be divided into a few stages. The very first transportation vehicles were in essence electric and they were well ahead of their combustion engine counterparts. In the early 1920"s to 1930"s there was actually a 3:1 domination in relation to EV registration when compared to their gasoline counterparts [5]. From 1930 onwards, gasoline vehicles overtook the electrical vehicle domination, because of the following reasons: 1) mass production at a reasonable cost which made personal transportation available to the general public; 2) a lack of charging infrastructure combined with the limited range for electrical vehicles; and 3) mass discovery of gasoline resulting in fuel being available in abundance and therefore cheap. From about 1935 till the 1970"s onward, there was political sensitivity with the Organisation of Petroleum Exporting Countries (OPEC), and when the oil-crisis hit in the 70"s only about 300 Electric Vehicles were purchased for fleet testing by the US postal service.

Most other countries did not consider EVs as an option, but purely focused on reducing fuel consumption. Other from this small order, the political and economic environment at the time did not stimulate any further development. Towards the end of the 90"s, a similar, but slightly larger design for EVs, to be tested in a fleet setting, was performed. Several manufacturers either altered existing vehicles to sell them as EV, or designed new vehicles. Even though these vehicles proved to be working well, all of them were rather quickly discontinued, and most of them have meanwhile been destroyed and recycled.

According to a prediction for the next 30 years made in 1999 [6] the hybrid as well as the hydrogen vehicle would have the longest market future, because they deliver the same functionality as our current vehicles at a higher fuel economy for a higher cost.

The only issue to resolve is to find the correct balance between cost and economy. Considering we are about half way through this 30 year prediction, it seems that hydrogen vehicles are still in the prototype stage, however, Toyota has already announced to provide a production model on the market in 2015 [7]. On the other hand, hybrid vehicles have started to take off, as can be seen from the large set of vehicles already provided currently, as well as the ones in the pipeline to market [8]. While Hybrid Electric Vehicles might still have a slightly larger sale price, over their lifetime, they benefit from a better fuel economy when compared to combustion engine based cars [9]. On top of that, they also benefit from the increase in social responsibility in relation to environmental concerns and consequently government has started to support this technology more substantially than it ever did in the past.

The further success of the electrical vehicle now depends on a variety of different factors, not all of them technical. First of all, there is the requirement for good education around EVs to ensure that people understand their benefits over the existing systems. Secondly, there is the financial support by the government to overcome the initial higher investment requirements related to this "new" technology hitting the market. The duration of this financial incentive might enable the technology to take off or, if stopped too early, stop market penetration while still at a crucial stage. As in relation to the social responsibility, the benefit of the electrical vehicle is that the "fuel", i.e. electricity, can be generated in a variety of different ways.

As mentioned further on in this chapter, unless the fuel production is green, the electrical vehicle does not justify some of its currently green image, but considering that electricity generation is changing that will certainly become the case in the not so distant future. On a technical side there are challenges in relation to: energy storage, range expansion and a charging infrastructure to be dealt with. However, the energy stored within the vehicle could actually become a benefit in relation to the increased use of renewable energy sources. The problem with this kind of sources is that they are less "reliable" in the sense of not providing a continuous supply. For example, sunlight cannot be guaranteed, and the same applies to wind. In most cases, the time when electricity is generated will not be the time it is actually needed; therefore temporary storage of energy is essential. EVs could provide the ideal energy storage required as we move to more renewable energy sources [10]. This approach is generally referred to as Vehicle-to-Grid (V2G) integration and has been studied in several countries that plan to introduce more renewable energy generation [11]. The principle is generally that the EV form a flexible load to the network so that it can charge when there is an excess of energy and when required they can put energy back into the grid to compensate for a shortage. To allow this to happen, certain intelligence would be required that could either form part of the grid or of the actual electrical vehicle itself [12]. This intelligence would not only need to take the grid into account, but also the specifics of the EV storage system, e.g. battery type, and the required urgency of the actual charging [13].

The influence of ICT on vehicles has been significant, and this will be even more the case for EVs, which will be even more ICT driven/controlled [14]. The move towards more ICT in current vehicles has had several reasons. Firstly, there is the engine management system, which due to the increasing performance and exhaust requirements has led to the need for a more precise control of the actual combustion cycle. Hence, current engines have many sensors to measure a variety of characteristics in order to improve the overall engine performance.

The second reason has been the continuous drive to increase "luxury" within the vehicle, which includes items like multimedia equipment, navigation systems among systems like driving aids (i.e. cruise control) and so on. Although several of these systems have already been available for a number of years, they are currently all being integrated and combined with one another to improve the overall user experience and friendliness. This development is combined with the continuous drive towards more intelligent vehicles, which started with the so-called drive by wire, and is currently being driven by a move towards fully autonomous vehicles. The latter make extensive use of different "measuring" mechanisms generally combined with a large amount of processing power in order to replace the human driver. Although a large skepticism exists when it comes to allowing computers to drive our vehicles, it is said that if computers would be driving, then that would eliminate all our traffic jams due to the possibility of more efficiently usage of the road [15].

One approach that could overcome some of the initial problems with these drive-by-wire systems is the move towards so-called semi-automatic systems, where a car would platoon onto a set of cars [16]. A professional driver drives the vehicle at the front of the platoon and other vehicles would join a platoon based on their destination. This would allow for a driver to be able to relax/work part of the journey whereas a certain part might still need to be driven by him/her-self. This would not only provide benefits to economy, but could also lead to a reduction in exhaust, as long as the energy required to deal with the processing for this autonomous driving is not excessive. Reducing the amount of energy required for autonomous driving could come from more intelligent processing, which could be achieved by either different algorithms and/or processing platforms.

The problem with current processing platforms is that for most algorithms used for autonomous driving, a large set of calculations need to be performed before an actual result is obtained, and this result can then still have a significant number of false positives. To overcome these problems more intelligent systems would be required. Even though neural networks have promised such intelligence, they have not shown to be practical for a variety of reasons. Considering current technological developments, there are however a number of promising paths as in relation to achieving this intelligence [17, 18, 19] however a detailed study of this subject would be beyond the scope of this chapter.

3. EFFICIENCY OF EXISTING VEHICLES

3.1. Fuel Efficiency

Energy or fuel efficiency for vehicles generally looks at the conversion from energy source or fuel into kinetic energy or vehicle movement. Considering this aspect, electrical vehicles are said to be running at a much higher efficiency than fossil fuel powered cars. This does apply if one considers the definition of fuel to be the fuel on board of the vehicle.

For fossil fuel powered vehicles, the fuel or energy source would be considered to be either petrol or diesel. Depending on which one is being used, the efficiency of turning that into kinetic energy is about 25% or 30% respectively [2, 3]. In contrast with that, electrical vehicles have an efficiency of somewhere between 80 and 90% [1], if one considers the conversion of electrical energy into kinetic energy. This higher efficiency is due to the fact that the electrical motor has been improved to reach these high efficiencies. So even if the energy density of fossil fuels may not be bad, there is a substantial amount of energy that is wasted, since it is converted into heat. Although the electrical vehicle seems to imply that these losses are smaller, it is essential to look at the complete picture, which requires an investigation of the efficiency of the actual generation of electricity.

Electricity generation can be performed in a variety of different ways and depending on the method used, Electrical vehicles will be more or less efficient and environmentally friendly. In order to make a fair comparison between fossil fuel vehicles and electrical vehicles, we will assume that the electricity is generated from fossil fuels, which could be through coal or gas powered power stations. Since coal was used for steam engines and most of these have made it to the museum by now, we shall look at gases as an example fuel for our calculation. Electricity generation from conventional fossil fuels has an efficiency of between 33 and 60 %, depending on whether the source fuel is coal/oil or gas [20, 21]. The reason for this range being rather large is that for pure steam driven power generation the efficiency would be about 30 to 40%. However, if a gas powered station makes use of the socalled combined cycle then efficiencies up to 60% can be obtained, which is due to the extraction of energy from the exhaust gases as shown in Figure 1.

Before the electricity can be stored in the batteries of the electric vehicle, it still needs to travel a long way.

First of all, it needs to be distributed across the grid, where it incurs a certain amount of transport losses. These transport losses would add up to an average of about 8% [22] depending on factors like distance from power station and so on. Hence, the efficiency of generating and transporting the electrical energy to the vehicle, assuming average values for any range provided, would be: 0.50*0.92=0.46. This leads to an overall efficiency from fuelsource, i.e. gas, into kinetic energy of: 0.46*0.85=0.39, which still assumes a perfect efficiency for the charging and discharging of the vehicle"s batteries. The overall efficiency of a battery again depends on a variety of factors, namely the type of battery, type of charger, number of cycles it has been charged/discharged, and so on. However, efficiency numbers of 85% are not uncommon for the types of batteries used in current electrical vehicles. This results in the overall efficiency from gas through to electricity and all the way to kinetic energy to become: 0.39*0.85=0.33 or 33%, which is only marginally better than our diesel powered car, but quite a bit better than the petrol powered car. However, we have not quite compared like with like.

In order to be completely correct, the efficiency number calculated above should be compared with a fossil fuelled car that combusts natural gas, which generally has a slightly higher efficiency than when it runs on normal petrol, although the difference is not massive. Exact numbers of gas powered vehicles are not wide-spread due to the small number of vehicles running on this fuel, but it is generally known that natural gas has a higher efficiency and therefore an engine combusting gas will have a slightly higher efficiency. This higher efficiency, which relates to the higher fuel burning efficiency of gas, can be derived from its octane number. The octane number for gasoline is known to be 95, whereas natural gas would have an octane number of about 130. These numbers can be taken as a measure of how the efficiency of the engine will change depending on which fuel it burns. Therefore, the efficiency improvement can be calculated to be: (130/95)*0.25=0.34. Therefore, as from an efficiency perspective, is seems that burning natural gas in a combustion engine directly is as efficient as using it to generate electricity, which is then used to drive an electric vehicle.

One might wonder why these efficiency values are so close to one another, especially if one considers that when using electrical energy, there are quite a few energy conversions being performed. Well, other from the fact that the efficiency numbers for our electricity generation are rather ambitious when it comes to the power station efficiency, the engine in our vehicle will continuously be accelerating and decelerating, which affects fuel efficiency in a negative way. This clearly provides an opportunity for improving energy efficiency, which will be looked at in more detail later on.

The above comparison has been fully based around one particular fuel, namely gas. There are however other fuels, like diesel, bio/diesel and so on. Using diesel as a fuel provides a fairly similar story, only that diesel engines are generally slightly more efficient due to the higher calorific value of the fuel itself. Bio-fuels on the other hand will again slightly change the efficiency aspects, and are discussed in more detail towards the end of this chapter.

Figure 1. Working principle of combined cycle gas plant: (1) Electrical Generators; (2) Steam Turbine; (3) Condenser; (4) Pump; (5) Heat Exchanger; (6) Gas turbine [23].

Besides the burning of fossil fuels, there are also other methods of generating electricity, which might provide an overall better efficiency. The first one that would need to be considered due to it being rather wide spread is nuclear power, which is quite often classified as a controversial method of energy generation, because, as much as it is being phased into some countries [24], it is being phased out of others [25, 26]. Trying to determine overall efficiency of a nuclear power station is not that straight-forward, since the general measure of looking at input fuel value in comparison to electrical energy output cannot simply be applied, since it is rather difficult to determine the energy value of the input fuel. Due to its controversial nature and the produced waste, depending on the source being listened to it will either be considered as a green way of generating energy or not.

However, considering that sustainability is often described as: "living on this planet without leaving any trace for future generations", one could question whether nuclear power does comply with this definition, due to the waste products generated. Therefore, in order to move to a more green and sustainable transport, nuclear power does not seem to be compliant with this image.

This brings us to green energy generation, which would generally comprise: solar, wind and hydro as energy sources. Either directly or indirectly most of these types of energy relate back to the energy provided to this planet by the sun, and considering that the sun provides more energy in one minute to this planet than we currently consume in one year, it seems that this is the more prominent way towards sustainable energy collection. Considering that solar energy is still provided to us free of charge and has no negative impact on the environment, efficiencies should not play such a major role when considering this energy source other than from the possible return on investment. It is however also interesting to compare these newer ways of energy collection with existing energy generation methods. Since solar and wind are the easier ones to install everywhere around our planet, these will be looked at in more detail. The efficiency for solar panels largely depends on the type of solar panel being used, however average efficiencies would range between 10 and 30% [2]. Considering that the theoretical efficiency limits of a single cell are at 30%, multi-layered structures will be needed to reach higher efficiencies.

Depending on the number of layers, these efficiencies can go up to 86% for concentrated sunlight [27]. On the other hand, wind turbines would have a maximum efficiency close to 60 %, but on average most practical turbines would have an efficiency of 35-45% [28]. Hence, as in comparison with fossil fuel generated electricity, these efficiency percentages are not that different, however these technologies might allow for the electricity to be generated closer to the actual consumption and consequently reduce the distribution losses, not to mention that the energy came completely free in the first place. However, these sources are generally not continuous and cannot be switched on/off, therefore a storage system is of significant importance, for which EVs could be useful. When one considers the electricity to be produced from a renewable source, then the overall efficiency from solar energy to kinetic energy produced in the electrical vehicle is not that different from using fossil fuels. However, since the actual generation is that much more sustainable, this also means that the vehicle as such will also become much greener, as will be worked out in detail later on.

Needless to say that this is an incomplete list of energy generation methods, but considering that this book deals with electrical vehicles and not with energy generation, this would bring us outside the scope of this book. It is however important to notice that an electrical vehicle is not much more efficient than the current fossil-fuelled vehicles.

Conversion Process	Current Efficiency (%)	Physical/Theoretical Limit (%)
Gas/Coal to electricity	30 - 40	-
Gas (combined cycle) to electricity	60	-
Solar - Photo Voltaic	10 - 30	86
Wind Turbine	35 - 45	60
Fuel Cell	60 - 85	-
Combustion Engine - Petrol	25	-
Combustion Engine - Diesel	30	-
Combustion Engine - Gas	34	-
Electrical Motor	80 - 90	-

Table 1. Efficiencies of energy conversions

However, if the electricity generation is greener, then using an electrical vehicle suddenly becomes more beneficial, but more on this later. A summary of the different efficiencies can be found in Table 1.

3.2. Increasing Efficiency

In an ideal world, all efficiencies would be 100%, but as seen in the previous section, this is rarely the case for any type of energy conversion. The question is however whether these efficiencies can be improved, as has been the case for the electrical motor where a significant amount of research has led to efficiencies of up to 90% and sometimes even higher. When considering electrical vehicles there are however a large set of factors involved, and the largest improvements would clearly be reached by looking at the factors, which currently have the lowest efficiencies.

In order to improve overall efficiency, the first factor that needs to be considered is the actual energy generation. Vehicles running on fossil fuels have been around for quite a while and although some improvements are still being discovered, very few of those actually reach the mass-market for a variety of reasons. When looking at electricity generation based on fossil fuels, then one can notice that this is largely based around steam turbine technology, which has been running close to optimal efficiency for quite a while now. On the other hand, there are the green technologies, which are still being improved on continuously, although also these methods will run against their physical limitations one day sooner or later.

The next factor where improvement is possible is the actual electricity transportation or distribution. With losses being on average only 8%, there is very little room for optimisation. However, any optimisation that can be achieved would have to come from introducing more localised generation, and therefore shortening the distance between generation and consumption.

Another factor is the actual energy storage in batteries. Battery technology, especially liteon, is continuously improving, which results in improved charge/discharge cycle efficiencies. However, the efficiencies for this factor were already quite high and can therefore only improve marginally.

When considering the actual use of vehicles, then a significant amount of energy is used to accelerate the vehicle itself. However, in many traffic situations, like city traffic, deceleration is as frequent as acceleration, which is generally reflected in higher fuel consumption. As much as acceleration generally consumes energy, deceleration tends to waste energy, because during braking, kinetic energy is converted into thermal energy.

Considering an average journey, the majority of fuel will be used for acceleration and only a minor amount of energy will be required to keep the vehicle at speed. However, at some point in time, the vehicle will need to decelerate again and rather than wasting the kinetic energy it would be useful if this energy can be stored to reuse during acceleration.

Obviously any of these conversions from kinetic to storage and back to kinetic might not have perfect efficiency, but overall, the effect of this temporary storage should still be better than converting all kinetic energy into heat. Hybrid vehicles perform this conversion by using the electrical motor as a generator during deceleration in order to recharge their batteries. This system is however mounted besides a conventional hydraulic break system for several reasons, the first being safety, the second one being purely technology related. A single generator would namely not be able to bring the vehicle to a full stand still in a sufficiently short timeframe for what could be described as emergency braking.

Hence, to be able to use regenerative braking to stop a vehicle completely a few problems need to be overcome. Firstly, there is the safety aspect, which leads to the requirement that the vehicle needs to be able to stop in case of emergency. This problem can be overcome by over-designing the system so that it can deal with emergency cases, including when the vehicle is fully loaded or even overloaded. Vehicle load plays an important part in braking since the force that needs to be countered during the braking action depends on vehicle mass and speed. Secondly, an electrical generator has certain efficiency at a certain rpm. This means that below this threshold rpm, the generator will no longer convert kinetic energy into electrical energy and therefore no longer extract kinetic energy from the system at all. This means that for low rpms, it becomes more difficult to bring the system to a full stop.

Considering that from the safety aspect the system needs to be overdesigned that would mean that unless the vehicle is overloaded and running at maximum speed, the vehicle users" would experience a shock when the generator is switched on for braking because the generator will extract the maximum amount of kinetic energy and therefore immediately perform a maximum braking force, whereas the actual mass of the system is lower and possibly also its velocity, and therefore the vehicle would slow down very quickly. This step response approach is clearly far off from the more gradual braking that currently applies within our vehicles, therefore a more gradual mechanism for the regenerative braking needs to be developed.

The work presented in [4] considers these two problems and presents a solution that allows for fully regenerative breaking. The problem of generator efficiency is overcome by using multiple generators, each one being connected to the main driving shaft through a different gear ratio. This method ensures that there is always at least one generator within its optimal rpm range and therefore able to extract energy from the system. Figure 2 shows the difference between using a single and multiple generators for regenerative braking, in this case: 4. One should keep in mind that this diagram is slightly simplified, since the braking slopes for the multiple generator case on the right-hand side should have slightly different slopes depending on how many generators are extracting energy. This is due to the fact that the amount of energy extracted from these generators depends on the rpm they run at, and also on whether they are switched on or off, since the current model would not switch them all on at the same point in time, but introduce them depending on the braking demand as will be shown next. Figure 2 clearly indicates that multiple generators lead to a better efficiency, since a full range of vehicle speeds can be dealt with.

The second problem can be overcome by connecting/disconnecting the load to these generators at a certain frequency. When this pulsing is done quickly enough then the average load presented to the generator depends on the length of the period that the generator is connected to the load as shown in Figure 3. Figure 3 clearly indicates that if this pulse is shorter, then the average load to the generator is lower, whereas if the pulse becomes longer then the average load increases. This increased load also means that more kinetic energy will be extracted from the moving vehicle and therefore the braking force will be higher.

This approach has certain additional benefits that extend beyond the provision of fully regenerative braking. The main advantage of this system is that there are benefits during the actual acceleration since the different gear ratios now ensure a more continuous acceleration independent of vehicle speed. There is also likely to be a higher overall efficiency due to the fact that multiple generators are being used and each one of them is supposed to run at optimal efficiency at some point in time. Calculating the overall efficiency will depend on many different factors, which include: speed of vehicle, required slow-down, efficiency of the individual generators and their gear ratios. Intuitively considering that there are multiple generators one has to keep in mind that not all of them will be running at optimal efficiency. However, overall the efficiency of the system should be higher than that of a single generator. This higher efficiency does however come at the price of requiring more materials, since one requires not only multiple generators, which from a generator perspective would be similar in price to a single larger generator, but one also requires a larger mechanical construction, since each generator needs to be connected to the main shaft with a specific gear.

Figure 2. Efficiency difference between using a single and multiple generators for regenerative braking.

Figure 3. Pulsing the connection between load and generator allows for a different average load to be presented to the generator.

In order to fully judge the impact of these extra materials, one would actually need to perform a full life-cycle analysis.

3.3. The Environmental Impact of Electrical Vehicles

Ideally, one would perform a full life-cycle analysis of electrical vehicles and compare it with other types of vehicles to draw a proper conclusion on environmental impact of each of the different types of vehicles [29, 30]. For example, if one looks at electrical vehicles, then there would be one or multiple electrical motors, and each of these motors would require a certain amount of copper, which is a limited, finite resource. On top of that, an electrical vehicle will generally have a substantial battery bank installed, and these batteries will have a limited life span, while probably making use of some limited resource materials like liteon, among others, depending on which battery technology is being used. All of this will play a factor in a full life cycle analysis of such vehicles, however looking at such an analysis, it seems that electrical vehicles are not too bad off [29, 30].

One aspect of difference that will play part during the actual transportation is the energy density of the fuel, since for fossil fuel cars the fuel is put in a tank and as the tank empties the weight reduces. For an electrical vehicle, the weight of the storage system, i.e. batteries, will not change, independent of the amount of energy stored. On top of that, batteries are on average rather heavy and this weight also needs to be transported along with the rest of the weight of the vehicle and will therefore have a negative impact on the overall fuel consumption. Additionally, one should know that batteries have a lower energy density than fossil fuels [31]. Although all of these factors play a role when looking at the full carbon footprint of a vehicle, this particular section will however only look at the carbon footprint related to the fuel used for transportation.

The current measure for carbon footprint is generally related to the produced amount of CO2. For vehicles this would be expressed as a certain amount of gram CO² produced per kilometre. Needless to say that such a measure is limited, since it is not only CO2that causes harm to the environment. Hence, there are debates on whether CO2is a correct measure of "damage to the environment", but these discussions are beyond this book and for the purpose of this exercise we will assume this can be considered as an appropriate measure. Due to the heavy weighting of CO2in most studies, this section will first focus on the aspect of CO² after which it will look at the other harmful by-products of burning fossil fuel.

As mentioned before, the environmental impact of an electrical vehicle largely depends on the manner in which electricity is generated, which then again changes across the different continents. Figure 4 shows the amount of CO² produced per kWh of heat and electricity on the different continents. From this figure, one can clearly notice that energy production as a whole is not very environmentally friendly, and the amount of CO2in the world has been increasing. Considering that the electricity production does play a vital part in how green an electrical vehicle is; it will be essential to ensure greener production to ensure that electrical vehicles are greener than fossil fuel vehicles.

As an example, let us compare the amount of CO² generated for a similar size vehicle powered by either fossil fuels or electricity. Considering that the average CO² for all new cars in UK is 158 g CO2/100km [33] and that not all vehicles are new, a value of 160 g CO2/100km is probably quite an optimistic measure.

Figure 4. CO² emissions per kWh of Electricity and Heat Output for the EU-27 level compared to the world, United States, the Russian Federation and China [32].

Over time this number should go down though due to the target set within EU to reach 120 g CO2/100km by 2012. Considering that 2012 is, at the time of writing, just turned past and this target has not been achieved, it seems that quite hard work will be required to catch up with what now seems to have been an ambitious target. On the other hand, this average size car would probably have a consumption of about 7 l/100km and this would relate to about 70 kWh/100km based on the fact that one litre of fuel relates to approximately 10kWh of energy [2].

When one considers the numbers from Figure 4 as an indication of the amount of CO² produced per kWh of energy, and taking that the electrical vehicle would consume the same amount of energy to move forward, then our first conclusion would be that the electrical vehicle is way more polluting than our fossil fuelled vehicle. There is, however, good news, which lies in the aspect that for a similar size vehicle we will need the same amount of energy for kinetic movement. After all, the required amount of energy depends on: the weight of vehicle, the front surface of the vehicle which is a measure of how much air needs to be moved and so on. Considering that in the fossil fuelled vehicle only about 25 to 30% of fuel is converted into actual kinetic energy, we can reduce the amount of electrical energy required to be the same percentage of this 70 kWh, which would be between 17 and 21 kWh for 100 km. Taking that from generation to actual energy usage an electrical vehicle is about 33% efficient, that would mean that the amount of energy to be generated by the power station is: 19/0.33=58 kWh, if we make use of average values. Since our current electricity generation has a CO² production of at least 400 to 500 g/kWh (See Figure 5, for data about Europe), this would mean a total of 58*450=26100 g CO2/100 km for an electric vehicle, which seems to indicate that electrical vehicles are way less environmentally friendly than their fossil fuelled counterparts. Considering that the CO² production of the electrical vehicle is mainly due to the actual electricity generation, if this generation becomes greener, then the vehicle as such will also become greener.

The burning process of fossil fuel, does however not only lead to the production of CO2. Depending on the type of fuel, and the conditions of the burning process the actual byproducts found in the exhaust would vary slightly. However, most commonly the other byproducts are: NOx, Carbon Monoxide, Small Particles, Hydrocarbon and Oxynitride, which are all in varying degrees responsible for a reduction in air quality, including smog, acid rain and consequently influencing health [7]. NO^x in combination with volatile organic compounds leads to the creation of ozone, which when close to the ground can lead to health problems especially for people with asthma and can even hurt plants and crops. NO^x is a byproduct from any burning of coal, gasoline or any other fossil fuel, and would therefore be generated in any combustion engine or power station running on fossil fuel. Carbon Monoxide is mainly produced due to inefficient burning of fuel within combustion engines, it is therefore highly essential that combustion engines are well maintained and tuned properly. Small Particles have made it to the press quite a few times in relation to diesel engines producing a substantial amount of them, which has consequently led to exhaust filters becoming a requirement and/or catalysts being installed, the latter of which not only eliminating the small particles from the exhaust [34]. Small particles are also formed when burning fuel in power stations. Hydrocarbons, Oxynitride among other pollutants are all side effects of burning fossil fuel, quite often in none ideal conditions.

Considering that an Electrical Vehicle requires the production of electricity, so in terms of air pollution, the pollution would depend on how the electricity is generated. If the electricity is generated from fossil fuels, then the difference is that in case of a power station, the pollution is going to be created local to the power station rather than where the actual vehicle is used.

Figure 5. Estimated CO² emission factors for public electricity production in EU-27, 2008 [32].

According to [35] when the electricity is generated from fossil fuel, then electrical vehicles do not benefit over internal combustion vehicles, which is fairly understandable since it is in essence the same fuel being burned and taking into account the overall efficiency from source to end, the electrical vehicle is worse off than the one with an internal combustion engine. This leads to the same conclusion as with the CO² calculation, which is that electrical vehicles will only have a smaller environmental impact if the electricity is generated from a renewable/green source.

4. ALTERNATIVES

The previous section indicated that electrical vehicles are not really better than our current fossil fuelled vehicles, and one of the main reasons for that is the actual electricity generation. The results presented clearly indicated the influence of fuels being used therefore one of the questions is whether there are any other fuels with better efficiency and lower environmental impact. The other option would clearly be to reduce actual fuel consumption of the vehicle itself and/or reduce the fuel consumed per person for transport, which could be achieved by moving towards smaller, lighter vehicles overall and/or more mass transport.

4.1. Hybrid Vehicles

One of the reasons that has prevented electrical vehicles from breaking through into the mass market is their limited range on a full charge. Although the majority of distances travelled are on average between 15 and 45 km [36, 37], the limited ranges offered by electrical vehicles, combined with the limited charging locations has led to a limited market penetration for electrical vehicles. On the other hand, adding in electrical motors and batteries to a car with a fossil fuelled engine, generally referred to as hybrid vehicles, is starting to know a fairly good uptake but that did take quite a number of years.

Hybrid vehicles can be classified into two types. The first type is the serial type, where the electrical motor is driving the vehicle and a combustion engine combined with a generator provides the electricity. The energy source within this serial type does not necessarily have to be a combustion engine, but could also be powered by hydrogen, which will be discussed in more detail further on. The second type of hybrid vehicle is the parallel type, and here the combustion engine as well as the electrical motor(s) can be used to drive the vehicle. An example of the latter type is the Toyota Prius.

Needless to say that both types have their respective advantages and disadvantages. The benefits of the serial type lies in the fact that the combustion engine can be designed and run at a certain optimal rpm, which stands in contrast to using the combustion engine to drive the vehicle itself. In the latter case, the engine needs to have good torque over a wide range of rpm, which is not always easily achieved and inherently implies a disadvantage when it comes to engine efficiency. An additional benefit is that due to the continuous energy source being available, the battery bank can be kept small, whereas the range of the vehicle itself is not being limited by the storage capacity of the battery bank.

A parallel hybrid benefits from being able to combine the advantages of both a motor and an engine. The advantage of an electrical motor is that it has high torque as soon as it is switched on and therefore allows for fast acceleration. On the other hand, an engine will have higher torque at higher rpms and will therefore be better for acceleration at higher speeds. The range of a parallel hybrid will not be too different from a fossil fuelled vehicle. Depending on how much electrical energy one wants to use, a smaller or larger battery bank will need to be installed.

When comparing the consumption of normal vehicles with that of hybrid vehicles, then the latter are said to run with higher fuel efficiency, generally about 20 % [38, 39]. This has multiple reasons: firstly, they recover some of the brake energy and secondly, the combustion engine in e.g. the Toyota Prius makes use of the Atkinson cycle, which is known to have an improved thermal efficiency. Another motor that has an improved thermal efficiency is the Reinhardt Turbine since it applies the principle of the combined cycle applied in some gas powered stations to the combustion engine [40]. This implies that the heat produced during the combustion cycle is used to drive another piston, which consequently increases the overall efficiency of the engine to over 50%, which in the future can possibly be improved up to 75%. The construction of the engine itself is easier than the current engines since there is no requirement for a crankshaft, while the principle can be applied to any type of combustion engine and therefore increases the efficiency of a combustion engine running on any type of fuel.

An engine used in a hybrid vehicle generally has slightly different requirements, as for example it may be running only to drive a generator. Hence, the requirements for the engine are slightly different, and the engine no longer needs to provide high output values in a range of rpms. Therefore, it is likely that new engine designs, such as the shock wave engine [41, 42] which is about 3.5 times more efficient than the normal combustion engine, produces 95% less carbon dioxide and can run on any type of fuel: gas, petrol, hydrogen will find its way into hybrid vehicles. On top of that, the shock wave engine has a smaller weight, is less expensive to produce and has lower maintenance requirements.

4.2. Bio-Fuels

Together with the drive towards greener energy production and transport, there has been a drive towards using more bio-fuels. The definition of bio-fuels has however also been diversified. In the beginning, the term bio-fuel would be used if waste was turned into a fuel. For example, in the early days when wood-chip waste or saw dust was used for running a boiler system, it would be considered a bio-fuel since it made use of a waste product. Now the demand for wood-chips and sawdust has overtaken the available waste and therefore there are specialised factories producing sawdust and wood chips to fulfill the demand. This raises questions as to whether fuel produced in this way can still be considered to be a bio-fuel, since the fuel is no longer a by-product of another process that is being recycled, but a product that is produced especially and actually requires quite some energy to be produced. More generally, the current definition of bio-fuel would be based on the principle that a fuel would be considered to be a bio-fuel if it compensates the CO² produced when it is burned during the actual growing and production of the fuel itself. This definition implies a zero carbon impact when performing a full life-cycle analysis, or in other words introduces the natural recycling principle to fuels and energy.

Bio-fuels have caused another controversy, which lies in the fact that their production can take up quite a number of fields normally used for food crops. Considering the rise in population on one hand and therefore food demands, it seems rather contradictory that legislation requires for an increasing percentage of bio-fuel to be produced. This brings us to a slightly different and probably less harmful definition for a bio-fuel, namely: a bio-fuel is a fuel that is either a by-product of, or a product that can easily be produced in nature around us without affecting current crop production and does not provide any harmful output when being burned.

Bio-fuels would currently be classified into first and second-generation fuels. The first generation fuels would be based on sugar, starch or vegetable oil, and hence comprise fuels like: bio-diesel, bio-alcohol, bio-gas among others. The second-generation bio-fuels, on the other hand, would be made from sustainable feedstock, and comprise fuels like: wood diesel, bio-methanol, bio-hydrogen, algae-fuel and the like.

Bio-diesel can be produced from a variety of different sources and is the lowest emission diesel available. Depending on the source it is created from, the viscosity can however be slightly different and can therefore only be used in the older type diesel engines without being diluted. The latest generation diesel engines with direct injection require a fuel with a certain viscosity and therefore bio-diesel has to be mixed with fossil fuel based diesel to ensure a consistent viscosity so that the engine"s injection system does not get damaged. One benefit of bio-diesel is that it cleans the engine from any deposits that settled there from burning normal diesel. Hence, especially when an engine is changed from normal diesel to bio-diesel the filters will need to be changed more regularly since they need to deal with all this deposit.

Bio-ethanol/propanol/buthanol would generally be produced from the fermentation of sugars or starches. The resulting fuel can be used in any normal gas-oil combustion engine generally mixed with normal gas-oil except for buthanol, which can be burned on its own. When compared to normal petrol, this fuel holds a lower energetic value and therefore more fuel will be required for the same energy output.

Wood gas was particularly popular as a fuel during the Second World War [43]. When wood is brought to a high temperature, it releases a gas that once purified from water vapour and other impurities is highly flammable. This gas can be burned in a normal gas-oil type engine with slight adjustments to the air-fuel ratio. The by-product of extracting the gas from the wood is that the wood turns into charcoal, which can then again be used to heat other wood. Considering that wood contains a burnable gas, it will be no surprise that one can also extract oil from it to burn in a petrol engine.

Bio methanol is a rather controversial fuel. Although it can be produced cheaper than ethanol, it has a much lower energy density. On top of that, it is toxic and therefore needs to be dealt with, with care, not to harm the user. Algae fuels are generally considered to overcome the food production problems associated with other bio-fuels. The added advantage of algae is that they produce fuels very quickly, and can be grown in an industrial environment to fulfill the current fuel requirements. A few fuels that can be produced by algae are ethanol/methanol and glycol. The latter of which is a bio-fuel with extremely low emissions and therefore a perfect solution for many situations other from road transport [44].

Although a variety of bio-fuels can be produced, quite a few of them have lower energetic values and therefore more fuel will be required for the same output. The main benefit however lies in the fact that most will have a reduced CO² output in comparison to fossil fuels.

4.3. Hydrogen

Hydrogen can be used as a fuel in two different ways: one can either combust it, which results in a slightly higher efficiency than burning gas-oil [45], or it can be used to produce electricity by providing it to a fuel cell, which acts as a catalyst. In this case, the catalyst converts hydrogen and air into water and electricity. The latter method is currently more commonly used in transportation. In essence this also means that hydrogen is actually only used as an energy carrier within the vehicle and therefore replaces the batteries. Generally, hydrogen will be stored under high pressure to ensure a proper amount of energy being carried on board the vehicle, but this requires strong storage tanks which obviously also have a certain weight associated with them. This weight will depend on the size of the tank, materials being used for its construction and so on, but overall, the energy density of hydrogen is smaller than that of fossil fuels, therefore to reach a similar range more storage will be required. With hydrogen being used as a store for electrical energy, one also needs to compare the storage efficiency in comparison to batteries. This efficiency would be somewhere between 60 and 85 % depending on the fuel cell type used (Table 2). Considering that batteries are at the latter level already, there seems to be little advantage to using hydrogen, especially considering the fact that, because it is highly flammable, there are extra problems for use in a transport context.

4.4. Compressed Air

Compressed air vehicles are not a new invention. They were used in the early 20th century in different locations around the planet. One of the main problems with compressed air vehicles is the actual compression/decompression cycles that are required. During compression there is generally heat produced and during decompression cold is produced, both reduce the efficiency of the overall process.

Fuel Cell Type	Electrolyte	Operating Temperature	Efficiency	Typical Stack Size
Alkaline Fuel Cell (AFC)	Potassium hydroxide solution	90-100°C	60%	10-100 kW
Polymer Electrolyte Membrane Fuel Cell (PEMFC)	Perfluoro Sulfonic Acid	50-100°C, typically 80°C	40-60%	<1 kW-100 kW
Phosphoric Acid Fuel Cell (PAFC)	Phosphoric Acid	150-200°C	40%	400 kW
Molten Carbonate Fuel Cell (MCFC)	Molten mixture of alkali metal carbonates	600-700°C	45-50%	300 kW-3 MW
Solid Oxide Fuel Cell (SOFC)	Oxide ion conducting ceramic	700-1000°C	60%	1 kW-2 MW

Table 2. Fuel cell technologies [46]

In order to increase this efficiency, several techniques are available. Firstly, there is the option of regenerating the heat produced during compression for decompression [47], alternatively one can compress air stored in a tank by adding/removing a fluid to the tank and hence use a hydraulic system for compression and decompression.

Before moving to purely hydraulic systems, there are still a number of advantages about compressed air storage that need to be mentioned. When compared to an electrical vehicle with batteries, the life expectancy of a compressed air vehicle is longer, there are no nasty chemicals being used and the storage capacity does not degrade over time. The actual construction is also easier, and although most tank systems would be based on metal, these come with a certain weight and therefore reduce the fuel efficiency in relation to weight. However, new materials are likely to enable other tank constructions, and such lighter materials could then improve on the fuel efficiency.

4.5. Hydraulic Vehicles

As mentioned previously, one of the problems with current fossil fuelled vehicles is that their engines have to be designed to deal with a large range of requirements and therefore need to have high torque across a large rpm-range. Having to over-design the system leads to inefficiencies when it is used for the majority of the time. Therefore, it would be better when the vehicle has an energy source that can deal with the average demand and delivers the peak demands in a different way. The hydrid [48] is a system build around this model. It makes use of a hydraulic transmission system, and even though this is less efficient than a mechanical transmission, overall, the system still has a higher efficiency. The hydrid also overcomes the complexity and extra requirements of hybrid vehicles, since these generally require a double transmission system to enable the electrical and combustion engine to power the vehicle. On the other hand, the hydrid uses the engine as an energy source to fill up the hydraulic accumulator, similar to the serial type hybrid. However, once the storage system is full, the engine would switch off and the car would drive completely based on the energy from the accumulator. This accumulator is in essence a set of tanks where liquid is used to compress a gas and therefore indirectly uses compressed air as storage. An added advantage of this hydraulic system is that this system can also be used in reverse mode to perform regenerative braking, which to some extend compensates the higher losses of the not mechanical transmission system. Overall, the hydrid system results in a 50% fuel efficiency when compared to a fossil fuelled vehicle, which consequently also halves the grammes of CO2/100km. The hydrid system can also be combined with an electric motor to gain the advantages of both systems, which are: the hydraulic system, which has a good power density, but low energy density, whereas the electrical system has a poor power density, but good energy density [49].

4.6. Public Transport

Considering that one of the influencing factors on fuel consumption for a vehicle is the size of the front of the vehicle and therefore the required air displacement, it will be no surprise that public transport can reach much higher efficiencies per person since there are more people in the same vehicle. From this aspect it will also be no surprise that trains are one of the most efficient ways of transportation. However public transport can only be efficient by having a limited number of stops, which also means it cannot provide the same comforts of a personal vehicle.

CONCLUSION

The current trend seems to propose that electrical vehicles form the solution towards the pollution problem of current fossil fuelled vehicles. However, the environmental friendliness of electrical vehicles largely depends on how the electricity itself is generated. As long as electricity is mainly generated from fossil fuels, like in most Western countries, electrical vehicles do not provide a benefit in this aspect. However, they do provide benefits when it comes to the integration of ICT and automation of the vehicle, as well as allowing for an increase in transport efficiency due to energy recovery during deceleration. It was shown that fully regenerative breaking is essentially not hard to achieve and even brings benefits for the actual acceleration of the vehicle.

Current hybrid vehicles already benefit from combining the two technologies. However there are a number of improvements still possible to this technology. The first one has already been made available, by creating the plug-in Hybrid, which allows for, preferably green electricity to be used to charge the vehicle. A second improvement lies in the fuel being burned in the engine in relation to extending the vehicle"s range. The required improvement here is related to the CO² production of the fuel being burned, which can be solved by using a different fuel. Replacing the engine with a Fuel cell would provide a better link to the third problem, which is that of electrical energy storage. Currently, batteries are used for this storage, whereas other principles, like compressed air, hydraulics and hydrogen would provide for higher efficiencies and storage densities. In any case the energy storage provided within EVs would come in handy when such vehicles are linked to the grid as a buffer for storing locally generated renewable energy.

REFERENCES

• [1] Binesti, D.; Core Losses and Efficiency of Electrical Motors Using New Magnetic Materials. IEEE Transactions on Magnetics, vol. 32, no. 5, pp. 4887-4889, 1996.
• [2] Mackay, D.J.C.; Sustainable Energy - Without the Hot Air. UIT Cambridge, 2009, www.withouthotair.com/download.html.
• [3] Bandivadekar, A.; Bodek, K.; Cheah, L.; Evans, C.; Groode, T.; Heywood, J.; Kasseris, E.; Kromer, M.; Weiss, M.; On the Road in 2035: Reducing Transportation"s Petroleum Consumption and GHG Emissions. MIT Report, 2008. http://mitei.mit.edu/system/files/ On%20the%20Road%20in%202035_MIT_July%202008.pdf.

• [4] Melis, W.J.C.; Chishty, O.; Fully Regenerative Braking and Improved Acceleration for Electrical Vehicles. International Journal of Sustainable Energy Development (IJSED), vol. 2, no. 1 and 2, pp. 75-80, 2013.

• [5] Situ, L.; Electric Vehicle Development: The Past, Present & Future. 3rd International Conference on Power Electronics Systems and Applications, pp. 1-3, 2009.

• [6] Chan, C.C.; The Past, Present and Future of Electric Vehicle Development. IEEE International Conference on Power Electronics and Drive Systems (PEDS "99), Hong Kong, pp. 11-13, 1999.

• [7] Toyota; Green Innovation. 2012. www.toyota.com/about/environmentreport2012/ green_innovation.html.

• [8] www.hybridcenter.org/hybrid-timeline.html.

• [9] Shen, Z.J.; Zhang, Y.B.; Yao, Y.Q.; Liu, B.W.; Electric Vehicle: The Hope of the Green Future. International Conference on Management and Service Science (MASS 2011), pp. 1-4, 2011.

• [10] Zahedi, A.; Electric Vehicle as Distributed Energy Storage Resource for Future Smart Grid. 22nd Australasian Universities Power Engineering Conference, pp. 1-4, 2012.

• [11] Pillai, J.R.; Bak-Jensen, B.; Thogersen, P.; Electric Vehicles to Support Large Wind Power Penetration in Future Danish Power Systems. IEEE Vehicle Power and Propulsion Conference, Seoul, Korea, pp. 1475-1479, 2012.

• [12] Mierau, M.;Kohrs R.;Wittwer, C.; A Distributed Approach to the Integration of Electric Vehicles into Future Smart Grids. 3rd IEEE PES International Conference and Exhibition on Innovative Smart Grid Technologies, pp. 1-7, 2012.

• [13] Shafiei, A.; Williamson, S.S.; Plug-in Hybrid Electric Vehicle Charging: Current Issues and Future Challenges. IEEE Vehicle Power and Propulsion Conference (VPPC),pp. 1- 8, 2010.

• [14] Buckl, C.; Camek, A.; Kainz, G.; Simon, C.; Mercep, L.; Stahle, H.; Knoll, A.; The Software Car: Building ICT Architectures for Future Electric Vehicles. IEEE International Electrical Vehicle Conference (IEVC), pp. 1-8, 2012.

• [15] Silberg, G.; Wallace, R.; Self-Driving Cars: The Next Revolution. KPMG LLP &Center for Automotive. Research Report, 2012. www.kpmg.com/US/en/IssuesAnd Insights/ArticlesPublications/Documents/self-driving-cars-next-revolution.pdf.

• [16] Coelingh, E.; Solyom, S.; All Aboard the Robotic Road Train. IEEE Spectrum, 2012, http://spectrum.ieee.org/green-tech/advanced-cars/all-aboard-the-robotic-road-train.

• [17] [17] Wolfram, S., A New Kind of Science. Wolfram Media Inc., 2002.

• [18] Melis, W.J.C.; Kameyama, M.; A Study of the Different Uses of Colour Channels for Traffic Sign Recognition on Hierarchical Temporal Memory. Fourth International Conference on Innovative Computing, Information and Control, pp. 111-114, 2009.

• [19] Chua, L.O.; The Fourth Element. Proceedings of the IEEE, vol. 100, no. 6, pp. 1920- 1927, 2012.

• [20] Sanpasertparnich, T.; Improved Thermal Efficiency of Coal-Fired Power Station: Monte Carlo Simulation. IEEE EIC Climate Change Technology, Ottawa, USA, pp. 1- 9, 2006.

• [21] http://en.wikipedia.org/wiki/Fossil-fuel_power_station.

• [22] Ferreira S.; Eco-sheet: 1 kWh Generated Using Different Energy Sources. Leonardo-Energy, 2008, www.leonardo-energy.org/eco-sheet-1-kwh-generated-using-differentenergy-sources.

• [23] http://en.wikipedia.org/wiki/Combined_cycle.

• [24] Siddique, A.H.; Lamont, L.A.; Al-Bloushi, H.K.; Nuclear Power for the Vision 2030 of Abu Dhabi and Improvement of Air Quality. World Congress on Sustainable Technologies, London, UK, pp. 14-18, 2012.

• [25] Yamaguchi, M.; Nuclear Power to be Phased out Following Fukushima Disaster. The Independent, September 15, 2012, www.independent.co.uk/news/world/asia/nuclearpower-to-be-phased-out-following-fukushima-disaster-8139966.html.

• [26] Evans, S.; Germany: Nuclear Power Plants to Close by 2022. BBC News, May 30, 2011, www.bbc.co.uk/news/world-europe-13592208.

• [27] De Vos, A.; Detailed Balance Limit of the Efficiency of Tandem Solar Cells. Journal of Physics D: Applied Physics, vol. 13, no. 5, pp. 839-846, 1980.

• [28] Shepherd, W.; Zhang, L.; Electricity Generation using Wind Power. World Scientific Publishing, 2011.

• [29] Aquirre, K.; Eisenhardt, L.; Lim, C.; Nelson, B.; Norring, A.; Slowik, P.; Tu, N.; Life Cycle Analysis Comparison of a Battery Electric Vehicle and a Conventional Gasoline Vehicle. 2012, www.environment.ucla.edu/media_IOE/files/BatteryElectricVehicle LCA2012-rh-ptd.pdf.

• [30] Hawkins, T.R.; Singh, B.; Majeau Bettez, G.; Stromman, A.H.; Comparative Environmental Life Cycle Assessment of Conventional and Electric Vehicles. Journal of Industrial Ecology, vol. 17, no. 1, pp. 53-64, 2012.

• [31] U.S. Energy Information Administration, Few Transportation Fuels surpass the Energy Densities of Gasoline and Diesel.2013, www.eia.gov/todayinenergy/detail.cfm ?id=9991.

• [32] www.eea.europa.eu/data-and-maps/indicators/ener002-co2-intensity-of-heat /assessment.

• [33] www.carpages.co.uk/co2/.

• [34] Deeba, M.; Dettling, J.C.;Lui, Y.K.; Four-Way Diesel Exhaust Catalyst and Method of Use. US6093378, 2000, http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2= HITOFF&d=PALL&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.htm&r=1&f=G&l=5 0&s1=6,093,378.PN.&OS=PN/6,093,378&RS=PN/6,093,378.

• [35] Ji, S.; Cherry, C.R.; Bechle, M.J.; Wu, Y.; Marshall, J.D.; Electric Vehicles in China: Emissions and Health Impacts. Environmental Science and Technology, vol. 46, no. 2, pp. 2018-2024, 2012.

• [36] FIA Foundation for the Automobile and Society; The Automobile and Society, Executive Summary. 2003, www.fiafoundation.org/publications/Documents/ auto_ society_exec_summary.pdf.

• [37] www.solarjourneyusa.com/EVdistanceAnalysis7.php.

• [38] Imai, S.; Takeda, N.; Horii, Y.; Total Efficiency of a Hybrid Electric Vehicle. Power Conversion Conference, Nagaoka, Japan, vol. 2, pp. 947-950, 1997.

• [39] Kaneko, T.; Shimada, M.; Kujiraoka, S.; Kojima, T.; Easy Maintenance and Environmentally-Friendly Train Traction System. Hitachi Hyoron, vol. 85, no. 8, pp. 549-552, 2003.

• [40] Reinhardt, G.T.; Thermal Engine such as e.g. an Internal Combustion Engine and/or Steam Engine. WO2011/042301, 2011, http://patentscope.wipo.int/search/en/ WO2011042301.

• [41] Knight, H.; Shock wave puts hybrid engines in a spin. New Scientist, vol. 209, no. 2803, p. 25, 2011.

• [42] Revkin, A.C.; The "Wave" of the (Car Engine) Future?. The New York Times, 2011, http://dotearth.blogs.nytimes.com/2011/04/11/the-wave-of-the-car-engine-future/? partner=rss&emc=rss.

• [43] Food and Agriculture Organization of the United Nations; Wood Gas as Engine Fuel. Food and Agriculture Organization of the United Nations, 1987, www.fao.org/docrep/ T0512E/T0512E00.htm.

• [44] Day, P.; McNeil, J.; Sirovski, F.; (E)Mission Impossible ?.The Chemical Engineer, vol. 839, pp. 33-35, 2011.

• [45] Berger, E.; BMW Hydrofen Near Zero Emission Vehicle Development. BMW Clean Energy CARB ZEV Technology Symposium, pp. 1-28, 2006.

• [46] US Department of Energy; Energy Efficiency & Renewable Energy: Fuel Cell Technologies Program. 2011, www1.eere.energy.gov/hydrogenandfuelcells/fuelcells/ pdfs/fc_comparison_chart.pdf.

• [47] http://lightsailenergy.com/tech.html.

• [48] Achten, P.; Vael, G.; Murrenhoff, H.; Low-Emission Hydraulic Hybrid for Passenger Cars.ATZ, 111, pp. 378-384, 2009.

• [49] Vael, G.; Eggenkamp, S.; Achten, P.; The E-Hydrid. The Twelfth Scandinavian International Conference on Fluid Power, Tampere, Finland, pp. 1-15, 2011.

INDEX

airports, 188 algae, 524

#

	Algeria, 483
20th century, 208, 525	algorithm, xiii, 15, 16, 29, 35, 81, 83, 84, 99, 100,
21st century, 411	102, 103, 110, 113, 121, 122, 124, 126, 127, 128,
3D images, 58	129, 130, 131, 132, 134, 135, 138, 157, 161, 163,
	202, 204, 215, 270, 279, 280, 281, 282, 283, 284,
A	287, 300
	alternative hypothesis, 35
abuse, 455	aluminium, 433, 443, 446, 450
AC motors, 260, 268, 380, 385	amplitude, 10, 11, 13, 14, 15, 32, 36, 38, 57, 58, 63,
accelerator, 326	70, 76, 239, 251, 279, 384, 385, 389, 390, 391,
access, xiv, 149, 151, 152, 161, 168, 170, 171, 172,	393, 394, 395, 403, 406
173, 183, 188, 192, 194, 195, 196, 197, 199, 202,	anatomy, 60
203, 208, 210, 214, 217, 219, 220, 237, 238, 240,	applied mathematics, 204
244, 504	aqueous alkali solution, 454
accessibility, 216, 219	arithmetic, 12, 22
acid, 322, 409, 411, 412, 414, 415, 426, 434, 454,	artificial intelligence, 267, 380
521	Asia, 188, 229
actuation, 410	assessment, 82, 149, 176, 199, 529
actuators, 115, 215	assets, 202
ad hoc network, 215	asthma, 410, 521
adaptation, xiii, 67, 68, 148, 150, 157, 213, 225, 317	asymmetry, 68, 395
ADC, 61, 62	atmosphere, xii, 42, 49, 212, 422, 454
adjustment, 155, 283, 370, 371	attachment, 192, 496, 501, 502, 504
administrators, 188	attacker, 190, 192, 199
Admittance, 483	Attenuation Factors, 3
adverse weather, 118	audit, 198
adware, 192	authentication, 242
aerosols, 49	authenticity, 195
Africa, 227	authority(s), 175, 189, 221
age, 241	automate, 190, 193
agencies, 189, 203, 209, 243, 410	Automatic incident detection, 148, 150
aggregation, 294	automation, 59, 437, 527
Aggregator, 291	automobile(s), xi, xv, 143, 197, 237, 410, 420, 506,
agility, 491, 498	529
agriculture, 188	automotive application(s), ix, xiii, 117, 252, 316,
air quality, 210, 521	317, 320, 321, 322, 334, 435
air temperature, 44

candidates, 132, 133, 437

car accidents, ix

autonomy, xii, 82, 83, 325, 380, 409, 410, 411, 413, 417, 420, 436 avoidance, 4, 5, 6, 51, 59, 74, 190, 242 awareness, 191, 223, 225 Azerbaijan, x, 111 B background noise, 35, 36, 38, 39 bandwidth, 10, 19, 32, 34, 37, 40, 41, 52, 65, 66, 76, 156, 169, 171, 183, 237, 239, 240 banking, 186, 188, 202 barriers, 167 base, xii, 107, 148, 150, 167, 168, 170, 211, 217, 218, 241, 262, 440, 444 battery, viii, 275, 276, 289, 292, 314, 322, 323, 324, 345, 346, 347, 348, 351, 411, 413, 427, 428, 437, 438, 439, 450, 451, 453, 454, 455, 456, 482, 516, 529 beams, 116, 155 behaviors, 215 Belgium, 233 benefits, 129, 172, 182, 189, 209, 210, 212, 265, 294, 486, 487, 511, 512, 518, 522, 523, 527 bias, 14, 65 Bible, 228 bio-fuel, 514, 524 blind spot, 5, 59, 213 blogs, 530 Bluetooth, 169, 171, 192 BMA, 163 bounds, 304 brain, 178 Brazil, 145, 350 breakdown, 242, 381, 385 burn, 524 business management, 190 businesses, 185, 187, 190, 217 buyer, 431 by-products, 519, 521 C cables, 171, 182 CAD, 444 cadmium, 415, 416, 417, 434, 454, 455 calibration, 159 CAM, 241 cancer, 410 carbon, 210, 212, 322, 420, 423, 425, 426, 519, 523, 524 carbon dioxide, 523 carbon emissions, 210, 212 carbon nanotubes, 426 case study(s), xvi, 84, 186, 187, 287, 297, 305 cash, 326 catalyst, 525 cathode materials, 426, 433 cation, 103, 110, 413 C-C, xv, 457, 459, 469, 477, 478, 479, 480 CEC, 231 cell phones, 454 ceramic(s), 418, 428, 483, 525 certificate, 199 certification, 176, 202 challenges, xii, 83, 114, 115, 117, 196, 217, 219, 221, 222, 223, 294, 503, 511 chemical(s), xii, xv, 188, 194, 413, 422, 433, 434, 454, 488, 526 chemical industry, 188 chemical properties, 413 chemical reactions, xii, xv chemical stability, 434 Chicago, 505 China, 78, 506, 520, 529 circulation, 157, 328, 392, 504 cities, 120, 172, 176, 183, 209, 213, 216, 244 citizens, 220, 224 civilization, 208 clarity, 338, 481 classes, 113, 122, 125, 127, 132, 134, 135, 136, 138, 139, 142, 168, 171, 196, 197, 212, 327 classification, 58, 82, 113, 119, 120, 121, 122, 123, 124, 125, 126, 127, 129, 130, 131, 132, 134, 135, 136, 137, 138, 141, 142, 155, 157, 211, 212, 214, 318, 494 clients, 190 climate, 150, 208 climate change, 208 clustering, 121, 130, 132, 134 clusters, 133, 245 CO2, 176, 177, 317, 347, 489, 519, 520, 521, 522, 523, 525, 526, 527 coal, 49, 50, 513, 521 coal dust, 49, 50 cobalt, 419, 432, 433, 434 coding, 166, 170 cognitive process, 218 collaboration, 219

collisions, 170, 215, 238, 244

color, 117, 133, 492

combustion, xi, xv, 250, 276, 295, 318, 410, 411, 412, 414, 434, 485, 486, 505, 509, 510, 511, 512, 514, 521, 522, 523, 524, 526 commerce, 186 commercial, 73, 166, 169, 172, 208, 317, 353, 354, 376, 417, 433, 454 communication(s), xi, xiv, 6, 8, 11, 82, 149, 167, 168, 169, 170, 178, 179, 182, 183, 187, 188, 189, 193, 195, 199, 208, 212, 215, 219, 222, 235, 236, 237, 238, 239, 240, 241, 243, 244, 245, 276, 324, 326 communication systems, 6, 82, 183, 212, 236 communication technologies, 236 communications channel, 182 community, xv, 172, 209, 215, 220, 222 compatibility, 103, 168, 183, 503 compensation, 5, 66, 74, 76 competition, 168, 225, 456 competitiveness, 208 compilation, 193, 202, 371, 390 complement, 118, 119, 279 complexity, 8, 38, 52, 113, 116, 119, 135, 161, 195, 196, 198, 207, 209, 219, 220, 224, 225, 325, 481, 486, 526 compliance, 191 composition, 420, 422 compounds, 426 compression, 525, 526 computation, 8, 52, 74, 124, 129, 134, 135, 161, 188, 190, 222, 283 computer, 6, 58, 60, 62, 71, 74, 98, 132, 185, 186, 187, 188, 190, 191, 192, 193, 194, 195, 196, 197, 199, 204, 205, 208, 237, 241, 436 computer systems, 186, 192 computing, 124, 128, 129, 130, 147, 150, 180, 188, 189, 190, 197, 341, 360 condensation, 47 conditioning, xii, 183, 251, 279 conductance, 458 conduction, 255, 257, 258, 328, 480, 481 conductivity, 425, 432, 433, 447 conductor, 157, 262 confidentiality, 195 configuration, xi, 65, 66, 94, 102, 154, 160, 180, 194, 250, 251, 269, 280, 295, 296, 364, 369, 370, 371, 372, 375, 376, 424, 443, 444, 445, 446, 480, 481, 505 conformity, 165 congress, 79, 184, 188, 204, 226, 311, 529 construction, xiv, 53, 74, 95, 215, 223, 249, 261, 265, 311, 321, 360, 364, 375, 416, 417, 481, 487, 497, 498, 502, 518, 523, 525, 526 consulting, 212 consumers, xi, 298 consumption, ix, 177, 208, 213, 293, 303, 317, 345, 410, 492, 510, 515, 516, 520, 523 containers, 169 Continental, 143, 354 Continuous Air interface Long and Medium range (CALM), 237 contour, 141, 162 control measures, 176 controversial, 515, 524 convergence, xv, 31, 217, 224, 284, 432 conversations, 169, 192 conversion rate, 416 cooling, xv, 47, 191, 265, 357, 362, 365, 416, 417, 431, 437, 438, 439, 440, 442, 443, 445, 447, 450, 500, 501, 502, 503, 504 cooperation, 198, 222 coordination, 60, 178, 182, 215 copper, 265, 319, 519 correlation(s), 5, 32, 35, 38, 39, 52, 74, 115, 148, 198 covering, 117, 171, 173, 189, 235 creativity, 491 credentials, 193 criminal activity, 188 criminals, xiii, 185 crises, 410 critical infrastructure, ix, xiii, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 202, 203, 204 crop(s), 521, 524 crop production, 524 cryptography, 198, 200, 204 Cryptography, 186 crystalline, 419, 420 crystals, 455 cultural influence, 225 culture, 222, 225 cure, 242 customers, 190, 494 Cyber security, 185 cyberspace, 204 cycles, 255, 266, 283, 294, 323, 324, 415, 416, 417, 418, 419, 422, 423, 426, 434, 435, 437, 513, 525 cycling, 323, 455

D

damages, 497 damping, 88, 499 danger, 219, 238

connectivity, 179, 190, 222, 238

consensus, 132, 211 conservation, 89

data analysis, 456 data availability, 209 data collection, 165 data communication, 165, 168 data gathering, 193 data generation, 161 data processing, 62, 161, 183, 241 data rates, 237 data transfer, 155, 170 database, 132, 136, 138, 139, 142 DDoS, 192 decision makers, 217, 297 decomposition, 381 decoupling, 263, 316 defects, 392 degradation, 28, 40, 41, 47, 48, 51, 52, 53, 294, 310, 385, 433 democratization, 381 dendrites, 455 denial, xiii, 185 denial of service attack, 185 Department of Energy, 410, 530 Department of Homeland Security, 191, 203 Department of Transportation, 184, 216 dependent variable, 474 deployments, 199 deposits, 524 depression, xv, 453, 457, 458, 459, 460, 480, 481 depth, xiii, xv, 82, 83, 101, 105, 135, 186 derivatives, 85, 92, 94, 109 designers, 320, 491 destruction, 393 detection system, 155, 156, 161, 165 detection techniques, 12, 51 Detection threshold, 5 deviation, xv, 151, 281, 297, 453 DHS, 203 diesel engines, 410, 514, 521, 524 differential equations, 98, 99 diffusion, 414, 423, 481 diodes, 253, 254, 318 Direct drive, 497 direct measure, 176, 177, 264 disaster, 529 discharges, 412 discretization, 305 discrimination, 74, 382 discs, 501 diseases, 410 dispersion, 167 displacement, 526 distortions, 68, 269 distributed computing, 222 distribution, 45, 122, 130, 173, 188, 190, 198, 203, 265, 276, 277, 278, 280, 288, 294, 299, 305, 310, 318, 326, 381, 383, 393, 395, 438, 448, 453, 477, 478, 479, 480, 481, 493, 498, 515, 516 distribution function, 305, 453, 477, 478, 479, 481 diversity, 207, 220, 223, 237, 454 DME, 435 doping, 433 Doppler shift, 10, 11, 239 DOT, 216 downlink, 171 drawing, 422, 454, 510 durability, 317, 323, 432, 435, 454, 455, 486, 498 Dynamic Modeling, 81 dynamic systems, 103 E economic development, 211, 221 economic policy, 208 economic progress, 208 economics, 310 ecosystem, 210 editors, ix, xv, xvi education, 204, 224, 511 educational software, 198 elaboration, 410 electric current, 358, 422, 435 electrical conductivity, 425 electricity, xi, xiii, xiv, 185, 186, 188, 190, 191, 202, 215, 250, 251, 276, 293, 294, 295, 302, 303, 304, 305, 306, 309, 310, 311, 326, 345, 381, 488, 509, 510, 511, 513, 514, 515, 516, 519, 520, 521, 522, 525, 527 electrodes, 413, 414, 415, 422, 425, 433, 435, 455 electrolysis, 414 electrolyte, 322, 413, 414, 415, 416, 418, 419, 420, 422, 424, 425, 433, 435, 454, 455 electromagnetic, xiv, 42, 63, 117, 157, 170, 242, 249, 261, 264, 265, 270, 381, 393, 394, 490 electronic communications, 186 electronic systems, 147, 454 electron(s), 413, 414, 420, 423, 435 e-mail, 192 emergency, xiii, 144, 147, 149, 150, 151, 152, 167, 168, 170, 178, 238, 243, 244, 319, 517 emission, 66, 74, 411, 490, 521, 524 employment, 11, 25, 38, 53, 212 employment growth, 212 empowerment, 225 EMS, 211 EMU, 314, 315, 316, 317, 321, 324, 325, 326, 328, 333, 334, 345

encryption, 171, 199, 202, 242 endurance, 82, 499 end-users, 172, 173 enemies, 191 energy consumption, 172, 176, 177 energy density, xii, 275, 276, 289, 318, 321, 322, 323, 325, 411, 412, 413, 415, 417, 420, 423, 425, 434, 435, 436, 437, 454, 455, 509, 513, 519, 524, 525, 526 energy efficiency, ix, xiv, 176, 177, 215, 313, 315, 316, 317, 318, 320, 321, 326, 328, 333, 334, 336, 337, 338, 340, 341, 342, 343, 344, 494, 497, 514 energy prices, 303, 305, 308, 309, 310 energy recovery, 413, 527 enforcement, 148 engineering, xii, xiii, xv, xvi, 6, 52, 95, 185, 191, 195, 196, 198, 457, 488 England, 53, 231 Environmental, 84, 226, 345, 347, 519, 529 environmental aspects, ix, 172, 208 environmental care, xv, 410 environmental conditions, 59, 83 environmental effects, 454 environmental impact, xv, 176, 317, 509, 510, 519, 522 environmental issues, 486 environmental protection, 504 environments, xiv, 4, 58, 59, 82, 83, 109, 115, 119, 120, 185, 189 EPS, 211 equality, 34 equilibrium, 83, 84, 89 equipment, ix, xi, 61, 148, 149, 150, 157, 161, 169, 170, 171, 176, 178, 179, 180, 183, 185, 190, 193, 212, 217, 276, 326, 434, 512 equity, 210, 211 ESD, 314, 315 ethanol, 524 ethylene, 435 EU, 173, 494, 520, 521 Europe, 110, 147, 169, 176, 184, 211, 212, 221, 228, 231, 240, 273, 347, 377, 416, 417, 520 European Commission, 227, 237 European Community, 227 European Parliament, 227 European Union, 221, 231 everyday life, 210 evidence, 148, 150, 178, 218 evolution, 114, 115, 119, 190, 192, 208, 210, 220, 360, 380, 392, 426, 455, 481 excitation, 265, 355, 356, 370, 374 execution, 195 exercise, 519 expertise, 209, 217, 381 exploitation, 199, 380 exporters, 208 external relations, 218 extraction, 69, 161, 165, 435, 458, 513 extracts, 115 extrusion, 432, 448, 450 F fabrication, 69, 360, 375 factories, xiii, 185, 523 false alarms, 26 false negative, 131 false positive, 131, 512 fault detection, xiii, 57, 58, 381, 397, 404 fault diagnosis, 168 feature detectors, 142 Federal Bureau of Investigation, 187 Federal Highway Administration, 184 feedstock, 524 fermentation, 524 FFT, 4, 8, 20, 21, 396 FGDC, 223 fiber(s), 82, 426, 456 fidelity, 81 field tests, 95 filters, 32, 122, 134, 255, 259, 490, 521, 524 financial, 188, 219, 298, 511 financial support, 511 Finland, 407, 530 firewalls, 192 first generation, 417, 486, 524 fitness, 223 flexibility, xii, 65, 160, 170, 223, 316, 323, 454 flight, 83, 116 flow field, 95 fluctuations, 18, 19, 298 fluid, 63, 84, 86, 88, 93, 94, 95, 96, 431, 436, 526 fog attenuation, 48, 49 food production, 188, 524 force, 89, 90, 91, 92, 93, 94, 95, 96, 102, 105, 108, 147, 148, 150, 185, 202, 220, 238, 243, 265, 295, 321, 324, 379, 382, 393, 490, 493, 501, 502, 517, 518 Ford, 318, 354, 410, 416, 417, 419, 494 forecasting, 303, 309, 311 formula, 89, 98, 99, 159, 201 FOV, 114, 117 framing, 162, 163 France, xvi, 379, 406, 407, 408, 505 fraud, 186, 202

freedom, 57, 60, 76, 85

freezing, 322 friction, 230, 393, 490, 501, 502, 504 fuel cell, ix, xii, xiii, xiv, xv, 249, 250, 251, 255, 318, 409, 417, 421, 488, 489, 525 fuel consumption, 149, 176, 177, 189, 212, 215, 412, 511, 517, 519, 522, 526 fuel efficiency, 513, 514, 523, 526 fusion, 8, 74, 75, 76, 77, 193, 488 fuzzy sets, 281, 282 G gas diffusion, 422 Gaussian membership, 281 gel, 322 General Accounting Office, 204 General Motors, 410, 417 Geographic Information System, GIS, 210, 218, 219, 220, 222, 227, 228, 232 geometry, xv, 95, 96, 322, 385, 388, 395, 432, 436, 444, 446, 448 Germany, 53, 54, 187, 202, 203, 216, 231, 311, 505, 529 global economy, 208 global warming, 294 globalization, 212 glue, 502 glycol, 524 goods and services, 210, 221 governance, ix, xiii, 207, 208, 209, 217, 220, 221, 225 governments, 185, 188, 191, 208, 225 GPS, 116, 165, 180, 240, 243, 244 graduate students, xvi graph, 136 graphite, 63, 65, 420, 425, 426, 432 gravitational force, 91, 93 gravity, 89, 91, 92, 93, 486, 498 Great Britain, 226, 227 Greece, 406 greenhouse, 410 grids, ix, xiii, 72, 130, 186, 187, 190, 191, 203, 313, 315, 326, 345 grouping, 130 growth, 6, 208, 210, 220 guidance, 84, 110, 152, 166, 177, 189, 212, 213, 214, 219 guidelines, 221 H habitat, 210 hacking, 198 halogen, 182 handwriting, 124 harbors, 82 harvesting, 315, 327 hazards, 216, 435 headache, 437 health, xv, 203, 210, 410, 521 health problems, 521 heart disease, 410 heat loss, 442 heat transfer, xv, 431, 432, 445, 500 height, 133, 155 highway system, 170 highways, 147, 149, 150, 151, 165, 183, 190 histogram, 129, 130 historical data, 176 history, 115, 191, 219, 223, 510 homes, 190 Hong Kong, 528 horses, 192 host, 187, 199, 200, 245, 435 hotel, 492 hotspots, 171, 243 housing, 68, 443 hub, 270, 315, 320, 497, 499, 500, 504 human, xiv, 82, 166, 167, 186, 194, 195, 207, 208, 209, 210, 212, 218, 219, 224, 241, 512 human behavior, xiv, 207, 224 human resources, 209, 218 humidity, 42, 116, 153 Hunter, 229 hunting, 67 hybridization, 323, 341 hydrogen, 215, 315, 317, 318, 417, 455, 511, 522, 523, 524, 525, 527 hydrogen gas, 455 hydroxide, 415, 416, 417, 422, 455, 525 hydroxyl, 423 hypermedia, 237 hypothesis, 16, 35, 36, 127, 133 I ICC, 235, 245 ICE, 250, 295, 296, 314, 318, 321, 326, 327, 410, 486, 488, 489, 490, 494 ICRA, 144, 145 Image generation, 73 Imaging, vii, 57, 58, 79 imbalances, 345 immobilization, 454

immunity, 170

improvements, 215, 220, 249, 261, 410, 435, 486, 516, 527 impulses, 71 impurities, 524 income, 326 independence, 82 individualization, 212 individuals, 185, 210, 217, 225 induction, xiv, 155, 156, 249, 251, 262, 263, 264, 333, 353, 354, 360, 361, 362, 363, 364, 369, 371, 372, 374, 375, 376, 377, 379, 381, 385, 389, 390, 392, 393, 395, 397, 400, 401, 402, 403, 404, 408 Induction machine, 261, 270, 354, 369 inductor, 34, 255, 256, 257, 258, 285, 360 industry(s), xii, xiii, xv, 6, 51, 57, 58, 82, 109, 165, 166, 185, 187, 202, 203, 217, 219, 220, 222, 224, 381, 409, 414, 418, 420, 421, 425, 426, 427, 490, 503 inertia, 88, 90, 91, 93, 96, 108, 365, 393, 395, 498 infection, 186, 194 Information and Communication Technologies, 148 information exchange, 171 information sharing, 165 information technology, xiv, 190, 220, 435 infrastructure, xi, xiv, 95, 148, 165, 166, 169, 173, 177, 178, 179, 180, 182, 187, 188, 190, 191, 202, 203, 204, 205, 209, 210, 212, 215, 216, 218, 222, 224, 225, 228, 235, 241, 489, 511 initial state, 287 injuries, 6 input signal, 13, 14, 375 inspections, 169 insulation, 262, 381, 385, 454, 456 integration, ix, xi, 8, 98, 110, 118, 133, 216, 217, 224, 225, 294, 316, 317, 326, 328, 503, 509, 510, 512, 527 integrity, 195, 242 intelligence, 82, 186, 188, 216, 222, 244, 512 intelligence gathering, 244 intelligent systems, xvi, 6, 212, 512 interface, xi, 8, 151, 167, 169, 171, 172, 173, 180, 193, 197, 198, 202, 237, 251, 252, 315 interference, xiv, 16, 28, 37, 40, 53, 238, 239, 249, 261, 270 interfering target, 20 international standards, 237 interoperability, 173, 221, 222, 223, 225, 241, 245, 320 intervention, 149, 152, 186 investment(s), 219, 221, 436, 511, 515 ions, 413, 414, 418, 419, 420, 425, 432, 435 Iran, xvi, 275, 293, 295, 303 iron, 262, 426, 433, 434 isolation, 168, 315, 326, 327 issues, x, xiii, 8, 131, 171, 185, 186, 190, 204, 208, 219, 241, 242, 269, 317, 344, 410, 417, 420, 432, 451, 454, 456, 502 Italy, 230, 407 iteration, 31, 98, 99, 106, 467 J Japan, 142, 169, 211, 228, 420, 433, 497, 498, 529 joints, 486 K kill, 192 KOH, 415, 482 Korea, 507, 528 Kosovo, 191 L lakes, 47 lamination, 262 languages, 71 laptop, 196 lasers, 132 latency, 236, 238 law enforcement, xiii, 183 laws, 89, 150, 203, 221 lead-acid batteries, 434 leakage, xii, 380, 383, 384, 392, 454 leaks, 195 learning, 193, 198, 208, 281, 283, 284 Least Developed Countries, 224 LED, 117, 176, 177 legislation, 524 legs, 267 life cycle, 321, 322, 519 life expectancy, 431, 526 lifetime, 323, 435, 511 light, 6, 10, 45, 47, 48, 59, 117, 160, 317, 368, 438 linear function, 30 linear model, 83 linear systems, 32, 34 liquids, 490 lithium, xv, 322, 323, 409, 419, 420, 423, 425, 426, 427, 428, 431, 432, 433, 434, 435, 437, 448 Lithium batteries, 431 living environment, 223 localization, xiii, 57, 69, 129, 149 loci, xv, 453, 457, 458, 465, 478

logistics, 215, 216

Microsoft, 192, 193, 194

longevity, 219, 231 low temperatures, 421 LPG, 486 luggage, 490, 491 luminosity, 164 Luo, 246, 271, 506 M machine learning, 129 magnet(s), 249, 252, 261, 265, 270, 333, 354, 355, 364, 365, 375, 400, 402, 487, 502, 500, 502, 503, 504, 505 magnetic field, x, xiv, 364, 379, 381, 382, 383, 384, 385, 390, 392, 394, 402, 403, 404 magnitude, 96, 259, 278, 287, 288, 345, 382, 391, 395, 404, 461 majority, 137, 138, 392, 488, 501, 505, 517, 522, 526 Malaysia, 376 malware, 186, 187, 191, 192, 193, 194, 202 man, 214, 242 MANETs, 242 manganese, 432, 434 manipulation, 88, 209, 502 manufacturing, 393, 456 mapping, 84, 123, 124, 125 market penetration, 241, 345, 509, 511, 522 market segment, 426, 492, 494 marketing, xi, 417 masking, 22, 23, 24, 52 mass, xv, 47, 63, 85, 88, 89, 91, 96, 108, 109, 116, 118, 119, 212, 216, 316, 354, 411, 425, 455, 456, 485, 487, 488, 490, 491, 498, 499, 500, 505, 510, 511, 516, 517, 522 materials, x, 62, 69, 262, 269, 317, 322, 410, 416, 422, 424, 425, 426, 433, 448, 450, 456, 518, 519, 525, 526 matrix, 31, 32, 87, 88, 90, 96, 104, 105, 120, 121, 122, 136, 138, 139, 180, 286, 425 matter, 151, 182, 208, 239 media, 77, 144, 197, 223, 529 medicine, xiii, 57, 58, 69 medium access control, 240 membership, 282 memory, 69, 71, 166, 190, 193, 196, 455, 456, 482 messages, 166, 167, 170, 171, 200, 214, 215, 238, 240, 241 metals, 419, 434 meter, 42, 155, 156, 157 methanol, 524 methodology, 58, 69, 177 microcrystalline, 455 microwaves, 167 migration, 220 military, 82, 109, 188, 194, 322 miniaturization, 179 mission(s), 81, 82, 84, 110 mixing, 194 mobile communication, 165, 242 mobile device, 172 mobile phone, 165, 191, 214 mobile robots, 57, 69 modelling, 241, 380, 395 models, 18, 19, 87, 95, 99, 102, 103, 106, 135, 155, 165, 175, 177, 189, 196, 198, 208, 210, 217, 222, 223, 241, 275, 276, 277, 281, 289, 318, 404, 431, 432 modifications, 35, 410, 415 modules, 155, 168, 180, 189, 192, 193, 195, 220, 242, 416, 419, 420, 438, 439, 440 moisture, 425 molecules, 42 momentum, 89, 90, 91, 93, 220 Monte Carlo method, 305 Monte Carlo Simulation, 528 morale, 188 morphology, 161 Moscow, 231, 435 motion control, 81, 84 motivation, 166, 426 motor control, ix, 251, 260, 270, 321 multilateration, 3 multimedia, 197, 237, 238, 512 multiples, 217 multiplication, 12 music, 166 N nanotechnology(s), 322, 323, 409, 425, 426 National Academy of Sciences, x national emergency, 168 national security, 188, 189 NATO, 53, 54, 482 natural evolution, 486 natural gas, 514 natural resources, 219 navigation, 53, 58, 78, 111, 143, 145, 146, 211, 213, 214, 229 navigation system, xii, 6, 183, 214, 243, 512 neglect, 189, 195 network operators, 216 networking, 241 neural network(s), 267, 512

Neuro-fuzzy, 280

neutral, 93, 267, 307, 489 next generation, 293, 322, 431 Ni-Cd batteries, 415, 416, 426, 454, 455, 456, 481 nickel, 409, 415, 417, 418, 419, 434, 437, 454, 455, 456 niobium, 433 nodes, 168, 188, 189, 199, 202, 300 noise power estimation, 11, 37, 52 nonlinear systems, 83 non-renewable resources, 210 normal distribution, 298 North America, 244 Norway, 227 NPS, 83, 110 nucleation, 455 null, 384, 388, 391, 404 numerical computations, 95

O

obstacles, 59, 116, 117, 119, 132, 133, 497 oceans, 218 octane, 514 octane number, 514 Office of Management and Budget, 227 oil, 188, 489, 490, 511, 513, 524, 525 one dimension, 124, 130 operating range, 450 operating system, 161, 183, 190, 191, 192, 193, 198 operations, xi, 82, 84, 185, 187, 192, 216, 260, 267, 270, 277, 398, 399, 400, 401, 402, 404 opportunities, xii, 219, 325, 326 optimal performance, 13 optimization, 60, 122, 135, 297, 305, 309, 316, 317, 494 organism, 410 originality, 382, 392 oscillation, 395 overlap, 130, 141 ownership, 209 oxidation, 413, 423 oxygen, 322, 414, 416, 420, 422, 423, 424, 425, 454, 455 oxygen absorption, 454

P

ozone, 521

Pacific, 229 parallel, xiv, 15, 60, 66, 86, 114, 120, 127, 134, 135, 173, 190, 223, 295, 296, 310, 313, 316, 317, 328,

329, 333, 334, 335, 344, 357, 362, 419, 438, 440, 441, 442, 456, 458, 488, 522, 523 parallel implementation, 135 parameter estimation, 83, 102, 104 partial differential equations, 84 participants, 115, 170, 172, 297 partition, 171 password, xiii, 185, 199, 202 pattern recognition, 58, 69 pedal, 319, 326 performance indicator, 319, 321 permeability, 158 permission, 345 permit, 149, 150, 155, 157, 177 permittivity, 49 personal communication, 165 personal computers, 191, 192 personal life, 216 PES, 358, 359, 360, 365, 367, 376, 528 Petroleum, 511, 527 phosphate(s), 433, 434 photovoltaic panels, 314, 315 physical interaction, 169 physical properties, xv, 432, 448, 449 physical structure, 252, 265 pipeline, 511 pitch, 85, 86, 92, 499 plants, 521 platform, 196, 219, 228, 241, 327 playing, 218 PM, 265, 314, 319, 333, 506 pneumonia, 410 Poland, 53, 54 polar, 386 polarity, 364, 387, 390, 392 police, 147, 150, 168 policy, 180, 188, 208, 209, 211, 221, 231, 232, 487 policy making, 209 politics, 149 pollutants, 317, 521 pollution, xi, 149, 189, 318, 410, 411, 521, 527 polymer, 322, 324, 334, 419 poor performance, 59 population, 202, 524 portability, 75, 76 Portugal, 272 postal service, 511 potassium, 415, 422, 455 power generation, 193, 414, 513 power plants, 194, 277, 489

precipitation, 155 preparation, 263 preservation, 210, 223

prevention, 242, 244 primary cells, 433 principles, 35, 36, 103, 224, 364, 527 prior knowledge, 28 private sector, 209 probability, 5, 11, 12, 14, 15, 16, 17, 18, 19, 20, 21, 25, 36, 37, 38, 39, 46, 47, 52, 104, 123, 125, 127, 128, 131, 134, 135, 204, 297, 298, 300, 303, 305, 306, 493, 494, 501, 505 probability density function, 18, 306, 493 probability distribution, 297, 298, 305 Probability of Detection, 46, 54 probe, 176, 177 producers, 303, 318, 497 product life cycle, 177 professionals, xii, xv, 212, 224 profit, xiv, 209, 293, 298, 302, 303, 309, 311 programming, xiv, 166, 197, 204, 293, 294, 297, 298, 300, 301, 303, 304, 307, 308, 309 project, 53, 59, 68, 77, 172, 173, 177, 208, 221 proliferation, 218 propagation, 8, 58, 241, 283 proportionality, 35 propylene, 435 protection, xiii, 179, 186, 187, 189, 194, 202, 203, 210, 242, 322, 416 prototype(s), 114, 115, 116, 119, 131, 132, 173, 353, 354, 376, 410, 417, 420, 431, 432, 436, 503, 511 public opinion, 202 publishing, 203 pumps, 251 PVA, 482

Q

quadratic programming, 122 quality of life, 211 quality of service, 167, 222, 240 quantification, 121 quantization, 71 Queensland, 198

R

racing, 498 radar, ix, xiii, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 18, 19, 20, 24, 28, 29, 30, 31, 32, 36, 38, 41, 42, 43, 44, 45, 47, 48, 49, 51, 52, 53, 59, 113, 116, 118, 120, 132, 144, 147, 148, 150, 155, 156, 242 radiation, 47, 63, 117, 155, 156 radio, 11, 155, 166, 167, 168, 169, 170, 171, 172, 182, 189, 212, 215, 236, 237, 238, 239, 240

radius, 10, 29, 63, 107, 130, 171, 491, 501 rain attenuation, 45, 46 rainfall, 45, 47, 51 ramp, 149, 152, 161, 215, 263 rate of change, 90 rating scale, 499 RB1, 66 reaction time, 236 reactions, xii, 413, 416, 425, 434, 454 reading, 13, 14 real numbers, 105 real time, 5, 11, 37, 52, 59, 62, 76, 134, 135, 147, 149, 150, 161, 182, 225 reality, 44, 116, 182, 194, 208, 453, 454, 473, 479, 504 reasoning, 218 recall, 113, 131, 135, 136, 137, 138, 139 reception, 63 recognition, xii, 11, 57, 58, 74, 124, 147, 148, 150, 154, 160, 161, 164, 165, 167, 183, 189, 242 recombination, 414 recommendations, 311 recovery, 82, 168, 182, 203, 412 rectification, 253 recurrence, 155 recycling, 422, 456, 524 redundancy, 168, 220, 244 reference frame, 86, 98, 387 reference system, 89 reflectivity, 116 regression, 103, 125 rejection, 129 relaxation, xv, 453, 454, 456, 457, 458, 459, 460, 461, 464, 465, 466, 467, 469, 472, 477, 478, 479, 480, 482, 483 relaxation process, 453, 457, 480 relaxation times, 480 relevance, 209, 236 reliability, ix, xii, xiv, 70, 118, 155, 178, 180, 194, 195, 196, 212, 249, 261, 263, 265, 276, 319, 322, 325, 333, 380, 410, 486, 502 remote sensing, 11 renewable energy, 310, 311, 511, 527 renewable fuel, 215

requirements, ix, xiv, 8, 82, 119, 167, 170, 172, 188, 195, 196, 202, 207, 208, 224, 249, 252, 255, 260, 261, 262, 315, 317, 319, 321, 322, 345, 410, 413, 426, 440, 441, 455, 485, 493, 494, 495, 503, 511, 512, 523, 524, 526 RES, 314, 315

research institutions, 486 researchers, xii, xv, xvi, 185, 198, 217, 220, 419, 423 reserves, 276, 311, 410, 426

residential, 291 resilience, 188 resistance, 34, 63, 255, 262, 323, 356, 360, 397, 414, 426, 455, 458, 461, 478, 480 resolution, 19, 61, 73, 76, 117, 118, 120, 147, 150, 180, 265 resources, xv, 149, 168, 179, 190, 193, 209, 210, 217, 218, 219, 220, 222, 223, 224, 225, 294, 436, 488, 489 response, xiv, 32, 64, 71, 75, 102, 149, 167, 169, 186, 190, 197, 198, 199, 203, 279, 282, 283, 311, 315, 321, 325, 326, 353, 355, 371, 372, 374, 375, 384, 456, 457, 459, 461, 464, 467, 469, 471, 472, 477, 517 response time, 167, 190, 315 responsiveness, 218, 219 restaurants, 166 restoration, 311 restrictions, 147, 148, 150, 189, 215 revenue, 243, 276 rights, 221, 244 rings, 262, 393 risk(s), 82, 133, 151, 185, 188, 190, 294, 297, 301, 302, 303, 304, 307, 309, 310, 311, 420, 434, 435, 503 risk assessment, 188 risk aversion, 307 risk management, 297 robotics, 57, 59, 96 Romania, xvi, 57, 78, 113, 143, 144, 147, 184, 185, 207, 313, 353, 377, 407, 409, 427 room temperature, 324, 449 root, 254, 300 rotations, 87, 493 routes, 147, 149, 151, 175, 177, 183, 239 routines, 160, 168 ROVs, 82 rules, 176, 191, 281, 317 Russia, 111, 231, 406 S safety, ix, xiii, 3, 4, 5, 6, 7, 8, 20, 51, 53, 59, 143, 144, 147, 148, 149, 150, 167, 169, 183, 188, 195, 210, 212, 214, 215, 216, 229, 236, 237, 238, 240, 244, 319, 322, 415, 421, 423, 426, 427, 432, 433, 435, 455, 456, 498, 517 Safety Management System, 211 salinity, 96 salts, 424, 435 sawdust, 523 scaling, 12, 14, 15, 16, 17, 25, 37 scattering, 45 schema, 65, 394 science, 220 scientific understanding, 208 scope, 208, 221, 223, 513, 515 SDSR, 203 sea level, 43, 44 second generation, 487 Second World, 524 sectoral policies, 221 security, ix, xiii, 84, 147, 176, 179, 183, 185, 186, 187, 188, 189, 190, 191, 192, 194, 195, 196, 198, 202, 203, 204, 205, 213, 239, 242, 244, 311, 420 security services, 188 semicircle, 458, 459 semiconductor(s), 118, 144, 257, 265 senses, 216, 375 sensing, ix, xiii, 62, 69, 82, 113, 114, 115, 116, 119, 142, 143 sensitivity, 4, 52, 62, 68, 117, 264, 285, 341, 381, 384, 389, 391, 511 sensor network, 8, 28, 29, 30, 31, 32, 53 servers, 179, 180, 183, 192, 199 service provider, 238 service quality, 210 services, xiii, 147, 150, 165, 166, 168, 172, 173, 174, 175, 176, 188, 199, 203, 204, 212, 215, 216, 219, 222, 223, 237, 238, 241, 242, 244, 276, 294, 310, 311, 325 shape, 58, 62, 67, 68, 74, 82, 95, 109, 116, 130, 194, 221, 372, 374, 445 shock, 426, 517, 523 shortage, 512 showing, 83, 131, 454, 459, 480, 492, 493 side effects, 191, 521 signalling, 154, 172, 237, 242 signals, 5, 8, 10, 14, 22, 23, 24, 25, 37, 38, 40, 41, 52, 60, 62, 65, 66, 67, 70, 71, 74, 76, 154, 166, 167, 169, 170, 171, 180, 216, 251, 252, 262, 263, 266, 267, 270, 285, 326, 363, 365, 366, 370, 372, 374, 375, 380, 381, 392 signal-to-noise ratio, 4 signs, 115, 116, 120, 146, 148, 149, 150, 161, 172, 183 silicon, 322, 426, 428 Singapore, 229 skewness, xv, 453, 478, 479, 480, 481 slaves, 171 Small particles, 521

smog, 58, 521 social interactions, 209

saturation, 389 Saudi Arabia, 235 savings, 176, 177

social responsibility, 208, 511 society, 167, 187, 208, 210, 216, 217, 219, 220, 223, 224, 225, 230, 410, 435, 486, 505, 529 sodium, 409, 418 software, 6, 59, 62, 74, 76, 95, 148, 149, 150, 154, 161, 168, 180, 185, 191, 192, 193, 194, 195, 196, 198, 204, 205, 217, 219, 220, 267, 436 solution, xiii, 28, 30, 31, 51, 52, 58, 60, 66, 69, 76, 92, 122, 147, 148, 160, 163, 179, 180, 182, 183, 188, 202, 220, 223, 297, 299, 354, 375, 414, 419, 422, 448, 450, 481, 487, 501, 502, 509, 510, 517, 524, 525, 527 solvents, 424, 435 Sonar, 58, 59, 77, 78, 110 South Africa, 506 South Korea, 3, 54, 310 Spain, 78, 79, 143, 146, 272, 310, 407, 431, 435 spam, 192 spatial information, 209, 217, 220 specialists, xii, 178, 185, 187, 209, 316, 317 species, 454 specifications, 182, 217, 223, 238, 242, 245, 269, 440, 491, 495 spectral component, 381, 389 spectroscopy, 481 speech, 11 Speed detection, 148, 150, 160 spending, 160, 240 spin, 117, 134, 529 sponge, 414 Spring, 232, 245 spyware, 192 St. Petersburg, 111, 406	439, 488, 509, 510, 511, 515, 516, 517, 519, 522, 525, 526, 527 strategic planning, 221 strategy use, 100 stress, 199, 261, 320, 385, 455 stress test, 199 stress testing, 199 structuring, 219 subscribers, 168 substitutes, 210 substitution, 198, 199 subtraction, 121 succession, 134, 161 suicide, 192 sulfur, 322 sulphur, 418, 426 Sun, 77, 111, 347, 428, 451 Super Bowl, 145 supervision, x, xiv, 168, 379, 380, 381, 382, 404 supervisor(s), 178, 182 supplier(s), 182, 183 surface area, 108, 425 surveillance, ix, xiii, 82, 147, 148, 149, 150, 151, 152, 179 Surveillance, vii, 79, 147, 150 survival, 219 survivors, 220 sustainability, xiii, 207, 208, 210, 217, 223, 224, 225, 226, 227, 271, 515 sustainable development, 208, 209, 210, 216, 222, 225, 226, 231 sustainable energy, 515 Switzerland, 226, 419
stability, 37, 66, 94, 102, 118, 220, 284, 294, 425,	symmetry, 91, 162
426, 432, 498, 499 stabilization, 334	synchronization, 8, 240 synthesis, 207, 270
stabilizers, 325
stakeholders, 217, 218
standard deviation, 37, 305, 478	T
standardization, 191, 222, 223, 320	tanks, 525, 526
starch, 524	tariff, 302
statistics, xiii, 5, 14, 16, 18, 35, 52, 148, 150, 161,	taxes, 161, 169
177, 199, 410	teachers, v
steel, 419	teams, 148, 150, 178
Steganography, 186	technical support, 180
stimulation, 82 stochastic model, 310	technician, 182
Stochastic programming, 294, 300, 309	techniques, xiii, 11, 38, 52, 57, 59, 81, 110, 117, 125,
stock, 503	152, 164, 165, 170, 183, 186, 194, 195, 204, 210,
storage, xi, xii, xiv, xv, 71, 180, 182, 183, 209, 219,	220, 252, 260, 261, 265, 270, 281, 316, 320, 328,
257, 275, 276, 277, 278, 280, 287, 289, 290, 294,	358, 380, 381, 391, 526
310, 315, 317, 318, 324, 334, 431, 432, 433, 437,	technological advances, 217, 223
	technological developments, ix, 380, 512

technological progress, 165

telecommunications, 165, 166, 420 Telematics, 235, 237 telephone(s), 165, 433 temperature, xv, 42, 43, 44, 95, 155, 262, 321, 357, 362, 418, 419, 431, 432, 435, 445, 447, 448, 449, 450, 455, 524 temporal variation, 14 terminals, 62, 183, 262, 397, 400 territorial, 221 territory, 209, 221, 293 terrorist attack, 203 test data, 136 testing, xiv, xv, 65, 95, 114, 125, 131, 136, 198, 224, 353, 371, 372, 375, 431, 487, 511 textbooks, 261, 265 texture, 58 theft, 202 thermal energy, 517 threats, xiii, 185, 187, 188, 190, 191, 192, 196, 203, 216 time frame, 98, 215 tin, 426 titanium, 432, 434 topology, 179, 215, 218, 239, 251, 258, 267, 269, 316, 318, 320, 321, 323, 324, 325, 326, 327, 328, 333, 334, 335, 336, 337, 338, 341, 344, 345, 412, 488, 490, 497, 500, 501 total costs, 177 toxic metals, 322 Toyota, 318, 354, 417, 420, 421, 496, 511, 522, 523, 528 toys, 454 tracks, 504 trade-off, 132, 194, 325 traffic information, 166 traffic violations, 148, 150 training, 122, 125, 126, 127, 128, 129, 136, 137, 219, 282, 283 trajectory, 83, 95, 243 transactions, xii, 186, 194 transducer, 62, 63, 64, 65, 66, 68, 70, 71, 75, 359 transformation, 86, 88, 263 transformation matrix, 86, 88 transistor, 67 transition metal, 419 transition period, 494 translation, 163 transmission, 8, 62, 71, 170, 171, 183, 188, 214, 237, 259, 294, 295, 490, 526 transparency, 204, 208 transportation infrastructure, 212 treatment, 397, 490 trial, 467 troubleshooting, 200 turbulence, 155 Turkey, 81, 249, 272 U UK, 54, 225, 226, 345, 376, 407, 505, 506, 509, 519, 529 Ultracapacitor, 314, 346, 348, 351 ultrasonic frequency, 57 ultrasound, xii, xiii, 57, 58, 61, 69, 70 underwater vehicles, ix, xii, xiii, 95 UNESCO, 231 unidirectional links, 239 uniform, 45, 72, 162, 443 United Kingdom, 187, 202, 203 United Nations, 208, 224, 225, 226, 227, 530 United States (USA), 53, 54, 77, 145, 146, 169, 187, 188, 203, 204, 205, 211, 227, 228, 229, 230, 240, 245, 272, 311, 410, 416, 417, 505, 520, 528 universities, 114, 198, 410 updating, 209 uplink, 171, 238 urban, 115, 119, 143, 151, 172, 173, 241, 244, 326, 369, 415, 486, 490, 491, 492, 495, 496 urban areas, 172, 369 urbanization, 212 user data, 242 V vacuum, ix valence, 413 validation, xv, 125, 161, 342, 379, 404 valve, 416 variables, 83, 102, 122, 264, 281, 282, 297, 298, 301, 303, 326 variations, 12, 49, 95, 155, 253, 258, 264, 276, 280, 288, 382, 393, 397, 434 vector, 30, 32, 87, 88, 89, 91, 92, 95, 97, 98, 99, 103, 104, 121, 122, 123, 124, 126, 127, 128, 129, 133, 134, 135, 163, 164, 185, 262, 263, 264, 282, 283, 285, 286, 380 vegetable oil, 524 velocity, 4, 5, 6, 7, 8, 9, 10, 18, 19, 22, 51, 83, 84, 85, 86, 87, 88, 89, 90, 92, 95, 97, 101, 105, 108, 135, 262, 393, 394, 444, 445, 450, 517 ventilation, 357, 362 versatility, 82 vibration, 63, 393, 499 virtualization, 183 viruses, 186, 187, 191, 192, 195, 196

viscosity, 524 Visibility, 48, 50 vision, 57, 58, 117, 146, 191, 208, 212, 213, 215, 216, 221, 225 visualization, 219 vocabulary, 130 volatile organic compounds, 521 Volkswagen, 494 vote, 125, 126 voting, 125, 126, 132, 137, 138, 140 vulnerability, 191, 199, 225

W

waste, xv, 210, 432, 487, 515, 517, 523 water, xiii, 42, 43, 44, 47, 49, 50, 84, 86, 95, 96, 115, 185, 188, 189, 193, 202, 203, 210, 414, 416, 422, 425, 439, 447, 448, 450, 489, 498, 501, 502, 524, 525 water quality, 210 water supplies, 203 water vapor, 425 waterways, 238 weapons, 186 wear, 382, 393 web, 173, 175, 191, 198, 220, 230

web browser, 191 weight ratio, 505 Western countries, 527 wholesale, 293 Wi-Fi, 171, 187, 202 wind turbines, 276, 315, 515 windows, 13, 15, 16, 26, 38, 130, 180 wireless networks, 202 wireless systems, 237 wires, 182, 397 Wisconsin, 146, 275 withdrawal, 421 wood, 523, 524 workflow, 223 working conditions, 178, 179, 492, 494, 496, 499 workstation, 180, 182 world order, 222 World Wide Web, 223 worldwide, 59 worms, 186, 192

X

Z

XML, 222

Zebra batteries, 418, 419 zirconium, 433 ZnO, 483