IET CONTROL ENGINEERING SERIES 92

Handbook of
Vehicle Suspension
Control Systems
Other volumes in this series:

Volume 8 A history of control engineering, 1800–1930 S. Bennett

Volume 18 Applied control theory, 2nd Edition J.R. Leigh
Volume 20 Design of modern control systems D.J. Bell, P.A. Cook and N. Munro (Editors)
Volume 28 Robots and automated manufacture J. Billingsley (Editor)
Volume 33 Temperature measurement and control J.R. Leigh
Volume 34 Singular perturbation methodology in control systems D.S. Naidu
Volume 35 Implementation of self-tuning controllers K. Warwick (Editor)
Volume 37 Industrial digital control systems, 2nd Edition K. Warwick and D. Rees (Editors)
Volume 39 Continuous time controller design R. Balasubramanian
Volume 40 Deterministic control of uncertain systems A.S.I. Zinober (Editor)
Volume 41 Computer control of real-time processes S. Bennett and G.S. Virk (Editors)
Volume 42 Digital signal processing: principles, devices and applications N.B. Jones and
J.D.McK. Watson (Editors)
Volume 44 Knowledge-based systems for industrial control J. McGhee, M.J. Grimble and
A. Mowforth (Editors)
Volume 47 A history of control engineering, 1930–1956 S. Bennett
Volume 49 Polynomial methods in optimal control and filtering K.J. Hunt (Editor)
Volume 50 Programming industrial control systems using IEC 1131-3 R.W. Lewis
Volume 51 Advanced robotics and intelligent machines J.O. Gray and D.G. Caldwell (Editors)
Volume 52 Adaptive prediction and predictive control P.P. Kanjilal
Volume 53 Neural network applications in control G.W. Irwin, K. Warwick and K.J. Hunt (Editors)
Volume 54 Control engineering solutions: a practical approach P. Albertos, R. Strietzel and
N. Mort (Editors)
Volume 55 Genetic algorithms in engineering systems A.M.S. Zalzala and P.J. Fleming (Editors)
Volume 56 Symbolic methods in control system analysis and design N. Munro (Editor)
Volume 57 Flight control systems R.W. Pratt (Editor)
Volume 58 Power-plant control and instrumentation: the control of boilers and HRSG
systems D. Lindsley
Volume 59 Modelling control systems using IEC 61499 R. Lewis
Volume 60 People in control: human factors in control room design J. Noyes and M. Bransby
(Editors)
Volume 61 Nonlinear predictive control: theory and practice B. Kouvaritakis and M. Cannon
(Editors)
Volume 62 Active sound and vibration control M.O. Tokhi and S.M. Veres
Volume 63 Stepping motors, 4th Edition P.P. Acarnley
Volume 64 Control theory, 2nd Edition J.R. Leigh
Volume 65 Modelling and parameter estimation of dynamic systems J.R. Raol, G. Girija and
J. Singh
Volume 66 Variable structure systems: from principles to implementation A. Sabanovic,
L. Fridman and S. Spurgeon (Editors)
Volume 67 Motion vision: design of compact motion sensing solution for autonomous
systems J. Kolodko and L. Vlacic
Volume 68 Flexible robot manipulators: modelling, simulation and control M.O. Tokhi and
A.K.M. Azad (Editors)
Volume 69 Advances in unmanned marine vehicles G. Roberts and R. Sutton (Editors)
Volume 70 Intelligent control systems using computational intelligence techniques A. Ruano
(Editor)
Volume 71 Advances in cognitive systems S. Nefti and J. Gray (Editors)
Volume 72 Control theory: a guided tour, 3rd Edition J.R. Leigh
Volume 73 Adaptive sampling with mobile WSN K. Sreenath, M.F. Mysorewala, D.O. Popa and
F.L. Lewis
Volume 74 Eigenstructure control algorithms: applications to aircraft/rotorcraft handling
qualities design S. Srinathkumar
Volume 75 Advanced control for constrained processes and systems F. Garelli, R.J. Mantz and
H. De Battista
Volume 76 Developments in control theory towards glocal control L. Qiu, J. Chen, T. Iwasaki and
H. Fujioka (Editors)
Volume 77 Further advances in unmanned marine vehicles G.N. Roberts and R. Sutton (Editors)
Volume 78 Frequency-domain control design for high-performance systems J. O’Brien
Volume 81 Optimal adaptive control and differential games by reinforcement learning
principles D. Vrabie, K. Vamvoudakis and F. Lewis
Volume 88 Distributed control and filtering for industrial systems M. Mahmoud
Volume 91 An introduction to fractional control D. Valério and J. Costa
 Handbook of
Vehicle Suspension
Control Systems
Edited by Honghai Liu,
Huijun Gao and Ping Li

The Institution of Engineering and Technology

Published by The Institution of Engineering and Technology, London, United Kingdom
The Institution of Engineering and Technology is registered as a Charity in England & Wales
(no. 211014) and Scotland (no. SC038698).
† The Institution of Engineering and Technology 2014
First published 2013

This publication is copyright under the Berne Convention and the Universal Copyright
Convention. All rights reserved. Apart from any fair dealing for the purposes of research
or private study, or criticism or review, as permitted under the Copyright, Designs and
Patents Act 1988, this publication may be reproduced, stored or transmitted, in any
form or by any means, only with the prior permission in writing of the publishers, or in
the case of reprographic reproduction in accordance with the terms of licences issued
by the Copyright Licensing Agency. Enquiries concerning reproduction outside those
terms should be sent to the publisher at the undermentioned address:

The Institution of Engineering and Technology

Michael Faraday House
Six Hills Way, Stevenage
Herts, SG1 2AY, United Kingdom
www.theiet.org

While the authors and publisher believe that the information and guidance given in
this work are correct, all parties must rely upon their own skill and judgement when
making use of them. Neither the authors nor publisher assumes any liability to anyone
for any loss or damage caused by any error or omission in the work, whether such
an error or omission is the result of negligence or any other cause. Any and all such
liability is disclaimed.
The moral rights of the authors to be identified as authors of this work have been
asserted by them in accordance with the Copyright, Designs and Patents Act 1988.

British Library Cataloguing in Publication Data

A catalogue record for this product is available from the British Library

ISBN 978-1-84919-633-8 (hardback)

ISBN 978-1-84919-634-5 (PDF)

Typeset in India by MPS Limited

Printed in the UK by CPI Group (UK) Ltd, Croydon
 Contents

Preface xi

1 State-of-the-art of vehicle intelligent suspension control system 1

Xiaomin Dong
Abstract 1
1.1 Introduction 1
1.2 Evaluation criterion of vehicle suspension performance 4
1.2.1 Ride comfort 5
1.2.2 Road holding 5
1.3 Modeling of vehicle suspension system 5
1.3.1 Road model 6
1.3.2 Intelligent quarter-car model 6
1.3.3 Intelligent half-car model 8
1.3.4 Intelligent full-vehicle model 10
1.3.5 Other nonlinear vehicle dynamic model 11
1.3.6 Nonlinear multibody dynamic model 14
1.3.7 Nonlinear uncertainty modeling 15
1.3.8 Nonlinear dynamic model with time delay 16
1.3.9 Nonlinear dynamics model with fault 17
1.3.10 The modeling of actuator 18
1.4 Control strategies 22
1.4.1 Linear control strategies 24
1.4.2 Nonlinear control strategies 24
1.4.3 Uncertainty control methods 25
1.4.4 Time delay control methods 26
1.4.5 Fault-tolerant control method 27
1.5 The method of validation 29
1.6 Final remarks and conclusions 31
Acknowledgments 33
References 33

2 Intelligence-based vehicle active suspension adaptive

control systems 39
Jiangtao Cao, Ping Li and Honghai Liu
Abstract 39
2.1 Introduction 39
vi Handbook of vehicle suspension control systems

2.2 Background 43
2.2.1 Active suspension system linear models and control 43
2.2.2 Nonlinearity and unmodeling dynamic description
of active suspension system 48
2.3 Adaptive fuzzy control 50
2.4 Adaptive fuzzy sliding-mode control 52
2.4.1 Alleviating SMC chattering 53
2.4.2 FL controller complementary to SMC for system
nonlinearity and uncertainty 55
2.5 Adaptive neural network control 56
2.6 Genetic algorithm-based adaptive optimization and control 58
2.7 Adaptive control integration 59
2.7.1 Adaptive neuro-fuzzy control 59
2.7.2 Adaptive genetic-based optimal fuzzy control 60
2.7.3 GA–NN combined control 62
2.8 Concluding remarks 62
Acknowledgments 63
References 63

3 Robust active control of an integrated suspension system 69

Haiping Du, James Lam, Weihua Li and Nong Zhang
Abstract 69
3.1 Introduction 69
3.2 Uncertain integrated system modelling 71
3.3 Robust control system design 75
3.3.1 Control objectives 75
3.3.2 Robust controller design 77
3.3.3 Force tracking control of electrohydraulic actuators 82
3.4 Numerical simulations 83
3.5 Conclusions 90
Acknowledgements 90
Appendix 91
References 94

4 An interval type-2 fuzzy controller for vehicle active

suspension systems 99
Jiangtao Cao, Ping Li and Honghai Liu
Abstract 99
4.1 Introduction 99
4.2 A nonlinear active suspension system 101
4.3 The interval type-2 T–S fuzzy control system 103
4.3.1 The general T–S fuzzy model and fuzzy control system 104
4.3.2 The interval type-2 T–S fuzzy control system 105
4.3.3 The proposed IT2 T–S fuzzy control system 107
 Contents vii

4.4 Stability analysis of the IT2 T–S fuzzy control system 109
4.5 Simulation examples 111
4.5.1 A numerical example 111
4.5.2 A half-vehicle active suspension system 113
4.6 Concluding remarks 120
References 120

5 Active control for actuator uncertain half-car suspension systems 125

Hongyi Li and Honghai Liu
Abstract 125
5.1 Introduction 125
5.2 Problem formulation 126
5.3 Main results 131
5.4 Simulation results 134
5.5 Conclusion 141
References 141

6 Active suspension control with finite frequency approach 143

Weichao Sun, Huihui Pan, Pinchao Wang and Huijun Gao
Abstract 143
6.1 Introduction 143
6.2 Problem formulation 144
6.3 State feedback controller design 148
6.4 Dynamic output feedback controller design 152
6.4.1 Finite frequency case 153
6.4.2 Entire frequency case 156
6.5 Simulation 158
6.5.1 State feedback case 158
6.5.2 Dynamic output feedback case 163
6.6 Concluding remarks 166
References 170

7 Fault-tolerant control for uncertain vehicle suspension

systems via fuzzy control approach 173
Hongyi Li and Honghai Liu
Abstract 173
7.1 Introduction 173
7.2 Problem formulation 175
7.3 Fault-tolerant fuzzy controller design 181
7.4 Simulation results 185
7.5 Conclusions 189
Appendix 190
References 193
viii Handbook of vehicle suspension control systems

8 H? fuzzy control of suspension systems with actuator

saturation 195
Dounia Saifia, Mohammed Chadli and Salim Labiod
Abstract 195
8.1 Introduction 195
8.2 Suspension systems model 197
8.2.1 Active quarter-car suspension model 197
8.2.2 Half-car suspension model 200
8.2.3 Full-car suspension model 204
8.3 Takagi–Sugeno fuzzy model of suspension systems 208
8.3.1 Takagi–Sugeno representation of active quarter-car
suspension 209
8.3.2 Takagi–Sugeno representation of active half-car
suspension 210
8.3.3 Takagi–Sugeno representation of active full-car
suspension 212
8.4 Validation of Takagi–Sugeno fuzzy model 216
8.4.1 Simulation parameters 216
8.4.2 Validation of Takagi–Sugeno fuzzy model 217
8.5 Actuator saturation 221
8.5.1 Types of saturation 223
8.5.2 Modelling of saturation effect 223
8.5.3 Saturated control and constrained control 224
8.6 Quadratic stabilization of Takagi–Sugeno fuzzy model 225
8.6.1 Convex analysis and linear matrix inequalities 225
8.6.2 Stability in the sense of Lyapunov 226
8.6.3 Attraction region 227
8.6.4 Quadratic stabilization via PDC control 228
8.7 H? approach 230
8.8 Analysis of PDC control with external disturbances and
actuator saturation 230
8.8.1 Constrained control 230
8.8.2 Saturated control 233
8.8.3 Optimization of the attraction region 234
8.9 Control design for a quarter-car active suspension system 235
8.10 Conclusion 239
References 243

9 Design of sliding mode controller for semi-active suspension

systems with magnetorheological dampers 247
Shigehiro Toyama, Makoto Yokoyama and Fujio Ikeda
Abstract 247
9.1 Introduction 248
9.2 Control of semi-active suspension systems with MR dampers 250
 Contents ix

9.2.1 Variable orifice damper 250

9.2.2 MR damper 252
9.3 Model-following sliding mode controller for semi-active
suspension systems 255
9.3.1 System model and problems 255
9.3.2 Sliding mode controller 257
9.3.3 Simulation results 259
9.4 Sliding mode controller with describing function method 261
9.4.1 Problem formulation 262
9.4.2 Integral sliding mode control 263
9.4.3 Redesign of relay input with describing function method 265
9.4.4 Simulation conditions 266
9.4.5 Accuracy of limit cycle of switching function 267
9.4.6 Improvement of deterioration caused by passivity constraint 268
9.4.7 Verification of robustness against parameter variation 269
9.5 VSS observer of semi-active suspension systems 272
9.5.1 Plant 273
9.5.2 Problem formulation 274
9.5.3 Design of VSS observer 275
9.5.4 Numerical simulations 278
References 283

10 Joint design of controller and parameters for active

vehicle suspension 287
Wei Zhan, Qingrui Zhang, Yinan Liu and Huijun Gao
Abstract 287
10.1 Overview 287
10.2 Problem formulation 289
10.3 Joint design of the system 290
10.4 Simulation results 293
10.5 Conclusion 299
Acknowledgement 300
References 300

11 System approach to vehicle suspension system control

in CAE environment 303
Vladimir M. Popović and Dragan D. Stamenković
Abstract 303
11.1 Introduction 303
11.2 Classification of mechatronic suspension systems 304
11.3 Design development process 306
11.4 Active suspension system modeling 308
11.4.1 Model of the system in state-space 309
11.4.2 Synthesis of active suspension digital system 313
x Handbook of vehicle suspension control systems

11.4.3 Active suspension control using PID controller 313

11.4.4 Active suspension control using neural network 318
11.5 Conclusions 323
References 324

12 Vibration control of vehicle engine-body systems with time

delay: an LMI approach 327
Hamid Reza Karimi
Abstract 327
12.1 Introduction 327
12.2 The vehicle engine-body system 330
12.3 Problem formulation 335
12.4 Main results 336
12.4.1 State-feedback control design 337
12.4.2 Output-feedback control design 344
12.5 Simulation results 346
12.6 Conclusion 351
References 353

13 Frequency domain analysis and design of nonlinear vehicle

suspension systems 357
Yue Chen, Xingjian Jing and Li Cheng
Abstract 357
13.1 Introduction 357
13.2 System model and the output frequency response function
(OFRF) method 360
13.2.1 System model 360
13.2.2 Determination of the system OFRF 362
13.2.3 Optimization and system analysis 366
13.2.4 Conclusion 371
13.3 Comparative studies 372
13.3.1 Existing nonlinear damping characteristics 372
13.3.2 Damping characteristics designed via the OFRF-based
analysis method 374
13.3.3 Comparative studies 376
13.3.4 Dynamic model verification 385
13.3.5 Conclusion 386
13.4 Application on a dynamic vehicle model 387
13.4.1 Dynamic vehicle model 387
13.4.2 Simulation study 389
13.4.3 Summary 395
13.5 Conclusion and future work 396
References 397

Index 401
 Preface

This book presents recent control theory and applications for vehicle suspension
systems. From the systematic point of view, a vehicle control system is composed
of key components such as modeling, sensing, controller, and actuator. It is not
our intention to cover all the technical details of recent control-related contribu-
tions in the context of vehicle suspensions, but priority has been given to recently
reported novel control methods and key challenges in the past decades. This book
consists of 13 self-contained chapters covering recent theoretical developments
and applications in active suspension systems.
The first two chapters review the state of the art of modeling and control
design, and computational intelligence approaches, respectively, in intelligent
suspension systems. Dong overviews the evaluation criterion of vehicle suspen-
sion performance, and also recent contributions to modeling vehicle suspensions,
focusing on control algorithms dealing with nonlinearity, uncertainty, time delay,
and faults. On the other hand, Cao et al. highlight computational intelligence
involved in active vehicle suspension control systems with an emphasis on the
problems that arose in practical implementations by their nonlinear and uncertain
properties. The two chapters form a joint force to point out the future directions:
first, a more precise vehicle suspension dynamic model, even a full vehicle
model, is required urgently to be developed; second, efforts are called for to close
the gap between vehicle modeling and implementation; third, an integration
of multi-objective optimization methods and computational intelligence, for
instance, fuzzy logic reasoning to handle the trade-off between riding comfort,
road handling quality, and other practical issues, is expected; finally, it is con-
firmed that it is essential to evaluate all kinds of control algorithms and hybrid
intelligent algorithms from the perspective of practical applications. The last two
chapters investigate control issues in the frequency domain. Sun et al. investigate
the problem of vehicle active suspension control with frequency band constraints.
The generalized Kalman–Yakubovich–Popov (GKYP) lemma is employed to
satisfy the time-domain constraints achieving better disturbance attenuation per-
formance over the concerned frequency range. Chen et al. give priority to how
to determine a proper stiffness and damping characteristics to meet various
requirements in practice. A nonlinear frequency domain analysis method
is introduced for nonlinear analysis and design of vehicle suspension systems
to obtain the relationship between the nonlinear suspension system output and
system parameters.
xii Handbook of vehicle suspension control systems

The majority of the chapters focus on recently reported advanced control

strategies, e.g., stability analysis, in the time domain. Du et al. present the study of
robust active control of an integrated vehicle suspension system that consists of
chassis suspension, seat suspension, and driver body models. It should be noted
that, considering the limited capability of actuators and the parameter uncertainties,
actuator saturation constraint and parameter uncertainties to driver body model are
included in the controller design process. Cao et al. report a novel interval type-2
fuzzy controller architecture for resolving nonlinear control problems of vehicle
active suspension systems. It integrates Takagi–Sugeno (T–S) fuzzy model, inter-
val type-2 fuzzy reasoning, the Wu–Mendel uncertainty bounds method, and
selected optimization algorithms together to secure stability of the constructed
switching routes between generated linear model control surfaces. Li et al. design a
non-fragile H-infinity controller for a class of active suspension systems with
actuator uncertainty. The Lyapunov stability theory employed guarantees asymp-
totic stability with a prescribed H-infinity disturbance attenuation level. Li and his
colleagues also investigate the problem of fault-tolerant fuzzy controller design. A
T–S fuzzy model approach is proposed to study the suspension systems with the
sprung and unsprung mass variations, the actuator fault, and other suspension
performances. The fault-tolerant fuzzy H-infinity controller has been designed such
that the resulting closed-loop T–S fuzzy system is asymptotically stable with
H-infinity performance, and simultaneously satisfies the constraint suspension
performance. Saifia et al. focus on H-infinity fuzzy control of suspension systems
under actuator saturation. A T–S approach is used to model the suspension system
(quarter, half, and full car) via state feedback under actuator saturation and external
disturbances by interpolation of different local linear models. Toyama et al. present
two sliding mode controllers of semi-active suspension systems with magne-
torheological dampers, which have undesirable nonlinear properties. One sliding
mode controller is based on the theory of model-following control. The other
sliding mode controller is designed by the describing function method so that a
switching function is enforced into a desired limit cycle instead of a perfect sliding
mode. Zhang et al. propose a joint optimization method, obtaining a desirable state-
feedback controller gain and the parameters of the system in the context of mod-
eling quarter-car active suspension system. It is stated that better control effects and
disturbance attenuation property can be achieved when the parameters of the sys-
tem are optimized by GA-based approach. Karimi studies a state- and out-feedback
H-infinite controller solution to resolve the problem of vibration control analysis
and synthesis of a vehicle engine-body vibration system. Additionally, Popovic and
Stamenkovic draw our attention to a tendency of replacing electromechanical
components by mechatronic systems with intelligent and autonomous properties
in motor vehicle industry. The key challenges associated with active suspension
and their actuators are size, weight, and energy consumption required to achieve
acceptable performance.
We wish to express our sincere thanks to many people who contributed to this
book. The first are the colleagues who contributed their excellent research in order
to bring the valuable materials for postgraduates, researchers, and practitioners.
 Preface xiii

We gently appreciate the book chapter contributors for their patience and time
taken to collaborate with us to eventually complete this book. Next there are
students and colleagues who asked many relevant questions during the book pre-
paration and assisted us in providing better explanations and to improve this
manuscript. Finally, we would like to sincerely appreciate the IET editorial staff.
Without their continuous support, assistance, and significant improvement in the
manuscript, this book would not be published as scheduled. We would also like to
thank the IEEE and the IET for granting the copyright permissions for some con-
tents of the published papers.
Honghai Liu, Portsmouth, UK
Huijun Gao, Harbin, China
Ping Li, Fushun, China
 Chapter 1
State-of-the-art of vehicle intelligent suspension
control system
Xiaomin Donga

Abstract

This chapter reviews the state-of-the-art of modeling and control design in intelli-
gent suspension systems. After briefly reviewing the evaluation criterion of vehicle
suspension performance, some new methods of modeling vehicle suspension are
summarized. These methods are beneficial in the development of intelligent sus-
pensions. Then some control algorithms of dealing with nonlinearity, uncertainty,
time delay, and fault are reviewed. The validation methods of control algorithms
are discussed. At last, final remarks and conclusions are given.

1.1 Introduction

The vehicle suspension system is responsible for ride comfort and road holding as
the suspension carries the vehicle body and transmits all forces between the body
and the road. A classical car suspension consists of a spring (coil spring, air spring
or leaf spring) and a damping element. The spring and damping coefficients are
chosen according to comfort, road holding, and handling specifications. However,
conventional suspensions can achieve a trade-off between ride comfort and road
holding since their spring and damping coefficients cannot be adaptively tuned
according to driving efforts and road conditions. They can achieve good ride
comfort and road holding only under the designed conditions. To avoid the trade-
off, intelligent suspension systems have been investigated since the 1980s with the
development of microprocessor, sensor, and actuator technologies; their spring and
damping coefficients can be controlled. According to the manner of generation of
the required control forces, intelligent suspension systems can be classified into
two main categories: active and semi-active ones. The active suspension usually
requires a substantial amount of external energy to generate the required control
forces. This is achieved by pneumatic or hydraulic actuators [61], which works in
parallel to a spring and damper. A fully active suspension system can potentially

a
State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing 400044, China
2 Handbook of vehicle suspension control systems

offer better performance than the passive and the semi-active counterparts. How-
ever, the practical applications are very limited due to their high energy con-
sumption, high cost, and complexity. In 1982, Lotus Cars developed several fairly
effective prototypes and introduced them to F1, which was banned later. Nissan
also introduced a low-bandwidth active suspension in circa 1990 as an option that
added an extra 20% to the price of luxury models. A recent fully active suspension
from Bose Corporation adopts the linear electric motor to replace the used
hydraulic or pneumatic actuators.
As comparison, semi-active suspension system can deliver versatility, adapt-
ability and higher performance of active suspension systems with a fraction of the
power consumption, yet keeping reliability comparable to that of passive counter-
parts. In a semi-active suspension the damping or spring coefficients can usually be
tuned in real time. Compared to active suspension systems, the semi-active control
devices cannot input mechanical energy into the controlled system and, therefore,
they do not have chance of the loss of stability (in the bounded input/bounded
output sense). Examples of such devices include air springs and switchable
shock absorbers, dampers with controllable fluids (e.g., electrorheological and
magnetorheological fluids), various self-leveling solutions, as well as systems like
hydro-pneumatic, hydrolastic, and hydragas suspensions. Mitsubishi developed the
world’s first production semi-active electronically controlled suspension system in
passenger cars; the system was first incorporated in the 1987 Galant model. Due to
their simplicity, fast response, and little energy requirement, semi-active suspension
systems based on magnetoreheological technology have received much attention
and begun to step into commercial applications. The systems have been applied on
some Cadillac models (Imaj, Seville, SRX, XLR, STS, DTS), the Chevrolet
Corvette, and most recently the Audi TT, the Audi R8, and Ferrari 599 GTB.
As intelligent suspension systems, active and semi-active suspensions belong to
the controlled suspension system, which consists of sensors, controllers, plants and
actuators. Each of the two suspension systems has different advantages and dis-
advantages. Xue et al. [71] reviewed research and development of automotive active
suspension and investigated comparisons between various vehicle suspensions from
the aspects of structure, weight, cost, ride comfort, handling performance, reliability,
dynamic performance, energy recovery, and commercial maturity. The comparison
results are given in Table 1.1.
The design of an intelligent suspension is actually a control engineering pro-
blem. The success of design of an intelligent suspension for improving the ride
comfort and road holding is determined by two steps like other control systems.
First step is to model an accurate dynamic model of vehicle suspension and the
second is to design and choose a suitable control strategy, which has significant
impact on the trade-off between ride comfort and road holding stability.
Vehicle dynamic modeling is an important step in the design of intelligent sus-
pension system. Generally speaking, the vehicle dynamic model of real vehicle is
some degrees of approximation. To describe approximation is a challenging problem.
In terms of the control design, a relatively simple vehicle model is reasonable.
 State-of-the-art of vehicle intelligent suspension control system 3

Table 1.1 Comparisons between various automotive suspensions [71]

Parameters Passive Semi-active Hydraulic or Electromagnetic

suspensions suspensions pneumatic active active
suspensions suspensions
Structure Simplest Complex Most complex Simple
Weight or volume Lowest Low High Highest
Cost Lowest Low Highest High
Ride comfort Bad Medium Good Best
Handling performance Bad Medium Good Best
Reliability Highest High Medium High
Dynamic performance Passive Passive Medium Good
Energy regeneration No No No Yes
Commercial maturity Yes Yes Yes No

A consequence of this is that the controller cannot achieve better control performance
in real-time control. A complex dynamic model can describe the dynamic behavior
of vehicle. The model may not be suitable to formulate a controller. With the
development of cognitive science and control theory (especially nonlinear theory),
the dynamic model also undergoes some changes in order to pursue better control
performance. Some complex dynamic behaviors are observed and added to the
model of vehicle. A real vehicle suspension system behavior has inherently sig-
nificant nonlinearity, some uncertainties, time delay and even actuator or system
fault. Once these complex factors are considered, the task of improving ride comfort
and road holding has become more challenging than ever. Therefore, the research of
control design in intelligent suspension systems has received much attention.
Like other physical systems, the nature of a real vehicle system is nonlinear.
The nonlinearity mainly comes from the suspension system itself or the actuator.
The nonlinearity of suspension includes nonlinear spring rates or damping coeffi-
cients, the nonlinear friction force between elements, nonlinear geometric con-
straints, etc. The dynamic behavior of actuators is usually nonlinear. For example,
the magneto-rheological (MR) damper (a promising semi-active damper) has sig-
nificant, hysteric nonlinear relation between the damping force and the velocity.
Besides the nonlinearity, a real vehicle suspension system also has uncertain-
ties. The uncertainties include parameter uncertainties and model uncertainties. The
mass of vehicle body sometimes is not constant for a passenger car or a truck.
There can be one or more passengers on the vehicle. The road disturbance may be
determined randomly. The model of lumped mass parameters is not enough to
describe the dynamic behavior of a vehicle.
Since an intelligent suspension system is also a closed-loop system, which has
sensors, controllers, plants, and actuator, time delay of measuring signal or actuator
must exist in the system. In most cases, the time delay can be neglected as it is very
small. However, sometimes the amplitude of the time delay can be comparable to the
control period, which means it cannot be neglected and needs be to carefully dealt
4 Handbook of vehicle suspension control systems

with. For an active suspension, the appearance of time delay will result in the loss
of stability. There is no loss of stability for a semi-active suspension for there is no
energy input. However, the appearance of time delay will significantly degrade the
control performance.
In addition, as a mechanical system, the controlled suspension system will
undergo all kinds of severe conditions in its long run. Inevitably, some elements
such as the sensors, plants, and actuator in the system will not work rightly. If faults
in these elements are not dealt with in time, the system may not assure its desired
performance or even lose its stability.
The mentioned modeling problems have become main obstacle in the
improvement of ride comfort and road holding for an intelligent suspension without
question. However, there is no general effective method to deal with all above-
discussed questions simultaneously. As a consequence, a wide range of control
schemes have been proposed in the literatures as the other step of intelligent sus-
pension design. These control strategies are studied to improve the ride comfort and
road holding while considering one or more aspects of nonlinearity, uncertainties,
time delay, and actuator or system fault.
Thanks to the development of nonlinear control theory, some nonlinear control
strategies are applied to intelligent suspension system such as feedback lineariza-
tion method, sliding mode control, adaptive control, etc. To accommodate the
uncertainties in the vehicle suspension system, some robust control algorithms are
also proposed by many researchers. In order to compensate the effect of time delay
on control performance, some time-delay control algorithms are applied on the
intelligent suspension systems through numerical simulation. With the develop-
ment of fault diagnose and identification techniques, fault-tolerant control algo-
rithms are proposed on the vehicle suspension system. These control strategies
achieve better control performance in the improvement of ride comfort and road
holding. However, there are few references to review the state-of-the-art in the
modeling and control of intelligent suspension to account for nonlinearity, uncer-
tainties, time delay, and actuator or system fault.
Therefore, the thrust of this study is to provide a comprehensive overview of
modeling and control from the aspects of nonlinearity, uncertainties, time delay,
and actuator or system fault. The study is organized as follows. Section 1.2 pro-
vides the evaluation criterion of vehicle suspension performance. Section 1.3
reviews the modeling of vehicle suspension system. The control strategies are
reviewed in Section 1.4. Section 1.5 summaries the method of validation. Finally,
we conclude the chapter in Section 1.6.

1.2 Evaluation criterion of vehicle suspension performance

To evaluate the performance of different intelligent suspension systems, some

evaluation indexes are established to assess the ride comfort and road holding.
They will be briefly reviewed here. For details, Griffin [19] can be referred.
 State-of-the-art of vehicle intelligent suspension control system 5

1.2.1 Ride comfort

The human body has varying sensitivity to vibrations, which is dependent on the
direction and frequency content of vibrations. Generally, the human body has more
sensitivity in the horizontal direction than in the vertical direction. For the vertical
vibration, the most sensitive frequency range is between 4 and 8 Hz, which is
corresponding to the resonant frequencies of the organs in abdominal cavity. The
sensitivity frequency range of horizontal vibration is 1–2 Hz [53]. There are four
methods to evaluate ride comfort objectively in the world at present. They are the
ISO 2631 standard, the British Standard – BS 6841, the VDI 2057, and the average
absorbed power. The ISO 2631 is mainly applied in Europe and the BS 6841 is used
in the United Kingdom. VDI 2057 is adopted by Germany and Austria, while the
average absorbed power is used by the United States of America.
In order to quantify the ride comfort of intelligent suspension systems, vibra-
tions of the vehicle body should be measured in two directions: vertical (i.e., heave)
and horizontal (i.e., roll, pitch, and yaw). The most commonly used measurement
method is the root mean square (RMS) acceleration, which is defined as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u ðT
u
u1 2
RMSðaÞ ¼ t a ðtÞdt ð1:1Þ
T
0

where T is the total sample time, a is the sprung mass acceleration, and t is the time.
To consider the effect of vibration frequency on ride comfort, a weighted form
of accelerations can be adopted.

1.2.2 Road holding

Road holding of a vehicle is important during maneuvers such as cornering, braking
or accelerating. It is related to the tire contact force influenced by two factors:
wheel and vehicle body vibrations. The variation of the contact force between the
tires and road surface corresponds directly to the road holding, which can be used
for the quantification. Generally, lower dynamic tire force means better road
holding, while higher dynamic tire force indicates worse road holding.

1.3 Modeling of vehicle suspension system

The intelligent suspension system is a complex dynamic system featuring with high
order, nonlinearity, uncertainty, time delay, and fault. An appropriate representa-
tion of vehicle body is the basis for the development of control scheme and vehicle
performance assessment. There are two kinds of models. One is the distributed
model and the other is the lumped parameter model. For the control design of
intelligent suspensions, the lumped parameter models are usually employed. The
dynamic model is usually constructed according to the Newton’s law, the theories
of Hamilton and Lagrange, or the method of bond graph. Generally, a vehicle body
6 Handbook of vehicle suspension control systems

has six degrees of freedom (DOF) motions: longitudinal translation, lateral trans-
lation, bounce or heave motion, roll motion, pitch motion, and yaw motion. Of
course, the motion of a vehicle should also contain the vibrations of its wheels.
These motions are coupled with each other. A vehicle consists of the sprung mass
(chassis), unsprung masses (wheels, axles, and linages), and elastic and dissipative
elements.
According to the requirement of control design, the vehicle dynamic model can
be modeled as a quarter-car model (two DOF), a half-car model (four DOF), and
full-vehicle model (seven DOF). The three dynamic models are often used for the
theoretical analysis and design of intelligent suspension systems.
In this section, the road model is first reviewed. Then a linear quarter-car
model, a linear half-car model, and a linear full car model are discussed. At last,
some nonlinear models with nonlinearity, uncertainty, time delay, and actuator or
system fault are also reviewed.

1.3.1 Road model

The modeling of the road profile is important for simulation of intelligent sus-
pension because it is the main excited source. Road inputs of vehicle dynamic
model can be classified into two types: deterministic road and random road. The
determined inputs include sine waves, square waves, step waves, triangular waves,
etc. These road inputs can be described by a mathematic formula. The repre-
sentation of random road is much more complex. The roughness of random road is
typically represented as a stationary Gaussian stochastic process of a given dis-
placement power spectral density (PSD) in m2/(cycle/m) [59, 77]. Zhao and Lu [74]
studied a kind of random road with road PSD.
 w
n
Gxr ðnÞ ¼ Gxr ðn0 Þ ð1:2Þ
n0
Figure 1.1 gives several samples of determined and random road.

1.3.2 Intelligent quarter-car model

The quarter-car model has been the bench model used in the study of control
algorithm for intelligent suspension system. Although the model is very simple and
is considering only vertical vibration motions of the sprung mass and the unsprung
mass, it is very useful in initial development. The model is shown in Figure 1.2 and
consists of a sprung mass, an unsprung mass, a spring, a damper, and an actuator
between the sprung mass and the unsprung mass. The model can simultaneously
represent the passive suspension and active/semi-active according to the state of the
actuator. If the actuator is neglected, the model is a passive suspension. The model
is an active suspension if the actuator can generate active control forces, while the
model is a semi-active suspension if the actuator can provide only damping forces.
The method is also adopted in the modeling of vehicle suspension in the following
sections.
 State-of-the-art of vehicle intelligent suspension control system 7

0.1
0.05 (a) Step
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.2
0.1 (b) Bump
0
Displacement (m)

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0.01
0 (c) Sine of 1.7 Hz
–0.01
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.01
0 (d) Sine of 14 Hz
–0.01
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.1
0.05
0 (e) Random
–0.05
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Figure 1.1 Road input [11]

zs
ms

ks ce Fc
zu
mu

kt
w

Figure 1.2 Quarter car model

According to the Newton’s law, the governing equations of the quarter-car

model can be given as
ms z€s þ ks ðzs  zu Þ þ ce ðz_s  z_u Þ þ Fc ¼ 0 ð1:3Þ
mu z€u  ce ðz_s  z_u Þ  Fc þ kt ðzu  wÞ ¼ 0 ð1:4Þ

where ms and mu are the sprung mass and unsprung mass; ce and ks denote the
damping and elastic coefficients; Fc is the variable damping force; kt is stiffness of
wheel; zs , zu , and w are the displacement of sprung mass, the displacement of
unsprung mass, and the road input, respectively.
8 Handbook of vehicle suspension control systems

1.3.3 Intelligent half-car model

To account for the pitch motion or roll motion, a half-car model is adopted by many
researchers, which is shown in Figure 1.3. The model is considering the vertical
vibration and pitch motions of vehicle body, and the vertical motion of front and
rear wheels.
The dynamic equations of this model can be written as

ms Z€s ¼ ðCsf þ Csr Þz_s þ ðCsf a  Csr bÞq_  Ksf ðzsf  zuf Þ
þ Csf zuf_  Ksr ðzsr  zur Þ þ Csr zur
_ þ Ffc þ Frc ð1:5Þ

Iy €q ¼ ðaCsf  bCsr Þ_z  ðCsf a2 þ Csr b2 Þq_ þ aKsf ðzsf  zuf Þ

 aCsf zuf_  bKsr ðzsr  zur Þ  bCsr zur
_  aFfc þ bFrc ð1:6Þ

muf z€uf ¼ Csf z_  aCsf q_ þ Ksf ðzsf  zuf Þ  Csf zuf_  Ktf ðZuf  Zrf Þ  Ffc
ð1:7Þ

mur z€ur ¼ Csr z_ þ bCsrq_ þ Ksr ðzsr  zur Þ  Csr zur

_  Ktr ðZur  Zrr Þ  Frc ð1:8Þ

in which, ms , muf , and mur are the vehicle body mass, the front unsprung mass, and
the rear unsprung mass, respectively; Iy is the pitch inertia of vehicle body; Ksf , Ksr ,
Ktf , are Ktr are the stiffness of front suspension spring, the stiffness of rear sus-
pension spring, the front tire stiffness, and the rear-tire stiffness, respectively; Csf
and Csr are the passive damping coefficients of front damper and the passive
damping coefficient of rear damper, respectively; a is the distance between the
center of vehicle body and the front wheel axis; b is the distance between the center
of vehicle body and the rear wheel axis; zs is vertical displacement of vehicle body;
q is the pitch angle of vehicle body; zuf is the vertical displacement of front
unsprung mass; zur is the vertical displacement of rear unsprung mass; zrf denotes
the road input of front wheel; zrr denotes the road input of rear wheel; Ffc is the
control force of front actuator; Frc is the control force of rear actuator;
zsf ¼ z  a sin q  z  aq, zsr ¼ z þ b sin q  z þ bq.

zs
q Iy
b a

Frc Ffc
Ksr Csr Ksf Csf

mur zur muf zuf

Ktr Ktf
zrr zrf

Figure 1.3 Intelligent half-car pitch model

 State-of-the-art of vehicle intelligent suspension control system 9

zs
j Ixx
b a

Flc Hc Frc
Ksl C Ksr Csr
sl

mul zul muf zur

Ktl Ktr
zrl zrr

Figure 1.4 Intelligent half-car roll model

To study the roll motion of vehicle body, a roll half-car model is sometimes
investigated, which is given in Figure 1.4.
The dynamic equations of this model can be given as

ms Z€s ¼ ðCsl þ Csr Þz_s þ ðCsl a  Csr bÞj_  Ksl ðzsl  zul Þ þ Csl z_ul
 Ksr ðzsr  zur Þ þ Csr zur
_ þ Flc þ Frc ð1:9Þ

€ ¼ ðaCsl  bCsr Þ_z  ðCsl a2 þ Csr b2 Þq_ þ aKsl ðzsl  zul Þ  aCsl z_ul
Ixx j
 
Hc
 bKsr ðzsr  zur Þ  bCsr zur
_  aFlc þ bFrc þ ms g sin q þ Mx
cos q
ð1:10Þ

mul z€ul ¼ Csl z_  aCsl j_ þ Ksl ðzsl  zul Þ  Csl zul

_  Ktl ðZul  Zrl Þ  Flc ð1:11Þ

mur z€ur ¼ Csr z_ þ bCsr j_ þ Ksr ðzsr  zur Þ  Csr zur

_  Ktr ðZur  Zrr Þ  Frc ð1:12Þ

where ms , mul , and mur are the vehicle body mass, the left unsprung mass, and the
right unsprung mass, respectively; Ixx is the roll inertia of vehicle body; Ksl , Ksr , Ktl ,
and Ktr are the stiffness of left suspension spring, the stiffness of right suspension
spring, the left-tire stiffness, and the right-tire stiffness, respectively; Csl and Csr are
the passive damping coefficients of left damper and the passive damping coeffi-
cient of right damper, respectively; a is the distance between the center of vehicle
body and the right track; b is the distance between the center of vehicle body and
the left track; zs is vertical displacement of vehicle body; j is the roll angle of
vehicle body; zul is the vertical displacement of left unsprung mass; zur is the ver-
tical displacement of right unsprung mass; zrl denotes the road input of left
wheel; zrr denotes the road input of right wheel; Flc is the control force of left
actuator; Frc is the control force of right actuator; zsl ¼ z  a sin j  z  aj,
zsr ¼ z þ b sin j  z þ bj.
10 Handbook of vehicle suspension control systems

1.3.4 Intelligent full-vehicle model

A full car model with four actuators is shown in Figure 1.5. The vehicle body itself
is assumed to be rigid and has seven DOF in heave, pitch, and roll directions. The
governing equation of the full vehicle is derived as follows:

ms€z ¼ Ffl þ Ffr þ Frl þ Frr ð1:13Þ

"  2 #  
Hc €q ¼ w Ffl  w Ffr þ w Frl  w Frr þ ms g Hc sin q þ Mx
Ixx þ ms
cos q 2 2 2 2 cos q
ð1:14Þ

Iyy j
€ ¼ aFfl þ aFfr  bFrl  bFrr ð1:15Þ

mufl z€ufl ¼ Ffl  ktfl ðzufl  zrfl Þ ð1:16Þ

mufr zufr
€ ¼ Ffr  ktfr ðzufr  zrfr Þ ð1:17Þ

murl zurl
€ ¼ Frl  ktrl ðzurl  zrrl Þ ð1:18Þ

murr zurr
€ ¼ Frr  ktrr ðzurr  zrrr Þ ð1:19Þ

where

Ffl ¼ kfl ðzfl  zufl Þ  cfl ðz_fl  zufl

_ Þ þ Fdfl

Ffr ¼ kfr ðzfr  zufr Þ  cfr ðz_fr  zufr

_ Þ þ Fdfr

Frl ¼ krl ðzrl  zurl Þ  crl ðz_rl  zurl

_ Þ þ Fdrl

Frr ¼ krr ðzrr  zurr Þ  crr ðz_rr  zurr

_ Þ þ Fdrr

b zrr
a crr Fdrr
krr
zfr w z zurr
mur
kfr cfr Fdfr ms ktrr
zufr x Ixx C.O.G
muf θ
ktfr w Fdrl
zrfr Hc
Iyy crl
y
Fdfl mur
cfl
ktrl
muf
ktfl

Figure 1.5 Mechanical model of the MR suspension system

 State-of-the-art of vehicle intelligent suspension control system 11

and
zfl ¼ z  w sin q  a sin j  z  wq  aj

zfr ¼ z þ w sin q  a sin j  z þ wq  aj

zrl ¼ z  w sin q þ a sin j  z  wq þ aj

zrl ¼ z þ w sin q þ a sin j  z þ wq þ aj

in which
Ffl ¼ kfl ðzfl  zufl Þ  cfl ðz_fl  zufl
_ Þ þ Fdfl

Ffr ¼ kfr ðzfr  zufr Þ  cfr ðz_fr  zufr

_ Þ þ Fdfr

Frl ¼ krl ðzrl  zurl Þ  crl ðz_rl  zurl

_ Þ þ Fdrl

Frr ¼ krr ðzrr  zurr Þ  crr ðz_rr  zurr

_ Þ þ Fdrr

zfl ¼ z  w sin q  a sin j  z  wq  aj

zfr ¼ z þ w sin q  a sin j  z þ wq  aj

zrl ¼ z  w sin q þ a sin j  z  wq þ aj

zrl ¼ z þ w sin q þ a sin j  z þ wq þ aj

Thus, the control damping force vector can be described as

_ Þ þ Fdfl ,  cfr ðz_fr  zufr
u ¼ ½cfl ðz_fl  zufl _ Þ þ Fdfr ,  crl ðz_rl  zurl
_ Þ þ Fdrl ,
T
 crr ðz_rr  zurr
_ Þ þ Fdrr 

which is determined by the control strategies.

1.3.5 Other nonlinear vehicle dynamic model

Intelligent suspension systems are inherently nonlinear. There are nonlinearities of
Coulomb friction, saturation, backlash, and hysteresis in a vehicle suspension
system. These nonlinearities usually cause undesirable behavior of the controlled
system, such as instabilities, limit cycles, bifurcation, chaos, etc. [65]. Therefore,
before formulating a nonlinear control method, a more accurate dynamic model
needs to be developed. Some researchers have begun to analyze the nonlinearities
in vehicle suspension system and some nonlinear characteristics are added to above
linear equations. McGee et al. [40] studied the nonlinearities of a quarter-car model
by using a simple technique. The technique is applied to actual laboratory shaker
forced response and vehicle road data. They thought the main sources of non-
linearities for a vehicle suspension system included quadratic and cubic stiffness
nonlinearities and Coulomb friction. Gao et al. studied a hydropneumatic slow
12 Handbook of vehicle suspension control systems

Suspension deflection Pressure

Controller
Body acceleration
zb
1

4 Gas strut
[[

6 Damper
3 7 7
valve

Damper
[[

valve
5
zr a
2 Current w9
zr s r Quarter car: Hydraulic system:
~ 1. Sprung mass (body) 9. Flow control valve
2. Unsprung mass (wheel) 10. Main accumulator
10 3. Coil spring 11. Motor
12 13 4. Top mount 12. Pump
11 M w 5. Bottom compliance 13. Relief valve
6. Piston-cylinder 14. Tank
14 7. Damper valve
8. Gas spring

Figure 1.6 Nonlinear quarter-car model [40]

active suspension. A nonlinear quarter-car mode was developed, which included a

gas strut model and a nonlinear dynamic flow control valve model. The model is
shown in Figure 1.6. The main benefit of this model is that it is considering the
interaction of the vehicle and hydraulic components and can overcome a major
shortfall of many vehicle suspension modeling approaches that simplify and
sometimes totally exclude the hydraulic elements.
Demir et al. [9] studied a five-DOF vehicle model. Besides the bounce motion
of the vehicle body, the pitch motion of the vehicle body, the vertical motion of
front and rear wheels, and the bounce motion of passenger seat are also considered.
The model is shown in Figure 1.7. The damping force and restoring force for the
front and rear suspension include linear nonlinear force of suspension systems.
They are defined as follows:

fsj ¼ fsjl þ fsjn ð1:20Þ

where j ¼ f or r. The f or r denotes the front or rear suspension system, respectively.

l and n stand for linear and nonlinear components of suspension forces, respec-
tively. The linear components of suspension system forces consist of linear spring
and damping forces in the following form:
_
fsjl ¼ ksj ½ðxb  xwj Þ  Lj q þ csj ½ðx_ b  x_ wj ÞLj q ð j ¼ f ; rÞ ð1:21Þ

where ksj is the linear spring coefficient, and csj is the linear damping coefficient.
 State-of-the-art of vehicle intelligent suspension control system 13

mp xp
Passenger seat

kp cp
Lp V

mb Xb
Vehicle body θ

Lr Lf

U U
ksr fsrn csr f ksf fsfn csf f

mwr xwr mwf xwf

Rear wheel Front wheel
n
fwr fwfn k
k wr wf
xrr xrf

Figure 1.7 Nonlinear half-vehicle model [9]

zs
b s
a ms
q
f
csf ksr csr
ksf
zufr zurl
muf ksf csf mur
zufl
muf kusr cusr
kusf cusf
kusf cusf zrl
zfr
zfl

Figure 1.8 Nonlinear full-vehicle model [76]

The nonlinear component of the suspension forces can be written as

fsjn ¼ ksjn ½ðxb  xwj Þ  Lj q3 þ m:fsuspj sign½ðx_ b  x_ wj Þ  Lj q

_ ð1:22Þ

where ksjn is the nonlinear spring coefficient, m is the Coulomb friction constant, and
the parameter fsuspj is set to be a fraction of the sprung mass weight.
Zhu and Ishitobi [76] developed a nonlinear seven-DOF vehicle model to
investigate the chaotic response, which is shown in Figure 1.8. The suspensions
between the sprung mass and unsprung masses are modeled as nonlinear spring and
nonlinear damper elements, while the tires are modeled as nonlinear springs with
viscous damping.
14 Handbook of vehicle suspension control systems

The modeled nonlinear spring of suspension has following characteristics:

Fsij ¼ ksij sgnðDsij ÞjDsij jnsij ði ¼ f ; r; j ¼ l; rÞ ð1:23Þ
where Fsij is the spring dynamic force, ksij is the equivalent stiffness, Dsij is the
deformation of the spring, and sgn() is the signum function.
The nonlinear damping forces of front and rear suspensions are given by

Fcij ¼ csi D_ uij ði ¼ f ; r; j ¼ l; rÞ ð1:24Þ

where the subscript c indicates damping of suspension. Fcij is the damping force
and D_ uij is the relative velocity between the extremes of the damper. The damping
coefficient csi is expressed by

csui D_ uij  0
csi ¼ ði ¼ f ; r; j ¼ l; rÞ ð1:25Þ
csdi D_ uij < 0
where csui and csdi are damping coefficients for tension and compression,
respectively.
The tire of the vehicle is also modeled by a nonlinear spring
Fusij ¼ kusi sgnðDusij ÞjDusij jnusi ði ¼ f ; r; j ¼ l; rÞ ð1:26Þ
where Fusij is the spring force, kusi is the equivalent stiffness, Dusij is the deforma-
tion, and nusi is the nonlinear coefficient of the tire spring.
The damping of the tires is assumed to be viscous, thus the damping force is
calculated as

Fucij ¼ cusi D_ usij ð1:27Þ

where cusi is the viscous damping coefficient, and D_ usij is the relative velocity of
extremes of the model of tires.

1.3.6 Nonlinear multibody dynamic model

With the development of computer technology, some more accurate and complicated
models are also developed to design intelligent suspension systems by using multi-
body techniques. One obvious advantage of these models is that the formulated
control strategy can directly be applied to real-time control without much effort of
tuning control parameters. These models, however, despite their sophistication, have
two drawbacks: parameters determination and long computation time. Bommer [4]
developed a nonlinear complete-car model containing a flexible trimmed body based
on measurement data and the engine and the suspension in Automatic Dynamic
Analysis of Mechanical Systems (ADAMS)/View, which is shown in Fig. 1.9. The
validation of the model is performed using measurements in the frequency domain. It
shows that this single nonlinear model is able to predict the vehicle’s dynamic
behavior for different excitation levels at the wheel, regardless whether the damper is
in the slip-phase or in the stick-phase. The results also show that the Coulomb friction
in the shock-absorber has significant influence on the vehicle’s dynamic behavior.
 State-of-the-art of vehicle intelligent suspension control system 15

z
y x

Figure 1.9 The complete-car model: including the flexible model of the trimmed
body and the rigid bodies for the suspension (left and right) and the
engine [4]

1.3.7 Nonlinear uncertainty modeling

Dealing with and understanding the effect of uncertainty are important tasks for
intelligent suspension control design. Reducing the effects of some forms of
uncertainty without catastrophically increasing the effects of other dominant forms
is the primary job of uncertainty control of intelligent suspension systems.
Some uncertainties of modeling technologies were adopted by many
researchers to formulate control strategy. Goh et al. [18] studied a case of a
MacPherson strut automotive suspension analysis and evaluated the uncertainties in
the modeling of this complex dynamic problem using a simplified analytic model
and a complex computational model. The probabilistic design methods were used
to characterize the variability in design variables. The sprung mass variation was
considered in the modeling of nonlinear uncertainty electrohydraulic active sus-
pension [16]. Agamennoni et al. [49] developed the nonlinear identification of a
nonlinear model as well as the uncertainty bounds of a magnetic suspension. They
proposed a cascade of a global linear fuzzy dynamic block followed by a piecewise
linear function. This model structure allowed a proper identification of the system
dynamic and a tight description of the uncertainties.
In Reference 43, the uncertainty is modeled as additive and multiplicative. To
account for the discrepancies between measurement data and analytical nominal
model for a large-gap magnetic suspension test bed [36], an uncertainty model is
developed through the method of minimum norm model validation. The computed
results show that uncertainty bounds, corresponding to a combination of additive
16 Handbook of vehicle suspension control systems

and diagonal input multiplicative uncertainty, can be directly obtained by calcu-

lating the smallest norm of the difference between the measured and the norm
model response.

1.3.8 Nonlinear dynamic model with time delay

While formulating a control algorithm, it is usually assumed that there is no time
delay for the state measurement and the response of the actuator. This is never
true in reality and there always exists a time delay for the state measurement and
the actuator action, such as pneumatic and hydraulic active suspension systems.
The total time delay generally includes the fixed time delay (caused by online data
acquisition, filtering, calculating control forces, and transmitting the control force
signals from the computer to the actuator) and the time delay of actuators. The
former can be reduced by using an advanced processing system and will usually be
neglected in the development of control algorithms. The latter usually depends on
the particular dynamics of actuators. For common electrohydraulic actuators, the
time delay will reach more than tens of milliseconds. The time delay may degrade
control performance and even render the instability of the controlled system. It is
not easy to consider the effect of time delay on the control scheme and to define
control forces in terms of delayed state vector. The introduction of time delay
converts the system differential equations into significantly nonlinear equations.
Before formulating the control strategy, a model with a time delay needs to be
developed.
Du and Zhang [14] developed a four-DOF half-car suspension model with the
time delay, which is shown in Figure 1.10.
The equations for this model [14] can be represented as

ms€z c ðtÞ þ ksf ½zsf ðtÞ  zuf ðtÞ þ csf ½_z sf ðtÞ  z_ uf ðtÞ þ ksr ½zsr ðtÞ  zur ðtÞ
þ csr ½_z sr ðtÞ  z_ ur ðtÞ ¼ uf ðt  tÞ þ ur ðt  tÞ ð1:28Þ

€ ðtÞ  l1 ksf ½zsf ðtÞ  zuf ðtÞ  l1 csf ½_z sf ðtÞ  z_ uf ðtÞ þ l2 ksr ½zsr ðtÞ  zur ðtÞ
Ij j
þ l2 csr ½_z sr ðtÞ  z_ ur ðtÞ ¼ l1 uf ðt  tÞ þ l2 ur ðt  tÞ ð1:29Þ

muf €z uf ðtÞ  ksf ½zsf ðtÞ  zuf ðtÞ  csf ½_z sf ðtÞ  z_ uf ðtÞ þ ktf ½zuf ðtÞ  zrf ðtÞ
¼ uf ðt  tÞ ð1:30Þ

mur€z ur ðtÞ  ksr ½zsr ðtÞ  zur ðtÞ  csr ½_z sr ðtÞ  z_ ur ðtÞ þ ktr ½zur ðtÞ  zrr ðtÞ
¼ ur ðt  tÞ ð1:31Þ

where ms is the mass of the car body; muf and mur are the unsprung masses on the
front and rear wheels, respectively; Ij is the pitch moment of inertia about the center
of mass; zuf (t) and zur(t) are the front and rear unsprung mass displacements,
respectively; zrf (t) and zrr(t) are the front and rear terrain height displacements,
respectively; ksf and ksr are the stiffness of the passive elements of the front and
rear wheels, respectively, and csf and csr are the damping coefficients of passive
 State-of-the-art of vehicle intelligent suspension control system 17

zc
l1 l2

ms, Iφ φ

zsf zsr

uf (t – τ) ur (t – τ)
ksf csf ksr csr

zuf zur
muf mur

ktf ktr

zrf zrr

Figure 1.10 Four-degrees-of-freedom half-car suspension model with a time

delay [14]

suspension elements for the front and rear wheels, respectively. Similarly, ktf and ktr
are the front- and rear-tire stiffness, respectively; uf (t) and ur(t) are the front and rear
actuator force inputs, respectively; and t is the actuator time delay.

1.3.9 Nonlinear dynamics model with fault

Intelligent suspension systems rely on sophisticated control to meet increased ride
comfort and road-holding requirements. A conventional control design for intelli-
gent suspension systems may result in an unsatisfactory performance, or even
instability, in the event of malfunctions in actuators, sensors, or other system
components, for example, Shi et al. [54]; Yang et al. [72]. Before formulating a
control strategy, a dynamic model with fault element needs to be developed. In our
prior work [10], we studied the output damping of the MR damper is reduced due to
the increase in temperature. To model the fault of MR dampers, four control
effective factors gi (i ¼ fl, fr, rl, rr) are adopted to characterize the degree of
damping force loss. gi ¼ 0 denotes the health status of the ith MR damper, and
gi ¼ 1 corresponds to the total failure of the ith MR damper. Obviously, 0 < gi < 1
represents the partial loss in the control effectiveness.
Considering the partial fault of MR dampers, the state space equation can be
written as

x_ ¼ Ax þ Buf þ Lw ð1:32Þ
18 Handbook of vehicle suspension control systems

where uf denotes the control damping force vector with partial fault of four MR
dampers and can be described as
 T
uf ¼ ð1  gfl Þufl , ð1  gfr Þufr , ð1  grl Þurl , ð1  grr Þurr ð1:33Þ
in which um (m ¼ fl, fr, rl, rr) denotes the damping output of the healthy MR damper.
Substituting (1.33) into (1.32) gets

x_ ¼ Ax þ Bf u þ Lw ð1:34Þ
where
2 3
gfl 0 0 0
60 gfr 0 0 7
Bf ¼ BðI  G4 Þ, G4 ¼ 6
40
7
0 grl 0 5
0 0 0 grr
To estimate the control effectiveness factors of MR dampers, (1.34) can also be
written as follows:
2 3
ufl
 6 ufr 7
x_ ¼ Ax þ Bu  b1 gfl b2 gfr b3 grl b4 grr 6 7
4 url 5 þ Lw ð1:35Þ
urr
where bi is the ith column vector of the matrix B.
Equation (1.35) can further be written as
x_ ¼ Ax þ Bu þ DðuÞg þ Lw ð1:36Þ

in which DðuÞ is defined by DðuÞ ¼ BU and

2 3 2 3
ufl 0 0 0 gfl
6 0 u 0 0 7 6 gfr 7
U ¼6 4 0
fr 7; g ¼ 6 7
0 url 0 5 4 grl 5
0 0 0 urr grr

Due to the difficulty in acquiring the true values of the loss in the control
effectiveness factor g of the MR dampers, the control effective factors can be
modeled as a random bias vector with a zero-mean white Gaussian noise sequence.

1.3.10 The modeling of actuator

The key part of an intelligent suspension is the controllable actuator. The precision
of actuator model is crucial for suspension analysis and design. Therefore, how to
build a more accurate model for an active or semi-active actuator is an important
issue. However, it is difficult to get the precise model of the actuator since the
actuator characteristic usually exhibits strong nonlinearity in many practical
applications. The nonlinearity must be considered during simulation to obtain
 State-of-the-art of vehicle intelligent suspension control system 19

realistic results. Due to big characteristic difference for different actuators, it is

difficult to describe the nonlinearity using a general method. In this study, two
types of the most promising actuators are reviewed. One is the electromagnetic
actuator and the other is the MR damper.

1.3.10.1 Model of electromagnetic actuators

Unlike other actuators, electromagnetic actuators are a kind of energy-regenerative
active actuators. It was first studied by Okada and Harada Harada. They used a
double-voltage charging circuit to regenerate electrical energy during high-speed
motion of the actuator. One year later, Suda and Shiiba [57] proposed a regen-
erative suspension. They used a rotary d.c. motor as actuator. The linear motion of
suspension was realized by the mechanism of the rack and pinion. Nakano et al.
[45] proposed a self-powered actuator, which generated power when the speed of
the armature was high and regenerated power would be utilized when the speed of
the armature was low. Martins et al. [39] investigated the prototype of a permanent
magnet’s linear actuator and studied the dynamic performances of the actuator in
low-frequency and high-frequency excitation. The Bose company [26] produced
linear electromagnetic motor actuators and mounted them on a Lexus sedan for
experiments on a real roadway. The results showed good vibration isolation and
body attitude control.
Huang et al. established a mathematical model for the developed electro-
magnetic suspension actuator based on measured parameters and test data.
The developed electromagnetic actuator comprises a brushless d.c. motor and a
ball–screw mechanism as shown in Figure 1.11.

1 2 3 4 5 6 7 8 9 10 11

1. I/O connectors 2. Position sensors 3. Stator core 4. Permanent magnet 5. Rotor axle
6. Recovery buffer 7. Armature winding 8. Ball nut 9. Screw 10. Compression buffer 11. Hanger

Figure 1.11 Electromagnetic actuator prototype

20 Handbook of vehicle suspension control systems

The permanent magnet is fixed on the rotor axle. During the vertical stroke of
the vehicle suspension system, the screw makes a translational movement and
drives the ball nut as well as the rotor axle to rotate. Vibration energy from uneven
road input is converted into electrical energy.
The circuit equations and electromagnetic torque of the brushless d.c. motor
can be expressed as
2 3 2 32 3 2 3 2 3 2 3
ua R 0 0 ia LM 0 0 ia ea
6 7 6 76 7 6 7 6 7 6 7
4 ub 5 ¼ 4 0 R 0 54 ib 5 þ 4 0 LM 0 5P 4 i b 5 þ 4 e b 5
uc 0 0 R ic 0 0 LM ic ec
ð1:37Þ

ea ia þ eb ib þ ec ic
Te ¼ ð1:38Þ
w

where u, i, and e are phase voltage, current, and back EMF, respectively; R is phase
resistance; L is phase self-inductance; M is phase mutual inductance; P is a deri-
vative operator; w is rotor speed; q is rotor angular position; and Te is electro-
magnetic torque. The subscripts a, b, and c represent different phases.
The generated electromagnetic torque is converted into vertical force by the
ball–screw mechanism and the rotational motion of the mechanism is converted
into the vertical stroke of the actuator, which can be expressed as
cot j
Fi ¼ Te ð1:39Þ
r
2p
w¼ v ð1:40Þ
l
where r is the effective radius for force conversion; j is the screw lead angle; l is
the lead of the ball–screw; and v is the vertical stroke speed.

1.3.10.2 Model of MR damper

As promising semi-active actuators, MR dampers have been extensively investi-
gated by many researchers [63] in the past two decades. The dampers feature fast
response characteristic to magnetic field and hence wide control bandwidth and
compact size. The applications of MR dampers have been found over the range
from civil structures such as buildings and bridges to automobiles and railway
vehicles.
In our prior work, we designed four telescopic dampers for a vehicle suspen-
sion, which is schematically shown in Figure 1.12. The MR damper has the same
appearance as that of the passive damper.
Due to the inherent hysteresis nonlinear nature of MR dampers, one of the
challenges in application of these devices to achieve high performance is to
develop an accurate model. [65] analyzed the state-of-the-art parametric dynamic
modeling, identification, and validation techniques for MR dampers.
 State-of-the-art of vehicle intelligent suspension control system 21

1 2 3 4 5 6

1. Piston Pole 2. Outer cylinder 3. Inner cylinder 4. Piston 5. Coil 6. MR fluid

(a)

(b)

Figure 1.12 The MR dampers of a passenger vehicle. (a) Schematic

configuration. (b) Photograph of four MR dampers

According to the methods to model the hysteresis [63], the parametric dynamic
models for MR dampers can be categorized into the Bingham model-based
dynamic models [56], the biviscous models [69], the viscoelastic–plastic models
[35], the stiffness–viscosity-elasto-slide models [68], the Bouc–Wen hysteresis
operator-based dynamic models [56], the Dahl hysteresis operator-based dynamic
models [23], the LuGre hysteresis operator-based dynamic models [25], the
hyperbolic tangent function-based models [30], the sigmoid function-based models
[64], the equivalent models [47], and the phase transition models [66].
Among all studied models, the polynomial model method is the easiest to
implement. Therefore, the polynomial model is adopted in our prior work, which
was firstly proposed by Choi et al. [6], which provided a convenient and effective
choice to calculate the desirable damping force in an open-loop control system. In
this model, the hysteresis loop is divided into two regions: positive acceleration
(lower loop) and negative acceleration (upper loop), which can be fitted by the
22 Handbook of vehicle suspension control systems

polynomial with the power of piston velocity. Thus, the damping force of the MR
damper can be written as
X
9
fMR ¼ ai vi i ¼ 0, 1 . . . , 9 ð1:41Þ
i¼0

The coefficient ai can be linearly approximated with respect to the input current as
follows: ai ¼ bi þ ci I, i ¼ 0, 1, . . . , 9.
Therefore, the damping force can be expressed by
X
9
fMR ¼ ðbi þ ci IÞvi ð1:42Þ
i¼0

where the coefficients bi and ci are obtained from the fitness of experimental data.
To verify the obtained polynomial model, the measurement and simulation
under five operating conditions are compared in Figure 1.13 with excitation fre-
quency 3.8197 Hz and amplitude 25 mm. From this figure, it can be seen that the
model of MR damper can accurately predict the behavior of the MR damper. Once
the piston velocity and the desired force determined by the control strategy are
known, the control current can be calculated:
P
fMR  9i¼0 bi vi
I¼ P9 ð1:43Þ
i¼0 ci v
i

where I is the MR damper input current and fMR is the desired damping force
determined by the later designed control strategy.

1.4 Control strategies

The control design is an important task for intelligent suspension design to achieve
good ride comfort and road holding. Once the actuator is determined, the control
performance of intelligent suspension system will significantly depend on the
choice of control strategy. Since the birth of the first intelligent suspension system,
all kinds of control strategies in control engineering are continuously applied to
intelligent suspension.
The first examination of semi-active control, skyhook control, was proposed
by Karnopp and coworkers [7]. They proposed a ‘‘skyhook’’ damper control
algorithm for a vehicle suspension system and demonstrated that this system
offered improved performance over a passive system when applied to a single-
degree-of-freedom system. To overcome the weakness of skyhook control in
depressing the vibration of the unsprung mass of a suspension, other researchers
investigated the groundhook control and hybrid control [1, 2]. Their results show
that skyhook control is much more effective in improving the ride comfort, the
groundhook control is effective in achieving better road-holding ability and
improving vehicle stability, and the hybrid control is a trade-off between the
 State-of-the-art of vehicle intelligent suspension control system 23

1.5
(a)

Damping force (kN) 0.5

Experiment
Simulated
0
I = 0.0 A
I = 0.4 A
–0.5 I = 0.8 A

I = 2.0 A
–1.1
–2.5 –2 –1 0 1 2 2.5
Displacement (cm)

(b) 1.5
Experiment
1 Simulated
Damping force (kN)

0.5

0 I = 0.0 A
I = 0.4 A
–0.5 I = 0.8 A

I = 2.0 A
–1.1
–0.6 –0.4 –0.2 0 0.2 0.4 0.6
Velocity (m/s)

Figure 1.13 Comparison of polynomial model and experimental results (3.8197 Hz,
25 mm): (a) force vs. displacement; (b) force vs. velocity

skyhook control and the groundhook control. Besides the three well-known semi-
active control strategies, some linear feedback control methods were also applied in
vibration control of a suspension.
The three classic semi-active control algorithms discussed above do not con-
sider the actual dynamic model of vehicle. However, the control performance is
very limited and some modern control algorithms are applied to the intelligent
suspension system.
The modern optimal control theory is very effective for linear system. If the
vehicle suspension model can be approximately simplified as a linear model, some
optimal control algorithms can be applied and achieve good control performance.
24 Handbook of vehicle suspension control systems

These control algorithms are mainly researched through simulation. Because of

their high sensor cost and linear simplification, the application to actual physical
system is very limited.
In recent years, with the development of nonlinear theory and control engi-
neering, some new physical phenomena are observed and added to the modeling
and control design. In reality, there are significant nonlinearity, uncertainty, time
delay, and fault in an intelligent suspension system. In this field, many literatures
have emerged prominently. It is necessary to review the advancements.
In the following subsection, the linear control methods are briefly reviewed
first. Then the control algorithms for nonlinearity, uncertainty, time delay and fault
will be summarized in turn.

1.4.1 Linear control strategies

Feedback control radically changes the dynamics of a system in terms of natural
frequencies, transient response, and stability. If the dynamic model of intelligent
suspension is linearly developed, some classical linear feedback control strategies
can naturally be applied to intelligent suspension systems. Many linear control
strategies are studied in literatures. Wang et al. [64] proposed a Proportional Inte-
gral (PI) control scheme for vibration attenuation of a quarter-vehicle model sub-
ject to idealized harmonic and transient base excitations. Davis and Thompson [8]
applied the linear optimal control to an active automobile suspension system by
inclusion of an integral constraint in the performance index. Full state feedback was
achieved by reconstructing the state variables from easily measured quantities. Two
main problems will be encountered in applying linear optimal control theory to the
control design of controlled suspension. The first problem is that some feedback
states are available, which are required from the reconstruction of an estimator
(filter). Another problem is that there is no zero steady-state axle to body deflection
in response to external body force due to payload variation, braking, accelerating,
or corning forces. Therefore, on the basis of singular value inequalities and optimal
control theory, a linear quadratic Gaussian controller was proposed by Mohamed
et al. [42] to a quarter-car active suspension system.
Considering the state or input constraints of controlled suspension system,
Paschedag [50] studied three different techniques (explicit model predictive control
(eMPC), optimal gain switching (OGS), and discontinuous variable structure con-
trol) and compared them under several aspects, namely range of applicability,
computational complexity, and performance.

1.4.2 Nonlinear control strategies

Controlled intelligent suspension systems are inherently nonlinear. The inherent
nonlinearities include Coulomb friction, saturation, backlash, and hysteresis. These
nonlinearities frequently cause undesirable behavior of the controlled system, such
as instabilities, limit cycles, bifurcation, and chaos [65]. Therefore, nonlinear
control methods must be developed to predict the system’s performance.
To pursue the better ride comfort and road holding of controlled nonlinear
suspension system, many nonlinear control methods are studied. These control
 State-of-the-art of vehicle intelligent suspension control system 25

methods include feedback linearization method, sliding mode control, adaptive

control, etc. It leads to very fruitful results. For example, Joo et al. [27] analyzed a
nonlinear feedback linearization control of electromagnetic suspension system
using Digital Signal Processor (DSP). Some tests were conducted for experimental
comparison of the feedback linearization control with the classical control
approach. The experimental results demonstrate that the feedback linearization
controller shows better performance than that of the classical state-feedback con-
troller using the linearization of small perturbation analysis method and it is robust
with respect to parameter variations and constant load disturbance. The feedback
linearization technique is also used to design stabilization control and tracking
control of the Electro-magnetic Suspension (EMS) system [29].
Wang et al. [65] studied a quarter-car model with hysteresis nonlinearity. The
behavior of the suspension system under road multifrequency excitations reveals
the existence of chaos in the system. The chaotic behavior is controlled by non-
linear feedback.
Litak et al. [37] used the Melnikov criterion to examine a global homoclinic
bifurcation and transition to chaos in the case of a quarter-car model excited
kinematically by the road surface profile. They found the critical Melnikov
amplitude of the road surface profile by introducing a harmonic excitation term and
damping as perturbations.
Hac and Fratini [20] found a vehicle with an active suspension using the
skyhook damping control law becomes unstable, while a vehicle with a semi-active
suspension exhibits limit cycles when road inputs have significant repetitive com-
ponent, when there are large control gains in the suspension control algorithm, and
when the velocities of the body are estimated by filtering the measured suspension
deflections. The research indicates although the semi-active suspension does not
supply energy to the system, limit cycles may develop when persistent excitations
from the road irregularities are present. They proposed an adaptive control scheme
to prevent the limit cycles.
Sims [55] studied limit cycle behavior of smart fluid dampers under closed-
loop control. In traditional opinion, the dissipative nature of semi-active dampers
ensures they always remain stable under closed-loop control, unlike their fully
active counterparts. Sims’ work demonstrates that such semi-active systems can
still produce potentially undesirable closed-loop behavior, which needs to be con-
sidered during controller design.

1.4.3 Uncertainty control methods

To accommodate the uncertainty of intelligent suspension system in simulation and
real test, a robust controller is usually formulated. For instance, based on fuzzy
control design approach, Du and Zhang [16] modeled the T–S fuzzy model of
nonlinear uncertainty electrohydraulic suspension. A fuzzy state-feedback con-
troller was designed for the obtained Takagi–Sugeno (T–S) fuzzy model with
optimized H? performance for ride comfort by using the parallel distributed
compensation (PDC) scheme. The numerical results showed that the designed
controller can achieve good suspension performance despite the existence of non-
linear actuator dynamics, sprung mass variation, and control input constraints.
26 Handbook of vehicle suspension control systems

Gao et al. [17] studied the problem of input-delayed robust H? state-feedback

controller design for a class of semi-active suspension systems with parameter
uncertainties and actuator time delays. Using linear matrix inequality constraints
method, the design of controller is cast into a convex multi-objective optimization
problem. In another reference [43], a robust controller H? was designed for the
defined plant with the uncertainty models. The robust controller assured high
quality of control despite the uncertainty models. Robust control of vibrations of a
rigid rotor was confirmed by experimental studies. A digital signal processor was
used to execute the control algorithm in real time. To demonstrate and validate the
minimum norm model validation approach for methodically constructing uncer-
tainty models for a real application [36], a series of robust control strategies were
designed, tested, and then compared in terms of stability robustness and disturbance
rejection performance. Results show that robust control design produced sig-
nificantly better performance in terms of stability and robust performance.

1.4.4 Time delay control methods

The time delay brings us the difficulty to design control strategy. For an active
suspension system, the time delay sometimes needs to be dealt with carefully,
because the time delay may cause the loss of stability. As a consequence, the
stability analysis of the system becomes important. The time delay compensation
becomes very important for an active suspension system with a time delay. How-
ever, the time delay only degrades the control performance for a semi-active sus-
pension system since the semi-active suspension only can dissipate the vibration
energy without introducing energy to the system. Therefore, the semi-active sus-
pension is inherently stable in this sense. It is necessary to improve the control
performance through time delay compensation.
Recently, stability analysis and controller synthesis have been presented for
linear systems with state delay or actuator delay. For example, the robust Hinf
control problem for uncertain linear system with multiple state delays was studied
by many researchers [5, 21, 22, 32, 67]. The similar method also was investigated
in quarter-vehicle, half-vehicle, and seat active suspension systems with actuator
delay [13, 14, 17]. In summary, there are two main approaches dealing with the
actuator delay problems. One is to design a controller using the integrated system
model where the actuator dynamics are included [62]. The other is to design a
controller by considering the actuator delay in the controller design process [13].
Xuan et al. [70] proposed the time delay control to the force loop controller of an
electrohydraulic active suspension. The numerical results show that the proposed
control scheme can effectively reduce the vibration from random road.
A robust gain scheduled control technique using time delay for the nonlinear
systems with model uncertainty and unexpected disturbances was proposed by
Sung and Shin [58]. The control method uses the past observation of the system’s
response and the control input to directly modify the control actions rather than to
adjust the controller gains or to identify system parameters. Simulation demon-
strated the benefits of this proposed scheme.
 State-of-the-art of vehicle intelligent suspension control system 27

In reality, the time delay may be dependent on many factors and be time-
varying in the control design for an intelligent suspension. Li [32] studied the
robust control design for vehicle active suspension systems with uncertainty in
time delay. By removing the assumptions from the state-of-the-art methods, state-
feedback and output-feedback controller design methods were proposed to design
less conservative state-feedback and output-feedback controller existence condi-
tions. It overcomes the challenges brought by generalized time-varying actuator
delay. To compensate the time delay uncertainty of MR dampers, a back propa-
gation network with four layers and nodes N1–N2–N3–N4 (in this study, N1 ¼ 4,
N2 ¼ 8, N3 ¼ 8, N4 ¼ 1) was adopted in our prior work [10].
Li et al. [34] proposed a fuzzy H? controller for active suspension systems
with actuator delay and fault. The sprung and unsprung mass variations, the
actuator delay and fault, and the suspension performance have all been taken into
account to construct the T–S fuzzy system for the control design objective. A
quarter-vehicle suspension model was used to validate the effectiveness of the
proposed design method. Simulation results demonstrated that the designed reliable
fuzzy controller has the capability of guaranteeing a better suspension performance
under sprung and unsprung mass variations, actuator delay, and fault.

1.4.5 Fault-tolerant control method

Because of the growing complexity of automated control systems, the control
system can meet some types of fault, especially faults from actuators and sensors.
Once an actuator or a sensor fault occurs in an intelligent suspension, the conven-
tional controllers cannot achieve better performance in comparison with the reli-
able and fault-tolerant controllers as discussed in [75], where the considered
actuator fault was described as a static behavior. The fault has different effect for an
active suspension system and a semi-active suspension system. For an active sus-
pension system, the fault usually will cause the loss of stability, whereas the control
performance will only be reduced for a semi-active suspension system in most
cases.
During the past few decades, many researches have attempted to resolve the
reliable and fault-tolerant control problems for dynamic systems with uncertainty
such as actuator and sensor faults, and a great number of theoretic results have also
been presented.
Generally speaking, fault-tolerant control system can be classified into two
types: passive and active. Passive fault-tolerant controllers are fixed and are
designed to be robust against a class of presumed faults, which is schematically
shown in Figure 1.14 [24]. In contrast to passive fault-tolerant controllers, active
fault-tolerant controllers react to the system component failures actively by
reconfiguring control actions so that the stability and performance of the system
can be maintained, which is shown in Figure 1.15 [73]. In certain circumstances,
degraded performance can be accepted.
Fault-tolerant control methods have been successfully applied in the vehicle
engineering area. Most of these control methods belong to passive fault-tolerant control.
28 Handbook of vehicle suspension control systems

Faults
N
U Actua- Sen- Y
Process
tors sors

Process
model

Model-based
Feature fault detection
generation
r, , x features
Normal Change
behavior detection
s analytical symptoms
Fault
diagnosis
f faults

Figure 1.14 General scheme of process model-based fault-detection and

diagnosis [24]

Fault detection
and diagnosis
(FDD)
Actuator W System V Sensor
faults faults faults
Command r Reconfigurable u z
(reference) feedforward Actuators Systems Sensors
governor controller -

Reconfiguration
mechanism
Reconfigurable
feedback
controller

Figure 1.15 A general structure of AFTCS [73]

For example, [75] proposed the reliable and fault-tolerant controller and found that
it has better performance than the conventional controllers when either the actuator
or the sensor faults occur in an active suspension system. Long et al. [38] studied an
active fault-tolerant control for a new structure of module suspension system of
electromagnetic system maglev train by control law reconfiguration. The numerical
results show that the active fault-tolerant control is more effective than the con-
ventional method. Qiu et al. [52] investigated fault estimation and compensation of
an active suspension system in finite-frequency domain. Mid et al. [41] studied
the sensor fault for truck pulling a trailer system, an automobile suspension, and
fileld controlled direct current motor system. When a senor fault is present, the
controller input is augmented to compensate the fault. Kim and Lee [28] proposed a
 State-of-the-art of vehicle intelligent suspension control system 29

Table 1.2 Fault flags

Falut category Flag type

Actuator faults Compressor or solenoid valve overheat flag
Compressor or solenoid valve overtime flag
Compressor or solenoid valve overtime fault flag
Compressor or solenoid valve driver fault flag
Plant faults Ignition inhibit flag
Severe lowering flag
Battery voltage fault flag
Severe vehicle roll flag
Sensor faults Vehicle speed sensor fault flag
Height sensor fault flag
Height sensor driver fault flag
Model-based height sensor fault flag

model-based fault-detection for the pressure sensor and the height sensors, which
are critical components in the closed-loop air suspension system. The detected
faults include actuator faults, plant faults, sensor faults, as shown in Table 1.2.
Simulation and test results showed that the proposed algorithm can determine
the target height to achieve the various control purposes and regulate the vehicle
height to the target height, even with time-varying model uncertainties. The fault-
detection and diagnosis algorithm can successively detect unknown system faults
and manage the faults to protect the system components and the driver from the
faults.
To compensate the effect of temperature on the MR dampers, an adaptive
sliding mode fault-tolerant control strategy was proposed in our prior work [10].
The performance of the proposed control scheme is evaluated and compared under
the bump road condition and presented in time domain. The results show that sig-
nificant gains are made in the presence of the partial fault of an MR damper. The
control scheme could reduce the effect of the partial fault of the MR damper on
system performance.

1.5 The method of validation

After formulating the controller, it is important to check the control performance of
intelligent suspension through different methods. These methods of validation
include numerical simulation, simulation of hardware in the loop, and the road test.
Compared to other two methods, the numerical simulation is the cheapest and
fastest method to validate the control performance. The method usually is based on
the MATLAB/SIMULINK. The method is the first step to validate the control
algorithm by many researchers [46, 51, 60]. Because the actual physical conditions
of intelligent suspension system are not considered in the numerical simulation,
limitation of this method can also not be avoided. For example, the ideal control
30 Handbook of vehicle suspension control systems

performance may not be assured in the road test again. Some control strategies such
as full state feedback cannot be implemented at all.
The method of hardware in the loop is a trade-off between the numerical
simulation and the road test.
It provides an effective platform by adding the complexity of the controlled
suspension system under control to the test platform. Some realistic physical ele-
ments are involved in the control loop [46, 60], for example, Batterbee et al. [3]
developed a skyhook damping controller for a two-degree-of-freedom quarter-car
model and investigated the controller performance by using the method, which is
shown in Figure 1.16. Lauwerys et al. [31] studied linear control of car suspension
using nonlinear actuator control on a quarter-car test setup.
The method of road test is the most expensive and time-consuming of the three
methods. However, the method is important to check the actual control perfor-
mance and tune the control parameters. Therefore, the method is also adopted by
some researchers. For example, Nahvi et al. (2009) selected four flat road surfaces
to evaluate the ride comfort of a passenger car. The road surfaces are shown in
Figure 1.17.

MR actuation Kepco
current amplifier
Carrera MR
damper
Transducers
(Load cell, LVDT)

IST Actuator
Position Desired MR
Manifold feedback damper
IST 8400
Servo-valves current
Controller

Measurement
Hydraulic power
data
supply

HOST PC RUNNING XPC Desired actuator

TARGET position
Model of ‘simulated’ isolator Upload HILS
TARGET PC
components and controller data RUNNING REAL-
Mc TIME
Damper force SIMULATION
K
(from hardware) N.I. DAQ Card Analogue inputs and outputs
Mw D/A Converter INTERFACE BOARD
A/D Converter
Cw Kw
Download C-
C-COMPILER code

Figure 1.16 Schematic diagram of the HILS experimental facility [3]

 State-of-the-art of vehicle intelligent suspension control system 31

(a) (b)

(c) (d)

Figure 1.17 Road surfaces: (a) highway, (b) pavement, (c) suburban, and
(d) bumpy [44]

To validate the actual fault-tolerant control algorithm for a four-corner closed-

loop air suspension system, the actual vehicle test was conducted with a test vehicle
having an electronically controlled air suspension system. Figure 1.18 shows the
schematics of the test environment.

1.6 Final remarks and conclusions

The intelligent suspensions can achieve good trade-off between ride comfort and
road holding. Two main problems of designing an intelligent suspension are
reviewed in this study. The problems include dynamic modeling and control
design. To achieve better ride comfort and road holding, some more precise
dynamic models are developed in accounting for the inherent nonlinearity, uncer-
tainty, time delay, and fault of intelligent suspension systems in literatures. Cor-
respondingly, nonlinear control methods, uncertainty robust control methods, time
delay control algorithms, and fault-tolerant control strategies are proposed by many
researchers to deal with the control problem caused by the inherent nonlinearity,
uncertainty, time delay, and fault of intelligent suspension systems. This study
provides an account of the state-of-the-art of intelligent suspension control system.
32 Handbook of vehicle suspension control systems

[SignalBus]
[SignalBus] [SignalBus]
SigBus CompRevCmd 1
SigBus CompRevCmd
Z target_i Z target_i
LevReq DesLSta AirFlowDirec
1 VehSig SigBus SigBus
Sensor & Z_f_i Z_f_i FlowDirecControl
Can signals
TargetModeDecision LevReqDecision LevReq DesLSta [DesLSta]
Signal Processing 1/2
[SignalBus] SigBus AirFlowDirec
[DesLSta] DesLSta FaultFlag FaultFlag DesLSta AirSpringCmd 2
AirSpringCmd
ActuatorCmd SigBus
FaultDetection LevelingManagement AirSpringControl
(failsafe) [SignalBus]
1/2

Height Control and Fail Safe Algorithm

Data Program
acquisition download

Actuator
driver

Signal
conditioner
dSpace
Microautobox Vehicle with CLAS System

Figure 1.18 Schematics of the test environment [28]

We provide some open questions and scientific problems for future research as
follows:
1. Developing more precise vehicle suspension dynamic model: Most of the
reported literatures on intelligent suspension control design are studied based
on the simplified lumped parameter model, such as the quarter-car model and
half-car model. With the development of computer technology, some precise
full car 3D models can easily be developed and be used to control design. In
addition, the tire model and human body model are also important and should
be incorporated into the full car model. At last, the precise model of actuators
is beneficial to control design. However, the dynamic model of actuators in
intelligent suspension systems is scarcely discussed in literatures.
2. Research on the difference between the design model and the real situation:
The problem is very important for implementing the control algorithm to
real vehicle suspension in real-time control. When some complex physical
characteristics are added to vehicle dynamic models such as nonlinearity,
uncertainty, time delay, and fault of intelligent suspension systems, how much
is the difference between the designed model and the real situation? There are
few papers about the issue.
 State-of-the-art of vehicle intelligent suspension control system 33

3. Evaluating all kinds of control algorithms from the perspective of practical

applications: With the development of control theory, some novel and complex
control algorithms are applied to control intelligent suspension systems. Vehi-
cle suspension system is a complex dynamic system with significant non-
linearity, uncertainty, time delay, and fault. How to choose a suitable control
algorithm for the vehicle suspension system is still a challenging problem. To
choose a suitable control method, it is necessary to develop an evaluation
system to compare different control algorithms according to some standards
such as the control performance, the application capability, and the cost of
implementation. The comparison results will be helpful to choose an appro-
priate control method for developing an intelligent suspension system.

Acknowledgments
This research is supported financially by the National Natural Science Foundation of
People’s Republic of China (Project No. 51275539, 60804018), and the Fundamental
Research Funds for the Central Universities (CDJZR12110058, CDJZR13135553)
and the Program for New Century Excellent Talents in University (NCET-13-0630).
These supports are gratefully acknowledged.

References

[1] Ahmadian, M., Simon, D.E. (2002). An analytical and experimental eva-
luation of magneto rheological suspensions for heavy trucks. Vehicle System
Dynamics 37: 38–49.
[2] Ahmadian, M., Vahdati, N. (2006). Transient dynamics of semiactive sus-
pensions with hybrid control. Journal of Intelligent Material Systems and
Structures 17(2): 145–153.
[3] Batterbee, D.C., Sims N.D. (2007). Hardware in the loop simulation of
magnetorheological dampers for vehicle suspension systems. Proceedings of
IMechE Part I: Journal of Systems and Control Engineering 221: 225–238.
[4] Bommer, A. (2006). A nonlinear vehicle model for comfort analysis.
Master’s thesis: 1–101.
[5] Cao, Y., Sun, Y., et al. (1998). Delay-dependent robust Hinf control for
uncertain systems with time varying delays. IEE Proceeding Control Theory
Applic 145(3): 338–344.
[6] Choi, S.B., Lee, H.S., et al. (2002). H-infinity control performance of a full-
vehicle suspension featuring magnetorheological dampers. Vehicle System
Dynamics 38(5): 341–360.
[7] Crosby, M.J., Harwood, R.A., Karnopp, D. (1974). Vibration control using
semi-active force generators. ASME Journal of Engineering for Industry.
96(2): 619–626
[8] Davis, B.R., Thompson, A.G. (1988). Optimal linear active suspension with
integral constraint. Vehicle System Dynamics 17(6): 357–366.
34 Handbook of vehicle suspension control systems

[9] Demir, O., Keskin, I., et al. (2012). Modeling and control of a nonlinear
half-vehicle suspension system: a hybrid fuzzy logic approach. Nonlinear
Dynamics 67: 2139–2151.
[10] Dong, X., Yu, M., Guan, Z. (2011). Adaptive sliding mode fault-tolerant
control for semi-active suspension using magneto-rheological dampers.
Journal of Intelligent Material Systems and Structures 22(15): 1–9.
[11] Dong, X., Yu, M., et al. (2010). Comparative research on semi-active control
strategies for magneto-rheological suspension. Nonlinear Dynamics 59:
433–453.
[12] Dong, X., Yu, M.M., et al. (2009). Neural network compensation of semi-
active control for magneto-rheological suspension with time delay uncer-
tainty. Smart Materials & Structures 18(1): 1–14.
[13] Du, H., Zhang, N. (2007). H? control of active vehicle suspensions with
actuator time delay. Journal of Sound and Vibration 301: 236–252.
[14] Du, H., Zhang, N. (2008a). Constrained H? control of active suspension for
a half-car model with a time delay in control. Proceedings of the Institution
of Mechanical Engineers Part D – Journal of Automobile Engineering 222:
665–684.
[15] Du, H., Zhang, N. (2008b). Constrained Hinf control of active suspension for
a half-car model with a time delay in control. Proceedings of the Institution
of Mechanical Engineers, Part D: Journal of Automobile Engineering 222:
665–685.
[16] Du, H., Zhang, N. (2009). Fuzzy control for nonlinear uncertain electro-
hydraulic active suspensions with Input constraint. IEEE Tractions on Fuzzy
Systems 17(2): 343–356.
[17] Gao, H., Zhao, Y., et al. (2010). Input-delayed control of uncertain seat
suspension systems with human-body model. IEEE Transctions on Control
System Technology 18(3): 591–601.
[18] Goh, Y.M., Booker, J.D., et al. (2005). Uncertainty modelling of a suspen-
sion unit. Proceedings of the Institution of Mechanical Engineers, Part D:
Journal of Automobile Engineering 219: 755–772.
[19] Griffin, M.J. (1990). Handbook of human vibration. London: Academic Press.
[20] Hac, A., Fratini, A.V. (1999). Elimination of limit cycles due to signal
estimation in semi-active suspensions. International Congress and Exposi-
tion Detroit, Michigan, March 1–4: 1–7.
[21] Han, Q. (2005). A new delay-dependent stability criterion for linear neutral
systems with norm-bounded uncertainties in all system matrices. Interna-
tional Journal of Systems Science 36: 469–475.
[22] He, Y., Wu, M., et al. (2004). Delay-dependent robust stability criteria for
uncertain neutral systems with mixed delays. Systems & Control Letters 51:
57–65.
[23] Ikhouane, F., Dyke, S.J. (2007). Modeling and identification of a shear mode
magnetorheological damper. Smart Materials and Structures 16: 605–616.
[24] Isermann, R. (2005). Model-based fault-detection and diagnosis – status and
applications. Annual Reviews in Control 29: 71–85.
 State-of-the-art of vehicle intelligent suspension control system 35

[25] Jimenez, R., Alvarez, L. (2002). Real time identification of structures with
magnetorheological dampers. Proceedings of 41st IEEE Conference on
Decision and Control 1: 1017–1022.
[26] Jones, W.D. (2005). Easy ride: Bose Corp. uses speaker technology to give
cars adaptive suspension. IEEE Spectrum 42(5): 12–14.
[27] Joo, S., Byun, J., et al. (1997). Design and analysis of the nonlinear feedback
linearizing controller for an EMS system. IEEE Tranactions on Control
Systems Technology 5(1): 135–144.
[28] Kim, H., Lee, H. (2011). Fault-tolerant control algorithm for a four-corner
closed-loop air suspension system. IEEE Transactions on Industrial Elec-
tronics 58(10): 4866–4879.
[29] Kushawaha, A.K. (2012). Stabilization and tracking control for electro-
magnetic suspension system using feedback linearization. 2012 Students
Conference on Engineering and Systems (SCES), Allahabad, Uttar Pradesh,
India: 1–6.
[30] Kwok, N.M., Ha, Q.P., Nguyen, T.H., Li, J., Samali, B. (2006). A novel
hysteretic model for magnetorheological fluid dampers and parameter
identification using particle swarm optimization. Sensors Actuators A 132:
144–151.
[31] Lauwerys, C., Swevers, J., et al. (2002). Linear control of car suspension
using nonlinear actuator control. Proceedings of ISMA 1: 55–62.
[32] Li, H. (2012). Robust control design for vehicle active suspension systems
with uncertainty. Ph.D. dissertation of University of Portsmouth: 1–206.
[33] Li, H., Chen, B., et al. (2008). Delay-dependent robust stability for stochastic
time-delay systems with polytopic uncertainties. International Journal of
Robust and Nonlinear Control 18: 1482–1492.
[34] Li, H., Liu, H., et al. (2012). Reliable fuzzy control for active suspension
systems with actuator delay and fault. IEEE Transactions on Fuzzy Systems
20(2): 342.
[35] Li, W.H., Yao, G.Z., Chen, G., Yeo, S.H., Yap, F.F. (2000). Testing and
steady state modeling of a linear MR damper under sinusoidal loading.
Smart Mater. Struct. 9: 95–102.
[36] Lim, K.B., Cox, D.E. (1998). Validation of an experimentally derived
uncertainty model. Journal of Guidance, Control, and Dynamics 21(3):
485–492.
[37] Litak, G., Borowiec, M., et al. (2008). Chaotic vibration of a quarter-car
model excited by the road surface profile. Communications in Nonlinear
Science and Numerical Simulation 13: 1373–1383.
[38] Long, Z., Xue, S., et al. (2007). New strategy of active fault-tolerant control
for suspension system of maglev train. Proceedings of the IEEE Interna-
tional Conference on Automation and Logistics, August 18–21, Jinan, China:
88–95.
[39] Martins, I., Esteves, J., et al. (2006). Permanent-magnets linear actuators
applicability in automobile active suspensions. IEEE Transactions on Vehi-
cular Technology 55(1): 86–94.
36 Handbook of vehicle suspension control systems

[40] McGee, C.G., Haroon, M., et al. (2005). A frequency domain technique for
characterizing nonlinearities in a tire-vehicle suspension system. Journal of
Vibration and Acoustics 127: 61–76.
[41] Mid, E.C., Faranadia, A.H., et al. (2011). Application of model based sensor
fault tolerant control system. 2011 IEEE International Conference on Con-
trol System, Computing and Engineering, Penang, Malaysia: 150–155.
[42] Mohamed, M.E., Zuhair, S.A. (1999). Linear quadratic gaussian control of a
quarter-car suspension. Vehicle System Dynamics: International Journal of
Vehicle Mechanics and Mobility 32(6): 479–497.
[43] Mystkowski, A., Gosiewski, Z. (2009). Uncertainty modeling in robust
control of active magnetic suspension. Solid State Phenomena 144: 22–26.
[44] Nahvi, H., Fouladi, M.H., Nor, M.J.M. (2009). Evaluation of whole-body
vibration and ride comfort in a passenger car. International Journal of
Acoustics and Vibration 14(3): 143–149.
[45] Nakano, K., Suda, Y., et al. (2003). Self-powered active vibration control using
a single electric actuator. Journal of Sound and Vibration 260(2): 213–235.
[46] Noor, H., Amer, R.R., et al. (2011). A review on control strategies for pas-
senger car intelligent suspension system. International Conference on Elec-
trical, Control and Computer Engineering, June 21–22, Pahang, Malaysia:
404–409.
[47] Oh, H.U. (2004). Experimental demonstration of an improved magneto-
rheological fluid damper for suppression of vibration of a space flexible
structure. Smart Materials and Structures 13: 1238–1244.
[48] Okada, Y., Harada, H. (1995). Active and regenerative control of electro-
dynamic vibration damper. Proceedings of the 1995 ASME Design Engineering
Technical Conference, Boston, MA: 595–602.
[49] Osvaldo, E., Agamennoni, I., et al. (2004). Nonlinear uncertainty model
of a magnetic suspension system. Mathematical and Computer Modelling
40(9–10): 1075–1087.
[50] Paschedag, T. (2006). Constrained optimal control: an application to semiactive
suspension systems. Control and Automation. MED ’06. 14th Mediterranean,
Università Politecnica delle Marche, Ancona, Italy: 1–8.
[51] Poussot-Vassal, C., Spelta, C., et al. (2011). Survey on some automotive semi-
active suspension control methods: a comparative study on a single-corner
model. 18th IFAC World Congress (IFAC WC 2011), Milan, Italy: 1–8.
[52] Qiu, J., Ren, M., et al. (2011). Active fault-tolerant control for vehicle active
suspension systems in finite-frequency domain. IET Control Theory and
Applications 5(13): 1544–1550.
[53] SAE. (1992). Measurement of whole body vibration of the seated operator of
off-highway work machines. SAE J1013, Society of Automotive Engineers,
Warrendale, PA.
[54] Shi, P., Boukas, E., et al. (2003). Robust disturbance attenuation for discrete-
time active fault tolerant control systems with uncertainties. Optimal Control
Applications and Methods 24: 85–101.
 State-of-the-art of vehicle intelligent suspension control system 37

[55] Sims, N.D. (2006). Limit cycle behavior of smart fluid dampers under closed
loop control. Journal of Vibration and Acoustics, Transactions of the ASME
128(4): 413–428.
[56] Spencer, B., Dyke, F.J.S.J., et al. (1997). Phenomenological model for
magnetorheological dampers. Journal of Engineering Mechanics ASCE 123:
230–238.
[57] Suda, Y., Shiiba, T. (1996). A new hybrid suspension system with active
control and energy regeneration. Vehicle System Dynamics 25(Suppl.):
641–654.
[58] Sung, H.K., Shin, C. (2006). A time delay-based gain scheduled control for
electromagnetic suspension system. The 19th International Conference on
Magnetically Levitated Systems and Linear Drives, September 13–15,
Dresden, Germany: 1–6.
[59] Taghirad, H., Esmailzadeh, E. (1998). Automobile passenger comfort
assured through LQG/LQR active suspension. Journal of Vibration Control
4(5): 603–618.
[60] Tang, C., Zhang, T. (2005). The research on control algorithm of vehicle
active suspension system. IEEE International Conference on Vehicular
Electronics and Safety – ICVES, Xian, PEOPLES R CHINA: 320–325.
[61] Thompson, A., Davis, B. (1988). Optimal linear active suspensions with
derivative constraints and output feedback control. Vehicle System Dynamics
17: 179–192.
[62] Thompson, A., Davis, B. (2001). Technical note: force control in electro-
hydraulic active suspensions revisited. Vehicle System Dynamics 35: 217–222.
[63] Wang, D.H., Liao, W.H. (2011). Magnetorheological fluid dampers: a
review of parametric modelling. Smart Materials and Structures 20: 1–34.
[64] Wang, E.R., Ma, X.Q., et al. (2003). Semi-active control of vehicle vibration
with MR-dampers. Proceedings of the 42nd IEEE Conference on Decision
and Control, December, Maui, HI.
[65] Wang, J., Wang, H., et al. (2011). Chaos control of vehicle nonlinear sus-
pension system with multi-frequency excitations by nonlinear feedback.
Applied Mechanics and Materials 55–57: 1156–1161.
[66] Wang, L.X., Kamath, H. (2006). Modelling hysteretic behaviour in magne-
torheological fluids and dampers using phase-transition theory. Smart
Materials and Structures 15: 1725–1733.
[67] Wang, Z., Liu, Y., et al. (2010). A note on control of a class of discrete-time
stochastic systems with distributed delays and nonlinear disturbances.
Automatica 46: 543–548.
[68] Wereley, N.M., Kamath, G.M., et al. (1999). Hysteresis modeling of
semi-active magnetorheological helicopter dampers. Journal of Intelligent
Material Systems and Structures 10: 624–633.
[69] Wereley, N.M., Pang, L., et al. (1998). Idealized hysteresis modeling of
electrorheological and magnetorheological dampers. Journal of Intellient
Material Systems and Structures 9: 642–649.
38 Handbook of vehicle suspension control systems

[70] Xuan, D.J., Kim, J.-W., et al. (2007). A study on active suspension system
using time delay control. International Conference on Control, Automation
and Systems, October 17–20, COEX, Seoul, Korea: 388–403.
[71] Xue, X.D.K., Cheng, W.E., et al. (2011). Study of art of automotive active
suspensions. 2011 4th International Conference on Power Electronics
Systems and Applications, Hong Kong: 360–366.
[72] Yang, G., Zhang, S., et al. (2002). Reliable control using redundant con-
trollers. IEEE Transactions on Automatic Control 43: 1588–1593.
[73] Zhang, Y., Jiang, J. (2008). Bibliographical review on reconfigurable fault-
tolerant control systems. Annual Reviews in Control 32: 229–252.
[74] Zhao, H., Lu, S. (1999). A vehicle’s time domain model with road input on
four wheels. Automotive Engineering 21(2): 112–117.
[75] Zhao, Y., Zhao, L., et al. (2010). Vibration control of seat suspension using
H? reliable control. Journal of Vibration and Control 16: 1859.
[76] Zhu, Q., Ishitobi, M. (2006). Chaotic vibration of a nonlinearfull-vehicle-
model. International Journal of Solids and Structures 43(3–4): 747–759.
[77] Zuo, L., Nayfeh, S.A. (2003). Structured H2 optimization of vehicle
suspensions based on multi-wheel models. Vehicle System Dynamics 40(5):
351–371.
 Chapter 2
Intelligence-based vehicle active suspension
adaptive control systems
Jiangtao Caoa,c, Ping Lib and Honghai Liuc

Abstract

This chapter reviews computational-intelligence-involved approaches in active

vehicle suspension control systems with a focus on the problems raised in practical
implementations by their nonlinear and uncertain properties. After a brief intro-
duction on active suspension models, the chapter explores state-of-the-art in fuzzy
inference systems, neural networks, genetic algorithms, and their combination for
suspension control issues. Discussion and comments are provided based on the
reviewed simulation and experimental results. The chapter is concluded with
remarks and future directions.

2.1 Introduction
A suspension system is one of the important components of a vehicle, which plays a
crucial role in handling performance and ride comfort characteristics of a vehicle.
A suspension system acts as a bridge between the occupants of a vehicle and the
road it rides on. It has two main functionalities, one is to isolate the vehicle body
with its passengers from external disturbance inputs, which mainly come from
irregular road surfaces. It always relates to riding quality. The other is to maintain a
firm contact between the road and the tires to provide guidance along the track. It is
called handling performance. In a conventional passive suspension system, which
comprises only springs and dampers, a trade-off is needed to resolve the conflicted
requirements of ride comfort and good handling performance. The reason is that
stiff suspension is required to support the weight of vehicle and to follow the track;
on the other hand, soft suspension is needed to isolate the disturbance from the
road. Hence, there exists a significantly growing interest in design and control of
active suspension systems for automotive engineers and researchers in the past

a
School of Automation, Northwestern Polytechnic University, Shaanxi, China
b
School of Information and Control Engineering, Liaoning Shihua University, Fushun, China
c
Institute of Industrial Research, the University of Portsmouth, Portsmouth PO1 3QL, UK
40 Handbook of vehicle suspension control systems

three decades. An active suspension system is characterized by employing some

kinds of suspension force generation such as pneumatic, magneto-rheological, or
hydraulic actuators. Practical applications of active suspension systems have been
facilitated by the development of microprocessors and electronics from the mid-
1980s [1–3]. Related surveys on theories and applications of active suspension
control systems were conducted in 1997 [4, 5], with fast-growing computational-
intelligence methodologies significantly driving recent advances in this research
area in the past decade.
The design of a vehicle active suspension control system (ASCS) is a long-
standing control engineering problem, which is rooted in multiple parameters’
optimization at real-time requirement. It includes ride comfort, body motion, road
handling, and suspension travel [6]. Ride comfort directly relates to the acceleration
sensed by passengers; body motion means bounce, pitch and roll of sprung mass are
created by cornering, acceleration, or deceleration; road handling is associated with
the contact forces of tires and the road surface; suspension travel refers to the
displacement between a sprung mass and an unsprung mass. It is really a challen-
ging issue for one active suspension system to simultaneously optimize all four sets
of parameters. Hence, how to handle related trade-offs is crucial for successfully
designing an ASCS. Research in the past three decades has shown that a linear
optimal control scheme provides an efficient way to design an active suspension
system, which can improve the vehicle ride and handling performance together
[4, 5]. This is based on the assumption that there exists a perfect (broad bandwidth)
actuator, which can generate the required force fast enough and the system can be
linearized within some opera regions. However, a real vehicle suspension system is
inherently nonlinear, even with some uncertainties. Therefore, adaptive control
schemes have to undertake the role of providing self-tuning feedback gains and to
take the aforementioned four sets of parameters into account to ensure optimal
operation of the system in different driving conditions and road surfaces [7–11].
A classical form of adaptive scheme for a vehicle active suspension system was
introduced in the late 1980s by Hać [7]. This is the starting point of the adaptive
control scheme, in which a set of feedback gains are varied by the change of power
spectral density of terrain roughness obtained by processing the measurement data.
Another comparison of adaptive Linear-Quadratic-Gaussian (LQG) and nonlinear
controllers for active suspensions was presented by Gordon et al. in Reference 9.
A model reference adaptive control scheme was proposed by Alleyne et al. [12],
which resulted in better performance than that of the active suspension system with
nonadaptive controller and passive suspension system. Also in this chapter, 10–30%
variances of sprung mass and stiffness coefficients are examined to check the
adaptation capability based on a single degree-of-freedom (DOF) quarter-vehicle
model. Sunwoo and Cheok proposed an explicit adaptive control for an active sus-
pension system, which is based on a self-tuning controller design [8]. It consisted
of online, low-order recursive parameters estimation, closed-form algebraic gain
computation, and manipulation of the control parameters. Some other works on
adaptive control of active suspension systems can be found in References 13–15. Up
to this point most researchers have dealt with a linear model in developing control
laws or using adaptive control scheme to conquer the limited nonlinear properties of
 Intelligence-based vehicle active suspension adaptive control systems 41

suspension systems. However, if the system is highly nonlinear over the range of
operation, its adaptive schemes may show severe limitations. For instance, if a
wheel stroke is so strong that the stiffness of a suspension is beyond the linear range,
it might be practically impossible to identify parameters through ordinary identifi-
cation [15–17]. In the early 1990s many studies began to consider nonlinearities,
uncertainties, and unmodeled parts of a real suspension system, which requires the
use of a nonlinear model and some nonlinear forms of control scheme [12, 18]. In
practice, these nonlinear models made ASCSs so complex and too challenging to
employ.
In industrial applications, control engineers often have to deal with complex
systems, having multiple variable and multiple parameter models with perhaps
nonlinear coupling. The conventional approaches for understanding and predicting
the behavior of such systems based on analytical techniques can be proved to be
inadequate, even at the initial stages of establishing an appropriate mathematical
model [19]. The computational environment used in such an analytical approach is
perhaps too categorical and inflexible in order to cope with the intricacy and the
complexity of the real-world industrial systems. It turns out that, in dealing with
such systems, one has to face a high degree of uncertainty and tolerate impreci-
sion. Trying to increase precision can be very costly. Thanks to significant
development of soft computing or computational intelligence in the past decades,
it has provided alternative ways to nonlinear system modeling and control. Gen-
erally speaking, the principal constituents of computing intelligence include fuzzy
logic (FL), artificial neural networks (ANNs), and evolutionary computing (EC).
FL is mainly concerned with imprecision and approximate reasoning, ANNs
mainly with learning and curve fitting, and EC mainly with global optimization
based on the natural selection and genetics. These intelligent computing meth-
odologies have resulted in the development of the ‘‘intelligent control’’ field,
which consists of novel control approaches based on FL, ANNs, EC, other tech-
niques induced from artificial intelligence, and their combination. These methods
provide an extensive freedom for control engineers to deal with practical problems
of vagueness, uncertainty, or imprecision. Convincingly, these intelligent methods
are good candidates for alleviating the problems associated with ACSCs [20].
Although in hard computing, imprecision and uncertainty are undesirable prop-
erties, computational intelligent approaches, also known as soft computing, pro-
vide the tolerance for imprecision and uncertainty, which is exploited to achieve a
practically acceptable solution at a reasonable cost, tractability, and high machine
intelligence quotient (MIQ). Zadeh argued that soft computing, rather than
hard computing, should be viewed as the foundation of machine intelligence. A
complete comparison of their capabilities in different application fields was con-
structed by Fukuda and Kubota in Table 2.1, together with those of control theory
and artificial intelligence [21].
This chapter reviews recent intelligent control approaches for active suspen-
sion systems. The chapter is organized as follows. Section 2.2 gives a revisit on
modeling of an active suspension system. Section 2.3 reviews adaptive fuzzy
control methods; Section 2.4 presents adaptive fuzzy sliding-mode control (SMC)
approaches; Section 2.5 revisits neural networks based control systems; and
Table 2.1 Comparison of capabilities of different adaptive methodologies [21]

Mathematical Learning Operator Real time Knowledge Nonlinearity Optimization

model data knowledge representation
Control theory Good or Unsuitable Needs other Good or suitable Unsuitable Unsuitable Unsuitable
suitable methods
Neural network Unsuitable Good or Unsuitable Good or suitable Unsuitable Good or suitable Fair
suitable
Fuzzy logic Fair Unsuitable Good or Good or suitable Needs other Good or suitable Unsuitable
suitable methods
Other artificial Needs other Unsuitable Good or Unsuitable Good or suitable Needs other Unsuitable
intelligence methods suitable methods
Genetic Unsuitable Good or Unsuitable Needs other Unsuitable Good or suitable Good or
algorithms suitable methods suitable
 Intelligence-based vehicle active suspension adaptive control systems 43

z
Yaw

C.G. Pitch
y

x Roll

Fig. 2.1 Six-degree-of-freedom vehicle model [5]

Section 2.6 presents adaptive genetic algorithm control methods. Section 2.7
describes combination methods based on neural networks, fuzzy inference, and
generic algorithms. Finally we conclude the chapter in Section 2.8 with discussions
and future work.

2.2 Background

A vehicle body is generally a rigid body with six-DOF motions shown in Fig. 2.1
[5], it consists of longitudinal, lateral, and heave motions and roll, pitch, and yaw
motions. These motions are restricted by suspension geometries in vehicles and are
more or less coupled with one another. Moreover, as the suspensions have a
mechanical structure with unsprung mass, coupling also occurs between the sprung
and unsprung masses. Regardless of such coupling problems, the reduced-order
mathematical model is useful for designing an ASCS. Therefore, a quarter-vehicle
model or a half-vehicle model is often used for theoretical analysis and design of
active suspension systems [4, 5].
In this section, a linear quarter-vehicle model and a linear half-vehicle model of
an active suspension system are introduced. Their linear quadratic (LQ) controllers
are designed based on the models; practical active suspension system models are
also analyzed in terms of nonlinear properties and uncertain dynamic disturbances.

2.2.1 Active suspension system linear models and control

2.2.1.1 Quarter vehicle active suspension system modeling
and LQ control design
The quarter-vehicle model was initially developed to explore active suspension
capabilities and gave birth to the concepts of skyhook damping and fast load
leveling, which are now being developed toward actual, large-scale production
applications. In this chapter, we define
mb : quarter body mass (or sprung mass) (kg);
mw : wheel mass (or unsprung mass) ðkgÞ;
Ks : suspension spring stiffness ðN=mÞ;
44 Handbook of vehicle suspension control systems

Kt : tire stiffness ðN=mÞ;

c: damping coefficient ðNs=mÞ;
G0 : road roughness coefficient ðm3 =cycleÞ;
U0 : vehicle original forward velocity ðm=sÞ;
f0 : low cut-off frequency ðHzÞ;
z0 : road displacement ðmÞ;
zw : wheel displacement ðmÞ;
zb : body displacement ðmÞ;
fa : actuator force (N).
The quarter-vehicle model is shown in Fig. 2.2. The dynamic differential
equations of this suspension system can be represented as

mb€z b ¼ fa þ cð_z w  z_ b Þ þ Ks ðzw  zb Þ ð2:1Þ

mw€z w ¼ fa  cð_z w  z_ b Þ  Ks ðzw  zb Þ  Kt ðzw  z0 Þ ð2:2Þ

The road surface is a natural changing condition for vehicle. For better riding
comfort, a perfect road surface model is necessary to design vehicle ASCS. There
are many possible ways to analytically describe the road inputs, which can be
classified as shock or vibration [4]. Shocks are the discrete events of relatively
short duration and high intensity, e.g., a pronounced bump or pothole on an
otherwise smooth road. Vibrations, on the other hand, are characterized by pro-
longed and consistent excitations that are called ‘‘rough’’ roads. In this section, the
rough road is considered. The International Organization for Standardization (ISO)
has proposed a series of standards of road roughness classification using the power
spectral density (PSD) values (ISO 1982), as shown in Table 2.2. Due to the ISO,
the road displacement PSD can be described as

w
n
GðnÞ ¼ Gðn0 Þ ð2:3Þ
n0

zb

mb

KS c
fa
zw

mw

Kt zo

Fig. 2.2 Two-degree-of-freedom quarter-vehicle model

 Intelligence-based vehicle active suspension adaptive control systems 45

Table 2.2 Road roughness values classified by ISO

(degree of roughness S(W)106 )

Road class Range Geometric mean

A (very good) <8 4
B (good) 8–32 16
C (average) 32–128 64
D (poor) 128–512 256
E (very poor) 512–2048 1024

Here, n is the space frequency (m1 ) and time frequency f is f ¼ nv (v is the

vehicle speed), n0 is the reference space frequency, GðnÞ is the road displacement
PSD, Gðn0 Þ is road roughness coefficient shown in Table 2.2, w is the linear fitting
coefficient, always w ¼ 2. Then based on the standard road surface description, the
road surface input model has been built through an inform filter by Gaussian White
noise and successfully used in many presented works [6, 22]. The equation of road
surface input is
pffiffiffiffiffiffiffiffiffiffiffi
z_ 0 ¼ 2pf0 z0 þ 2p G0 U0 w0 ð2:4Þ

where f0 is low cut-off frequency, G0 is road roughness coefficient, and w0 is a

Gaussian white noise.
Equations 2.1, 2.2, and 2.4 are combined to give the state space representation
of quarter-vehicle model:

X_ ¼ AX þ BU þ FW ð2:5Þ
Y ¼ CX þ DU ð2:6Þ

where

X ¼ ½z_ b z_ w zb zw z0  ð2:7Þ
Y ¼ ½€z b €z w zw  zb zw  z0  ð2:8Þ
U ¼ ½ fa ; W ¼ ½w0  ð2:9Þ

Based on the proposed model, linear optimal control theory is used to design
the active suspension controller here. For obtaining the better handling perfor-
mance and riding comfort, the performance index can be written as a weighted
sum of mean square values of output performance variables including body
acceleration, wheel-to-body displacement, and dynamic tire deflection. The weight
coefficients are q1 , q2 , and q3 :
ðT n o
1
J ¼ lim q1 ðzw  zb Þ2 þ q2 ðzw  z0 Þ2 þ q3€z 2b dt ð2:10Þ
T!1 T 0
46 Handbook of vehicle suspension control systems

Changing (2.10) into a general matrix format, it becomes

ðT
1
J ¼ lim ½X T QX þ U T RU þ 2X T NU dt ð2:11Þ
T!1 T 0

where Q, R, and N can be gotten from (2.1), (2.2), and (2.4). Assuming that
an optimal state observer, i.e., Kalman filter, is available to get a satisfactory
estimation of state vector X^ ; based on the separation theorem, an optimal control
force is

U ¼ R1 BT PX^ ¼ K X^ ð2:12Þ

where K represents the gain matrix; and P is the solution of the following classical
algebraic Riccati equation:

PA þ AT P  ðPB þ NÞR1 ðBT P þ N T Þ ¼ Q ð2:13Þ

2.2.1.2 Half vehicle active suspension system modeling

and LQ control design
The half-vehicle model including pitch and heave modes was represented to
simulate ride characteristics of a simplified whole vehicle, which leads to sig-
nificant improvement in ride and handling [23]. Let f and r denote the front and
rear, x and z be the longitudinal forward direction and vertical up direction,
respectively, in this chapter; we define

df : distance from the front axle to the center of gravity (m);

dr : distance from the rear axle to the center of gravity (m);
Ib : pitch inertia ðkgm2 Þ;
zf 0 : road displacement at the front wheel (m);
zr0 : road displacement at the rear wheel ðmÞ;
zwf : front wheel displacement ðmÞ;
zbf : front body displacement ðmÞ;
zwr : rear wheel displacement ðmÞ;
zbr : rear body displacement ðmÞ;
faf : front actuator force ðNÞ;
far : rear actuator force (N).

The half-vehicle model is shown in Fig. 2.3. With the assumption of a small
pitch angle, the following are obtained:

zbf ¼ zb  df  q; zbr ¼ zb þ dr  q ð2:14Þ

From (2.14), the pitch angle can be written as

zbr  zbf
q¼ ð2:15Þ
df þ dr
 Intelligence-based vehicle active suspension adaptive control systems 47

zbf df dr zbr

mb Ib zb

Ksf U0 Ksr
cf faf far cr

zwf zwr

Z
mwf mwr
zf 0 zr0
Ktf Ktr
X

Fig. 2.3 Half-vehicle suspension model

and hence the model equations of motion can be written as follows:

€z wf mwf ¼ Ktf ðzwf  zf 0 Þ  ½ faf þ cf ð_z wf  z_ bf Þ þ Ksf ðzwf  zbf Þ ð2:16aÞ

€z wr mwr ¼ Ktr ðzwr  zr0 Þ  ½ far þ cr ð_z wr  z_ br Þ þ Ksr ðzwr  zbr Þ ð2:16bÞ

€z b mb ¼ faf þ cf ð_z wf  z_ bf Þ þ Ksf ðzwf  zbf Þ

þ far þ cr ð_z wr  z_ br Þ þ Ksr ðzwr  zbr Þ ð2:16cÞ

€ b ¼ df ½ faf þ cf ð_z wf  z_ bf Þ þ Ksf ðzwf  zbf Þ

qI

þ dr ½ far þ cr ð_z wr  z_ br Þ þ Ksr ðzwr  zbr Þ ð2:16dÞ

Substituting (2.14) into (2.16c) and (2.16d), we have the following:

!
1 df2
€z bf ¼ þ ½ faf þ cf ð_z wf  z_ bf Þ þ Ksf ðzwf  zbf Þ
mb Ib


1 df dr
þ  ½ far þ cr ð_z wr  z_ br Þ þ Ksr ðzwr  zbr Þ ð2:17aÞ
mb Ib


1 df dr
€z br ¼  ½ faf þ cf ð_z wf  z_ bf Þ þ Ksf ðzwf  zbf Þ
mb Ib


1 dr2
þ þ ½ far þ cr ð_z wr  z_ br Þ þ Ksr ðzwr  zbr Þ ð2:17bÞ
mb Ib
48 Handbook of vehicle suspension control systems

Using filtered white noise w1 and w2 as the road inputs, the road input equations for
the front and rear wheels, respectively, are
pffiffiffiffiffiffiffiffiffiffiffi
z_ f 0 ¼ 2pf0 zf 0 þ 2p G0 U0 w1 ð2:18aÞ
pffiffiffiffiffiffiffiffiffiffiffi
z_ r0 ¼ 2pf0 zr0 þ 2p G0 U0 w2 ð2:18bÞ

So far we have a state vector as

Xhalf ¼ ½_z br z_ wr z_ bf z_ wf zbr zwr zbf zwf zr0 zf 0 T ð2:19Þ

Combining vehicle model equations of motion equations (2.15), (2.16a), (2.17a),

(2.17b), and road input equations (2.18a) and (2.18b), the system model and output
equation in state space form are obtained:

~ half þ BU
X_ half ¼ AX ~ half þ F
~ whalf ð2:20aÞ
~ half þ DU
Yhalf ¼ CX ~ half þ vhalf ð2:20bÞ

~ B;
where A; ~ C;~ D;
~ F~ are differential equation coefficient matrices, Xhalf is the state
vector, Yhalf is the output vector, here Yhalf was defined as (2.21), Uhalf is control
input matrix, whalf is road inputs, and vhalf is measurement noise.

Yhalf ¼ ½€z bf zbf  zwf zwf  zf 0 €z br zbr  zwr zwr  zr0 T ð2:21Þ
   
f w2
Uhalf ¼ af ; whalf ¼ ð2:22Þ
far w1

Based on the proposed model, linear optimal control theory is used here to
design the active suspension controller. For obtaining the better handling and riding
comfort, the performance index can be written as a weighted sum of mean square
values of output performance variables including body acceleration, wheel-to-body
displacement, and dynamic tires’ deflection. The weight coefficients are r1 , r2 , q1 ,
q2 , q3 , and q4 :
ð
1 T
J ¼ lim ½q1 ðzwf  zf 0 Þ2 þ q2 ðzbf  zwf Þ2
T!1 T 0
þ r1€z bf þ q3 ðzwr  zr0 Þ2 þ q4 ðzbr  zwr Þ2 þ r2€z br dt ð2:23Þ

Same as the quarter-vehicle, the optimal LQ control can be solved from Riccati
equation.

2.2.2 Nonlinearity and unmodeling dynamic description

of active suspension system
Many researchers have dealt with a linear model in developing control laws.
However, considering the inherent nonlinearities and uncertainties, it is not
 Intelligence-based vehicle active suspension adaptive control systems 49

sufficient to represent the real system with a linear model as Sections 2.2.1.1 and
2.2.1.2. In the early 1990s many studies began to consider nonlinearities, uncer-
tainties, and unmodeled parts of a real suspension system, which required the
use of a nonlinear model and some adaptive or robust form of control scheme
[4, 8, 9, 12, 13, 15, 24, 25]. In this section, the nonlinear properties are introduced
and the general nonlinear models of suspension systems are carried out.
As Hrovat remarked, for many operations, the linear system approximation
was appropriate; however, there were some situations that amplify the nonlinear
effects [4]. One is created by discrete-event disturbances, such as single bumps or
potholes, which can cause a highly nonlinear phenomenon. Another is dry friction.
Based on the quarter-vehicle model shown in Section 2.2.1.1, Kim et al. modeled
the connecting forces (e.g., spring force, damping force) as nonlinear functions
using measured data [15]. In Kim’s paper, the nonlinear spring properties are
mainly due to two parts. One is the bump stop, which restricts the wheel travel
within a given range and prevents the tire from contacting the vehicle body. The
other is the strut bushing, which connects the strut with the body structure and
reduces harshness from the road input. These two nonlinear effects can be included
in the spring force fs with nonlinear characteristics versus suspension rattle space
ðzw  zb Þ. Based on the measured data in Reference 15, Kim modeled the spring
force fs and the damping force by the high-order polynomial functions. The spring
force was described as a third-order polynomial function shown as

fs ¼ fsl þ fsn ¼ k1 Dx þ ðk0 þ k2 Dx2 þ k3 Dx3 Þ ð2:24Þ

where fsl is the linear part of spring force and fsn is the nonlinear part of spring
force. The coefficients can be obtained from fitting the experimental data.
Also the damping force fd was modeled as a second-order polynomial function
by fitting the measured data, shown as below:

fd ¼ fdl þ fdn ¼ c1 D_x þ c2 D_x 2 ð2:25Þ

where the fdl is the linear part and the fdn is the nonlinear part of damper force, the
coefficients can be obtained from fitting the experimental data.
Except for the nonlinear properties presented by the spring force and damping
force, the vertical tire force was highly nonlinear, especially when the load con-
dition seriously changed. Even the vertical tire force became zero when the tire lost
contact with the road. Kim et al. modeled the tire force as

ftl ¼ kt ðz0  zw Þ when ðz0  zw Þ > 0

ftn ¼ 0 when ðz0  zw Þ  0

where ftl denotes the linear tire force, and ftn denotes the nonlinear tire force.
In order to show the effect of the asymmetric tire stiffness on the response of
the quarter-car model, some simulation results were shown to investigate the effect
of nonlinear tire force under the different amplitudes of road input [15]. From the
results, it was clear that vehicle nonlinearities should be considered in developing a
50 Handbook of vehicle suspension control systems

more accurate system model, from which a more reliable control algorithm can be
developed.
In this chapter, two kinds of suspension system nonlinear models are provided
for the controller design and performance analysis. Considering the nonlinear
models shown in (2.24) and (2.25), the active suspension system can be written as a
MIMO nonlinear model:

X_ ¼ FðX Þ þ BU þ d ð2:26Þ
where FðX Þ is a nonlinear function including the nonlinear forces fs ; ft , and fd ; U is
the input of the suspension system; and d is the unknown external disturbance.
The other nonlinear model can be described as a hybrid model with linear part
and nonlinear part:

X_ ¼ AX þ BU þ d~ ð2:27Þ
where AX þ BU is the linear model of suspension system based on fsl , fdl , and ftl ;
d~ represents the nonlinear and uncertain model of suspension system.

2.3 Adaptive fuzzy control

The control performance of a traditional controller greatly depends on the accuracy
of the known system dynamic model according to Section 2.2.1.1. In order to meet
practical requirements in an active suspension system, it is crucial to derive or to
identify an appropriate model for the traditional controller design. Estimating
uncertain effects is even more challenging due to the random noise occurred by
road inputs. Hence, some model-free intelligent controllers were introduced to
solve these problems, e.g., fuzzy logic controller (FLC) [26–30]. The FLC is
credited with being an adequate methodology for designing robust controllers that
are capable of delivering a satisfactory performance in the face of uncertainty and
imprecision. As a result the FLC has become a popular approach to nonlinear and
uncertain system control in recent years.
There are different ways to construct FLCs for vehicle suspension control, with
the most common method to construct the FLCs by eliciting the fuzzy rules and its
membership functions based on experts’ knowledge or experience. Then there will
be linguistic and numerical uncertainties in the experts’ knowledge or experience.
However, it is hard to handle these uncertainties by using fixed fuzzy rules and
membership functions. In order to overcome this weakness, adaptive FLCs were
designed to self-tune the fuzzy rules or membership functions [26–29, 31, 32].
Recently, with the development of type 2 fuzzy reasoning and control theory
[33–35], Cao et al. [36] studied the adaptive type 2 fuzzy control and optimization
on active suspension system. In this section, the adaptive FLC designs and appli-
cations on active suspension systems are reviewed.
The key components of an FLC are a set of linguistic fuzzy control rules and
an inference engine to digest these rules. These fuzzy rules offer a transformation
between the linguistic control knowledge of an expert and automatic control
 Intelligence-based vehicle active suspension adaptive control systems 51

S1
S3 S2
Rule base

uk = uk–1 + ∆uk Defuzzy Fuzzy inference Fuzzy

∆
ce

Fig. 2.4 The adaptive FLC scheme in Reference 29

strategies of an actuator. Every fuzzy control rule is composed of an antecedent and

a consequent. The structures and parameters of control rules dominate the perfor-
mance of fuzzy control. From the control point of view, it is crucial that related
parameters or structures are modified automatically by evaluating the results of
fuzzy control. For instance, Huang and Chao [29] proposed an adaptive FLC for an
active suspension system. This adaptive FLC scheme is shown in Fig. 2.4. The
inputs of FLC are the vertical position error and error change of the vehicle sprung
mass. Its output is the control voltage increment. The antecedents’ membership
functions consist of 11 equal triangular-type functions. The voltage increment
membership function is a set of 15 equal triangular-type functions. Its self-tuning
property is implemented by adjusted scaling factors S1 ; S2 ; and S3 . That is to say
that the membership functions are adapted to improve the FLC performance. Its
121 fuzzy rules are employed to suppress the sprung mass vibration amplitude due
to road inputs.
In order to evaluate the fuzzy control system, a two-DOF quarter-vehicle
suspension model was established. The suspension mechanism includes a spring, a
mass, and a hydraulic control loop. A hydraulic servo system was used to generate
various road surfaces, and an optical linear scale and a linear potentiometer were
employed to measure the sprung mass and road surface vertical displacements,
respectively. Based on this realistic suspension model, the dynamic response of the
active suspension system was provided for vehicle ride performance on a rough
concave–convex road with 25 mm obstacles. The maximum displacement of the
vehicle body is less than 5 mm and it converges within 0:5 s. The control signal was
very smooth and easy to employ in the practical vehicle. However, its adjusted
scaling factors were chosen by experiments and many simulations, which limited
the flexible and adaptive abilities of the FLC. In order to overcome this problem,
researchers have compensated this type of adaptive FLCs by employing the non-
linear optimal algorithms: they have employed GA and/or ANNs to self-tune the
parameters of their membership functions and fuzzy rules. These kinds of adaptive
FLC will be covered in Section 2.7.
52 Handbook of vehicle suspension control systems

2.4 Adaptive fuzzy sliding-mode control

Sliding-mode control (SMC) nowadays enjoys a wide variety of application areas,
such as general motion control applications and robotics, process control, aerospace
applications, and vehicle active suspension systems. The main reason for this
popularity is its attractive properties including good control performance for non-
linear systems, applicability to multiple-input-multiple-output (MIMO) systems,
and well-established design criteria for discrete-time systems. Note that its most
significant property should belong to its robustness. Loosely speaking, when a
system is in a sliding mode, it is insensitive to parameter changes or external dis-
turbances [37]. However, SMCs also suffer from the following disadvantages in
practical application. First, it is the problem of chattering, which is the high-
frequency oscillations of the controller output, which is brought by the high-speed
switching for the establishment of a sliding mode. Chattering is very undesirable
and dangerous in practice because it may excite unmodeled high-frequency
dynamics resulting in unforeseen instabilities. Second, an SMC is extremely vul-
nerable to measure noise since its input depends on the sign of a measured variable
that is very close to zero. Third, the SMC may employ unnecessarily large control
signals to overcome the parametric uncertainties. Last, there exists difficulty in the
calculation of what is known as the equivalent control. The integration of an FL
system in an SMC has been witnessed in many successful applications where an
attempt to relieve the implementation difficulties of the SMC are made via the
addition of the FL system [37–39]. On the other hand, some significant research
works have originated due to different difficulties, i.e., the difficulties in carrying
out a rigorous stability analysis of FCs.
The design of an SMC involves two steps. The first step is to select switching
hyperplanes called sliding surface to prescribe the desired dynamic characteristics
of a controlled system. The second step is to design discontinuous control such that
the system enters a sliding surface and remains in it. Regarding the system given in
(2.26), the sliding surface S is selected generally as

SðX Þ ¼ GX ¼ 0 ð2:28Þ
where SðX Þ denotes a set of switching hyperplanes, and G is a constant q  n
matrix to be determined.
The main object in an SMC is to force the system states to the sliding surface.
Once the states are on the sliding surface, the system errors converge to zero with
an error dynamics dictated by the matrix G. More details about the sliding-mode
controller design can be found in reference paper [37]. Here the total control of
SMC is given as

U ¼ Ueq þ Usw ð2:29Þ

where Ueq is the equivalent control, and Usw is always called switch control.
Generally speaking, two steps are required for an SMC design, one is to select
an approximation model such that the system trajectory exhibits desirable behavior
 Intelligence-based vehicle active suspension adaptive control systems 53

when confined to the model; the other is to find feedback gains so that the system
trajectory intersects and stays on the approximation model. In practical systems,
these conditions will be constrained. In the last two decades, fuzzy logic has been
employed to improve SMC in terms of efficient and practical issues. Two types
of fuzzy SMC are introduced in this section. They are employed to solve two
SMC weaknesses, i.e., alleviation SMC chattering and modeling the nonlinear or
uncertain characteristics of practical systems.

2.4.1 Alleviating SMC chattering

Fuzzy logic is employed to self-tune the discontinuous switching control law in
order to overcome the chattering phenomenon in SMC. Consider the switching
control law in terms of (2.35), which has two parameters G and K to be optimized
[37]. Their effects on the system performance are shown in Fig. 2.5. Parameter G
determines the slope of the sliding line, which means the larger the G, the faster a
system response is. Due to the fact that an over-large value of G can cause over-
shoot or instability, it would be advantageous to adaptively vary its slope in such a
way that the slope is increased as the magnitude of its error gets smaller. Curve
labeled ‘‘1’’ corresponds to the case when K is large. The system states reach the
sliding line in a short time, but overshoot it by a considerable amount. Curve
labeled ‘‘2’’ reflects the case with a small K parameter. Neither curve 1 nor curve 2
is desired. Curve ‘‘3’’ can be obtained via fuzzy adaptive algorithms in which
parameter ‘‘K’’ is increased only when the states are close to its sliding line.
For instance, Chen et al. proposed a fuzzy adaptive sliding-mode controller for
an active suspension system [40]. The proposed quarter-car active suspension
model was as (2.27). For the design of SMC, a reduced-order dynamic model was
used and the state variables were x1 (the suspension deflection) and x2 (sprung mass
velocity). Its sliding surface was defined as

SðX Þ ¼ GX ¼ x2 þ lx1 ¼ 0; l > 0 ð2:30Þ

de de
dt dt
e

S=0
e

2
3

Fig. 2.5 Effects of parameters G and K [37]

54 Handbook of vehicle suspension control systems

Likewise, the SMC control Ueq and UN were chosen as

Ueq ¼ b1 ½a1 x1  ða2 þ lÞx2 ; UN ¼ b1 KsgnðSÞ ð2:31Þ

Note that the actual inputs of the proposed fuzzy adaptive SMC controller are
S and its derivation S_ . The output was the hitting control. Fuzzification and
defuzzification stood for an interface between the crisp values of the reality and the
linguistic values of the inference. The controller was organized at two levels. At
the basic level, the conventional fuzzy control rule sets and inference mechanism
were constructed to generate a fuzzy control scheme. At the supervising level, the
control performance was evaluated to modify system parameters, especially for
adaptively tuning its scaling factors. The proposed fuzzy control rules were out-
lined in Reference 40.
In order to investigate an active suspension performance based on the above-
mentioned fuzzy SMC, a pseudo-random disturbance road input was employed to
test robustness of the controller under the condition that spring mass disturbance
was increased by 30% and damping coefficient and spring stiffness were decreased
by 30% from the nominal values. The simulation results demonstrated that the
controlled suspension deflection was smaller than its counterpart of an LQR opti-
mal control but larger than that of a conventional SMC. Regarding the riding
quality, the fuzzy SMC achieved the best performance of sprung mass acceleration.
The simulation results also illustrated that road-handling ability maintained by the
fuzzy SMC outperformed that of an LQ controller and a conventional SMC. Similar
conclusions were also drawn for the perturbed conditions.
Additionally, Yun et al. also proposed a fuzzy adaptive sliding-mode controller
for an active suspension system [20]. The main difference from Chen’s research
is the way in which a sliding surface is constructed. In Yun’s paper, the sliding
surface was constructed on the basis of conventional sliding surface s and its
derivative s_ as
s ¼ s_ þ ls ð2:32Þ
where l was a positive value, and its Lyapunov stability condition must be satisfied:
V_ ¼ ss_ < 0 ð2:33Þ
The equivalent control can be obtained:

U_ eq ¼ ðGBÞ1 ½ðGA þ lGÞAZ þ ðGA þ lGÞBU  ð2:34Þ

U_ N ¼ ðGBÞ1 esgnðsÞ ð2:35Þ
Then the SMC control output was achieved:

U_ ¼ U_ eq þ U_ N ð2:36Þ
Finally it led to the controller output:

UðnÞ ¼ U ðn  1Þ þ U_ ðnÞ ð2:37Þ

 Intelligence-based vehicle active suspension adaptive control systems 55

‫׀‬σ‫׀‬
S‫׀‬σ‫׀‬
Fuzzy adaptive
SεF
‫׀‬σ‫׀‬ tuning controller εF
S‫׀‬σ‫׀‬

State variables Z Hitting control

law UN
+ U
+
Equivalent control
law UE

Fig. 2.6 The fuzzy adaptive controller scheme in Reference 20

The scheme of fuzzy adaptive tuning controller is shown in Fig. 2.6. The
simulations in the time domain and the frequency domain were carried out on a
quarter-car active suspension system. In the time domain analysis, the comparison
between an LQG controller and the fuzzy adaptive SMC controller showed that the
proposed controller can significantly decrease its sprung mass acceleration from
peak value to zero. However, the proposed fuzzy adaptive SMC simultaneously
needed higher active forces than the LQG controller. In the frequency domain
analysis, the fuzzy adaptive SMC improved the frequency response from the road
input to the sprung mass acceleration, especially in the range of frequency 4–8 Hz.
Based on ISO 2361, the human body is very sensitive to vertical vibration in the
range of frequency 4–8 Hz. That is to say that the proposed controller can sig-
nificantly improve the ride quality. Considering the existence of uncertain para-
meters, the sprung mass was assumed to change in a bounded range of 50%. The
simulation results demonstrated that the maximum acceleration of the active sus-
pension using the proposed controller was an average of 54% smaller than a passive
suspension system.

2.4.2 FL controller complementary to SMC for system

nonlinearity and uncertainty
Referring to a traditional SMC design, the equivalent control law always depends
on its system model, due to the fact that an exact system model is very expensive to
achieve for more complex nonlinear system. A practical method for a nonlinear
problem is linearized around given operation points such that the well-developed
linear control theory can be applied into the local region with apparent ease. But it
leads to a new problem of how to aggregate each locally linearized model into a
global model that represents the corresponding nonlinear system. Fuzzy logic offers
a solution to the problem without the need of a mathematical model and constant
gain limitation [41]. Huang et al. [42] proposed an adaptive fuzzy sliding-mode
controller (AFSMC) for an active vehicle suspension system. Fuzzy logic control
was employed to approximate the nonlinear function of equivalent control law, Ueq :
The voltage output of an actuator in each sampling step was derived from fuzzy
inference, instead of from the nominal model at the sliding surface. It significantly
diminished the chattering phenomenon of the traditional SMC.
56 Handbook of vehicle suspension control systems

The input signal of this type of fuzzy logic control was sliding surface variable,
S, in terms of its sprung mass position and velocity deviations. Its output signal was
control voltage, U, which was the output of the hydraulic servo actuator. Its fuzzy
input variable S consisted of 11 equal-span triangular membership functions, which
were employed for the fuzzy output variable U through 11 fuzzy inference rules.
The tunable consequent parameters of those peaks of the triangular membership
functions were initialized with zero by default. A novel online parameters’ tuning
algorithm was proposed to adjust the consequent parameters for monitoring the
system control performance. A quarter-car two-degrees-of-freedom active suspen-
sion system was designed and built for investigating its dynamic performance and
control effect. The suspension system was tested under three different conditions.
One was a rough road with 40 mm amplitude sinusoidal wave, and the other two
were a rough concave–convex road with a randomly dynamic 40 mm height and a
rough road with a random amplitude. The experimental results showed that the
proposed AFSMC had significantly suppressed the sprung mass position oscillation
amplitude. In addition, the control voltage was smooth and the converging
speed was fast.
Additionally, Kucukdemiral et al. proposed a fuzzy logic method to handle the
nonlinear system model and uncertain disturbance for an active suspension system
[43]. The control U was given by u ¼ ufz þ uvs  uvs , denoting the switching control,
which improved by a boundary layer, alleviated the chattering; ufz was obtained
from FLC with the input S. To evaluate the proposed controller, the simulation
environment was controlled as follows: vehicle speed was 72 km/h and two types
of road surfaces were employed for controller performance evaluation including a
standard bump-type surface profile with 10 cm length  10 cm height and a random
road profile generated to simulate stabilized road with 1 cm  1 cm pebbles. Four
types of controllers were employed on the active suspension system. When the
standard bump-type surface profile was used, the proposed controller clearly pro-
duced the shortest response time of 0.85 s and the lowest peak value of 0.4 cm.
Under the condition of random road input, the AFSMC had overwhelming success
over other controllers. Besides, since it has a single input FLC as the main con-
troller, the rule base of FLC drastically decreased when it was compared with
traditional FLCs.

2.5 Adaptive neural network control

Due to its nonlinear mapping and learning ability, neural networks (NNs) have been
one of the dominant methods for designing robust, adaptive, and intelligent control
systems [44]. For further information on neural network (NN) control systems
please refer to References 45 and 46.
An adaptive nonlinear controller is required for the nonlinearity and uncer-
tainty during operation in an active suspension system. For instance, Guo et al. [47]
designed an adaptive controller with an NN-based identifier to control a semi-active
suspension with a magneto-rheological damper based on a quarter-vehicle model.
 Intelligence-based vehicle active suspension adaptive control systems 57

TDL

TDL

Reference u Nonlinear q
NNC vehicle
suspension ei
eu TDL
qp
eq
NNI
TDL
ed
eu

Fig. 2.7 The scheme of indirect adaptive control based on NNs in Reference 47

The NN control system scheme is shown in Fig. 2.7. In principle, the direct neural
network control takes the error between the ideal reference signal and the system
response as the error of back propagation. However, this error does not offer good
information for updating the weights of neural networks because of potential
uncertainty on the nonlinear model with nature and random disturbance. In Guo’s
paper, an indirect adaptive NNs’ control strategy was proposed to approximate the
input error. The structure of neural network controller is shown in Fig. 2.7. The
NNC was the neural network controller, the NNI was the neural network identifier,
and the TDL was tapped delay. Due to the unavailability of the inverse model of the
nonlinear dynamic system, not only did the NNI trace the system response, but it
also calculated the back propagation error for the NNC. The topological structure of
the NNC consisted of three layers with 4  9  1 nodes, including one hidden layer.
The NNI structure was the same as the NNC. The sigmoid function served as the
activation function for both the hidden and the output layers; clearly the back
propagation algorithm was used to update the weights.
For evaluating the adaptive NN-based control system, numerical simulations
and experiments were carried out for the quarter-vehicle rig equipped with a
magneto-rheological damper. The fundamental natural frequency of the quarter-
vehicle model was chosen as 1.8 Hz and the road profile was given based on the
road classification of the ISO database. The numerical simulation and experiment
results convincingly showed the vertical acceleration of vehicle body to be con-
siderably reduced with the indirect NNs controller than the traditional NNs
controller. For example, the root-mean-square (rms) acceleration of the vehicle
body subject to the random road disturbance of C grade was reduced by 38.2%
when the direct NNs’ controller was used and by 55% when the indirect adaptive
NNs controller was implemented in numerical simulation. In comparison with
passive suspension, the semi-active suspension with indirect adaptive NNs con-
troller reduced the acceleration of the vehicle body under the sinusoidal road
excitation of C grade by 41% in the experiment. On the other hand, the indirect
adaptive NNC worked very fast since the neural networks included only single
58 Handbook of vehicle suspension control systems

hidden layer and the neural network identifier received good training before the
experiments took place.

2.6 Genetic algorithm-based adaptive optimization

and control
Genetic Algorithms (GAs), one kind of stochastic global optimization techniques,
have been successfully applied in a variety of research and industrial fields, espe-
cially in optimization and control [48–56]. For instance, GAs have demonstrated
their effectiveness in multi-peak problems with local optimum solutions with
approval in robust search around complex spaces. The main differences between
GAs and conventional optimization and search procedures are: (1) work with a
coding set of the parameters, not the parameters themselves; (2) search from a
population of points, not a single point, and are capable of handling large search
spaces; (3) use probabilistic transition rules, rather than deterministic ones [37].
However, it also needs to be pointed out that the main disadvantage of GAs is that
the optimal speed is too slow to use in real-time applications.
Considering the control strategy in active suspension systems, Baumal et al.
[52] utilized the GA in a five-DOF half-vehicle model. In their research, all the
involved parameters comprised one constraint optimal description with eight
unknown parameters and seven constraints. That means the active control and
passive mechanical parameters were the designed variables to be optimized. Two
active elements provided forces proportional to the absolute vertical velocity of the
points on the car body directly above the rear and front wheels. These devices,
characterized by proportionality constants cf and cr , were known as skyhook
dampers. The design variables were the set: fxg ¼ fk1 ; c1 ; k3 ; c3 ; cr ; k4 ; c4 ; cf g.
And the constraints were obtained from the three performances of vehicle sus-
pension systems: (1) ride comfort; (2) road-holding ability; and (3) the suspension
working space. Two constraints were for the body acceleration and the seat
acceleration. The other five constraints were for the deflections of seat, suspension
and tire. Given the optimized initial set, there were three steps to implement the
genetic algorithm. The algorithm stopped when the maximum fitness design com-
prised at least 30% of a newly created generation. The reproduction stage itself was
a simulation of the survival of the fittest designs. Moreover, in order to improve the
efficiency of the GA, the binary strings and fitness values for each unique design of
the current generation were stored in a linear search look-up table. If a design string
in the next generation matched one in the table, then the fitness did not have to be
re-calculated. This significantly avoided Gas’ weakness by improving computing
time, especially for expensive fitness evaluations. With five independent runs of the
GA, the optimal values were obtained and compared with the local optimization
search technique and the passive suspension design. The results showed that
the proposed GA can carry out the best parameters with the least computing
time among the three methods. The active and passive suspension system seat
acceleration responses were compared to evaluate its dynamics performance.
 Intelligence-based vehicle active suspension adaptive control systems 59

The response of the active system showed that the road disturbance had little effect
on the seat acceleration, and indicated that GAs had strong potential to incorporate
global optimization methods for suspension system design.
Tsao et al. [55] also proposed an active suspension force controller using GAs
with maximum stroke constraints based on their former research [53, 54]. In con-
trast to the traditional approach, the maximum absolute values of suspension
strokes were employed in the objective function to achieve better ride comfort
within the stroke limitation. GA was employed to search for the parameters of
damping ratios and spring constants to achieve an optimum trade-off among ride
comfort, handling quality, and suspension stroke limitation simultaneously. Two
driving conditions were tested on the active force controller. One was a steep ramp
road with forward speed V ¼ 100 m/s and the other was a sinusoidal bump road
with V ¼ 40 m/s. The simulations were carried out for the three cases in each
driving condition. Each case was ended after 500 generation runs. The comparisons
of the performance among these cases showed that the proposed force controller
using GA achieved great ride and handling quality while the suspension stroke was
restricted to be less than or equal to the passive system. In terms of the dynamic
performance, the heave and pitch angle motions of the suspension system were
shown and compared with the passive suspension system. Especially considering
the comparison of the suspension displacement, the summation of the quadratic
values of the suspension displacements in the active system was larger than that of
the passive suspension. But the maximum displacement was smaller and the
vibration had been absorbed during the transient period. These results can explain
why the maximum absolute value instead of the summation quadratic form of
suspension displacement can achieve better performance.

2.7 Adaptive control integration

Control strategies are reviewed in this section based on the combination of pre-
sented methodologies in previous sections.

2.7.1 Adaptive neuro-fuzzy control

Much attention has been paid to the combination of NNs and fuzzy systems [57]
with a focus on combining fuzzy systems with NN learning techniques, espe-
cially for NN-fuzzy controller. The advantage is that the fuzzy systems can
compensate the tuning ability of their rules by using the learning algorithms of
NNs; on the other hand, the NN system can also improve the transparency and
interpretability by rule-based fuzzy reasoning construction. Generally speaking,
an NN-fuzzy system can be viewed as a special three-layer feed-forward neural
network, and the fuzzy rules are trained by an NN algorithm. With both advan-
tages of neural network and fuzzy logic, the neuro-fuzzy system had been suc-
cessfully employed to solve a wide range of industry problems, especially on
nonlinear and uncertain systems.
60 Handbook of vehicle suspension control systems

For instance, Dong et al. employed an adaptive NN-fuzzy controller for a

quarter-vehicle magneto-rheological suspension system [58]. This controller con-
sisted of an FNNC and a TDC. The FNNC calculated the control force according to
error and the change of the error, and the TDC was a neural network model that
predicted for compensating the suspension’s time-delay. For the quarter-vehicle
model, the input was the damper force, the output was the sprung mass vertical
acceleration, and road input was treated as a disturbance. In the FNNC scheme,
where two linguistic variables were input into the network and seven fuzzy sets
were defined for each input as NB, NM, ZE, PS, PM, PB in the first layer; the
second layer included 14 neurons to correspond to all the fuzzy sets; and the third
layer contained 49 neurons to do the fuzzy reasoning based on the defined fuzzy
rules. The simulation and experimental results showed that the proposed FNNC
with TDC can significantly reduce the acceleration peak value and decreased
42.3% in comparison to a passive suspension system.
Additionally, Wu et al. proposed a fuzzy controller based on the neuron-fuzzy
model for a half-vehicle active suspension system [59]. The half-vehicle active
suspension was modeled as a nonlinear system including heave, pitch, and motion
of the front and rear wheels. The proposed neuron-fuzzy network was a self-
organizing inference network with six layers to derive the corresponding T–S fuzzy
model. The learning structure included both precondition and consequence identi-
fication of fuzzy IF–THEN rules. Based on supervised learning algorithms, the
parameters of linear equations in the consequent parts were adjusted by recursive
least squares (RLS) algorithms, and the parameters in the precondition part were
adjusted by back propagation algorithm to minimize a given cost function. Based
on the T–S fuzzy model, a fuzzy controller was designed to get the optimal active
force. The simulation results showed the proposed optimal fuzzy controller can
improve the ride comfort by minimizing both the displacements and accelerations
of the vehicle center and the pitch angle simultaneously.

2.7.2 Adaptive genetic-based optimal fuzzy control

Due to the fact that IF–THEN rules in a fuzzy inference system are not always
available, automatic design methods and rule acquisition procedures for fuzzy
systems are required and have been proposed mostly based on GA and/or NNs in
the past four decades. The key advantage of the hybrid system combining GA and
FL is that almost all the tasks of fuzzy system design can be accomplished auto-
matically. Thanks to the global optimal ability, FL parameters of inference rules
and membership functions are able to be determined by a hybrid system itself. For
GA-fuzzy control systems please refer to References 60–63.
Nawa et al. studied a GA-fuzzy control system with the aid of pseudo-
bacterial GAs (PBGAs) and employed this controller to an active suspension system.
Its encoding method is demonstrated in Fig. 2.8. Differing from the traditional
canonical binary encoding, the parameters were put into the chromosome, each
of which encoded the rules of fuzzy system. Since every rule contained the
 Intelligence-based vehicle active suspension adaptive control systems 61

X1 X2 X3 Y

c = 4.238 w = 3.432 c = 3.428 w = 1.213 c = 1.382 w = 0.381 c = 12.112 w = 9.141

Rule 1 Rule 2 Rule 3 Rule 4 Rule 5 Rule 6 Rule n

Fig. 2.8 An example of the fuzzy system encoded in a chromosome in Reference 64

information of antecedent and consequent variables, each chromosome encoded

the parameters of the membership functions. Triangular-type membership func-
tions were employed so that the parameters of the membership function were in
pairs of center and width as shown in Fig. 2.8. This encoding method gave a high
degree of-freedom for the GA, which can optimize the variables to be employed
in the rules, the rules themselves, and the parameters of membership functions.
Therefore, this encoding was desirable to simultaneously evolve the rules and the
membership functions, minimizing the probability of arriving at a local optimal
point.
The GA algorithm can be briefly described as follows: (1) generation of the
initial population; (2) genetic operations: mutation–evaluation–selection–replacement;
(3) crossover and production of the new generation. An adaptive method was used
in a crossover operation instead of randomly deciding the chromosomes’ cutting
points. The adaptive crossover operator took into account the moving average of
the degrees of truth values of the fuzzy rules when deciding where to cut the
chromosome. The moving average was defined as the average of the accumulated
truth values of the rules. The accumulated truth value of a fuzzy rule was the sum of
the truth values for each one of the entries in the training data. It was a measure of
quality. If a rule possessed a high value of accumulated truth, it meant that the rule
was intensively and frequently triggered during the evaluation process. Conse-
quently, this was an indication of the utility and possible effectiveness of that rule.
On the other hand, if a rule possessed a low value of accumulated truth, this was an
indication that the rule did not play an important role in the system. Four approa-
ches were employed on the semi-active suspension control system. The first
method was a GA with fixed membership functions as defined in Reference 49. The
second method is a GA with the possibility of defining the membership functions
and rules of a fuzzy controller simultaneously. The third approach used the PBGA
with a traditional crossover operator, and the fourth approach was the PBGA with
adaptive crossover operation. The simulation results showed the proposed adaptive
PBGA fuzzy controller worked well to find out better rules and obtained the best
performance of these four control strategies. The results also indicated that this
PBGA fuzzy controller focused more on the actuation but the encoding methods
increased the total number of membership functions in the system.
62 Handbook of vehicle suspension control systems

2.7.3 GA–NN combined control

A combination of a genetic algorithm and a neural network was employed to design an
active suspension controller by Tang and Zhang [65]. The GA searched for the opti-
mal acceleration of the vehicle body, which served as the objective output of the NN
control system. The neural network had two hidden layers and the input, hidden,
and output neurons were 1, 10, 3, and 1, respectively. An adaptive leaning rate was
applied to decrease the training by keeping the learning reasonably high, while
ensuring stable learning. The input of the NN was the time response of the acceleration
of the sprung mass, and the objective output was the optimized suspension control
force. The proposed GA–NN combined controller and an LQG controller were
employed to evaluate the control performance. The simulation results demonstrated
that the NN controller with optimal acceleration parameters computed by GA-based
optimization provided better ride comfort in the time domain.

2.8 Concluding remarks

Computational intelligence-based adaptive control approaches are required due to the
real-time, nonlinear, and uncertain nature properties of active suspension systems.
This chapter provided an account of state-of-the-art of adaptive ASCSs with intelli-
gent methodologies. Their advantages and disadvantages are concluded based on
theoretical analysis, analyzing simulations, and the experimental results of the
reviewed systems. In summary, the fuzzy control systems with learning and adaptive
capability can be used to solve most of modeling problems and uncertain disturbance
of active suspension systems. But the control stability analysis is also a bottleneck for
the application of fuzzy control systems. Sliding-mode controller with fuzzy logic
system has been studied for integrating the advantages of transferring human expert
knowledge and stability verification. However, these designs are always complex and
the tuning parameters are not easy to operate by the engineers. From the point of
adaptive ability, the NN and GA also have shown many advantages in suspension
systems by simulations and applications. In addition, the combination of these meth-
ods is expected to bring better performance. Simultaneously, these hybrid systems are
shown poor interpreting ability and difficult to be evaluated in same test case.
In what follows, we enumerate some open questions and scientific problems
that suggest future research:
1. Employing intelligent control based on the full-vehicle model or 3D model.
Most of the reported research on active suspension intelligent control has stu-
died the suspension performance under the quarter-vehicle or half-vehicle
model. A comprehensive consideration of a full car, 3D model will bring fur-
ther distinct functional and safety-related benefits. Also a full-vehicle model
will be convenient to integrate the other control subsystems, such as brake
control, steering control, anti-roll control, etc., to a hybrid intelligent system
and will benefit analysis of a unit vehicle performance.
2. Integration of multi-objective optimization methods and fuzzy logic reasoning.
Considering the trade-off between riding comfort and road-handling quality,
 Intelligence-based vehicle active suspension adaptive control systems 63

the optimal objective will be changed with the requirements of different road
surfaces. Then the adaptive multi-objective optimization methods with high
real-time computing efficiency need to be researched, especially for the
application of race vehicle active suspension systems.
3. Pursuing the balance of accuracy and interpreting ability in a hybrid ASCS
such as NN–GA–Fuzzy controllers, GA–NN controllers, or NN–GA–Fuzzy–
sliding-mode controller. Though hybrid intelligent systems have been widely
investigated in many domains, their future will lie in the careful integration
of the best constituent technologies beyond simply combining individual
methods.
4. Evaluating hybrid intelligent control methodology from the perspective of
practical applications. It is necessary to build an evaluating system to compare
the different intelligent systems according to application requirements such as
computing cost, number of tuning parameters, the interface to faulty diagnosis,
etc. The comparison results will be beneficial to hybrid intelligent control
system applications and to direct the future researches.

Acknowledgments

The authors would like to acknowledge financial support from British Council and
PML Flightlink Ltd. The authors would like to acknowledge the support from
National Natural Science Foundation of China, the grant No. 61203021.

References

[1] D. Hrovat, ‘‘Ride Improvements with Unsprung Mass Dynamic Absorbers

for Active Suspension Vehicles,’’ Ford Motor Company Research Reports,
vol. SR-87-120, 1987.
[2] A. Thompson and B. Davis, ‘‘Optimal Linear Active Suspensions with
Derivative Constraints and Output Feedback Control,’’ Vehicle System
Dynamics, vol. 17, no. 4, pp. 179–192, 1988.
[3] A. Thompson and B. Davis, ‘‘A Technical Note on the Lotus Suspension
Patents,’’ Vehicle System Dynamics, vol. 20, no. 6, pp. 381–383, 1991.
[4] D. Hrovat, ‘‘Survey of Advanced Suspension Developments and Related
Optimal Control Applications,’’ Automatica, vol. 33, no. 10, pp. 1781–1817,
1997.
[5] M. Nagai, ‘‘Recent Researches on Active Suspensions for Ground Vehi-
cles,’’ JSME International Journal. Series C, Mechanical systems, machine
elements and manufacturing, vol. 36, pp. 161–170, 1993.
[6] H. Taghirad and E. Esmailzadeh, ‘‘Automobile Passenger Comfort Assured
Through LQG/LQR Active Suspension,’’ Journal of Vibration and Control,
vol. 4, no. 5, pp. 603–618, 1998.
[7] A. Hać, ‘‘Adaptive Control of Vehicle Suspension,’’ Vehicle System
Dynamics, vol. 16, no. 2, pp. 57–74, 1987.
64 Handbook of vehicle suspension control systems

[8] M. Sunwoo and K. Cheok, ‘‘An Application of Explicit Self-tuning Con-

troller to Vehicle Active Suspension Systems,’’ Decision and Control, 1990.
Proceedings of the 29th IEEE Conference on, pp. 2251–2257, 1990.
[9] T. Gordon, C. Marsh, and M. Milsted, ‘‘A Comparison of Adaptive LQG and
Nonlinear Controllers for Vehicle Suspension Systems,’’ Vehicle System
Dynamics, vol. 20, no. 6, pp. 321–340, 1991.
[10] R. Saeks, C. Cox, J. Neidhoefer, P. Mays, and J. Murray, ‘‘Adaptive Control
of a Hybrid Electric Vehicle,’’ Intelligent Transportation Systems, IEEE
Transactions on, vol. 3, no. 4, pp. 213–234, 2002.
[11] A. Vahidi and A. Eskandarian, ‘‘Research Advances in Intelligent Collision
Avoidance and Adaptive Cruise Control,’’ Intelligent Transportation Systems,
IEEE Transactions on, vol. 4, no. 3, pp. 143–153, 2003.
[12] A. Alleyne, P. Neuhaus, and J. Hedrick, ‘‘Application of Nonlinear Control
Theory to Electronically Controlled Suspensions,’’ Vehicle System Dynam-
ics, vol. 22, no. 5, pp. 309–320, 1993.
[13] A. Alleyne and J. Hedrick, ‘‘Nonlinear Adaptive Control of Active Suspen-
sions,’’ Control Systems Technology, IEEE Transactions on, vol. 3, no. 1,
pp. 94–101, 1995.
[14] C. Kim, ‘‘A Comparative Study of Active Suspension Systems Using
Adaptive Self-Tuning Control and Sliding Mode Control Schemes,’’ Master
Thesis, Mechanical and Aerospace Engineering Department, North Carolina
State University, 1996.
[15] C. Kim and P. Ro, ‘‘A Sliding Mode Controller for Vehicle Active Sus-
pension Systems with Non-linearities,’’ Proceedings of IMechE Part D:
Journal of Automobile Engineering, vol. 212, no. 2, pp. 79–92, 1998.
[16] M. Sunwoo, K. Cheok, and N. Huang, ‘‘Model Reference Adaptive Control
for Vehicle Active Suspension Systems,’’ Industrial Electronics, IEEE
Transactions on, vol. 38, no. 3, pp. 217–222, 1991.
[17] L. Palkovics and P. Venhovens, ‘‘Investigation on Stability and Possible
Chaotic Motions in the Controlled Wheel Suspension System,’’ Vehicle
System Dynamics, vol. 21, no. 1, pp. 269–296, 1992.
[18] J. Slotine and W. Li, Applied Nonlinear Control, Englewood Cliffs, NJ:
Prantice-Hall, 1991.
[19] O. Kaynak, K. Erbatur, and M. Ertugnrl, ‘‘The Fusion of Computationally
Intelligent Methodologies and Sliding-mode Control – A Survey,’’ Industrial
Electronics, IEEE Transactions on, vol. 48, no. 1, pp. 4–17, 2001.
[20] Y. Zhang, Y. Zhao, H. Yang, and L. Chen, ‘‘A Dynamic Sliding-Mode
Controller with Fuzzy Adaptive Tuning for an Active Suspension System,’’
Proceedings of IMechE Part D: Journal of Automobile Engineering,
vol. 221, pp. 417–428, 2007.
[21] T. Fukuda and N. Kubota, ‘‘Intelligent Robotic Systems: Adaptation, Learn-
ing, and Evolution,’’ Artificial Life and Robotics, vol. 3, no. 1, pp. 32–38,
1999.
[22] F. Yu, J. Zhang, and D. Crolla, ‘‘A Study of a Kalman Filter Active Vehicle
Suspension System Using Correlation of Front and Rear Wheel Road Inputs,’’
 Intelligence-based vehicle active suspension adaptive control systems 65

Proceedings of IMechE Part D: Journal of Automobile Engineering, vol. 214,

no. 5, pp. 493–502, 2000.
[23] J. Cao, H. Liu, P. Li, and D. Brown, ‘‘Study on Active Suspension Control
System Based on an Improved Half-Vehicle Model,’’ International Journal
of Automation and Computing, vol. 4, no. 3, pp. 236–242, 2007.
[24] L. Li, F. Wang, and Q. Zhou, ‘‘Integrated Longitudinal and Lateral Tire/
Road Friction Modeling and Monitoring for Vehicle Motion Control,’’
Intelligent Transportation Systems, IEEE Transactions on, vol. 7, no. 1,
pp. 1–19, 2006.
[25] D. Liaw, H. Chiang, and T. Lee, ‘‘Elucidating Vehicle Lateral Dynamics
Using a Bifurcation Analysis,’’ Intelligent Transportation Systems, IEEE
Transactions on, vol. 8, no. 2, pp. 195–207, 2007.
[26] E. Yeh and Y. Tsao, ‘‘A Fuzzy Preview Control Scheme of Active Suspen-
sion for Rough Road,’’ International Journal of Vehicle Design, vol. 15,
no. 1, pp. 166–180, 1994.
[27] S. Huang and C. Lin, ‘‘Application of a Fuzzy Enhance Adaptive Control on
Active Suspension System,’’ International Journal of Computer Applica-
tions in Technology, vol. 20, no. 4, pp. 152–160, 2004.
[28] M. Rao and V. Prahlad, ‘‘A Tunable Fuzzy Logic Controller for Vehicle-active
Suspension Systems,’’ Fuzzy Sets and Systems, vol. 85, no. 1, pp. 11–21, 1997.
[29] S. Huang and H. Chao, ‘‘Fuzzy Logic Controller for a Vehicle Active Sus-
pension System,’’ Proceedings of IMechE Part D: Journal of Automobile
Engineering, vol. 214, no. 1, pp. 1–12, 2000.
[30] T. Terano, K. Asai, and M. Sugeno, Fuzzy Systems Theory and Its Applica-
tions, San Diego, CA: Academic Press Professional, Inc., 1992.
[31] J. Yang, J. Li, and Y. Du, ‘‘Adaptive Fuzzy Control of Lateral Semi-active
Suspension for High-Speed Railway Vehicle,’’ Lecture Notes in Computer
Science, Computational Intelligence, vol. 4114, pp. 1104–1115, 2006.
[32] R. Lian, B. Lin, and W. Sie, ‘‘Self-organizing Fuzzy control of Active Sus-
pension Systems,’’ International Journal of Systems Science, vol. 36, no. 3,
pp. 119–135, 2005.
[33] H. Hagras, ‘‘A Hierarchical Type-2 Fuzzy Logic Control Architecture for
Autonomous Mobile Robots,’’ Fuzzy Systems, IEEE Transactions on, vol. 12,
no. 4, pp. 524–539, 2004.
[34] R. Sepúlveda, O. Castillo, P. Melin, A. Rodrguez-Daz, and O. Montiel,
‘‘Experimental Study of Intelligent Controllers Under Uncertainty Using
Type-1 and Type-2 Fuzzy Logic,’’ Information Sciences, vol. 177, no. 10,
pp. 2023–2048, 2007.
[35] J. Mendel, ‘‘New Results About the Centroid of an Interval Type-2 Fuzzy
Set, Including the Centroid of a Fuzzy Granule,’’ Infromation Sciences,
vol. 117, pp. 360–377, 2007.
[36] J. Cao, H. Liu, P. Li, and D. Brown, ‘‘Adaptive Fuzzy Logic Controller for
Vehicle Active Suspensions with Interval Type-2 Fuzzy Membership
Functions,’’ 2008 IEEE World Congress on Computational Intelligence,
FUZZ2008, pp. 83–89, 2008, Hongkong, 2008.
66 Handbook of vehicle suspension control systems

[37] O. Kaynak, Computational Intelligence: Soft Computing and Fuzzy-Neuro

Integration with Applications, Springer, Berlin German, 1998.
[38] M. Efe, O. Kaynak, and B. Wilamowski, ‘‘Stable Training of Computation-
ally Intelligent Systems by Using Variable Structure Systems Technique,’’
Industrial Electronics, IEEE Transactions on, vol. 47, no. 2, pp. 487–496,
2000.
[39] B. Yoo and W. Ham, ‘‘Adaptive Fuzzy Sliding Mode Control of Nonlinear
System,’’ Fuzzy Systems, IEEE Transactions on, vol. 6, no. 2, pp. 315–321,
1998.
[40] C. Ting, T. Li, and F. Kung, ‘‘Design of Fuzzy Controller for Active Sus-
pension System,’’ Mechatronics, vol. 5, no. 4, pp. 365–383, 1995.
[41] C. Chen and M. Chang, ‘‘Optimal Design of Fuzzy Sliding-mode Control: A
Comparative Study,’’ Fuzzy Sets and Systems, vol. 93, no. 1, pp. 37–48,
1998.
[42] S. Huang and W. Lin, ‘‘Adaptive Fuzzy Controller with Sliding Surface for
Vehicle Suspension Control,’’ Fuzzy Systems, IEEE Transactions on, vol. 11,
no. 4, pp. 550–559, 2003.
[43] I. Kucukdemiral, S. Engin, V. Omurlu, and G. Cansever, ‘‘A Robust Single
Input Adaptive Sliding Mode Fuzzy Logic Controller for Automotive Active
Suspension System,’’ Lecture Notes in Artificial Intelligence, vol. 3613,
pp. 981–986, 2005.
[44] X. Feng, C. Lin, T. Yu, and N. Coleman, ‘‘Intelligent Control Design and
Simulation Using Neural Networks,’’ AIAA Guidance, Navigation and
Control Conference, vol. Part 1, pp. 294–299, 1997.
[45] M. Agarwal, ‘‘A Systematic Classification of Neural-network-based Con-
trol,’’ Control Systems Magazine, IEEE, vol. 17, no. 2, pp. 75–93, 1997.
[46] V. Vemuri, ‘‘Artificial Neural Networks in Control Applications,’’ Advances
in Computers, vol. 36, pp. 203–254, 1993.
[47] D. Guo, H. Hu, and J. Yi, ‘‘Neural Network Control for a Semi-Active
Vehicle Suspension with a Magnetorheological Damper,’’ Journal of
Vibration and Control, vol. 10, no. 3, pp. 461–471, 2004.
[48] L. Davis, ‘‘Handbook of Genetic Algorithms,’’ Van Nostrand Reinhold New
York, 1991.
[49] D. Goldberg, Genetic Algorithms in Search, Optimization, and Machine
Learning, Readomg, MA: Addison-Wesley, 1989.
[50] D. Goldberg, ‘‘A Meditation on the Computational Intelligence and Its
Future,’’ Illigal Report No. 2000019, Department of General Engineering,
University of Illinois at Urbana-Champaign, 2000.
[51] A. Moran and M. Nagai, ‘‘Optimal Preview Control of Rear Suspension
Using Nonlinear Neural Networks,’’ Vehicle System Dynamics, vol. 22,
no. 5, pp. 321–334, 1993.
[52] A. Baumal, J. McPhee, and P. Calamai, ‘‘Application of Genetic Algorithms
to the Design Optimization of an Active Vehicle Suspension System,’’
Computer Methods in Applied Mechanics and Engineering, vol. 163, no. 1,
pp. 87–94, 1998.
 Intelligence-based vehicle active suspension adaptive control systems 67

[53] Y. Tsao and R. Chen, ‘‘Parameters Searching for Force Control of Active
Suspension Design by Using Genetic Algorithm,’’ Automatic Control Con-
ference, Taipei, pp. 695–699, 1997.
[54] Y. Tsao and R. Chen, ‘‘Force Control for Active Suspension Design with a
Half Car Model by Using Genetic Algorithms,’’ Proceedings of Fourth
International Symposium on Advanced Vehicle Control, AVEC’98, Nagoya,
Japan, pp. 243–248, 1998.
[55] J. Tsao and R. Chen, ‘‘The Design of an Active Suspension Force Controller
Using Genetic Algorithms with Maximum Stroke Constraints,’’ Proceedings
of IMechE Part D: Journal of Automobile Engineering, vol. 215, no. 3,
pp. 317–327, 2001.
[56] J. Holland, Adaptation in Natural and Artificial Systems, Ann Arbor, MI:
University of Michigan Press, 1975.
[57] A. Nürnberger, D. Nauck, and R. Kruse, ‘‘Neuro-Fuzzy Control Based on the
NEFCON-Model: Recent Developments,’’ Soft Computing – A Fusion of
Foundations, Methodologies and Applications, vol. 2, no. 4, pp. 168–182,
1999.
[58] X. Dong, M. Yu, C. Liao, W. Chen, H. Zhang, and S. Huang, ‘‘Adaptive
Fuzzy Neural Network Control for Transient Dynamics of Magneto-
Rheological Suspension with Time-Delay,’’ Lecture Notes in Computer
Science, ISNN 2006, vol. 3972, pp. 1046–1051, 2006.
[59] S. Wu, C. Wu, and T. Lee, ‘‘Neural-Network-Based Optimal Fuzzy Control
Design for Half-Car Active Suspension Systems,’’ Intelligent Vehicles
Symposium, Proceedings, IEEE, pp. 376–381, 2005.
[60] F. Herrera, M. Lozano, and J. Verdegay, ‘‘Tuning Fuzzy Logic Controllers
by Genetic Algorithms,’’ International Journal of Approximate Reasoning,
vol. 12, no. 3–4, pp. 299–315, 1995.
[61] F. Herrera, M. Lozano, and J. Verdegay, ‘‘A Learning Process for Fuzzy
Control Rules Using Genetic Algorithms,’’ Fuzzy Sets and Systems, vol. 100,
no. 1–3, pp. 143–158, 1998.
[62] A. Homaifar and E. McCormick, ‘‘Simultaneous Design of Membership
Functions and Rule Sets for Fuzzy controllers Using Genetic Algorithms,’’
Fuzzy Systems, IEEE Transactions on, vol. 3, no. 2, pp. 129–139, 1995.
[63] L. Magdalena, O. Cordon, F. Gomide, F. Herrera, and F. Hoffmann, ‘‘Ten
Years of Genetic Fuzzy Systems: Current Framework and New Trends,’’
Fuzzy Sets and Systems, vol. 141, no. 1, pp. 5–31, 2004.
[64] N. Nawa, T. Furuhashi, T. Hashiyama, and Y. Uchikawa, ‘‘A Study on the
Discovery of Relevant Fuzzy Rules Using Pseudobacterial Genetic Algo-
rithm,’’ Industrial Electronics, IEEE Transactions on, vol. 46, no. 6,
pp. 1080–1089, 1999.
[65] C. Tang and T. Zhang, ‘‘The Research on Control Algorithms of Vehicle
Active Suspension System,’’ Vehicular Electronics and Safety, 2005. IEEE
International Conference on, pp. 320–325, 2005.
 Chapter 3
Robust active control of an integrated
suspension system
Haiping Dua, James Lamb, Weihua Lic and Nong Zhangd

Abstract

This chapter presents the study of robust active control of an integrated vehicle sus-
pension system that consists of chassis suspension, seat suspension, and driver body
models. This integrated system has five control inputs and ten control outputs and each
control input may require different feedback signals and have different saturation
limits. Taking the measurement available variables as feedback signals, an H? static
output feedback controller is designed to improve vehicle ride comfort performance in
terms of driver head acceleration under the constraints of actuator saturation, sus-
pension deflection limitation, and road holding capability. The parameter uncertain-
ties to the driver body are considered in the controller design procedure. The controller
design conditions, which are expressed as linear matrix inequalities (LMIs), are
derived by dealing with each control input separately under a common Lyapunov
function so that a feasible solution can be found. Furthermore, force tracking control
strategy is applied to implement the proposed control system using electrohydraulic
actuators. The improvement of ride comfort is evaluated by using numerical simula-
tions on the driver head acceleration responses under a typical road disturbance.

3.1 Introduction

Vehicle chassis suspension has been adopted for all passenger and commercial
vehicles to provide ride comfort, road holding, and other functions [1]. A vehicle
chassis suspension design will generally encounter conflicting requirements on ride
comfort, suspension deflection limitation, and road holding capability. To date,

a
School of Electrical, Computer and Telecommunications Engineering, University of Wollongong,
Wollongong, Australia
b
Department of Mechanical Engineering, University of Hong Kong, Pokfulam Road, Hong Kong
c
School of Mechanical, Materials and Mechatronic Engineering, University of Wollongong, Wollongong,
NSW, Australia
d
Faculty of Engineering, University of Technology, Sydney, Australia
70 Handbook of vehicle suspension control systems

three main kinds of suspensions, that is, passive, active, and semi-active suspen-
sions [2, 3], have been proposed to deal with these conflicting requirements. A
passive suspension is simple, reliable, and cost-effective. However, it cannot pro-
vide controllable damping force such that its performance is inevitably limited.
While maintaining the geometric and dynamical properties of a passive suspension
structure, active and semi-active devices are considered to be incorporated in
modern suspension structures to meet the conflicting requirements. Active and
semi-active suspensions are attracting more attention in both academia and industry
for improving vehicle ride comfort and road holding capability [2–13].
In addition to vehicle chassis suspension, seat suspension has also been
adopted in vehicles, in particular, for commercial industrial, agricultural, and other
transport purposes [14], to provide driver ride comfort, to reduce driver fatigue due
to long hour driving or exposure to severe working environment such as rough road
condition and to improve driver safety and health [15]. Study on optimisation and
control of seat suspensions for reducing vertical vibration has been an active topic
for decades. Similar to vehicle suspension, passive, semi-active, and active seat
suspensions have also been proposed [16]. The study on passive seat suspension
mainly focuses on parameter optimisation for the spring stiffness and the damping
coefficient [17, 18]. The study on active seat suspension mainly focuses on
developing advanced control strategies or applying different types of actuators to
improve seat suspension performance with taking account of issues like actuator
saturation, load variation, time delay, and reliability, for instance [19–24].
No matter whether it is vehicle chassis suspension or seat suspension, one of
the common objectives is to improve ride comfort performance. However, it is
found from literature that most of the current research on active/semi-active seat
suspension and active/semi-active vehicle chassis suspension are conducted sepa-
rately. It therefore naturally motivates the idea of integrally controlling both sus-
pensions in order to provide an enhanced ride comfort performance. To date, only a
few studies [25, 26] consider both vehicle suspension and seat suspension together
to study vehicle or seat suspension optimisation problem. On the other hand, for
vehicle chassis suspension study, sprung mass acceleration is normally used as
performance index to evaluate vehicle ride comfort in the frequency range which is
sensitive to human comfort, and for seat suspension study, a driver or passenger
rigid dummy body acceleration is often used to evaluate ride comfort. However,
neither sprung mass nor rigid dummy body can precisely reflect human biodynamic
properties. Preliminary studies [27, 22] show that it is necessary to involve a
complicated biomechanical model of human body in a seated posture to gain a
good insight into the ride comfort performance when designing a seat suspension.
Therefore, developing an integrated model which includes vehicle suspension, seat
suspension, and human body model, and designing an integrated control system
will be significant to improve human body ride comfort when sitting in a vehicle.
In this chapter, an integrated system that includes a full-car chassis suspension
(7 degree-of-freedom (DOF)), a seat suspension (2 DOF), and a driver body model
(4 DOF) is developed first. Based on this integrated model, an H? static output
feedback controller is then designed to generate the desired control forces to reduce
 Robust active control of an integrated suspension system 71

driver head acceleration under the constraints of actuator saturation, suspension

deflection limitation, and road holding capability. A static output feedback controller
is considered because not all the state variables, in particular, the state variables in
relation to the human body model, are measurement available in practice [28, 29].
In addition, the mass of the driver body may be varied when a driver’s physical
condition is changed or a different driver who has a different weight is driving the
vehicle, and parameter uncertainties may happen to the stiffness and damping coef-
ficient of each segment of driver body, of which values are in fact hard to be measured
accurately in practice; the parameter uncertainties to the driver body model will be
considered in the controller design procedure. As this is a high-order system with
multiple inputs (five control inputs: four for vehicle suspension and one for seat
suspension) and multiple outputs (ten control outputs: driver head acceleration, four
vehicle suspension deflections, one seat deflection, four tyre deflections), and each
control input may require different feedback signals and have different saturation
limitations, each control input will be dealt with separately under a common
Lyapunov function to derive the controller design conditions, which are expressed as
linear matrix inequalities (LMIs). Furthermore, the electrohydraulic actuators will be
chosen to provide the required forces due to their high power-to-weight ratio and low
cost. A two-loop control strategy, in which the outer-loop is used to provide the
desired forces which are calculated by the above designed controller and the inner-
loop is used to make the electrohydraulic actuators track the desired forces, will be
applied for the implementation of the proposed suspension. Numerical simulations
are finally used to validate the effectiveness of the proposed control system.
This chapter is organised as follows. In section 3.2, the integrated uncertain
system model is developed. In section 3.3, the control system design is presented.
The simulation results are discussed in section 3.4. Finally, conclusions are sum-
marised in section 3.5.
The notation used throughout the chapter is standard. For a real symmetric
matrix W, the notation of W > 0 (W < 0) is used to denote its positive- (negative-)
definiteness. kk refers to either the Euclidean vector norm or the induced matrix
2-norm. I is used to denote the identity matrix of appropriate dimensions. To simplify
notation, * is used to represent a block matrix which is readily inferred by symmetry.

3.2 Uncertain integrated system modelling

The integrated system model includes a full-car chassis suspension model, a seat
suspension model, and a 4-DOF driver body model as shown in Figure 3.1, where ms
is the sprung mass, which represents the car chassis; mfl , mfr , mrl , and mrr are the
unsprung masses for the front left (fl), front right (fr), rear left (rl), and rear right (rr)
wheels, respectively; mf is the seat frame mass; mc is the seat cushion mass; the driver
body is composed of four mass segments, that is, thighs m1 , lower torso m2 , high
torso m3 and head m4 ; where arms and legs are combined with the upper torso and
thighs, respectively. If and Iq are moment of inertia of the vehicle body for roll and
pitch motion, respectively; f and q are roll and pitch angles; zs , zufl , zufr , zurl , zurr ,
72 Handbook of vehicle suspension control systems

z4
Head m4
Fs k4 c4
Outputs Outputs z3
Upper torso m3
Ffr Controller Frr
Ffl k3 c3
u = Kx Frl z2
Lower torso m2

x1 x2 xi k2 c2
z1
Inputs Thighs m1
k1 c1
v zc
kc cc z Seat cushion mc
c
Fs Right Human body model
tf
zsfr zps zs zsrr
Front φ rx
tf csf ms, If, Iθ Rear csr
ksf Ffr ry Frr
lf θ lr ksr
mur zurr
zufr zsrl
zsfl Left ktr ktr
csf csr
ksf Ffl wfr Frl wrr
ksr
muf zufl xvi mur zurl
zsi
ktf
wfl Ps Fi wrl

Servo-value zui
Hydraulic cylinder
Electrohydraulic actuator

Figure 3.1 Integrated full-car suspension, seat suspension, and driver body model

Table 3.1 Parameters of the seat-driver suspension model

css damping of seat suspension kss stiffness of seat suspension

cc damping of seat cushing kc stiffness of seat cushion
c1 damping of buttock and thighs k1 stiffness of buttocks and thighs
c2 damping of lumber spine k2 stiffness of lumber spine
c3 damping of seat k3 stiffness of thoracic spine
c4 damping of cervical spine k4 stiffness of cervical spine

zf , zc , and z14 are the displacements of the corresponding masses; zrfl , zrfr , zrrl , and
zrrr are the road displacement inputs to each wheel; rx and ry represent the seat
position in relation to the centre of mass. csf , csr and ksf , ksr are damping and stiffness
of the car suspension system; ktf , ktr and ctf , ctr stand for compressibility and
damping of the pneumatic tyres; cs , css , c14 , ks , kss and k14 are defined in
Table 3.1. Ffl , Ffr , Frl , Frr , and Fs represent the controllable damping forces applied
to the car suspension and the seat suspension.
 Robust active control of an integrated suspension system 73

The state-space equation of the integrated nominal system is given as

x_ ¼ Ax þ Bw w þ Bu ð3:1Þ

T
where x ¼ pT p_ T is the state vector,

p ¼ ½ xu q x d T

q ¼ ½ zs q f T

xu ¼ ½ zufl zufr zurl zurr T

xd ¼ ½ zf zc z1 z2 z3 z4 T

T T
u ¼ FaT FsT is the control input vector, Fa ¼ ½ Ffl Ffr Frl Frr  , and
T
w ¼ ½ zrfl zrfr zrrl zrrr  is the road disturbance vector. Please see the appendix
for the details of the derivation.
In practice, all the actuators are limited by their physical capabilities, and
hence, actuator saturation needs to be considered for active control of seat sus-
pension [23] and car chassis suspension [30]. Taking actuator saturation into
account, (3.1) is modified as

x_ ¼ Ax þ Bw w þ Bu ð3:2Þ

where u ¼ sat(u) and sat(u) is a saturation function of control input u defined as

8
>
>ulim if u < ulim
<
satðuÞ ¼ u if ulim p u p ulim ð3:3Þ
>
>
:
ulim if u > ulim

where ulim is the control input limit.

In addition, the mass of the driver body may be varied when a driver’s physical
condition is changed or a different driver who has a different weight is driving the
vehicle. To make the controller have adequate performance despite the changes of
the driver’s mass, the variation of the driver’s mass will be considered. Referring to
the driver model used in this chapter, it can be seen that the driver’s mass is
composed
P of the masses of thighs, lower torso, high torso and head, that is,
m ¼ 4i¼1 mi : It is reasonable to assume that the mass variation ratio of each seg-
ment of the driver body is equal and the driver’s mass is actually varied in a range
of ½mmin , mmax , where mmin and mmax are the possible minimum and maximum
driver masses. Therefore, it is not difficult to represent the uncertain driver mass
appeared in the model as
1 1 1
¼ h1 þ h2 ð3:4Þ
m mmin mmax
74 Handbook of vehicle suspension control systems

where h1 and h2 are defined as

1=m  1=mmax 1=mmax  1=m

h1 ¼ , h2 ¼ ð3:5Þ
1=mmin  1mmax 1=mmin  1=mmax
P2
It can be seen that hi X P0; i ¼ 1; 2; and i¼1 hi ¼ 1: If we Pdefine mmin ¼
ð1  dÞm ¼ dmin m ¼ dmin 4i¼1 mi , mmax ¼ ð1 þ dÞm ¼ dmax m ¼ dmax 4i¼1 mi , where
0 < d < 1; dmin ¼ 1  d, and dmax ¼ 1 þ d, the vehicle model in (3.2) with uncertain
driver mass can be expressed as

X
2
x_ ¼ hi Ai x þ Bw w þ Bu ð3:6Þ
i¼1

where matrices Ai , i ¼ 1; 2; are obtained by replacing mj , j ¼ 1; 2; 3; 4; in matrix

A with dmin mj and dmax mj , respectively.
On the other hand, parameter uncertainties may happen to the stiffness and
damping coefficient of each segment of driver body, of which values are in fact
hard to be measured accurately in practice. To describe these uncertainties in the
model, a norm-bounded method can be used. If we use the norm-bounded matrices
DAi to represent the parameter uncertainties due to the uncertainties in the stiffness
and damping coefficients, system (3.6) is written as

X
2
x_ ¼ hi ðAi þ DAi Þx þ Bw w þ Bu ð3:7Þ
i¼1

where DAi ¼ Ha FEi represents the uncertainty caused by the uncertain stiffness
and damping coefficients on matrix Ai , Ha and Ei are known constant matrices
with appropriate dimensions, which can be defined in terms of the locations and
variation ranges of the uncertain parameters appeared in the matrix Ai , and F is
an unknown matrix function bounded by F T F p I: For description simplicity,
P P P
we define Ah ¼ 2i¼1 hi Ai , DAh ¼ 2i¼1 hi DAi ¼ 2i¼1 hi Ha FEi ¼ Ha FEh , where
P
Eh ¼ 2i¼1 hi Ei , and A
^ h ¼ Ah þ DAh , then, (3.7) is expressed as

^ h x þ Bw w þ Bu
x_ ¼ A ð3:8Þ

To deal with the saturation problem in the controller design process, the
following lemma will be used.

Lemma 3.1 [31]. For the saturation constraint defined by (3.3), as long as jujp ulim
e ,
we have
 
 
u  1 þ e up 1  e kuk ð3:9Þ
 2  2
 Robust active control of an integrated suspension system 75

and hence
     
1þe T 1þe 1e 2 T
u u u u p u u ð3:10Þ
2 2 2
where 0 < e < 1 is a given scalar.
To apply Lemma 3.1 in the next section, system (3.8) is further written as
 
^ h x þ Bw w þ 1 þ e Bu þ B u  1 þ e u
x_ ¼ A
2 2
^ h x þ Bw w þ 1 þ e
¼A Bu þ Bd ð3:11Þ
2
where d ¼ u  1þe
2 u:
To derive the main result, the following lemma is also used.
Lemma 3.2 [32]. For any matrices (or vectors) X and Y with appropriate dimen-
sions, we have

X T Y þ Y T X p eX T X þ e1 Y T Y ð3:12Þ
where e > 0 is any scalar.

3.3 Robust control system design

3.3.1 Control objectives
For the suspension design with a driver body model, the performance on ride
comfort will be mainly described by the driver head acceleration [24, 22] and,
therefore, the driver head acceleration
^ hx
zg1 ¼ €z 4 ¼ C ð3:13Þ
P2 P2 P2
where C ^ h ¼ Ch þ DCh , Ch ¼
i¼1 hi Ci , DCh ¼ i¼1 hi DCi ¼ i¼1 hi Hc FEi ¼
Hc FEh , Ci is the last row of Ai matrix and DCi represents the uncertainty, will be
defined as one of the control outputs.
On the other hand, for vehicle suspension systems, besides the ride comfort
performance, the suspension deflection limitation and the road holding ability
should be considered as well. To keep the suspension deflections to be within
their limits, the car suspension deflections zsi  zui , i 2 ffl, fr, rl, rrg, and the seat
suspension deflection zc  zps , are required to be small. Similarly, to keep the
wheel contact with ground, the wheel vertical displacements zui , i 2 ffl, fr, rl, rrg,
are required to be small so that a good road holding performance can be achieved.
Thus, we define the suspension deflections and wheel displacements as another two
control outputs, that is
T
zg2 ¼ ½ zsfl  zufl zsfr  zufr zsrl  zurl zsrr  zurr zc  zps  ¼ C2 x ð3:14Þ
76 Handbook of vehicle suspension control systems

and

zg3 ¼ ½ zufl zufr zurl zurr T ¼ C3 x ð3:15Þ

where C2 and C3 can be defined appropriately in terms of the state vector x. For
example, the first row of C2 can be defined as ½ 1 0 0 0 1 lf tf ½0118 ,
and the first row of C3 is defined as ½ 1 ½0125 .
As the three control objectives are conflicting to each other, they cannot be
optimised at the same time. To compromise these control objectives and convert
the multiple objective problem into a single objective problem in the controller
design procedure, the final control output is defined as
2 3 2 3 02 3 2 31
zg1 ^h
C Ch DCh
6 7 6 7 B6 7 6 7C
z¼6 a 2 z 7¼6
6 a C
7
7x ¼ B6 a2 C2 7 þ 6 0 7Cx
4 g2 5 4 2 2
5 @4 5 4 5A
a3 zg3 a 3 C3 a3 C3 0
02 3 2 31
Ch Hc FEh
B6 7 6 7C
¼B6 7 6 0 ^ zx
7Cx ¼ ðCz þ DCz Þx  C
@4 a2 C2 5 þ 4 5A ð3:16Þ
a3 C3 0

where a2 and a3 are weighting parameters used to provide trade-off among zg1
to zg3 :
" #T

 X
2 X
2
T T ^i
Cz ¼ ChT a2 C2T a3 C3 ¼ hi CiT a2 C2T a3 C3T ¼ hi C
i¼1 i¼1

T
T
^ i ¼ CT
C a2 C2T a3 C3T , ^ c FEh ,
DCz ¼ H ^ c ¼ HT
H 0 0
i c

To achieve good suspension performance and make the controller performing

adequately for a wide range of road disturbances, the L2 gain between the road
disturbance input w and the control output z, which is defined as

kzk2
kTzw k1 ¼ sup ð3:17Þ
w2 6¼0 kwk2

Ð1 Ð1
where kzk22 ¼ 0 zT ðtÞzðtÞdt and kwk22 ¼ 0 wT ðtÞwðtÞdt are chosen as the perfor-
mance measure. A small value of kTzw k1 generally means a small value of driver
head acceleration under the energy bounded road disturbances and reasonable
suspension deflection and road holding requirement. The control objective is to
design a controller such that the closed-loop system is asymptotically stable and the
performance measure (3.17) is minimised.
 Robust active control of an integrated suspension system 77

3.3.2 Robust controller design

In practice, not all the state variables are measurable, in particular, for high DOF
human body model where most of the state variables, such as torso displacements and
velocities, are not measurable or not suitable for measurement when a driver is driving.
Therefore, a static output feedback controller, which only uses available measure-
ments, will be designed for the integrated system to achieve the required objectives.
The static output feedback controller is designed as
2 3
K1 Cs1
6 7
6 K2 Cs2 7
6 7
u ¼ 6 . 7x ð3:18Þ
6 .. 7
4 5
K5 Cs5
where Ki , i ¼ 1; 2; . . . , 5; are the feedback gain matrices to be found, Csi ,
i ¼ 1; 2; . . . , 5; are the constant matrices used to define the available state vari-
ables for different control inputs.
To find the controller gain matrices Ki in (3.18), we now define a Lyapunov
function for system (3.8) as
V ðxÞ ¼ xT Px ð3:19Þ
where P > 0: By differentiating (3.19) and using (3.8), we obtain:

V_ ðxÞ ¼ x_ T Px þ xT P_x
 T
¼ A ^ h x þ Bw w þ 1 þ e Bu þ Bd Px
2
 
T ^ 1þe
þ x P A h x þ Bw w þ Bu þ Bd ð3:20Þ
2
Since there are five inputs in the system, and each input may use different
feedback signals and have different saturation limits, we will describe each input
separately in deriving the controller design conditions to reduce the possible
conservatism. The control input vector is defined as u ¼ ½Ffl , Ffr , Frl , Frr , Fs T ¼
½u1 , u2 , u3 , u4 , u5 T , where ui is the ith input, i ¼ 1; 2; . . . , 5; and hence
P P
Bu ¼ 5i¼1 bi ui ¼ 5i¼1 bi Ki Csi x can be obtained, where bi is the ith column of
matrix B. Then, (3.20) is replaced by
V_ ðxÞ ¼ x_ T Px þ xT P_x
( ) T
X5 
^ h x þ Bw w þ 1 þ e i
¼ A bi ui þ bi di Px
i¼1
2
( )
X 5 
T ^ 1 þ ei
þ x P A h x þ Bw w þ bi ui þ bi di
i¼1
2
78 Handbook of vehicle suspension control systems

where di ¼ u i  1þe
2 ui : By using Lemma 3.1, Lemma 3.2, and ui ¼ Ki Csi x,
we have
( " T #)
T X5
1 þ e i 1 þ ei
V_ ðxÞ p xT A^ P þ PA ^h þ bi Ki Csi P þ Pbi Ki Csi x
h
i¼1
2 2

X
5

þ wT BTw Px þ xT PBw w þ i diT di þ 1 T T

i x Pbi bi Px
i¼1
( " T #)
X
5
1 þ ei 1 þ ei
p xT ^TP
A ^h þ
þ PA bi Ki Csi Pþ x
Pbi Ki Csi
h
2
i¼1
2
"   #
X5
1  ei 2 T
T T T 1 T T
þ w Bw Px þ x PBw w þ i ui ui þ i x Pbi bi Px
i¼1
2

¼ xT Qx þ wT BTw Px þ xT PBw w ð3:21Þ

where
" T
X
5
1 þ ei 1 þ ei
^ T P þ PA
Q¼A ^h þ bi Ki Csi Pþ Pbi Ki Csi
h
i¼1
2 2
 2 
1  ei T T 1 T
þ i Csi Ki Ki Csi þ i Pbi bi P
2
and 0 < ei < 1 are given scalars, i , i ¼ 1; 2, . . . , 5; are any positive scalars.
Adding zT z  g2 wT w, g > 0 is a performance index, to the two sides of (3.21)
yields:
2 3

 QþC ^ TC^ z PBw " x #
z
V_ ðxÞ þ zT z  g2 wT w p xT wT 4 5
T 2 w
Bw P g I
" #

 x
¼ x T wT P ð3:22Þ
w
" #
QþC ^TC ^ PBw
z z
where P ¼ : It is now deduced from (3.22) that if P < 0;
BTw P g2 I
then, V_ ðxÞ þ zT z  g2 wT w < 0; and then, kTzw k1 < g with the initial condition
xð0Þ ¼ 0 [33]. When the road disturbance is zero, that is, w ¼ 0; it can be inferred
from (3.22) that if P < 0; then V_ ðxÞ < 0; and the system (3.8) with the controller
(3.18) is quadratically stable.

By pre- and post-multiplying P with diag P1 ; I and its transpose, respec-
tively, and defining Q ¼ P1 , Wi Csi ¼ Csi Q and Yi ¼ Ki Wi , the condition of
P < 0 is equivalent to
 Robust active control of an integrated suspension system 79
" #
^TC
Y þ QC ^
z zQ Bw
<0 ð3:23Þ
BTw g2 I
where
T X5  1 þ ei 1 þ ei
^ þA
Y ¼ QA ^ hQ þ CsiT YiT bTi þ bi Yi Csi
h i¼1 2 2
 2 
1 1  ei T T T
þ i Csi Yi Yi Csi þ ei bi bi
2
By using Schur complement equivalence, (3.23) is equivalent to
2 T 3
Y1 T T
Cs1 Y1 T T
Cs2 Y2  T T
Cs5 Y5 ^
QC Bw
z
6  2 7
6 7
6  2
1 1e I 0 0 0 0 0 7
6 1 7
6 7
6  2 7
6   2 2
I 0 0 0 0 7
6 1e2 7
6 7
6 7<0
6 .. 7
6    . 0 0 0 7
6  2 7
6 7
6     5 2
I 0 0 7
6 1e5 7
6 7
6 7
4      I 0 5
      g2 I
ð3:24Þ
P5
1þei 
where Y1 ¼ QA ^T þ A
^ hQ þ CsiT YiT bTi þ 1þe T
2 bi Yi Csi þ i bi bi : (3.24) is
i
h i¼1 2
further expressed as
2 3
Y2 T T
Cs1 Y1 T T
Cs2 Y2  T T
Cs5 Y5 QðCz þ DCz ÞT Bw
6  2 7
6 7
6  1 2
I 0 0 0 0 0 7
6 1e1 7
6  2 7
6 7
6   2 2
I 0 0 0 0 7
6 1e2 7
6 7
6 7<0
6 .. 7
6    . 0 0 0 7
6  2 7
6 7
6     5 2
I 0 0 7
6 1e5 7
6 7
6      I 0 7
4 5
      g2 I
ð3:25Þ
P5
1þei
where Y2 ¼ QðAh þ DAh ÞT þ ðAh þ DAh ÞQ þ i¼1 2 CsiT YiT bTi þ 1þe 2 bi Yi Csi þ
i

i bi bTi : We now need the following lemma to derive the result.

80 Handbook of vehicle suspension control systems

Lemma 3.3 [34]. Given appropriately dimensioned matrices S1 , S2 , S3 , with

ST1 ¼ S1 , then

S1 þ S3 DS2 þ ST2 DST3 < 0

holds for all D satisfying DT D I if and only if for  > 0
S1 þ S3 ST3 þ 1 ST2 S2 < 0

In fact, inequality (3.25) is equivalent to

S1 þ S3 FS2 þ ST2 F T ST3 < 0 ð3:26Þ

where
2 T T T T T T
3
Y3 Cs1 Y1 Cs2 Y2  Cs5 Y5 QCzT Bw
6  2 7
6 2 7
6  1 1e1 I 0 0 0 0 0 7
6 7
6  2 7
6 2 7
6   2 I 0 0 0 0 7
6 1e2 7
6 7
S1 ¼ 6 .. 7
6    . 0 0 0 7
6 7
6  2 7
6     5 2
I 0 0 7
6 1e5 7
6 7
6      I 0 7
4 5
2
      g I
X5  
1 þ ei T T T 1 þ ei
Y3 ¼ QATh þ Ah Q þ Csi Yi bi þ bi Yi Csi þ i bi bTi ; ST3 ¼
i¼1 2 2
h i
HaT 0    0 H ^ Tc 0 ; S2 ¼ ½ Eh Q 0    0 0 0 : By using Lemma 3.3,
we can see that the inequality (3.26) is satisfied if the following inequality holds for
h > 0:
2 3
T T T T T T
QCzT
6 Y4 Cs1 Y1 Cs2 Y2  Cs5 Y5 QEhT 7
Bw
6 þhHa H^T 7
 2 c
6 7
6   2
0 0 0 0 0 0 7
6 1 1e1 I 7
6  2 7
6 2
7
6   2 1e I 0 0 0 0 0 7
6 2 7
6 .. 7<0
6 7
6 
6
  .  0 2 0 0 0 7
7
6      2
I 0 0 0 7
6 5 1e5 7
6 7
6      ^T
I þ hHc H 0 0 7
6 c 7
4       2
g I 0 5
       hI
ð3:27Þ
P5
1þei 
where Y4 ¼ QATh þ Ah Q þ hHa HaT þ i¼1 2 CsiT YiT bTi þ 1þe T
2 bi Yi Csi þ i bi bi :
i
 Robust active control of an integrated suspension system 81
P P ^ j , Eh ¼ P2 hj Ej , and the
By the definitions Ah ¼ 2j¼1 hj Aj , Cz ¼ 2j¼1 hj C j¼1
P2
fact that hj X 0 and j¼1 hj ¼ 1, (3.27) is equivalent to
2 T
3
QC^
T T T T T T j
6 Y5 Cs1 Y1 Cs2 Y2  Cs5 Y5 Bw QEjT 7
6 þhHa H^T 7
6 c 7
6  2 7
6 7
6  1 2
I 0 0 0 0 0 0 7
6 1e1 7
6 7
6  2 7
6 2 7
6   2 1e2 I 0 0 0 0 0 7
6 7
6 7
6 .. 7<0
6    . 0 0 0 0 7
6 7
6  2 7
6     5 2
I 0 0 0 7
6 1e5 7
6 7
6 7
6      I þ ^T
hHc H 0 0 7
6 c 7
6 7
6       2
g I 0 7
4 5
       hI
j ¼ 1, 2,
ð3:28Þ
P5
1þei 
where Y5 ¼ QATj þ Aj Q þ hHa HaT þ i¼1 2 CsiT YiT bTi þ 1þe T
2 bi Yi Csi þ i bi bi :
i

In addition, from (3.18), the constraint jui j p ulim

ei can be expressed as
i

ulim i
jKi Csi xjp ð3:29Þ
ei
 2 
Let WðKi Þ ¼ xjjx Csi Ki Ki Csi xj p ulim
T T T
ei
i
, the equivalent condition for an
ellipsoid WðP, rÞ ¼ fxjxT Px p rg being a subset of WðKi Þ, that is WðP, rÞ

WðKi Þ, is given as (see [35])

 1  
P T T ulim i 2
Ki Csi Csi Ki p ð3:30Þ
r ei
By Schur complement equivalence, inequality (3.30) can be written as
2  2  1 3
ulim i P
Ki C si
6 e i r 7
6 7X0 ð3:31Þ
4  1  1 5
P
r CsiT KiT P
r I

Using the definitions Q ¼ P1 , Wi Csi ¼ Csi Q, and Yi ¼ Ki Wi , inequality (3.31) is

equivalent to
2  2 3
ulim i
6 ei Y C
i si 7
4 5X0 ð3:32Þ
CsiT YiT r1 Q
82 Handbook of vehicle suspension control systems

It is observed that the static output feedback controller design is the feasibility
problem of LMIs (3.23) and (3.32) with equality constraint Wi Csi ¼ Csi Q. The
equality constraint Wi Csi ¼ Csi Q can be equivalently converted to [36, 37]
h i
tr ðWi Csi  Csi QÞT ðWi Csi  Csi QÞ ¼ 0 ð3:33Þ

By introducing the condition

ðWi Csi  Csi QÞT ðWi Csi  Csi QÞ p mI ð3:34Þ

where m > 0; it is then equivalent to

 
mI ðWi Csi  Csi QÞT
p0 ð3:35Þ
Wi Csi  Csi Q I

by means of the Schur complement equivalence. If we assume m as a very small

positive number, then (3.35) is also an LMI and can be numerically solved.
We now state the controller design problem as follows: for given numbers
g > 0; ei > 0; r > 0; m  0, and ulim i , i ¼ 1; 2; . . . ; 5; the system (3.7) with con-
troller (3.18) is quadratically stable and kTzw k1 < g if there exist matrices Q > 0;
Wi , Yi , and scalars ei > 0; h > 0; i ¼ 1; 2; . . . ; 5; such that LMIs (3.28), (3.32),
and (3.35) are feasible. Moreover, the feedback gain matrices are obtained as
Ki ¼ Yi Wi1 , i ¼ 1; 2; . . . ; 5:
It is noticed that (3.28), (3.32), and (3.35) are linear matrix inequalities (LMIs)
with respect to g2 , hence, to minimise the performance measure g, the controller
design problem can be modified as a minimisation problem of

min g2 s:t: LMIs ð3:23Þ; ð3:32Þ and ð3:35Þ ð3:36Þ

This minimisation problem is a convex optimisation problem and can be solved by

using some standard software.

3.3.3 Force tracking control of electrohydraulic actuators

The above designed controllers will be used to calculate the desired forces needed
from the electrohydraulic actuators as shown in Figure 3.1.
The electrohydraulic actuator dynamics can be expressed as [38–41]
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sgnðxvi ÞFi
F_ i ¼ bFi  aA2s ð_z si  z_ ui Þ þ ga As Ps  xvi ð3:37Þ
As
1
x_ vi ¼ ðxvi þ vi Þ ð3:38Þ
t
where xvi is the spool valve displacement, vi is the control input voltage to
the servo-valve, and i denotes fl, fr, rl, rr, and s. As is the actuator ram area. Ps isffi
pffiffiffiffiffiffiffiffiffi
the hydraulic supply pressure. a ¼ 4be =Vt , b ¼ aCtm , and ga ¼ aCd wa 1=ra ,
 Robust active control of an integrated suspension system 83

where be is the effective bulk modulus, Vt is the total actuator volume, Ctm is the
coefficient of total leakage due to pressure, Cd is the discharge coefficient, wa is the
spool valve area gradient, ra is the hydraulic fluid density, and t is the time con-
stant of the spool valve dynamics.
Referring to [42], the control input voltage to the servo-valve of each actuator
is defined as
vi ¼ t_x vi þ xvi ð3:39Þ
It can be obtained from (3.37) that
xvi ¼ hi F_ i þ gi ð3:40Þ
where
1 aA2s ð_z si  z_ ui Þ þ bFi
hi ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , gi ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sgnðxvi ÞFi sgnðxvi ÞFi
g a A s Ps  g a As P s 
As As

By defining
Ui ¼ F_ di  Kp ðFi  Fdi Þ ð3:41Þ
where Fdi is the desired force generated by the designed controller and Kp is a
positive proportional coefficient, and let Ui ¼ F_ i , it can be seen from (3.41) that
e_ þ Kp e ¼ 0 ð3:42Þ

where e ¼ Fi  Fdi is the force tracking error, and the actuator force will expo-
nentially converge to the desired force when time approaches to infinity.
Substitute (3.41) into (3.40) and then (3.39), we have

vi ¼ ðth_ i þ hi ÞUi þ thi U_ i þ gi þ tg_ i ð3:43Þ

Therefore, the valve control input voltage can be calculated from (3.43), where Ui
is obtained from (3.41) based on the desired control force and the force tracking
error.

3.4 Numerical simulations

The effectiveness of the proposed active control of the integrated system for
improving driver ride comfort is evaluated by numerical simulations. The para-
meters for the full-car suspension are listed in Table 3.2 [20, 43], the parameters for
the seat suspension and driver body model parameters are listed in Table 3.3 [27]
and the electrohydraulic actuator parameters are listed in Table 3.4 [43].
In terms of the measurement availability of the signals in practice, the relative
velocities between sprung and unsprung masses z_ si  z_ ui , the unsprung mass dis-
placements zui , and velocities z_ ui , i 2 ffl, fr, rl, rrg will be used as feedback signals
84 Handbook of vehicle suspension control systems

Table 3.2 Parameter values of the full-car suspension model

Parameter Value Unit Parameter Value Unit

ms 1460 kg csf 946 Nm/s
Iq 2460 kgm2 csr 946 Nm/s
If 460 kgm2 ksf 20,580 N/m
muf 40 kg ksr 20,580 N/m
mur 35.5 kg ktf, ktr 175,500 N/m
lf 1.011 m tf 0.761 m
lr 1.803 m tr 0.755 m
rx 0.3 m ry 0.25 m

Table 3.3 Parameter values of the seat and driver body model

Mass Value Damping Value Spring Value

(kg) coefficient (Ns/m) stiffness (N/m)
mf 15 css 830 kss 31,000
mc 1 cc 200 kc 18,000
m1 12.78 c1 2,064 k1 90,000
m2 8.62 c2 4,585 k2 162,800
m3 28.49 c3 4,750 k3 183,000
m4 5.31 c4 400 k4 310,000

Table 3.4 Parameter values of electrohydraulic actuator

Parameter a b g ps As t
Unit N/m5 s1 N/m5/2/kg1/2 Pa m2 s
13 9
Value 4.515  10 1 1.545  10 10342500 3.35  104 0.003

for the car suspension control. For the seat suspension control, the relative velocity
z_ f  z_ ps and the relative displacement zf  zps between cabin floor and seat frame
are used as feedback signals. By solving the minimisation problem of (3.36), the
static output feedback controller gain matrices for the integrated system model
(3.8) are designed as

K1 ¼ 105  ½0:2275x1 0:0062x14 0:0862x18 0:0872x19 0:0656x20 

ð3:44Þ

K2 ¼ 105  ½0:0590x2 0:0025x15 0:0545x18 0:0551x19 0:0415x20 

ð3:45Þ
 Robust active control of an integrated suspension system 85

K3 ¼ 105  ½ 0:1401x3 0:0025x16 0:0429x18 0:0773x19 0:0324x20 

ð3:46Þ

K4 ¼ 105  ½ 0:0868x4 0:0014x17 0:0348x18 0:0627x19 0:0263x20 

ð3:47Þ

K5 ¼ 105  ½ 3:9667x8 0:0163x18 0:0049x19 0:0041x20 0:0463x21 

ð3:48Þ

To validate the system performance in time domain, a typical road disturbance,

that is bump road disturbance, will be considered in the simulation and applied to
the vehicle wheels.
The ground displacement for an isolated bump in an otherwise smooth road
surface is given by
8   
>
> a 2pv0 l
< 1  cos t , 0ptp
zr ðtÞ ¼ 2 l v 0 ð3:49Þ
>
> l
: 0; t>
v0
where a and l are the height and the length of the bump and v0 is vehicle forward
speed. We choose a ¼ 0:08 m, l ¼ 2 m, and v0 ¼ 30 km/h in the simulation.

0.1
Front right
Rear right
0.09
Front left
Rear left
0.08

0.07
Displacement (m)

0.06

0.05

0.04

0.03

0.02

0.01

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)

Figure 3.2 Bump road disturbances

86 Handbook of vehicle suspension control systems

The bump road disturbances are shown in Figure 3.2, where the road disturbances
to the front and rear wheels have same peak amplitude with a time delay of
(lf þ lr Þ=v0 ; the road disturbances to the left and right wheels are applied with
different amplitude to excite the roll motion of the vehicle [20].
The bump responses of the driver head acceleration for the integrated system
are compared in Figure 3.3, in which Passive means no controller has been used in
the system, and Active means the designed controllers are applied to the system. It
can be seen from Figure 3.3 that the designed active control system achieves an
improved performance on ride comfort in terms of the maximum peak value of
driver head acceleration compared to the passive system. The comparison on car
suspension deflections are shown in Figure 3.4, from which we can see that the
active control system generates similar car suspension deflections compared to the
passive system. Further comparison on the tyre deflections are shown in Figure 3.5.
It can be seen from Figure 3.5 that both systems generate similar tyre deflections.
This means that the active control system keeps the similar road holding perfor-
mance to passive system while it achieves better ride comfort performance. The
desired forces and the corresponding actual forces which are generated from elec-
trohydraulic actuators are shown in Figure 3.6 for the chassis suspension and in
Figure 3.7 for seat suspension. It can be seen that the applied force tracking control

3
Passive
Active
2

1
Head acceleration (m/s2)

–1

–2

–3

–4
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)

Figure 3.3 Bump responses on driver head acceleration for passive and active
suspension systems
 Robust active control of an integrated suspension system 87

Front-left Front-right
0.06 0.1
Passive Passive
0.04 Active Active

Displacement (m)
Displacement (m) 0.05
0.02
0
0
–0.05
–0.02

–0.04 –0.1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)
Rear-left Rear-right
0.04 0.1
Passive Passive
0.02 Active Active

Displacement (m)
Displacement (m)

0.05
0
0
–0.02
–0.05
–0.04

–0.06 –0.1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)

Figure 3.4 Bump responses on car suspension deflections for passive and active
suspension systems

Front-left Front-right
0.06 0.12
0.05 Passive 0.1 Passive
Displacement (m)

Active Active
Displacement (m)

0.04 0.08
0.03 0.06
0.02 0.04
0.01 0.02
0 0
–0.01 –0.02
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)
Rear-left Rear-right
0.06 0.12
0.05 Passive 0.1 Passive
Displacement (m)

Active
Displacement (m)

Active
0.04 0.08
0.03 0.06
0.02 0.04
0.01 0.02
0 0
–0.01 –0.02
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)

Figure 3.5 Bump responses on tyre deflections for passive and active suspension
systems
88 Handbook of vehicle suspension control systems

Front-left Front-right
1500 1500
Actual force Actual force
1000 1000
Desired force Desired force
Force (N)

Force (N)
500 500

0 0

–500 –500

–1000 –1000
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)

Rear-left Rear-right
1000 1000
Actual force Actual force
500 Desired force 500 Desired force
Force (N)

Force (N)

0 0

–500 –500

–1000 –1000
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (s) Time (s)

Figure 3.6 Chassis suspension forces

Seat
700
Actual force
Desired force
600

500

400
Force (N)

300

200

100

–100
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (s)

Figure 3.7 Seat suspension forces

 Robust active control of an integrated suspension system 89

strategy is able to make the actuator output force track the desired force well, and
thus guarantees the designed system performance. It can also see that the seat
suspension assists the car chassis suspension to improve ride comfort performance
with less force output.
The robustness of the designed control system is checked by intentionally
changing the parameter values of the driver body model in the simulation. The
variations to driver body mass are considered first. Several different driver body
masses ranging from 35 kg to 75 kg are chosen to evaluate the ride comfort
performance in terms of the peak value of driver head acceleration under bump
road input. The comparison between the passive suspension and the designed
active suspension is made in Figure 3.8. It can be seen that the designed active
suspension can nearly keep the same performance and always outperforms the
passive suspension no matter the change of driver body mass. Then, the stiffness
and damping coefficients are modified to be 5% of their nominal values. Four
cases with different stiffnesses and damping coefficients are generated. Including
the nominal case, five cases are compared in Figure 3.9. It can be seen from
Figure 3.9 that the designed active suspension can achieve good ride comfort
performance compared to the passive suspension even when there exist para-
meter uncertainties. The robustness of the active suspension is validated to be
effective.

4.5
Passive
4 Active

3.5

3
Peak acceleration (m/s2)

2.5

1.5

0.5

0
35 40 45 50 55 60 65 70 75
Body mass (kg)

Figure 3.8 Comparison on peak value of driver head acceleration under bump
input for different driver body masses
90 Handbook of vehicle suspension control systems

4
Passive
Active
3.5

3
Peak acceleration (m/s2)

2.5

1.5

0.5

0
1 2 3 4 5
Case

Figure 3.9 Comparison on peak value of driver head acceleration under bump
input for different cases

3.5 Conclusions
In this chapter, an integrated active suspension has been developed. The inte-
grated suspension model is constructed first and then the integrated controller is
designed. As some state variables are not measurable in practice, a static output
feedback controller design method is presented. Considering the limited cap-
ability of actuators and the parameter uncertainties, actuator saturation con-
straint and parameter uncertainties to driver body model are included in the
controller design process. Numerical simulations are used to validate the per-
formance of the designed active suspension system. The results have shown that
the integrated suspension can provide improved ride comfort performance
compared to the passive suspension no matter the parameter variations of the
driver body model.

Acknowledgements

The support of this work by UOW URC Small Grant and GRF HKU 7140/11E is
gratefully acknowledged.
 Robust active control of an integrated suspension system 91

Appendix
For small angles f and q, sin f  f, cos f  1; sin q  q, cos q  1; we have the
following approximate linear relationships:
zps ¼ zs  rx q þ ry f ð3:50Þ
zsfl ¼ zs  lf q þ tf f ð3:51Þ
zsfr ¼ zs  lf q  tf f ð3:52Þ
zsrl ¼ zs þ lr q þ tr f ð3:53Þ
zsrr ¼ zs þ lr q  tr f ð3:54Þ

The passive forces generated between the sprung masses and the unsprung masses
can be expressed as
Fsfl ¼ ksf ðzsfl  zufl Þ þ csf ð_z sfl  z_ ufl Þ ð3:55Þ
Fsfr ¼ ksf ðzsfr  zufr Þ þ csf ð_z sfr  z_ ufr Þ ð3:56Þ
Fsrl ¼ ksr ðzsrl  zurl Þ þ csr ð_z srl  z_ url Þ ð3:57Þ
Fsrr ¼ ksr ðzsrr  zurr Þ þ csr ð_z srr  z_ urr Þ ð3:58Þ

The passive force generated between cabin floor and seat frame is
Fss ¼ kss ðzf  zps Þ þ css ð_z f  z_ ps Þ ð3:59Þ

The dynamic vertical motion of equations for the car unsprung masses are
mufl€z ufl ¼ Fsfl þ Ffl  ktf ðzufl  zrfl Þ ð3:60Þ
mufr€z ufr ¼ Fsfr þ Ffr  ktf ðzufr  zrfr Þ ð3:61Þ
murl€z url ¼ Fsrl þ Frl  ktr ðzurl  zrrl Þ ð3:62Þ
murr€z urr ¼ Fsrr þ Frr  ktr ðzurr  zrrr Þ ð3:63Þ

Similarly, for small angles f and q, the dynamic vertical and rotational motion of
equations for the car chassis can be linearly formalised as
ms€z s ¼ Fsfl  Fsfr  Fsrl  Fsrr  Ffl  Ffr  Frl  Frr þ Fss þ Fs ð3:64Þ
€ ¼ lf Fsfl þ lf Fsfr  lr Fsrl  lr Fsrr þ lf Ffl þ lf Ffr  lr Frl  lr Frr  rx Fss  rx Fs
Iq q
ð3:65Þ
€ ¼ tf Fsfl þ tf Fsfr  tf Fsrl þ tf Fsrr  tf Ffl þ tf Ffr  tf Frl þ tf Frr þ ry Fss þ ry Fs
If f
ð3:66Þ
92 Handbook of vehicle suspension control systems

The dynamic vertical motion of equations for the seat suspension and driver body
are given by

mf €z f ¼ kc ðzc  zf Þ þ cc ð_z c  z_ f Þ  Fss  Fs ð3:67Þ

mc€z c ¼ kc ðzc  zf Þ  cc ð_z c  z_ f Þ þ k1 ðz1  zc Þ þ c1 ð_z 1  z_ c Þ ð3:68Þ
m1€z 1 ¼ k1 ðz1  zc Þ  c1 ð_z 1  z_ c Þ þ k2 ðz2  z1 Þ þ c2 ð_z 2  z_ 1 Þ ð3:69Þ
m2€z 2 ¼ k2 ðz2  z1 Þ  c2 ð_z 2  z_ 1 Þ þ k3 ðz3  z2 Þ þ c3 ð_z 3  z_ 2 Þ ð3:70Þ
m3€z 3 ¼ k3 ðz3  z2 Þ  c3 ð_z 3  z_ 2 Þ þ k4 ðz4  z3 Þ þ c4 ð_z 4  z_ 3 Þ ð3:71Þ
m4€z 4 ¼ k4 ðz4  z3 Þ  c4 ð_z 4  z_ 3 Þ ð3:72Þ

The motion equations of the integrated model in terms of mass, damping, and
stiffness matrices can be formalised as
Mu€x u ¼ Bs ð_x s  x_ u Þ þ Ks ðxs  xu Þ þ Kt ðw  xu Þ þ Fa ð3:73Þ

Ms €
q ¼ LBs ð_x u  x_ s Þ þ LKs ðxu  xs Þ  LFa

þ Lp ðkss Lf xd  kss LTp q þ css Lf x_ d  css LTp qÞ

_ þ Lp Fs ð3:74Þ

Md €x d ¼ Bd x_ d þ Kd xd þ LTf kss LTp q þ LTf css LTp q_  LTf Fs ð3:75Þ

where
2 3
mufl 0 0 0 2 3
6 0 ms 0 0
6 mufr 0 0 7 7
Mu ¼ 6 7, Ms ¼ 4 0 Iq 05
4 0 0 murl 0 5 0 0 If
0 0 0 murr

2 3
mf 0 0 0 0 0
2 3
6 0 mc 0 0 0 0 7 csf 0 0 0
6 7
6 0 0 m1 0 0 0 7 6 0 csf 0 0 7
6 7 6 7
Md ¼ 6 7, Bs ¼ 6 7
6 0 0 0 m2 0 0 7 4 0 0 csr 0 5
6 7
4 0 0 0 0 m3 0 5 0 0 0 csr
0 0 0 0 0 m4

2 3 2 3
ksfl 0 0 0 ktf 0 0 0
6 0 ksfr 0 0 7 6 0 ktf 0 0 7
6 7 6 7
Ks ¼ 6 7, Kt ¼ 6 7
4 0 0 ksrl 0 5 4 0 0 ktr 0 5
0 0 0 ksrr 0 0 0 ktr
 Robust active control of an integrated suspension system 93
2 3
css  cc cc 0 0 0 0
6
6 cc cc  c1 c1 0 0 0 77
6
6 0 c1 c1  c2 c2 0 0 77
Bd ¼ 6 7
6 0 0 c2 c2  c3 c3 0 7
6 7
4 0 0 0 c3 c3  c4 c4 5
0 0 0 0 c4 c4

2 3
kss  kc kc 0 0 0 0
6 kc kc  k1 k1 0 0 0 7
6 7
6 0 k1 k1  k2 k2 0 0 7
6 7
Kd ¼ 6 7
6 0 0 k2 k2  k3 k3 0 7
6 7
4 0 0 0 k3 k3  k4 k4 5
0 0 0 0 k4 k4

2 3 2 3
1 1 1 1 1
6 7 6 7
L ¼ 4 lf lf lr l r 5, Lp ¼ 4 rx 5, Lf ¼ ½ 1 0 0 0 0 0
tf tf tr tr ry

q ¼ ½ zs q f T

xs ¼ ½ zsfl zsfr zsrl zsrr T , xu ¼ ½ zufl zufr zurl zurr T

xd ¼ ½ zf zc z1 z2 z3 z4 T

w ¼ ½ zrfl zrfr zrrl zrrr T , Fa ¼ ½ Ffl Ffr Frl Frr T

By using the following relationships between vectors

zf ¼ Lf xd

zps ¼ LTp q

xs ¼ LT q

Fss ¼ kss Lf xd  kss LTp q þ css Lf x_ d  css LTp q_

(3.73)–(3.75) can be further grouped as

p þ Bp p_ þ Kp p ¼ Kw w þ Kf u
Mp € ð3:76Þ
94 Handbook of vehicle suspension control systems

where
2 3
Mu 0 0
p ¼ ½ xu q x d T , Mp ¼ 4 0 Ms 0 5
0 0 Md

2 3
Bs Bs LT 0
6 7
Bp ¼ 6
4 LBs LBs LT þ Lp css LTp Lp css Lf 7
5
0 LTf css LTp Bd

2 3
Ks þ Kt Ks LT 0
6 7
Kp ¼ 6
4 LKs LKs LT þ Lp kss LTp Lp kss Lf 7
5
0 LTf kss LTp Kd

2 3 2 3
Kt I 0  
6 7 6 7 Fa
Kw ¼ 4 0 5, Kf ¼ 4 L Lp 5, u¼
Fs
0 0 LTf

T
By defining the state vector as x ¼ pT p_ T , we can write the dynamic
equations (3.76) into a state-space form as
x_ ¼ Ax þ Bw w þ Bu ð3:77Þ
     
0 I 0 0
where A ¼ , Bw ¼ , B¼ :
Mp1 Kp Mp1 Bp Mp1 Kw Mp1 Kf

References

[1] M. S. Fallah, R. Bhat, and W. F. Xie. ‘New model and simulation of

Macpherson suspension system for ride control applications’. Vehicle System
Dynamics, 47(2):195–220, 2009.
[2] D. Hrovat. ‘Survey of advanced suspension developments and related opti-
mal control applications’. Automatica, 33(10):1781–1817, 1997.
[3] R. A. Williams. ‘Automotive active suspensions’. Proceedings of the Insti-
tution of Mechanical Engineers, Part D: Journal of Automobile Engineering,
211(6):415–444, 1997.
[4] H. Du, J. Lam, and K. Y. Sze. ‘Design of non-fragile H? controller for active
vehicle suspensions’. Journal of Vibration and Control, 11:225–243, 2005.
[5] H. Du, K. Y. Sze, and J. Lam. ‘Semi-active H? control of vehicle suspension
with magneto-rheological dampers’. Journal of Sound and Vibration, 283:
981–996, 2005.
 Robust active control of an integrated suspension system 95

[6] H. Gao, J. Lam, and C. Wang. ‘Multi-objective control of vehicle active

suspension systems via load-dependent controllers’. Journal of Sound and
Vibration, 290(3):654–675, 2006.
[7] E. Guglielmino, T. Sireteanu, C. W. Stammers, G. Ghita, and M. Giuclea.
Semi-active Suspension Control – Improved Vehicle Ride and Road
Friendliness. Springer, London, 2008.
[8] M. Zapateiro, N. Luo, H. R. Karimi, and J. Vehi. ‘Vibration control of a class
of semiactive suspension system using neural network and backstepping
techniques’. Mechanical Systems and Signal Processing, 23(6):1946–1953,
2009.
[9] H. Gao, W. Sun, and P. Shi. ‘Robust sampled-data H? control for vehicle
active suspension systems’. IEEE Transactions on Control Systems Tech-
nology, 18(1):238–245, 2010.
[10] W. Sun, H. Gao, and O. Kaynak. ‘Finite frequency H? control for vehicle
active suspension systems’. IEEE Transactions on Control Systems Tech-
nology, 19(2):416–422, 2011.
[11] M. Zapateiro, F. Pozo, H. R. Karimi, and N. Luo. ‘Semiactive control
methodologies for suspension control with magnetorheological dampers’.
IEEE/ASME Transactions on Mechatronics, 17(2):370–380, 2012.
[12] H. Li, H. Liu, H. Gao, and P. Shi. ‘Reliable fuzzy control for active sus-
pension systems with actuator delay and fault’. IEEE Transactions on Fuzzy
Systems, 20(2):342–357, 2012.
[13] W. Sun, Y. Zhao, J. Li, L. Zhang, and H. Gao. ‘Active suspension control
with frequency band constraints and actuator input delay’. IEEE Transac-
tions on Industry Electronics, 59(1):530–537, 2012.
[14] S.-B. Choi, M.-H. Nam, and B.-K. Lee. ‘Vibration control of a MR seat
damper for commercial vehicle’. Journal of Intelligent Material Systems and
Structures, 11:936–944, 2000.
[15] I. J. Tiemessen, C. T. J. Hulshof, and M. H. W. Frings-Dresen. ‘An overview
of strategies to reduce whole-body vibration exposure on drivers: A sys-
tematic review’. International Journal of Industrial Ergonomics, 37:245–256,
2007.
[16] H. Du, W. Li, and N. Zhang. ‘Semi-active variable stiffness vibration control
of vehicle seat suspension using an MR elastomer isolator’. Smart Materials
and Structures, 20(10):1–10, 2011.
[17] Y. Wan and J. M. Schimmels. ‘Improved vibration isolating seat suspension
designs based on position-dependent nonlinear stiffness and damping char-
acteristics’. Journal of Dynamic Systems, Measurement, and Control, 125(3):
330–338, 2003.
[18] C.-M. Lee, A. H. Bogatchenkov, V. N. Goverdovskiy, Y. V. Shynkarenko,
and A. I. Temnikov. ‘Position control of seat suspension with minimum
stiffness’. Journal of Sound and Vibration, 292:435–442, 2006.
[19] J.-D. Wu and R.-J. Chen. ‘Application of an active controller for reducing
small-amplitude vertical vibration in a vehicle seat’. Journal of Sound and
Vibration, 274:939–951, 2004.
96 Handbook of vehicle suspension control systems

[20] M. Bouazara, M. J. Richard, and S. Rakheja. ‘Safety and comfort analysis of

a 3-D vehicle model with optimal non-linear active seat suspension’. Journal
of Terramechanics, 43:97–118, 2006.
[21] I. Maciejewski, L. Meyer, and T. Krzyzynski. ‘The vibration damping
effectiveness of an active seat suspension system and its robustness to varying
mass loading’. Journal of Sound and Vibration, 329(19):3898–3914, 2010.
[22] Y. Zhao, L. Zhao, and H. Gao. ‘Vibration control of seat suspension using H?
reliable control’. Journal of Vibration and Control, 16(12):1859–1879, 2010.
[23] Y. Zhao, W. Sun, and H. Gao. ‘Robust control synthesis for seat suspension
systems with actuator saturation and time-varying input delay’. Journal of
Sound and Vibration, 329(21):4335–4353, 2010.
[24] W. Sun, J. Li, Y. Zhao, and H. Gao. ‘Vibration control for active seat
suspension systems via dynamic output feedback with limited frequency
charateristic’. Mechatronics, 21:250–260, 2011.
[25] O. Gundogdu. ‘Optimal seat and suspension design for a quarter car with
driver model using genetic algorithms’. International Journal of Industrial
Ergonomics, 37:327–332, 2007.
[26] A. Kuznetsov, M. Mammadov, I. Sultan, and E. Hajilarov. ‘Optimization of
a quarter-car suspension model coupled with the driver biomechanical
effects’. Journal of Sound and Vibration, 330:2937–2946, 2011.
[27] S.-B. Choi and Y.-M. Han. ‘Vibration control of electrorheological seat
suspension with human-body model using sliding mode control’. Journal of
Sound and Vibration, 303:391–404, 2007.
[28] H. Du, J. Lam, and K. Y. Sze. ‘Non-fragile output feedback H? vehicle
suspension control using genetic algorithm’. Engineering Applications of
Artificial Intelligence, 16(8):667–680, 2003.
[29] H. Du and N. Zhang. ‘Designing H?/GH2 static-output feedback controller
for vehicle suspensions using linear matrix inequalities and genetic algo-
rithms’. Vehicle System Dynamics, 46(5):385–412, 2008.
[30] H. Du, N. Zhang, and J. Lam. ‘Parameter-dependent input-delayed control
of uncertain vehicle suspensions’. Journal of Sound and Vibration, 317(3):
537–556, 2008.
[31] J. H. Kim and F. Jabbari. ‘Actuator saturation and control design for build-
ings under seismic excitation’. Journal of Engineering Mechanics, 128(4):
403–412, 2002.
[32] K. Zhou and P. P. Khargonekar. ‘An algebraic Riccati equation approach to
H? optimization’. Systems & Control Letters, 11:85–91, 1988.
[33] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan. Linear Matrix
Inequalities in System and Control Theory. SIAM, Philadelphia, PA, June
1994.
[34] H. Gao and T. Chen. ‘Network-based H? output tracking control’. IEEE
Transactions on Automatic Control, 53(3):655–667, 2008.
[35] Y.-Y. Cao and Z. Lin. ‘Robust stability analysis and fuzzy-scheduling con-
trol for nonlinear systems subject to actuator saturation’. IEEE Transactions
on Fuzzy Systems, 11(1):57–67, 2003.
 Robust active control of an integrated suspension system 97

[36] Y. Niu, J. Lam, X. Wang, and D. W. C. Ho. ‘Observer-based sliding mode

control for nonlinear state-delayed systems’. International Journal of
Systems Science, 35(2):139–150, 2004.
[37] D. W. C. Ho and Y. Niu. ‘Robust fuzzy design for nonlinear uncertain sto-
chastic systems via sliding-mode control’. IEEE Transactions on Fuzzy
Systems, 15(3):350–358, 2007.
[38] A. Alleyne and J. K. Hedrick. ‘Nonlinear adaptive control of active suspen-
sions’. IEEE Transactions on Control Systems Technology, 3(1):94–101, 1995.
[39] P. C. Chen and A. C. Huang. ‘Adaptive sliding control of active suspension
systems with uncertain hydraulic actuator dynamics’. Vehicle System
Dynamics, 44(5):357–368, 2006.
[40] A. Alleyne and R. Liu. ‘On the limitations of force tracking control for
hydraulic servosystems’. Journal of Dynamic Systems, Measurement, and
Control, 121(2):184–190, 1999.
[41] H. Du and N. Zhang. ‘Fuzzy control for nonlinear uncertain electrohydraulic
active suspensions with input constraint’. IEEE Transactions on Fuzzy Systems,
17(2):343–356, 2009.
[42] M. M. Fateh and S. S. Alavi. ‘Impedance control of an active suspension
system’. Mechatronics, 19(1):134–140, 2009.
[43] S. B. Choi and S. S. Han. ‘H? control of electrorheological suspension
system subjected to parameter uncertainties’. Mechtronics, 13:639–657,
2003.
 Chapter 4
An interval type-2 fuzzy controller for vehicle
active suspension systems1
Jiangtao Caoa,b, Ping Lia and Honghai Liub

Abstract

A novel interval type-2 fuzzy controller architecture is proposed for resolving

nonlinear control problems of vehicle active suspension systems. It integrates
Takagi–Sugeno (T–S) fuzzy model, interval type-2 fuzzy reasoning, the Wu–Mendel
uncertainty bounds method, and selected optimization algorithms in order to con-
struct the switching routes between generated linear model control surfaces. The
stability analysis of the proposed approach is presented. The proposed method is
implemented into a numerical example and a case study on a nonlinear half-vehicle
active suspension system. The simulation results demonstrate the effectiveness and
efficiency of the proposed approach.

4.1 Introduction
Vehicle suspension systems, as one kind of typical nonlinear system, play a crucial
role in riding comfort, safety handling, and road damage minimization and con-
tribute significantly to the overall vehicle performance. It is evident that trade-offs
have to be taken in order to achieve an overall better performance for all types of
suspension systems including passive, semi-active, and active suspensions [1].
Growing attention has been paid to active suspensions in recent years, mainly due
to its less physical constraints, flexible structure, and intelligent methodology to
deal with random vibrations. It is also evident that different developing control
algorithms of the core part of active suspension systems have significantly con-
tributed to improve suspension performance [2, 3].

1
The work described in this chapter was supported in part by the Program for Liaoning Excellent
Talents in University (2008RC32), and in part by the Program for Liaoning Science and Technology
Innovative Research Team in University (LT2010058) and in part by the University of Portsmouth.
a
School of Information and Control Engineering, Liaoning Shihua University, Fushun, China
b
School of Creative Technologies, University of Portsmouth, Portsmouth, UK
100 Handbook of vehicle suspension control systems

In order to control an active suspension, the control algorithms must be able to

deal with mechanical nonlinear dynamics and to be operated under the imprecise
and uncertain conditions, which are mainly caused by random natural road sur-
faces. The mechanism behind fuzzy logic controllers (FLCs) is credited with being
a feasible methodology for designing robust controllers that are able to deliver a
satisfied performance in face of nonlinearity, uncertainty, and imprecision [4].
Hence, FLCs have become one kind of popular approaches for active suspension
system in recent years. There are different ways to construct FLCs for vehicle
suspension control systems. It is common to construct an FLC by eliciting the
corresponding fuzzy rules and defining their membership functions based on expert
knowledge or industrial experience. More importantly, contributions have been
made to improve the basic fuzzy control structure with computational intelligent
strategies. It is evident that integrating the fuzzy controllers with such other intel-
ligent methods as neural networks and genetic algorithms does improve the control
performance in uncertain scenarios [5–11].
From the viewpoint of application, all existing suspension control systems
employ type-1 fuzzy sets to build a type-1 fuzzy control system. It limits introdu-
cing uncertain factors from linguistic rules through predefined membership func-
tions. In order to overcome the weakness, type-2 fuzzy sets have been recently
proposed with their more general fuzzy membership functions and potential ability
to solve real-world uncertain scenarios [12–14]. The concept of the type-2 fuzzy set
was originally introduced by Zadeh [15] as an extension of the ordinary fuzzy set.
Then it has been significantly developed from theoretical research to applications
in the past decade. A fuzzy logic system (FLS) consisting of at least one type-2
fuzzy set is called a type-2 FLS [16]. In comparison with the type-1 FLS, a type-2
FLS has the twofold advantages as follows. First, it has the capability of directly
handling the uncertain factors of fuzzy rules caused by expert experience or lin-
guistic description. Second, it is efficient to employ a type-2 FLS to cope with
scenarios in which it is difficult or impossible to determine an exact membership
function and related measurement of uncertainties. These strengths have made
researchers to consider type-2 FLS as the preference for real-world applications.
Interval type-2 fuzzy system (IT2 FS) is one of the main branches of type-2
fuzzy systems. It has been widely studied and utilized for real-time control sys-
tems mainly due to its lower computational cost [3, 17–19]. There are many types
of methods for IT2 fuzzy systems to aggregate IT2 fuzzy values to crisp outputs.
Here, two of them are considered. One is represented by Karnik–Mendel algo-
rithm, which involves type-reduction (TR) [20, 21]. The other is represented by
Wu–Mendel uncertainty bounds, which do not involve TR [18]. The first type of
method calculates the exact solutions monotonically and super-exponentially fast
with simple formula and they can be run in parallel, but the time delay caused by
algorithmatic iteration is the bottleneck for real-time applications. On the other
hand, the second type of method replaces the TR by four uncertainty bounds.
These bounds only depend on the lower and upper firing levels of each rule and
the centroid of each rule’s consequent set. Wu and Mendel [18] calculated the
 An interval type-2 fuzzy controller for vehicle active suspension systems 101

final crisp outputs of an IT2 FLS by averaging the uncertainty bounds. That
means that it is computationally efficient that the mean operator and four uncer-
tainty bounds are combined to estimate their defuzzified outputs. It is obvious
that existing aggregation operators could play the mean method’s role in the
defuzzification [16] to achieve better performance for individual applications.
Some new researches on defuzzification of type-2 fuzzy sets are presented in
References 22 and 23.
Inspired by Mendel’s work [16], this chapter has proposed a general structure
to aggregate uncertainty bounds to the defuzzified outputs through further optimi-
zation structure. It integrates the Takagi–Sugeno (T–S) fuzzy model, the interval
type-2 fuzzy reasoning, the Wu–Mendel uncertainty bounds, and selected optimi-
zation algorithms together to construct the switching routes among the generated
linear model control surfaces. Considering the uncertainty bounds and further
optimization algorithms, based on the common quadratic Lyapunov functions, the
stability analysis of closed-loop control system is presented. By integrating T–S
fuzzy model into the proposed architecture of the IT2 FLS, the further optimization
module rebuilds the transfer routes between generated linear control surfaces and
integrates the control performance and other practical requirements into the
defuzzification interface. For the evaluation purpose, the proposed structure is
implemented into a numerical example and a case study on a half-vehicle active
suspension system with convincing results.
The rest of the chapter is organized as follows. Section 4.2 presents a nonlinear
model of a half-vehicle active suspension system. Section 4.3 proposes an IT2
T–S fuzzy control system with further optimization structure. Section 4.4 analyzes
the stability of the closed-loop control system with proposed architecture. Some
simulations are given in Section 4.5. Concluding remarks and future work are
discussed in Section 4.6.

4.2 A nonlinear active suspension system

A vehicle body is generally a rigid body with six-degrees of freedom (DOF)
motions; it consists of longitudinal, lateral, and heave motions and roll, pitch, and
yaw motions [2]. These motions are restricted by geometrical constraints of a
vehicle suspension, and they are also coupled with each other to certain degree.
Regardless of such coupling problems, the reduced-order mathematical model is
commonly employed for designing an active suspension control system. Therefore,
a quarter-vehicle model or a half-vehicle model is often used for theoretical ana-
lysis and design of active suspension systems.
A half-vehicle model including pitch and heave modes is represented to
simulate ride characteristics of a simplified whole vehicle. Based on the half-
vehicle model, many active control strategies have been designed to improve the
ride comfort and handling performance of vehicle suspensions [24]. Let f and r
denote the front and rear wheels, and x and z denote the longitudinal forward
102 Handbook of vehicle suspension control systems

zbf df dr zbr

mb Ib zb

Ks f U0 Ksr
cf faf far cr

zwf zwr

Z
mwf mwr
zf0 zr0
Ktf Ktr
X

Fig. 4.1 A half-vehicle suspension model in Reference 24

direction and vertical up direction. The notation is provided below for the half-
vehicle model as shown in Fig. 4.1:
df : distance from the front axle to the center of gravity (m);
dr : distance from the rear axle to the center of gravity (m);
mb : half-body mass (or sprung mass) (kg);
mwf : front-wheel mass (or unsprung mass) (kg);
mwr : rear-wheel mass (or unsprung mass) (kg);
Ksf : front suspension spring stiffness (N/m);
Ksr : rear suspension spring stiffness (N/m);
Ktf : front-tire stiffness (N/m);
Ktr : rear-tire stiffness (N/m);
cf : front damping coefficient (Ns/m);
cr : rear damping coefficient (Ns/m);
faf : front actuator force ðNÞ;
far : rear actuator force ðNÞ;
U0 : original velocity (m/s);
zb : car body displacement (m);
Ib : pitch inertia (kgm2);
zf 0 : road displacement at the front wheel (m);
zr0 : road displacement at the rear wheel (m);
zwf : front-wheel displacement (m);
zbf : front-body displacement (m);
zwr : rear-wheel displacement (m);
zbr : rear-body displacement (m).
As Hrovat remarked [1], the linear system approximation was appropriate for
some operations; however, there were some situations that amplified the nonlinear
effects. These nonlinear effects could be created by dry friction and discrete-event
 An interval type-2 fuzzy controller for vehicle active suspension systems 103

disturbances (e.g., single bumps or potholes). In order to simulate the real suspen-
sion system and to evaluate the potential application of proposed control method, a
nonlinear model is better to precisely describe the real system dynamics than linear
models. Meanwhile, for vehicle suspension systems, the high-order polynomial
functions are better than the proportional functions to describe the real spring and
damper forces. Based on the method in Reference 25, the connecting forces (e.g.,
spring force and damping force) can be modeled as the nonlinear functions using
measured data. The spring force fs is estimated by high-order polynomial functions:

fs ¼ fsl þ fsn ¼ k1 Dz þ ðk0 þ k2 Dz2 þ k3 Dz3 Þ ð4:1Þ

where fsl is the linear term of the spring force and fsn is the nonlinear term of the
spring force. The coefficients are obtained by fitting the equation to experimental
data. The damping force fd is also modeled as a second-order polynomial function
by fitting the measured data:

fd ¼ fdl þ fdn ¼ c1 D_z þ c2 D_z 2 ð4:2Þ

where fdl is the linear term and fdn is the nonlinear term of the damper force;
likewise, the coefficients are obtained by the measured data fitting.
Besides the nonlinear properties presented by the spring force and the damping
force, the vertical tire force is also highly nonlinear, especially when there are
substantial load changes. Even the vertical tire force becomes zero when the tire
loses contact with the road surface. The tire force is modeled as
ftl ¼ kt ðz0  zw Þ when ðz0  zw Þ > 0
ftn ¼ 0 when ðz0  zw Þ  0
where ftl denotes the linear tire force and ftn denotes the nonlinear tire force.
The nonlinear model of an active suspension system is provided below for the
controller design and performance analysis. Considering the nonlinearity shown in
(4.1) and (4.2), the active suspension system can be written as a multiple-input-
multiple-output (MIMO) nonlinear model:

X_ ¼ FðX ;UÞ ð4:3Þ

where X denotes the state matrices, which includes the displacements and velocity
of the vehicle body (i.e., Z_ b and Zb ), suspension (i.e., Z_ w and Zw ), and road surface
input (i.e., Z_ 0 and Z0 ); U denotes the actuator force matrices (i.e., faf and far );
FðX ;U Þ is a nonliner function that presents the suspension nonlinear dynamic
description and can be obtained by integrating the linear model in Reference 3 and
the nonlinear forces (i.e., fs , ft , and fd ).

4.3 The interval type-2 T–S fuzzy control system

Extensive studies, particularly fuzzy control strategies, have been conducted to
overcome the nonlinearity and uncertainty of active suspension systems. First, a
brief introduction on general T–S fuzzy control is presented in this section, then the
104 Handbook of vehicle suspension control systems

interval membership functions, type-2 reasoning methods, and proposed optimi-

zation structure are demonstrated, and finally the section is concluded with a novel
IT2 T–S fuzzy control system.

4.3.1 The general T–S fuzzy model and fuzzy control system
Considering a T–S fuzzy model, a fuzzy control system is represented as the general form:
RðlÞ : IF z1 is F1l and z2 is F2l , . . . , and zn is Fnl , THEN xðt þ 1Þ is g l ðX ;UÞ
where l 2 L :¼ 1; 2; . . . ;m: ð4:4Þ
where RðlÞ denotes the lth fuzzy inference rule, m denotes the number of fuzzy
rules, Fjl ð j ¼ 1; 2; . . . ;nÞ denote the type-1 fuzzy sets, zðtÞ :¼ ½z1 ; z2 ; . . . ; zn  denote
measurable variables, xðtÞ 2 <n denotes the state vector, and the T–S consequent
terms gil are defined as
gl ðX ;U; ql Þ ¼ Al xðtÞ þ Bl uðtÞ þ al
ð4:5Þ
l 2 L :¼ 1; 2; . . . ; m
where Al , Bl , and al are the parameter matrices of the lth local model, and uðtÞ 2 <p
denotes the input vector.
The fuzzy control scheme is chosen as the parallel distributed compensation
control and it is defined as follows:

RðrÞ : IF z1 is F1r and z2 is F2r , . . . , and zn is Fnr THEN uðtÞ is Kr xðtÞ

r 2 L :¼ 1; 2; . . . ;m ð4:6Þ
where Kr stands for the rth local linear control gain.
By using a singleton fuzzifier, product inference operator, and center average
defuzzifier, with the affine terms al  0; the closed-loop fuzzy control system can
be rewritten as
X
m X
m
xðt þ 1Þ ¼ ml mr ðAl þ Bl Kr ÞxðtÞ ð4:7Þ
l¼1 r¼1

It is assumed that ml and mr , the normalized membership functions, both satisfy as

follows:
xl ðzÞ
ml ¼ X
m
xi ðzÞ
i¼1
Y
n
ð4:8Þ
xl ðzÞ ¼ Fil ðzi Þ
i¼1
X
m
ml  0, ml ¼ 1
l¼1

where Fil ðzi Þ is the membership grade of zi in fuzzy set Fil :

 An interval type-2 fuzzy controller for vehicle active suspension systems 105

4.3.2 The interval type-2 T–S fuzzy control system

Though the priority in real-time applications has been given to the IT2 FLS in that
it has the capability of handling higher-order uncertainty factors in terms of cheaper
computational cost and simple structure, it has difficulties in interpreting related
uncertainty scenarios in the IT2 fuzzy reasoning. Hence, an IT2 T–S fuzzy control
system is presented in this section in order to analyze the problem of how to bound
the potential uncertainty in the type-2 reasoning process.
In order to present the general structure of the IT2 T–S FLS, the lth rule in (4.4)
can be rewritten as the following format:
~ l and z2 is F
RðlÞ : IF z1 is F ~ l , . . . , and zn is F
~l THEN xðt þ 1Þ ¼ Al xðtÞ þ Bl uðtÞ
1 2 n

ðl 2 L :¼ 1; 2; . . . ;mÞ ð4:9Þ
where, ~ li
F is an interval type-2 fuzzy set of rule l, which corresponds to a mem-
bership function shown in Fig. 4.2. Its firing strength of the lth rule belongs to the
following interval set:
wl ðxÞ 2 ½w l ðxÞ; w l ðxÞ; l ¼ 1; 2; . . . ; m ð4:10Þ
where
w l ðxÞ ¼ m F~ l ðxÞ  m F~ l ðxÞ     m F~ l ðxÞ ð4:11Þ
1 2 m

w l ðxÞ ¼ m F~ l ðxÞ  m F~ l ðxÞ     m F~ l ðxÞ ð4:12Þ

1 2 m

in which, m F~ l ðxÞ and m F~ l ðxÞ denote the lower and upper membership grades,
i i

respectively. Then the inferred IT2 T–S fuzzy model is defined as

X
m
xðt þ 1Þ ¼ ða  w l ðxÞ þ b  w l ðxÞÞðAl x þ Bl uÞ
l¼1
X
m
¼ ~ l ðxÞðAl x þ Bl uÞ
w ð4:13Þ
l¼1

μA(x)

1
UMF(A) UMF(A)
Embedded FS
Embedded FS

LMF(A)

Fig. 4.2 An example of an IT2 fuzzy membership function, where UMF stands
for the upper membership function and LMF stands for the lower
membership function [19]
106 Handbook of vehicle suspension control systems

where
~ l ðxÞ ¼ a  w l ðxÞ þ b  w l ðxÞ 2 ½0; 1
w
Xm
ð4:14Þ
~ l ðxÞ ¼ 1
w
l¼1

Herein, the values of a and b are both set as 0.5, according to Reference 26.
In order to control a nonlinear plant based on the IT2 T–S fuzzy model described
in (4.13), an IT2 T–S fuzzy controller is designed and its fuzzy rules are given as
~ r and z2 is F
RðrÞ : IF z1 is F ~ r , . . . , and zn is F
~r, ~ r xðtÞ
THEN uðtÞ is K
1 2 n

ðr 2 L :¼ 1; 2; . . . ;mÞ ð4:15Þ
where K~ r stands for the rth local linear control gain. The output of this controller is
defined as
X
m  
uðtÞ ¼ f wLr ðxÞ; wU ~
r ðxÞ K r  x ð4:16Þ
r¼1

here
w r ðxÞ
wLr ðxÞ ¼ X
m ð4:17Þ
ðw r ðxÞ þ w r ðxÞÞ
r¼1
w r ðxÞ
wU
r ðxÞ ¼ X
m ð4:18Þ
ðw r ðxÞ þ w r ðxÞÞ
r¼1

wLr and wU
r are satisfied with
X
m  
wLr ðxÞ þ wU
r ðxÞ ¼ 1 ð4:19Þ
r¼1

and the value of f ðwLr ðxÞ; wUr ðxÞÞ depends on the TR methods and belongs to an
interval.
The TR method is employed in this section, which is based on minimax
uncertainty bounds [16, 18]. Let us assign wLr ðxÞ þ wU L U
r ðxÞ=2 to f ðwr ðxÞ; wr ðxÞÞ
and substitute it into (4.16), we obtain the following:
h i
uðtÞ 2 uðOÞ ðtÞ; uðMÞ ðtÞ ð4:20Þ

where
X
m
w i Ki x
uðOÞ ðtÞ ¼ i¼1
Xm ð4:21aÞ
i
w
i¼1
 An interval type-2 fuzzy controller for vehicle active suspension systems 107

X
m
w i Ki x
uðMÞ ðtÞ ¼ i¼1
Xm ð4:21bÞ
i
w
i¼1

Then uncertainty bounds can be calculated by (4.22a) and (4.22b). The lower
bound u c ðxÞ is assigned to be equal to the upper bound u c ðxÞ if only one rule is fired
(i.e., m ¼ 1),
n o
u c ðtÞ ¼ min uðOÞ ðtÞ; uðMÞ ðtÞ ð4:22aÞ
2 3
Xm X m Xm
i i i i
6 ðw  w Þ w ðKi  K1 Þx w ðKm  Ki Þx 7
6 i¼1 i¼1 i¼1 7
u c ðtÞ ¼ u c ðtÞ  6
6X m Xm X
m X m
7
7 ð4:22bÞ
4 w i
w i i
w ðK  K Þx þ w ðK  K Þx5
i
i 1 m i
i¼1 i¼1 i¼1 i¼1

The crisp output of controller is

1 
uðtÞ
u ðtÞ þ u c ðtÞ ð4:23Þ
2 c

Generally speaking, the T–S fuzzy model is employed to describe a global

nonlinear system in terms of a set of local linear models that are smoothly con-
nected by fuzzy membership functions. Herein, the IT2 T–S fuzzy model supplies
an alternative way to build the bounded interval switching routes among local
linear models. That is to say, under the principle of the IT2 T–S fuzzy system, a
system model has the capability of switching among local linear models in an
interval route constrained by certain bounds involving with uncertainty and non-
linear dynamics. Equation (4.23) demonstrates that the center route of the interval
terms is employed to build its global system model. It is crucial to identify a
suitable structure within the principle in order to improve the performance of
optimizing switching routes from bounded intervals.

4.3.3 The proposed IT2 T–S fuzzy control system

A novel control structure is proposed in this section, which aims to rebuild the
switching routes between local linear controllers by integrating the IT2 fuzzy rea-
soning findings with control performance optimization. The structure is shown in
Fig. 4.3.
By adopting the proposed structure, the crisp outputs of the IT2 T–S FLS can
be recalculated as
n o
u c ¼ min uðOÞ ðtÞ; uðMÞ ðtÞ ð4:24aÞ
n o
u c ¼ max uðOÞ ðtÞ; uðMÞ ðtÞ ð4:24bÞ
108 Handbook of vehicle suspension control systems

Interval type-2 fuzzy controller

Further performance index

(Interval MFs) Uncertain Outputs

Inputs Optimization Plant
Fuzzification Rules bounds

Boundary
Inference
centroids

Feedback

Fig. 4.3 The structure of proposed IT2 T–S fuzzy controller

G ¼ f ð~
u ðtÞÞ
  ð4:24cÞ
u ðtÞ 2 u c þ Duc ; u c þ 2Duc ; . . . ;u c
~

u c  u c
Duc ¼ ð4:25Þ
n
where uðOÞ ðtÞ and uðMÞ ðtÞ can be calculated from (4.21a) and (4.21b), n denotes the
re-sampling number, G denotes the further optimization goal, and f is defined as a
performance function of the system with variable ~u ðtÞ. The control output ~u ðtÞ can
be solved from (4.24c) by off-the-shelf optimization algorithms.
With the above information, the systematic control procedure of proposed
framework is obtained as follows.
● Step 1: Determine all the state variables, their interval type-2 fuzzy MFs, and
fuzzy rules.
● Step 2: With the control plant and required control aims, design the optimi-
zation task and choose the related proper optimization method.
● Step 3: Obtain the system inputs, the interval outputs are calculated with the
interval type-2 fuzzy inference and the type-reduction method by (4.13),
(4.16), (4.21a), and (4.22b).
● Step 4: Calculate the fuzzy control outputs by further optimization structure
with equations (4.24a) and (4.25).
● Step 5: Perform the control outputs on the plant, the system inputs are updated,
and the system performance in further optimization part are also recalculated.
● Step 6: Return to the Step 3 to do the next interval type-2 fuzzy reasoning.
Recycle this process until the expected system performance is obtained.
In comparison with the conventional IT2 T–S FLS, the proposed structure builds
a more general framework to represent the defuzzifier processing. If an optimal goal
of the proposed IT2 T–S FLS can be described by (4.23), the convergence of the
optimization method is guaranteed, and the general method is shrunk to the same
form as the conventional IT2 T–S FLS. However, under the proposed structure, the
crisp output of the IT2 T–S FLS represents twofold information. One is the fuzzy
 An interval type-2 fuzzy controller for vehicle active suspension systems 109

rules extracted from expert knowledge or industrial experience. The other is the
further optimal goal that is required in practical issues or is impossible to be com-
bined into the fuzzy rules. Optimization algorithms can be selected in the terms of
domain-dependent goals and practical requirements. For the purpose of evaluating
the proposed method, a numerical example and a case study on a nonlinear half-
vehicle active suspension system are implemented in Section 4.5.

4.4 Stability analysis of the IT2 T–S fuzzy control system

Stability is one of the most important issues in analysis and design of control sys-
tems. Stability analysis of fuzzy control system has been more difficult because the
system is essentially nonlinear. Exploring the existed stability analysis results of
typical fuzzy control systems, for type-1 T–S fuzzy control systems, the systematic
approaches of stability analysis have been well developed in view of powerful
conventional control theory and techniques [27]. However, there are only a few
attentions to be given to the stability of type-2 fuzzy systems [28–33].
In this section, we analyze the stability of the proposed closed-loop IT2 T–S
fuzzy control system, which is formed by the IT2 T–S fuzzy model in (4.13) and
the proposed controller. For formulating the system in a closed format and without
losing the generality, the control output can be rewritten as
X
m
uðtÞ
ðaw i þ ð1  aÞw i ÞKi x; a 2 ½0; 1 ð4:26Þ
i¼1

Then the closed-loop IT2 T–S fuzzy control system can be described as follows:
X
m X
m
xðt þ 1Þ ¼ Gij ðAi þ Bi Kj ÞxðtÞ ð4:27Þ
i¼1 j¼1

where Gij denotes the fixed membership grade from the IT2 antecedents and T–S
consequent; it is described as
 
Gij ¼ aw i þ ð1  aÞw i w ~j
~ j ¼ wi w ð4:28Þ

where w i , w i , and w
~ j are defined in (4.11), (4.12), and (4.14).
For the further stability analysis, (4.27) can be represented as a general
uncertain system:
X
m X
m
xðt þ 1Þ ¼ G0 xðtÞ þ ~ j DGij xðtÞ
wi w
i¼1 j¼1
X
m X
m
¼ G0 xðtÞ þ ~ i DGii xðtÞ þ
wi w ~ j DFij xðtÞ
wi w
i¼1 i<j

¼ fG0 þ W DðtÞZ gxðtÞ ð4:29Þ

110 Handbook of vehicle suspension control systems

Here

1X m
G0 ¼ ðAi þ Bi Ki Þ ð4:30Þ
m i¼1

DGij ¼ Ai þ Bi Kj  G0
DGii ¼ Qii Fii SiiT ð4:31Þ
DFij ¼ DGij þ DGji ¼ Qij Fij SijT ; i<j

Q and S are unitary matrices, and W 2 <n g , DðtÞ 2 <g g , Z 2 <g n ,

g ¼ ½n m ðm þ 1Þ=2, the matrices W and Z are as follows:
 
W ¼ Q1 Q2    Qm
 
Z ¼ S1 S2    Sm ð4:32Þ
 e e 
DðtÞ ¼ block  diag F 1 F 2    Fem

where

Q i ¼ ½Qii Qiiþ1    Qir 

S i ¼ ½Sii Siiþ1    Sir 
e ð4:33Þ
Fi ¼ block  diag ½eii Fii eiiþ1 Fiiþ1  eir Fir 
 
eii ¼ wi w
~ j 2 wiw
~ j; wiw~j

Based on (4.32), the matrices M and N are defined as follows:

h i
d d d
M ¼ N ¼ block  diag F 1 F 2    F r ð4:34Þ

where
 
d dii diiþ1 dir
Fi ¼ block  diag Fii Fiiþ1  Fir
2 2 2 ð4:35Þ
dij ¼ max wi w
~j

Considering the IT2 membership grade of antecedents wi 2 ½w i ; w i , if the

interval bounds are fixed, then the stability of IT2 T–S fuzzy system can be covered
by the stability results of type-1 T–S fuzzy systems [34]. The stability analysis
result is summarized in Theorem 4.1.
Theorem 4.1. The equilibrium of an IT2 T–S fuzzy control system is given in
(4.27), formed by the TR method and control structure in Section 4.3.3, and the
proposed control system is quadratically stable in the large if and only if one of the
following conditions are satisfied.
 An interval type-2 fuzzy controller for vehicle active suspension systems 111

1. There exists a positive definite matrix P such that

PðG0 þ WMZÞ þ ðG0 þ WMZÞT P þ PWNN T W T P þ Z T Z < 0

2. If defined
" #
G0 þ WMZ WNN T W T
H¼
ZTZ ðG0 þ WMZÞ

the condition is

Reli ðHÞ 6¼ 0; i ¼ 1; 2; . . . ; 2 n:

3. There exists a positive definite matrix P such that

2 3
PðG0 þ WMZÞ þ ðG0 þ WMZÞT P PWN ZT
6 7
4 N TW TP I 0 5<0
Z 0 I

Remark 4.1. Since the type-reduction method is used to aggregate the IT2 centroids
to bounds, the closed-loop system is reduced to type-1 T–S FLS and its stability
analysis is similar to the type-1 T–S FLS. It can be also deduced that if all the
subsystems related to the interval bounds are asymptotically stable, then the IT2
FLS is asymptotically stable. However, these stability conditions are only sufficient
conditions with strong constraints.
Remark 4.2. Consider the proposed control structure in Fig. 4.3, algorithms can be
selected further to indirectly tune Gij by optimizing the uncertainty of lower and
upper bounds. Theorem 4.1 proves that, if one of the conditions in theorem is
satisfied, the optimization algorithms do not affect the closed-loop stability of the
proposed IT2 T–S FLS.

4.5 Simulation examples

The proposed approach is implemented into a numerical example and a case study
on a half-vehicle active suspension model in this section.

4.5.1 A numerical example

A numerical example is conducted to demonstrate the effect of the novel IT2
T–S FLS.
Consider a T–S fuzzy model with two following rules:

R1 : If x is A1 , then y ¼ 0:2x þ 9
R2 : If x is A2 , then y ¼ 0:6x þ 2
112 Handbook of vehicle suspension control systems

This example has been used in Takagi’s paper [35] to illustrate the T–S model
fuzzy reasoning results. Here, for demonstrating the reasoning process of the IT2
FLS, the proposed interval membership functions are implemented in the same
example. The membership functions of A1 and A2 are shown in Fig. 4.3. The
figure includes not only the interval membership functions shown as a and c, but
also the crisp membership functions denoted as b from paper [35].
Fig. 4.4 demonstrates the simulation results generated by a type-1 T–S fuzzy
system, an IT2 T–S fuzzy system, and the proposed IT2 T–S fuzzy system with
further optimization. First, translating route 1 was obtained by using type-1 mem-
bership functions b in Fig. 4.5 and type-1 fuzzy reasoning method in Reference 35;
routes 2 and 3 were produced by employing boundary membership functions a and
c in Fig. 4.5 and the same type-1 reasoning method. It illustrated that the sliding
route from one line to the other was nonlinear and with a boundary field between
the lower and upper routes. It implied that type-1 fuzzy reasoning can model
the nonlinear switch routes between two linear surfaces but cannot deal with
uncertainties in the switching routes.
By using the IT2 membership functions and K–M algorithm, the switching
route 4 was obtained. It is different from all the type-1 fuzzy reasoning results, and

10

9.5

5
9
4
8.5

8 2

1
7.5
3
7

6.5

5.5
2 3 4 5 6 7 8 9 10 11

Fig. 4.4 Interval fuzzy reasoning results (1: the type-1 reasoning middle route, 2: the
type-1 reasoning lower route, 3: the type-1 reasoning upper route, 4: the
K–M IT2 reasoning route, and 5: the proposed IT2 reasoning route).
 An interval type-2 fuzzy controller for vehicle active suspension systems 113

1 A1 A2

a a
b b
c c

0 1 2 3 4 5 6 7 8 9 10

Fig. 4.5 Interval membership functions (‘‘a’’ denotes the lower MFs, ‘‘b’’ denotes
the crisp MFs, and ‘‘c’’ denotes the upper MFs)

piecewise near to route 2 (i.e., the input value in References 3, 4, 7, and 8) and
route 1 (i.e., the input value in References 5 and 6). Route 4 also shows the IT2 T–S
FLS has the inherent ability to build more complex switch routes than those of the
type-1 FLS. Route 5 was produced by employing the proposed IT2 T–S FLS with
the optimal goal in (4.40). A switching route 5 was optimally generated among the
linear control surfaces to track the sigmoid function.
2:8
f ðxÞ ¼ þ 6:8 ð4:36Þ
1þ e2:5ðx5:08Þ
The reasoning results demonstrated that the proposed structure can make use of
interval type-2 fuzzy reasoning ability and the proposed structure to rebuild the
switching lines between linear surfaces and have potential ability to deal with the
high nonlinear and uncertain.

4.5.2 A half-vehicle active suspension system

The proposed method in Section 4.3 was implemented into a half-vehicle active
suspension system as shown in Fig. 4.1 whose mathematical model is given in
Section 4.2. Parameters of the model are provided in Table 4.1. The vehicle body
velocities (i.e., z_ bf , z_ br ) and displacements (i.e., zbf and zbr ) are chosen as inputs,
and the actuator forces (i.e., Uf and Ur ) are chosen as outputs. The interval

Table 4.1 The parameters of half-vehicle active suspension

mb (kg) I b (kg  m2 ) mwf (kg) mwr (kg) d f (m) d r (m)

1794.4 3443.05 187.15 440.04 1.15 1.60
cf 1 (m) cf 2 (m) k0f ðN Þ k1f ðN =mÞ k2f ðN =m2 Þ k3f ðN =m3 Þ
1190 426 136 60824 10865 104
c1r (m) c2r (m) k0r ðN Þ k1r ðN =mÞ k2r ðN =m2 Þ k3r ðN =m3 Þ
1000 215 146 18615 3665 384
114 Handbook of vehicle suspension control systems

NL NM S PM PL
1

–1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1

(a) Membership functions of the vehicle body velocities (i.e., z˙bf , z˙br)

N S P
1

–1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1

(b) Membership functions of displacements of the vehicle body (i.e., zbf , zbr)

Fig. 4.6 IT2 fuzzy membership functions of the four FLC input variables

membership functions of the inputs are provided in Fig. 4.6. The consequents are
linear control outputs as given in (4.37). The vehicle speed is 20 m/s.
 
Uf
U¼ ¼ KX ð4:37Þ
Ur

where
" #
k11 k12 k13 k14
K¼ ð4:38Þ
k21 k22 k23 k24
2 3
Z_ bf
6 Z_ 7
6 7
X ¼ 6 br 7 ð4:39Þ
4 Zbf 5
Zbr

Based on the linear model of half-vehicle suspension system and LQ control

strategy, the control gains K can be solved [24]. There are totally 18 fuzzy rules for
 An interval type-2 fuzzy controller for vehicle active suspension systems 115

the half-vehicle suspension control system as shown in Table 4.2. The local linear
controller gains are as below:
 
28:9 4095:1 69:3 66341:6
K0 ¼
3568:9 30:58 19019:6 73:34
 
28:9 4095:1 69:3 76365:6
K1 ¼
3568:9 30:58 26874:6 73:34
 
28:9 4095:1 69:3 70365:6
K2 ¼
3568:9 30:58 22874:66 73:34
 
28:9 4095:1 69:3 60541:6
K3 ¼
3568:9 30:58 15474:66 73:34
 
28:9 4095:1 69:3 57541:6
K4 ¼
3568:9 30:58 13474:66 73:34

The fuzzy rules and local linear controller are mainly designed to reduce the body
accelerations with the aim to improve the riding comfort. Herein the proposed
new control structure with a cost function in (4.40) is employed to save the
actuator’s energy.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
GðUf ;Ur Þ ¼ min ðq1€z 2bf þ q2€z 2br þ q3 Uf2 þ q4 Ur2 Þ ð4:40Þ

where, in order to prioritize the parameters, q1 is set as 1, q2 is set as 1, q3 is set as

10, and q4 is set as 10. Due to the real-time requirement and its nonlinearity, a
particle swarm optimization (PSO) algorithm proposed in Reference 36 was inte-
grated into the structure. And the re-sample number is set as 20 by default, the
inertia weight is 0.5, and the acceleration coefficients for local optimization and
global optimization are 2.
For evaluation propose, a linear quadratic controller (LQ controller) [24] and
an IT2 fuzzy controller with Wu–Mendel uncertain bounds [3] are also designed to
compare with the proposed approach.
The class C road surface, one kind of poor road surfaces, is used as random
road input, where the road roughness 2:56 104 m3 /cycle according to the
ISO (1982) classification using the power spectral density (PSD) is shown in
Fig. 4.7.
The simulation results are shown in Figs. 4.8–4.11. Regarding the accelera-
tions of the front body and rear body, the proposed method has achieved better
performance on riding comfort than the other two methods. On the other hand, the
linear controller has not satisfied the criteria related to the nonlinearity of the sus-
pension system. Figs. 4.10 and 4.11 have demonstrated the comparison of the
control force; it is evident that the proposed method requires lower force than the
conventional IT2 method, but higher force than the LQ controller.
Table 4.2 The rules of fuzzy controller

z_ bf z_ br zbf zbr U z_ bf z_ br zbf zbr U

S or NM S or NM S S K0 X NL NL S S K4 X
PM or PL S or NM S S K1 X S or NM S or NM P or N P or N K3 X
NL S or NM S S K3 X PM or PL S or NM P or N P or N K0 X
S or NM PM or PL S S K1 X NL S or NM P or N P or N K3 X
PM or PL PM or PL S S K3 X PM or PL S or NM P or N P or N K0 X
NL PM or PL S S K3 X PM or PL PM or PL P or N P or N K1 X
S or NM NL S S K3 X NL PM or PL P or N P or N K3 X
PM or PL NL S S K0 X S or NM NL P or N P or N K2 X
PM or PL NL P or N P or N K1 X NL NL P or N P or N K2 X
 An interval type-2 fuzzy controller for vehicle active suspension systems 117

0.03

0.02

0.01
The random road input (m)

−0.01

−0.02

−0.03

−0.04
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 4.7 Random road inputs

1.5

1
The front-body accelerations (m/s2)

0.5

−0.5

−1

−1.5 LQ controller
IT2 method
Proposed method
−2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 4.8 The front-body accelerations with random road inputs

118 Handbook of vehicle suspension control systems

1.5
The rear-body accelerations (m/s2)

0.5

−0.5

−1
LQ controller
−1.5 IT2 method
Proposed method
−2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 4.9 The rear-body accelerations with random road inputs

2000

1500
The front control force (N)

1000

500

−500

LQ controller
−1000
IT2 method
Proposed method
−1500
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 4.10 The front control force with random road inputs

From the statistic evaluation point, two kinds of performance criteria are used
to compare the vehicle active suspension control performance. One is the root mean
square (RMS) value, which presents the vehicle ride comfort and handling per-
formance from time domain [1]. Another is the ride index of body vibration, which
focuses on the ride comfort from frequency weighted vibrating accelerations [37].
The comparison of RMS values for vertical accelerations and control forces are
shown in Table 4.3.
 An interval type-2 fuzzy controller for vehicle active suspension systems 119

1000

800

600
The rear control force (N)

400

200

−200

−400

−600 LQ controller
−800 IT2 method
Proposed method
−1000
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Fig. 4.11 The rear control force with random road inputs

Table 4.3 The RMS values’ comparison of body accelerations and

control forces

FVA (m=s2 ) RVA (m=s2 )

LQ controller 0.5067 0.5652
IT2 method 0.2195 0.2122
Proposed method 0.1811 0.1608
Front control forces (N) Rear control forces (N)
LQ controller 191.85 176.09
IT2 method 715.23 350.42
Proposed method 519.93 261.14

FVA: Front vehicle accelerations and RVA: Rear vehicle accelerations.

Table 4.4 The crest factors’ comparison of body accelerations

Random road input Front vehicle Rear vehicle

LQ controller 3.1422 3.3526
IT2 method 2.8366 2.2031
Proposed method 2.4235 2.1125

Based on the ISO criteria of vehicle ride comfort evaluation, the crest factors
of body vertical and rotational vibration are determined in Table 4.4. The ride index
calculation method in Reference 38 is used to compare the vehicle ride comfort
with LQ controller, IT2 fuzzy control system, and proposed fuzzy control system.
Table 4.5 shows the comparison of ride index.
120 Handbook of vehicle suspension control systems

Table 4.5 The ride index comparison of vehicle body

Random road input Front vehicle Rear vehicle

LQ controller 0.3621 0.3765
IT2 method 0.2874 0.2930
Proposed method 0.2655 0.2512

Regarding the RMS accelerations and ride index of the front body and rear
body, the proposed method has achieved better performance on ride comfort than
the other two methods. Simultaneously, the simulation results showed the proposed
method need higher control force than LQ control system.

4.6 Concluding remarks

This chapter has presented a novel IT2 T–S fuzzy control system for active sus-
pensions in order to resolve their nonlinear dynamics, to understand in-depth
uncertainty embedded in fuzzy linguistic rules and reasoning, and to meet real-time
requirement as a whole. The proposed approach has integrated the IT2 membership
functions, T–S fuzzy model, Wu–Mendel uncertain bounds, and further optimiza-
tion algorithms into one control framework; it has been implemented into case
studies and been compared with existing IT2 fuzzy control system. It is evident that
the proposed approach outperforms an LQ controller and an IT2 fuzzy controller.
The simulation results have shown that the proposed IT2 T–S fuzzy control system
can not only effectively handle the system uncertainty and improve control per-
formance, but also save actuator energy with an added optimization module.
The priority of future work is given to the real-time applications on an electric
vehicle suspension system. Also the in-depth interpretation on the uncertainty
bounds will be investigated based on the interval reasoning. The stability condi-
tions of proposed closed-loop systems will be further relaxed. Future work is also
aimed at a bit more ambitious target of developing a fuzzy inference engine that has
the capability of conducting reasoning in terms of qualitative and probabilistic
information [39, 40].

References
[1] D. Hrovat, ‘‘Survey of advanced suspension developments and related
optimal control applications,’’ Automatica, vol. 33, no. 10, pp. 1781–1817,
1997.
[2] M. Nagai, ‘‘Recent researches on active suspensions for ground vehicles,’’
Japan Society of Mechanical Engineers, vol. 36, pp. 161–170, 1993.
[3] J. Cao, H. Liu, P. Li, and D. Brown, ‘‘An interval type-2 fuzzy logic con-
troller for quarter-vehicle suspensions,’’ Proc. IMechE Part D: J. Automobile
Engineering, vol. 222, no. 8, pp. 1361–1373, 2008.
 An interval type-2 fuzzy controller for vehicle active suspension systems 121

[4] C. Ting, T. Li, and F. Kung, ‘‘Design of fuzzy controller for active suspen-
sion system,’’ Mechatronics, vol. 5, no. 4, pp. 365–383, 1995.
[5] S. Huang and W. Lin, ‘‘Adaptive fuzzy controller with sliding surface for
vehicle suspension control,’’ Fuzzy Systems, IEEE Transactions on, vol. 11,
no. 4, pp. 550–559, 2003.
[6] S. Kumarawadu and T. Lee, ‘‘Neuroadaptive combined lateral and longitudinal
control of highway vehicles using RBF networks,’’ Intelligent Transportation
Systems, IEEE Transactions on, vol. 7, no. 4, pp. 500–512, 2006.
[7] S. Huang and C. Lin, ‘‘Application of a fuzzy enhance adaptive control on
active suspension system,’’ International Journal of Computer Applications
in Technology, vol. 20, no. 4, pp. 152–160, 2004.
[8] I. Kucukdemiral, S. Engin, V. Omurlu, and G. Cansever, ‘‘A robust single
input adaptive sliding mode fuzzy logic controller for automotive active
suspension system,’’ Lecture Notes in Artificial Intelligence, vol. 3613,
pp. 981–986, 2005.
[9] S. Wu, C. Wu, and T. Lee, ‘‘Neural-network-based optimal fuzzy control
design for half-car active suspension systems,’’ Intelligent Vehicles Sympo-
sium. Proceedings. IEEE, pp. 376–381, 2005.
[10] R. Saeks, C. Cox, J. Neidhoefer, P. Mays, and J. Murray, ‘‘Adaptive control
of a hybrid electric vehicle,’’ Intelligent Transportation Systems, IEEE
Transactions on, vol. 3, no. 4, pp. 213–234, 2002.
[11] R. Lian, B. Lin, and W. Sie, ‘‘Self-organizing fuzzy control of active sus-
pension systems,’’ International Journal of Systems Science, vol. 36, no. 3,
pp. 119–135, 2005.
[12] H. Hagras, ‘‘A hierarchical type-2 fuzzy logic control architecture for
autonomous mobile robots,’’ Fuzzy Systems, IEEE Transactions on, vol. 12,
no. 4, pp. 524–539, 2004.
[13] R. Sepúlveda, O. Castillo, P. Melin, A. Rodrguez-Daz, and O. Montiel,
‘‘Experimental study of intelligent controllers under uncertainty using type-1
and type-2 fuzzy logic,’’ Information Sciences, vol. 177, no. 10, pp. 2023–2048,
2007.
[14] O. Castillo and P. Melin, Type-2 fuzzy logic: theory and applications,
1st ed. Heidelberg, Germany: Springer-Verlag, January 2008, vol. 223
(ISBN: 978-3-540-76283-6).
[15] L. Zadeh, ‘‘The concept of a linguistic variable and its application to
approximate reasoning part i,’’ Information Sciences, vol. 8, pp. 199–249,
1975.
[16] J. Mendel, ‘‘Advances in type-2 fuzzy sets and systems,’’ Information
Sciences, vol. 177, no. 1, pp. 84–110, 2007.
[17] Q. Liang and J. Mendel, ‘‘Interval type-2 fuzzy logic systems: theory and
design,’’ Fuzzy Systems, IEEE Transactions on, vol. 8, no. 5, pp. 535–550,
2000.
[18] H. Wu and J. Mendel, ‘‘Uncertainty bounds and their use in the design of
interval type-2 fuzzy logic systems,’’ Fuzzy Systems, IEEE Transactions on,
vol. 10, no. 5, pp. 622–639, 2002.
122 Handbook of vehicle suspension control systems

[19] J. Mendel, H. Hagras, and R. John, ‘‘Standard background material about

interval type-2 fuzzy logic systems that can be used by all authors.’’
[Online]. Available: http://ieee-cis.org/files/standards.t2.win.pdf
[20] N. Karnik and J. Liang, ‘‘Type-2 fuzzy logic systems,’’ Fuzzy Systems, IEEE
Transactions on, vol. 7, no. 6, pp. 643–658, 1999.
[21] N. Karnik and J. Mendel, ‘‘Centroid of a type-2 fuzzy set,’’ Information
Sciences, vol. 132, nos. 1–4, pp. 195–220, 2001.
[22] S. Coupland and R. John, ‘‘A fast geometric method for defuzzification of
type-2 fuzzy sets,’’ Fuzzy Systems, IEEE Transactions on, vol. 16, no. 4,
pp. 929–941, 2008.
[23] M. Nie and W. Tan, ‘‘Towards an efficient type-reduction method for
interval type-2 fuzzy logic systems,’’ 2008 IEEE International Conference
on Fuzzy systems, pp. 1425–1432, 2008.
[24] J. Cao, H. Liu, P. Li, and D. Brown, ‘‘Study on active suspension control
system based on an improved half-vehicle model,’’ International Journal of
Automation and Computing, vol. 4, no. 3, pp. 236–242, 2007.
[25] C. Kim and P. Ro, ‘‘A sliding mode controller for vehicle active suspension
systems with nonlinearities,’’ Proceedings of IMechE Part D: Journal of
Automobile Engineering, vol. 212, no. 2, pp. 79–92, 1998.
[26] Q. Liang and J. Mendel, ‘‘Equalization of nonlinear time-varying channels
using type-2 fuzzyadaptive filters,’’ Fuzzy Systems, IEEE Transactions on,
vol. 8, no. 5, pp. 551–563, 2000.
[27] G. Feng, ‘‘A survey on analysis and design of model-based fuzzy control
systems,’’ Fuzzy systems, IEEE Transactions on, vol. 14, no. 5, pp. 676–697,
2006.
[28] N. Cázarez, S. Cárdenas, L. Aguilar, and O. Castillo, ‘‘Lyapunov stability on
type-2 fuzzy logic control,’’ IEEE-CIS International Seminar on Computa-
tional Intelligence, México Distrito Federal, México, pp. 1–10, 2005.
[29] O. Castillo, N. Cázarez, D. Rico, and L. Aguilar, ‘‘Intelligent control of
dynamic systems using type-2 fuzzy logic and stability issues,’’ Interna-
tional Mathematical Forum, vol. 1, no. 28, pp. 1371–1382, 2006.
[30] O. Castillo, L. Aguilar, N. Cázarez, and S. Cárdenas, ‘‘Systematic design of
a stable type-2 fuzzy logic controller,’’ Applied Soft Computing Journal,
vol. 8, no. 3, pp. 1274–1279, 2008.
[31] H. K. Lam and L. D. Seneviratne, ‘‘Stability analysis of interval type-2
fuzzy-model-based control systems,’’ Systems, Man and Cybernetics, Part B,
IEEE Transactions on, vol. 38, no. 3, pp. 617–628, 2008.
[32] B. M. Begian, W. W. Melek, and J. Mendel, ‘‘Stability analysis of type-2
fuzzy systems,’’ IEEE World Congress on Computational Intelligence,
pp. 947–953, 2008.
[33] J. Morales, O. Castillo, and J. Soria, ‘‘Stability on type-1 and type-2 fuzzy
logic systems,’’ Soft Computing for Hybrid Intelligent Systems, pp. 29–51,
2008.
[34] K. Tanaka, T. Ikeda, and H. Wang, ‘‘Robust stabilization of a class of
uncertain nonlinear systems viafuzzy control: quadratic stabilizability, H8
 An interval type-2 fuzzy controller for vehicle active suspension systems 123

control theory, and linear matrix inequalities,’’ Fuzzy Systems, IEEE

Transactions on, vol. 4, no. 1, pp. 1–13, 1996.
[35] T. Takagi and M. Sugeno, ‘‘Fuzzy identification of systems and its appli-
cations to modeling and control,’’ Systems, Man, and Cybernetics, IEEE
Transations on, vol. 15, pp. 116–132, 1985.
[36] J. Kennedy and R. Eberhart, ‘‘Particle swarm optimization,’’ Neural Networks.
Proceedings, IEEE International Conference on, vol. 4, 1995.
[37] ISO2631-1, ‘‘Mechanical vibrations and shock – evaluation of human
exposure to whole-body vibration part 1: general requirements,’’ Interna-
tional Standardization Organization, 1997.
[38] M. Montazeri-Gh and M. Soleymani, ‘‘Genetic optimization of a fuzzy
active suspension system based on human sensitivity to the transmitted
vibrations,’’ Proceedings of the Institution of Mechanical Engineers, Part D,
Journal of Automobile Engineering, vol. 222, no. 8, pp. 1769–1780, 2008.
[39] H. Liu, D. Brown, and G. Coghill, ‘‘Fuzzy qualitative robot kinematics,’’
Fuzzy Systems, IEEE Transactions on, vol. 16, no. 3, pp. 808–822, 2008.
[40] Z. Liu and H. Li, ‘‘A probabilistic fuzzy logic system for modelling and
control,’’ Fuzzy Systems, IEEE Transations on, vol. 13, no. 6, pp. 848–859,
2005.
 Chapter 5
Active control for actuator uncertain half-car
suspension systems1
Hongyi Lia and Honghai Liub

Abstract

This chapter designs a non-fragile H1 controller for a class of active suspension

systems with actuator uncertainty. By using Lyapunov stability theory, a non-
fragile controller is designed for the purpose of ensuring that the resulting active
suspension system is asymptotically stable with a prescribed H1 disturbance
attenuation level. The designed non-fragile H1 controller is constructed via convex
optimization by guaranteeing its sufficient condition in terms of feasible linear
matrix inequalities (LMIs). Simulation results are given to show the effectiveness
of the proposed control approach.

5.1 Introduction

As is well known, a vehicle suspension model takes an important part in modern

vehicles for improving vehicle suspension performance. Three main suspension
performances ride comfort, road handling and suspension deflection are often
employed to evaluate the suspension performance. In the past few decades, many
research projects have been conducted to improve vehicle suspension systems
performance [9]. It should be mentioned that enhancing ride comfort needing larger
suspension stroke and smaller damping of wheel-hop mode results in a degradation
in ride safety [3]; the problem, hence, is still wide open for groundbreaking scientific
contribution and feasible solutions. Recently, the researchers have paid increasing
attention to solve the trade-off between the conflicting objectives [8].

1
This work was partially supported by a joint PhD scholarship of University of Portsmouth and Protean
Electric Ltd, the National Science Foundation of P. R. China (61203002), the Program for New Century
Excellent Talents in University, and the Program for Liaoning Excellent Talents in University (LR2013053).
a
College of Engineering, Bohai University, Jinzhou, Liaoning, China
b
Intelligent Systems & Biomedical Robotics Group, School of Creative Technologies, University of
Portsmouth, UK
126 Handbook of vehicle suspension control systems

Recently, passive [14], semi-active [10] and active suspension systems [2, 15]
have been proposed to achieve the vehicle requirement performance and accom-
modating the trade-off. It is widely accepted that active suspension is the effective
way to improve suspension performance due to its flexibility of dealing with the
conflicting parameters; further interesting research has been reported to address the
active suspension systems design problems [17].
Many active control methods have been developed to handle the trade-off
between conflicting performance by utilizing different control techniques such as
fuzzy logic and neural network control [1], gain scheduling control [13], linear
optimal control [5], adaptive control [6] and H1 control [3, 7]. In [3], it was shown
that H1 control strategy for active suspension systems can be feasible solutions to
manage the trade-off by compromising the requirements being achieved for the
better combination performances. However, it should be noted that this type of
control method used to improve the system’s performance and robustness is pre-
sented mathematically with an assumption that the proposed controllers can be
accurately realized to a certain extent.
On the other hand, controller-involved uncertainties or faults can be caused by
many reasons, for instance finite word length in digital systems, ageing of controller
devices and imprecision inherent in analogue systems, which may lead to additional
tuning of parameters in the final controller implementation [11, 12, 18–20]. In [21],
the authors study the problem of non-fragile H1 controller for a quarter-car active
suspension system with actuator uncertain. However, there are few results on non-
fragile control for actuator uncertain half-vehicle suspension systems.
Based on the above observation, this chapter designs a non-fragile H1 con-
troller for half-vehicle suspension systems with actuator uncertainty. The main
aims of this chapter are to improve the suspension performances. It presents linear
matrix inequality (LMI)-based sufficient conditions for existence of desired non-
fragile H1 controllers, under which the expected non-fragile H1 controller can be
obtained via convex optimization.
Notation: Rn denotes the n-dimensional Euclidean space and the notation
P > 0 ( 0) means that P is real symmetric and positive definite (semi-definite).
The superscript ‘T’ is used to stand for matrix transposition. In symmetric block
matrices, we use an asterisk ðÞ to represent a term that is induced by symmetry
and diagf. . .g stands for a block-diagonal matrix. The notation symðAÞ is used to
stand for A þ AT : The space of square-integrable vector functions over ½0; 1Þ is
denoted by L2 ½0; 1Þ, and for w ¼ fwðtÞg 2 L2 ½0; 1Þ, its norm is given by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Ð1 2
kwk2 ¼ t¼0 jwðt Þj dt : Matrices, if their dimensions are not explicitly stated, are
assumed to be compatible for algebraic operations.

5.2 Problem formulation

In this chapter, we consider the problem of non-fragile control for the actuator
uncertain half-vehicle suspension systems. First, we introduce the half-vehicle
 Active control for actuator uncertain half-car suspension systems 127

Zc
l1 l2

j
m s, Ij

zsf zsr

uf (t) ur(t)

ksf csf Controller ksr csr Controller

zuf muf zur mur

ktf ktr

zrf zrr

Figure 5.1 A half-vehicle model

model as shown in Figure 5.1, in which zsf ðtÞ is used to denote the front body
displacement, zsr ðtÞ stands for the rear body displacement, l1 is the distance
between the front axle and the centre of mass, l2 is the distance between the rear
axle and the centre of mass, jðtÞ is the pitch angle and zc ðtÞ is the displacement of
the centre of mass. The mass of the car body is denoted by ms , the unsprung
masses on the front and rear wheels are denoted by muf and mur , the pitch moment
of inertia about the center of mass is denoted by Ij , the front and rear unsprung
mass displacements are denoted by zuf ðtÞ and zur ðtÞ respectively. zrf ðtÞ and zrr ðtÞ
stand for the front and rear terrain height displacements, while csf and csr are the
stiffness of the passive elements of the front and rear wheels respectively. ksf and
ksr are the front and rear tyre stiffness, and uf ðtÞ and ur ðtÞ are the front and rear
actuator force inputs respectively. With the assumption of a small pitch angle jðtÞ
in [4], one has

zsf ðtÞ ¼ zc ðtÞ  l1 jðtÞ ð5:1Þ

zsr ðtÞ ¼ zc ðtÞ þ l2 jðtÞ ð5:2Þ
128 Handbook of vehicle suspension control systems

Then, we have
   
ms€z c ðtÞ þ ksf zsf ðtÞ  zuf ðtÞ þ csf z_ sf ðtÞ  z_ uf ðtÞ
þ ksr ½zsr ðtÞ  zur ðtÞ þ csr ½z_ sr ðtÞ  z_ ur ðtÞ ¼ uf ðtÞ þ ur ðtÞ
   
€ ðtÞ  l1 ksf zsf ðtÞ  zuf ðtÞ  l1 csf z_ sf ðtÞ  z_ uf ðtÞ
Ij j
þ l2 ksr ½zsr ðtÞ  zur ðtÞ þ l2 csr ½z_ sr ðtÞ  z_ ur ðtÞ ¼ l1 uf ðtÞ þ l2 ur ðtÞ
    ð5:3Þ
muf €z uf ðtÞ  ksf zsf ðtÞ  zuf ðtÞ  csf z_ sf ðtÞ  z_ uf ðtÞ
 
þ ktf zuf ðtÞ  zrf ðtÞ ¼ uf ðtÞ
mur€z ur ðtÞ  ksr ½zsr ðtÞ  zur ðtÞ  csr ½z_ sr ðtÞ  z_ ur ðtÞ
þ ktr ½zur ðtÞ  zrr ðtÞ ¼ ur ðtÞ

From (5.1)–(5.3), one can see that

€z sf ðtÞ ¼ €z c ðtÞ  l1 j
€ ðt Þ
    
¼ a1 uf ðtÞ  ksf zsf ðtÞ  zuf ðtÞ  csf z_ sf ðtÞ  z_ uf ðtÞ
þ a2 fur ðtÞ  ksr ½zsr ðtÞ  zur ðtÞ  csr ½z_ sr ðtÞ  z_ ur ðtÞg
ð5:4Þ
€ ðt Þ
€z sr ðtÞ ¼ €z c ðtÞ  l2 j
    
¼ a2 uf ðtÞ  ksf zsf ðtÞ  zuf ðtÞ  csf z_ sf ðtÞ  z_ uf ðtÞ
þ a3 fur ðtÞ  ksr ½zsr ðtÞ  zur ðtÞ  csr ½z_ sr ðtÞ  z_ ur ðtÞg

where

1 l2 1 l1 l2 1 l2
a1 ¼ þ 1, a2 ¼  , a3 ¼ þ 2
ms Ij ms Ij ms Ij

We define the following state variables: x1 ðtÞ ¼ zsf ðtÞ  zuf ðtÞ is the suspen-
sion deflection of the front car body, x2 ðtÞ ¼ zsr ðtÞ  zur ðtÞ is the suspension
deflection of the rear car body, x3 ðtÞ ¼ zuf ðtÞ  zrf ðtÞ is the tyre deflection of the
front car body, x4 ðtÞ ¼ zur ðtÞ  zrr ðtÞ is the tyre deflection of the rear car body,
x5 ðtÞ ¼ z_ sf ðtÞ is the vertical velocity of the front car body, x6 ðtÞ ¼ z_ sr ðtÞ is the
vertical velocity of the rear car body, x7 ðtÞ ¼ z_ uf ðtÞ is the vertical velocity of the
front wheel and x8 ðtÞ ¼ z_ ur ðtÞ is the vertical velocity of the rear wheel. After
z_ ðtÞ
choosing the disturbance input wðtÞ ¼ rf and the variables as
z_ rr ðtÞ
 T
xðtÞ ¼ xT1 ðtÞ xT2 ðtÞ xT3 ðtÞ xT4 ðtÞ xT5 ðtÞ xT6 ðtÞ xT7 ðtÞ xT8 ðtÞ
" #
uf ðtÞ
uðtÞ ¼
ur ðtÞ
 Active control for actuator uncertain half-car suspension systems 129

we can express the dynamic equations in (5.3) and (5.4) as the following state-
space form:

x_ ðtÞ ¼ AxðtÞ þ BuðtÞ þ B1 wðtÞ ð5:5Þ

where
2 3
0 0 0 0 1 0 1 0
6 0 0 0 0 0 1 0 1 7
6 7
6 0 0 0 0 0 0 1 0 7
6 7
6 0 0 0 0 0 0 0 1 7
6 7
6 a1 ksf a2 ksr 0 0 a1 csf a2 csr a1 csf a2 csr7
A¼6
6 a2 ksf
7
7
6 a3 ksr 0 0 a2 csf a3 csr a2 csf a3 csr7
6 ksf ktf csf csf 7
6 0  0 0  0 7
6 muf muf muf muf 7
6 7
4 ksr ktr csr csr 5
0 0  0 0 
mur mur mur mur
2 3T
1
60 0 0 0 a1 a2 
muf
0 7
B¼6
4
7
1 5
0 0 0 0 a2 a3 0 
mur
 T
0 0 1 0 0 0 0 0
B1 ¼
0 0 0 1 0 0 0 0
ð5:6Þ

For the active suspension control design problem, it should be noticed that ride
comfort, road holding ability and suspension deflection are three key performance
characteristics to be considered. First of all, it is widely accepted that ride comfort
is closely related to the vertical acceleration experienced by the car body. In this
study, both the heave and the pitch accelerations are chosen as the performance
output vector, namely
 
€z c ðtÞ
z1 ðtÞ ¼
€ ðt Þ
j

To ensure that the designed controllers must have the capability of performing the
suspension system, subject to the mechanical constraints of vehicle components
and passenger comfort generation, the inequalities as below are provided to guar-
antee the suspension deflection constraint:

jzsf ðtÞ  zuf ðtÞj  zf max , jzsr ðtÞ  zur ðtÞj  zr max ð5:7Þ

where zf max and zr max denote the maximum front and rear suspension deflection
hard limits respectively.
130 Handbook of vehicle suspension control systems

Moreover, to ensure a firm uninterrupted contact of the wheels with the road, it
is very reasonably assumed that the dynamic tyre loads should not exceed the static
tyre loads for both the front and rear wheels:

jksf zuf ðtÞ  zrf ðtÞ j  Ff , jksr ðzur ðtÞ  zrr ðtÞÞj  Fr ð5:8Þ

where Ff and Fr stand for static type loads that can be calculated by

Fr ðl1 þ l2 Þ ¼ ms gl1 þ mur gðl1 þ l2 Þ ð5:9Þ

Ff þ Fr ¼ ms þ muf þ mur g ð5:10Þ

Of the above conditions, the conditions in (5.7) and (5.8) are chosen as con-
straint output; the active vehicle suspension system can be rewritten as follows:

x_ ðtÞ ¼ AxðtÞ þ B1 wðtÞ þ BuðtÞ

z1 ðtÞ ¼ C1 xðtÞ þ D1 uðtÞ ð5:11Þ
z2 ðtÞ ¼ C2 xðtÞ

where A, B1 and B are defined in (5.5), and

2 ksf ksr csf csr csf csr 3

  0 0  
6 ms ms ms ms ms ms 7
C1 ¼ 6
4 l1 ksf
7
l2 ksr l1 csf l2 csr l1 csf l2 csr 5
 0 0  
Ij Ij Ij Ij Ij Ij
2 3
1
6 zf max 0 0 0 0 0 0 07
6 7
6 7
2 1 1 3 6 1 7
6 0 0 0 0 0 0 07
6 ms ms 7
7,
6 zr max 7
7
D1 ¼ 6 C2 ¼ 6
4 l1 l2 5 6 ksf 7
 6 0
6 0 0 0 0 0 077
Ij Ij 6 Ff 7
6 7
4 ksr 5
0 0 0 0 0 0 0
Fr
ð5:12Þ
As we know that actuator uncertainties exist in many cases and can be caused
by many factors such as buffer size in digital systems and ageing of controller
devices for active suspension systems. Taking advantage of the fact that the non-
fragile feedback controller is insensitive to gain changes of feedback control, we
construct the following state feedback controller:

uðtÞ ¼ KxðtÞ ð5:13Þ

 Active control for actuator uncertain half-car suspension systems 131

where K is to be designed for the objective of non-fragile control problem. In this

chapter, the controller uncertainty is considered in the following form:
DK ðtÞ ¼ HF ðtÞE ð5:14Þ
where H and E are known constant matrices with appropriate dimensions, and
F ðtÞ is unknown matrix functions with the property F T ðtÞF ðtÞ  I:
Then, the closed-loop system with the non-fragile controller (5.13) can be
written as

x_ ðtÞ ¼ AxðtÞ þ B1 wðtÞ

z1 ðtÞ ¼ C 1 xðtÞ ð5:15Þ
z2 ðtÞ ¼ C2 xðtÞ
where

A ¼ A þ BðK þ DK ðtÞÞ, C 1 ¼ C1 þ D1 ðK þ DK ðtÞÞ

It is generally assumed that w 2 L2 ½0; 1Þ, without loss of generality,
kwk22  wmax < 1 is obtained. In short, the objective of this chapter is to design a
non-fragile controller in (5.13) such that
1. the closed-loop system is asymptotically stable;
2. under zero initial condition, the closed-loop system guarantees that
kz1 k2 < gkwk2 for all non-zero w 2 L2 ½0; 1Þ, where g > 0 is a prescribed
scalar;
3. the following control output constraints are guaranteed:
jfz2 ðtÞgq j  1; q ¼ 1; 2; 3; 4 ð5:16Þ

5.3 Main results

In the above proposed control strategy, the multiple requirements, namely ride
comfort, road holding and suspension deflection, are formulated in a unified fra-
mework. Note that the lemma below is employed for developing our main results in
the next section.
Lemma 5.1 [16]. Let Q, Y and ^ are real matrices of appropriate dimensions, and a
time-varying matrix F ðtÞ satisfying F T ðtÞF ðtÞ  I, then

Q þ YF ðtÞ^ þ ^T F T ðtÞYT < 0

if and only if there exists a scalar e > 0 such that

Q þ e1 YYT þ e^T ^ < 0:

In this section, we design H1 non-fragile state-feedback controller such that

the closed-loop system in (5.15) is asymptotically stable and can also ensure a
132 Handbook of vehicle suspension control systems

prescribed gain from disturbance wðtÞ to performance output z1 ðtÞ while keeping
the output constraints in (5.16) satisfied. First, we have the following proposition.
Proposition 5.1. For the closed-loop system in (5.15) and given matrix K, if there
exist matrix P > 0 and q ¼ 1; 2; 3; 4 satisfying
2
T 3
sym PA PB1 C 1
6 7
4  g2 I 0 5 < 0 ð5:17Þ
  I
" pffiffiffi #
P rfC2 gTq
<0 ð5:18Þ
 I
then a non-fragile controller in the form of (5.13) is proposed such that
1. the closed-loop system is asymptotically stable;
2. the performance kTz1 w k1 < g is minimized subject to output constraints in
(5.16) with the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 ,
where Tz1 w denotes the closed-loop transfer function from the road disturbance
wðtÞ to the control output z1 ðtÞ:
Proof. In this proof, we consider the asymptotic stability of the system in (5.15)
under the condition of wðtÞ ¼ 0. Then, we give the H1 performance analysis
condition based on the stability criterion. Hence the following Lyapunov function
is proposed for the system (5.15) as
V ðtÞ ¼ xT ðtÞPxðtÞ ð5:19Þ

Then, we have V_ ðtÞ ¼ x ðtÞ symðPAÞ xðtÞ. It can be seen from (5.17) in Theorem 5.1
T

that symðPAÞ < 0, which implies that the system in (5.15) with wðtÞ ¼ 0 is
asymptotically stable. The H1 performance of the active suspension system is
presented under zero initial condition. It is not difficult to obtain
zT1 ðtÞz1 ðtÞ  g2 wT ðtÞwðtÞ þ V_ ðtÞ ¼ xT ðtÞQxðtÞ ð5:20Þ
where
 T

 T
x ðt Þ ¼ x ðt ÞT T
w ðt Þ , Q ¼ symðPA þ C 1 C 1 PB1 ð5:21Þ
 g2 I
By using Schur complement to (5.17), one has Q < 0; which means

zT1 ðtÞz1 ðtÞ  g2 wT ðtÞwðtÞ þ V_ ðtÞ < 0 ð5:22Þ

for any non-zero w 2 L2 ½0; 1Þ, which indicates J < 0. Hence kz1 k2 < gkwk2 is
guaranteed for any non-zero w 2 L2 ½0; 1Þ.
In the following part, the problem the output constraints in (5.16) is con-
sidered. Based on [7] and (5.16), it is obtained that the constraints are guaranteed, if

rP2 fC2 gTq fC2 gq P2 < I,

1 1
q ¼ 1; 2; 3; 4: ð5:23Þ
 Active control for actuator uncertain half-car suspension systems 133

By Schur complement, (5.23) is equivalent to (5.25) in Proposition 5.1, and the

proof is completed. &
We will have the following Theorem 5.1 to handle the uncertainty existing in
Proposition 5.1.

Theorem 5.1. For the closed-loop system in (5.15) and given matrix K, if there
exist matrix P > 0, and scalar e, ðq ¼ 1; 2; 3; 4Þ satisfying
2 3
W PB1 C1T þ K T DT1 eET PBH
6 7
6  g2 0 0 0 7
6 7
6 D1 H 7
6  I 0 7<0 ð5:24Þ
6 7
4   eI 0 5
    eI
" pffiffiffi #
P rfC2 gTq
<0 ð5:25Þ
 I

where

W ¼ symðPðA þ BK ÞÞ: ð5:26Þ

then
1. the closed-loop system is asymptotically stable;
2. the performance kTz1 w k1 < g is minimized subject to output constraints in
(5.16) with the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 .

Proof. From (5.17), it can be seen that

2
T 3 2 3
sym PA PB1 C1 symðPðA þ BK ÞÞ PB1 ðC1 þ D1 K ÞT
6 7 6 7
4  g2 I 0 5¼4  g2 I 0 5
  I   I
02 3 1
PBH
þ sym@4 0 5F ðtÞ½ E 0 0 A < 0
DH

On the other hand, by using Schur complement, (5.24) is equivalent to

2 3
symðPðA þ BK ÞÞ PB1 ðC1 þ D1 K ÞT
6 7
4  g2 0 5
 I 
2 32 2 3T 3T
PBH PBH E
þ e1 4 0 54 0 5 þ e4 0 5 ½ E 0 0 < 0
DH DH 0
134 Handbook of vehicle suspension control systems

which implies the condition (5.17) holds via Lemma 5.1. Therefore, it is obtained
that the conditions in Proposition 5.1 are satisfied. &
Based on the proposed criterion in Theorem 5.1, the existence condition of
non-fragile H1 controller is presented in the following theorem.

Theorem 5.2. For the closed-loop system in (5.15), if there exist matrices P > 0
and K and scalar e ðq ¼ 1; 2; 3; 4Þ such that the following LMIs hold:
2 T
3
W B1 PC1T þ K DT1 PET eBH
6 7
6  g2
6 0 0 0 7 7
6 7
6  I 0 eD1 H 7 < 0 ð5:27Þ
6 7
6 7
4   eI 0 5
    eI
" pffiffiffi #
P rP fC2 gTq
<0 ð5:28Þ
 I

where

W ¼ sym AP þ BK ,
then a non-fragile controller in the form of (5.13) exists such that
1. the closed-loop system is asymptotically stable;
2. the performance kTz1 w k1 < g is minimized subject to output constraint in
(5.16) with the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 .
Moreover, if inequalities (5.27) and (5.28) have a feasible solution, then the
1
controller in (5.13) is given by K ¼ K P :
1 1
Proof. We define K ¼ K P , P ¼ P and e ¼ e 1 . By utilizing Schur comple-
ment, (5.27) and (5.28) are equivalent to (5.24) and (5.25) by performing con-
gruence transformations with
diagfP, I, I, I, I, I g, diagfP, I g
respectively. Therefore, all the conditions in Theorem 5.1 are satisfied. The proof is
completed. &

5.4 Simulation results

In this section, simulation results show the effectiveness of the proposed method.
First, we provide the half-vehicle model parameters [4] in Table 5.1.
A non-fragile H1 controller is developed for the half-car active suspension
system in (5.15) by employing the parameters in Table 5.1. It is assumed that
the maximum allowable suspension strokes zmax f ¼ zmax r ¼ 0:08 m. r ¼ 1 is set
 Active control for actuator uncertain half-car suspension systems 135

Table 5.1 Systems parameter values for the half-car suspension model

ms muf ksf ktf csf l1

690 kg 40 kg 18,000 N/m 200,000 N/m 1,000 Ns/m 1.3 m
Ij mur ksr ktr csr l2
1,222 kgm2 45 kg 22,000 N/m 200,000 N/m 1,000 Ns/m 1.5 m

according to the previous study in [3]. For actuator uncertainties in (5.13), we

consider the following two possible cases:
1. Case I:
   
1 0 sinðtÞ 0
H¼ , F ðt Þ ¼ ,
0 1 0 sinðtÞ
 
0:3 0:5 0 0:1 1 0:8 0:9 1
E¼
2:1 2:5 2 1 2 0:8 0:2 2:5
2. Case II:
   
0:5 0 cosðtÞ 0
H¼ , F ðt Þ ¼ ,
0 0:5 0 cosðtÞ
 
0:8 0 0:8 0:1 1 1:8 5:9 2
E¼
1 3:5 2:5 1 2 0:8 0:2 2:5

According to Theorem 5.1, utilizing convex optimization method, we obtain

the minimum guaranteed closed-loop H1 performance indexes as 5.5419 and
5.5421 for Case I and Case II respectively. And the admissible control gain
matrices can be obtained as follows:

Kcase1 ¼ 105
 
1:0317 0:0848 0:6468 0:0176 0:1963 0:0329 0:0053 0:0055

0:2684 0:8364 0:0146 0:6233 0:0224 0:1540 0:0016 0:0050
5
Kcase2 ¼ 10
 
1:0294 0:0824 0:6466 0:0154 0:1960 0:0327 0:0052 0:0054

0:2647 0:8498 0:0149 0:6376 0:0217 0:1557 0:0015 0:0051

To check the effectiveness of the proposed non-fragile controllers, it is expected

that the desired controller can satisfy the following: (1) the first control output z1 ðtÞ
including the heave acceleration €z c ðtÞ and the pitch acceleration j€ ðtÞ is as small as
possible; (2) the suspension deflection is below the maximum allowable suspension
strokes zf max ¼ 0:08 m and zr max ¼ 0:08 m, which shows z2 ðtÞ1 < 1 and z2 ðtÞ2 < 1;
(3) the controlled output defined in (5.11) satisfies z2 ðtÞ3 < 1 and z2 ðtÞ4 < 1.
136 Handbook of vehicle suspension control systems

In order to evaluate the suspension characteristics with respect to ride comfort,

vehicle handling and working space of the suspension, the variability of the road
profiles is taken into account. In the context of active suspension performance,
road disturbances can be generally assumed as shocks. Shocks are discrete
events of relatively short duration and high intensity, caused by, for example, a
pronounced bump or pothole on an otherwise smooth road surface. In this work,
this case of road profile is considered first to reveal the transient response
characteristic, which is given by
8   
> A 2pV L
< 1  cos t , if 0  t 
zrf ðtÞ ¼ 2 L V ð5:29Þ
>
: 0; L
if t >
V
where A and L are the height and the length of the bump. Assume A ¼ 0:05 m,
L ¼ 2 m and the vehicle forward velocity as V ¼ 20 km/h. In this section, we
assume that the road condition zrr ðtÞ for the rear wheel is same as the front wheel
but with a time delay of ðl1 þ l2 Þ=V : Figure 5.2 illustrates the corresponding
ground velocities for the front and rear wheels.
For Case I, Figures 5.3–5.5 show the responses of the heave accelerations and
the pitch acceleration, the front and rear suspension deflection constraints and the
front and rear tyre deflection constrains of the passive and closed-loop systems.
One can see from Figure 5.3 that the values of the heave acceleration €z c ðtÞ and the

0.5
Front wheel
0.4 Rear wheel

0.3

0.2
Ground velocity (m/s)

0.1

−0.1

−0.2

−0.3

−0.4

−0.5
0 1 2 3 4 5
Time (s)

Figure 5.2 Ground disturbance

 Active control for actuator uncertain half-car suspension systems 137

Heave acceleration (m/s2)

Passive
2 K

−2

−4
0 1 2 3 4 5

1
Pitch acceleration (m/s2)

Passive
0.5 K

−0.5

−1
0 1 2 3 4 5
Time (s)

Figure 5.3 Responses of the heave accelerations and the pitch acceleration

Suspension deflection constraint (front)

1
Passive
0.5 K

−0.5

−1
0 1 2 3 4 5
Time (s)

Suspension deflection constraint (rear)

0.1
Passive
0.05 K

−0.05

−0.1
0 1 2 3 4 5
Time (s)

Figure 5.4 Responses of the front and rear suspension deflection constraints
138 Handbook of vehicle suspension control systems

The relation dynamic tyre load (front)

0.4
Passive
0.2 K

−0.2

−0.4
0 1 2 3 4 5
Time (s)

The relation dynamic tyre load (rear)

0.4
Passive
0.2 K

−0.2

−0.4
0 1 2 3 4 5
Time (s)

Figure 5.5 Responses of tyre deflection constraints

500
Active force (front) (N)

−500

−1000
0 1 2 3 4 5
Time (s)

1000
K
Active force (rear) (N)

500

−500

−1000
0 1 2 3 4 5
Time (s)

Figure 5.6 Responses of active forces

 Active control for actuator uncertain half-car suspension systems 139

Heave acceleration (m/s2)

Passive
2 K

−2

−4
0 1 2 3 4 5

1
Pitch acceleration (m/s2)

Passive
0.5 K

−0.5

−1
0 1 2 3 4 5
Time (s)

Figure 5.7 Responses of the heave accelerations and the pitch acceleration

Suspension deflection constraint (front)

1
Passive
0.5 K

−0.5

−1
0 1 2 3 4 5
Time (s)

Suspension deflection constraint (rear)

0.1
Passive
0.05 K

−0.05

−0.1
0 1 2 3 4 5
Time (s)

Figure 5.8 Responses of the front and rear suspension deflection constraints
140 Handbook of vehicle suspension control systems

The relation dynamic tyre load (front)

0.4
Passive
0.2 K

−0.2

−0.4
0 1 2 3 4 5
Time (s)

The relation dynamic tyre load (rear)

0.4
Passive
0.2 K

−0.2

−0.4
0 1 2 3 4 5
Time (s)

Figure 5.9 Responses of tyre deflection constraints

500
Active force (front) (N)

−500

−1000
0 1 2 3 4 5
Time (s)

1000
K
Active force (rear) (N)

500

−500

−1000
0 1 2 3 4 5
Time (s)

Figure 5.10 Responses of active forces

 Active control for actuator uncertain half-car suspension systems 141

pitch acceleration j € ðtÞ in closed-loop systems are much less than the the open-loop
systems and an improved ride comfort has been achieved by using the non-fragile
H1 controller under the actuator uncertainties.
Furthermore, Figure 5.4 demonstrates that both the suspension stroke con-
straints x1 ðtÞ=zf max < 1 and x2 ðtÞ=zr max < 1 can be guaranteed, which yields
z2 ðtÞ1 < 1 and z2 ðtÞ2 < 1. Figure 5.5 further validates that the relation dynamic
front tyre load ktf x3 ðtÞ=Ff and rear tire load ktr x4 ðtÞ=Fr are all below 1, which
means z2 ðtÞ3 < 1 and z2 ðtÞ4 < 1. Figure 5.6 shows the responses of the front and
rear active forces. From these figures, we know that the closed-loop system is
asymptotically stable with the guaranteed output constraints and has a much better
performance than the open-loop system with actuator uncertainties. Furthermore,
for Case II, it also can be seen from Figures 5.7–5.10 that the non-fragile controller
can improve the suspension performances.

5.5 Conclusion
This chapter has considered the problem of non-fragile H1 controller design for
half-vehicle active suspension systems with actuator uncertainty. Based on Lya-
punov stability theory, a non-fragile H1 controller has been designed such that the
closed-loop system is satisfied with constraint performance and also guaranteed
asymptotically stable with a prescribed H1 disturbance attenuation level. Simula-
tion results have shown the effectiveness of the proposed approach.

References
[1] N. Al-Holou, T. Lahdhiri, D. Joo, J. Weaver, and F. Al-Abbas. ‘Sliding
mode neural network inference fuzzy logic control foractive suspension
systems’. IEEE Transactions on Fuzzy Systems, 10(2):234–246, 2002.
[2] J. Cao, H. Liu, P. Li, and D. Brown. ‘State of the art in vehicle active sus-
pension adaptive control systems based on intelligent methodologies’. IEEE
Transactions on Intelligent Transportation Systems, 9(3):392–405, 2008.
[3] H. Chen and K. Guo. ‘Constrained H1 control of active suspensions: An
LMI approach’. IEEE Transactions on Control Systems Technology, 13(3):
412–421, 2005.
[4] H. Du and N. Zhang. ‘Constrained H1 control of active suspension for a
half-car model with a time delay in control’. Proceedings of the Institution of
Mechanical Engineers, Part D: Journal of Automobile Engineering, 222(5):
665–684, 2008.
[5] M. El Madany and Z. Abduljabbar. ‘Linear quadratic Gaussian control of a
quarter-car suspension’. Vehicle System Dynamics, 32(6):479–497, 1999.
[6] I. Fialho and G. Balas. ‘Road adaptive active suspension design using linear
parameter-varying gain-scheduling’. IEEE Transactions on Control Systems
Technology, 10(1):43–54, 2002.
[7] H. Gao, W. Sun, and P. Shi. ‘Robust sampled-data H1 control for vehicle
active suspension systems’. IEEE Transactions on Control Systems Technology,
18(1):238–245, 2010.
142 Handbook of vehicle suspension control systems

[8] T. Gordon, C. Marsh, and M. Milsted. ‘A comparison of adaptive LQG

and nonlinear controllers for vehicle suspension systems’. Vehicle System
Dynamics, 20(6):321–340, 1991.
[9] D. Hrovat. ‘Survey of advanced suspension developments and related opti-
mal control applications’. Automatica, 33(10):1781–1817, 1997.
[10] D. Hrovat, D. Margolis, and M. Hubbard. ‘An approach toward the optimal
semi-active suspension’. Journal of Dynamic Systems, Measurement, and
Control, 110:288, 1988.
[11] B. Jiang, K. Zhang, and P. Shi. ‘Integrated fault estimation and accommoda-
tion design for discrete-time Takagi–Sugeno fuzzy systems with actuator
faults’. IEEE Transactions on Fuzzy Systems, 19(2):291–304, 2011.
[12] L. Keel and S. Bhattacharyya. ‘Robust, fragile or optimal’? Proceedings of
the 1997 American Control Conference, 1997. vol. 2, 1997.
[13] C. Sivrioglu and I. Cansever. ‘LPV gain-scheduling controller design for a
non-linear quarter-vehicle active suspension system’. Transactions of the
Institute of Measurement and Control, 31(1):71–95, 2009.
[14] J. Tamboli and S. Joshi. ‘Optimum design of a passive suspension system of
a vehicle subjected to actual random road excitations’. Journal of Sound and
Vibration, 219(2):193–205, 1999.
[15] C. Ting, T. Li, and F. Kung. ‘Design of fuzzy controller for active suspen-
sion system’. Mechatronics, 5(4):365–383, 1995.
[16] Y. Wang, L. Xie, and C. De Souza. ‘Robust control of a class of uncertain
nonlinear systems’. Systems & Control Letters, 19(2):139–149, 1992.
[17] M. Yamashita, K. Fujimori, K. Hayakawa, and H. Kimura. ‘Application of
H1 control to active suspension systems’. Automatica, 30(11):1717–1729,
1994.
[18] G. Yang and J. Liang Wang. ‘Non-fragile protect H1 control for linear
systems with multiplicative controller gain variations’. Automatica, 37(5):
727–737, 2001.
[19] G. Yang and J. Wang. ‘Non-fragile H1 output feedback controller design for
linear systems’. Journal of Dynamic Systems, Measurement, and Control,
125:117, 2003.
[20] B. Zhang, S. Zhou, and T. Li. ‘A new approach to robust and non-fragile
H1 control for uncertain fuzzy systems’. Information Sciences, 177(22):
5118–5133, 2007.
[21] J. Zhang, P. Shi, and J. Qiu. ‘Non-fragile guaranteed cost control for
uncertain stochastic nonlinear time-delay systems’. Journal of the Franklin
Institute, 346(7):676–690, 2009.
 Chapter 6
Active suspension control with finite
frequency approach
Weichao Sun, Huihui Pan, Pinchao Wang and Huijun Gao

Abstract

In this chapter, the problem of vehicle active suspension control with frequency
band constraints is investigated. According to the online availability of state mea-
surements, both state feedback and dynamic output feedback control problems
are solved, based on the generalized Kalman–Yakubovich–Popov (KYP) lemma.
Compared with the traditional entire frequency approach for active suspension
systems, the finite frequency approach proposed in this chapter achieves better
disturbance attenuation performance for the chosen frequency range, and meantime
the constraints required by real situation are guaranteed in the controller design.
The effectiveness and merits of the proposed method are verified by a number of
simulations with several types of road disturbances.

6.1 Introduction
Nowadays, one of the popular methods dealing with active suspension control is
based on state-space realization [1–10]. When all the states are online measur-
able, state feedback is an excellent choice, as it can make use of full information,
and thus the closed-loop performance can be enhanced to its full potential.
Though state feedback control is a powerful strategy, it is based on the premise
that all the state variables are online measurable, which sometimes introduces
higher cost and additional complexity by measuring all the states. In the cases
where not all the state variables can be measured online, output feedback control
is an alternative, which can conduct effective control according to part of the
measured states. In other words, output feedback strategy requires less sensors,
compared with the state feedback counterparts, and has been investigated in
many studies.

Research Institute of Intelligent Control and Systems, Harbin Institute of Technology, Harbin, China
144 Handbook of vehicle suspension control systems

Concerning active suspension systems, various control strategies have been

introduced aiming at improving the ride comfort performance [11–17]. It is worth
mentioning that most researchers design controllers for suspension systems over
the entire frequency range, and the existing results pay little attention to the fact
that active suspension systems just belong to certain frequency band. From the
ISO2361, the human body is more sensitive to vibrations of 4–8 Hz in the vertical
direction, and human’s organs will resonate with the vibrations in this frequency
domain. Hence, the development of the finite frequency control is significant for
active suspension systems.
In this chapter, the problem of vehicle active suspension control with fre-
quency band constraints is considered, where state feedback control and dynamic
output feedback control cases are taken into account in a unified framework. The
quarter-car model is employed as the object of research. By using the generalized
KYP lemma [18–23], the finite frequency problems are transformed into a set of
linear matrix inequalities (LMIs) to be solved. In addition, the time-domain con-
straints, which represent performance requirements for vehicle suspensions (road
holding and suspension deflection), are guaranteed in the controller design. Under
the assumption that all the state variables are online measurable, a state feedback
H1 controller over the finite frequency range is designed at first to meet the finite
frequency characteristic of the active suspension systems. In order to cope with the
real situation, a dynamic output feedback controller of order equal to the plant is
also designed, where an effective multiplier expansion is used to convert the con-
dition of controller design to an LMI condition through a linearizing change of
variables. The desired controllers can be obtained by solving a set of LMIs using
standard numerical algorithms [24]. The effectiveness of the proposed approach is
shown by a number of simulations.
Notation: For a matrix P, PT , P1 and P? denote its transpose, inverse, and
orthogonal complement, respectively; the notation P > 0 (0) means that P is real
symmetric and positive definite (semi-definite); and ½Ps means P þ PT : For a
vector or matrix, fgi (i ¼ 1; 2; . . .) represents the ith line of the vector or matrix,
and kGk1 denotes the H1 -norm of transfer function matrix GðsÞ: In symmetric
block matrices or complex matrix expressions, we use an asterisk ðÞ to represent a
term that is induced by symmetry and diagf. . .g stands for a block-diagonal matrix.
The space of square-integrable vector functions over ½0;1Þ is denoted by L2 ½0;1Þ,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Ð1 2
and for w ¼ fwðtÞg 2 L2 ½0;1Þ, its norm is given by kwk2 ¼ t¼0 jwðt Þj dt :

6.2 Problem formulation

By considering the vertical dynamics and taking into account the vehicle’s sym-
metry, a suspension can in the first step be reduced to the so-called quarter-car
model as shown in Fig. 6.1.
This model has been used extensively in the literature and captures many
important characteristics of more detailed models. In this chapter, the effect of
 Active suspension control with finite frequency approach 145

Body ms
zs

Fd
Active
u
suspension
Fs

Unspring mass mu zu

Wheel
zr Ft Fb

Fig. 6.1 Quarter-car model with an active suspension

Table 6.1 Parameters of the active suspension system

ms Sprung mass mu Unsprung mass

cs Damping of the suspension system ks Stiffness of the suspension system
kt Compressibility of the pneumatic tire ct Damping of the pneumatic tire
zs Displacements of the sprung masses zu Displacements of the unsprung masses
zr Road displacement input u Active input of the suspension system

actuator dynamics is neglected and the actuator is modeled as an ideal force gen-
erator. The parameters are defined in Table 6.1.
The ideal dynamic equations of the sprung and unsprung masses are given as
ms€z s ðtÞ þ cs ½_z s ðtÞ  z_ u ðtÞ þ ks ½zs ðtÞ  zu ðtÞ ¼ uðtÞ
mu€z u ðtÞ þ cs ½_z u ðtÞ  z_ s ðtÞ þ ks ½zu ðtÞ  zs ðtÞ ð6:1Þ
þ kt ½zu ðtÞ  zr ðtÞ þ ct ½_z u ðtÞ  z_ r ðtÞ ¼ uðtÞ

Define the following state variables:

x1 ðtÞ ¼ zs ðtÞ  zu ðtÞ, x2 ðtÞ ¼ zu ðtÞ  zr ðtÞ, x3 ðtÞ ¼ z_ s ðtÞ, x4 ðtÞ ¼ z_ u ðtÞ
ð6:2Þ
where x1 ðtÞ denotes the suspension deflection, x2 ðtÞ is the tire deflection, x3 ðtÞ is
the sprung mass speed, and x4 ðtÞ denotes the unsprung mass speed. We define the
disturbance inputs as wðtÞ ¼ z_ r ðtÞ: Then, by defining

xðtÞ ¼ ½ x1 ðtÞ x2 ðtÞ x3 ðtÞ x4 ðtÞ T

146 Handbook of vehicle suspension control systems

the dynamic equations in (6.1) can be rewritten in the following state-space form:
x_ ðtÞ ¼ AxðtÞ þ B1 wðtÞ þ BuðtÞ ð6:3Þ
where
2 3 2 3
0 0 1 1 0 2 3
6 7 6 7 0
6 0 0 0 1 7 6 0 7 6 7
6 7 6 7 6 1 7
6 k cs cs 7 6 1 7 6 7
A ¼ 6 s 0  7, B¼6 7, B1 ¼ 6 7
6 ms ms ms 7 6 ms 7 6 0 7
6 7 6 7 4 ct 5
4 ks ku cs cs þ ct 5 4 1 5
   mu
mu mu mu mu mu
ð6:4Þ
In designing the control law for a suspension system, we need to consider ride
comfort. It is widely accepted that ride comfort is closely related to the body
acceleration in frequency band 4–8 Hz. Consequently, in order to improve ride
comfort it is important to keep the transfer function from the disturbance inputs,
wðtÞ, to car body acceleration, €z s ðtÞ, as small as possible over the frequency band
4–8 Hz.
Due to the disturbances caused by road bumpiness, a firm uninterrupted con-
tact of wheels with road is important for vehicle handling and is essentially related
to ride safety. In order to make sure the car safety, we should ensure the firm
uninterrupted contact of wheels to road, and the dynamic tire load should be small,
that is
kt jzu ðtÞ  zr ðtÞj < ðms þ mu Þg
In addition, the structural features of the vehicle also constrain the amount of
suspension deflection, that is
jzs ðtÞ  zu ðtÞj  zmax
where zmax is the maximum suspension deflection.
In order to satisfy the performance requirements, the controlled outputs are
defined by
z1 ðtÞ ¼ €z s ðtÞ
 
zs ðtÞ  zu ðtÞ kt jzu ðtÞ  zr ðtÞj T ð6:5Þ
z2 ðtÞ ¼
zmax ðms þ mu Þg
Therefore, the vehicle suspension control system can be described by
x_ ðtÞ ¼ AxðtÞ þ BuðtÞ þ B1 wðtÞ
z1ðtÞ ¼ C1 xðtÞ þ D1 uðtÞ
ð6:6Þ
z2 ðtÞ ¼ C2 xðtÞ
yðtÞ ¼ CxðtÞ
 Active suspension control with finite frequency approach 147

where C is the measuring matrix, and

 
ks cs cs
C1 ¼  0  , D1 ¼ m1
ms ms ms s

2 3
1
6 zmax 0 0 07
6 7
C2 ¼6 7
4 kt 5
0 0 0
ðms þ mu Þg

Denote GðjwÞ as the transfer function from the disturbance inputs wðtÞ to the
controlled output z1 ðtÞ: The finite frequency H1 control problem is to design a
controller such that the closed-loop system guarantees

sup kGðjwÞk1 < g ð6:7Þ

w1 <w<w2

where g > 0 is a prescribed scalar, and w1 , w2 represent the upper and lower
bounds, respectively, of the concerned frequency. In addition, from the safety and
mechanical structure point of view, the constraints

jfz2 ðtÞgi j  1; i ¼ 1; 2 ð6:8Þ

are guaranteed.
To facilitate the presentation, we introduce the essential lemmas. For the sake
of brevity, all the proofs of the lemmas have been omitted.
Lemma 6.1 (Projection lemma [25]). Let G, L, Q be given. There exists a matrix
F satisfying

GFL þ ðGFLÞT þ Q < 0

if and only if the following two conditions hold

T ? ?T
G? QG? < 0; LT QLT <0

Lemma 6.2 (Reciprocal projection lemma [25]). Let P be any given positive
definite matrix. The following statements are equivalent:
1. Y þ S þ S T < 0:
2. The LMI problem
 
Y þ P  ½ X s ST þ X T
<0
 P

is feasible with respect to X :

148 Handbook of vehicle suspension control systems

Lemma 6.3 (Generalized KYP lemma [20]). Consider the linear system ðA, B,
C, DÞ: Given a symmetric matrix P, the following statements are equivalent:
1. The finite frequency inequality
 T  
GðjwÞ GðjwÞ
P < 0; w 1  w  w 2
I I
2. There exist symmetric matrices P and Q satisfying Q > 0 and
  T 
G½P, Q, C, D C D <0
 I

where G½P, Q, C, D is defined as

 T   
A B Q P þ j wcQ A B
G½P, Q, C, D ¼
I 0 P  jw c Q w 1 w 2 Q I 0
 T 
þ 0 T
C P12
 ½D P12 s þ P22
and P12 and P22 are the bounds’ right and lower right block matrices of P:

6.3 State feedback controller design

In this section, our purpose is to design a controller that can ensure the transfer
function from the disturbance inputs wðtÞ to the controlled output z1 ðtÞ as small as
possible over the frequency band 4–8 Hz, while the time-domain constraints in
(6.8) are guaranteed. Here, it is assumed that all the state variables can be mea-
sured, and we are interested in designing a state feedback controller:
uðtÞ ¼ KxðtÞ, ð6:9Þ
where K is the state feedback gain matrix to be designed. By combining (6.9) with
(6.6), the closed-loop system is given by
x_ ðtÞ ¼ AxðtÞ þ BwðtÞ
z1 ðtÞ ¼ CxðtÞ þ DwðtÞ ð6:10Þ
z2 ðtÞ ¼ C2 xðtÞ
where
   
A B A þ BK B1
¼
C D C1 þ D1 K 0
For the active suspension systems, in accordance with the requirements, the
constrained H1 control problem is formulated to minimize the H1 norm from wðtÞ
to z1 ðtÞ under the time-domain constraints (6.8) over the fixed frequency band
w1  w  w2 .
 Active suspension control with finite frequency approach 149

Theorem 6.1. Let positive scalars g, h, and r be given. A state feedback controller
in the form of (6.9) exists, such that the closed-loop system in (6.10) is asympto-
tically stable with wðtÞ ¼ 0; and satisfies kGðjwÞkw11 <w<w2 < g for all nonzero
w 2 L2 ½0;1Þ, while the constraints in (6.8) are guaranteed with the disturbance
energy under the bound wmax ¼ ðr  V ð0ÞÞ=h, if there exist symmetric matrices
P, P1 > 0; Q > 0 and general matrix F satisfying
2 3
½F s F T A þ P1 FT FT B
6  P1 0 0 7
6 7< 0 ð6:11Þ
4   P1 0 5
   hI
2 3
Q P þ jwc Q  F 0 0
6  w w2 Q þ ½F T A F T B T
6 C 7 7< 0
4 
1 s ð6:12Þ
 g2 I 0 5
   I
 pffiffiffi 
I rfC2 gi
< 0; i ¼ 1; 2 ð6:13Þ
 P1

where wc ¼ ðw1 þ w2 Þ=2 is a given scalar.

Proof. By using Schur complement, inequality (6.11) is equivalent to
" #
1 T T
F BB F þ F T P1
1 F  ½Fs F T A þ P1
h <0 ð6:14Þ
 P1

Performing the congruence transformation to inequality (6.14) by diagfF 1 , P1

1 g,
with F :¼ W 1, inequality (6.14) can be transformed to the following inequality:
2 3
1 T 1 1 T
4 h BB þ P1  ½W s AP1 þ W 5< 0 ð6:15Þ
 P1
1

By using Lemma 6.2, inequality (6.15) is equivalent to

1 T 1 T
AP1
1 þ P1 A þ B B < 0
h
T
with Y ¼ 1h BB and S T ¼ AP1
1 . Clearly, we have

T 1 T
A P 1 þ P1 A þ P 1 B B P 1 < 0 ð6:16Þ
h
which can guarantee
T
A P 1 þ P1 A < 0
150 Handbook of vehicle suspension control systems

From the standard Lyapunov theory for the continuous-time linear system, the
closed-loop system (6.10) is asymptotically stable with wðtÞ ¼ 0:
Rewrite inequality (6.12) as

J X J T þ HPH T þ ½GFLs < 0 ð6:17Þ

where
2 3
I 0  
Q P þ jw c Q
J ¼ 40 I 5, X¼
P  jw c Q w 1 w 2 Q
0 0
   T
I 0 0 C 0
P¼ , H¼
0 g2 I 0 0 I
 T
G ¼ I A B , L ¼ ½0 I 0
Then, according to Lemma 6.1, inequality (6.17) holds if and only if

W T ðJ X J T þ HPH T ÞW < 0
ð6:18Þ
U ðJ X J T þ HPH T ÞU T < 0

where
 T  T

I 0 0 0
W¼ , U ¼ AT I
0 0 I B 0 I
Note that inequality (6.18) can be transformed to the following form:
 T    T  
A B A B C 0 C 0
X þ P <0 ð6:19Þ
I 0 I 0 0 I 0 I
which can be further transformed to
 T  
Lþ C 0 C 0 <0 ð6:20Þ
where
 T    
A B A B 0 0
L¼ X þ
I 0 I 0 0 g2 I
By using Schur complement and Lemma 6.3, we can obtain
 T  
GðjwÞ GðjwÞ
P < 0; w1  w  w2 ð6:21Þ
I I
which is exactly the finite frequency H1 performance index inequality in (6.7).
Denote V ðtÞ ¼ xT ðtÞP1 xðtÞ as the energy function, whose derivative is
obtained as

V_ ðtÞ ¼ 2xT ðtÞP1 AxðtÞ þ 2xT ðtÞP1 BwðtÞ

 Active suspension control with finite frequency approach 151

Noting that

1 T
2xT ðtÞP1 BwðtÞ  xðtÞT P1 BB P1 xðtÞ þ hwðtÞT wðtÞ, 8h > 0
h

we have
 
_V ðtÞ  xðtÞT A T P1 þ P1 A þ 1 P1 BB T P1 xðtÞ þ hwðtÞT wðtÞ ð6:22Þ
h

According to the inequality in (6.16), inequality (6.22) guarantees

V_ ðtÞ  hwðtÞT wðtÞ ð6:23Þ

Integrating both sides of inequality (6.23) from 0 to t results in

ðt
V ðtÞ  V ð0Þ  h wT ðtÞwðtÞdt  hkwk22 ¼ hwmax
0

This shows that

xT ðtÞP1 xðtÞ  V ð0Þ þ hwmax ¼ r ð6:24Þ

Consider

max jfz2 ðtÞgi j2 ¼ max xT ðtÞfC2 gTi fC2 gi xðtÞ

t0 t0
1
Using the transformation xðtÞ ¼ P12 xðtÞ, from inequality (6.24) it follows that
x T ðtÞxðtÞ  r: Hence,
 1 
 1
max jfz2 ðtÞgi j2  r  lmax P1 2 fC2 gTi fC2 gi P1 2
t0

where lmax ðÞ represents the maximum eigenvalue. Then, the constraints in (6.8)
hold if
1 1
rP1 2 fC2 gTi fC2 gi P1 2 < I ð6:25Þ

which, by Schur complement, are equivalent to (6.13). The proof is completed. &
Since expressions like (6.11) and (6.12) involve coupling terms, the resulting
feasibility problem is nonlinear. Hence, it cannot be handled directly by LMI
optimization. In order to solve the nonlinear problem, define

J1 ¼ diagfF 1, F 1, F 1, Ig

J2 ¼ diagfF 1, F 1, I, Ig
J3 ¼ diagfI, F 1 g
152 Handbook of vehicle suspension control systems

Then, we perform a congruence transformation to (6.11)–(6.13), respectively, by

the full rank matrix J1T , J2T , and J3T on the left, and J1 , J2 , and J3 on the right.
Defining

Q ¼ ðF 1 ÞT QF 1, P ¼ ðF 1 ÞT PF 1

P 1 ¼ ðF 1 ÞT P1 F 1, K ¼ KF 1, F ¼ F 1

the following theorem is obtained.

Theorem 6.2. Let positive scalars g, h, and r be given. A state feedback controller
in the form of (6.9) exists, such that the closed-loop system in (6.10) is asympto-
tically stable with wðtÞ ¼ 0; and satisfies kGðjwÞkw11 <w<w2 < g for all nonzero
w 2 L2 ½0;1Þ, while the constraints in (6.8) are guaranteed with the disturbance
energy under the bound wmax ¼ ðr  V ð0ÞÞ=h, if there exist matrices P, P 1 > 0;
Q > 0 and general matrix F satisfying inequalities (6.26)–(6.28).
2   3
 F s AF þ BK þ P 1 F B1
6  P 1 0 0 7
6 7<0 ð6:26Þ
4   P 1 0 5
   hI
2 3
Q P þ jwc Q  F 0 0
6  F C1T þ K DT1 7
T T
6 w 1 w2 Q þ ½AF þ BK  s B 1 7<0 ð6:27Þ
4   g2 I 0 5
   I
 pffiffiffi 
I rfC2 gi F
 0; i ¼ 1; 2 ð6:28Þ
 P 1

Moreover, if inequalities (6.26)–(6.28) have a set of feasible solutions, the control

gain K in (6.9) is given by
1
K ¼ KF
Remark 6.1. Note that the linear matrix inequality (6.27) has complex variables.
According to Reference 24, the LMI in complex variables can be converted to an
LMI of larger dimension
 inreal variables. This means that inequality S1 þ jS2 < 0
S1 S2
is equivalent to <0
S2 S1

6.4 Dynamic output feedback controller design

In this section, we are interested in designing a dynamic output feedback controller,

such that the H1 norm of the closed-loop systems is minimized in the specific
frequency domain, while respecting the constraints in (6.8) within bounds.
 Active suspension control with finite frequency approach 153

For the active suspension system (6.6), a dynamic output feedback controller
KðsÞ is given as follows:
_
hðtÞ ¼ AK hðtÞ þ BK yðtÞ
ð6:29Þ
uðtÞ ¼ CK hðtÞ þ DK yðtÞ
Substituting (6.29) into (6.6), and defining
 
xðtÞ
xc ðtÞ ¼ ð6:30Þ
hðtÞ
the closed-loop system admits the realization
x_ c ðtÞ ¼ A xc ðtÞ þ B wðtÞ
z1 ðtÞ ¼ C 1 xc ðtÞ ð6:31Þ
z2 ðtÞ ¼ C 2 xc ðtÞ
where
" # " #
A þ BDK C BCK B1
A :¼ , B :¼
BK C AK 0 ð6:32Þ
C 1 :¼ ½ C1 0 , C 2 :¼ ½ C2 0
The transfer function of the closed-loop system from the disturbance input wðtÞ to
the controlled output z1 ðtÞ is defined as follows:

Gð jwÞ ¼ C 1 ðjwI  A Þ1 B

6.4.1 Finite frequency case

In this subsection, a dynamic output feedback controller is designed in the finite
frequency domain, so that the closed-loop system in (6.31) is asymptotically stable,
and satisfies

sup kGðjwÞk1 < g ð6:33Þ

w 1 <w<w 2

while respecting the constraints in (6.8) within bounds.

Theorem 6.3. For given positive scalars g, h, and r, the closed-loop system in (6.31)
is asymptotically stable, and satisfies supw1 <w<w2 kGðjwÞk1 < g, while respecting
the constraints in (6.8) with the disturbance energy under the bound
wmax ¼ ðr  V ð0ÞÞ=h, if there exist symmetric matrices P, Ps > 0; Q > 0 and
general matrix W satisfying inequalities (6.34)–(6.36).
2 3
½W s W T A þ Ps W T W T B
6  Ps 0 0 7
6 7<0 ð6:34Þ
4   Ps 0 5
   hI
154 Handbook of vehicle suspension control systems
2   3
w1 w2 Q  A T W s P  jwc Q þ W T W T B C T1
6
6  Q 0 0 7 7<0 ð6:35Þ
4   g2 I 0 5
   I
 pffiffiffi 
I rfC 2 gi
< 0; i ¼ 1; 2 ð6:36Þ
 Ps

Proof. Since the proof is similar to that in Theorem 6.1, here we omit it. &
Expressions in Theorem 6.3 are nonconvex due to the product terms of the
multiplier W , the controller parameters, and coefficient matrices. In order to solve
the nonconvex problem, we carry on the following transformation.
In accordance with the partition of A in (6.32), we introduce a partition of W
and its inverse W 1 in the form:
   
X Y 1 M G
W¼ , W ¼ ð6:37Þ
U V H L
From the literature [25], there is no loss of generality in assuming that U and H are
invertible. Define
   
X I I M
D1 ¼ , D2 ¼
U 0 0 H
and note that
 
X XM þ YH
W D2 ¼ ¼ D1 ð6:38Þ
U UM þ VH
Define

J 1 ¼ diag DT2 , DT2 , DT2 , I

J 2 ¼ diag DT2 , DT2 , I, I

J 3 ¼ diag I, DT2
Pre- and post-multiplying (6.34)–(6.36) by J 1 , J 2 , and J 3 and their transposes,
respectively, and defining
Q ¼ DT2 QD2 ,
P ¼ DT2 PD2 , P s ¼ DT2 Ps D2
 
T T XTA þ B ^KC ^K
A
A ¼ D2 W A D2 ¼ ^ K C AM þ BC ^K
A þ BD
 T 
X B1
B ¼ DT2 W T B
B1
C 1 ¼ C 1 D2 ¼ ½ C1 C1 M 
C 2 ¼ C 2 D2 ¼ ½ C2 C2 M 
 T 
X Z
W ¼ DT2 W D2 ¼
I M
 Active suspension control with finite frequency approach 155

with the following linearizing changes of variables:

^ K ¼ X T AM þ X T BDK CM þ U T BK CM þ X T BCK H þ U T AK H
A
^ K ¼ X T BDK þ U T BK
B
ð6:39Þ
^ K ¼ CK H þ DK CM
C
^ K ¼ DK
D
Z ¼ X TM þ UTH ð6:40Þ
we can give the following theorem.
Theorem 6.4. Give positive scalars g, h, and r: A dynamic output feedback con-
troller in the form of (6.29) exists, such that the closed-loop system in (6.31)
is asymptotically stable, and satisfies supw1 <w<w2 kGðjwÞk1 < g, while respecting
the constraints in (6.8) with the disturbance energy under the bound
wmax ¼ ðr  V ð0ÞÞ=h, if there exist symmetric matrices P, P s > 0; Q > 0, and
general matrices W , A ^k, B ^ k, D
^k, C ^ k , M, X, and Z satisfying
2  T 3
W s A þ P s W B
6  P s 0 0 7
6 7<0 ð6:41Þ
4   P s 0 5
   hI
2   T T 3
w1 w2 Q  A s P  jw c Q þ W B C 1
6  Q 0 0 7
6 7<0 ð6:42Þ
4   2
g I 0 5
   I
 pffiffiffi  
I r C2 i
<0 ð6:43Þ
 P s
^k,
Moreover, if the above inequalities have a feasible solution, then the matrices A
B ^ k, D
^k, C ^ k , M, X, and Z can be obtained. According to (6.39) and (6.40), we will
compute the controller by

DK ¼ D^K
^ K  DK CMÞH 1
C K ¼ ðC
ð6:44Þ
^ K  X T BDK Þ
BK ¼ U T ðB
^ K  X T AM  X T BDK CM  U T BK CM  X T BCK HH 1
AK ¼ U T ½A
Remark 6.2. When we calculate the controller, the matrices U and H, which cannot
be directly obtained by the Theorem 6.4, are needed and they should be chosen
such that

UTH ¼ Z  XTM

It is worth mentioning that the factorization of U T H can always be achieved so that

the invertible matrices U and H are deduced. In this chapter, the two invertible
matrices are obtained by using the singular value decomposition approach. &
156 Handbook of vehicle suspension control systems

6.4.2 Entire frequency case

In order to highlight the advantages of controller designed in the finite frequency
domain, we develop another dynamic output feedback controller in the entire fre-
quency domain, based on the method proposed in the literature [26]. In this sub-
section, a dynamic output feedback controller is designed in the entire frequency
domain, so that the closed-loop system in (6.31) is asymptotically stable, and under
zero initial condition, the closed-loop system guarantees that kz1 k2 < gkwk2 for all
nonzero w 2 L2 ½0;1Þ: We give the conclusion of designing entire frequency con-
troller in the following corollary.
Corollary 6.1. Let positive scalars r and g be given. If there exists symmetric
matrix Pc > 0 satisfying
2 3
½Pc A s Pc B C T1
4  g2 I 0 5<0 ð6:45Þ
  I
 pffiffiffi 
I rfC 2 gi
<0 ð6:46Þ
 Pc

then a stabilizing dynamic output feedback controller in the form of (6.29) exists,
such that
1. the closed-loop system in (6.31) is asymptotically stable;
2. under zero initial condition, the closed-loop system guarantees that
kz1 k2 < gkwk2 for all nonzero w 2 L2 ½0;1Þ;
3. the constraints in (6.8) are guaranteed with the disturbance energy under the
bound wmax ¼ ðr  V ð0ÞÞ=h:
Proof. Since the results can easily be obtained, the proof has been omitted here. &
Hereafter, we will show how to transform (6.45) and (6.46) into the forms that
can be solved directly. Partition the matrix Pc and its inverse P1
c in the form
   
Yc Nc X c Mc
Pc ¼ , P1c ¼ MT ð6:47Þ
NcT # c #

where ‘‘#’’ represents this position can be arbitrary. From the literature [26], there
is no loss of generality in assuming that Nc and Mc are invertible. Define
   
Xc I I Yc
Dc1 ¼ , D c2 ¼
McT 0 0 NcT

and note that

Pc Dc1 ¼ Dc2 ð6:48Þ

Define

Jc1 ¼ diagfDc1 , I, Ig, Jc2 ¼ diagfI, Dc1 g

 Active suspension control with finite frequency approach 157

Pre- and post-multiplying (6.45) and (6.46) by Jc1 and Jc2 and their transposes,
respectively, and defining
 
AXc þ BC ^ c A þ BD ^ cC
A c ¼ DTc1 Pc A D2 ¼ ^c ^ cC
A Yc A þ B
 
B1
B c ¼ DTc1 Pc B ¼
Yc B1

C c1 ¼ C 1 Dc1 ¼ ½ C1 Xc C1 
C c2 ¼ C 2 Dc1 ¼ ½ C2 Xc C2 
 
Xc I
P c ¼ DTc1 Pc Dc1 ¼
I Yc
with the following linearizing changes of variables:
^ c ¼ Yc AXc þ Yc BD
A ^ c CXc þ Nc BK CXc þ Yc BCK M T þ Nc AK M T ð6:49Þ
c c
^ ^
B c ¼ Yc BD c þ Nc BK
^ c ¼ CK M T þ D
C ^ c CXc
c
^ c ¼ DK
D ð6:50Þ
we can give the following corollary:
Corollary 6.2. Let scalar g > 0 be given. If there exist matrices Yc > 0; Xc > 0,
^ c, B
and general matrices A ^ c , and D
^ c, C ^ c satisfying
2  T 3
Ac B c C c1
4  s g2 I 0 5<0 ð6:51Þ
  I
 pffiffiffi  
I r C c2 i
<0 ð6:52Þ
 P c
then a stabilizing dynamic output feedback controller in the form of (6.29) exists,
such that
1. the closed-loop system in (6.31) is asymptotically stable;
2. under zero initial condition, the closed-loop system guarantees that kz1 k2
< gkwk2 for all nonzero w 2 L2 ½0;1Þ;
3. the constraints in (6.8) are guaranteed with the disturbance energy under the
bound wmax ¼ ðr  V ð0ÞÞ=h:
Moreover, if inequalities (6.51) and (6.52) have a feasible solution, then we
will compute the controller by
DK ¼ D^c
^c  D
C K ¼ ðC ^ c CXc ÞM T
c
1 ^ ^ cÞ ð6:53Þ
BK ¼ Nc ðB c  Yc BD
^ c  Yc AXc  Yc BD
AK ¼ Nc1 ½A ^ c CXc  Nc BK CXc  Yc BCK M T M T
c c
158 Handbook of vehicle suspension control systems

Remark 6.3. As the same lines with Remark 6.2, the matrices Nc and Mc , which
cannot be directly obtained by the Corollary 6.2, should satisfy

Nc McT ¼ I  Yc Xc
Here, we also obtain the two invertible matrices by using the singular value
decomposition approach. &

6.5 Simulation
In this section, we will apply the above approach to design a finite frequency H1
controller based on the quarter-car model described in Section 6.2. The quarter-car
model parameters are listed in Table 6.2.

6.5.1 State feedback case

For subsequent comparison, a state feedback H1 controller in the finite frequency
domain for the system in (6.6) is designed first, based on the assumption that all the
state variables can be measured. Under zero initial conditions, by solving the matrix
inequalities (6.26)–(6.28) for matrices P, P 1 > 0, and Q > 0 with the optimized
parameter g > 0 and w1 ¼ 4 Hz (8p rad/s), w2 ¼ 8 Hz (16p rad/s), r ¼ 5; h ¼ 1000;
zmax ¼ 100 mm, the minimum guaranteed closed-loop H1 performance obtained is
gmin ¼ 4:3914
1
Then, an admissible control gain matrix is given based on K ¼ K F :

K¼ 105  ½1:1169 0:7377 0:4488 0:0088 

For description in brevity, we denote this finite frequency controller as Controller I
hereafter.
Then, we give another H1 state feedback controller, which is designed over
the entire frequency range, and the minimum guaranteed closed-loop H1 perfor-
mance obtained is
gmin ¼ 8:7929
Then, an admissible control gain matrix is given as

K¼ 104  ½ 1:0567 6:2035 0:2689 0:1081 

and we denote this controller as Controller II for brevity.

Table 6.2 Quarter-car model parameters

ms mu ks kt cs ct
320 kg 40 kg 18 kN/m 200 kN/m 1 kNs/m 10 Ns/m
 Active suspension control with finite frequency approach 159

After obtaining the finite frequency controller (Controller I) and the entire
frequency controller (Controller II), we will compare the two controllers to illus-
trate the performance of the closed-loop suspension system in finite frequency
domain. By the simulation, the responses of the open-loop system, the closed-loop
system, which is composed of Controller I, and the closed-loop system, which is
composed of Controller II, are compared in Fig. 6.2. In this figure, the solid and
dotted lines are the responses of the closed-loop system with Controller I and
Controller II, respectively, and the dashed line is the response of the passive sys-
tem. From the figure, we can see that the finite frequency controller yields the least
value of H1 norm over the frequency range 48 Hz, compared with the passive
system and the closed-loop system with the entire frequency controller, which
clearly shows that an improved ride comfort has been achieved.
Evaluation of the vehicle suspension performance is based on the examina-
tion of three response quantities, that is, the body acceleration of the specific
frequency domain, the suspension deflection between the wheel and the car body,
and the tire deflection. In order to evaluate the suspension characteristics with
respect to three performance requirements, we give the disturbance signal as
follows to clarify the effectiveness of our finite frequency controller. Consider the
case of an isolated bump in an otherwise smooth road surface; the disturbance
inputs are given by

Asinð2pftÞ, if 0  t  T
wðtÞ ¼ ð6:54Þ
0; if t > T

25

20
Maximum singular values

15

10

0
10–1 100 101 102
Frequency (Hz)

Fig. 6.2 The curves of maximum singular values (dashed line: open-loop system;
dotted line: system with controller II; and solid line: system with
controller I)
160 Handbook of vehicle suspension control systems

where A, f, and T represent the amplitude, frequency, and period of vibration,

respectively. Assume A ¼ 0:5 m, f ¼ 5 Hz (among the frequency band 48 Hz),
and T ¼ 1=f ¼ 0:2 s, which implies the disturbance energy is 0:025 m2s. The
time-domain response of body vertical acceleration for the active suspension sys-
tem is shown in Fig. 6.3, where the solid line and the dashed line are the responses
of body vertical acceleration with the finite frequency controller and the entire
frequency controller, respectively. We can clearly see that the value of the body
acceleration with the finite frequency controller is less than that with the entire
frequency controller. In addition, Figs. 6.4 and 6.5 show that the ratio x1 ðtÞ=zmax
and the relation dynamic tire load kt x2 ðtÞ=ðms þ mu Þg are below 1, which means
the time-domain constraints are guaranteed by the designed controller.
From Theorem 6.2, the energy of the disturbance signal is bounded by
0:005 m2 s (computed by r=h, under zero initial conditions). As mentioned pre-
viously, the disturbance energy in our simulation is 0:025 m2 s, much larger than
the theoretically allowable; nevertheless, the hard constraints are respected on the
whole. This reveals the conservatism of the design.
Random vibrations are consistent and typically specified as random process
with a given ground displacement power spectral density (PSD) of
 W
n
Gq ðnÞ ¼ Gq ðn0 Þ ð6:55Þ
n0
where n is the spatial frequency and n0 is the reference spatial frequency of
n0 ¼ 0:1(1/m); Gq ðn0 Þ stands for the road roughness coefficient; W ¼2 is the road
roughness constant. Related to the time frequency f , we have f ¼nV with V for the

Body acceleration (m/s2)

3
Entire frequency
2 Finite frequency

–1

–2

–3

–4
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)

Fig. 6.3 The time-domain response of body acceleration

 Active suspension control with finite frequency approach 161

The ratio of suspension deflection and the maximum limitation

0.15

0.1

0.05

–0.05

–0.1

–0.15

–0.2

–0.25

–0.3

–0.35
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)

Fig. 6.4 The ratio of suspension deflection and the maximum limitation
(( zs  zu )=zmax )

Relation dynamic tire load

0.4

0.3

0.2

0.1

–0.1

–0.2

–0.3

–0.4
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)

Fig. 6.5 Relation dynamic tire load (kt ( zs  zu )=( ms þ mu ))

162 Handbook of vehicle suspension control systems

vehicle forward velocity. According to (6.55), we can obtain the PSD ground dis-
placement:
V
Gq ðf Þ ¼ Gq ðn0 Þn20 ð6:56Þ
f2

Correspondingly, PSD ground velocity is given by

_ Þ ¼ ð2pf Þ2 Gq ðf Þ ¼ 4pGq ðn0 Þn20 V

Gqðf ð6:57Þ

which is only related with the vehicle forward velocity. When the vehicle forward
velocity is fixed, the ground velocity can be viewed as a white-noise signal. Select
the road roughness as Gq ðn0 Þ ¼ 256  106 m3 , which is corresponded to D grade
(poor) according to ISO2361 standards, to generate the random road profile. Set the
vehicle forward velocity as V þ 45 km/h, and as expected, it is observed from
Fig. 6.6 that the closed-loop system with finite frequency controller realizes a better
ride comfort, compared with system for the frequency range 4–8 Hz (since the
closed-loop system with Controller I has lower PSD body acceleration than that

0.7
Open-loop system
Entire frequency
Finite frequency
0.6

0.5
PSD body acceleration (m2/s3)

0.4

0.3

0.2

0.1

0 –2 1
10 10 100 101 102
Frequency (Hz)

Fig. 6.6 The power spectral density of body acceleration (dashed line: open-loop
system; dotted line: system with controller II; solid line: system with
controller I)
 Active suspension control with finite frequency approach 163

with Controller II, and smaller PSD body acceleration value results in better ride
comfort), where PSD body acceleration can be calculated by

Gz1 ðf Þ ¼ jGðjwÞj2 Gqðf

_ Þ ð6:58Þ
To check more random road profiles, we select the road roughness as
Gq ðn0 Þ ¼ 16106 m3 (B grade, good), Gq ðn0 Þ ¼ 64106 m3 (C grade, average),
and Gq ðn0 Þ ¼ 1024106 m3 (E grade, very poor), respectively. From Figs. 6.7–6.9,
it can be observed that the closed-loop system with Controller I realizes a better ride
comfort in spite of the different road roughness.

6.5.2 Dynamic output feedback case

Here, we will give the simulation of the closed-loop system with a dynamic output
feedback H1 controller in the finite frequency domain. The process is similar with
the state feedback case. First, the closed-loop system with a dynamic output feed-
back H1 controller in the finite frequency domain can be obtained, based on the
proposed method in Section 6.4, and we denote this closed-loop system as system
S1 for brevity. After setting C ¼ ½ I33 031  (we assume that there are three
states measured) and solving the matrix inequalities (6.41)–(6.43) for symmetric
matrices P, P s > 0; Q > 0 and general matrices W , A ^k, B
^k, C^ k, D
^ k , M, X, and Z

PSD body acceleration (m2/s3)

0.04
Open-loop system
Entire frequency
Finite frequency
0.035

0.03

0.025

0.02

0.015

0.01

0.005

0
10 2 10 1 100 101 102
Frequency (Hz)

Fig. 6.7 The power spectral density of body acceleration (B grade)

164 Handbook of vehicle suspension control systems

PSD body acceleration (m2/s3)

0.16
Open–loop system
Entire frequency
0.14 Finite frequency

0.12

0.1

0.08

0.06

0.04

0.02

0
10–2 10–1 100 101 102
Frequency (Hz)

Fig. 6.8 The power spectral density of body acceleration (C grade)

PSD body acceleration (m2s3)

2.5

Open-loop system
Entire frequency
Finite frequency
2

1.5

0.5

0
10–2 10–1 100 101 102
Frequency (Hz)

Fig. 6.9 The power spectral density of body acceleration (E grade)

 Active suspension control with finite frequency approach 165

with given scalars g > 0 and w1 ¼ 4 Hz, w2 ¼ 8 Hz, h ¼ 10,000; the optimal
guaranteed closed-loop H1 performance obtained is
gmin ¼ 2:8648
For subsequent comparison, we can get another closed-loop system with a
dynamic output feedback H1 controller over the entire frequency range, according
to the Corollary 6.2, and set it as system S2 . After solving the matrix inequalities in
Corollary 6.2, we obtain the optimal guaranteed closed-loop H1 performance:
gmin ¼ 6:0245
Fig. 6.10 shows the curves of maximum singular values, where the open-loop
system (passive mode), the closed-loop systems S1 (active finite frequency mode),
and the closed-loop system S2 (active entire frequency mode) are compared.
In Fig. 6.10, the dash–dot/dot/solid line represents the curve of maximum singular
values in the open-loop system/system S2 /system S1 , respectively. From this figure,
we can see that the closed-loop system with finite frequency controller has the least
value of H1 norm over the frequency range 48 Hz, compared with the passive
system and the closed-loop system with an entire frequency controller, which means
an improved ride comfort has been achieved by the finite frequency controller.

25
Passive
Entire frequency
Finite frequency

20

4 Hz 8 Hz
Maximum singular values

15

10

0
10–2 10–1 100 101 102
Frequency (Hz)

Fig. 6.10 The curves of maximum singular values (dashed line: open-loop system;
dotted line: system S2 ; and solid line: system S1 )
166 Handbook of vehicle suspension control systems

3
Passive
Entire frequency
2.5 Finite frequency

1.5
Body acceleration (m/s2)

0.5

–0.5

–1

–1.5
0 0.5 1 1.5 2 2.5 3
Time (s)

Fig. 6.11 The time-domain response of body acceleration

Choose the signal in (6.54) as the disturbance input, and assume A ¼ 0:5 m,
f ¼ 5 Hz, and the time-domain responses of body vertical acceleration are given in
Fig. 6.11 where the solid/dotted line is the response of body vertical acceleration
with the finite/entire frequency controller, and the dashed–dotted line respects the
response of the passive system. It is seen from this figure that the magnitude for the
body acceleration is significantly decreased. In particular, reduced accelerations
indicate that the ride comfort is improved, and the result confirms the efficiency of
the finite frequency controller. Figs. 6.12 and 6.13 show the time-domain con-
straints of the active suspension systems, and we can see that the time-domain
constraints are guaranteed by the designed controller, clearly.
Like the simulation in state feedback case, Figs. 6.14–6.17 tell us that our
designed controller is also effective under the random input signals.

6.6 Concluding remarks

In this chapter, the state feedback and dynamic output feedback H1 controllers for
active suspension systems with frequency band constraints have been designed in a
unified framework to improve ride comfort. The key idea of designing the proposed
controllers is to use the generalized Kalman–Yakubovich–Popov (GKYP) lemma.
At the same time, the time-domain constraints have also been guaranteed in the
controller design. Simulation results show that the finite frequency controllers
 Active suspension control with finite frequency approach 167

0.05

The ratio of suspension deflection and the maximum limitation

0

–0.05

–0.1

–0.15

–0.2

–0.25

–0.3

–0.35

–0.4

–0.45
0 0.5 1 1.5 2 2.5 3
Time (s)

Fig. 6.12 The ratio of suspension deflection and the maximum limitation

0.5

0.4

0.3
Relation dynamic tire load

0.2

0.1

–0.1

–0.2

–0.3

–0.4
0 0.5 1 1.5 2 2.5 3
Time (s)

Fig. 6.13 The relation dynamic tire load

168 Handbook of vehicle suspension control systems

0.08
Power spectral density Entire frequency
0.06 Finite frequency

0.04

0.02

0
10–2 10–1 100 101 102

0.7
Power spectral density

0.6
0.5
0.4
0.3
0.2
0.1
0
10–2 10–1 100 101 102
Frequency (Hz)

Fig. 6.14 The power spectral density of body acceleration (system S1 : solid line;
system S2 : dashed line)

Gq(n0) = 16 × 10–6m3
× 10–3
5
Entire frequency
4.5 Finite frequency

3.5
Power spectral density

2.5

1.5

0.5

0
10–2 10–1 100 101 102
Frequency (Hz)

Fig. 6.15 The power spectral density of body acceleration (B grade)

 Active suspension control with finite frequency approach 169

Gq(n0) = 64 × 10–6m3
0.02
Entire frequency
Finite frequency
0.018

0.016

0.014
Power spectral density

0.012

0.01

0.008

0.006

0.004

0.002

0
10–2 10–1 100 101 102

Fig. 6.16 The power spectral density of body acceleration (C grade)

Gq(n0) = 1024 × 10–6m3

0.35
Entire frequency
Finite frequency

0.3

0.25
Power spectral density

0.2

0.15

0.1

0.05

0
10–2 10–1 100 101 102

Fig. 6.17 The power spectral density of body acceleration (E grade)

170 Handbook of vehicle suspension control systems

achieve better disturbance attenuation performance over the concerned frequency

range than those designed in the entire frequency.

References
[1] D. Hrovat. Survey of advanced suspension developments and related optimal
control applications. Automatica, 33(10):1781–1817, 1997.
[2] H.-J. Kim, H. S. Yang and Y.-P. Park. Improving the vehicle performance
with active suspension using road-sensing algorithm. Computers and Struc-
tures, 80:1569–1577, 2002.
[3] J. Wang and D. A. Wilson. Mixed GL2 /H2 /GH2 control with pole
placement and its application to vehicle suspension systems. International
Journal of Control, 74(13):1353–1369, 2001.
[4] J. Marzbanrad, G. Ahmadi, H. Zohoor and Y. Hojjat. Stochastic optimal
previewcontrol of a vehicle suspension. Journal of Sound and Vibration,
275:973–990, 2004.
[5] P. S. Els, N. J. Theron, P. E. Uys and M. J. Thoresson. The ride comfort vs.
handling compromise for off-road vehicles. Journal of Terramechanics,
44:303–317, 2007.
[6] R. A. Williams. Automotive active suspensions. Proceedings of the Institu-
tion of Mechanical Engineers Part D: Journal of Automobile Engineering,
211:415–444, 1997.
[7] S. Türkay and H. Akcay. Aspects of achievable performance for quarter-car
active suspensions. Journal of Sound and Vibration, 311:440–460, 2008.
[8] S.-J. Huang and H.-Y. Chen. Adaptive sliding controller with self-tuning
fuzzy compensation for vehicle suspension control. Mechatronics,
16:607–622, 2006.
[9] T. Yoshimura, A. Kume, M. Kurimoto and J. Hino. Construction of an active
suspension system of a quarter car model using the concept of sliding mode
control. Journal of Sound and Vibration, 239(2):187–199, 2001.
[10] Y. P. He and J. McPhee. Multidisciplinary design optimization of mecha-
tronic vehicles with active suspensions. Journal of Sound and Vibration,
283:217–241, 2005.
[11] C. Kaddissi, J. P. Kenné and M. Saad. Drive by wire control of an electro-
hydraulic active suspension a backstepping approach. Proceedings of the
2005 IEEE Conference on Control Applications, Toronto, Canada, August
28–31, 2005.
[12] D. Hrovat. A class of active LQG optimal actuators. Automatica, 18(1):
117–119, 1982.
[13] H. Du, J. Lam and K. Y. Sze. Design of non-fragile H1 controller for active
vehicle suspensions. Journal of Vibration and Control, 11:225–243, 2005.
[14] I. Fialho and G. J. Balas. Road adaptive active suspension design using
linear parameter-varying gain-scheduling. IEEE Transactions on Control
Systems Technology, 10(1):43–54, 2002.
 Active suspension control with finite frequency approach 171

[15] M. Yamashita, K. Fujimori, K. Hayakawa and H. Kimura. Application of

H1 control to active suspension systems. Automatica, 30(11):1717–1729,
1994.
[16] N. Al-Holou, T. Lahdhiri, D. S. Joo, J. Weaver and F. Al-Abbas. Sliding mode
neural network inference fuzzy logic control for active suspension systems.
IEEE Transactions on Fuzzy Systems, 10(2):234–246, 2002.
[17] N. Yagiz and Y. Hacioglu. Backstepping control of a vehicle with active
suspensions. Control Engineering Practice, 16:1457–1467, 2008.
[18] B. Hencey and A. G. Alleyne. A KYP Lemma for LMI Regions. IEEE
Transactions on Automatic Control, 52(10):1926–1930, 2007.
[19] H. Khatibi, A. Karimi and R. Longchamp. Fixed-order controller design
for polytopic systems using LMIs. IEEE Transactions on Automatic Control,
53(1):428–434, 2008.
[20] J. Collado, R. Lozano and R. Johansson. On Kalman–Yakubovich–Popov
Lemma for stabilizable systems. IEEE Transactions on Automatic Control,
46(7):1089–1093, 2001.
[21] T. Iwasaki and S. Hara. Generalized KYP Lemma: unified frequency domain
inequalities with design applications. IEEE Transactions on Automatic
Control, 50(1):41–59, 2005.
[22] T. Iwasaki and S. Hara. Feedback control synthesis of multiple frequency
domain specifications via generalized KYP lemma. International Journal of
Robust Nonlinear Control, 17:415–434, 2007.
[23] T. Iwasaki and S. Hara. Dynamic output feedback synthesis with general
frequency domain specications. Proceedings of the 16th IFAC World
Congress, Elsevier, Prague, Czechoslovakia, 2005.
[24] P. Gahinet, A. Nemirovskii, A. J. Laub and M. Chilali. LMI control toolbox
user’s Guide. The math. Works Inc., Natick, MA, 1995.
[25] P. Apkarian, H. D. Tuan and J. Bernussou. Continuous-time analysis,
eigenstructure assignment, and H2 synthesis with enhanced linear matrix
inequalities (LMI) characterizations. IEEE Transactions on Automatic
Control, 42(12):1941–1946, 2001.
[26] C. Scherer, P. Gahinet and M. Chilali. Multi-objective output-feedback
control via LMI optimization. IEEE Transactions on Automatic Control,
42(7):896–911, 1997.
 Chapter 7
Fault-tolerant control for uncertain
vehicle suspension systems via
fuzzy control approach1
Hongyi Lia and Honghai Liub

Abstract
This chapter investigates the problem of fault-tolerant fuzzy controller design for
active vehicle suspension systems. We use Takagi–Sugeno (T–S) fuzzy model
approach to study the suspension systems with the sprung and unsprung masses’
variation, the actuator fault, and other suspension performances. A novel fault-tolerant
fuzzy controller is designed such that the resulting T–S fuzzy system is asymptotically
stable and has a prescribed H1 performance under given constraints. Finally, some
simulation results based on a quarter-vehicle suspension model are provided to
demonstrate the effectiveness of the approaches of the proposed design.

7.1 Introduction

Recently, a vehicle system has played a significant role in evaluating the vehicle
dynamics performance since a suspension component can support the vehicle
weight, provide effective isolation of the chassis from road excitations, keep tire
contact with the ground, and maintain the wheels in appropriate position on the
road surface. The main roles of a vehicle suspension system are to adequately
guarantee the stability of the vehicle and to provide as much comfort as possible for
the passengers by serving the basic function of isolating passengers from road-
induced vibration and shocks [1, 2]. Considerable attention has been paid and
efforts have been made to the challenging issue of how to optimize the required

1
This work was partially supported by a joint PhD scholarship of University of Portsmouth and Protean
Electric Ltd, the National Science Foundation of P. R. China (61203002), the Program for New Century
Excellent Talents in University, and the Program for Liaoning Excellent Talents in University (LR2013053).
a
College of Engineering, Bohai University, Jinzhou 121013, Liaoning, China
b
Intelligent Systems & Biomedical Robotics Group, School of Creative Technologies, University of
Portsmouth, Portsmouth PO5 4BP, UK
174 Handbook of vehicle suspension control systems

suspension performances, namely, ride comfort, road handling, and suspension

deflection. It is evident that many vehicle models and controller design methods
have been reported in References [3–9].
However, all the aforementioned suspension control results are under a full-
reliability assumption that all control components of the systems are in ideal working
conditions. With the growing complexity of automated control systems, various faults
are likely to be encountered, especially faults from actuators and sensors. Recently,
fault-tolerant control problems have received considerable attention for dynamic
systems with uncertainty such as actuator and sensor faults. Additionally, a great
number of theoretic results have also been presented. In Reference [10], the authors
show that the conventional controllers cannot achieve better performance in com-
parison with the fault-tolerant controllers. Therefore, it is very interesting to design a
fault-tolerant controller such that the system stability and performance of the active
suspension closed-loop system can be tolerated with sensor or actuator faults.
It is well known that the active suspension system has the capability to enhance
vehicle dynamics by relaxing external impact such as the impact of road surface on
vehicle travel comfort. In terms of its control design, uncertainty of vehicle sprung
and unsprung masses such as its loading conditions should be taken into account to
meet vehicle travel performance criteria. For instance, the polytopic parameter
uncertainties was employed to model the varying vehicle sprung or unsprung
masses [11–13]. The parameter-dependent controllers were proposed for the
quarter-car suspension systems with sprung mass variation [11]. The parameter-
independent sampled-data H1 controller design strategy was presented to handle
both sprung and unsprung mass variations in a case study of a quarter-car suspen-
sion system [13].
The state of the art in suspension control design in these scenarios, however,
could not provide feasible performance for uncertain active suspension systems
with actuator fault. Clearly, there is a need for a new controller design method that
has the capability of satisfying the control condition. On the other hand, since fuzzy
sets were proposed by Zadeh [14], fuzzy logic control has developed into a
conspicuous and successful branch of automation and control theory. The Takagi–
Sugeno (T–S) fuzzy model has been proved as an effective theoretical method and
practical tool for representing complex nonlinear systems and applications [15–18].
This chapter considers fault-tolerant fuzzy control for uncertain active
suspension systems with actuator fault based on the T–S fuzzy model approach. By
using the Lyapunov stability theory, a fault-tolerant fuzzy H1 controller is
designed to guarantee the system’s asymptotic stability and H1 performance,
simultaneously satisfying the constraint performances. Finally, simulation results
show the effectivenes of the proposed methods.
Notation: The superscript T denotes matrix transposition. Rn denotes the
n-dimensional Euclidean space. kk1 denotes the H1 norm for matrices. The
notation P > 0 ð 0Þ is used to denote a symmetric and positive definite (semi-
definite) matrix. In symmetric block matrices or complex matrix expressions, an
asterisk (*) is employed to represent a term that is readily induced by symmetry and
 Fault-tolerant control for uncertain vehicle suspension systems 175

diagf. . .g stands for a block-diagonal matrix. symðAÞ is used to denote A þ AT for

the sake of simplicity. Matrices, if the dimensions are not explicitly stated, are assumed
to be compatible for algebraic operations. The space of square-integrable vector
functions over ½0; 1Þ is denoted by L2 ½0; 1Þ, and for w ¼ fwðtÞg 2 L2 ½0; 1Þ, w
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Ð1 2
norm is denoted by kwk2 ¼ t¼0 jwðt Þj dt .

7.2 Problem formulation

In this chapter, we consider a quarter-vehicle model for designing an active sus-

pension controller. In Figure 7.1, ms is used to denote the sprung mass, which
represents the car chassis; mu is the unsprung mass, which represents mass of the
wheel assembly; uðtÞ stands for the active input of the suspension system; zs and zu
denote the displacements of the sprung and unsprung masses, respectively; zr is
used to denote the road displacement input; cs and ks are damping and stiffness of
the suspension system, respectively; kt and ct stand for compressibility and
damping of the pneumatic tire, respectively. Then, we have the dynamic equation
of the suspension model:

mu€z u ðtÞ þ cs ½_z u ðtÞ  z_ s ðtÞ þ ks ½zu ðtÞ  zs ðtÞ þ kt ½zu ðtÞ  zr ðtÞ
þ ct ½_z u ðtÞ  z_ r ðtÞ ¼ uðtÞ ð7:1Þ
ms€z s ðtÞ þ cs ½_z s ðtÞ  z_ u ðtÞ þ ks ½zs ðtÞ  zu ðtÞ ¼ uðtÞ

Denote x1 ðtÞ ¼ zs ðtÞ  zu ðtÞ as the suspension deflection, x2 ðtÞ ¼ zu ðtÞ  zr ðtÞ as
the tire deflection, x3 ðtÞ ¼ z_ s ðtÞ as the sprung mass speed, x4 ðtÞ ¼ z_ u ðtÞ as the
unsprung mass speed, and wðtÞ ¼ z_ r ðtÞ as the disturbance input. Then, we can
rewrite the systems in (7.1) as

x_ ðtÞ ¼ AðtÞxðtÞ þ B1 ðtÞwðtÞ þ BðtÞuðtÞ ð7:2Þ

zs
Vehicle ms
u(t)
Suspension ks cs Controller
zu
Wheel mu

Tire kt ct zr

Figure 7.1 A quarter-car model

176 Handbook of vehicle suspension control systems

where
2 3
0 0 1 1
6 7
6 0 0 0 1 7
6 7
6 ks cs cs 7
A ðt Þ ¼ 6  0  7
6 ms m s m s 7
6 7
4 ks kt cs cs þ ct 5
 
mu mu mu mu
2 3 2 3
0 0
6 7 6 7
6 0 7 6 1 7
6 7 6 7
6 1 7 6 7
B ðt Þ ¼ 6 7 , B 1 ðt Þ ¼ 6 7
6 ms 7 6 0 7
6 7 6 7
4 1 5 4 ct 5

mu mu
xðtÞ ¼ ½ x1 ðtÞ x2 ðtÞ x3 ðtÞ x4 ðtÞ T

Performances of suspension systems, such as ride comfort and suspension

deflection, road holding are the fundamentals being taken into account. It is widely
accepted that ride comfort can be generally quantified by the body acceleration in
the vertical direction in the context of a quarter-vehicle model; hence, it is prac-
tical to choose body acceleration, €z s ðtÞ, as the first control output. It indicates that
one of the objectives is to minimize the vertical acceleration, €z s ðtÞ, to secure
vehicle travel comfort.
Recall H1 control method, the value of H1 norm is defined as an upper bound
of the root mean square gain, the main objective is to minimize the H1 norm of the
transfer function from the disturbance wðtÞ to the control output z1 ðtÞ ¼ €z s ðtÞ with
an emphasis on improving the ride comfort. Meanwhile, the following required
performances have to be taken into account as well:

(I) The suspension deflection cannot exceed a maximum value constrained by

mechanical structure, that is,

jzs ðtÞ  zu ðtÞj  zmax ð7:3Þ

where zmax is the maximum suspension deflection.

(II) The dynamic tire load has to be less than the static tire load in order to ensure
a firm, uninterrupted contact of the wheels on the road:

kt ðzu ðtÞ  zr ðtÞÞ < ðms þ mu Þg ð7:4Þ

Based on the above criteria, the body acceleration €z s ðtÞ is chosen as the per-
formance control output, and the suspension stroke zs ðtÞ  zu ðtÞ and relative
dynamic tire load kt ðzu ðtÞ  zr ðtÞÞ=ðms þ mu Þg are chosen as the second control
 Fault-tolerant control for uncertain vehicle suspension systems 177

output z2 ðtÞ. Therefore, the following system is derived to present the active vehicle
suspension system:

x_ ðtÞ ¼ AðtÞxðtÞ þ B1 ðtÞwðtÞ þ BðtÞuðtÞ

z1 ðtÞ ¼ C1 ðtÞxðtÞ þ D1 ðtÞuðtÞ ð7:5Þ
z2 ðtÞ ¼ C2 ðtÞxðtÞ

where AðtÞ, B1 ðtÞ, and BðtÞ are defined in (7.2), and

 
ks cs cs 1
C1 ðtÞ ¼  0  , D1 ðtÞ ¼
ms ms ms ms
2 1 3
0 0 0 ð7:6Þ
6 zmax 7
C2 ðtÞ ¼ 64
7
5
kt
0 0 0
ðms þ mu Þg

Note that the suspension system in (7.5) is a model with uncertainty in that the
sprung mass ms and the unsprung mass mu vary in the given ranges, in which ms
and mu denote ms ðtÞ and mu ðtÞ, respectively. In the meantime, the actuator fault
should be taken into account since the suspension performance could be affected by
these factors. It leads to the system as

x_ ðtÞ ¼ AðtÞxðtÞ þ B1 ðtÞwðtÞ þ BðtÞuf ðtÞ

z1 ðtÞ ¼ C1 ðtÞxðtÞ þ D1 ðtÞuf ðtÞ ð7:7Þ
z2 ðtÞ ¼ C2 ðtÞxðtÞ

Considering the fault channel from controller to actuator,

uf ðtÞ ¼ ma uðtÞ ð7:8Þ

ma is used to represent the possible fault of the corresponding actuator uf ðtÞ.

m
 a  ma  m^ a , where m
 a and m
^ a are constant scalars and used to constrain lower
and upper bounds of the actuator faults. Three following cases are considered
corresponding to three different actuator conditions:
1. a ¼ m
m ^ a ¼ 0, then ma ¼ 0; which implies that the corresponding actuator
f
u ðtÞ is completely failed.
2. m
a ¼ m ^ a ¼ 1, thus we obtain ma ¼ 1; which represents the case of no fault in
the actuator uf ðtÞ.
3. 0<m a < m ^ a < 1, which means that there exists partial fault in the corre-
sponding actuator uf ðtÞ.
The sprung mass ms ðtÞ and the unsprung mass mu ðtÞ are uncertainties, which
vary in a given range, that is, ms ðtÞ 2 ½ms min , ms max  and mu ðtÞ 2 ½mu min , mu max : It is
to say that the uncertainty scenarios of the mass ms ðtÞ are bounded by its minimum
value ms min and its maximum value ms max . In addition, the mass mu ðtÞ is bounded
178 Handbook of vehicle suspension control systems

by its minimum value mu min and its maximum value mu max : Next, we obtain the
values of ms1ðtÞ and mu1ðtÞ from ms ðtÞ 2 ½ms min , ms max  and mu ðtÞ 2 ½mu min , mu max :
Then we have

1 1 1 1
max ¼ ^ s,
¼: m min ¼ ¼: m
s
ms ðtÞ ms min ms ðtÞ ms max
1 1 1 1
max ¼ ^ u , min
¼: m ¼ ¼: m
u
mu ðtÞ mu min mu ðtÞ mu max

The sector nonlinear method [18] is employed to represent 1=ms ðtÞ and 1=mu ðtÞ by

1
¼ M 1 ð x 1 ð t ÞÞ m
^ s þ M2 ðx1 ðtÞÞm
s
ms ðtÞ
1
¼ N1 ðx2 ðtÞÞm
^ u þ N 2 ð x 2 ð t ÞÞ m
u
mu ðtÞ

where x1 ðtÞ ¼ 1=ms ðtÞ and x2 ðtÞ ¼ 1=mu ðtÞ are premise variables,

M1 ðx1 ðtÞÞ þ M2 ðx1 ðtÞÞ ¼ 1

N1 ðx2 ðtÞÞ þ N2 ðx2 ðtÞÞ ¼ 1

The membership functions M1 ðx1 ðtÞÞ, M2 ðx1 ðtÞÞ, N1 ðx2 ðtÞÞ, and N2 ðx2 ðtÞÞ can be
calculated as

1 1
ms m
^s 
m ðt Þ ms ðtÞ
M 1 ð x 1 ð t ÞÞ ¼ s , M2 ðx1 ðtÞÞ ¼
^s  m
m s ^s  m
m s
1 1
mu m
^u 
m ðt Þ mu ðtÞ
N1 ðx2 ðtÞÞ ¼ u , N 2 ð x 2 ð t ÞÞ ¼
^u  m
m u ^u  m
m u

The member functions are labeled as Heavy, Light, Heavy and Light as shown in
Figure 7.2. Then, the system with uncertainty in (7.7) is represented by the fol-
lowing fuzzy model:

(a) (b)
M2(ξ1(t)) M1(ξ1(t)) N2(ξ2(t)) N1(ξ2(t))
1 1
Light Heavy Light Heavy

0 ξ1(t) 0 ξ2(t)
0 0

Figure 7.2 (a) Membership functions M1 ðx1 ðtÞÞ and M2 ðx1 ðtÞÞ. (b) Membership
functions N1 ðx2 ðtÞÞ and N2 ðx2 ðtÞÞ
 Fault-tolerant control for uncertain vehicle suspension systems 179

Model Rule 1: IF x1 ðtÞ is Heavy and x2 ðtÞ is Heavy,

THEN
x_ ðtÞ ¼ A1 xðtÞ þ B1 uf ðtÞ þ B11 wðtÞ
z1 ðtÞ ¼ C11 xðtÞ þ D11 uf ðtÞ
z2 ðtÞ ¼ C21 xðtÞ

matrices A1 , B1 , B11 , C11 , D11 , and C21 are obtained by replacing ms1ðtÞ and mu1ðtÞ
with matrices AðtÞ, BðtÞ, B1 ðtÞ, C1 ðtÞ, D1 ðtÞ, and C2 ðtÞ with m
^ s and m
^ u , respectively.

Model Rule 2: IF x1 ðtÞ is Heavy and x2 ðtÞ is Light,

THEN
x_ ðtÞ ¼ A2 xðtÞ þ B2 uf ðtÞ þ B12 wðtÞ
z1 ðtÞ ¼ C12 xðtÞ þ D12 uf ðtÞ
z2 ðtÞ ¼ C22 xðtÞ

matrices A2 , B2 , B12 , C12 , D12 , and C22 are obtained by replacing ms1ðtÞ and mu1ðtÞ
with matrices AðtÞ, BðtÞ, B1 ðtÞ, C1 ðtÞ, D1 ðtÞ, and C2 ðtÞ, with m
^ s and m
 u , respectively.

Model Rule 3: IF x1 ðtÞ is Light and x2 ðtÞ is Heavy,

THEN
x_ ðtÞ ¼ A3 xðtÞ þ B3 uf ðtÞ þ B13 wðtÞ
z1 ðtÞ ¼ C13 xðtÞ þ D13 uf ðtÞ
z2 ðtÞ ¼ C23 xðtÞ

matrices A3 , B3 , B13 , C13 , D13 , and C23 are obtained by replacing ms1ðtÞ and mu1ðtÞ with
matrices AðtÞ, BðtÞ, B1 ðtÞ, C1 ðtÞ, D1 ðtÞ, and C2 ðtÞ with m
 s and m
^ u , respectively.

Model Rule 4: IF x1 ðtÞ is Light and x2 ðtÞ is Light,

THEN
x_ ðtÞ ¼ A4 xðtÞ þ B4 uf ðtÞ þ B14 wðtÞ
z1 ðtÞ ¼ C14 xðtÞ þ D14 uf ðtÞ
z2 ðtÞ ¼ C24 xðtÞ
matrices A4 , B4 , B14 , C14 , D14 , and C24 are obtained by replacing ms1ðtÞ and mu1ðtÞ with
matrices AðtÞ, BðtÞ, B1 ðtÞ, C1 ðtÞ, D1 ðtÞ, and C2 ðtÞ with m
 s and m
 u , respectively.
180 Handbook of vehicle suspension control systems

Fuzzy blending allows to infer the overall fuzzy model as follows:

X
4  
x_ ðtÞ ¼ hi ðxðtÞÞ Ai xðtÞ þ Bi uf ðtÞ þ B1i wðtÞ
i¼1
X
4  
z1 ðtÞ ¼ hi ðxðtÞÞ C1i xðtÞ þ D1i uf ðtÞ ð7:9Þ
i¼1
X
4
z2 ðtÞ ¼ hi ðxðtÞÞC2i xðtÞ
i¼1

where

h1 ðxðtÞÞ ¼ M1 ðx1 ðtÞÞ  N1 ðx2 ðtÞÞ

h2 ðxðtÞÞ ¼ M1 ðx1 ðtÞÞ  N2 ðx2 ðtÞÞ
h3 ðxðtÞÞ ¼ M2 ðx1 ðtÞÞ  N1 ðx2 ðtÞÞ
h4 ðxðtÞÞ ¼ M2 ðx1 ðtÞÞ  N2 ðx2 ðtÞÞ

It
P4is apparent that the fuzzy weighting functions hi ðxðtÞÞ satisfy hi ðxðtÞÞ  0;
i¼1 hi ðxðtÞÞ ¼ 1. In order to design a fuzzy fault-tolerant controller, parallel
distributed compensation is adapted and the following fuzzy controller is obtained:

Control Rule 1: IF x1 ðtÞ is Heavy and x2 ðtÞ is Heavy,

THEN uðtÞ ¼ Ka1 xðtÞ:
Control Rule 2: IF x1 ðtÞ is Heavy and x2 ðtÞ is Light,
THEN uðtÞ ¼ Ka2 xðtÞ:
Control Rule 3: IF x1 ðtÞ is Light and x2 ðtÞ is Heavy,
THEN uðtÞ ¼ Ka3 xðtÞ:
Control Rule 4: IF x1 ðtÞ is Light and x2 ðtÞ is Light,
THEN uðtÞ ¼ Ka4 xðtÞ:

Hence, the overall fuzzy control law is represented by

X
4
uðtÞ ¼ hj ðxðtÞÞKaj xðtÞ ð7:10Þ
j¼1
P4
where Kaj ðj ¼ 1; 2; 3; 4Þ are the local control gains and uðtÞ ¼ j¼1 hj ðxðtÞÞKaj xðtÞ:
For simplicity, the following notations will be used:

hi ¼: hi ðxðtÞÞ, hj ¼: hj ðxðtÞÞ
 Fault-tolerant control for uncertain vehicle suspension systems 181

Applying the fuzzy controller (7.10) to system (7.9) yields the closed-loop system:
4 X
X 4   
x_ ðtÞ ¼ hi hj Ai þ Bi ma ðtÞKaj xðtÞ þ B1i wðtÞ
i¼1 j¼1

X
4 X
4   
z1 ðtÞ ¼ hi hj C1i þ D1i ma Kaj xðtÞ ð7:11Þ
i¼1 j¼1

X
4
z2 ðtÞ ¼ hi C2i xðtÞ
i¼1

The fuzzy system in (7.11) is established based on the practically measurable

sprung ms ðtÞ and unsprung mu ðtÞ. The sector nonlinearity method [18] is employed
to analyze the variation of the sprung ms ðtÞ and unsprung mu ðtÞ and present the T–S
fuzzy system in (7.11).

7.3 Fault-tolerant fuzzy controller design

This section focuses on designing fault-tolerant fuzzy control i for the active sus-
pension system with actuator fault. The fault-tolerant fuzzy controller can ensure
that the closed-loop system in (7.11) is asymptotically stable, and it also guarantees
a prescribed gain from disturbance wðtÞ to performance output z1 ðtÞ, under the
condition that the suspension stroke and tire deflection constraints are satisfied.
First, the following lemma is presented.
Lemma 7.1. For a time-varying diagonal matrix FðtÞ ¼ diagfs1 ðtÞ, s2 ðtÞ, . . . , sp ðtÞg
and two matrices R and S with appropriate dimensions, if jFðtÞj  V , where
V > 0 is a known diagonal matrix, then for any scalar e > 0; it is true that
RFS þ S T FT RT  eRVRT þ e1 S T VS
Next, the following scalars are introduced that will be used in the development
later in this chapter. Ma0 ¼ ðm a þ m ^ a Þ=2, La ¼ ½ma  Ma0 =ma0 , and Ja ¼
^a  m
ðm  a Þ=ðm
^a þ m a Þ. Thus, one has ma ¼ Ma0 ðI þ La Þ and LTa La  JaT Ja  I.
Then, it leads to the following theorem.
Theorem 7.1. Consider the closed-loop system in (7.11). For given matrices Kaj , if
there exist matrices P > 0 with appropriate dimensions and positive scalars
eij > 0 ði, j ¼ 1; 2; 3; 4Þ such that the following linear matrix inequalities (LMIs)
hold for q ¼ 1; 2:
2 T
3
symðPAi þ PBi Ma0 Kaj Þ PB1i C1i þ KajT Ma0 DT1i PBi eij Ma0 KajT
6 7
6  g2 I 0 0 0 7
6 7
6 7
6   I D1i 0 7<0
6 7
6    1
eij Ja 0 7
4 5
    eij Ja1
ð7:12Þ
182 Handbook of vehicle suspension control systems
" pffiffiffi #
P rfC2i gTq
<0 ð7:13Þ
 I
Furthermore,
1. The closed-loop system is robustly asymptotically stable.
2. The performance kTz1 w k1 < g is minimized subject to output constraints with
the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 , where Tz1 w
denotes the closed-loop transfer function from the road disturbance wðtÞ to the
control output z1 ðtÞ:
The proof of Theorem 7.1 is given in the appendix.
In what follows, the fault-tolerant fuzzy H1 controller existence condition is
presented for the active suspension system in (7.11); it is based on fault-tolerant
fuzzy H1 performance analysis criterion in Theorem 7.1.
Theorem 7.2. Consider the closed-loop system in (7.11). If there exist matrices
P > 0 and Yaj with appropriate dimensions and positive scalars e ij > 0
ði,j ¼ 1; 2; 3; 4Þ such that the following LMIs hold for q ¼ 1; 2:
2 T
3
symðAi P þ Bi Yaj Þ B1i PC1i þ YajT DT1i e ij Bi YajT
6 7
6  g2 I 0 0 0 7
6 7
6 7
Yij ¼ 6   I e ij D1i 0 7<0
6 7
6    e ij Ja1
0 7
4 5
    e ij Ja1
ð7:14Þ
" pffiffiffi #
P rP fC2i gTq
<0 ð7:15Þ
 I

then a fault-tolerant controller in the form of (7.10) exists, such that

1. The closed-loop system is asymptotically stable.
2. The performance kTz1 w k1 < g is minimized subject to output constraints with
the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 .
Moreover, if inequalities (7.14) and (7.15) have a feasible solution, then the
1
control gain Kaj in (7.10) is given by Kaj ¼ Ma0
1
Yaj P :
The proof of Theorem 7.2 is given in the appendix.
Fuzzy H1 controller design is presented for active suspension systems based
on T–S fuzzy model method. If there is no actuator fault in the active suspension
system, then we obtain
x_ ðtÞ ¼ AðtÞxðtÞ þ B1 ðtÞwðtÞ þ BðtÞuðtÞ
z1 ðtÞ ¼ C1 ðtÞxðtÞ þ D1 ðtÞuðtÞ ð7:16Þ
z2 ðtÞ ¼ C2 ðtÞxðtÞ
 Fault-tolerant control for uncertain vehicle suspension systems 183

Based on the above-presented fuzzy modeling, the overall fuzzy model is inferred
as follows:
X
4
x_ ðtÞ ¼ hi ðxðtÞÞ½Ai xðtÞ þ Bi uðtÞ þ B1i wðtÞ
i¼1
X
4
z1 ðtÞ ¼ hi ðxðtÞÞ½C1i xðtÞ þ Di uðtÞ ð7:17Þ
i¼1
X
4
z2 ðtÞ ¼ hi ðxðtÞÞC2i xðtÞ
i¼1

In addition, the overall fuzzy control law is represented by

X
4
uðtÞ ¼ hj ðxðtÞÞKsj xðtÞ ð7:18Þ
j¼1

For the case of the standard controller (7.18), the closed-loop system is given by
X
4 X
4  
x_ ðtÞ ¼ hi hj Ai xðtÞ þ Bi Ksj xðtÞ þ B1i wðtÞ
i¼1 j¼1

X
4 X
4  
z1 ðtÞ ¼ hi hj C1i xðtÞ þ D1i Ksj xðtÞ ð7:19Þ
i¼1 j¼1

X
4
z2 ðtÞ ¼ hi C2i xðtÞ
i¼1

Employing the similar method proposed earlier in this section, the fuzzy H1 con-
troller design condition as below is derived from Theorem 7.2.
Corollary 7.1. The closed-loop system (7.19) is asymptotically stable with an H1
disturbance attenuation level g, if there exist matrices P > 0 and Ysj ðj ¼ 1; 2; 3; 4Þ
with appropriate dimensions such that the following LMIs hold for q ¼ 1; 2:
2 T 3
symðAi P þ Bi Ysj Þ B1i PC1i þ YsjT DT1i
6 7
Y ij ¼ 4  g2 I 0 5<0 ð7:20Þ
  I
" pffiffiffi #
P rP fC2i gTq
<0 ð7:21Þ
 I
Then a standard controller in the form of (7.18) exists, such that
1. The closed-loop system is asymptotically stable.
2. The performance kTz1 w k1 < g is minimized subject to output constraint.
Moreover, if inequalities (7.20) and (7.21) have a feasible solution, then the
1
control gain Ksj in (7.18) is given by Ksj ¼ Ysj P :
184 Handbook of vehicle suspension control systems

Remark 7.1. In the chapter, the conservativeness of Theorem 7.1 will be reduced if
we again obtain from the term (A.2),
X
4 X
4
T
V_ ðtÞ þ zT1 ðtÞz1 ðtÞ  g2 wT ðtÞwðtÞ  hi hj x ðtÞPij x ðtÞ
i¼1 j¼1
!
T X
4 X
3 X
4  
¼ x ðt Þ h2i Pii þ hi ðxðtÞÞhj ðxðtÞÞ Pij þ Pji x ðt Þ
i¼1 i¼1 j¼iþ1

Then, based on Theorem 7.1, by using the Schur complement, we can have the
following much less conservative corollary.
Corollary 7.2. Consider the closed-loop system in (7.11). If there exist matrices
P > 0 and Yaj with appropriate dimensions and positive scalars e ij > 0
ði, j ¼ 1; 2; 3; 4Þ such that the following LMIs hold for q ¼ 1; 2:
Yii < 0; i ¼ 1; 2; . . . ; 4
Yij þ Yji < 0; i < j, i, j ¼ 1; 2; . . . ; 4
" pffiffiffi #
P rP fC2i gTq
<0
 I
where Yij is defined in Theorem 7.1. Then a fault-tolerant controller in the form of
1
(7.10) exists and Kaj ¼ Ma0
1
Yaj P such that
1. The closed-loop system is asymptotically stable.
2. The performance kTz1 w k1 < g is minimized subject to output constraints with
the disturbance energy under the bound wmax ¼ ðr  V ð0ÞÞ=g2 .
Similarly, the conservativeness of Corollary 7.1 will be reduced by the fol-
lowing corollary.
Corollary 7.3. Consider the closed-loop system in (7.19). The closed-loop system
(7.19) is asymptotically stable with an H1 disturbance attenuation level g, if there
exist matrices P > 0 and Ysj ðj ¼ 1; 2; 3; 4Þ with appropriate dimensions such that
the following LMIs hold for q ¼ 1; 2:

Y ii < 0; i ¼ 1; 2; . . . ; 4
Y ij þ Y ji < 0; i < j, i, j ¼ 1; 2; . . . ; 4
" pffiffiffi #
P rP fC2i gTq
<0
 I
where Y ij is defined in Corollary 7.1. Then a standard controller in the form of
(7.18) exists, such that
1. The closed-loop system is asymptotically stable.
2. The performance kTz1 w k1 < g is minimized subject to output constraint.
Moreover, if inequalities (7.20) and (7.21) have a feasible solution, then the
1
control gain Ksj in (7.18) is given by Ksj ¼ Ysj P :
 Fault-tolerant control for uncertain vehicle suspension systems 185

7.4 Simulation results

In this section, we utilize a quarter-vehicle active suspension system to show the
effectiveness and merit of the proposed approach. First, the parameters of
the quarter car are listed in Table 7.1. The sprung mass ms ðtÞ is assumed to set in
the range ½943 kg, 1003 kg and the unsprung mass mu ðtÞ in ½110 kg; 118 kg. In
this study, the maximum allowable suspension stroke is set as zmax ¼ 0:08 m with
r ¼ 1. It is assumed that there exists the following actuator fault, namely,
 a ¼ 0:001 and m
m ^ a ¼ 0:5. For the uncertain active suspension system in (7.11), by
applying the convex optimization to Corollary 7.2, the fault-tolerant fuzzy con-
troller gain matrices are found:

Ka1 ¼ 105  ½ 0:4097 1:7146 2:1528 0:3107 

5
Ka2 ¼ 10  ½ 0:5271 1:9489 2:2148 0:3351 
5
ð7:22Þ
Ka3 ¼ 10  ½ 0:5113 1:8143 2:2044 0:3077 
Ka4 ¼ 105  ½ 0:4090 1:7387 2:1530 0:3104 
Similarly, the fuzzy controller gain matrices from Corollary 7.3 can be found
as below:

Ks1 ¼ 107  ½ 1:6214 0:1972 0:3708 0:0266 

7
Ks2 ¼ 10  ½ 1:5460 0:1859 0:3538 0:0252 
7
ð7:23Þ
Ks3 ¼ 10  ½ 1:8240 0:2084 0:4204 0:0295 
Ks4 ¼ 107  ½ 1:4860 0:1795 0:3401 0:0243 
In order to confirm the effectiveness and advantages of the fault-tolerant fuzzy
controller compared with the standard fuzzy controller in the case of possible
actuator fault, we consider the road disturbance as below. In the context of active
suspension performance, the road disturbance can be generally assumed as discrete
events of relatively short duration and high intensity, caused by, for example, a
pronounced bump or pothole on an otherwise smooth road surface. The road sur-
face is represented by
8

> A 2pV L
< 1  cos t , if 0  t 
zr ðtÞ ¼ 2 L V ð7:24Þ
>
: 0; L
if t >
V
where A and L are the height and the length of the bump, respectively. Assume
A ¼ 50 mm, L ¼ 6 m, and the vehicle forward velocity as V ¼ 35 (km/h).

Table 7.1 Quarter-car model parameters

ks kt cs ct
42,720 N/m 101,115 N/m 1,095 Ns/m 14:6 Ns/m
186 Handbook of vehicle suspension control systems

2
Passive
Ksi
1.5

1
Body acceleration (m/s2)

0.5

−0.5

−1

−1.5

−2

−2.5
0 1 2 3 4 5 6 7 8 9 10

Figure 7.3 Responses of body vertical accelerations

First, when there is no actuator fault, the standard fuzzy controller (7.18) with
the parameters in (7.23) is effective for the uncertain suspension system. In addi-
tion, we hope that this fuzzy controller can guarantee that: (1) the sprung mass
acceleration z1 ðtÞ is as small as possible; (2) the suspension deflection is below the
maximum allowable suspension stroke zmax ¼ 0:08 m, which means that x1 ðtÞ=zmax
is below 1; (3) the relation dynamic tire load kt x2 ðtÞ=ðms ðtÞ þ mu ðtÞÞg < 1. The
variational sprung mass ms ðtÞ and the variational unsprung mass mu ðtÞ are set
as: ms ðtÞ ¼ 973 þ 30 sinðtÞ kg and mu ðtÞ ¼ 114 þ 4 cosðtÞ kg, respectively, for
deriving the fuzzy membership functional hi ðxðtÞÞ ði ¼ 1; 2; 3; 4Þ. By using the
standard fuzzy H1 controller with the parameters in (7.23), we derive the corre-
sponding closed-loop fuzzy system. Figures 7.3 and 7.4 depict the responses of
body vertical accelerations and the actuator force for the open- (e.g., passive) and
closed-loop (e.g., active) systems. Figure 7.5 demonstrates the response of
suspension stroke, and Figure 7.6 shows the response of tire deflection constraint
for both the passive and the active systems. It is observed from Figure 7.3 that
the proposed fuzzy H1 control strategy reduces the sprung mass acceleration
significantly in comparison with the passive suspension under the same road dis-
turbance. The designed controller can achieve the less value of the maximum body
acceleration for the active suspension system than the passive system, and pas-
senger acceleration in the active suspension system is reduced significantly, which
guarantees better ride comfort. In addition, it can be seen that, from Figures 7.5
and 7.6, the suspension deflection constraint x1 ðtÞ=zmax < 1 and the relation
dynamic tire load constraint kt x2 ðtÞ=ðms ðtÞ þ mu ðtÞÞg < 1 are guaranteed, which
implies the road holding capability is ensured by the desired fuzzy controller.
In Figures 7.7–7.10, the responses to the open- and closed-loop systems with
the actuator fault via the standard fuzzy controller Ksi and fault-tolerant fuzzy
 Fault-tolerant control for uncertain vehicle suspension systems 187

1500
Kai
1000

500
Active force (N)

−500

−1000

−1500

−2000
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.4 Response of active force

The ratio of suspension deflection and the maximum limitation

0.6
Passive
Ksi
0.4

0.2

−0.2

−0.4

−0.6

−0.8
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.5 Responses of suspension deflection constraint

controller Kai (i ¼ 1; 2; 3; 4) are given. These figures show that the less value of the
maximum body acceleration is achieved for the active suspension system, the
suspension deflection constraint x1 ðtÞ=zmax < 1 is guaranteed and the relation
dynamic tire load kt x2 ðtÞ=ðms ðtÞ þ mu ðtÞÞg is below 1 in comparison with the
passive suspension system, by utilizing the standard fuzzy H1 controller Ksi and
188 Handbook of vehicle suspension control systems

0.2
Passive
Ksi
0.15
The relation dynamic tire load

0.1

0.05

−0.05

−0.1

−0.15

−0.2
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.6 Responses of tire stroke constraint

2
Passive
Ksi
1.5
Kai
1
Body acceleration (m/s2)

0.5

−0.5

−1

−1.5

−2

−2.5
0 1 2 3 4 5 6 7 8 9 10

Figure 7.7 Responses of body vertical accelerations

fault-tolerant controller Kai (i ¼ 1; 2; 3; 4). However, it should be noticed that the

fault-tolerant fuzzy controller can achieve less value of the maximum body accel-
eration than the standard H1 controller for the active suspension system with
actuator fault in these simulation results. We can see from Figures 7.7–7.10 that Kai
(i ¼ 1; 2; 3; 4) is capable to provide a much steadier control force in fault condition
than the standard controller Ksi (i ¼ 1; 2; 3; 4).
 Fault-tolerant control for uncertain vehicle suspension systems 189

The ratio of suspension deflection and the maximum limitation

0.6
Passive
Ksi
0.4 Kai

0.2

−0.2

−0.4

−0.6

−0.8
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.8 Responses of suspension deflection constraint

0.2
Passive
Ksi
0.15 Kai
The relation dynamic tire load

0.1

0.05

−0.05

−0.1

−0.15

−0.2
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.9 Responses of tire stroke constraint

7.5 Conclusions
In this chapter, we have investigated the problem of fault-tolerant fuzzy H1 control
for active suspension systems with actuator fault. The sprung and unsprung mass
variations, the actuator fault, and the suspension performance have been included to
190 Handbook of vehicle suspension control systems

1500
Ksi
Kai
1000

500
Active force (N)

−500

−1000

−1500

−2000
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 7.10 Response of active force

construct the T–S fuzzy system for the control design objective. The fault-tolerant
fuzzy H1 controller has been designed such that the resulting closed-loop T–S
fuzzy system is asymptotically stable with H1 performance, and simultaneously
satisfies the constraint suspension performance. Simulation results have shown the
effectiveness and the merit of the proposed design method.

Appendix

A.1 Proof of Theorem 7.1

Consider the Lyapunov functional as follows:

V ðtÞ ¼ xT ðtÞPxðtÞ ðA:1Þ

The derivative of V ðtÞ along the solution of system (7.11) is expressed as

V_ ðtÞ  2xT ðtÞP_x ðtÞ

To develop H1 performance analysis criterion, the system (7.11) is stable with

wðtÞ ¼ 0; then the H1 performance index is satisfied. The following inequalities is
obtained:
4 X
X 4  
V_ ðtÞ ¼ hi hj xT ðtÞfsym PAi þ PBi ma Kaj gxðtÞ
i¼1 j¼1
 Fault-tolerant control for uncertain vehicle suspension systems 191

Via Lemma 7.1, it is found from ma ¼ Ma0 ðI þ La Þ that

     
sym PAi þ PBi ma Kaj ¼ sym PAi þ PBi Ma0 Kaj þ sym PBi Ma0 La Kaj
  T T
 sym PAi þ PBi Ma0 Kaj þ e1
ij PBi Ja Bi P þ eij Kaj Ma0 Ja Ma0 Kaj

From (7.12), we know that V_ ðtÞ < 0, so the system in (7.11) is asymptotically
stable. Next, the H1 performance of the system in (7.11) is established under zero
initial conditions. First, the Lyapunov functional is defined as shown in (A.1). It is
not difficult to achieve

V_ ðtÞ þ zT1 ðtÞz1 ðtÞ  g2 wT ðtÞwðtÞ

X
4 X
4
 hi hj xT ðtÞfsymðPAi þ PBi ma Kaj Þ
i¼1 j¼1

þ ðC1i þ D1i ma Kaj ÞT ðC1i þ D1i ma Kaj ÞgxðtÞ

4 X
X 4
þ hi hj xT ðtÞ2PB1i wðtÞ  g2 wT ðtÞwðtÞ
i¼1 j¼1
4 X
X 4
T
¼ hi hj x ðtÞPij x ðtÞ, ðA:2Þ
i¼1 j¼1

where
" #
 T T
T Qij PB1i
x ðt Þ ¼ x ðt Þ w ðt Þ , Pij ¼
 g2
   T  
Qij ¼ sym PAi þ PBi ma Kaj þ C1i þ D1i ma Kaj C1i þ D1i ma Kaj

On the other hand,

2 T 3
symðPAi þ PBi ma Kaj Þ PB1i C1i þ KajT Ma0 DT1i
6 7
Pij ¼ 4  g2 0 5
  I
2 T 3
symðPAi þ PBi ma Kaj Þ PB1i C1i þ KajT Ma0 DT1i
6 7
4  g2 0 5
  I
2 3 2 3T
PBi PBi
6 7 6 7
þ e1
ij 4 0 5Ja 4 0 5 þ eij ½ Kaj 0 0 T Ma0 Ja Ma0 ½ Kaj 0 0
D1i D1i
192 Handbook of vehicle suspension control systems

According to the Schur complement and the above method, we develop

V_ ðtÞ þ zT1 ðtÞz1 ðtÞ  g2 wT ðtÞwðtÞ < 0 ðA:3Þ

for all nonzero w 2 L2 ½0; 1Þ: Under zero initial conditions, we have V ð0Þ ¼ 0 and
V ð1Þ  0: Integrating both sides of (A.3) yields kz1 k2 < gkwk2 for all nonzero
w 2 L2 ½0; 1Þ, and the H1 performance is established.
Inequality (A.3) guarantees V_ ðtÞ  g2 wT ðtÞwðtÞ < 0: Integrating both sides of
the above inequality from zero to any t > 0, we obtain
ðt
V ðtÞ  V ð0Þ < g2 wT ðsÞwðsÞ ds < g2 kwk22 ðA:4Þ
0

From the definition of the Lyapunov functional in (A.1), we obtain that

xT ðtÞPxðtÞ < r with r ¼ g2 wmax þ V ð0Þ: The following inequalities hold
 
X 4 
2  T T 
max jfz2 ðtÞgq j  max  hi x ðtÞfC2i gq fC2i gq xðtÞ
t>0 t>0  
i¼1 2
 
X 4 
 T 1
1 T  1 1 
¼ max  hi x ðtÞP2 P 2 fC2i gq fC2i gq P 2 P2 xðtÞ
t>0  
i¼1 2
!
X
4
hi P2 fC2i gTq fC2i gq P2 , q ¼ 1; 2
1 1
< r  qmax
i¼1

where qmax ðÞ represents maximal eigenvalue. From the above inequalities, it
implies that the constraints are guaranteed, if

X
4
hi P2 fC2i gTq fC2i gq P2 < I
1 1
r ðA:5Þ
i¼1

which means

X
4 
hi r  P2 fC2i gTq fC2i gq P2  I < 0
1 1

i¼1

which is guaranteed by the feasibility of (). The proof is completed. &

A.2 Proof of Theorem 7.2

Performing corresponding congruence transformation to (7.14) by
n o
1
diag P , I, I, e 1
ij I, e 1
ij I
 Fault-tolerant control for uncertain vehicle suspension systems 193

together with the change of matrix variables defined by

1 1 1
P ¼ P , Kj ¼ Ma0 Yj P , eij ¼ e 1
ij

It is concluded that the condition in (7.12) holds. On the other hand, (7.15) is
equivalent to (7.13) by performing a simple congruence transformation with
1
diagfP , Ig. Therefore, all the conditions in Theorem 7.1 are satisfied. The proof
is completed. &

References
[1] J. Cao, H. Liu, P. Li, and D. Brown. State of the art in vehicle active sus-
pension adaptive control systems based on intelligent methodologies. IEEE
Transactions on Intelligent Transportation Systems, 9(3):392–405, 2008.
[2] D. Hrovat. Survey of advanced suspension developments and related optimal
control applications. Automatica, 33(10):1781–1817, 1997.
[3] N. Al-Holou, T. Lahdhiri, D. Joo, J. Weaver, and F. Al-Abbas. Sliding mode
neural network inference fuzzy logic control for active suspension systems.
IEEE Transactions on Fuzzy Systems, 10(2):234–246, 2002.
[4] A. Alleyne and J. Hedrick. Nonlinear adaptive control of active suspensions.
IEEE Transactions on Control Systems Technology, 3(1):94–101, 1995.
[5] H. Chen and K. Guo. Constrained H1 control of active suspensions: an LMI
approach. IEEE Transactions on Control Systems Technology, 13(3):412–421,
2005.
[6] M. ElMadany and Z. Abduljabbar. Linear quadratic Gaussian control of a
quarter-car suspension. Vehicle System Dynamics, 32(6):479–497, 1999.
[7] I. Fialho and G. Balas. Road adaptive active suspension design using linear
parameter-varying gain-scheduling. IEEE Transactions on Control Systems
Technology, 10(1):43–54, 2002.
[8] C. Sivrioglu and I. Cansever. LPV gain-scheduling controller design for a
non-linear quarter-vehicle active suspension system. Transactions of the
Institute of Measurement and Control, 31(1):71–95, 2009.
[9] M. Yamashita, K. Fujimori, K. Hayakawa, and H. Kimura. Application of
H1 control to active suspension systems. Automatica, 30(11):1717–1729,
1994.
[10] Y. Zhao, L. Zhao, and H. Gao. Vibration control of seat suspension using
H1 reliable control. Journal of Vibration and Control, 16(12):1859, 2010.
[11] H. Du, N. Zhang, and J. Lam. Parameter-dependent input-delayed control
of uncertain vehicle suspensions. Journal of Sound and Vibration, 317(3–5):
537–556, 2008.
[12] H. Gao, J. Lam, and C. Wang. Multi-objective control of vehicle active
suspension systems via load-dependent controllers. Journal of Sound and
Vibration, 290(3–5):654–675, 2006.
194 Handbook of vehicle suspension control systems

[13] H. Gao, W. Sun, and P. Shi. Robust sampled-data H1 control for vehicle
active suspension systems. IEEE Transactions on Control Systems Tech-
nology, 18(1):238–245, 2010.
[14] L. Zadeh. Fuzzy sets. Information and Control, 8(3):338–353, 1965.
[15] G. Feng. A survey on analysis and design of model-based fuzzy control
systems. IEEE Transactions on Fuzzy Systems, 14(5):676–697, 2006.
[16] C. Lin, G. Wang, and T. Lee. LMI approach to analysis and control of
Takagi–Sugeno fuzzy systems with time delay. Springer Verlag, 2007.
[17] M. Sugeno. An introductory survey of fuzzy control. Information Sciences,
36(1–2):59–83, 1985.
[18] K. Tanaka and H. Wang. Fuzzy control systems design and analysis: a linear
matrix inequality approach. Wiley-Interscience, 2001.
 Chapter 8
H? fuzzy control of suspension systems with
actuator saturation
Dounia Saifiaa, Mohammed Chadlib and Salim Labioda

Abstract

This chapter focuses on H? fuzzy control of suspension systems under actuator

saturation. The Takagi–Sugeno (T–S) approach is used to model the suspension
system (quarter, half and full cars) by interpolation of different local linear models.
A nonlinear state feedback control parallel distributed compensation (PDC) is
employed for designing control system. The main idea of this controller consists in
designing a linear feedback control for each local linear model. To address the
input saturation problem, both constrained and saturated control input cases are
proposed. In the two cases, H? stabilization conditions are derived using Lyapunov
method. Moreover, a controller design with the largest domain of attraction is
formulated and solved as a linear matrix inequality optimization problem. An
application to quarter-car suspension system is given. Our simulation results show
that both saturated and constrained controls can stabilize the resulting closed-loop
suspension quarter car via PDC control and eliminate the effect of external dis-
turbances despite the presence of saturation. Indeed, the main roles of car suspen-
sion systems which consist of improving ride comfort of passengers and the road
holding capacity of the vehicle are achieved.

8.1 Introduction
The main roles of vehicle suspension systems consist of improving ride comfort of
passengers and the road holding capacity of the vehicle. There are three types of
suspension systems: passive, semi-active and active [1]. The passive suspension
(Figure 8.1(a)) consists of energy storage mechanical components (spring) and
dissipation (damping). The semi-active suspension (Figure 8.1(b)) is constituted by
a spring and a damper controlled by an exogenous signal. The damper can only
dissipate energy in a frequency range varying with the type of the used semi-active

a
LAJ, Faculty of Science and Technology, University of Jijel BP. 98, Ouled Aissa, 18000 Jijel, Algeria
b
University of Picardie Jules Verne, MIS (EA 4290) 33 rue Saint-Leu - 80039, Amiens – France.
196 Handbook of vehicle suspension control systems

ms ms ms

fk fb fk fb

mus mus mus

kr kr kr

(a) (b) (c)

Figure 8.1 Different types of suspensions

actuator. The active suspension (Figure 8.1(c)) is composed of a force generating

actuator placed between the sprung and unsprung masses [2].
Among these types of suspension systems, active suspensions have received a
lot of attention since they seem to provide the best performances. Actually, many
control approaches such as crone control [3, 4], impedance control [5], back-
stepping control [6], singular perturbation [7], H?-based controller [8, 9], adaptive
sliding controller [10] and fuzzy control [11–13] are applied to control the active
suspension. However, in the aforementioned papers, authors have not taken into
account the inherent nonlinearity of actuator saturation. The limitations of actuators
result in constraints and saturation on their amplitudes and/or their velocities. For
example, electrohydraulic actuators [14, 15] are more used in active suspension
where the input voltage is limited in practice.
Suspension systems are nonlinear systems [7]. The conventional linear con-
trols will result in poor dynamic performance or system instability. In this chapter,
we will use Takagi–Sugeno (T–S) fuzzy system [15–20] to model suspension
systems. The control design is based on state feedback in parallel distributed
compensation (PDC) scheme. The idea is that for each local linear model, a linear
feedback control is designed [21–22, 16]. Recently, H? approach appears very
attractive for control design for T–S fuzzy model in the presence of external dis-
turbances [8, 14, 23–24]. Most of these works use linear matrix inequality (LMI)
techniques to solve the control problem of a nonlinear system [25]. Based on these
works, we address, via PDC control, the H? stabilization problem of a nonlinear
system with actuator saturation. Lyapunov approach is used to stabilize multiple
model subject to actuator saturation and external disturbances. Both constrained
and saturated control designs are developed. The H? stabilization problem with
the larger attraction domain is formulated and solved as LMI problem.
This chapter is organized as follows. Sections 8.2 and 8.3 present the physical
model of suspension systems and its T–S fuzzy modelling. Sections 8.4–8.6 give
mathematical description of actuator saturation, quadratic stabilization of T–S
fuzzy model and H? approach respectively. Sections 8.7 and 8.8 give the H? state
 H? fuzzy control of suspension systems with actuator saturation 197

feedback stabilization conditions in term of LMI optimization problem. Finally, in

section 8.9, the quarter-car suspension system is used to show the effectiveness of
the proposed method.
Notation: Ir denotes the set f1, 2, . . . , rg, < denotes the set of real number
and <nm the set of all n  m real matrices. M > ð , < , Þ 0 is used to denote a
symmetric positive definite (positive semi-definite, negative definite, negative
semi-definite respectively) matrix. * denotes the symmetric block matrix, X þ ðÞ
denotes X þ X T ,  denotes the multiplication and co denotes the convex hull.

8.2 Suspension systems model

The main role of a vehicle suspension system (see Figure 8.3) is to ensure pas-
senger comfort and road handling of the vehicle. The suspension should act as a
filter to remove unwanted vibration frequencies to ensure road holding and pas-
senger comfort.
The choice of a dynamic model depends on vehicle suspension movements
that are of interest. The main movements generally considered are (see Figure 8.2)
the vertical, lateral and the longitudinal movements, i.e. the roll ðjÞ, yaw ðyÞ
and pitch ðqÞ of the various elements constituting the vehicle suspension system
(Figure 8.2).
The purpose of modelling is to better understand this type of suspension in order
to develop effective control strategies based on different constraints on the auto-
motive field. In this section, we are interested in modelling active suspension systems.

8.2.1 Active quarter-car suspension model

The simplest model is the quarter-car suspension model; it only affects the vertical
movement of the vehicle body and the wheel. When looking for comfort, we can
consider only the vertical displacement of the body and not that of the wheel. This
model is called 1 DOF (degree of freedom) quarter-car model. If the handling is

z
ψ Vertical movement

φ
y
θ
x Lateral
movement
Longitudinal
movement Wheel
direction

Figure 8.2 Different types of movements

198 Handbook of vehicle suspension control systems

Zs
ms

ks
bc u

Zu
mu

kt

Zr

Figure 8.3 The active quarter-car suspension with 2 DOF

considered, it should also take into account the vertical movement of the wheel.
This model is called 2 DOF quarter-car suspension model.
Figure 8.3 shows the active quarter-car suspension with 2 DOF and force
generating actuator [11], where zs and zu represent the vertical displacements of the
sprung mass and unsprung mass respectively. zr is the vertical profile of road sur-
face. ms and mu denote the quarter sprung and unsprung masses respectively. ks and
kt are the coefficient of suspension sprung and tyre sprung respectively. bc is the
damming ratio of the damper.
Applying Newton’s second law to the quarter-car model shown in Figure 8.3
gives the following equations of motion [26]:
X
m€x ¼ Fi ð8:1Þ
ms€z s ¼ fk  fb þ u
ð8:2Þ
mu€z u ¼ fk þ fb  ft  u
where:
fk is the force exerted by the spring suspension.
fb is the force exerted by the damper.
ft is the force exerted by the tyre.
u is the control force.
Case of linear model: In the linear models of suspension, the effects of springs and
dampers are assumed linear.
fk ¼ ks ðzs  zu Þ
fb ¼ bc ðz_ s  z_ u Þ ð8:3Þ
ft ¼ kt ðzu  zr Þ
 H? fuzzy control of suspension systems with actuator saturation 199

Thus, the linear model can be given by

ms€z s ¼ ks ðzs  zu Þ  bc ðz_ s  z_ u Þ þ u
ð8:4Þ
mu€z u ¼ ks ðzs  zu Þ þ bc ðz_ s  z_ u Þ  kt ðzu  zr Þ  u
where:
€z s is the vertical acceleration of sprung mass (body).
€z u is the vertical acceleration of unsprung mass (wheel).
z_ s is the sprung mass vertical velocity.
z_ u is the wheel velocity.
zs  zu is the suspension deflection.
zu  zr is the tyre deflection.
If we assume a state vector as
2 3 2 3
x1 zs
6 x2 7 6 zu 7
x¼6 7 6 7
4 x3 5 ¼ 4 z_ s 5 ð8:5Þ
x4 z_ u
a state space representation of this system is given as the following:
x_ ðtÞ ¼ AxðtÞ þ B1 wðtÞ þ B2 uðtÞ ð8:6Þ
where:
0 1
0 0 1 0
B 0 0 0 1 C
B C
B ks ks bc bc C
A ¼B
B  ms  C
B ms ms ms C C
@ ks ðk s þ k t Þ bc bc A
 
mu mu mu mu
2 3
2 3 0
0
6 0 7
6 0 7 6 7
6 7 6 1 7
B1 ¼ 6 7
6 0 7, B2 ¼ 6 6 ms 7
7
4 kt 5 6 7
4 1 5
mu 
mu
Case of nonlinear model: We consider the nonlinearity effect of suspension spring
[7]; the forces are given by
fk ¼ ks ðzs  zu Þ þ ps ðzs  zu Þ3
ð8:7Þ
fb ¼ bc ðz_ s  z_ u Þ
Substituting in (8.2), the resulting system is a nonlinear system given by the
following equations:
ms€z s ¼ ks ðzs  zu Þ  ps ðzs  zu Þ3  bc ðz_ s  z_ u Þ þ u
ð8:8Þ
mu€z u ¼ ks ðzs  zu Þ þ ps ðzs  zu Þ3 þ bc ðz_ s  z_ u Þ  kt ðzu  zr Þ  u
with ps ¼ 10% ks [17].
200 Handbook of vehicle suspension control systems

8.2.2 Half-car suspension model

The half-car suspension model is used to take into account, in addition to vertical
movements, movements of roll or pitch of the body. This model represents a side
view of a vehicle.
Figure 8.4 represents a simplified model of the bicycle type in which the pitch
movement is included [26]. Assuming that the pitch angle q (Figure 3.4) is small,
we have:

zsf ¼ z  a sin q  z  aq
ð8:9Þ
zsr ¼ z þ b sin q  z þ bq

and the application of second Newton’s law, the equations of motion are given by
( P
m€x ¼ Fi
P ð8:10Þ
€ ¼ Mi
Jq
€
ms€z sf ¼ ms€z  ms aq
€
ms€z sr ¼ ms€z þ ms bq
€
ms€z sf ¼ Ff þ Fr  ms aq
€
ms€z sr ¼ Ff þ Fr þ ms bq
ð8:11Þ
muf €z uf ¼ Ff  ftf
mur€z ur ¼ Fr  ftr
ms€z ¼ Ff þ Fr
€ ¼ a  Ff þ b  Fr
Jq

Fr
Ff
b ms

z θ Reference
br a
kr zuf kf bf
zuf
ur uf
mur
–ur –uf muf

ktr zrr ktf zrf

Figure 8.4 Half-car suspension

 H? fuzzy control of suspension systems with actuator saturation 201

The mathematical formulas of forces applied to the sprung mass are given by
Ff ¼ fkf  fbf þ uf
ð8:12Þ
Fr ¼ fkr  fbr þ ur
Case of linear model: In this case, the forces are given by
fki ¼ ki ðzsi  zui Þ
fbi ¼ bi ðz_ si  z_ ui Þ ð8:13Þ
fti ¼ kti ðzui  zri Þ, i ¼ f f, rg
We replace these forces in (8.10) and obtain:
   
€z sf ¼ akf zsf  zuf  abf z_ sf  z_ uf þ bkr ðzsr  zur Þ
þ bbr ðz_ sr  z_ ur Þ þ auf  bur
   
€z sr ¼ bkf zsf  zuf þ bbf z_ sf  z_ uf
þ gkr ðzsr  zur Þ  gbr ðz_ sr  z_ ur Þ  buf
      1
€z uf ¼ kf zsf  zuf þ bf z_ sf  z_ uf  ktf zuf  zrf  uf
muf ð8:14Þ
1
€z ur ¼ kr ðzsr  zur Þ þ br ðz_ sr  z_ ur Þ  ktr ðzur  zrr Þ  uur
mur
   
€ ¼ akf zsf  zuf þ abf z_ sf  z_ uf  bkr ðzsr  zur Þ
q
J J J
bbr
 ðz_ sr  z_ ur Þ  auf þ bur
J
2 2
with a ¼ m1s þ aJ , b ¼ m1s  ab 1 b
J , g ¼ ms þ J , and

€z sf , €z sr are the vertical acceleration of sprung mass at front and rear sides
respectively.
€z uf , €z ur are the vertical acceleration of unsprung mass at front and rear sides
respectively.
zsr is the movement of the sprung mass at rear side.
z_ sr is the velocity of the sprung mass at rear side.
 
z  zuf is the suspension deflection at front side.
 sf 
zuf  zrf is the tyre deflection at front side.
ðzur  zrr Þ is the tyre deflection at rear side.
zuf is the movement of the unsprung mass at front side.
z_ uf is the velocity of the unsprung mass at front side.
ðzsr  zur Þ is the suspension deflection at rear side.
zur is the movement of the unsprung mass at rear side.
z_ ur is the velocity of the unsprung mass at rear side.
z_ is the vertical velocity of car body.
q_ is the angular velocity (pitch) of the body.
202 Handbook of vehicle suspension control systems

uf and ur are the control inputs.

ms is the half sprung mass.
mf , mr are the unsprung mass at front and rear sides respectively.
Jq is the moment of inertia of pitch.
a, b are the distance from the front and rear axle respectively, in the centre of
gravity of the vehicle.
kf , kr are spring stiffness of front and rear suspensions.
rf , rr are the stiff anti-roll bar at front and rear sides.
bf , br are damping coefficient of the front and rear suspensions.
ktf , ktr are the stiffness of the front and rear tyres.
T
Considering that zrf and zrr are the external perturbations, wðtÞ ¼ ½ wf wr ¼
T
½ zrf zrr .
For the state space representation of this system, state variables are chosen as
the following:
2 3 2 3
x 1 ðt Þ zsf  zuf
6 x 2 ðt Þ 7 6 zuf 7
6 7 6 7
6 7 6 7
6 x 3 ðt Þ 7 6 z_ uf 7
6 7 6 7
6 x4 ðtÞ 7 6 ðzsr  zur Þ 7
x ðt Þ ¼ 6 7 ¼6
6 x 5 ðt Þ 7 6
7
7 ð8:15Þ
6 7 6 zur 7
6 7 6 7
6 x 6 ðt Þ 7 6 z_ ur 7
6 7 6 7
4 x 7 ðt Þ 5 4 z_ 5
x 8 ðt Þ q_

Then, a system of eight equations is obtained:

x_ 1 ¼ x3 þ x7  ax8
x_ 2 ¼ x3
kf ktf bf bf abf ktf 1
x_ 3 ¼ x1  x2  x3 þ x7  x8 þ wf  uf
muf muf muf muf muf muf muf
x_ 4 ¼ x6 þ x7  bx8
x_ 5 ¼ x6
kr ktr br br bbr ktr 1
x_ 6 ¼ x4  x5  x6 þ x7 þ x8 þ wr  ur
mur mur mur mur mur mur mur
  ð8:16Þ
kf bf kr br bf þ br
x_ 7 ¼  x1 þ x3  x4 þ x6  x7
ms ms ms ms ms
 
abf  bbr 1 1
þ x8 þ uf þ ur
ms ms ms
 
akf abf bkr bbr abf  bbr
x_ 8 ¼ x1  x3  x4 þ x6 þ x7
J J  J J J
bf a2 þ br b2 a b
 x8  uf þ ur
J J J
 H? fuzzy control of suspension systems with actuator saturation 203

The linear state space representation of the half-car model is of the form:
x_ ðtÞ ¼ AxðtÞ þ B1 wðtÞ þ B2 uðtÞ ð8:17Þ
with the following matrices:
2 3
0 0 1 0 0 0 1 a
6 0 0 0 0 7
6 0 0 1 0 7
6 bf abf 7
6 kf ktf bf 7
6   0 0 0  7
6 muf muf muf m uf m uf 7
6 7
6 0 0 0 0 0 1 1 b 7
6 7
6 7
6 0 0 0 0 0 1 0 0 7
A ¼6 7
6 kr k b b bb 7
6 0 0 0 
tr

r r r 7
6 mur mur mur mur mur 7
6 7
6 7
6 kf bf kr br ðbf þ br Þ ðabf  bbr Þ 7
6 0  0  7
6 ms ms ms ms ms ms 7
6 7
4 ak abf bkr 5
f bbr ðabf  bbr Þ ðbf a2 þ br b2 Þ
0   0 
J J J J J J
2 ktf 3T
0 0 0 0 0 0 0
6 muf 7
B1 ¼46 7
ktr 5
0 0 0 0 0 0 0
mur
2 1 1 a 3T
0 0  0 0 0 
6 muf ms J7
B2 ¼6 7
4 1 1 b 5
0 0 0 0 0 
mur ms J
Case of nonlinear model: We consider the nonlinearity effect of spring, so the
forces are given by
fki ¼ ki ðzsi  zui Þ þ psi ðzsi  zui Þ3
fbi ¼ bi ðz_ si  z_ ui Þ ð8:18Þ
where i ¼ ff , rg.
The system of equations (8.10) becomes:
 
€z sf ¼ a fkf  fbf þ uf þ bðfkr  fbr þ ur Þ
 
€z sr ¼ b fkf  fbf þ uf þ gðfkr  fbr þ ur Þ
1  
€z uf ¼ fkf þ fbf  uf  ftf
muf
1 ð8:19Þ
€z ur ¼ ð fkr þ fbr  ur  ftr Þ
mur
   
€ ¼ 1 a fkf þ fbf  uf þ bðfkr  fbr þ ur Þ
q
J 
€z ¼ fkf  fbf þ uf þ ðfkr  fbr þ ur Þ
204 Handbook of vehicle suspension control systems

8.2.3 Full-car suspension model

Different models can be developed to simulate the behaviour of all suspensions of
the vehicle depending on the complexity and desired realism. A complete model of
7 DOF implying only the vertical movement roll and pitch, because they generate
directly a vertical deflection of suspensions, is considered in the following
paragraph.
We consider in our study a model developed in [18]. This model includes three
complete vehicle movements (roll, pitch of the body and vertical displacements of
each of the four wheels independently). The chassis is assumed to be rigid. Each
suspension connects to a four-wheel independent to one of the four corners of the
chassis. A suspension consists of a spring, a damper and an actuator (Figure 8.5).
Two stabilizer bars, placed at the front and rear, are used to reduce the roll angle of
the body.
Under the following assumptions:
– rigid chassis,
– small displacement around the equilibrium point,
the model proposed by Park and Kim [18] is given by

q ¼ TFr þ Tu
Ms €
ð8:20Þ
Mu€z u ¼ Fs  Ft  u
Fr ¼ Fkr  Fb
ð8:21Þ
Fs ¼ Fks  Fb
By using (8.21) in (8.20), we get:

Ms €
q ¼ T ðFkr  Fb þ uÞ
ð8:22Þ
Mu€z u ¼ Fks þ Fb  Ft  u

Z
2tf
2tr

X C.G

φ y
θ
zs2

1 3
zu2
lf lr
zr2
2 4

Figure 8.5 Full-car suspension

 H? fuzzy control of suspension systems with actuator saturation 205

with the following notations:

q ¼ ½zc , q, j T is the vertical displacement at a gravity centre, the roll angle and
the pitch angle respectively of the sprung mass.
zu ¼ ½zu1 , zu2 , zu3 , zu4 T is the pushing force generated by each actuator.
zs ¼ ½zs1 , zs2 , zs3 , zs4 T is the vertical displacement of each corner of the
sprung mass.
zr ¼ ½zr1 , zr2 , zr3 , zr4 T is the profile of the road under each wheel.
 T
u ¼ uf 1 , uf 2 , ur3 , ur4 is the forces generated by each actuator.
The matrices used in (8.22) are
 T
Ms ¼ diag ms , Jq , Jj
 T
Mu ¼ diag mf , mf , mr , mr
 T
Bs ¼ diag bf , bf , br , br
 T
kss ¼ diag kf , kf , kr , kr
 T
kt ¼ diag ktf , ktf , ktr , ktr
0 1
kf þ rf =2 rf =2 0 0
B r =2 kf þ rf =2 0 0 C
B f C
Ksr ¼ B C
@ 0 0 kr þ rr =2 rr =2 A
0 0 rr =2 kr þ rr =2
0 1
1 1 1 1
@
T ¼ tf tf tr tr A
lf lf lr lr

where the following notations are used:

ms is the sprung mass.
mf and mr are unsprung mass at front and rear sides respectively.
Jq and Jj are the moment of inertia of pitch and roll of the sprung mass
respectively.
lf and lr are the distances from the front and rear axle in the gravity centre of
the vehicle.
tf and tr are half way to the front and rear of each corner of the vehicle.
kf and kr are the spring stiffness of front and rear suspensions.
rf and rr are the stiff anti-roll bar at front and rear sides.
bf and br are the damping coefficients of the front and rear suspensions.
ktf and ktr are the stiffness of the front and rear tyres.
Travel from all corners of the sprung mass needed to calculate the deflection
of each of the four suspensions. The vector zs represents the position of each
206 Handbook of vehicle suspension control systems

frame corner. The body is assumed to be rigid; vectors are connected by the fol-
lowing equation:
8
>
> zsf 1 ¼ zc  tf sinðqÞ  lf sinðjÞ
>
< z ¼ z þ t sinðqÞ  l sinðjÞ
sf 2 c f f
ð8:23Þ
>
> z sr1 ¼ zc  tr sin ð q Þ þ l r sinðjÞ
>
:
zsr2 ¼ zc þ tr sinðqÞ þ lr sinðjÞ
For small displacement angles ðq, jÞ
8
> zsf 1 ¼ zc  tf q  lf j
>
<z ¼ z þ t q  l j
sf 2 c f f
>
> z ¼ zc  tr q þ lr j
: sr1
zsr2 ¼ zc þ tr q þ lr j
0 1 0 1
zsf 1 1 tf lf 0 1
B zsf 2 C B 1 tf zc
lf C
C@ q A
zs ¼ B C B
@ zsr1 A ¼ @ 1 tr lr A
j
zsr2 1 tr lr
zs ¼ T T q ð8:24Þ
Case of linear model: We consider linear effects of springs and dampers as follows:
Fkr ¼ ksr ðzs  zu Þ
Fks ¼ kss ðzs  zu Þ
ð8:25Þ
Fb ¼ bs ðz_ s  z_ u Þ
Ft ¼ ks ðzu  zr Þ
The system (8.20) becomes:

Ms €
q ¼ T ðksr ðzu  zs Þ þ bs ðz_ u  z_ s Þ þ uÞ
ð8:26Þ
Mu€z u ¼ kss ðzs  zu Þ þ bs ðz_ s  z_ u Þ þ kt ðzr  zu Þ  u
By replacing (8.24) in (8.26), we get:

q ¼ TBs T T q_ þ TBs z_ u  Tksr T T q þ Tksr zu þ Tu
Ms €
ð8:27Þ
Mu€z u ¼ Bs T T q_  Bs z_ u þ kss T T q  ðkss þ kt Þzu þ kt zr  u
or in a matrix form as
     
Ms 034 q
€ TBs T T TBs q_
¼
043 Mu €z u Bs T T
Bs z_ u
 T      
Tksr T Tksr q 034 T
þ þ zr þ u
kss T T ðkss þ kt Þ zu kt I44
ð8:28Þ
We define:
z ¼ ½q, zu T ð8:29Þ
 H? fuzzy control of suspension systems with actuator saturation 207

then the system (8.27) can be written as

Mz€z ¼ Bz z_ þ kz z þ kr zr þ ku u ð8:30Þ
with:
  !
Ms 034 TBs T T TBs
Mz ¼ , Bz ¼
043 Mu Bs T T Bs
!  
Tksr T T Tksr 034
Kz ¼ T
, Kr ¼
kss T kt þ kss kt
 
T
Ta ¼
044
Or in state space representation as
_ ¼ AxðtÞ þ B1 wðtÞ þ B2 uðtÞ
xðtÞ ð8:31Þ
with
2 3 2 3
x1 zc
6 x2 7 6 q 7
6 7 6 7
6 x3 7 6 j 7
6 7 6 7
6 x4 7 6 zuf 1 7
6 7 6 7
6 x5 7 6 zuf 2 7
6 7 6 7
6 x6 7 6 zur1 7
6 7 6 7
6 x7 7 6 zur2 7
6 7¼6 7 ð8:32Þ
6 x8 7 6 z_ c 7
6 7 6 7
6 x9 7 6 q_ 7
6 7 6 7
6 x10 7 6 j_ 7
6 7 6 7
6 x11 7 6 z_ uf 1 7
6 7 6 7
6 x12 7 6 z_ uf 2 7
6 7 6 7
4 x13 5 4 z_ ur1 5
x14 z_ ur2

and the following matrix:

   
077 I77 074
A¼ , B1 ¼
Mz1 kz Mz1 Bz Mz1 kr
  ð8:33Þ
074
B2 ¼
Mz1 Ta
Case of nonlinear system: In this case, the following nonlinear forces are considered:

Fkr ¼ ksr ðzs  zu Þ 1 þ 0:1ðzs  zu Þ2

Fks ¼ kss ðzs  zu Þ 1 þ 0:1ðzs  zu Þ2 ð8:34Þ

Fb ¼ bs ðz_ s  z_ u Þ
Ft ¼ ks ðzu  zr Þ
208 Handbook of vehicle suspension control systems

The dynamic model in [18] is obtained by substituting (8.34) in (8.20).

Remark 8.1. Here, we consider the suspension spring. We can take other non-
linearity effects, such as tire spring effect and damper (see [7]).

8.3 Takagi–Sugeno fuzzy model of suspension systems

The T–S fuzzy systems provide an effective representation of complex nonlinear
systems described by a set of linear models. This approach has been extensively
used for stability analysis and control of nonlinear systems [14–16, 20–26].
This approach represents a nonlinear system as follows:
8
> X r
>
> x
_ ð t Þ ¼ mi ðxðtÞÞðAi xðtÞ þ B1i wðtÞ þ B2i sðtÞÞ
>
< i¼1
ð8:35Þ
>
> Xr
>
>
: zðtÞ ¼ mi ðxðtÞÞðCi xðtÞ þ D1i wðtÞ þ D2i sðtÞÞ
i¼1

where xðtÞ 2 Rn is the system state vector, sðtÞ 2 Rm is the control input, zðtÞ 2 Rnz
is the controlled output variable, wðtÞ 2 =2 is the disturbance variable with
=2 ¼ fwðtÞ 2 Rnw jkwk2  w, w
0g and xðtÞ 2 Rq is the decision variable vector
assumed measurable, and the normalized activation function mi ðxðtÞÞ in relation
with the ith sub-model is such that:
8X r
>
< mi ðxðtÞÞ ¼ 1
i¼1 ð8:36Þ
>
:
0  mi ðxðtÞÞ  1 8 i 2 Ir
In general, there are three methods for obtaining the T–S fuzzy model, by
identification method, linearization around different operating points or by trans-
formation of sector nonlinearities. In the first method, the Takagi–Sugeno fuzzy
system structure is first chosen then identification techniques are used to estimate
the model’s parameters. The second method is based on linearization of a nonlinear
system around different operating points. In this method, the T–S fuzzy system can
be given by interpolation of numerous local models with activation functions
chosen in advance. The last method gives an exact representation of a nonlinear
system. This method is based on bounds of nonlinear functions and gives a mini-
mum number of local models. Here, we will use the last method to represent the
suspension system.
The following lemma will be used in the sequel of chapter.
Lemma 8.1 [15]. Let f ðxðtÞÞ:R ! R is a bounded function, and there exist always
two functions, h1 ðxðtÞÞ and h2 ðxðtÞÞ, and two constants a and b such that:
f ðxðtÞÞ ¼ a  h1 ðxðtÞÞ þ b  h2 ðxðtÞÞ ð8:37Þ
 H? fuzzy control of suspension systems with actuator saturation 209

with:
h1 ðxðtÞÞ þ h2 ðxðtÞÞ ¼ 1,
ð8:38Þ
h1 ðxðtÞÞ  0, h2 ðxðtÞÞ  0
In this case, the obtained T–S fuzzy model represents exactly the nonlinear sys-
tem for x 2 Rn with 2nl local models where nl represents the number of local models.
In this section we will describe the T–S fuzzy model representation of active
quarter-car, half-car and full-car suspension systems.

8.3.1 Takagi–Sugeno representation of active quarter-car

suspension
Consider the model of active quarter-car suspension with nonlinearity effect of
suspension spring given by (8.8), and let:
z1 ¼ ðzs  zu Þ ð8:39Þ
If z1 2 ½a, a then z21 2 ½0, a2 and:
   
max z2 ¼ a2 , min z2 ¼ 0
By using Lemma 8.1:

z2 ¼ h1 ðxÞa2 þ h2 ðxÞ0
with:

z2  minðz2 Þ z2
h1 ðxÞ ¼ ¼
maxðz2 Þ  minðz2 Þ a2
ð8:40Þ
maxðz2 Þ  z2 a2  z2
h2 ðxÞ ¼ ¼
maxðz Þ  minðz Þ
2 2 a2

By choosing (8.5), the active quarter-car suspension system can be represented by

the following T–S fuzzy model:
X
2
xðtÞ
_ ¼ mi ðAi xðtÞ þ B1 wðtÞ þ B2 uðtÞÞ ð8:41Þ
i¼1
0 1
0 0 1 0
B C
B 0 0 0 1 C
B C
B ks ð1 þ 0:1a2 Þ 2
ks ð1 þ 0:1a Þ bc bc C
A1 ¼ B
B  C
C
B ms ms ms ms C
B C
@ ks ð1 þ 0:1a2 Þ ðks ð1 þ 0:1a2 Þ þ kt Þ bc bc A
 
mu mu mu mu
210 Handbook of vehicle suspension control systems
0 1
0 0 1 0
B 0
B 0 0 1 C C
B k C
B ks bc bc C
A2 ¼ B  s  C
B ms ms ms ms C
B C
@ ks ðk s þ k t Þ bc bc A
 
mu mu mu mu
2 3 2 3
0 0
6 0 7 6 0 7
6 7 6 7
6 1 7 6 7
6
B2 ¼ 6 7, B1 ¼ 6 0 7
7
6 ms 7 6 7
4 kt 5
4 1 5
 mu
mu

where wðtÞ ¼ zr ðtÞ and uðtÞ is the actuating force.

8.3.2 Takagi–Sugeno representation of active half-car suspension

Similar to the case of quarter-car suspension system, let us define:

z1i ¼ ðzsi  zui Þ, i ¼ ff , r g

 
z2i  min z2i
Mi1 ðxÞ¼
maxðz2i Þ  minðz2i Þ
 
2
max z2i  z2i ð8:42Þ
Mi ðxÞ ¼
maxðz2i Þ  minðz2i Þ

i ¼ ff , rg

The active half-car suspension system can be represented by the following T–S
fuzzy model:

X
4
x_ ðtÞ ¼ mi ðAi xðtÞ þ B1 wðtÞ þ B2 uðtÞÞ ð8:43Þ
i¼1

m1 ðtÞ ¼ Mf1 ðxðtÞÞMr1 ðxðtÞÞ

m2 ðtÞ ¼ Mf1 ðxðtÞÞMr2 ðxðtÞÞ

m3 ðtÞ ¼ Mf2 ðxðtÞÞMr1 ðxðtÞÞ

m4 ðtÞ ¼ Mf2 ðxðtÞÞMr2 ðxðtÞÞ

 H? fuzzy control of suspension systems with actuator saturation 211

with:
2 3
0 0 1 0 0 0 1 a
6 7
6 0 0 1 0 0 0 1 0 7
7
6
6 kf Sf ktf bf bf 7
abf
6   0 0 0  7
6 m muf muf muf 7
muf
6 uf 7
6 7
6 0 0 0 0 0 1 1 b 7
6 7
6 7
6 7
A1 ¼6 0 0 0 0 0 1 0 0 7
6 ktr br br bbr 7
6 0 0 0 0 7
6   7
6 kS bf kr Sr m ur mur mur mur 7
6 f f 0  7
6
6 ms ms ms br ðbf þ br Þ ðabf  bbr Þ 77
6 0  7
6 akf Sf abf bkr Sr ms ms ms 7
6 0   7
4 J 2 2 5
J J bbr ðabf  bbr Þ ðb a þ b b Þ
f r
0 
J J J
2 3
0 0 1 0 0 0 1 a
6 7
6 0 0 1 0 0 0 1 0 7
6 7
6 k f Sf ktf bf bf abf7
6   0 0 0  7
6 m muf muf muf muf7
6 uf 7
6 7
6 0 0 0 0 0 1 1 b 7
6 7
6 7
6 7
A2 ¼6 0 0 0 0 0 1 0 0 7
6 7
6 0
6 0 0 0  ktr 
br br bbr 7
7
6 kS bf kr mur mur mur mur 7
6 f f 0  7
6 br ðbf þ br Þ ðabf  bbr Þ 7
6 ms ms ms 0  7
6 7
6 akf Sf abf bkr ms ms ms 7
6 0   7
4 J J J bbr ðabf  bbr Þ 2
ðb a þ b b Þ2 5
f r
0 
J J J
2 3
0 0 1 0 0 0 1 a
6 7
6 0 0 1 0 0 0 1 0 7
6 7
6 k f Sf ktf bf bf abf 7
6   0 0 0  7
6 m muf muf muf muf 7
6 uf 7
6 7
6 0 0 0 0 0 1 1 b 7
6 7
6 7
6 7
A3 ¼6 0 0 0 0 0 1 0 0 7
6 ktr br br bbr 7
6 0 0 0 0 7
6   7
6 k bf k r Sr m ur mur mur mur 7
6 f 0  7
6 br ðbf þ br Þ ðabf  bbr Þ 7
6 ms ms ms 0  7
6 7
6 akf abf bkr Sr ms ms ms 7
6 0   7
4 J J J bbr ðabf  bbr Þ 2
ðb a þ b b Þ2 5
f r
0 
J J J
212 Handbook of vehicle suspension control systems
2 3
0 0 1 0 0 0 1 a
6 7
6 0 0 1 0 0 0 1 0 7
6 7
6 kf Sf ktf bf bf abf 7
6   0 0 0  7
6m m muf muf muf 7
6 uf uf 7
6 7
6 0 0 0 0 0 1 1 b 7
6 7
6 7
6 7
A4 ¼ 6 0 0 0 0 0 1 0 0 7
6 k b b bbr 7
6 0 0 0 0  tr  r r 7
6 7
6 k bf kr m ur m ur m ur mur 7
6 f 0  7
6 br ðbf þ br Þ ðabf  bbr Þ 7
6 ms ms ms 0  7
6 7
6 akf abf bkr m s ms ms 7
6 0   7
4 J J J bbr ðabf  bbr Þ 2
ðb a þ b b Þ2 5
f r
0 
J J J

2 ktf 3t
0 0 0 0 0 0 0
6 muf 7
B1 ¼6
4
7
5
ktr
0 0 0 0 0 0 0
mur
2 1 1 a 3t
0 0  0 0 0 
6 muf ms J7
B2 ¼6
4
7
1 1 b 5
0 0 0 0 0 
mur ms J

with Si ¼ ð1 þ psi a2 Þ, i ¼ ff , rg.

If in addition, one takes psf ¼ 10%kf , psr ¼ 10%kr , it follows that
Si ¼ ð1 þ 0:1a2 Þ, i ¼ ff , rg.

8.3.3 Takagi–Sugeno representation of active full-car suspension

The active full-car suspension system can be represented by the following T–S
fuzzy model:
X
16
x_ ðtÞ ¼ mi ðAi xðtÞ þ B1 wðtÞ þ B2 uðtÞÞ ð8:44Þ
i¼1

m1 ðtÞ ¼ Mf11ðxðtÞÞMf12 ðxðtÞÞMr1

1 1
ðxðtÞÞMr2 ðxðtÞÞ

m2 ðtÞ ¼ Mf11 ðxðtÞÞMf12 ðxðtÞÞMr1

1 2
ðxðtÞÞMr2 ðxðtÞÞ

m3 ðtÞ ¼ Mf11 ðxðtÞÞMf12 ðxðtÞÞMr1

2 1
ðxðtÞÞMr2 ðxðtÞÞ

m4 ðtÞ ¼ Mf11 ðxðtÞÞMf12 ðxðtÞÞMr1

2 2
ðxðtÞÞMr2 ðxðtÞÞ
 H? fuzzy control of suspension systems with actuator saturation 213

m5 ðtÞ ¼ Mf11 ðxðtÞÞMf22 ðxðtÞÞMr1

1 1
ðxðtÞÞMr2 ðxðtÞÞ

m6 ðtÞ ¼ Mf11 ðxðtÞÞMf22 ðxðtÞÞMr1

1 2
ðxðtÞÞMr2 ðxðtÞÞ

m7 ðtÞ ¼ Mf11 ðxðtÞÞMf22 ðxðtÞÞMr1

2 1
ðxðtÞÞMr2 ðxðtÞÞ

m8 ðtÞ ¼ Mf11 ðxðtÞÞMf22 ðxðtÞÞMr1

2 2
ðxðtÞÞMr2 ðxðtÞÞ

m9 ðtÞ ¼ Mf21 ðxðtÞÞMf12 ðxðtÞÞMr1

2 2
ðxðtÞÞMr2 ðxðtÞÞ

m10 ðtÞ ¼ Mf21 ðxðtÞÞMf12 ðxðtÞÞMr1

1 1
ðxðtÞÞMr2 ðxðtÞÞ

m11 ðtÞ ¼ Mf21 ðxðtÞÞMf12 ðxðtÞÞMr1

1 2
ðxðtÞÞMr2 ðxðtÞÞ

m12 ðtÞ ¼ Mf21 ðxðtÞÞMf12 ðxðtÞÞMr1

2 1
ðxðtÞÞMr2 ðxðtÞÞ

m13 ðtÞ ¼ Mf21 ðxðtÞÞMf22 ðxðtÞÞMr1

2 2
ðxðtÞÞMr2 ðxðtÞÞ

m14 ðtÞ ¼ Mf21 ðxðtÞÞMf22 ðxðtÞÞMr1

1 1
ðxðtÞÞMr2 ðxðtÞÞ

m15 ðtÞ ¼ Mf21 ðxðtÞÞMf22 ðxðtÞÞMr1

1 2
ðxðtÞÞMr2 ðxðtÞÞ

m16 ðtÞ ¼ Mf21 ðxðtÞÞMf22 ðxðtÞÞMr1

2 1
ðxðtÞÞMr2 ðxðtÞÞ
z1i ¼ ðzsi  zui Þ, i ¼ ff , r g ð8:45Þ

z2ij  min z2ij

Mij1 ðxÞ ¼

max z2ij  min z2ij

max z2ij  z2ij ð8:46Þ

Mij2 ðxÞ ¼

max z2ij  min z2ij

i ¼ ff , rg, j ¼ f1, 2g
   
074 074
B1 ¼ , B2 ¼
Mz1 kr Mz1 Ta
 
077 I77
A1 ¼
Mz1 kz1 Mz1 Bz
!
Tksr T T Tksr Ek1
Kz1 ¼
kss Ek1 T T kt þ kss Ek1
       
Ek1 ¼ diag 1 þ 0:1a2f 1 þ 0:1a2f 1 þ 0:1a2r 1 þ 0:1a2r
 
077 I77
A2 ¼
Mz1 kz2 Mz1 Bz
214 Handbook of vehicle suspension control systems
!
Tksr T T Tksr Ek2
Kz2 ¼
kss Ek2 T T kt þ kss Ek2
      
Ek2 ¼ diag 1 þ a2f 1 þ a2f 1 þ a2r 1
 
077 I77
A3 ¼
Mz1 kz3 Mz1 Bz
!
Tksr T T Tksr Ek3
Kz2 ¼
kss Ek3 T T kt þ kss Ek3
     
Ek3 ¼ diag 1 þ 0:1a2f 1 þ 0:1a2f 1 1 þ 0:1a2r
 
077 I77
A4 ¼
Mz1 kz4 Mz1 Bz
!
Tksr T T Tksr Ek4
Kz4 ¼
kss Ek4 T T kt þ kss Ek4
   
Ek4 ¼ diag 1 þ a2f 1 þ a2f 1 1Þ
 
077 I77
A5 ¼
Mz1 kz5 Mz1 Bz
!
Tksr T T Tksr Ek5
Kz5 ¼
kss Ek5 T T kt þ kss Ek5
     
Ek5 ¼ diag 1 þ 0:1a2f 1 1 þ 0:1a2r 1 þ 0:1a2r
 
077 I77
A5 ¼
Mz1 kz5 Mz1 Bz
!
Tksr T T Tksr Ek5
Kz5 ¼
kss Ek5 T T kt þ kss Ek5
     
Ek5 ¼ diag 1 þ 0:1a2f 1 1 þ 0:1a2r 1 þ 0:1a2r
 
077 I77
A6 ¼
Mz1 kz6 Mz1 Bz
!
Tksr T T Tksr Ek6
Kz6 ¼
kss Ek6 T T kt þ kss Ek6
    
Ek6 ¼ diag 1 þ 0:1a2f 1 1 þ 0:1a2r 1
 H? fuzzy control of suspension systems with actuator saturation 215
 
077 I77
A7 ¼
Mz1 kz7 Mz1 Bz
!
Tksr T T Tksr Ek7
Kz7 ¼
kss Ek7 T T kt þ kss Ek7
   
Ek7 ¼ diag 1 þ 0:1a2f 1 1 1 þ 0:1a2r
 
077 I77
A8 ¼
Mz1 kz8 Mz1 Bz
!
Tksr T T Tksr Ek8
Kz8 ¼
kss Ek8 T T kt þ kss Ek8
  
Ek8 ¼ diag 1 þ 0:1a2f 1 1 1
 
077 I77
A9 ¼
Mz1 kz9 Mz1 Bz
!
Tksr T T Tksr Ek9
Kz9 ¼
kss Ek9 T T kt þ kss Ek9
      
Ek9 ¼ diag 1 1 þ 0:1a2r 1 þ 0:1a2r 1 þ 0:1a2r
 
077 I77
A10 ¼
Mz1 kz10 Mz1 Bz
!
Tksr T T Tksr Ek10
Kz10 ¼
kss Ek10 T T kt þ kss Ek10
     
Ek10 ¼ diag 1 1 þ 0:1a2r 1 þ 0:1a2r 1
 
077 I77
A11 ¼
Mz kz11 Mz1 Bz
1

!
Tksr T T Tksr Ek11
Kz11 ¼
kss Ek11 T T kt þ kss Ek11
    
Ek11 ¼ diag 1 1 þ 0:1a2r 1 1 þ 0:1a2r
 
077 I77
A12 ¼
Mz1 kz12 Mz1 Bz
 
Tksr T T Tksr Ek12
Kz12 ¼
kss Ek12 T T kt þ kss Ek12
216 Handbook of vehicle suspension control systems
   
Ek12 ¼ diag 1 1 þ 0:1a2r 1 1
 
077 I77
A13 ¼
Mz1 kz13 Mz1 Bz
!
Tksr T T Tksr Ek13
Kz13 ¼
kss Ek13 T T kt þ kss Ek13
    
Ek13 ¼ diag 1 1 1 þ 0:1a2r 1 þ 0:1a2r
 
077 I77
A14 ¼
Mz1 kz14 Mz1 Bz
!
Tksr T T Tksr Ek14
Kz14 ¼
kss Ek14 T T kt þ kss Ek14
   
Ek14 ¼ diag 1 1 1 þ 0:1a2r 1
 
077 I77
A15 ¼
Mz kz15 Mz1 Bz
1

!
Tksr T T Tksr Ek15
Kz15 ¼
kss Ek15 T T kt þ kss Ek15
  
Ek15 ¼ diag 1 1 1 1 þ 0:1a2r
 
077 I77
A16 ¼
Mz1 kz16 Mz1 Bz
!
Tksr T T Tksr Ek16
Kz16 ¼
kss Ek16 T T kt þ kss Ek16

Ek15 ¼ diag ð1 1 1 1Þ

8.4 Validation of Takagi–Sugeno fuzzy model

In this section, we present the validation of T–S fuzzy model representation of the
quarter-car suspension system.

8.4.1 Simulation parameters

The system parameters are chosen as follows [7]:

ms ¼ 250 kg, mu ¼ 30 kg, ks ¼ 15,000 N=m, kt ¼ 150,000 N=m,

bc ¼ 1,000 Ns=m and ps ¼ 0:1 ks .
 H? fuzzy control of suspension systems with actuator saturation 217

8.4.2 Validation of Takagi–Sugeno fuzzy model

The open-loop T–S fuzzy model and vehicle suspension plant responses with
external disturbance (Figure 8.6) are shown in Figures 8.7–8.13.
These figures show that the T–S fuzzy system has the same behaviour as the
nonlinear system. This means the satisfactory approximation ability of the T–S
fuzzy model.

0.4

0.35

0.3

0.25

0.2
w(t)

0.15

0.1

0.05

–0.05

–0.1
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.6 The external disturbance wðtÞ

1
0.9
M2
0.8 M1
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
–0.5 –0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5

Figure 8.7 Membership functions of the T–S fuzzy model

218 Handbook of vehicle suspension control systems

0.6

0.5

0.4

0.3
x1(t)

0.2

0.1

–0.1

–0.2
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.8 The vertical displacements of the sprung mass x1ðtÞ (solid curve:
nonlinear plant; star curve: T–S fuzzy model)

0.5

0.4

0.3

0.2
x2(t)

0.1

–0.1

–0.2
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.9 The vertical displacements of the unsprung mass x2 ðtÞ (solid curve:
nonlinear plant; star curve: T–S fuzzy model)
 H? fuzzy control of suspension systems with actuator saturation 219

2.5

1.5

0.5
x3(t)

–0.5

–1

–1.5

–2

–2.5
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.10 Speed of the sprung mass x3ðtÞ (solid curve: nonlinear plant; star
curve: T–S fuzzy model)

20

15

10

5
x4(t)

–5

–10

–15

–20
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.11 Velocity of the unsprung mass x4ðtÞ (solid curve: nonlinear plant;
star curve: T–S fuzzy model)
220 Handbook of vehicle suspension control systems

0.5

0.4

0.3

0.2

0.1
z1 = x1 – x2

–0.1

–0.2

–0.3

–0.4

–0.5
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.12 Suspension deflection (solid curve: nonlinear plant; star curve: T–S
fuzzy model)

0.4

0.3

0.2

0.1
x2 (t) – w(t)

–0.1

–0.2

–0.3

–0.4
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.13 The wheel deflection (solid curve: nonlinear plant; star curve: T–S
fuzzy model)
 H? fuzzy control of suspension systems with actuator saturation 221

8.5 Actuator saturation

The active suspension comprises a force-generating actuator placed between the
sprung and unsprung masses. In practice, this force is limited. For example, elec-
trohydraulic actuators [5, 14] are more used in active suspension (Figure 8.2) with
an input voltage that is limited in practice.
The actuator saturation can degrade the performance of the closed-loop system
and often make the stable closed-loop system unstable.

Example 8.1. Consider the following linear system:

x_ ¼ AxðtÞ þ BuðtÞ ð8:47Þ
with
   
0 1 0
A¼ , B¼
1 0 1
The open-loop system ðx_ ¼ AxðtÞÞ is not stable, its eigenvalues are eig fAg ¼
f1; 1g.
We consider the linear state feedback defined by
uðtÞ ¼ KxðtÞ ð8:48Þ
with K ¼ ½ 13 7 , the linear control (8.48) can stabilize the linear system (8.48).
The eigenvalues of the closed-loop system are eig fA þ BK g ¼ f3; 4g. This
means that the origin is asymptotically stable.
We suppose that the system (8.26) is subject to the actuator saturation with the
following constraints:
5  satðuðtÞÞ  5 ð8:49Þ
The closed-loop system composed of (8.47)–(8.49) is given by
x_ ðtÞ ¼ AxðtÞ þ BsatðuðtÞÞ ð8:50Þ
where:
8
< 5 Kx  5
satðuðtÞÞ ¼ Kx 5  Kx  5 ð8:51Þ
:
5 Kx  5
In the presence of saturation, undesirable equilibrium points at:

xe1 ¼ ½ 5 0 T and xe2 ¼ ½ 5 0 T ðsee Figure 8:15Þ

Figure 8.14 illustrates the trajectory of the closed-loop system in the state
space with different initial condition:
xe1 ¼ ½ 3 2 T , xe2 ¼ ½ 1 3 T , xe3 ¼ ½ 3 3 T , xe4 ¼ ½ 2 3 T
This figure illustrates that in the presence of saturation, there is initial condition
points that are not stable as x0 ¼ ½ 2 3 T .
222 Handbook of vehicle suspension control systems

0
x2

–1

–2

–3

–4

–5
–5 –4 –3 –2 –1 0 1 2 3 4 5
x1

Figure 8.14 The trajectory of closed-loop system in state space with different
initial conditions

sat(ui,ui)

u
umax

Linear
–umin umax ui

Saturation Saturation

–umin

Figure 8.15 Saturation function

Thus, the implementation of control laws designed without taking into account
the saturation effect may have undesirable consequences on the system behaviour.
This problem has been receiving increasing attention for control of both linear
[27–33] and nonlinear systems [17, 2–23, 34–36]. Generally, the saturation limits are
avoided by designing low gain control laws and considering a bounded set of initial
 H? fuzzy control of suspension systems with actuator saturation 223

system states [21–22]. However, this method often gives low levels of performance
[21]. Alternatively, the problem is dealt with by estimating the domain of attraction
of the closed-loop system in the presence of actuator saturation [17, 21, 23, 33–36].
The saturation function is defined by
2 3
satðu1 Þ
6 .. 7
satðuÞ ¼ 4 . 5 ð8:52Þ
satðum Þ

8.5.1 Types of saturation

In the literature, there are two types of actuator saturations:
– Rate actuator saturations: In this type of saturation, the actuator is subject to
rate constraint as [34]
Du i  DuðtÞ  Du i ð8:53Þ
– Amplitude actuator saturations: In this type of saturation, the actuator is sub-
ject to amplitude constraint (see Figure 8.15) as
u i  uðtÞ  u i ð8:54Þ
 
In most cases, the saturation of amplitude is symmetric u i ¼ u i .
In this case, each component of sðtÞ, the saturation function, can be repre-
sented by
8
<u i ,
> if u i  uðtÞ
satðui , u i Þ ¼ ui ðtÞ if u i  uðtÞ  u i ð8:55Þ
>
:
u i if uðtÞ  u i
where:
u i and ui denote the ith element of u and u respectively.
u 2 Rm denotes the saturation level, u i and ui denote the ith element of u and uðtÞ.

8.5.2 Modelling of saturation effect

In most cases, the saturation nonlinearity term is transformed into dead zone non-
linearity [34] or it is represented by a polytopic model [17, 20, 23, 21, 36].
To model the saturation effect, we will use the polytopic representation.
Let the dead zone function of Figure 8.16:
8
<ui  u i if ui > u i
yðui Þ ¼ 0 if u i  ui  u i i 2 Im ð8:56Þ
:
ui þ ui if ui  u i
The saturation defined by (8.55) is converted into dead zone according to the
equation:
satðui Þ ¼ ui  yðui Þ ð8:57Þ
The following lemma will be used in the development of the results of this work.
224 Handbook of vehicle suspension control systems

ψ (ui)

–u–i ui

Figure 8.16 The dead zone function

Lemma 8.2 [29]. Let fðaÞ be a nonlinear operator that consists of m normalized
dead zone functions. Let j vi j < u 2 <m with i 2 Im , there is a matrix L ¼
diag ðl1 , l2 , . . . , lm Þ, li 2 ½ 1 0 , i 2 Im such that:
fðaÞ ¼ ðIm  LÞðu  vÞ ð8:58Þ
with:
v ¼ HxðtÞ, H 2 <mm
As the matrix L belongs to a polytope consisting of 2m vertices defined by
E ¼ diag ðl1 , l2 , . . . . . . :, lm Þ, li 2 ½ 1 0 , i 2 Im ð8:59Þ
the following lemma is derived directly from (8.57) and (8.58).
Lemma 8.3 [21]. Let E be the set of m  m diagonal matrices whose diagonal ele-
ments are 1 or 0. Suppose that jvi j  u i for all i 2 Im , where vi and ui denote

the element of nðtÞ 2 Rm and uðtÞ 2 Rm respectively. If xðtÞ 2 \rj¼1 = Hj for
xðtÞ 2 Rn , then:
8
X2m  
>
>
>
> satðuðtÞ, u Þ ¼ as Es uðtÞ þ E s nðtÞ
>
>
>
> s¼1
> 2m
< X
as ¼ 1, 0  as  1 ð8:60Þ
>
>
>
> i¼1
>
> Xr
>
> n ð t Þ ¼ mj ðxðtÞÞ Hj xðtÞ
>
:
j¼1
  
= Hj ¼ xðtÞ 2 Rn j jhji xj  u i ð8:61Þ
Es denotes all elements of E, E s ¼ I  Es , Hj is m  n matrix and hji is the ith row
of the matrix Hj .
Thus, if m ¼ 2, one has
       
0 0 1 0 0 0 1 0
E 2 co , , ,
0 0 0 0 0 1 0 1

8.5.3 Saturated control and constrained control

Generally, the design of control laws in the presence of actuator saturation is treated
either by synthesizing a constraint control law or by designing a saturated
 H? fuzzy control of suspension systems with actuator saturation 225

control law. In the first approach, the control law is designed so that the actuator is
never saturated and the problem is treated by determining a set of initial conditions
of the system states that prevents saturation [21–22, 36]. Consequently, the control
input signal remains in its linear region (see Figure 8.16) and one has
u i  uðtÞ  u i and uðtÞ ¼ satðuðtÞÞ ð8:62Þ

However in the second approach, the actuator may saturate and the problem is
treated by the estimation of an attraction domain inside which any system initi-
alization does not generate instability in the presence of saturation [17, 20, 21, 23,
35–36]. In this case, a model in the form of (8.60) that represents the nonlinear
effect of saturation is necessary for the analysis and the synthesis of the control law.

8.6 Quadratic stabilization of Takagi–Sugeno fuzzy model

The simple structure of the T–S fuzzy model as interpolation of local linear models
enables many researchers to exploit it in the stability analysis and stabilization of
nonlinear systems. Often the study of stability and stabilization uses the Lyapunov
theory, in particular the second Lyapunov method. Stability properties are then
derived by solving a set of LMIs. The LMIs [25], if they are proved to admit a
solution, can be solved using tools from the field of convex optimization.

8.6.1 Convex analysis and linear matrix inequalities

A strict (non-strict respectively) LMI has the form [25]:
X
n
FðxÞ ¼ F0 þ xi Fi > 0 ðresp: 0Þ ð8:63Þ
i¼1

with:
P0 and Pi , i 2 I n 2 Rpp 2 Rpp are a family of given symmetric matrices
and x ¼ ðx1 , x2 , . . . , xn ÞT 2 Rn . The constraint FðxÞ > 0 is called convex
constraint LMI.
Among the existing techniques for transforming nonlinear matrix inequalities
optimization problem to an LMI are as follows:
Schur complement: It transforms convex nonlinearities in LMIs.
Lemma 8.4 [25]. Consider the three matrices QðxÞ ¼ QðxÞT, RðxÞ ¼ RðxÞT and SðxÞ
affine in the variable x, the following LMIs are equivalent:
RðxÞ > 0, QðxÞ  SðxÞRðxÞ1 S T ðxÞ > 0
!
QðxÞ SðxÞ ð8:64Þ
T
>0
SðxÞ RðxÞ

Example 8.2. The standard constraint:

kH ðxÞk  1 ð8:65Þ
226 Handbook of vehicle suspension control systems

where H ðxÞ 2 Rnm is affine in x, is equivalent to:

I  H ðxÞH T ðxÞ
0 ð8:66Þ
The transformation of the LMI constraint by the Schur complement is given by
 
I ðÞ

0 ð8:67Þ
H T ðx Þ I
Property of congruence: W is a symmetric positive definite matrix with appropriate
dimension and X is a nonsingular matrix with compatible dimensions.
If W
0, then:

X T WX
0 ð8:68Þ

Example 8.3. Let the following matrix:

2 3
W11 W12    W1n
6 
6 W22    W2n 7 7
6 .
6 . .. .. .. 7
70 ð8:69Þ
4 . . . . 5
 . . .    W12
The following holds also:
2 3T 2 W W12  W1r
32
m1 I
3
m1 I 11
6m I 7 6 W22
76 7
W2r 76 m2 I 7 X
6 2 7 6   r Xr
6 . 7 6 76
6 . 7¼
7
mi 2 Wii þ mi mj Wij < 0
6 . 7 6 . .. .. .. 7
4 . 5 6 4 .
. . . . 7 6 . 7
54 . 5 i¼1 ij
mr I  ...    W1r mr I
ð8:70Þ
LMI approaches are very attractive in the stability analysis of multiple models.
Indeed, the optimization problems in terms of LMIs are convex optimization pro-
blems. Convexity is a very important property and has two major advantages:
● There is not a local solution of the objective function to be optimized; the
obtained solution corresponds to a unique global solution.
● The computational cost for finding a solution is reasonable.
Note that if a problem is not convex, the optimal solution obtained from the
convex formulation is not necessarily optimal for this problem.

8.6.2 Stability in the sense of Lyapunov

The principle of stability according to Lyapunov is based on the dynamic behaviour
of the system in terms of its total energy. This energy is usually represented by a
function V ðxðtÞÞ of the system state x. The sign of this function and its time
 H? fuzzy control of suspension systems with actuator saturation 227

derivative in some neighbourhood of the equilibrium point provides information on

system stability. The main result shows that for an autonomous system with con-
tinuous time, the origin is a globally asymptotically stable equilibrium point if there
exists a positive function satisfying the following conditions:
8
< iÞ V ðxðtÞÞ
0
iiÞ V_ ðxðtÞÞ  0 ð8:71Þ
:
iiiÞ V ð0Þ ¼ 0 and V ð1Þ ! 1
The quadratic functions class plays a very important role in the study of sta-
bility and stabilization of controllers and/or observers [37–48]. In this work, the
quadratic function is of the form:
V ðxðtÞÞ ¼ xT ðtÞPxðtÞ ð8:72Þ
nn
for a symmetric positive definite matrix P 2 R .
Despite the success that has known quadratic approach, obtaining stable condi-
tions seems, for once, very conservative. Efforts to introduce menus are relaxations of
the constraints that the results reported in [39, 49].

8.6.3 Attraction region

The notion of the Lyapunov function is related to the region of attraction, defined
as the largest set in the state space of the region of attraction in which any trajectory
that begins in this region converge to the origin. If a system is globally asympto-
tically stable, then the region of attraction of the origin is the entire state space. The
exact determination of this region by the analytical method is a difficult task if not
impossible [29]. However, it is possible to determine regions of asymptotic stability
using Lyapunov domain and positively invariant set.
Lemma 8.4 [7]. For a constant r > 0, we define the set:
WðxÞ ¼ fx 2 <n jV ðxÞ  rg ð8:73Þ
The set WðxÞ is contractively invariant (Figure 8.17) if:
V_ ðxÞ < 0, 8 x 2 WðxÞjf0g ð8:74Þ
Therefore, if a set WðxÞ is contractively invariant, it is inside the domain of
attraction.
Note that if we choose the quadratic Lyapunov function (8.72), the set (8.73) is
an ellipsoid defined by
 
eðP, rÞ ¼ x 2 Rn xT Px  1 ð8:75Þ
Thus, using Lemma 8.4, an ellipsoid (8.75) is said to be contractively invariant set if:

V_ ðxÞ < 0, 8x 2 eðP, 1Þf0g ð8:76Þ
Therefore, if an ellipsoid is contractively invariant, it is inside the domain of
attraction [21–22].
228 Handbook of vehicle suspension control systems

x1

x2

x3

Figure 8.17 Example of contractively invariant set

 
Lemma 8.5 [21]. An ellipsoid eðP, rÞ is inside \rj¼1 = Hj if and only if for all
i, j 2 Ir2
 T
hji ðPÞ1 hji  u 2i ð8:77Þ

8.6.4 Quadratic stabilization via PDC control

In this work, the controller is a nonlinear state feedback which shares the same
activation functions of the T–S fuzzy model (8.1) and has following form:
X
r
u ðt Þ ¼ mj ðxðtÞÞKj xðtÞ ð8:78Þ
j¼1

where Kj 2 Rmn is the local controller matrix to be determined.

Without saturation and external perturbation, the closed-loop system com-
posed of (8.35) and (8.78) is described by
(  
x_ ðtÞ ¼ Am þ B2m Km xðtÞ
  ð8:79Þ
zðtÞ ¼ C1m þ D2m Km xðtÞ
X
r X
r
Am ¼ mi ðxðtÞÞAi , C1m ¼ mi ðxðtÞÞC1i
i¼1 i¼1
ð8:80Þ
Xr
B2m ¼ mi ðxðtÞÞB2i
i¼1

The first condition of stabilization is given by [41]

GiiT P þ PGii  0, 8i 2 In
   
Gij þ Gji T Gij þ Gji ð8:81Þ
PþP  0, ij
2 2
where
Gij ¼ Ai þ Bi Kj ð8:82Þ

Proof.
Xr X r  
2 Gij þ Gji
xðtÞ
_ ¼ mi ðxðtÞÞGii xðtÞ þ 2 mi ðxðtÞÞmj ðxðtÞÞ xðtÞ ð8:83Þ
i¼1 ij
2
 H? fuzzy control of suspension systems with actuator saturation 229

V_ ðxðtÞÞ ¼ x_ T ðtÞPxðtÞ þ xT ðtÞPx_ ðtÞ

2X r 3
2
6 m i ð x ð t Þ ÞGii x ð t Þ 7
6 i¼1 7
¼ x T ð t Þ6
6 X r   7 x ðt Þ
7
4 Gij þ Gji 5
þ2 mi ðxðtÞÞmj ðxðtÞÞ x ðt Þ
ij
2

If (8.81) is verified then

Xr Xr  
Gij þ Gji
mi 2 ðxðtÞÞGii xðtÞ þ 2 mi ðxðtÞÞmj ðxðtÞÞ x ðt Þ
i¼1 ij
2

and V_ ðxðtÞÞ  0, which ends the proof.

Remark 8.2. The LMI conditions (8.81) can be obtained by the following change
in variables.
Q ¼ P1 , Fi ¼ Ki P1 , 8i 2 Ir ð8:84Þ
For conditions (8.81), it is not necessary to have all crossed stable models.
However, if the number of rules is quite high, it is difficult to find a common
Lyapunov matrix ensuring the stability of the dominant models (Ai þ Bi Ki ) and
cross models (Ai þ Bi Kj ). The relaxed stabilization conditions for multiple model
systems were reported to relax the conservatism of the conditions (8.81). In [41],
the relaxation is based on the maximum number of active local models at each
time. Kim and Lee [49] were given an approach to relax the conservatism of the
previous works [41] by introducing symmetrical matrices.
The possibility of introducing matrices that are not necessarily symmetrical off-
diagonal matrix blocks admits more great freedom (or dimension) in guaranteeing
the stability of the multiple model control systems. This relaxation is carried out by
using the property of congruence (8.68) and is summarized in the following theorem.
Theorem 8.1 [39]. The system (8.79) is quadratically stabilizable via the PDC
control (8.78) if there exist symmetric matrix Wii P ¼ PT
0 and matrices Ki ,
Wij ¼ WjiT such that:

iÞ GiiT P þ PGii  Wii 8i 2 In

   
Gij þ Gji T Gij þ Gji
iiÞ PþP  Wij þ WjiT
2 2
ij
2 3
W11 W12    W1n ð8:85Þ
6 7
6  W22    W2n 7
6 7
iiiÞ 6
6 .. .. ..
7
.. 7  0
6 . . . . 7
4 5
 . . .    W12
Proof (see [39]).
230 Handbook of vehicle suspension control systems

8.7 H? approach
The H? approach is used to analyse and to synthesize controllers/observers
achieving an optimal level of disturbance attenuation and to guarantee the stability
of the closed-loop system. In the literature, several works have been developed for
the synthesis of control law T–S models with uncertainties in adopting the H?
approach (see, e.g., [50–54] and references therein). The idea of this approach is to
minimize the H? norm which represents the maximum value ratio between the
output signal energy (controlled output) and the input signal energy (disturbance
input) given by the following definitions:
Definition 8.1 [25]. Given a positive scalar g, the T–S fuzzy system (8.35) is said to
be stable with g-attenuation, if, for all any wðtÞ 2 =2 , the system (8.35) is stable
and the response zðtÞ under zero initial condition satisfies the following equation:
 2  
zðtÞ  g2 wðtÞ2 ð8:86Þ
2 2

In this chapter, to guarantee the stability of the closed-loop system (8.35) and
to eliminate the effect of external disturbances, in the presence of actuators
saturation, we use =2 -norm. This means:
ð
1
 2
zðtÞ ¼ zðtÞT zðtÞdt
2
0
ð8:87Þ
ð
1
 2
zðtÞ ¼ T
zðtÞ zðtÞdt
2
0

8.8 Analysis of PDC control with external disturbances and

actuator saturation
The problem of estimating the domain of attraction with an optimal level of
disturbance attenuation is equivalent to the following optimization problem [24]:

min g
ð8:88Þ
V_ ðtÞ þ zðtÞT zðtÞ  g2 wðtÞT wðtÞ < 0

8.8.1 Constrained control

In this section, our objective is to design a PDC controller to avoid saturation limits.
In this case we determine:
X
r
sðtÞ ¼ uðtÞ ¼ mj ðxðtÞÞKj xðtÞ ð8:89Þ
j¼1

such that:
r  
x ðt Þ 2 \=
j¼1
Kj ð8:90Þ
 H? fuzzy control of suspension systems with actuator saturation 231
  
with = Kj ¼ xðtÞ 2 Rn j jkij xj  u i for all i 2 Im , j 2 Ir and kij is the ith row of
the matrix Kj .
Using:

X
r
B1m ¼ mi ðxðtÞÞB1i
i¼1
X
r
D1m ¼ mi ðxðtÞÞD1i ð8:91Þ
i¼1
X
r
D1m ¼ mi ðxðtÞÞD1i
i¼1

The closed-loop system composed of (8.1) and (8.78) is described by

(  
x_ ðtÞ ¼ Am þ B2m Km xðtÞ þ B1m wðtÞ
  ð8:92Þ
zðtÞ ¼ C1m þ D2m Km xðtÞ þ D1m wðtÞ

Theorem 8.2. For a given system (8.35) and a given state feedback control (8.89),
the ellipsoid eðP, rÞ is contractively invariant set of the closed-loop system (8.92)
and achieves a disturbance rejection level g, if there exists a symmetric positive
definite matrix Q and matrices Fj 2 Rmn , solution of the LMI problem:

mingQ, Fj ð8:93Þ
" 2 j#
ui fi
 j T  0, for all i 2 Im , j 2 Ir ð8:94Þ
fi Q

2 3
Ai Q þ B2i Fj þ ðÞ  
6 7
Sij ¼ 6  7
T
4 B1i g2 I 50 for all i, j 2 Ir2 ð8:95Þ
T
C1i Q þ D2i Fj D1i I

The control law (8.78) is given by Kj ¼ Fj Q1 .

Proof. By (8.90) and using Lemma 8.5, we have for all i 2 Im and j 2 Ir :
 
kij ðPÞ1 kij T  u 2i ð8:96Þ

Let there be the change of variable (8.84), thus the inequality (8.96) can be written
as follows:
 T
u 2i Q  fij fij  0 ð8:97Þ

where fij is the ith row of the matrix Fj .

232 Handbook of vehicle suspension control systems

By Schur complement [25], the last inequality can be written as LMI (8.94).
Using (8.11), we have
T   T   
z z¼ C1m þ D2m Km x þ D1m w C1m þ D2m Km x þ D1m w
" #T 2  T 3 "   #T " #
x C1m þ D2m Km C1m þ D2m Km x
¼ 4 5
w T
D1m D1m w

then

Sm ¼ V_  g2 w w þ z z
T T

" #T 2  T 3" #
x Am þ B2m Km P þ ðÞ  x
¼ 4 5 T
þz z
w T
B1m P 2
g I w
" #T 8 2  T 3
x < Am þ B2m Km P þ ðÞ 
¼ 4 5
w : T
B1m P g2 I
2 9
T 3 2  T 3T >" #
= x
C1m þ D2m Km C1m þ D2m Km
þ4 54 5
D
T
D
T >
; w
1m 1m

By Schur complement (see Lemma 8.4), we have

2 T  3
P Am þ B2m Km þ ðÞ  
6 T 7
Sm ¼ 6
4 B1m P g2 I  7 5
  T
C1m þ D2m Km D1m I

Pre-and post-multiplying the first inequality by G ¼ diagðQ, I, IÞ, we have

2  3
Am þ B2m Km Q þ ðÞ  
6 T 7
GSm G ¼ 6
4 B1m g2 I  7
5
  T
C1m þ D2m Km Q D1m I

By using (8.17), we get:

2 3
Am Q þ B2m Fm þ ðÞ  
6 T 7
GSm G ¼ 6 B1m g2 I  7
4 5
T
C1m Q þ D2m Fm D1m I

To guarantee the inequality in optimization problem (8.88), it suffices to verify

GSm G  0, which ends the proof.
 H? fuzzy control of suspension systems with actuator saturation 233

8.8.2 Saturated control

In this control design, our goal is to design a PDC controller in the presence of
saturation. Differently to the first case, we will permit the saturation limits. Thus,
we have
sðtÞ ¼ satðuðtÞ, u Þ ð8:98Þ
Using Lemma 8.3, the closed-loop system composed of (8.1) and (8.78) is
described by
(   
x_ ðtÞ ¼ Am þ B2m Ea Km þ E a Hm xðtÞ þ B1m wðtÞ
   ð8:99Þ
zðtÞ ¼ C1m þ D2m Ea Km þ E a Hm xðtÞ þ D1m wðtÞ
  P r Pm
such that xðtÞ 2 \rj¼1 = Hj and with Hm ¼ mi ðxðtÞÞ Hi , Ea ¼ 2s¼1 as Es and
P2m i¼1
E a ¼ s¼1 as E s .

Theorem 8.3. For a given system (8.35) and a given state feedback control (8.98),
the ellipsoid eðP, rÞ is a contractively invariant set of the closed-loop system (8.99)
and achieves a disturbance rejection level g, if there exist a symmetric positive
matrix Q and matrices Fj 2 Rmn , Zj 2 Rmn , solution of the LMI problem:

mingQ, Fj , Zj ð8:100Þ
" #
u2 zji
 j i T  0, for all i 2 Im , j 2 Ir ð8:101Þ
zi Q

2 3
Ai Q þ B2i Es Fj þ B2i E s Zj þ ðÞ  
6 7
Ssij ¼ 6  7 2
T
4 B1i g2 I 5  0 for all i, j 2 Ir
T
C1i Q þ D2i Es Fj þ B2i E s Zj D1i I
ð8:102Þ

The control law (8.78) is given by Kj ¼ Fj Q1 and the matrix Hj ¼ Zj Q1 .
Proof. By (8.61) and using Lemma 8.5, we have
 T
hji ðPÞ1 hji  u 2i ð8:103Þ
Let:

Zj ¼ Hj Q ð8:104Þ

By (8.104) and (8.84), the inequality (8.103) can be written as follows:

 T
u 2i Q  zji zji  0 ð8:105Þ
234 Handbook of vehicle suspension control systems

where zij is the ith row of matrix Zj . By Schur complement, the last inequality can
be written as LMI (8.101). For LMI (8.102), the proof is obtained by using (8.100)
and following the same demonstration of constrained control.

8.8.3 Optimization of the attraction region

In this section, we are interested in estimating the largest domain of attraction, the
maximization of the ellipsoid eðP, 1Þ. The largest of ellipsoid is formulated as the
following optimization constraints [21]:

sup a
aXR  eðP, 1Þ
where XR  Rn is a prescribed bounded convex set containing the origin. There are
two cases that depend upon the shape of the reference set. The first case, XR , is an
ellipsoid:
 
X R ¼ x 2 R n x T N x  1 ð8:106Þ
where N > 0 can be chosen, for example, as an identity with compatible dimen-
sion. The other case is a polyhedron defined by

XR ¼ co x10 , x20 , . . . , xl0 ð8:107Þ
with x10 , x20 , . . . , xl0 are a priori points given in Rn . If XR is given by (8.103), the
set aXR is defined by
 
aXR ¼ x 2 Rn xT N x  a2
aXR  eðP, 1Þ is equivalent to a2 N  P  0.
By Schur complement, this last inequality can be written in LMI form as follows:
 
dN 1
0 ð8:108Þ
1 Q
with d ¼ a2 . If XR is given by (8.107), the set aXR is defined by

aXR ¼ co ax10 , ax20 , . . . , axl0
aXR  eðP, 1Þ is equivalent to:
 T
a xi0 ðPÞa xi0  1 8i 2 Il
or in LMI form by
 
d xi0
 i T 0 8i 2 Il ð8:109Þ
x0 Q

The stabilization problem of state feedback with a g-attenuation under actuator

saturation, in a domain of attraction as larger as possible, can be formulated as the
following LMI conditions.
 H? fuzzy control of suspension systems with actuator saturation 235

Theorem 8.4 (constrained control case). For a given system (8.1) and a given state
feedback control (8.78), the ellipsoid eðP, rÞ is contractively invariant set of the
closed-loop system (8.11) and achieves a disturbance rejection level g, with a
domain of attraction as larger as possible, if there exist a symmetric positive matrix
Q and matrices Fj 2 Rmn , solution of the LMI problem:
8
> mingQ, Fj
<
LMI ð8:94Þ, LMI ð8:95Þ
>
:
LMI ð8:108Þ or LMI ð8:109Þ
with d as minimum as possible.
Theorem 8.5 (saturated control case). For a given system (8.1) and a given state
feedback control (8.78), the ellipsoid eðP, rÞ is contractively invariant set of the
closed-loop system (8.92) and achieves a disturbance rejection level g, with a
domain of attraction as larger as possible, if there exist a symmetric positive matrix
Q and matrices Fj 2 Rmn , Zj 2 Rmn solution of the LMI problem:
8
> mingQ, Fj , Zj
<
LMI ð101Þ, LMI ð102Þ
>
:
LMI ð108Þ or LMI ð109Þ
with d as minimum as possible.
To relax the above results (Theorems 8.3 and 8.4) by reducing the number of
LMI conditions and introducing slack variables, we can use theorem in [39] by
replacing (8.95) and (8.102) by
Yii þ Wii  0, i 2 Ir
Yij þ Yij þ Wij þ WijT  0, i < j 2 Ir
0 1
W11    W1r
B C
B .. .. C
B  . . C
0
@ A
  W1r
where Yij ¼ Sij for LMI (8.95) and Yij ¼ Ssij for LMI (8.102) in. Notice that in this
case the LMI conditions are reduced to r(r þ 1)/2 instead of r2 which allow with the
slack variables Wij to introduce more of relaxation.

8.9 Control design for a quarter-car active suspension system

In this section, our objective is to design a PDC controller to guarantee the stability
of the vertical displacements of the quarter-car sprung mass (zs is the control
variable) and to eliminate the effect of external disturbances (Figure 8.7) by con-
strained control and saturated control.
236 Handbook of vehicle suspension control systems

Now, choosing XR described by (8.106) and solving the optimization problem

of Theorems 8.3 and 8.4 for different saturation levels, we obtain Table 8.1. The
result of Table 8.1 shows the influence of saturation levels on the stability domain
of the closed-loop system; rejection level g decreases and the stability region
expands with the increase in saturation level.
For example, consider the saturation level u ¼ 5KN ; solving the optimization
problem of Theorems 8.3 and 8.4, we have the results shown in Table 8.2.
For the saturation level u ¼ 5KN and amax ¼ 0:134 (see Table 8.1), with the
presence of saturation, this control (both saturated and constrained) can stabilize
the suspension car for initial conditions x01 2 ½ 0:134m 0:134m with x02 ¼
x03 ¼ x04 ¼ 0.
To illustrate these results, first, the simulation results are carried out without
the disturbance and the initial conditions: x01 ¼ 0:134 and x01 ¼ 0:134 with
x02 ¼ x03 ¼ x04 ¼ 0.
The system responses are shown in Figures 8.18–8.20 for saturated control and
in Figures 8.21 and 8.22 for constrained control.
These figures demonstrate that, despite the presence of actuator saturation, the
two controllers (saturate and constraint) can stabilize the suspension car in all
intervals given by Table 8.1.

Table 8.1 Minimized H1 parameters

u(KN) 2 3 5 10 20
Saturated g 1.5060 1.2601 0.9902 0.7023 0.4955
a 0.1279 0.1282 0.1345 0.1508 0.1753
Constrained g 1.5037 1.2602 0.9901 0.7026 0.4949
a 0.1280 0.1286 0.1344 0.1510 0.1752

Table 8.2 Parameters for saturated and constrained controls

2 3
0:0000 0:0000 0:0000 0:0013
6 0:0000 0:0005 0:0018 0:0101 7
Saturated control Q ¼ 1e þ 0076
4 0:0000
7
0:0018 0:0076 0:0610 5
0:0013 0:0101 0:0610 2:4716
H1 ¼ H2 ¼ 1e þ 004½1:3690 0:7182 0:2026 0:0028

K1 ¼ K2 ¼ 1e þ 007½1:4018 1:1632 0:3182 0:0038

2 3
0:0000 0:0000 0:0000 0:0012
6 0:0000 0:0005 0:0019 0:0107 7
Constrained control Q ¼ 1e þ 0076
4 0:0000
7
0:0019 0:0077 0:0633 5
0:0012 0:0107 0:0633 2:4936
K1 ¼ K2 ¼ 1e004½1:3711 0:7185 0:2026 0:0028
 H? fuzzy control of suspension systems with actuator saturation 237

0.15

0.1

0.05

0
x1(t)

–0.05

–0.1

–0.15

–0.2
0 0.5 1 1.5
Time (s)

Figure 8.18 The vertical displacements of the sprung mass x1 ðtÞ in saturated
control case (solid curve: x01 ¼ 0:134; dashed curve: x01 ¼ 0:134)

× 106
2

1.5

0.5
u(t)

–0.5

–1

–1.5

–2
0 0.5 1 1.5
Time (s)

Figure 8.19 Trajectory of the input control uðtÞ in saturated control

238 Handbook of vehicle suspension control systems

5,000

4,000

3,000

2,000

1,000
sat(u(t))

–1,000

–2,000

–3,000

–4,000

–5,000
0 0.5 1 1.5
Time (s)

Figure 8.20 Trajectory of the saturate input control satðuðtÞÞ in saturated control

0.15

0.1

0.05

0
x1(t)

–0.05

–0.1

–0.15

–0.2
0 0.5 1 1.5
Time (s)

Figure 8.21 Vertical displacements of the sprung mass x1ðtÞ in constrained

control case (solid curve: x01 ¼ 0:134; dashed curve: x01 ¼ 0:134)
 H? fuzzy control of suspension systems with actuator saturation 239

2,000

1,500

1,000

500
u(t)

–500

–1,000

–1,500

–2,000
0 0.5 1 1.5
Time (s)

Figure 8.22 Trajectory of the input control uðtÞ in constrained control

Now, in order to compare the two methods proposed, we will simulate the
active suspension with the external disturbance (Figure 8.7) and the initial condi-
tions x01 ¼ 0:134. Figures 8.18 and 8.21 show the trajectory of vertical displace-
ments of the sprung mass in the case of constrained and saturated control
respectively. Figures 8.20 and 8.21 show the input control signal and saturated
input signal, respectively, in the case of saturated control. Figure 8.22 shows the
input control signal in the case of constrained control.
The figures demonstrate that, despite the presence of actuator saturation and
the external disturbance, the two controls (saturated control and constrained
control) can stabilize the suspension car. But, if we compare the x1ðtÞ trajectory in
Figures 8.23–8.25, we can conclude that the saturated control (Figures 8.26–8.27)
give better stabilization performance of the system as reported in the open lit-
erature such as [21, 34]. However, saturated control presents chattering (e.g. see
Figure 8.27 between the instants 5 s and 5.5 s) that cannot be supported by some
actuators, especially actuators with velocity saturation (see [10]), the constrained
control gives smooth control signal (see Figure 8.28 between the instants 5 s
and 5.5 s).

8.10 Conclusion

This chapter has presented an H? control design of T–S model via state feedback
under actuator saturation and external disturbances. The T–S model was first used
240 Handbook of vehicle suspension control systems

0.6

0.5

0.4

0.3
x1(t)

0.2

0.1

–0.1

–0.2
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.23 Vertical displacements of the sprung mass x1ðtÞ in saturated control
case (solid curve: without control; dashed curve: with saturated
control; dashdot curve: with constrained control)

1
x3(t)

–1

–2

–3
0 2 4 6 8 10
Time (s)

Figure 8.24 Speed of the sprung mass x3ðtÞ (solid curve: without control; dashed
curve: with saturated control; dashdot curve: with constrained
control)
 H? fuzzy control of suspension systems with actuator saturation 241

60 80
60
40
40 20
0

20 –20
–40
4 4.05 4.1 4.15 4.2 4.25 4.3
dx3/dt

40
–20 20
0
–20
–40 –40
–60
–80
–60 6 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8

0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.25 Acceleration of the sprung mass dx3 /dt (solid curve: without control;
dashed curve: with saturated control; dashdot curve: with
constrained control)

× 105
5

4
1× 10
–5 0.5
0
u(t)

–0.5
–10
–1
–1.5
–2
–15
–2.5
–3
4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6
–20
1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.26 Trajectory of the input control uðtÞ in saturated control case
242 Handbook of vehicle suspension control systems

5,000
–900
4,000 –950

3,000 –1,000
–1,050
2,000
–1,100
1,000
–1,150
5 5.05 5.1 5.15 5.2 5.25 5.3 5.35 5.4 5.45 5.5
sat(u(t))

–1,000

–2,000

–3,000

–4,000

–5,000
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.27 Trajectory of the saturate input control satðuðtÞÞ in saturated control
case

2,000

1,000

0
u(t)

–1,000
–1,185

–1,190
–2,000
–1,195

–1,200

–3,000 –1,205

–1,210

–1,215
5 5.05 5.1 5.15 5.2 5.25 5.3 5.35 5.4 5.45 5.5
–4,000
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Figure 8.28 Trajectory of the input control uðtÞ in constrained control case
 H? fuzzy control of suspension systems with actuator saturation 243

to describe the nonlinear system. Then, the PDC controller was designed to guar-
antee the closed-loop system stability with H? performance. The control saturation
design was dealt with using two methods: the first was the constrained control input
method, whereas the second was a saturated control input method. Based on
Lyapunov approach, the problem of H? stabilization in the presence of actuator
saturation was formulated as an LMI optimization problem in the two cases solved
easily by using existing numerical tools. Finally, the simulation results on an active
suspension system demonstrated the advantage of the proposed control and illu-
strated clearly the benefits of these methods.

References

[1] H. Liu, K. Nonami, and T. Hagiwara, ‘Active following fuzzy output feed-
back sliding mode control of real-vehicle semi-active suspensions’, Journal
of Sound and Vibration, vol. 314, pp. 39–52, 2008.
[2] D. Sammier, ‘La modélisation et la commande de suspension de véhicule
automobile’, thesis, Polytechnic Institut national de Grenoble, France, 9
November 2001.
[3] X. Moreau, A. Rizzo, and A. Oustaloup, ‘Improvement of hydractive sus-
pension hard mode confort thanks to a low frequency active CRONE system.
Part 2: control part and simulation results’, International Design Engineering
Technical Conference & Computers and Information in Engineering, 2009.
[4] A. Oustaloup, X. Moreau, and M. Nouillant, ‘The CRONE suspension’,
Control Engineering Practice, vol. 4, no. 8, pp. 1101–1108, 1996.
[5] M. M Fateh and S. S. Alavi, ‘Impedance control of an active suspension
system’, Mechatronics, vol. 19, pp. 134–140, 2009.
[6] N. Yagiz and Y. Hacioglu, ‘Backstepping control of a vehicle with active
suspensions’, Control Engineering Practice, vol. 16, pp. 1457–1467, 2008.
[7] Y. Ando and M. Suzuki, ‘Control of active suspension systems using
the singular perturbation method’, Control Engineering Practice, vol. 4,
pp. 287–293, 1999.
[8] M. Chadli, A. Rabhi, and A. El Hajjaji, ‘Observer-based H? fuzzy control
for vehicle active suspension’, 16th IEEE Mediterranean Conference on
Control and Automation, Ajaccio, France, 2008.
[9] M. Yamashita, F. Kazuo, K. Hayakawa, and H. Kimura, ‘Application
of H? control to active suspension systems’, Automatica, vol. 30, no. 11,
pp. 1717–1729, 1994.
[10] S. J. Huang and H. Y. Chen, ‘Adaptive sliding controller with self-tuning
fuzzy compensation for vehicle suspension control’, Mechatronics, vol. 16,
pp. 607–622, 2006.
[11] C. S. Ting, T.-H.S. Li, and F.-C. Kung, ‘Design of fuzzy controller for active
suspension system’, Mechatronics, vol. 5, pp. 365–383, 1995.
[12] F. J. D’Amato and D. E. Viassolo, ‘Fuzzy control for active suspensions’,
Mechatronics, vol. 10, pp. 897–920, 2000.
244 Handbook of vehicle suspension control systems

[13] M. M. M. Salem and A. A. Aly, ‘Fuzzy control of a quarter-car suspension

system’, World Academy of Science, Engineering and Technology, vol. 53,
pp. 258–263, 2009.
[14] H. Du and N. Zhang, ‘Fuzzy control for nonlinear uncertain electrohydraulic
active suspensions with input constraint’, IEEE Transactions on Fuzzy System,
vol. 17, pp. 343–356, 2009.
[15] R. Murray-Smith and T. Johansen, A Multiple Model Approaches to Mod-
eling and Control, Taylor & Francis, London, UK, 1997.
[16] M. Chadli, A. Akhenak, J. Ragot, and D. Maquin ‘State and unknown input
estimation for discrete time multiple model’, Journal of the Franklin Insti-
tute, vol. 346, no. 6, pp. 593–610, 2009.
[17] D. Saifia, M. Chadli, and S. Labiod, ‘Static output feedback stabilization of
multiple models subject to actuators saturation’, 12th IFAC LSS Symposium,
11–14 July, Villeneuve d’Ascq, France, 2010.
[18] H. J. P. and Y. S. Kim, ‘Decentralized variable structure control base on a
full-car model’, Proceedings of the IEEE International Conference on
Control Applications, Trieste, Italy, 1–4 September, 1998.
[19] L. Jung-Shan and H. Chiou-Jye Huang, ‘Nonlinear back-stepping active
suspension design applied to a half-car model’, Vehicle System Dynamics,
vol. 42, no. 6, pp. 373–393, 2004.
[20] D. Saifia, M. Chadli, S. Labiod, and H. R. Karimi, ‘Fuzzy control of DC-DC
converters with input constraint’, Mathematical Problems in Engineering,
doi: 10.1155/2012/973082, 2012.
[21] Y. Y. Cao and Z. Lin, ‘Robust stability analysis and fuzzy-scheduling
control for nonlinear systems subject to actuator saturation’, IEEE Trans-
actions on Systems, vol. 11, pp. 57–67, 2003.
[22] H. Han, ‘T–S fuzzy controller in consideration of input constraint’, Second
International Conference on Innovative Computing, Information and Con-
trol, 5–7 September, Kumamoto, Japan, 2007.
[23] S. H. Kim, C. H. Lee, and P. G. Park, ‘H? state-feedback control for
fuzzy systems with input saturation via fuzzy weighting-dependent Lyapunov
functions’, Computers and Mathematics with Applications, vol. 57, pp. 981–990,
2009.
[24] J. C. Lo and M. L. Lin, ‘Robust H? nonlinear control via fuzzy static
output feedback’, IEEE Transactions on Circuits and System, vol. 50,
pp. 1494–1502, 2003.
[25] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, Linear Matrix
Inequalities in System and Control Theory, Philadelphia, PA: SIAM, 1994.
[26] M. Chadli and A. Elhajjaji, ‘Observer-based robust fuzzy control of non-
linear systems with parametric uncertainties – Comment on’, Fuzzy Sets and
Systems, vol. 157 no. 9, pp. 1276–1281, 2006.
[27] D. Henrion, Stabilité des systèmes linéaires incertains à commande contra-
inte,’ doctoratethesis, laboratoire d’analyse et d’architecture des systèmes du
CNRS, Université Paul Sabatier de Toulouse, 1999.
 H? fuzzy control of suspension systems with actuator saturation 245

[28] T. Hu and Z. Lin, Control Systems with Actuator Saturation. Analysis and
Design, BirkhHâuser, Boston, vol. xvi, p. 392, 2001.
[29] D. Henrion and S. Tarbouriech, ‘LMI relaxations for robust stability of lin-
ear systems with saturating controls’, Automatica, vol. 35, pp. 1599–1604,
1999.
[30] N. Kapoor and P. Daoutidis, ‘Stabilization of nonlinear processes with
input constraints’, Computers and Chemical Engineering, vol. 24, pp. 9–21,
2000.
[31] J. M. Jomas da Silva Jr. and S. Tarbouriech, ‘Anti-windup design region of
stability for discrete-time linear systems’, System & Control Letters, vol. 55,
pp. 184–192, 2006.
[32] T. Hu, A. R. Teel, and L. Zaccarian, ‘Anti-windup synthesis for linear
control systems with input saturation: Achieving regional, nonlinear per-
formance’, Automatica, vol. 44, pp. 512–519, 2008.
[33] D. Dai, T. Hu, A. R. Teel, and L. Zaccarian, ‘Piecewise quadratic Lyapunov
function for system with dead zones or saturation’, System & Control Letters,
vol. 58, pp. 365–371, 2009.
[34] T. Zhang, G. Feng, H. Liu, and J. Lu, ‘Piecewise fuzzy anti-windup dynamic
output feedback control of nonlinear processes with amplitude and rate
actuator saturations’, IEEE Transactions on Fuzzy System, vol. 17, no. 2,
pp. 253–263, 2009.
[35] D. Saifia, M. Chadli, S. Labiod, and T. M. Guerra, ‘Robust H? static output
feedback stabilization of T–S fuzzy systems subject to actuator saturation’,
International Journal of Control, Automation, and Systems, vol. 10, no. 3,
pp. 1–10, 2012.
[36] D. Saifia, M. Chadli, and S. Labiod, ‘H? control of multiple model subject to
actuator saturation: Application to quarter-car suspension system’, Analog
Integrated Circuits and Signal Journal. doi: 10.1007/s10470-011-9628-3, 2011.
[37] M. Chadli and H. R. Karimi, ‘Robust observer design for unknown inputs
Takagi–Sugeno models’, IEEE Transactions on Fuzzy Systems, vol. 21,
no. 1, pp. 158–164, 2013.
[38] M. Chadli and T.-M. Guerra, ‘LMI solution for robust static output feedback
control of Takagi–Sugeno fuzzy models’, IEEE Transactions on Fuzzy
Systems, vol. 20, no. 6, doi: 10.1109/TFUZZ.2012.2196048, 2012.
[39] L. Xiaodong and Z. Qingling, ‘New approaches to H? controller designs
based on fuzzy observers for T–S fuzzy systems via LMI’, Automatica, vol. 39,
pp. 1571–1582, 2003.
[40] M. Chadli, S. Aouaouda, H. R. Karimi, and P. Shi, ‘Robust fault tolerant
tracking controller design for a VTOL aircraft’, Journal of the Franklin
Institute, vol. 350, pp. 2627–2645, 2013.
[41] K. Tanaka, T. Ikede, and H. O. Wang, ‘An LMI approach to fuzzy controller
designs based on the relaxed stability conditions’, Proceedings of the IEEE
International Conference on Fuzzy Systems, Barcelona, Spain, pp. 171–176,
1997.
246 Handbook of vehicle suspension control systems

[42] K. Tanaka, T. Ikede, and H. O. Wang, ‘Fuzzy regulators and fuzzy obser-
vers: Relaxed stability conditions and LMI-based design’, IEEE Trans-
actions on Fuzzy Systems, vol. 6, pp. 250–265, 1998.
[43] K. Tanaka, T. Ikeda, and H. O. Wang, ‘A unified approach to controlling
chaos via an LMI-based fuzzy control system design’, IEEE Transactions on
Circuit Systems, vol. 45, pp. 1021–1040, 1998.
[44] K. Tanaka and T. Kosaki, ‘Design of a stable fuzzy controller for an
articulated vehicle’, IEEE Transactions on Systems, Man and Cybernetics B,
vol. 27, pp. 552–558, 1997.
[45] K. Tanaka and M. Sano, ‘A robust stabilization problem of fuzzy control
systems and its application to backing up control of a truck-trailer’, IEEE
Transactions on Fuzzy Systems, vol. 2, no. 1, pp. 119–134, 1994.
[46] K. Tanaka, H. O. Wang, ‘Fuzzy Control Systems Design and Analysis: A
Linear Matrix Inequality Approach. John Wiley and Sons, Hoboken
New Jersey, USA, 2001.
[47] M. C. M. Teixeira, E. Assumção, R. G. Avellar, ‘On relaxed LMI-based for
fuzzy regulators and fuzzy observers’, IEEE Transactions on Fuzzy System,
vol. 11, no. 5, pp. 613–623, 2003.
[48] H. O. Wang, K. Tanaka, and M. F. Griffin, ‘An approach to fuzzy control of
nonlinear systems: Stability and design issues’, IEEE Transactions on Fuzzy
Systems, vol. 4, no. 1, pp. 14–23, 1996.
[49] E. Kim and H. Lee, ‘New approaches to relaxed quadratic stability condition
of fuzzy control systems’, IEEE Transactions on Fuzzy Systems, vol. 8,
no. 5, pp. 523–533, 2000.
[50] L. K. Wang and L. Xiao-Dong, ‘Robust H? fuzzy control for discrete-time
nonlinear systems’, International Journal of Control, Automation, and Sys-
tems, vol. 8, no. 1, pp. 118–126, 2010.
[51] H. N. Wu, ‘Robust H2 fuzzy output feedback control for discrete-time
nonlinear systems with parametric uncertainties’, International Journal of
Approximate Reasoning, vol. 46, pp. 151–165, 2007.
[52] J. Yoneyama, ‘Robust H? control analysis and synthesis for Takagi–
Sugeno general uncertain fuzzy systems’, Fuzzy Sets and Systems, vol. 157,
pp. 2205–2223, 2006.
[53] H. Zhang and F. Gang, ‘Stability analysis and H? controller design of
discrete-time fuzzy large-scale systems based on piecewise Lyapunov
functions’, IEEE Transactions on System, Man and Cybernetics B, vol. 38,
no. 5, 2008.
[54] S. Z. Shaosheng, G. Feng, L. James, and X. Shengyuan, ‘Robust H? control
for discrete time fuzzy systems via basis-dependent Lyapunov functions’,
Automatica, vol. 40, pp. 823–829, 2004.
 Chapter 9
Design of sliding mode controller for
semi-active suspension systems with
magnetorheological dampers
Shigehiro Toyamaa, Makoto Yokoyamab
and Fujio Ikedac

Abstract
This chapter presents two sliding mode controllers of semi-active suspension
systems with magnetorheological dampers, which have undesirable nonlinear
properties. One sliding mode controller is based on the theory of model-following
control. In the model-following controller, a desired semi-active suspension system
is chosen as the reference model to be followed, and the control law is determined
so that an asymptotically stable sliding mode occurs in the error dynamics between
the plant and the reference model states. The advantages of this controller are as
follows: (1) measurement of the damping force is not required, (2) the reference
model specifies the desired performance considering the passivity constraint of
the damper, and (3) it is entirely possible to maintain the sliding mode and achieve
high robustness against the nonlinear properties of the damper. The other sliding
mode controller is designed by the describing function method so that a switching
function is enforced into a desired limit cycle instead of a perfect sliding mode.
Although the proposed sliding mode controller cannot generate the limit cycle as
desired because of the passivity constraint of controllable dampers, restricting the
switching function in the vicinity of the origin can suppress the deterioration due to
the passivity constraint, such as increase in jerk of the sprung mass.
Moreover, a method for designing an observer is introduced for semi-active
suspension systems using variable structure system theory, which provides a highly
robust property against modeling errors and disturbances in the context of the
matching conditions. The structure of the introduced observer is designed to be
robust against road variations, which can be seen as nonstationary system

a
Department of Electronic Control Engineering, Nagaoka National College of Technology, Niigata,
Japan
b
Department of Mechanical and Production Engineering, Niigata University, Niigata, Japan
c
Department of Mechanical Engineering, Nagaoka National College of Technology, Niigata, Japan
248 Handbook of vehicle suspension control systems

disturbances. Although this structure basically requires the actual damping force to
be measured, it is estimated using a model of the damper. Thus, the effect of the
estimation errors of the damping force on the state estimation is discussed in
detail, and the sufficient conditions for stability of the observer are given using
Lyapunov’s theory. As a result, both the structure and the design process of the
proposed observer are simplified in comparison with existing ones.

9.1 Introduction
Suspension systems are located between wheel axles and vehicle bodies or
frames. Their purpose is to support the weight of vehicles, to cushion against
bumps and holes in the road, and to maintain traction between the tires and the
road. Springs and dampers are the two main components of suspension systems.
The springs support the weight of the vehicle and its load, and absorb road
shocks. The dampers help to control or absorb spring action.
It is difficult for passive suspension systems to fulfill two functions at the same
time, such as providing good ride comfort while also securing good driving stability
from road disturbances. For example, an increase in the damping coefficient of
the damper decreases the resonance problems at handling and improves driving
stability. However, ride comfort simultaneously deteriorates, since the gain of the
vibration transfer characteristics in the high-frequency range rises.
To go beyond the limitations of passive suspension systems, active control
methods of body motion were devised. These are called active suspension systems
and are equipped with active devices such as hydraulic actuators. Active devices
can give active force generated with external energy besides passive force corre-
sponding to springs and dampers. In the 1980s, the development of active sus-
pension systems was energetic with the progression of electronics and control
valves. In 1983, Team Lotus equipped a Formula One racing car, Lotus 92, with an
active suspension system [1]. In 1989, two models with active suspension systems
were marketed for the first time, by Nissan Motor [2, 3]. However, the spread of
active suspension systems could not be materialized, since practical requirements
such as improvement of durability, security of fail-safe, and reduction of cost and
power consumption remain insufficient.
In equipping controllable dampers with a simple mechanism that needs no
external energy sources, semi-active suspension systems are superior to active ones
for their practical requirements. In particular, the development of variable orifice
dampers has been pursued vigorously. The electronic control unit of semi-active
suspension systems signals a solenoid or small electric motor at the top of each
controllable damper. The motor then rotates slightly, opening or closing various-
sized orifices in the piston. This changes the resistance to fluid flow within the
controllable dampers.
For the primary controllable damper systems developed in the early 1980s,
feedforward control, which classifies driving operations and changes the damping
 Design of sliding mode controller for semi-active suspension systems 249

characteristics of dampers in advance, was mainly used [4]. The default damping
characteristics are set to soft to obtain good ride comfort. For steering and accel-
erating or braking, feedforward systems are set to hard to control rolling, diving,
and squatting. However, predicting driving conditions and disturbance is challen-
ging for such feedforward systems, since a misjudged driving condition and dis-
turbance can result in an opposing control action.
Feedback control systems to change damping characteristics according to the
measured states of vehicles have also been studied. Feedback systems are known to
show superior performance that is considered impossible with the feedforward
systems. However, one key problem is ‘‘hunting’’ in the high-frequency range
caused by time delay in the orifice switching of variable orifice dampers.
As a new controllable damper to supersede variable orifice dampers, the
application of functional fluids such as magnetorheological (MR) fluids, which
show strong interaction with electromagnetic fields, has attracted attention. In the
1960s, MR fluids of colloidal-solution-type that dispersed minute particles of fer-
romagnetics were developed. Since then, applications of MR fluids, such as seals,
bearings, and dampers, have been developed and improved so as to meet the
demands of practical use.
MR dampers have been studied not only for their application to the semi-active
suspension systems of vehicles but also in various other fields of vibration control
[5–7], since they offer large force and have weak temperature dependence. Because
the response speed of MR dampers is faster than that of variable orifice dampers, an
improvement of suspension performance in the high-frequency range can be
expected. Because the mechanism of MR dampers is almost the same as that of
normal dampers except for the addition of a coil near the orifice, cost and failure
rate should both be reduced. However, MR dampers show strong nonlinearity such
as viscous characteristics of Bingham fluids and hysteresis. Thus, there is a desire
for a design method that permits such nonlinearity.
Various attempts have been made to apply control theories to semi-active
suspension systems [8–18]. One is the sliding mode control theory in which high-
frequency switching of a relay input enforces the plant into sliding modes. A plant
in sliding modes is known to have high robustness against modeling errors and
disturbances, and is also known for the simplicity of the controller, which can be
designed if the upper bound of the modeling error is known [19]. Applied to the
controller for semi-active suspension systems, the sliding mode control could be
expected to provide robustness against changes in the number of passengers, fric-
tion of suspension mechanics, and undesired nonlinearity of controllable dampers
[20–22].
This chapter presents two sliding mode controllers that are able to overcome
the above problem. One proposed sliding mode controller is based on the theory of
model-following control [21]. In the model-following controller, a desired semi-
active suspension system is chosen as the reference model to be followed, and the
control law is determined so that an asymptotically stable sliding mode occurs in
the error dynamics between the plant and the reference model states. The other
250 Handbook of vehicle suspension control systems

proposed sliding mode controller is designed by the describing function method so

that a limit cycle of the switching function in the vicinity of the origin occurs
instead of a perfect sliding mode [22]. As a result, the effect of the passivity
constraint can be improved by avoiding frequent switching of the control input,
while the robustness of the proposed controller can be maintained. A method for
designing an observer is also proposed for semi-active suspensions using the vari-
able structure system (VSS) theory, which provides a highly robust property against
modeling errors and disturbances in the context of the matching conditions. The
structure of the proposed observer is designed to be robust against road variations,
which can be viewed as nonstationary system disturbances [23].

9.2 Control of semi-active suspension systems

with MR dampers
9.2.1 Variable orifice damper
Variable orifice dampers have a solenoid or small electric motor at the top. The
motor rotates slightly, opening or closing various-sized orifices in the piston. This
changes the resistance to fluid flow within those dampers. Figure 9.1 shows the
typical damping characteristics of variable orifice dampers. From this figure, it can
be seen that the damping coefficient c(t) is regarded as the control input of the
semi-active suspension systems with variable orifice dampers.
A controller design of the quarter-car model shown in Figure 9.2 is demon-
strated as follows. The equations of motion for the plant are given by

m2 y€2 ¼ k2 ðy2  y1 Þ  cðtÞðy_ 2  y_ 1 Þ ð9:1Þ


m1€y 1 ¼ k2 ðy2  y1 Þ þ cðtÞðy_ 2  y_ 1 Þ  k1 ðy1  d Þ  c1 y_ 1  d_ ð9:2Þ

where m2 and m1 denote the sprung mass and the unsprung mass, respectively;
k2 and k1 denote the stiffness; c1 denotes the damping coefficient; and
y2(t), y1(t), and d(t) denote the sprung-mass displacement, the unsprung-mass

Damping force

c(t)

0 Piston speed

Figure 9.1 Damping characteristics of variable orifice damper

 Design of sliding mode controller for semi-active suspension systems 251

y2
m2

k2 c(t)
y1
m1

k1 c1
d

Figure 9.2 Quarter-car model of semi-active suspension with variable orifice

damper

displacement, and the road disturbance, respectively. Defining the state variable
vector as
2 3 2 3
x1 y2
6 x2 7 6 y_ 2 7
x¼6 7 6 7
4 x3 5 ¼ 4 y1 5 ð9:3Þ
x4 y_ 1
the state space model of the plant is given by
x_ ¼ Ax þ BcðtÞx þ Gx ð9:4Þ
where each matrix is given by
T
x ¼ ½d d_  ,
2 3
0 1 0 0
6 k2 k2 7
6 0 7
6 m2 0 m2 7
A ¼6
6 0
7
6 0 0 1 7 7
4 k ðk2 þ k1 Þ c1 5
2
0  
m1 m1 m1
2 3
0 0 0 0
6 1 1 7
60  0 7
6 m m2 7
B ¼6
60
2 7
6 0 0 0 77
4 1 1 5
0 0 
m1 m1
2 3
0 0
6 0 0 7
6 7
6 7
G ¼6 0 0 7
6 7
4 k1 c1 5
m1 m1
252 Handbook of vehicle suspension control systems

The plant (9.4) is characterized by the bilinear term c(t)x, which is a linear com-
bination of the control input c(t) and the state variable vector x. For this bilinear
system, various controller design methods such as Karnopp’s law [8], H? control
[11, 17], and optimal control have been proposed. As an example of this type of
controller, Karnopp’s law is introduced as follows. To realize the skyhook damper,
the damping force should satisfy the following condition:

cðtÞðx2  x4 Þ ¼ cs x2 ð9:5Þ

where cs is the damping coefficient of the skyhook damper. Then, the damping
coefficient c(t) regarded as the control input of the semi-active suspension system is
given by
cs x2
cðtÞ ¼ ð9:6Þ
x2  x4

From (9.6), it can be seen that the control input can take the negative sign. How-
ever, the damping coefficient c(t) of the controllable damper cannot be negative
because of the passivity condition that the direction of the damping force is subject
to the sign of the piston speed of the damper. Therefore, in considering the passivity
condition, Kanopp suggested an approximation of the skyhook damper as follows:
8 cx
< s 2 when x2 ðx2  x4 Þ > 0
cðtÞ ¼ x2  x4 ð9:7Þ
:
0 when x2 ðx2  x4 Þ  0

The disadvantage of such approximation methods is ‘‘hunting,’’ caused by two

factors. One factor is that the control input can change drastically when the control
input c(t) is approximated to zero at the moment the sign of the passivity condition
changes to negative. The other factor is the actuator response. In particular, a
response-delay effect can appear when multistep revolution of the control rod is
required to change the damping coefficient by a large amount.

9.2.2 MR damper
Figure 9.3 shows a typical structure of the MR damper enclosing MR fluid. This
fluid is a colloidal solution dispersing magnetic particles in carrier oil. Under
influence of a magnetic field, the magnetic particles form clusters as shown in
Figure 9.4. The cluster attraction grows considerably with increasing intensity of
the magnetic field. The electromagnets, which are arranged close to the annular
orifices, play the role of controlling the damping characteristics. Figure 9.5 shows
the damping characteristics of MR dampers. As shown in Figure 9.6, the char-
acteristics can be modeled on the following simple physical model:

_ uÞ ¼ c0 Dx_ þ q  u  sgnðDxÞ
fd ¼ gðDx, _ ð9:8Þ

where fd is the damping force, c0 is the damping coefficient of the viscous friction,
and Dx_ is the piston speed of the dampers. The control input u is the current to the
 Design of sliding mode controller for semi-active suspension systems 253

Accumulator Diaphragm Annular orifice

Electromagnet MR fluid

Figure 9.3 Structure of MR damper

Carrier oil

Magnetic
particles

(a)

(b)

Figure 9.4 Mechanism of MR effect: (a) magnetic field OFF; (b) magnetic
field ON

Damping force

fd

0 Piston speed

Figure 9.5 Damping characteristics of MR damper

254 Handbook of vehicle suspension control systems

Viscous Controllable
friction coulomb friction

Figure 9.6 Simple physical model of MR damper

Controller

Ideal damping force

Control fd
law Inverse
Some information model
of suspension

Measured piston speed

Control input Damping force

u fd
MR
damper

Piston speed

Figure 9.7 Control scheme using inverse model of MR damper

electromagnets, and q is a coefficient showing the relation between the control

input and the strength of the controllable coulomb friction.
Figure 9.7 shows the control scheme using an inverse model of the MR dam-
per. From information on the suspension, the control law calculates the ideal
damping force fd . From the ideal damping force and the measured piston speed, the
inverse model of the MR damper calculates the control input u. When the damping
characteristics are modeled by (9.8), the control input u can be described by
8
< 1 ðf  c DxÞsgnðD xÞ if ðfd  c0 DxÞD
_ x_ > 0
0 _ _
u¼ q d ð9:9Þ
: if ðfd  c0 DxÞD
_ x_  0
0
where the passivity condition is added to the inverse model of (9.8). Without the
modeling error with the actual damping characteristics and damping model (9.8),
 Design of sliding mode controller for semi-active suspension systems 255

the damping force fd could be equal to the ideal damping force. However, it is
difficult to obtain an accurate damper model analytically, since the actual damping
characteristics of MR dampers are known to have a strong nonlinearity, such as
hysteresis. The undesired nonlinearity forces us to employ a simple physical model
like (9.8), and the control law, which calculates the ideal damping force f d as
shown in Figure 9.7, should be designed in order to have the robustness against the
modeling error.

9.3 Model-following sliding mode controller for semi-active

suspension systems
MR dampers have intrinsic nonlinear properties in addition to a passivity con-
straint. In order to take full advantage of the dampers, robust control schemes must
be developed. Various control strategies have been proposed, including sliding
mode control. Previous researchers who have investigated sliding mode controllers
have attempted to enforce the plant dynamics in the sliding mode. It appears that
this strategy did not always yield a satisfactory performance, since it was difficult
to maintain the sliding mode in the plant dynamics owing to the passivity con-
straint or modeling errors [24, 25]. An alternative way of using the sliding mode
controller is in a force feedback loop, which is incorporated to induce the damper
to generate the desired force [26]. Note that this strategy requires measurement of
the damping force.
In this section of the chapter, a sliding mode controller is introduced based on
the theory of model-following control. The basic idea is to enforce the tracking
error dynamics between the plant and the reference model states, instead of the
plant dynamics, in the sliding mode. The reference model in this controller is
essential. Intuitively, if the reference model is under the same constraints as those
of the plant, the plant can physically follow the reference model and the error
dynamics can be enforced in the sliding mode. In a broader sense this may be
viewed alongside the idea of the matching condition. In the present work, a semi-
active suspension based on the skyhook damper system is used as the reference
model, and the effectiveness of the controller is investigated by numerical
simulation.

9.3.1 System model and problems

The left side of Figure 9.8 shows a model of the quarter-car semi-active suspension
system under consideration. The goal of this investigation is to improve the fre-
quency response of the sprung mass in the lower-frequency range, which is closely
linked to ride quality, without deteriorating the response in the higher-frequency
range. Therefore, using the state variable vector
" # " #
x1 y2
x¼ ¼
x2 y_ 2
256 Handbook of vehicle suspension control systems

Controller
Reference model

State State c0(t)

y2 vector x vector
m2 SMC xr m2
fd
k2 fd
u Inverse k2
y1 model y1
m1

k1 c1
d

Figure 9.8 Quarter-car model and control scheme

in which y2 is the sprung-mass displacement and y_ 2 is the sprung-mass velocity, the

following single degree of freedom model is used to design the controller:
x_ ¼ Ax þ bfd þ gy1 þ bw ð9:10Þ
fd ¼ gðx2  y_ 1 , uÞ ð9:11Þ
where y1 is the unsprung-mass displacement, w is the disturbing force to the sprung
mass, and fd is the damping force, which is a nonlinear function of the control input
u (current) and the piston speed of the damper x2  y_ 1 . Each matrix is described as
follows:
2 3
0 1
A ¼ 4 k2 5
 0
m2
2 3
0
b¼4 1 5
m2
2 3
0
g¼ 4 k 2
5
m2
Typical characteristics of the MR damper are shown in Figure 9.9, where the
hatched regions represent the physical constraints that prevent the damper from
generating the damping force in addition to passivity. Furthermore, the damping
force is not strictly linear with respect to the current, which is the control input.
Various nonlinear dynamical models have been proposed.
One of the most popular schemes to determine the damping force is given by
(
cS x2 if x2 ðx2  y_ 1 Þ > 0
fd ¼ ð9:12Þ
0 if x2 ðx2  y_ 1 Þ  0
 Design of sliding mode controller for semi-active suspension systems 257

where fd is the ideal damping force, and cS is the desired damping coefficient of the
skyhook system. This scheme proposed by Karnopp [8], which is a type of clipped-
optimal control, is very effective in practice if the actual damper can exactly gen-
erate the desired force fd . It is straightforward in that, for the ideal force given by
(9.12), we can determine the control input with the inverse function of gð,Þ based
on the measurement of the piston speed. However, this control strategy is not
always successful, because the actual damper has nonlinear properties, as men-
tioned above (simulation results are shown below).

9.3.2 Sliding mode controller

We consider the model-following control, as shown in Figure 9.8, in which the
reference model is not an ideal skyhook damper system (active suspension) but an
approximate one (semi-active suspension) with an ideal controllable damper. It
should be noted that the reference model is a system that the plant can physically
follow if the physical constraints as shown by the hatching in Figure 9.9 can be
neglected. The reference model is described by
x_ r ¼ Ar xr þ gy1 ð9:13Þ
where xr is the state variable vector of the reference model, and consists of the
sprung-mass displacement and the velocity, similar to the plant. In (9.13), the
matrix Ar is given by
2 3
0 1
Ar ¼ 4 k 2 c0 ðtÞ 5
 
m2 m2
where
(
cs if x2 ðx2  y_ 1 Þ > 0
c0 ðtÞ ¼ ð9:14Þ
0 if x2 ðx2  y_ 1 Þ  0

Damping force (N)

u = 2.0 (A)
1.5 (A)
1.0 (A)
1,786
0.5 (A)
0.0 (A)
286
0.02 Piston speed (m/s)

Figure 9.9 Typical characteristics of MR damper

258 Handbook of vehicle suspension control systems

Letting
e ¼ x  xr ð9:15Þ
be the tracking error vector in the state variables, the tracking error dynamics is
given by
e_ ¼ Ae ðtÞe þ KðtÞx þ bðfd þ wÞ ð9:16Þ
where
2 3
0 1
Ae ðtÞ ¼ 4 k2 c0 ðtÞ 5
 
m2 m2
2 3
0 0
KðtÞ ¼ 4 c0 ðtÞ 5
0
m2
In guaranteeing asymptotically stable error dynamics, sliding mode control is
highly promising. Although many types of design methods have been proposed for
sliding mode control, we investigated the following method which is simple and
effective. First, let us define a switching function (surface) of the error vector:
s ¼ ½ h1 1  e ¼ he ð9:17Þ
If the error dynamics is in sliding mode, the switching function s is given by
s ¼ s_ ¼ 0 ð9:18Þ
Assuming the disturbance w ¼ 0, the equivalent control is given by
fdeq ¼ ½ k2 m2 k2 þ c0 ðtÞ  e  c0 ðtÞx2 ð9:19Þ
This equivalent control and a relay control are used as the desired control force, i.e.:

f þ gsgnðsÞ for fd ðx2  y_ 1 Þ > 0
fd ¼ deq ð9:20Þ
0 for fd ðx2  y_ 1 Þ  0
For (9.20), note that the desired force is constrained by the passivity condition.
Finally, the control input is determined by
u ¼ g1 ðx2  y_ 1 , fd Þ ð9:21Þ
where g 1 ð, Þ is the inverse model of the MR damper given by (9.11).
For sliding mode existence, consider a Lyapunov function candidate as follows:
1
V ðsÞ ¼ s2 ð9:22Þ
2
and the following condition:
d 1d 2
V ðsÞ ¼ ðs Þ  hjsj ð9:23Þ
dt 2 dt
 Design of sliding mode controller for semi-active suspension systems 259

where h is a strictly positive constant. Letting Dfd ¼ fd  fd be the damping force

error, and using (9.16), (9.17), (9.19), and (9.20), the sliding condition (9.23) becomes

g  hðhbÞ1  ðw þ Dfd ÞsgnðsÞ ð9:24Þ

Therefore, since ðhbÞ1 ¼ mS , the following condition suffices:

g  hmS þ jw þ Dfd j ð9:25Þ

Owing to the passivity constraint, it is impossible to guarantee that this condition

holds all the time, even if the maximum absolute value of the disturbance
including the damping force error can be estimated in advance. However, recal-
ling that for the plant, the reference model is physically easy to follow, in general,
the error state can be expected to be near the switching surface. Simulation
results in the following section will reveal that this conjecture is adequate in
practice.
Assuming the existence of the sliding mode, the error dynamics is described by
e_ 1 ¼ h1 e1 ð9:26Þ

Thus, during the sliding mode, the error dynamics is solely governed by the para-
meter h1 of the switching function and is robust against disturbances that satisfy the
matching condition.

9.3.3 Simulation results

In order to investigate the performance of the sliding mode controller, numerical
simulations were carried out. The parameter values used in the simulations are as
follows:
m2 ¼ 400 ðkgÞ, m1 ¼ 40 ðkgÞ, k2 ¼ 15,791 ðN=mÞ
k1 ¼ 157,913 ðN=mÞ, c1 ¼ 126 ðNs=mÞ
cs ¼ 3,554 ðNs=mÞ, h1 ¼ 10, g ¼ 1,000

The characteristics of the MR damper were assumed to be as in Figure 9.9, the road
input was white Gaussian with mean zero, and the sampling period for control was
1 (ms).
Figure 9.10 shows the response of the semi-active suspension system by
Karnopp’s clipped-optimal control with the inverse model of the damper. As seen
from this figure, this controller was unable to achieve the performance of the
desired approximate skyhook damper system (reference model). This means that
this control strategy is not robust against the damping force error Dfd , which cor-
responds to the hatched regions in Figure 9.9.
As shown in Figure 9.11(a) and (b), the sliding mode controller yielded
almost perfect tracking to the reference model. Figure 9.11(c) and (d) show that
the error dynamics was approximately in sliding mode, as expected. The power
spectral density of the acceleration of the sprung mass is shown in Figure 9.12.
260 Handbook of vehicle suspension control systems

Reference model
(m)
0.02

–0.02
Plant
0 2 4 6
Time (s)
(a)

(m/s) Reference model

0.2

Plant
–0.2
0 2 4 6
Time (s)
(b)

Figure 9.10 Conventional controller: (a) body displacement; (b) body velocity

Reference model (m/s) Reference model

(m)
0.2
0.02

0 0

–0.02 Plant
Plant –0.2
0 2 4 6 0 2 4 6
Time (s) Time (s)
(a) (b)

(N)
0.1
1,000
0 0

–1,000 –0.1
0 2 4 6 0 2 4 6
Time (s) Time (s)
(c) (d)

Figure 9.11 Sliding mode controller: (a) body displacement; (b) body velocity;
(c) damping force; (d) switching function
From this, it can be seen that the sliding mode controller yielded lower mass
acceleration than another controller, mentioned above, in the low-frequency range.
On the other hand, in the frequency range higher than the resonant frequency of
the unsprung mass, this situation was reversed owing to chattering generated by
 Design of sliding mode controller for semi-active suspension systems 261

((m/s2)2/Hz)
20

Conventional
15 SMC

10

0
10–1 100 101 102
Frequency (Hz)

Figure 9.12 Power spectral density of body acceleration

the relay input used in the sliding mode controller. Though various schemes
to reduce this chattering have been proposed, in general, there is some type of
trade-off between a high degree of robustness and reduction of chattering in the
sliding mode control.

9.4 Sliding mode controller with describing function method

Controllable dampers employed as actuators are passive devices, and the damping
force acts as a resistance force against expansion and contraction of the dampers.
In the controller design of a semi-active suspension system, it is necessary to
consider the passivity condition that the direction of the damping force is subject
to the sign of the piston velocity of the damper. In many studies, control laws are
designed first on the assumption that the plant has an active suspension system.
Then, approximate control laws, to which is added the passivity condition of
controllable dampers, are implemented at the time of application to the semi-
active suspension system. For example, Karnopp’s law [8], used frequently for
semi-active suspension systems, is an approximate control law for the skyhook
damper system, which requires a fully active control. For the passivity constraint,
the control input is approximated as a minimum control input. However, the
approximation for the passivity constraint can cause sprung-mass jerk to increase,
since the damping force can change suddenly when the sign of the piston velocity
changes.
For this type of problem, a nonlinear H? controller that considers semi-active
suspension systems as bilinear systems has been proposed [17]. A linear H?
controller that considers the Lissajous figure of the damping force has also been
suggested [9]. These controllers can improve jerks. However, no studies of sliding
262 Handbook of vehicle suspension control systems

mode control that take into account the passivity constraint have been conducted
as yet.
This section of the chapter presents a sliding mode control law which is able to
overcome the above problem. With the proposed sliding mode controller designed
by the describing function method, a limit cycle of the switching function in the
vicinity of the origin occurs instead of a perfect sliding mode. As a result, the effect
of the passivity constraint can be improved by avoiding frequent switching of the
control input, while the robustness of the proposed controller can be maintained.

9.4.1 Problem formulation

Figure 9.13 shows a quarter-car model of the semi-active suspension system con-
sidered as the plant in this section. The sprung mass is supported with a semi-active
suspension system consisting of a spring and a controllable damper. In the figure,
m2 and m1 denote the sprung mass and the unsprung mass, respectively; k2 and k1
denote the stiffness; c2 and c1 denote the damping coefficient; and y2(t), y1(t), and
d(t) denote the sprung-mass displacement, the unsprung-mass displacement, and
the road disturbance, respectively. The other sliding mode controller, which cal-
culates the ideal damping force fd ðtÞ, is designed for this plant. The motion
equations for the quarter-car model are given by
 
m2€y 2 ðtÞ ¼ k2 fy2 ðtÞ  y1 ðtÞg  c2 fy_ 2 ðtÞ  y_ 1 ðtÞg  fd ðtÞ þ wðtÞ ð9:27Þ
 
m1€y 1 ðtÞ ¼ k2 fy2 ðtÞ  y1 ðtÞg þ c2 fy_ 2 ðtÞ  y_ 1 ðtÞg þ fd ðtÞ þ wðtÞ
 
 k1 fy1 ðtÞ  dðtÞg  c1 y_ 1 ðtÞ  d_ ðtÞ ð9:28Þ

where k1 fy1 ðtÞ  dðtÞg denotes variation of the road holding force, and w(t)
denotes model uncertainties and disturbances in the suspension system. The pas-
sivity of the damper brings about the following constraint:

f ðtÞ if f ðtÞfy_ 2 ðtÞ  y_ 1 ðtÞg > 0
fd ðtÞ ¼ ð9:29Þ
0 if f ðtÞfy_ 2 ðtÞ  y_ 1 ðtÞg  0

y2
m2
fd
k2
y1 C2
m1

k1 C1
d

Figure 9.13 Quarter-car model with semi-active suspension system

 Design of sliding mode controller for semi-active suspension systems 263

where f (t) denotes a fully active control force derived from a control law, and
y_ 2 ðtÞ  y_ 1 ðtÞ denotes the piston velocity of the damper. The passivity constraint
(9.29) means that the fully active control force f (t) can be generated by the con-
trollable damper if the condition f ðtÞfy_ 2 ðtÞ  y_ 1 ðtÞg is positive, otherwise f (t) is
obliged to be approximated as the minimum control input, which is considered to
be zero in this section. The goal here is to improve the jerk increase of the sprung
mass that accompanies a rapid change in the damping force while ensuring
robustness against modeling error. Next, a sliding mode control law to derive the
fully active control force f (t) is introduced.

9.4.2 Integral sliding mode control

For the quarter-car model shown in Figure 9.13, the state space model of the
sprung-mass system is described as

_
xðtÞ ¼ AxðtÞ þ bff ðtÞ þ wðtÞg þ gdðtÞ ð9:30Þ

where the state vector x(t) and each matrix are given as

x ð t Þ ¼ ½ x 1 ð t Þ x 2 ð t Þ T ¼ ½ y 2 ð t Þ y_ 2 ðtÞT
2 3
0 1
A ¼ 4 k2 c2 5
 
m2 m2
23
0
b¼4 1 5

m2
2 3
0 0
g ¼ 4 k2 c2 5
m2 m2

dðtÞ ¼ ½ y1 ðtÞ y_ 1 ðtÞT

Applying the integral sliding mode control theory [19] to the state space model
(9.30), the fully active control force f (t) is defined as

f ðtÞ ¼ f0 ðtÞ þ f1 ðtÞ ð9:31Þ

where f0(t) is a state feedback control input leading to an ideal closed-loop system as

f0 ðtÞ ¼ FxðtÞ ð9:32Þ

264 Handbook of vehicle suspension control systems

As an example of this type of state feedback control input, Karnopp’s law is widely
known as

c x ðtÞ if x2 ðtÞfy_ 2 ðtÞ  y_ 1 ðtÞg > 0
f0 ðtÞ ¼ s 2 ð9:33Þ
0 if x2 ðtÞfy_ 2 ðtÞ  y_ 1 ðtÞg  0
where cs denotes the damping coefficient of the skyhook damper. In (9.31), f1(t) is a
relay input to reject the effect of modeling error as follows:
f1 ðtÞ ¼ g  signðsðtÞÞ ð9:34Þ
where the relay gain g > 0 is determined in order to satisfy the existence condition
of a sliding mode. The switching function sðtÞ is defined by
sðtÞ ¼ zðtÞ  CxðtÞ ð9:35Þ
where C ¼ ½ a 1 , a  0, and zðtÞ is an auxiliary variable, whose dynamics can
be described by
z_ ðtÞ ¼ C fAxðtÞ þ bf0 ðtÞ þ gdðtÞg ð9:36Þ
The sliding mode caused by the control law (9.31), (9.32), (9.34)–(9.36) can be
analyzed with the equivalent control method [19]. First, we set the time derivative
of the switching function sðtÞ equal to zero in order to show the equivalent control
input f1eq ðtÞ of the relay input (9.34).
_
sðtÞ ¼ z_ ðtÞ  C 
_
xðtÞ
 
¼ z_ ðtÞ  C AxðtÞ þ b f0 ðtÞ þ f1eq ðtÞ þ wðtÞ þ gdðtÞ ¼ 0 ð9:37Þ
Solving (9.37), we can describe the equivalent control input f1eq ðtÞ as
f1eq ðtÞ ¼ wðtÞ ð9:38Þ
Substituting the equivalent control input (9.38) in (9.30) and (9.31), we can
describe the closed-loop dynamics of the sliding mode as
_
xðtÞ ¼ AxðtÞ þ bf0 þ gdðtÞ ð9:39Þ
From (9.38) and (9.39), it can be seen that the equivalent control input could
eliminate the effect of the disturbance wðtÞ and that the ideal dynamics derived
from the state feedback control f0(t) in (9.32) could be achieved in the sliding mode.
Then, with Lyapunov’s stability theorem, the existence condition of the sliding
mode may be treated as follows. When a function
1
V ðsðtÞÞ ¼ sðtÞ2 ð9:40Þ
2
is chosen as a candidate of Lyapunov’s function, the time derivative of V ðsðtÞÞ
along the solution of (9.37) is given by
d 1d
V ðsðtÞÞ ¼ ðsðtÞ2 Þ ¼ sðtÞ  sðtÞ
_ <0 ð9:41Þ
dt 2 dt
 Design of sliding mode controller for semi-active suspension systems 265

For the existence of the sliding mode in the sense of Lyapunov stability, the relay
gain g should satisfy the following inequality:
g > jwðtÞj ð9:42Þ

9.4.3 Redesign of relay input with describing function method

Replacing the relay input f1 ðtÞ given by (9.34), a new control input is introduced as

w2n
F1 ðsÞ ¼ F2 ðsÞ ð9:43Þ
s2 þ 2Vwn s þ w2n

f2 ðtÞ ¼ g1 signðsðtÞÞ  g2 signðsðtÞÞ

_ ð9:44Þ

where s denotes the Laplace operator; F1 ðsÞ and F2 ðsÞ denote the Laplace transform
of f1 ðtÞ and f2 ðtÞ, respectively; and g1 , g2 , V, and wn are design parameters for
generating a limit cycle of the switching function sðtÞ: Figure 9.14 is a block
diagram of the proposed sliding mode control system. In this system, the transfer
function GðsÞ from f2 ðtÞ to sðtÞ is given as

w2n
GðsÞ ¼ ð9:45Þ
m2 sðs2 þ 2Vwn s þ w2n Þ

Then, the proposed control system may be transformed into a nonlinear feedback
system shown in Figure 9.15. With the describing function method [27], the
existence condition of a limit cycle of the switching function sðtÞ may be
described as
1 þ N ðaÞGðsÞ ¼ 0 ð9:46Þ

d Plant
g
f1 + 1 x
b C
++ ++ s
A
σ
–
F +
f0
A z
++ 1
b C
+ s
g
ω2n f2 −γ1 sign (s (t))
s2 + 2ςωns + ω2n − γ2 sign (s (t))

Figure 9.14 Block diagram of proposed control system

266 Handbook of vehicle suspension control systems

Linear part G(s)

f2 ω2n σ
– m2s(s + 2ςωns + ω2n)
2

Nonlinear part
γ1 sign (σ) + γ2 sign (σ)

Figure 9.15 Analysis model for existence condition of limit cycle

where a denotes the amplitude of the limit cycle, and N ðaÞ denotes the describing
function of the nonlinear part in the proposed control system, and may be described
as follows [28]:
4
N ðaÞ ¼ ðg þ g2 jÞ ð9:47Þ
pa 1
Let s ¼ jw in the existence condition (9.46) for a limit cycle, then
1
GðjwÞ ¼  ð9:48Þ
N ðaÞ
The describing function method states that if (9.48) has a solution (a, w), then there
may be a periodic solution of the system with frequency and amplitude close to w
and a, respectively. The existence condition (9.48) can be concretely described as
 
2wn ðpam2 Vw2 þ 2wn g1 Þ þ j pam2 wðw2n  w2 Þ þ 4w2n g2 ¼ 0 ð9:49Þ
From (9.49), the design parameters of the control input (9.43) and (9.44) should be
determined as follows when V is set as a constant in the first step:
g1 > m2 pVaw=2 ð9:50Þ
wn ¼ pam2 Vw2 =2g1 ð9:51Þ
2
g2 ¼ pam2 wðw  w2n Þ=4w2n ð9:52Þ

9.4.4 Simulation conditions

In order to investigate the effectiveness of the proposed sliding mode controller,
numerical simulations were carried out with the numerical analysis software
MATLAB/Simulink. The proposed sliding mode controller (hereafter called the
proposed SMC) was compared with Karnopp’s law and the sliding mode controller
with the relay input (SMC with relay). Table 9.1 lists the parameters for the
numerical simulations. The natural frequency of the sprung mass is 7.5 (rad/s), and
that of the unsprung mass is 75 (rad/s). Table 9.2 presents the design parameters of
the proposed SMC for each desired limit cycle of the switching function s(t). These
parameters were determined to satisfy (9.50)–(9.52) when z was set to 1.0 in the
 Design of sliding mode controller for semi-active suspension systems 267

Table 9.1 Simulation parameters

Simulation Value Simulation Value

parameter parameter
m2 (kg) 300 m1 (kg) 30
k2 (N/m) 17,000 k1 (N/m) 170,00
c2 (Ns/m) 100 c1 (Ns/m) 452
cs (Ns/m) 3,193 a 0
g 100

Table 9.2 Design parameters for each desired limit cycle of

switching function

Desired limit cycle Design parameters

w a g1 g2 z wn
3
12.3 1.00  10 80 549 1.0 0.9
25.1 1.00  103 80 265 1.0 3.7
37.7 1.00  103 80 171 1.0 8.4
50.2 1.00  103 80 123 1.0 14.8
62.8 1.00  103 80 93.3 1.0 23.2
75.4 1.00  103 80 72.3 1.0 33.5
87.9 1.00  103 80 56.5 1.0 45.5
100 1.00  103 80 44.3 1.0 58.9
113 1.00  103 80 33.5 1.0 75.2
126 1.00  103 80 24.2 1.0 93.5

first step. The road disturbance velocity was a band-limited white Gaussian noise
signal, and the sampling period for control was 1 ms.

9.4.5 Accuracy of limit cycle of switching function

First, numerical simulations were carried out to investigate the accuracy of the limit
cycle led by the proposed SMC when the passivity constraint (9.29) was dis-
regarded, that is, in the case of fully active control. The accuracy of the limit cycle
should be verified, since the describing function method is an approximation
method for finding a periodic solution. Table 9.3(a) presents the accuracy of the
limit cycles in the case of fully active control. As an example, Figure 9.16(a) shows
the time response of the switching function for w ¼ 126 (rad/s). From Table 9.3(a)
and Figure 9.16(a), it can be seen that the fully active control without the passivity
constraint (9.29) can generate approximately the desired limit cycle of the switching
function sðtÞ: Although the relay gain g1 was set to be constant in this study, higher
accuracy can be obtained by adjusting the parameter for each desired limit cycle.
Table 9.3(b) presents the accuracy of the limit cycle led by the proposed SMC
with the passivity constraint (9.29), that is, in the case of semi-active control. As
an example, Figure 9.16(b) shows the time response of the switching function for
268 Handbook of vehicle suspension control systems

Table 9.3(a) Accuracy of limit cycle of switching function: fully

active control

Desired limit cycle Actual limit cycle Error (%)

w a w a w a
12.3 1.00  103 14.7 8.88  104 17.0 11.2
25.1 1.00  103 25.8 9.06  104 2.5 9.4
37.7 1.00  103 39.3 9.18  104 4.2 8.2
50.2 1.00  103 51.5 9.34  104 2.5 6.6
62.8 1.00  103 64.1 9.47  104 2.0 5.3
75.4 1.00  103 76.0 9.65  104 0.8 3.5
87.9 1.00  103 88.6 9.80  104 0.7 2.0
100 1.00  103 99.3 9.99  104 1.3 0.1
113 1.00  103 111 1.00  103 1.7 0.0
126 1.00  103 124 1.00  103 1.5 0.0

Table 9.3(b) Accuracy of limit cycle of switching function:

semi-active control

Desired Actual Error (%)

w a w a w a
3 3
12.3 1.00  10 6.1 1.30  10 51.2 30.0
25.1 1.00  103 2.5 3.70  103 90.2 270
37.7 1.00  103 7.4 6.80  103 80.5 580
50.2 1.00  103 31.9 7.20  103 36.5 620
62.8 1.00  103 28.2 5.70  103 55.1 470
75.4 1.00  103 43.0 4.30  103 43.0 330
87.9 1.00  103 60.1 4.00  103 31.6 300
100 1.00  103 60.1 3.30  103 40.2 230
113 1.00  103 60.1 2.80  103 46.8 180
126 1.00  103 60.1 2.40  103 52.2 140

w ¼ 126 (rad/s). From Table 9.3(b), it can be seen that there are more than 30%
errors for each desired limit cycle, particularly for the error of amplitude a. How-
ever, the time response of the switching function shown in Figure 9.16(b) can be
gently restrained in the vicinity of the origin as intended. The design parameters
can be adjusted so that the actual limit cycle can be enforced as desired.

9.4.6 Improvement of deterioration caused by passivity constraint

How such a restriction in the vicinity of the origin can improve the deterioration
due to the passivity constraint was investigated. The simulation parameters were
considered to be nominal. While the acceleration of the sprung mass could be
obtained equivalently in every control law, jerk of the sprung mass and variation of
the road holding force were subject to the effect of the passivity constraint,
 Design of sliding mode controller for semi-active suspension systems 269

× 10–3
2

–1

–2
0 0.5 1 1.5 2
Time (s)
(a)
× 10–3
2

–1

–2
0 0.5 1 1.5 2
Time (s)
(b)

Figure 9.16 Limit cycle of switching function: (a) fully active control;
(b) semi-active control

particularly in the SMC with relay. Table 9.4 describes the effectiveness of the
proposed SMC in comparison to the SMC with relay. Although the average value
of variation of the road holding force increased slightly, the average jerk of
the sprung mass decreased dramatically. Figure 9.17 shows the Lissajous figure
of the ideal control input for both the sliding mode controllers. It can be seen that
the proposed SMC was able to suppress the switching of the control input that
accompanied the passivity constraint, compared to the SMC with relay. As
improvement of performance indexes, the switching frequency of the control input
due to the passivity constraint and the rate of time for which the control input
cannot be applied by the passivity constraint were considered, as shown on the
right-hand side of Table 9.4. In the case of the SMC with relay, the switching
frequency was 54.6 (1/s) and the rate of time was 75.9 (%). From these results, it
can be seen that restricting the switching function in the vicinity of the origin can
improve the deterioration due to the passivity constraint.

9.4.7 Verification of robustness against parameter variation

Now the robustness of the proposed SMC for the parameter variation problem is
illustrated. The parameter variation was assumed so that the sprung mass m2
270 Handbook of vehicle suspension control systems

Table 9.4 Improvement of deterioration caused by passivity constraint

effectiveness of proposed SMC in comparison with SMC with relay)

Desired limit cycle Jerk (%) Variation of Switching Rate of time

road holding frequency due concerning
force (%) to passivity passivity
constraint (%) constraint (%)
w a RMS Max. RMS Max.
3
12.3 1.00  10 69.7 11.0 0.1 0.0 44.1 65.3
25.1 1.00  103 66.6 4.4 0.0 0.2 40.3 65.3
37.7 1.00  103 66.5 6.6 0.2 0.6 44.7 69.6
50.2 1.00  103 61.8 9.4 þ0.5 0.0 54.0 67.7
62.8 1.00  103 61.5 12.2 þ1.1 þ2.8 41.0 65.3
75.4 1.00  103 58.7 11.5 þ2.5 0.2 34.4 64.7
87.9 1.00  103 58.4 8.6 þ5.5 þ11.8 32.6 62.5
100 1.00  103 56.6 1.0 þ6.6 þ11.5 30.8 56.0
113 1.00  103 54.0 þ5.9 þ9.1 þ7.6 28.8 62.2
126 1.00  103 52.6 þ18.3 þ9.2 þ7.3 42.3 45.1
Average 60.6 4.1 þ3.4 þ4.0 39.3 62.4

400
Ideal control input (N)

Passive
200 constraint

–200 Passive
constraint

–400
–0.4 –0.2 0 0.2 0.4
Piston velocity (m/s)

(a)

400
Ideal control input (N)

Passive
200 constraint

–200 Passive
constraint

–400
–0.4 –0.2 0 0.2 0.4
Piston velocity (m/s)
(b)

Figure 9.17 Lissajous figure of ideal control input: (a) SMC with relay;
(b) proposed SMC
 Design of sliding mode controller for semi-active suspension systems 271

decreased by 25% and the stiffness k2 increased by 25% compared with the nom-
inal values.
Figure 9.18 shows the power spectral density of the sprung-mass acceleration
for each control law. The desired frequency of the limit cycle in the proposed SMC
was 25.1 (rad/s). From Figure 9.18(a), it can be seen that the parameter variation

–10
Nominal
Parameter
–20 variation

–30

–40

–50
10–1 100 101 102
Frequency (Hz)
(a)
–10 Nominal
Parameter
–20 variation

–30

–40

–50
10–1 100 101 102
Frequency (Hz)
(b)

–10
Nominal
Parameter
–20 variation

–30

–40

–50
10–1 100 101 102
Frequency (Hz)
(c)

Figure 9.18 Power spectral density of sprung-mass acceleration under parameter

variation (compared with Karnopp’s law under nominal conditions):
(a) Karnopp’s law; (b) SMC with relay; (c) proposed SMC
272 Handbook of vehicle suspension control systems

400 400

Ideal control input (N)

Ideal control input (N)

Passive Passive
200 200
constraint constraint

0 0

–200 Passive –200 Passive

constraint constraint
–400 –400
–0.4 –0.2 0 0.2 0.4 –0.4 –0.2 0 0.2 0.4
Piston velocity (m/s) Piston velocity (m/s)

Figure 9.19 Lissajous figure of ideal damping force

Table 9.5 Variation of road holding force under

parameter variation (compared with
Karnopp’s law under nominal conditions)

Controller Variation of road holding force

RMS (%) Max. (%)

Karnopp’s law 6.8 þ10.3
SMC with relay 2.9 þ1.7
Proposed SMC 4.8 þ0.5

increased the spectral density around the natural frequency of the sprung mass
according to Karnopp’s law. From Figure 9.18(b), it can be seen that the SMC with
relay was able to achieve robustness against the parameter variation, since the
quasi-sliding mode could occur. However, chattering occurred in the high-
frequency range upon switching the relay input (9.34). Moreover, from Figure 9.18(c),
it can be seen that the proposed SMC was able to show robustness against para-
meter variation without the effect of chattering in the high-frequency range, since
the switching number due to the passivity constraint was reduced as shown in
Figure 9.19. Table 9.5 presents the variation in the road holding force with para-
meter variation in comparison with Karnopp’s law under nominal conditions. From
these results, it can be seen that the proposed sliding mode controller was able to
achieve higher robustness than the other two controllers.

9.5 VSS observer of semi-active suspension systems

In most researches on the control of semi-active suspension systems, the plant

states used for calculation of the control input are assumed to be known. However,
there are technical challenges in the use of some plant states. One problem is how
to estimate the vertical velocity of the vehicle body. Although the feedback control
 Design of sliding mode controller for semi-active suspension systems 273

of the tire state is reported in some of the literature as providing good suspension
performance [13, 29], it is difficult to measure tire information such as tire
deflection and tire velocity. For this type of problem, it is essential to use state
estimation observers to limit costs in the automobile industry. The observer design
of semi-active suspension systems requires considerable ingenuity, since such
suspension systems have strong nonlinearity and road disturbance exists as system
disturbance. A bilinear disturbance-isolation observer [30] and a linear gain-
scheduling observer [31] have been proposed. The former observer is not practical,
since it needs to estimate road disturbance, and the robustness of this observer,
designed through complicated parameter settings, cannot be guaranteed. The latter
type of observer is designed for sprung-mass systems. Although the observer
design method can be applied to unsprung-mass systems, gain scheduling of the
observer may be extremely difficult in all road conditions. To resolve this type of
state estimation problem of semi-active suspension systems, this section of the
chapter presents a method for designing a VSS observer.

9.5.1 Plant
Figure 9.20 shows the static characteristics of typical semi-active devices such as
variable orifice dampers and MR dampers. Variable orifice dampers can control
their viscous resistance as shown in Figure 9.20(a). MR dampers can control their
yield strength as shown in Figure 9.20(b).

Damping force

fc

u=0 f0
0 Suspension stroke velocity
(a)

Damping force

fc

u=0
f0
0 Suspension stroke velocity

(b)

Figure 9.20 Characteristics of semi-active devices: (a) variable orifice damper;

(b) MR damper
274 Handbook of vehicle suspension control systems

y2
m2
fc
k2 c2
u Controller
Δy2 y1
m1

Δy1 d k1

Figure 9.21 One-wheel model with two degrees of freedom

In this study, we assume the plant to be a one-wheel model with two degrees
of freedom as shown in Figure 9.21. The motion equations for the plant are
given by
m2€y 2 ¼ k2 Dy2  c2 Dy_ 2  fc ð9:53Þ
m1€y 1 ¼ k1 Dy1 þ k2 Dy2 þ c2 Dy_ 2 þ fc ð9:54Þ
where y2, y1, and d denote sprung-mass displacement, unsprung-mass displace-
ment, and road disturbance, respectively; Dy2 and Dy1 denote the relative dis-
placements as shown in Figure 9.21; c2 denotes the viscous resistance containing
the minimum damping force f0 with no control input; and fc denotes variable
damping force and is assumed to show the following nonlinearity
fc ¼ g ðDy_ 2 , uÞ: ð9:55Þ

9.5.2 Problem formulation

Define the state variable vector x as

x ¼ ½ x1 x2 x3 x4 T ¼ ½ y_ 2 Dy2 y_ 1 Dy1 T ð9:56Þ

Then, the state space model of the plant can be described as

x_ ¼ Ax þ Bfc þ G d_ ð9:57Þ
fc ¼ gðx1  x3 , uÞ ð9:58Þ
where
2 3
c2 =m2 k2 =m2 c2 =m2 0
6 7
6 1 0 1 07
A¼6
6 c =m
7
4 2 1 k2 =m1 c2 =m1 k1 =m1 7
5
0 0 1 0
T
B ¼ ½ 1=m2 0 1=m1 0
T
G ¼ ½0 0 0 1 
 Design of sliding mode controller for semi-active suspension systems 275

The output y is given by

y ¼ Cx þ Dfc ð9:59Þ

The output y is selected so that (C, A) is observable.

Now, the problem of the state estimation of a semi-active suspension system
is considered. From (9.57), it can be seen that the road disturbance velocity d_
exists as a system disturbance. The statistical property of the disturbance d_ can
change widely according to driving conditions such as bumps and driving on
highways. Even if an optimal observer such as the Kalman filter is applied under
the assumption that the variable damping force fc is measurable, it is difficult to
optimize the observer against the nonstationary disturbance. Moreover, the state
estimation remains difficult because of the nonlinearity of the variable damping
force fc shown in (9.58). From (9.58), it can be seen that the damping force fc is
expressed by the function of the control input u and the piston speed, which is a
linear combination of some of the state variables. Even if the damping force fc is
linear with respect to the control input and the piston speed, the plant can be
described as a bilinear system. On the other hand, robustness against various
model uncertainties such as change in the number of passengers is required for
state estimation.

9.5.3 Design of VSS observer

When the VSS observer proposed by Walcott and Zak [32] is directly applied to the
observer design of the plant, two variable structure terms for the damping force fc
and the road disturbance velocity d_ , which are the second and third terms of the
right-hand side of (9.57), respectively, are required. However, this type of observer
design may cause the chattering problem, since not employing nonlinear models of
the controllable damper increases the upper bound of the uncertainty. Thus, as
shown in Figure 9.22, we propose a VSS observer employing a nonlinear damper
model that estimates the variable damping force fc as

^_ ¼ AO x
x ^ þ ðB  KDÞ^f c þ Ky þ S ðx
^ , y, rÞ ð9:60Þ
^f c ¼ g^ ð^x 1  ^x 3 , uÞ ð9:61Þ
^ x þ D^f c
y ¼ C^ ð9:62Þ

where

AO ¼ A  KC ð9:63Þ
8
<G s r if s 6¼ 0
x , y, rÞ ¼
Sð^ ksk ð9:64Þ
: if s ¼ 0
0
s ¼ Hð^
y  yÞ ð9:65Þ
276 Handbook of vehicle suspension control systems

y + 1 x̂
K + s
+ +
Ao
S (x̂,y,ρ) C
+ − y
S H
σ +
Switching
function D

fˆc Damper model xˆ1 − xˆ 3

B−KD ĝ u

Figure 9.22 Proposed VSS observer

The gain matrix K and the variable structure term (9.64) are determined to satisfy
the following conditions:

1. All eigenvalues of the matrix AO have negative real parts.

2. For a symmetric positive-definite matrix Q, the matrix H satisfies the fol-
lowing equation:
C T H T ¼ PG ð9:66Þ
where P is the unique symmetric positive-definite solution of the Lyapunov
equation
ATO P þ PAO ¼ Q ð9:67Þ
3. A scalar function r satisfies the following condition:
r  kd_ k ð9:68Þ
Then, the dynamics of the estimation error e is given by
e_ ¼ AO e þ ðB  KDÞð^fc  fc Þ  G d_ þ S ðx
^ , y, rÞ ð9:69Þ

The effect of the state estimation on the estimation error ^fc  fc of the variable
damping force in (9.61) should be considered. First, in the case of MR dampers, the
asymptotic stability of the state estimation error is analyzed. The damping force fc
shown in Figure 9.20(b) can be described by
fc ¼ gmr ðuÞ  sgnðx1  x3 Þ ð9:70Þ
where gmr(u) is the absolute value of the damping force, which is controlled by the
input u, and the sign of the damping force is equal to the suspension stroke velocity.
The MR damper model must exactly estimate the damping force fc against the
estimation error of the suspension stroke velocity, unless the estimated suspension
stroke velocity has a different sign to the true velocity. Then, the dynamics of the
estimation error e given by (9.69) can be regarded as
e_ ¼ AO e  G d_ þ S ðx
^ , y, rÞ ð9:71Þ
 Design of sliding mode controller for semi-active suspension systems 277

When a function

V ¼ eT Pe ð9:72Þ

is chosen as a candidate for Lyapunov’s function on the estimation error e, the time
derivative of the function (9.72) is given by

 eT PBHCe
V_ ¼ eT AT0 P þ PA0 e  2 r  2eT PG d_
kHCek
¼ eT Qe  2kHCekr  2eT C T H T d_  eT Qe  2kHCekr þ 2kHCekr
¼ eT Qe < 0 ð9:73Þ

With Lyapunov’s stability theorem, the estimation error e is asymptotically stable.

In most practical cases, however, the effect on modeling errors of the damper
model, which arise from perturbation according to the driving conditions, cannot be
disregarded. Now, the effect of the damper characteristics on the perturbation
should be considered, when Lyapunov’s function is also given by (9.72). Assuming
that the modeling error of the damper model ^fc  fc exists, the time derivative of
the function (9.72) is given by

 eT PBHCe  
V_ ¼ eT AT0 P þ PA0 e  2 r  2eT PG d_  2eT P ðB  KDÞ ^fc  fc
kHCek
< e Qe þ 2kP ðB  KDÞkkekj^fc  fc j
T

< lmin ðQÞkek2 þ 2kPðB  KDÞkkekj^fc  fc j

ð9:74Þ

where lmin ðQÞ is the minimal eigenvalue of the matrix Q. The sensitivity a of the
absolute value of the damping force estimation error to the suspension stroke
velocity estimation error leads to the following equation:

pffiffiffi
j^f c  fc j ¼ aje1  e3 j ¼ aj½ 1 0 1 0 ej < a 2kek ð9:75Þ

From (9.74) and (9.75), it can be seen that the following condition should be
satisfied so that the time derivative of Lyapunov’s function V_ is negative:
pffiffiffi
2lmin ðQÞ
0<a< ð9:76Þ
4kP ðB  KDÞk

From (9.76), it can be considered that MR dampers acting like Bingham fluids
allow a larger modeling error for the damping force than variable orifice dampers
that behave like Newton fluids, as shown in Figure 9.23. In other words, MR
dampers can increase the degree of freedom in designing the observer.
278 Handbook of vehicle suspension control systems

Controllable damper force

Slope α

Estimated
Damper
model
Actual
Actual characteristics

0 Actual Estimated
Suspension stroke velocity
(a)

Controllable damper force

Slope α
Damper model
Estimated
Actual
Actual
characteristics

0 Actual Estimated
Suspension stroke velocity
(b)

Figure 9.23 Sensitivity of variable damping force on estimation error of

suspension stroke velocity

Table 9.6 Simulation parameters

m2 (kg) 350 m1 (kg) 35

k2 (N/m) 23300 k1 (N/m) 233000
c2 (Ns/m) 250 c1 (Ns/m) 571
r 5 e 8
H [9 0.1] b 1500
F [9.77  103 2.15  102 3.63  10 3.71  103]

T
0:2121 0:5528 25:3898 0:0556
K
0:0149 0:0040 0:2121 0:0159

9.5.4 Numerical simulations

This section describes results of numerical simulations to verify the effectiveness of
the proposed VSS observer. As measured outputs from the plant, the acceleration of
the body and the acceleration of the tire with sensor noises are assumed. Table 9.6
 Design of sliding mode controller for semi-active suspension systems 279

presents the simulation parameters, and the sampling time of the controller is 1 ms.
An MR damper, which has the following damping characteristics, is assumed as the
semi-active device:
fc ¼ bu  sgnðx1  x3 Þ ð9:77Þ

–8
0 1 2 3
Time (s)

Figure 9.24 Switching function s

(m/s)
0.05 Plant
VSS obsv.
0

–0.05
0 1 2 3
Time (s)
(a)
(m)
0.02

0
Plant
–0.02 VSS obsv.
0 1 2 3
Time (s)
(b)
(m/s)
1.5

0
Plant
–1.5 VSS obsv.
0 1 2 3
Time (s)
(c)
(m)
0.01

0
Plant
–0.01 VSS obsv.
0 1 2 3
Time (s)
(d)

Figure 9.25 Example of state estimation with proposed VSS observer: (a) body
velocity; (b) suspension stroke; (c) tire velocity; (d) tire deflection
280 Handbook of vehicle suspension control systems

where u is the control input of the MR damper. For the semi-active control law, the
linear optimal state feedback control, whose feedback gain F is shown in Table 9.6,
is employed as
8
< 1 Fx if Fxðx  x Þ > 0
1 3
u¼ b ð9:78Þ
:
0 if Fxðx1  x3 Þ < 0

The observer gain K in (9.60) is selected as shown in Table 9.6 so that the Kalman
filter is designed when the VSS term is neglected. Coping with the chattering pro-
blem caused by the switching and the sensor noises, the following VSS term with the
saturation function is substituted for the original VSS term (9.64):

(m/s)
0.02

0
Linear obsv.
VSS obsv.
–0.02
0 1 Time (s) 2 3
(a)
(m)
0.008

0
Linear obsv.
–0.008 VSS obsv.
0 1 Time (s) 2 3
(b)
(m/s)
1.5

0
Linear obsv.
–1.5 VSS obsv.
0 1 2 3
Time (s)
(c)
(m)
0.005

Linear obsv. VSS obsv.

–0.005
0 1 2 3
Time (s)
(d)

Figure 9.26 Estimation errors of the proposed VSS observer compared to those of
the linear observer: (a) body velocity; (b) suspension stroke; (c) tire
velocity; (d) tire deflection
 Design of sliding mode controller for semi-active suspension systems 281
8
> s
< G r if ksk  e
x , y, rÞ ¼
Sð^ ksk ð9:79Þ
> s
: G r if ksk < e
e
In these numerical simulations, the estimation accuracy of the proposed VSS
observer is compared with that of the Kalman filter without the VSS term, which
makes use of the variable damping force measured by sensors instead of using the

(m/s)
0.01

0
Plant
VSS obsv.
–0.01
1 2 3
Time (s)
(a)
(m)
0.002
0
–0.002
–0.004 Plant
VSS obsv.
–0.006
1 2 3
Time (s)
(b)
(m/s)
0.02 Plant
VSS obsv.
0.01
0
–0.01
1 2 3
Time (s)
(c)
(m)
0.002

–0.002
Plant
VSS obsv.
–0.004
1 2 3
Time (s)
(d)

Figure 9.27 Example of state estimation with the proposed VSS observer against
modeling error not satisfying the matching condition: (a) body velocity;
(b) suspension stroke; (c) tire velocity; (d) tire deflection
282 Handbook of vehicle suspension control systems

damper model. In such steady linear observers, we measure the damping force or
model the damping characteristics as an orifice damper whose damping coefficient
is constant. The former, corresponding to the object of comparison in this section,
is inferior to the proposed VSS observer from a cost perspective.
Figures 9.24 and 9.25 show the estimation results of the proposed VSS
observer when the road disturbance velocity is a normal distribution white noise
signal. From Figure 9.24, it can be seen that the quasi-sliding mode occurs, since
the time response of the switching function s takes a value near zero. As a result,
accurate estimation results are obtained since most effects of the road disturbance
cannot be seen by the matching condition of the VSS term. That is, the proposed
VSS observer has robustness against variation in statistical properties of the road
disturbance. Figure 9.26 compares the estimation errors of the proposed VSS

(m/s)
0.02
0
–0.02 Linear obsv.
–0.04 VSS obsv.
1 2 3
Time (s)
(m) (a)

0.020 Linear obsv.

VSS obsv.
0.010
0
1 2 3
Time (s)
(m/s) (b)
0.05
0
–0.05 Linear obsv.
–0.10 VSS obsv.
1 2 3
Time (s)
(m) (c)
0.003
Linear obsv.
0.002
VSS obsv.
0.001
0
–0.001
1 2 3
Time (s)
(d)

Figure 9.28 Estimation errors of the proposed VSS observer compared with those
of the linear observer against modeling error not satisfying the
matching condition: (a) body velocity; (b) suspension stroke; (c) tire
velocity; (d) tire deflection
 Design of sliding mode controller for semi-active suspension systems 283

observer to those with the Kalman filter. From this figure, it can be seen that the
estimation error of the body’s vertical velocity in the range of the natural frequency
for the sprung-mass system is particularly outstanding, since the linear observer is
not optimized for such road disturbance conditions. To improve the estimation
accuracy in this range, the Kalman filter must have gain scheduling corresponding
to variation in the road disturbance.
Next, the road disturbance is assumed to be a step function with a 1 cm mag-
nitude. Modeling errors of the sprung mass and the tire model as examples
of modeling errors that do not satisfy the matching condition are also assumed.
Figures 9.27 and 9.28 show the estimation results in the case where the sprung mass
m2 changes to 410 (kg) and an attenuation element with the damping coefficient c1
is added to the tire dynamics. From these results, it can be seen that the proposed
VSS observer is able to retain the appropriate estimation accuracy for the possible
modeling errors that do not satisfy the matching condition of the VSS term, thus
supporting the design principle proposed in this study.

References
[1] M. S. Jones, ‘A short history of Moog’s involvement in motorsport’, Moog
Newsletters, no. 10, pp. 1, 2, 2006.
[2] Y. Kawasaki, Y. Fukunaga, S. Hasegawa, Y. Okuyama, T. Kurozu,
‘Development of hydraulic active suspension’, Journal of Society of Auto-
motive Engineers of Japan, vol. 44, no. 1, pp. 120–127, 1990 (in Japanese).
[3] T. Oka, A. Ochiai, Y. Katoh, ‘New Nissan Infinity Q45’, Journal of Society
of Automotive Engineers of Japan, vol. 44, no. 1, pp. 132–136, 1990 (in
Japanese).
[4] Kayaba Industry Co., Ltd., Automotive suspension, Sankaido, Tokyo, Japan,
1994 (in Japanese).
[5] A. Fukukita, M. Takahashi, ‘Semi-active control for base isolated structure
using response evaluator subjected to long-period earthquake ground motion’,
Journal of System Design and Dynamics, vol. 5, no. 8, pp. 1674–1686, 2011.
[6] Y. Li, J. Li, B. Samali, ‘Dynamic performance of a novel magnetorheolo-
gical pin joint’, Journal of System Design and Dynamics, vol. 5, no. 5,
pp. 706–715, 2011.
[7] T. Kikuchi, K. Kobayashi, ‘Design and development of cylindrical MR fluid
brake with multi-coil structure’, Journal of System Design and Dynamics,
vol. 5, no. 7, pp. 1471–1484, 2011.
[8] D. Karnopp, M. J. Crosby, R. A. Harwood, ‘Vibration control using semi-
active force generators’, ASME, Journal of Engineering for Industry, vol. 94,
pp. 619–626, 1974.
[9] T. Suzuki, M. Takahashi, ‘Semi-active suspension control system design for
vibration reduction of passenger’s body based on Lissajous figure of
damping force’, Journal of System Design and Dynamics, vol. 5, no. 2,
pp. 279–295, 2011.
284 Handbook of vehicle suspension control systems

[10] S. H. Zareh, A. A. A. Khayyat, ‘Fuzzy inverse model of magnetorheological

dampers for semi-active vibration control of an eleven-degrees of freedom
suspension system’, Journal of System Design and Dynamics, vol. 5, no. 7,
pp. 1485–1497, 2011.
[11] M. Takahashi, T. Kumamaru, K. Yoshida, ‘Integrated controller design for
automotive semi-active suspension considering vehicle behavior with steering
input’, Journal of System Design and Dynamics, vol. 4, no. 5, pp. 712–724,
2010.
[12] S. M. Savaresi, C. Spelta, ‘A single-sensor control strategy for semi-active
suspensions’, IEEE Transactions on Control Systems Technology, vol. 17,
no. 1, pp. 143–152, 2009.
[13] V. Sankaranarayanan, M. E. Emekli, B. A. Gilvenc, L. Guvenc, E. S. Ozturk,
E. S. Ersolmaz, I. E. Eyol, M. Sinal, ‘Semiactive suspension control of
a light commercial vehicle’, IEEE/ASME Transactions on Mechatronics,
vol. 13, no. 5, pp. 598–604, 2008.
[14] M. Canale, M. Milanese, C. Novara, ‘Semi-active suspension control using
‘‘fast’’ model-predictive techniques’, IEEE Transactions on Control Systems
Technology, vol. 14, no. 6, pp. 1034–1046, 2006.
[15] J. J. H. Paulides, L. Encica, E. A. Lomonova, A. J. A. Vandenput, ‘Design
considerations for a semi-active electromagnetic suspension system’, IEEE
Transactions on Magnetics, vol. 42, no. 10, pp. 3446–3448, 2006.
[16] N. E. Nawa, T. Furuhashi, T. Hashiyama, Y. Uchikawa, ‘A study on the dis-
covery of relevant fuzzy rules using pseudobacterial genetic algorithm’, IEEE
Transactions on Industrial Electronics, vol. 46, no. 6, pp. 1080–1089, 1999.
[17] S. Ohsaku, M. Sanpei, E. Shimizu, K. Tomida, ‘Nonlinear H? state feed-
back controller for semi-active controlled suspension’, Journal of the Society
of Instrument and Control Engineers, vol. 39, no. 2, pp. 126–129, 2000 (in
Japanese).
[18] J. Kitade, K. Yoshida, ‘Design of controller for automotive semi-active
suspension considering both ride comfort and driving stability’, Transac-
tions of the Japan Society of Mechanical Engineers, Series C, vol. 74,
no. 739, pp. 603–609, 2008 (in Japanese).
[19] V. Utkin, J. Guldner, J. Shi, Sliding mode control in electromechanical
systems, Taylor & Francis, London, UK, 1999.
[20] H. Liu, K. Nonami, T. Hagiwara, ‘A fuzzy sliding mode controller of real
vehicle semi-active suspensions: comparison with output feedback sliding
mode control’, Transactions of the Japan Society of Mechanical Engineers,
Series C, vol. 72, no. 720, pp. 2352–2359, 2006 (in Japanese).
[21] M. Yokoyama, J. K. Hedrick, S. Toyama, ‘A sliding mode controller for
semi-active suspension systems’, Transactions of the Japan Society of
Mechanical Engineers, Series C, vol. 67, no. 657, pp. 1449–1454, 2001 (in
Japanese).
[22] S. Toyama, F. Ikeda, ‘A sliding mode control of semi-active suspension
systems with describing function method’, Journal of System Design and
Dynamics, vol. 6, no. 13, pp. 263–272, 2012.
 Design of sliding mode controller for semi-active suspension systems 285

[23] S. Toyama, M. Yokoyama, K. Aida, ‘State estimation using VSS observer for
semi-active suspensions’, Transactions of the Japan Society of Mechanical
Engineers, Series C, vol. 65, no. 639, pp. 2352–2359, 1999 (in Japanese).
[24] T. Kawabe, O. Isobe, Y. Watanabe, Y. Miyasato, S. Hanba, ‘Sliding mode
control of semi-active suspension’, Proceedings of 38th Automatic Control
Allied Meeting, pp. 247–248, 1995.
[25] Y. Kubo, ‘Semi-active control performance of electro-rheological fluid
shock absorber’, Proceedings of 5th Symposium on Motion and Vibration
Control, pp. 345–348, 1997.
[26] A. Alleyne, J. K. Hedrick, ‘Nonlinear adaptive control of active suspensions’,
IEEE Transactions on Control Systems Technology, vol. 3, pp. 94–101, 1995.
[27] H. K. Khalil, Nonlinear systems, Prentice-Hall, Upper Saddle River, NJ,
USA, 1996.
[28] I. Boiko, L. Fridman, M. I. Castellanos, ‘Analysis of second-order sliding
mode algorithms in the frequency domain’, IEEE Transactions on Automatic
Control, vol. 49, no. 6, pp. 946–950, 2004.
[29] K. Yi, J. K. Hedrick, ‘Dynamic tire force control by semiactive suspensions’,
Transactions of the ASME, Journal of Dynamic Systems, Measurements, and
Control, vol. 115, no. 3, pp. 465–474, 1993.
[30] A. Hac, ‘Design of disturbance decoupled observer for bilinear systems’,
Transactions of the ASME, Journal of Dynamic Systems, Measurement and
Control, vol. 114, pp. 556–562, 1992.
[31] H. Nakai, K. Yoshida, S. Ohsaku, Y. Motozono, ‘Design of practical observer
for semiactive suspensions’, Transactions of the Japan Society of Mechanical
Engineers, Series C, vol. 63, no. 615, pp. 3898–3904, 1997 (in Japanese).
[32] B. Walcott, S. Zak, ‘State observation of nonlinear uncertain dynamical sys-
tems’, IEEE Transactions on Automatic Control, vol. 32, no. 2, pp. 166–170,
1987.
 Chapter 10
Joint design of controller and parameters
for active vehicle suspension
Wei Zhan, Qingrui Zhang, Yinan Liu and Huijun Gao

Abstract

This chapter considers a joint design problem for a vehicle active suspension system
to optimize system parameters and a controller with constraints simultaneously.
A model of quarter-car system with an active suspension is presented first. In view of
the practical constraints of the acting force and suspension stroke, a mixed H1 =GH2
control is employed to attenuate the vibration of the system. By considering that both
the controller and parameters of the system have great effect on the control perfor-
mance, this chapter uses genetic algorithm (GA) to search for the desirable para-
meters and obtain the corresponding controller jointly. Simulations are given to
demonstrate the effectiveness and superiority of the joint design approach in com-
parison with the open-loop, non-optimized and partly optimized systems.

10.1 Overview
For the vehicle active suspension systems, the factors which significantly influence
the vibration attenuation effect are not only the designed controllers, but also the
parameters of the suspensions and the tyres. Optimizing the parameters and con-
trollers of the system simultaneously can improve the performance. In the last
20 years, lots of efforts have been made to develop appropriate approach to the
optimization of the parameters. However, only a few researchers have focused on
the joint design which combined controller design and parameter optimization.
Many basic concepts and methods for parameter optimization of suspensions
have been proposed in the 1980s and 1990s. Basic concepts which are useful for
vehicle suspension tuning have been derived in [1–3]. In [4], the suspension or
control system parameters have been made adaptive to changing conditions. The
optimal values of parameters have been found by a numerical procedure in [5]. In
[6], the component designs of a vehicle suspension system have been optimized.

Research Institute of Intelligent Control and Systems, Harbin Institute of Technology, Harbin, Heilongjiang,
China
288 Handbook of vehicle suspension control systems

In the first decade of the twenty-first century, new ideas and techniques have been
employed to the parameter optimization methods. In [7], the optimal settings of
vehicle suspension parameters have been obtained either symbolically or numerically.
The global optimization technique DIRECT has been used in [8] to develop a method
to optimize suspension systems. In order to adequately present the physical char-
acteristics of the practical vehicle suspension, design approaches to nonlinear models
have been developed by many researchers [9–12]. Semi-active control strategy with a
nonlinear actuator has been considered in [13]. Parameters of vehicle suspension
system controlled by magnetorheological (MR) fluid dampers have been optimized.
Many optimization methods have been used to optimize the parameters of the
suspension system, especially for the genetic algorithm (GA). In [14], a one-degree-
of-freedom (1-DOF) vibration isolator mount has been optimized by a GA method
which is extended to the optimization of a quarter-car suspension. Parameters of a
4-DOF quarter car seat and suspension system have been determined by GA in [15].
In [16], the optimal car suspension has been designed by GA with constraints.
For the design of suspension system, many practical constraints should be
considered, such as the suspension stroke and the actuator force [17–20]. By con-
sidering control with constraints and parameter optimization comprehensively, a
joint design method has been proposed in [21]. The coefficients of skyhook dam-
pers and passive mechanical parameters of a vehicle suspension system have been
both determined by GA. However, the controllers were just passive dampers with
constant coefficients. No semi-active or active controller design approach with
appropriate constraints was referred to in [21]. A combined design approach has
been proposed in [41] to optimize the suspension parameters and controller
simultaneously or sequentially. The acting force and stroke has been constrained by
introducing a linear quadratic performance index. The weighting coefficients have
to be adjusted by observing the results to achieve appropriate constraints. In this
chapter, a mixed H1 =GH2 control will be employed to set the bounds of acting
force and stroke. Moreover, the joint design in this chapter will consider not only
active controller and suspension parameters, but also tyre parameters.
In the rest of this chapter, the state-space model of vehicle active suspension
system is established and an active vibration control approach with constraints will
be employed. We will use GA to optimize the controller and the parameters
simultaneously. Simulations are provided to illustrate the effectiveness and super-
iority of the joint design method.
Notation: Rn means the n-dimensional Euclidean space; for a matrix A, the
notation A > 0ð<0Þ denotes that A is real symmetric and positive (negative) defi-
nite; AT and A1 are used to denote its transpose and inverse respectively. I and 0
are used to denote the identity and the zero matrices respectively, of appropriate
dimensions. In symmetric block matrices or complex matrix expressions, an
asterisk () is used to represent a term that is induced by symmetry and symðAÞ is
used to denote A þ AT . Txy denotes the transfer function from x to y and diag
{M1 , M2 , . . . , Mr } stands for a block-diagonal matrix with diagonal blocks
M1 , M2 , . . . , Mr . Matrices whose dimensions are not explicitly stated are assumed
to be compatible for algebraic operations.
 Joint design of controller and parameters for active vehicle suspension 289

cs ks ct kt
w z
Quarter-car
u model x

Joint
design
approach

Controller

Figure 10.1 Control system diagram

Vehicle ms Zs

ks cs u
Suspension

Wheel mu Zu

kt ct
Tyre
Zr

Figure 10.2 Model of a quarter-car with an active suspension

10.2 Problem formulation

In this section, the joint design problem is formulated. The control system diagram
is presented in Figure 10.1. w represents the input disturbance of the road; z is the
output vector which contains the system performance and constraints; x represents
the state vector of the system; u is the acting force of the actuator. In the diagram
we can see that a state-feedback controller needs to be designed:
u ¼ Kx ð10:1Þ
where K is a constant feedback gain. However, it is not the whole task of the system
design. Some parameters of the quarter-car model are also the objectives of the off-
line optimization approach. The problem is to find an appropriate method which
can optimize the controller and parameters of the system jointly, while the vibration
caused by the disturbance is suppressed effectively and the constraints in a practical
system are satisfied.
The quarter-car model in the control system diagram will be considered first.
Such a model with active suspension has been established in [20] and is shown in
Figure 10.2. The influence of actuator dynamics has been neglected in the model.
290 Handbook of vehicle suspension control systems

In this figure, zr represents the road displacement input; u is the acting force of the
suspension system; ms represents the chassis mass of the quarter car; mu represents
mass of the wheel assembly; cs and ks stand for damping and stiffness of the
suspension system respectively; kt and ct are compressibility and damping of
the pneumatic tyre respectively; zs and zu are the displacements of the chassis and
wheel masses respectively. Besides the active force u, system parameters cs , ks , ct and
kt also have great influence on the vibration attenuation performance of the system.
Suppose
x1 ¼ zs  zu , x2 ¼ zu  zr , x3 ¼ z_ s , x4 ¼ z_ u , w ¼ z_ r
The equations governing the motion can be written as
ms x_ 3 þ cs ðx3  x4 Þ þ ks x1 ¼ u ð10:2Þ
mu x_ 4 þ ct ðx4  wÞ þ kt x2  cs ðx3  x4 Þ  ks x1 ¼ u ð10:3Þ
In general, the following aspects should be taken into consideration for the
design of the suspension system:
1. In view of promoting the comfort of the passengers, €z s which represents body
acceleration should be suppressed.
2. Because of the limited space of the quarter-car mechanical structure, the
maximum allowable suspension stroke has to be constrained as
jx1 j  zmax ð10:4Þ
3. By considering that the actuators have limited power, the acting force for the
suspension system should be confined as
juj  umax ð10:5Þ

According to the above three requirements, the following output variables can
be defined:
z1 ¼ €z s
x1
z2 ¼
zmax ð10:6Þ
u
z3 ¼
umax
We can see that the first output needs to be minimized, while the other two are
constraints. Therefore, the problem in this chapter can be formulated as follows: to
develop an appropriate strategy to design the controller gain K and system para-
meters cs , ks , ct and kt , so that the first output is minimized while the constraints in
(10.4) and (10.5) are satisfied.

10.3 Joint design of the system

In this section, a state-space model of vehicle active suspension system is presented
first. Then a controller and parameters for the active suspension system are
designed simultaneously by using the conclusions in [22]. A mixed H1 =GH2
 Joint design of controller and parameters for active vehicle suspension 291

controller is obtained with a group of given parameters cs , ks , ct and kt . Then we

provide a method based on GA to search for the desirable parameters of the system.
Suppose x ¼ ½ x1 x2 x3 x4 T . For the whole system, according to (10.2),
(10.3), and (10.6), the state-space model can be given as [20]
x_ ¼ Ax þ B1 w þ B2 u
z1 ¼ C1 x þ D12 u
ð10:7Þ
z2 ¼ C2 x
z3 ¼ D3 u
in which
2 3 2 3
0 0 1 1 2 3 0
0
6 7 6 7
6 0 0 0 1 7 6 7 6 0 7
6 7 6 1 7 6 7
6 ks cs cs 7 6 7 6 1 7
A ¼6
6 0  7, B 1 ¼ 6
7
7
6 0 7, B2 ¼ 6
6
7
7
6 ms ms ms 7 6 7 6 ms 7
6 7 4 cs 5 6 7
4 ks kt cs cs þ ct 5 4 1 5
  mu 
mu mu mu mu mu
 
ks cs cs 1
C1 ¼  0  , D12 ¼
ms ms ms ms
 
1 1
C2 ¼ 0 0 0 , D3 ¼
zmax umax

After presenting the state-space model of the system, we then design the
controller of the suspension system.
In order to reduce the body acceleration, H1 control is employed, which has
been widely studied and applied in the field of vibration attenuation. In [23–26],
H1 control has been studied for building vibration suppression. References [20]
and [27] have dealt with the problem of H1 control in finite frequency domain for
active suspensions. H1 control has been applied for active suspensions with T–S
fuzzy models in [28]
 and [29].
Suppose Twz1 1 represents the H1 norm from w to z1 . According to the
Bounded
  Real Lemma [30], for system (10.7) we have: for a prescribed g > 0,
Twz  < g is satisfied if and only if there exist X1 > 0 and W1 such that
1 1

2 3
symðAX1 þ B2 W1 Þ B1 ðC1 X1 þ D12 W1 ÞT
6 7
4  gI 0 5<0 ð10:8Þ
  gI
where W1 ¼ KX1
For the two constraints in (10.4) and (10.5),
 generalized H2 (GH2 ) performance
will be an appropriate index. Suppose Twzr g represent the GH2 norm from w to
292 Handbook of vehicle suspension control systems
 
zr ðr ¼ 2; 3Þ. According to [31], we have: for prescribed gr > 0, Twzr g < gr is
satisfied if and only if there exist Xr > 0 and Wr such that
" #
symðAXr þ B2 Wr Þ B1
<0
 I
" # ð10:9Þ
g2r I Cr Xr þ Dr Wr
<0
 Xr

where Wr ¼ KXr :
Specify X ¼ X1 ¼ Xr and W ¼ W1 ¼ Wr ðr ¼ 2;  3Þ. Then we have:
 for
prescribed g > 0, gr > 0, system (10.7) satisfies Twz1 1 < g and
 Twz  < g if
r g r
and only if there exist X > 0 and W such that [22]
2 3
symðAX þ B2 W Þ B1 ðC1 X þ D12 W ÞT
6 7
6  I 0 7<0
4 5
  g21 I
" #
symðAX þ B2 W Þ B1 ð10:10Þ
<0
 I
" #
g2r I Cr X þ Dr W
<0
 X

where W ¼ KX :
In this chapter we set gr ¼ 1ðr ¼ 2; 3Þ, because in the simulation the norm of
the input is much smaller than 1. The problem to seek a desirable controller gain
can be described as follows:
max  g
ð10:11Þ
s:t: X > 0; gr ¼ 1; and ð10Þ

By defining Xe and W e as the values of X and W when g is minimized, we can

express the desirable controller gain as

e ¼W
K eXe 1 ð10:12Þ
According to the problem we have formulated for the quarter-car system in the
former section, our task includes not only designing the controller, but also seeking
the desirable parameters cs , ks , ct and kt . With prescribed parameters cs , ks , ct and
kt , we have obtained an optimal g and corresponding controller gain K. Next a
search should be done in a 4s-dimensional space for the minimized g and desirable
 Joint design of controller and parameters for active vehicle suspension 293

parameters cs , ks , ct and kt . By considering physical realizability, the parameters cs ,

ks , ct and kt should be constrained within certain ranges.
Genetic algorithm is appropriate to accomplish such a task because the ranges
of the parameters are specified naturally in the procedure of the algorithm. GA has
been widely used in control systems. In [25, 32, 33], GA has been used to H1
controller design. In [34], a GA-based fuzzy PI/PD controller has been proposed for
a vehicle active suspension system. In [35], a fuzzy logic controller (FLC) whose
parameters were tuned by GA has been proposed for the active suspension system
control. In [36], the design of a passive vehicle suspension system has been handled
in the framework of nonlinear optimization, which was solved by GA. In [37],
multi-objective optimization has been performed by using a GA. Further details
about GA can be found in the textbook [38].
In this chapter, a standard GA approach is used to find the desirable parameters
cs , ks , ct and kt by solving the maximization problem in (10.11). The main steps of
the joint design approach are presented in Figure 10.3.

10.4 Simulation results

In this section, simulations are used to illustrate the performance of the designed
system. The controller and the parameters of the suspension system are designed
with the approach presented in the former section. Here quarter-car parameters in
[20] are used. The mass coefficients are ms ¼ 320 kg and mu ¼ 40 kg. The sus-
pension parameters are cs ¼ 1 kNs/m and ks ¼ 18 kN/m. The tyre parameters are
ct ¼ 10 Ns/m and kt ¼ 200 kN/m.
To compare the control effects, we consider four systems. The first one is an
open-loop system with no actuator (System I). The next three systems are closed-
loop, whose controllers are H1 controllers designed based on the method in the
former section. The second one is a non-optimized system with the suspension and
tyre parameters given in [20] (System II). The third one is a partly optimized sys-
tem with the tyre parameters in [20] and its suspension parameters are optimized by
the GA-based approach (System III). The last one is an entirely optimized system
whose suspension and tyre parameters are all obtained by the GA-based joint
optimization approach (System IV).
For System II, g ¼ 9:5195; and the corresponding gain is
e ¼ 103  ½ 1:5551
K 0:0482 1:1001 0:0010  ð10:13Þ
For the GA to search the parameters of the system, the initial population is
chosen as N ¼ 120, the crossover rate pc ¼ 0:6, the mutation rate pm ¼ 0:05 and
the intervals of cs , ks , ct and kt are ½0:7; 1:3 kNs/m, ½12; 24 kN/m, ½7; 13 Ns/m
and ½170; 230 kN/m respectively. The constraints are set as umax ¼ 2500 N and
zmax ¼ 0:4 m respectively. For System III, g ¼ 8:9916, and the corresponding gain is
e ¼ 103  ½ 4:7256
K 2:6947 1:7195 0:0082  ð10:14Þ
and the desirable parameters are cs ¼ 775 Ns/m and ks ¼ 12 kN/m.
294 Handbook of vehicle suspension control systems

Start

Specify the bounds of

the parameters cs ks ct kt

Encode cs ks ct kt to a binary strings

Generate an initial population of N

chromosomes randomly

Decode the initial population into the

practical values of csr ksr ctr ktr (r = 1, 2 , . . . , N)

Solve the problem in (10.11) for

each set of parameters csr ksr ctr ktr

~
Obtain –γr and corresponding Kr

Associate every group of parameters

with suitable fitness value

Cross over with a probability pc

Mutate with a probability pm

N Convergent condition
is satisfied?

Y
Output the results γ, K, cs, ks, ct, kt

End

Figure 10.3 Main steps of the joint design

For System IV, g ¼ 8:5102, and the corresponding gain is

e ¼ 103  ½ 4:3753
K 2:6264 1:6726 0:0101 ð10:15Þ
and the desirable parameters are cs ¼ 734 Ns/m, ks ¼ 12 kN/m, ct ¼ 13 Ns/m and
kt ¼ 170 kN/m.
In the following, comparisons are made to illustrate the effectiveness and
advantages of the joint design. The frequency responses of each system are
 Joint design of controller and parameters for active vehicle suspension 295

compared first. In Figure 10.4, the dashed line with dots and the dashed line
without dots are the responses of System I and System II respectively; the dotted
and solid lines are the responses of System III and System IV respectively. The
figure shows that the controller design approach is effective. The peak values of
each closed-loop systems are much smaller than the open-loop system. The vibra-
tion attenuation performance over the frequency range 0.6–2 Hz is very evident.
When the parameters of the suspension are optimized (System III and System IV),
the frequency responses over the frequency range 1–8 Hz are suppressed obviously.
It is worth regarding that the human body is much sensitive to vibrations of 4–8 Hz
in the vertical direction as per ISO 2361 [20]. Therefore, such promotion of
vibration suppression effect is very significant to improve the comfort of human
body. Moreover, when the parameters of both the suspension and tyre are opti-
mized (System IV), the frequency responses over the frequency range 10–11 Hz are
suppressed in comparison with System III. The superiority of System IV is more
obvious in the following simulations in time domain.
After the comparison of the frequency responses among the four systems,
simulations of the closed-loop systems (Systems II, III and IV) in time
domain are given with different excitations. The first excitation is an isolated
bump on an otherwise smooth road surface; the corresponding ground displace-
ment is given by

25
Openloop
Non-optimized
20 Partly optimized
Entirely optimized
Maximum singular value

15

10

0
10−1 100 101
Time (s)

Figure 10.4 Frequency response of body acceleration

296 Handbook of vehicle suspension control systems
8   
>a 2pv0 l
>
<2 1  cos t , 0t
l v0
zr ðtÞ ¼ ð10:16Þ
>
> l
: 0; t>
v0
where a and l are the height and the length of the bump, and v0 is the vehicle
forward velocity. According to [39], the coefficients are given as a ¼ 0:07 m,
l ¼ 0:8 m and v0 ¼ 0:856 m/s. Another road test profile [40] with sinusoidal
components is employed as an excitation; the corresponding road displacement is
given as
zr ðtÞ ¼ 0:0254 sin 2pt þ 0:005 sin 10:5pt þ 0:001 sin 21:5pt ð10:17Þ
The responses of three output variables with each excitation are plotted.
Figures 10.5–10.7 show the response curves with the first excitation. Figures 10.8–
10.10 show the response curves with the second excitation.
These figures clearly show that the System IV whose parameters are entirely
optimized has the best vibration attenuation performance. For System IV the body
acceleration responses with each excitation have smaller peak values and faster
vibration attenuation rate. Such superiority is particularly obvious for the curve
with the second excitation which has high-frequency components.
The suspension strokes and acting forces are also shown in the figures to verify
the effectiveness of the constraints to these variables. The corresponding figures show

Body acceleration
0.8
Non-optimized
0.6 Partly optimized
Entirely optimized
0.4
Acceleration (m/s2)

0.2

−0.2

−0.4

−0.6

−0.8
0 2 4 6 8 10
Time (s)

Figure 10.5 Bump response of body vertical acceleration

Joint design of controller and parameters for active vehicle suspension 297

Suspension stroke
× 10−3
3
Non-optimized
Partly optimized
2
Entirely optimized

1
Stroke (m)

−1

−2

−3

−4
0 2 4 6 8 10
Time (s)

Figure 10.6 Bump response of suspension stroke

Acting force
10
Non-optimized
Partly optimized
Entirely optimized
5

0
u (N)

−5

−10

−15
0 2 4 6 8 10
Time (s)

Figure 10.7 Bump response of acting force

298 Handbook of vehicle suspension control systems

Body acceleration
0.3
Non-optimized
Partly optimized
0.2
Entirely optimized

0.1
Acceleration (m/s2)

−0.1

−0.2

−0.3

−0.4
0 2 4 6 8 10
Time (s)

Figure 10.8 Road test response of body vertical acceleration

Suspension stroke
× 10−3
1.5
Non-optimized
Partly optimized
1 Entirely optimized

0.5
Stroke (m)

−0.5

−1

−1.5
0 2 4 6 8 10
Time (s)

Figure 10.9 Road test response of suspension stroke

 Joint design of controller and parameters for active vehicle suspension 299

Acting force
6
Non-optimized
4 Partly optimized
Entirely optimized

0
u (N)

−2

−4

−6

−8

−10
0 2 4 6 8 10
Time (s)

Figure 10.10 Road test response of acting force

that the strokes and forces are constrained within certain ranges. For a practical sys-
tem, the constraints umax and zmax should be adjusted appropriately according to the
character of the actuator, the space for the suspension and the estimation of the
probable input signals.

10.5 Conclusion
In this chapter a model of quarter-car system with active suspension has been
established and a joint optimization method for obtaining a desirable state-feedback
controller gain and the parameters of the system has been employed. A mixed
H1 =GH2 control has been applied to attenuate the vibration of the system with
constrained acting force and suspension stroke. GA has been used to search the
desirable parameters and obtain the corresponding controller gain.
A quarter-car system with two kinds of excitations has been used in the
simulations to compare the control effects. The simulation result in frequency
domain has demonstrated the effectiveness of the controller design approach with
constraints, which has attenuated the vibration and reduced the response obviously.
It has also shown that better control effects and disturbance attenuation property
can be achieved when the parameters of the system are optimized by GA-based
approach. The simulation results in time domain have clearly shown that the opti-
mization of the parameters can promote the performance of the system and the
constraints of stroke and force are effective.
300 Handbook of vehicle suspension control systems

Acknowledgement
This work was supported by National Natural Science Foundation of China under
Grants 61203035.

References
[1] G. Mastinu, ‘Passive automobile suspension parameter adaptation’, in
Proceedings of the IMechE Conference – Advanced Suspensions. London:
Institution of Mechanical Engineers, 1988.
[2] T. Gillespie, Fundamentals of Vehicle Dynamics. Warrendale, PA: Society
of Automotive Engineering, 1992.
[3] J. C. Dixon, Tyres, Suspension and Handling. Cambridge: Cambridge Uni-
versity Press, 1996.
[4] R. S. Sharp and D. A. Crolla, ‘Road vehicle suspension system design – a
review’, Vehicle System Dynamics, vol. 16, no. 3, pp. 167–192, 1987.
[5] P. Pintado and F. Benitez, ‘Optimization for vehicle suspension I: Time
domain’, Vehicle System Dynamics, vol. 19, pp. 273–288, 1990.
[6] J. D. Castillo, P. Pintado, and F. Benitez, ‘Optimization for vehicle suspension
II: Frequency domain’, Vehicle System Dynamics, vol. 19, pp. 331–352, 1990.
[7] M. Gobbi and G. Mastinu, ‘Analytical description and optimization of the
dynamic behaviour of passively suspended road vehicles’, Journal of Sound
and Vibration, vol. 245, no. 3, pp. 457–481, 2001.
[8] K. Deprez, K. Maertens, and H. Ramon, ‘Comfort improvement by hydro-
pneumatic suspension passive and semi-active using global optimization
technique’, in Proceedings of the American Control Conference, Anchorage,
AK, USA, 2002.
[9] T. Sireteanu and N. Stoia, ‘Damping optimization of passive and semi-active
vehicle suspension by numerical simulation’, Proceedings of the Romanian
Academy Series A, vol. 4, no. 2, pp. 121–127, 2003.
[10] G. Verros, S. Natsiavas, and C. Papadimitriou, ‘Design optimization of quarter-
car models with passive and semi-active suspensions under random road
excitation’, Journal of Vibration and Control, vol. 11, pp. 581–606, 2005.
[11] K. Deprez, D. Moshou, and H. Ramon, ‘Comfort improvement of a non-
linear suspension using global optimization and in situ measurements’,
Journal of Sound and Vibration, vol. 284, pp. 1003–1014, 2005.
[12] M. Bouazara, M. Richard, and S. Rakheja, ‘Safety and comfort analysis of a
3-D vehicle model with optimal non-linear active seat suspension’, Journal
of Terramechanics, vol. 43, pp. 97–118, 2006.
[13] C. W. Zhang, J. P. Ou, and J. Q. Zhang, ‘Parameter optimization and ana-
lysis of a vehicle suspension system controlled by magnetorheological fluid
dampers’, Structural Control and Health Monitoring, vol. 13, pp. 885–896,
2006.
 Joint design of controller and parameters for active vehicle suspension 301

[14] R. Alkhatib, G. N. Jazar, and M. Golnaraghi, ‘Optimal design of passive

linear suspension using genetic algorithm’, Journal of Sound and Vibration,
vol. 275, pp. 665–691, 2004.
[15] O. Gundogdu, ‘Optimal seat and suspension design for a quarter car with
driver model using genetic algorithms’, International Journal of Industrial
Ergonomics, vol. 37, pp. 327–332, 2007.
[16] K. Deb and V. Saxena, ‘Car suspension design for comfort using genetic
algorithms’, in International Conference on Genetic Algorithms, 1997.
[17] A. Zaremba, R. Hampo, and D. Hrovat, ‘Optimal active suspension design
using constrained optimization’, Journal of Sound and Vibration, vol. 207,
no. 3, pp. 240–253, 1997.
[18] G. Georgiou, G. Verros, and S. Natsiavas, ‘Multi-objective optimization of
quarter-car models with a passive or semi-active suspension system’, Vehicle
System Dynamics, vol. 45, no. 1, pp. 77–92, 2007.
[19] A. Shirahatti, P. Prasad, P. Panzade, and M. Kulkarni, ‘Optimal design of pas-
senger car suspension for ride and road holding’, Journal of the Brazilian Society
of Mechanical Sciences and Engineering, vol. 30, no. 1, pp. 66–76, 2008.
[20] W. Sun, H. Gao, and O. Kaynak, ‘Finite frequency H1 control for vehicle
active suspension systems’, IEEE Transactions on Control Systems Tech-
nology, vol. 19, pp. 416–422, 2011.
[21] A. E. Baumal, J. J. McPhee, and P. H. Calamai, ‘Application of genetic algo-
rithms to the design optimization of an active vehicle suspension system’,
Computer Methods in Applied Mechanics and Engineerlng, vol. 163,
pp. 87–94, 1998.
[22] W. Zhan, Y. Cui, Z. Feng, K. C. Cheung, J. Lam, and H. Gao, ‘Joint opti-
mization approach to building vibration control via multiple active tuned
mass dampers’, Mechatronics, vol. 23, no. 3, pp. 355–368, 2013.
[23] C. Wu, J. Yang, and W. Schmitendorf, ‘Reduced-order H1 and LQR control
for wind-excited tall buildings’, Engineering Structures, vol. 20, no. 3,
pp. 222–236, 1998.
[24] I. N. Kar, K. Seto, and F. Doi, ‘Multimode vibration control of a flexible
structure using H1 -based robust control’, IEEE/ASME Transactions on
Mechatronics, vol. 5, pp. 23–31, 2000.
[25] H. Du and N. Zhang, ‘H1 control for buildings with time delay in control via
linear matrix inequalities and genetic algorithms’, Engineering Structures,
vol. 30, pp. 81–92, 2008.
[26] Y. Chen, W. Zhang, and H. Gao, ‘Finite frequency H1 control for building
under earthquake excitation’, Mechatronics, vol. 20, pp. 128–142, 2010.
[27] W. Sun, Y. Zhao, J. Li, L. Zhang, and H. Gao, ‘Active suspension control
with frequency band constraints and actuator input delay’, IEEE Transac-
tions on Industrial Electronics, vol. 59, pp. 530–537, 2012.
[28] H. Li, J. Yu, C. Hilton, and H. Liu, ‘Adaptive sliding mode control for
nonlinear active suspension systems using T–S fuzzy model’, IEEE Trans-
actions on Industrial Electronics, 2012.
302 Handbook of vehicle suspension control systems

[29] H. Li, H. Liu, H. Gao, and P. Shi, ‘Reliable fuzzy control for active sus-
pension systems with actuator delay and fault’, IEEE Transactions on Fuzzy
Systems, vol. 20, pp. 342–357, 2012.
[30] B. D. O. Anderson and S. Vongpanitlerd, Network Analysis and Synthesis.
Prentice-Hall Networks Series, Englewood Cliffs, NJ: Prentice-Hall, 1973.
[31] S. Boyd, L. E. Ghaoui, E. Feron, and V. Balakishnan, Linear Matrix
Inequalities in System and Control Theory. Philadelphia, PA: SIAM, 1994.
[32] Y. Arfiadi and M. Hadi, ‘Optimal direct (static) output feedback controller
using real coded genetic algorithms’, Computers and Structures, vol. 790,
pp. 1625–1634, 2001.
[33] H. Du, J. Lam, and K. Y. Sze, ‘Non-fragile output feedback H1 vehicle
suspension control using genetic algorithm’, Engineering Applications of
Artificial Intelligence, vol. 16, pp. 667–680, 2003.
[34] Y.-P. Kuo and T.-H. S. Li, ‘GA-based fuzzy PI/PD controller for automotive
active suspension system’, IEEE Transactions on Industrial Electronics,
vol. 46, no. 6, pp. 1051–1056, 1999.
[35] M. Montazeri-Gh and M. Soleymani, ‘Genetic optimization of a fuzzy active
suspension system based on human sensitivity to the transmitted vibrations’,
Proceedings of the Institution of Mechanical Engineers, Part D: Journal of
Automobile Engineering, vol. 222, pp. 1769–1780, 2008.
[36] L. Sun, X. Cai, and J. Yang, ‘Genetic algorithm-based optimum vehicle
suspension design using minimum dynamic pavement load as a design cri-
terion’, Journal of Sound and Vibration, vol. 301, pp. 18–27, 2007.
[37] J. S. Hwang, S. R. Kim, and S. Y. Han, ‘Kinematic design of a double
wishbone type front suspension mechanism using multi-objective optimiza-
tion’, in 5th Australasian Congress on Applied Mechanics, ACAM, Brisbane,
Australia, 2007.
[38] R. L. Haupt and S. E. Haupt, Practical Genetic Algorithms. Wiley, 2004.
[39] H. Metered, P. Bonello, and S. Oyadiji, ‘An investigation into the use of
neural networks for the semi-active control of a magnetorheologically
damped vehicle suspension’, Proceedings of the Institution of Mechanical
Engineers, Part D: Journal of Automobile Engineering, vol. 224, no. 7,
pp. 829–848, 2010.
[40] P. C. Chen and A. C. Huang, ‘Adaptive sliding control of active suspension
systems with uncertain hydraulic actuator dynamics’, Vehicle System
Dynamics, vol. 44, no. 5, pp. 357–368, 2006.
[41] S. F. Alyaqout, P. Y. Papalambros, and A. Galip Ulsoy, ‘Combined design
and robust control of a vehicle passive/active suspension’, International
Journal of Vehicle Design, vol. 59, 315–330, 2012.
 Chapter 11
System approach to vehicle suspension system
control in CAE environment
Vladimir M. Popović and Dragan D. Stamenković

Abstract

In recent years, the motor vehicle industry has shown a tendency of replacing
electro-mechanical components by mechatronic systems with intelligent and auton-
omous properties. The integration of hardware components and implementation of
advance control function characterize this replacement. In this text, we have applied
the system approach and system engineering methods in the initial phase of vehicle
active suspension development. An emphasis has been placed upon the interrelations
between computer-aided simulation and other elements of development process. The
benefits of the application of active suspension simulation are numerous: reduction of
time to market, the new and improved functions of mechatronic components/devices,
and the increased system reliability. In suspension model development, we used CAD/
CAE tools, as well as the multipurpose simulation programs. For simulation, we used
the quarter-car model. The modeling was carried out through the state-space equation,
after which we designed two variants of controller for the suspension system –
proportional-integral-derivative (PID) controller and neural network controller.

11.1 Introduction
Considerable effort aiming at reducing the cost, the ever increasing expectations of
users when it comes to reliability, the greater complexity of modern technical sys-
tems, and other requests help us draw the conclusion that the approach based on
system engineering principles remains the only way to stay competitive on the market
[1]. Intensifying the integration of electronic components into areas that were earlier
strictly mechanical, the degree of change within technologies, all result in increased
demands for dealing with the problems on the basis of what is offered by system
engineering. In recent years, the motor vehicle industry has started developing an
unusual trend – the mechatronic systems with intelligent and autonomous abilities.
Development of integrated mechatronic systems should, in its own way, play the key
role in the motor vehicle industry [2]. More and more mechatronics is being accepted

Faculty of Mechanical Engineering, University of Belgrade, Kraljice Marije 16, Belgrade, Serbia
304 Handbook of vehicle suspension control systems

as design methodology for all motor vehicle systems. Verification of proper func-
tionality through simulation can improve reliability and reduce component design
time. Replacement of simple electro-mechanical components by ‘‘intelligent’’
mechatronic systems is characterized by two aspects: integration of hardware com-
ponents and implementation of advanced control functions (functional and algorithm
integration). Hence, mechanical components are operated through basic sensor
feedback of low level and intelligent processing of advanced level information.
The goal in active suspension control research is to improve the ride perfor-
mance, generally quantified by sprung mass acceleration, while maintaining an
acceptable level of suspension stroke and tire deflection as packaging and handling
measures. Ride comfort, road holding ability, and suspension deflection are the
three main performance criteria in any vehicle suspension design [3]. Finally, all
control concepts aim at introducing additional forces to the suspension system to
reduce roll, pitch, and stroke movements, as well as body and wheel vibrations. So,
the basic control approach is similar for all actuation systems. Then, the physical
structure of the considered system determines the further controller design and also
directly limits the achievable system’s performance [4].

11.2 Classification of mechatronic suspension systems

Mechatronic suspension systems are classified, according to the type of actuators
used, system complexity, and energy consumption, into the following three groups:
Slowly variable/adaptive suspension systems – functioning of this type of sys-
tems is based on a slow variation of the damper and spring characteristics.
These characteristics are varied based on the vehicle velocity and/or vehicle
load. In the first case, the center of gravity is lowered with the increase in
velocity to ensure better road holding. In the second case, the distance
between the vehicle chassis and the road is kept constant to compensate for
changes in vehicle loading. This is usually realized by using air springs.
Components of these systems cannot supply energy to the system, so gen-
erated forces depend on the direction of relative motion of system elements.
Semi-active suspension systems – these systems are based on a fast adjustment
of the damper and/or spring characteristics. As with the slowly variable/
adaptive suspension systems, generated forces in semi-active suspension
systems depend on the direction of travel of suspension system elements.
Semi-active suspension systems with conventional semi-active dampers are
based on pneumatic, hydraulic, or hydro-pneumatic components. In such
systems, the characteristics of damper are changed by changing the volume
of fluid in damper. This can be done by varying the cross-sectional area of
the bypass between the two chambers of the damper’s piston. In this way a
single damper can be set to multiple performance curves. Depending on the
way the cross-sectional area is changed, the damper can have fixed step or
continuous characteristics. The second type of semi-active suspension sys-
tems is one with electro-rheological or magneto-rheological dampers. Such
systems base their functioning on electro-rheological or magneto-rheological
System approach to vehicle suspension system control in CAE environment 305

properties of liquids. By applying electric or magnetic field, viscosity of fluid

in the damper is altered to attain desirable performance. Electro-rheological
dampers are characterized by a fast response to an electric field, while
magneto-rheological dampers are characterized by high yield strength
and stable behavior over wide temperature range. Major drawback of these
systems is a lack of long-term stability of the fluid properties.
Active suspension systems – these systems comprise passive components (con-
ventional springs and dampers) and active components (actuators). Hydraulic,
hydro-pneumatic, pneumatic, or electro-mechanical actuators generate suspen-
sion forces independently of the relative motion of the chassis and wheels. Active
suspension systems are complex, are expensive, and demand a lot of energy.
Active components of active suspension systems can be combined with semi-
active dampers to make a good compromise between the cost, complexity, and
energy efficiency. Presented classification is also shown in Table 11.1 [4, 5].

Table 11.1 Classification of suspension systems

System System Force range Operating Energy

representation range demand
Passive F No actuators 0W

Y1
Ẏ1

Slowly variable/adaptive F <1 Hz Low

Y1
Ẏ1

Semi-active F 0–40 Hz Low

Y1
Ẏ1

Active partially loaded F 0–5 Hz Medium

Y1
Ẏ1

Active fully loaded F 0–30 Hz High

Y1
Ẏ1
306 Handbook of vehicle suspension control systems

11.3 Design development process

Design procedure, which is to be carried out in the case of mechatronic systems, is
very demanding – it is necessary that system engineering, within the field of
mechanics, electronics, and computer science, forms a completely integrated sys-
tem. Hence, intelligent testing technologies, supported by CAE, which provides
numerical simulation models, are to be employed during component development
and their qualification. An emphasis should be placed upon the interaction between
computer-aided system simulation and experimental testing techniques through
intelligent information processing. CAE is a technology that enables computer
analysis of the design, created within CAD technology. CAE technology enables a
computational analysis of the design created using CAD technology. Application of
CAE technology contributes to a better linkage between design, testing, and design
improvement, the three phases in development process that have, so far, been
almost separate [6]. Figure 11.1 shows a diagram of design development process
[7]. After theoretical modeling and experimental system identification, the fol-
lowing design steps are taken for control system analysis and model-based con-
troller design. The demands for efficient optimization and testing include:
● Software development environment with accurate, firm connections among dif-
ferent CAE tools, such as CAD, FEM, and MBS (multi-body simulation) soft-
ware, and CACE (computer-aided control engineering) tools. An example of
such an environment is shown in Figure 11.2 [8]. On the basis of this diagram, we
can determine the role of MATLAB (and Simulink) in the framework of the
complete process of modeling and simulation of the new technical system;
● Experimental, high-performance platform, which, in combination with software
development environment, completely automates the implementation process
from off-line simulation to engaging complex controllers with real-time operation.

DESIGN DEVELOPMENT PROCESS

CAD STATION

Create Build/test Final

Design Design
drawing & prototype to optimized
specification concept solid model verify model design

CAE at
CAE at CAE at
concept
CAD stage test stage
stage

CAE STATION

DESIGN MODEL OPTIMIZATION

Figure 11.1 Design development process

System approach to vehicle suspension system control in CAE environment 307

FEM Physical
Nastran modeling

linearization
CAD
ProEngineer/
CATIA
MBS CACE
Adams/ MATLAB
Simpack Simulink

symbolic code
Computer aided
Computer-aided
CAD control system
modeling
analysis/design

Figure 11.2 Software development environment with connections among different

CAE tools

x1 Body mass

Active components
Actuator

Passive Signal Energy

k1 b1 Controller
components input

x2 Sensing
Suspension mass
device

k2 b2
Z
Road profile

Figure 11.3 Quarter-car model

Interactive testing of mechatronic components using advanced hardware

simulators as feedback, plays an important role in improving the design devel-
opment process. The idea is to build in mechatronic components into virtual
environment in which vehicle movement, external load, and the surrounding
mechanical systems are simulated in real time. A complex model of vertical
vehicle dynamics has been developed using CAE tools and multipurpose simu-
lation programs. On the basis of reduced models, a multivariable, powerful con-
trol system should be designed for active suspension control. Figure 11.3 [7, 8]
308 Handbook of vehicle suspension control systems

shows the corresponding quarter-car model with a passive spring that bears static
load, which is to be used for suspension system simulation.

11.4 Active suspension system modeling

Various approaches have been proposed to improve the performance of active sus-
pension designs, such as linear optimal control, fuzzy logic, and neural network
control, adaptive control, H? control, nonlinear control, LQG (linear quadratic
Gaussian) control, skyhook control algorithm, etc. [3, 4, 7–11]. Also, many
approaches are presented to deal with the multi-objective requirements of vehicle
suspensions. In particular, H? control of active suspensions is intensively discussed
in the context of robustness and disturbance attenuation [3]. An interesting solution
is a fuzzy sliding mode controller to control an active suspension system and eval-
uate its control performance [12–15]. The fuzzy sliding mode controller employed
the error of the sprung mass position and the error change to establish a sliding
surface, and then introduced the sliding surface and the change of the sliding surface
as input variables of a traditional fuzzy controller in controlling the suspension
system [16]. One of the possible solutions is vibration control of vehicle active
suspension system using a new robust neural network control system [17]. It consists
of a robust feedback controller and feedforward neural network predictive controller.
In the design of a suspension system, the following are used: quarter-car model
[3, 4, 7–11, 14, 16], half-car model [18], and full-car model [17, 19]. We have used
the quarter-car model (Figure 11.3) to simplify the problem to a one-dimensional
spring-damper system. The vehicle represents a complex oscillatory system with a
great number of degrees of freedom. Vehicle oscillations are caused mainly by road
disturbance. In reviewing vehicle oscillations we have adopted the following
starting points:
● the vehicle is in the rectilinear motion at a constant speed;
● the wheels are always in contact with the road and that is a one-spot contact;
● road disturbance is the same on the left wheel and the right wheel and the car is
symmetrical relative to the longitudinal axis;
● mass distribution coefficient is approximately 1.
Although the real characteristics of the suspension system (k1, b1) are non-
linear, we have adopted constant values for them which, to a certain extent, enable
the linearization of the model. Also, the step function that we used for simulation
has certain limitations. Namely, if the tire of the vehicle is excited by the step
function, the tire will jump and the contact with the driving surface will be lost.
Furthermore, the damping of the suspension damper when compressed is many
times smaller that the damping when the damper is extended. These limitations will
be further discussed in the conclusion. The data that we used during the simulation
refer to the bus. The simulation was performed using MATLAB interactive envir-
onment. The designations on Figure 11.3 have the following meanings [7, 8]:
● body mass m1 ¼ 2,250 kg;
● suspension mass m2 ¼ 290 kg;
System approach to vehicle suspension system control in CAE environment 309

● spring constant of suspension system k1 ¼ 72,000 N/m;

● spring constant of wheel and tire k2 ¼ 450,000 N/m;
● damping constant of suspension system b1 ¼ 315 Ns/m;
● damping constant of wheel and tire b2 ¼ 13,500 Ns/m;
● control force Fa – force from the controller we are going to design.
When the vehicle is experiencing any road disturbance, the body should not
have large oscillations, and the oscillations should dissipate quickly. This, at the
same time, is our principal task. Since the distance x1 – Z is very difficult to
measure and the deformation of the tire x2 – Z is negligible, we will use the distance
x1 – x2 instead of x1 – Z as the output in our problem. The road disturbance Z will be
simulated by a step input. This step could represent the vehicle coming out of a
pothole. We want to design a feedback controller so that the output x1 – x2 has an
overshoot less than 5% and a settling time shorter than 5 s. Dynamic properties of
the system can, in time range, be most suitably defined by the values of parameters
that determine the system response (overshoot and settling time) [20].

11.4.1 Model of the system in state-space

The assessment of the quality of automatic regulation of a system behavior essen-
tially boils down to estimating the error between a predetermined value and the real
value of the controlled variable. Knowing this error at any point would give a
complete information on the features of the observed system. Due to the diversity
of the laws of the system input change, which might occur in the normal work
regime, such an approach, based on estimating the current values of error, is
inappropriate speaking from the aspect of practice. Therefore, it is preferable to
make an estimation of the relevant system characteristics on the basis of the fea-
tures it manifests while being disturbed by the typical input signals [20]. On the
basis of Figure 11.3 and the Newton’s law, we can obtain the following dynamic
equations, which represent the mathematical model of the dynamic system [7, 8]:

m1€x 1 ¼ b1 ðx_ 1  x_ 2 Þ  k1 ðx1  x2 Þ þ Fa ð11:1Þ


m2€x 2 ¼ b1 ðx_ 1  x_ 2 Þ þ k1 ðx1  x2 Þ þ b2 Z_  x_ 2 þ k2 ðZ  x2 Þ  Fa ð11:2Þ

When talking about the new ways for system analysis and synthesis, that avoid
the problem of solving algebraic and differential equations, one mainly bears in
mind the system analysis and synthesis in state-space. Representation of the system
in the form of state-space equation is much easier to derive from differential
equations than by using Laplace transforms. To be a valid state-space representa-
tion, the derivative of all states must be in terms of inputs and the states themselves.
Now, let us choose the states that we shall be using. To begin, let us divide (11.1)
and (11.2) by m1 and m2, respectively, and introduce the substitute Y1 ¼ x1 – x2.
Note that Z_ appears in (11.4):
b1 _ k1 Fa
€x 1 ¼  Y1  Y1 þ ð11:3Þ
m1 m1 m1
310 Handbook of vehicle suspension control systems

b1 _ k1 b2
_  k2 Fa
€x 2 ¼ Y1 þ Y1 þ Z  x_ 2 þ ðZ  x 2 Þ  ð11:4Þ
m2 m2 m2 m2 m2
The first state will be x1. Since no derivatives of the input appear in the
equation for €x 1 , we choose x_ 1 for the second state. Then, we choose as the third
state the difference between x1 and x2. After algebraic calculations, we will deter-
mine what the fourth state should be. So we subtract (11.4) from (11.3) to get an
expression for Y€ 1 :
   
€ b1 b1 _ k1 k1 b2
_ 
€x 1  €x 2 ¼ Y 1 ¼  þ Y1  þ Y1  Z  x_ 2
m1 m2 m1 m2 m2
 
k2 1 1
 ðZ  x 2 Þ þ F a þ ð11:5Þ
m2 m1 m2
Since we cannot use second derivatives in the state-space representation, we
integrate (11.5) to get Y_ 1 :
  ð  
b1 b1 b2 k1 k1
Y_ 1 ¼  þ Y1  ðZ  x2 Þ þ  þ Y1
m1 m2 m2 m1 m2
 
k2 1 1
 ðZ  x 2 Þ þ F a þ dt ð11:6Þ
m2 m1 m2
No derivatives of the input appear in this equation, and Y_ 1 is expressed in
terms of states and inputs only, except for the integral. Let us call the integral Y2.
Assuming that x2 ¼ x1 – Y1, from (11.6) we get the state-space equation for Y1:
 
_Y 1 ¼  b1 þ b1 Y1  b2 ðZ  x1 þ Y1 Þ þ Y2 ð11:7Þ
m1 m2 m2
Then we shall substitute the derivative of Y1 into (11.3):
   
b1 b2 b1 b1 b1 þ b2 k1 b1 b2 Fa b1
€x 1 ¼  x1 þ þ  Y1 þ Zþ  Y2
m1 m2 m1 m1 m2 m1 m1 m2 m1 m1
ð11:8Þ
The state variables are x1, x_ 1 , Y1, and Y2. The matrix from (11.8) is:
2 3 2 3
0 1 0 0 0 0
2 3 6 b b   72 3
x_ 1 6 1 2 b 1 b 1 b 1 þ b2 k1 b1 7 x 1 6 1 b1 b2 7
6 0 þ   7 6 7
6 7 6 m1 m2 m1 m1 m2 m1 m1 76 7 6 m1
7
m1 m2 7" #
6 €x 1 7 6 76 x_ 1 7 6 7 Fa
6 7 6   76 7 6 6
6 7 ¼ 6 b2 b 1 b 1 þ b2 76 7þ 6 b2 7
7
6 Y_ 1 7 6 0  þ 1 76 7 0 
4 5 6 m2 m1 m2 74 Y1 5 6 6 m2 77
Z
6 7 6 7
Y_ 2 6   7 Y2 4 1
4 k2 k1 k1 þ k2 5 1 k2 5
0  þ 0 þ 
m2 m1 m2 m1 m2 m2
ð11:9Þ
System approach to vehicle suspension system control in CAE environment 311
2 3
x1
6 7 " #
6 x_ 1 7 Fa
6 7
Y ¼ ½0 0 1 0 6 7 þ ½0 0 ð11:10Þ
6 Y1 7 Z
4 5
Y2

We can put the above state-space equations (11.9) and (11.10) into MATLAB
by defining the four matrices of the standard state-space equations: X_ ¼ AX þ BZ;
Y ¼ CX þ DZ. Thus, we have created a new m-file, formed on the basis of state-
space equation. By adding ‘step(A,B,C,D,1)’ command into the m-file and acti-
vating it in the MATLAB command window, we get an open-loop response for the
unit step actuated force. Normally, new functions are added to the MATLAB
vocabulary by expressing them in terms of existing functions. The existing com-
mands and functions that compose the new function reside in a text file called an
m-file [21]. M-files can be either scripts or functions. Scripts are simply files
containing a sequence of MATLAB statements. Functions make use of their own
local variables and accept input arguments. The name of an m-file begins with an
alphabetic character and has a filename extension of ‘‘.m.’’
Figure 11.4 shows that the system is underdumped. People sitting in the bus
will feel a small amount of oscillation and the steady-state error is about 0.01 mm.
However, the bus needs an unacceptably long time to reach the steady state – the
settling time is rather long. The solution to this problem lies in including a feedback

•10–5 Step response

2.5

2
Amplitude

1.5

0.5

0
0 10 20 30 40 50 60
Time (s)

Figure 11.4 Open-loop response for the unit step actuated force
312 Handbook of vehicle suspension control systems

controller into the block diagram of the system. By the term ‘‘controller’’ we also
imply the controller and the actuator. Adding the command ‘step(A,0.1*B,C,D,2)’
into the m-file we get the open-loop response to 0.1 m step disturbance.
Figure 11.5 shows that, when the bus passes a 10 cm high bump on the road,
the bus body will oscillate for an unacceptably long time (50 s), and with a much
larger amplitude than the initial impact. The big overshoot and the slow settling
time will cause damage to the suspension system. As we have already stated, the
solution to this problem is to add a feedback controller into the system to improve
the performance. The block diagram of the thus obtained closed-loop system is
presented in Figure 11.6.

Step response
0.08

0.06

0.04

0.02
Amplitude

–0.02

–0.04

–0.06

–0.08

–0.1

0 10 20 30 40 50
Time (s)

Figure 11.5 Open-loop response to 0.1 m step disturbance

+
r=0 + X1 – X2
Controller Plant
+ Fa
–

Figure 11.6 Closed-loop system block diagram

System approach to vehicle suspension system control in CAE environment 313

11.4.2 Synthesis of active suspension digital system

Apart from the PID algorithm and neural networks, which we will use for con-
trolling an active vehicle suspension system, in literature we can find many modern
solutions for control systems and actuators themselves:
● H? control with actuator time delay [3];
● Impedance control for the vehicle suspension system by electro-hydraulic
actuator – the electro-hydraulic actuator comprises a servo-valve and a
hydraulic cylinder [9];
● Hybrid control technique applied to a vehicle active suspension system using
skyhook and adaptive neuro active force control [11]. The overall control
system essentially comprises four feedback control loops, namely the inner-
most proportional–integral control loop for the force tracking of the pneumatic
actuator, the intermediate skyhook, and active force control loops for the
compensation of the disturbances and the outermost PID control loop for the
computation of the optimum target/commanded force. Pneumatic actuators
have the advantages of a low cost, a high power-to-weight ratio, ease of
maintenance, and having a readily available and cheap power source;
● The different kinds of rotational actuators, electric leveling actuators, electro-
mechanical actuators, and electric damper actuators [10];
● The optimization technique used for the optimization of the controller para-
meters and the spring rates based on a genetic algorithm [10]. The main
advantage of this method and all other nongradient-based optimization meth-
ods is that they do not need the derivatives of the objective function, which are
difficult to calculate from a numeric simulation model;
● Knowledge-based fuzzy logic controllers, variable structure controllers, and
backstepping controllers [19];
● Possible solutions are dampers taking advantage of the electro-rheological or
magneto-rheological properties of liquids. These systems are mainly applied
with semi-active vehicle suspensions [4, 22];
● Somewhat older solutions are hydraulic, hydro-pneumatic, electro-mechanical,
and pneumatic systems. Most of these systems also comprise passive elements
such as conventional springs and dampers or hydro-pneumatic damping, and
springing components involved in the actuation system [4];
● In the integrated intelligence type implementation, actuator intelligence and
actuation are combined into a single material. A piezoceramic actuator is a
typical example of this type [23].

11.4.3 Active suspension control using PID controller

First, we will consider controlling our suspension system using the PID controller.
Actuators may have undesirable static and dynamic characteristics introducing
many problems for the PID control loop. For instance, limit cycles caused by
actuator backlash cannot be removed by adjusting PID parameters. Therefore, a
nonlinearity compensation algorithm using an inverse model and a simple position
controller is implemented in order to maintain the actuator characteristics to be as
314 Handbook of vehicle suspension control systems

linear as possible within the attainable actuation range [23]. PID controller has been
most widely employed because of its simple structure and the effective use in
industry [24, 25]. Despite many advances for PID controller, this structure has
constant gain parameters and is not good to decrease velocity control error [17].
When it comes to this text, the simplest conventional PID controllers are obtained
through an easy discretization of analogous equivalents. By the application of
z-transformation we get the function of incremental PID controller discrete transfer
in the form of:
 
U ðzÞ T =TI TD
1

Dp ðzÞ ¼ ¼K 1þ þ 1  z ð11:11Þ
E ðz Þ 1  z1 T
or
KI

Dp ðzÞ ¼ KP þ þ KD 1  z1 ð11:12Þ
1z 1

where KP ¼ K, KI ¼ KT/TI, and KD ¼ KTD/T are known as P-action, I-action, and

D-action factors, respectively (K is the factor of gain; TI and TD are time constants
of integral and differential actions).
The structure of the system with a digital PID controller is given in
Figure 11.7 [20]. Note that, without the block with dashed lines in Figure 11.7(a),
the regulation contour contains an incremental PID controller. With this block,
however, this contour becomes equivalent to the contour in Figure 11.7(b), which
contains a positional PID regulation law. Positional type of digital law seems
natural; however, the incremental kind is more readily used. Incremental algo-
rithm must be used if it is immediately connected to the executive organ of the
stepping motor type, and it is also convenient for it works easily even in the case
when the executive organ output is in border line position. Also, it is important to
emphasize that there are numerous methods of on-hardware optimization and
improvement of the performance of open-loop stepping motor system dynamics
[26]. The principal defect of incremental algorithm is ‘‘not being able to see’’ the
position of executive organ output. Hence, when proportional executive or inte-
grating (but included by feedback) organ is used, the positional form of PID
algorithm must be used, or the incremental form must be converted into the
positional, which essentially boils down to implementation of the block with
dashed lines in Figure 11.7(a). By adopting a smart actuator technology, standard
PID control techniques can be used successfully in the presence of undesirable
actuator characteristics. The smart actuator could be applied to online PID con-
troller retuning using a standard pole-placement technique to counteract degraded
actuator performance [23].
Our task is to design a digital controller, in state-space, for suspension system
control. First, we should convert continual into discrete model, and then use pole-
placement method, as one of the possible solutions, in order to design a controller.
The state-space model is given in (11.9) and (11.10). The first step in converting
the continual into its discrete equivalent is the choice of suitable sampling time T.
System approach to vehicle suspension system control in CAE environment 315

r e u 1 u m c
KI D/A Object
T 1 – z–1 T
I-action P-action

KP(z–1 – 1)
T

KD(z–2 – 2z–1 + 1)
T
D-action

(a)

r e KI ui u m c
D/A Object
T 1 – z–1 T
I-action P-action
KP
T

KD(1 – z–1)
T
D-action

(b)

Figure 11.7 Structure of the system with a digital PID controller: (a) incremental
type and (b) positional type

This is a highly important step, since road disturbance very quickly influences the
output. Because the controller can ‘‘see’’ only the consequence of the disturbance
after a complete sampling time, we shall have to choose a sampling time T, short
enough, such that the (x1 – x2) output will not exceed the initial requirement of 5%
in one selection time. To choose a selection time, we should carefully consider the
initial part of the system response graph. Figure 11.5 tells us that the output soon
assumes negative values, and then starts to oscillate. We shall simulate only the
beginning of this response by adjusting the time vector in 0–0.005 range. The
response to 0.1 m step disturbance is simulated by multiplying B matrix by 0.1. Into
the already formed m-file we added the command ‘step(A,.1*B,C,D,2,0:0.0001:
.005)’ and got the open-loop response to 0.1 m step disturbance. This graph shows
that the spring with the deflection rate k1 compresses rather quickly, and that our
system exceeds the initial requirement of 5 mm response to 0.1 m step disturbance
after only a little more than 0.001 s. Therefore, we adjusted T to 0.0005 s to give
316 Handbook of vehicle suspension control systems

the controller a chance to react. After we have chosen the sampling time, we can
convert the system into the discrete form. We can use MATLAB to convert the
above presented state-space model of the system, using A, B, C, and D matrices,
into a discrete state-space model (using Ad, Bd, Cd, and Dd) by ‘c2dm’ command.
This command normally has six arguments: four space matrices, sampling time T,
and the type of circuit hold. In this example we used the zero-order hold – ‘zoh’.
Let us add the following commands into our m-file:

T ¼ 0.0005;
[Ad Bd Cd Dd] ¼ c2dm(A, B, C, D, T, ‘zoh’);

In this way, we get the response that represents a new discrete state-space
model of the system. Also, in active control of vehicle suspension systems, the time
delay of the system is another important issue that needs careful treatment to
avoid poor performance or even possible instability of the closed-loop system.
Unavoidable time delays may appear in the controlled channel, particularly in the
digital controller as it carries out the calculations associated with complex sophis-
ticated control law, and in sensors and actuators hardware such as hydraulic
actuators where the delays are taken by the actuators to build up the required
control force [3]. Though the delay time may be short, it can nevertheless limit the
control performance or even cause the instability of the system when the delay
appears in the feedback loop.

11.4.3.1 Adding an integrator

The next step is adding an integrator into the system, so that the system response in
the stationary state is zero. We will add this integrator in series with the plant. This
will result in adding other system states. We shall add an integrator by representing
it in state-space and by using the ‘series’ command. This command takes A, B, C,
and D matrices of the two systems, links them in a serial connection as arguments
and brings us back a new set of A, B, C, and D matrices. The integrator in discrete
state-space may be represented in any time T in the following way:

xðk þ 1Þ ¼ xðk Þ þ Tuðk Þ ð11:13Þ

and
T
y ðk Þ ¼ x ðk Þ þ u ðk Þ ð11:14Þ
2
To do this, let us add the following command into the m-file:

Ai ¼ 1; Bi ¼ 1; Ci ¼ T; Di ¼ T/2;
[Ada Bda Cda Dda] ¼ series(Ad, Bd, Cd, Dd, Ai, Bi, Ci, Di);

MATLAB will respond to us with a new set of matrices, which are the result of
integrator action, with matrix dimension 5 instead of the earlier dimension 4.
System approach to vehicle suspension system control in CAE environment 317

Ada ¼
1.0000 0 0 1.0000 0
0 1.0000 0.0005 –0.0000 –0.0000
0 –0.0034 1.0000 –0.0125 –0.0001
0 0.0232 0.0000 0.9762 0.0005
0 0.7652 0.0002 –0.9051 0.9998
Bda ¼
0 0
0.0000 0.0000
0.0000 0.0034
0.0000 –0.0232
0.0000 –0.7652
Cda ¼
1.0e–003 * (0.5000 0 0 0.2500 0)
Dda ¼
0 0

Unfortunately, the output of this equation is a new integral value. We must

change the matrix output Cda, in which case we get:

Cda ¼ [Cd 0] Cda ¼ 0 0 1 0 0

The controller structure is similar to that of the state-space controller with

continual systems. We shall now use the ‘place’ command, so as to calculate the
gain matrix K, which will provide us the desired poles of the closed-loop regulation
system. First, we should decide where to position the closed-loop poles. Since we
are to determine the position of all the five closed-loop poles, we must be very
selective when it comes to this problem. To be more precise, we can define them so
that they cancel all the system zeros, and at the same time give us the desired
response. First we should track all the zeros by converting digital state-space
equations into the transfer function, and then by determining the square root of the
numerator. We shall use the ‘ss2tf ’ command which takes state-space matrices and
the chosen input as arguments, and gives us the transfer function of the numerator
and denominator as the output. Let us add the following code into the m-file:
[num, den] ¼ ss2tf (Ad, Bd, Cd, Dd, 1);
zeros ¼ roots(num);
We get the following response:
zeros ¼ 0.9987 þ 0.0065i
0.9987 – 0.0065i
–0.9929
We shall select these three zeros as three out of five desired poles of the
closed-loop system. As the forth pole we will select the value of 0.9992 as the pole
318 Handbook of vehicle suspension control systems

location after approximately 10,000 samples (or 5 s). The last pole will be selected
with z ¼ 0.2, for this is quick enough to be neglected. Let us add the following code
into our m-file:

p1 ¼ .97 þ .13i; p2 ¼ .97 – .13i; p3 ¼ –.87

p1 ¼ zeros(1); p2 ¼ zeros(2); p3 ¼ zeros(3)
p4 ¼ .9992; p5 ¼ .5
K ¼ place(Ada, Bda * [1;0], [p1 p2 p3 p4 p5])

MATLAB gives us the following response:

place: ndigits ¼ 15
K ¼ 1.0eþ008 * (0.0082 0.3458 0.0054 9.8881 0.0096)

We shall use ‘dstep’ command to simulate the closed-loop response. Since

multiplying the state-space vector by K in our controller, gives one signal only, it is
necessary to add a series of zeros to K vector by multiplying it by [1 0]T. We shall
perform the simulation by the negative value of the step disturbance –0.1 m, in
order to get the positive value of the deflection for practical reasons. Let us add the
following code:

yout ¼ dstep(Ada – Bda * [1 0]’ * K – .1 * Bda, Cda, –

– .1 * Dda, 2, 10001);
t ¼ 0:.0005:5;
stairs(t, yout);

In Figure 11.8 we can see the closed-loop response to 0.1 m step disturbance.
The overshoot is less than 5 mm, and the settling time is not more than 5 s.

11.4.4 Active suspension control using neural network

As mentioned previously, PID controller is one of many different means of con-
trolling the active suspension. Neural networks can be efficiently trained not only
to identify the nonlinear dynamics of active suspension system, but also to control
it [27, 28]. Semi-active suspension systems can also be controlled using the neural
networks. For example, in Reference 29 the task of the neural network is to gen-
erate the control voltage based on the piston displacement and velocity to attain the
desired magneto-rheological damper force. A neural network controller based on a
Taylor series approximation of nonlinear suspension system dynamics can be
trained using a semi-stochastic parameter optimization method [30]. Also, neural
networks can be combined with other techniques, such as fuzzy logic [31, 32], to
control active suspension systems.
Artificial neural networks are fast, effective, and reliable algorithms for the
modeling of complex systems that cannot be modeled, or are hard to model using
conventional physical and mathematical models. They comprise group of artificial
neurons, and can be trained to perform a desired function by adjusting the weights of
the connections between these neurons in such a way that a particular network input
leads to a specific network output. Neurons in the neural network are usually arranged
System approach to vehicle suspension system control in CAE environment 319

Step response
0.6

0.4

0.2

0
Amplitude

–0.2

–0.4

–0.6

–0.8

–1

–1.2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (s)

Figure 11.8 Closed-loop response to 0.1 m step disturbance

in more than one layer. Outputs from neurons in one layer are inputs for the neurons in
the next layer. Input layer comprises nothing but scalar network inputs. The last layer
in the network is called the output layer. Outputs from this layer are also network
outputs. All layers between the input layer and output layer are called hidden layers.
Each neuron consists of an input function (most commonly summation) and an
S-shaped (log-sigmoid, hyperbolic tangent sigmoid, etc.) transfer function. Every
neuron has one additional input with a fixed value of one called bias. Network
outputs are calculated from its inputs as follows: every neuron input (including
bias) is multiplied by the corresponding weight value; in every neuron, inputs
weighted in such a way are then processed by the input function (summed if input
function is summation, multiplied if input function is multiplication, etc.) and the
S-shaped transfer function; these operations give the neuron output that will be the
input for the neurons in the next layer, or, if it is the neuron in output layer,
a network output. An example of neural network architecture is presented in
Figure 11.9 and abbreviated notation in Figure 11.10.
Input vector for the network presented in Figures 11.9 and 11.10 is a four-
column vector since the network is designed to control the active suspension by
changing the control force based on the four inputs:
8 9
>
> x1 > >
>
> >
< x_ 1 >
=
p¼ ð11:15Þ
>
> Y1 > >
>
> >
: > ;
Y2
320 Handbook of vehicle suspension control systems

Inputs Hidden layer Output layer

iw1,11,1 n11 a11

Σ lw2,11,1
x1 b11
n12 a12
Σ
x1
b12
n21 Fa
Σ
Y1 b21

Y2
n16 a16 lw2,12,6
iw1,16,4 Σ
b16

Figure 11.9 Neural network architecture

Input Hidden layer Output layer

f1 f2
p a1 a2 = y
IW1,1 6⫻1 LW2,1 1⫻1
4⫻1
6⫻4 n1 1⫻6 n2
+ 6⫻1 + 1⫻1
2
1 b1 1 b
4 6⫻1 6 1⫻1 1

Figure 11.10 Neural network architecture (abbreviated notation)

This network is now going to serve as an example to demonstrate how trained

network calculates its outputs according to the given inputs. The presented network
comprises one hidden layer having six neurons. That means that the input vector
will be multiplied by 6  4 input weight matrix IW1,1 and then summed with bias
vector b1. The result of this summation, vector n1, will then be processed by (in this
case, a log-sigmoid) transfer function. Output vector from the hidden layer a1 is the
input vector for the output layer, where it will be multiplied by the layer weight
matrix LW2,1 and then summed with bias vector b2 to form the vector n2, which
will be again processed by the transfer function that gives the output layer output
vector that is also the network output vector y:

a 2 ¼ y ¼ Fa ð11:16Þ
System approach to vehicle suspension system control in CAE environment 321

These operations are defined by the following two equations:


a1 ¼ f 1 IW1;1 p þ b1 ð11:17Þ



 
a2 ¼ y ¼ f 2 LW2;1 a1 þ b2 ¼ f 2 LW2;1 f 1 IW1;1 p þ b1 þ b2 ð11:18Þ

Figure 11.11 shows neural network creation procedure. The first step is to
identify the input parameters. The choice of network inputs and outputs depends on
the analysis for which the network is constructed, since neural network inputs and
outputs represent inputs and outputs, respectively, in the analysis that needs to be
done.
When network inputs and outputs are defined, training, validation, and test
data have to be prepared. These data comprise a sufficient number of input–output
relations, where for every combination of input parameters there are corresponding
output values. Prepared data then need to be divided into three sets: training,
validation, and test set.
Data from the training set will be used for the training of neural network.
Before the training can start, the data have to be pre-processed by normalizing. It
can be done by scaling the values of all parameters in the dataset (both inputs and
outputs) to a range from 0 to 1, where 1 corresponds to the maximum expected value
of a single parameter. Pre-processing also makes trained network response analysis
easier. Afterward, the outputs from the network are easily reverse-transformed back
to the original units.
When data are ready for the training, choice of number of hidden layers and
neurons, type of training, and its parameters has to be made. Choice must be made
according to the problem complexity – if the number of neurons is less than
required, approximation of the function will not be accurate enough and the outputs
will not relate to the inputs the way they should; on the other hand, if the number of
neurons in the hidden layer is greater than required, the phenomenon of over-fitting

Identification of input parameters

Preparation of training, validation, and test data

Choice of network architecture and training parameters

Corrections

Network training

Network validation and testing

Network use

Figure 11.11 Neural network model creation procedure

322 Handbook of vehicle suspension control systems

Simulated road disturbance

x1(t), x1(t), Y1(t), Y2(t) Fa(t)

CAE model

Overshoot Settling time

Real road disturbance Neural network training

x1(t), x1(t), Y1(t), Y2(t) Neural network controller Fa(t)

Figure 11.12 Neural network controller training scheme

can arise (this phenomenon results in large testing errors emergence, regardless of
the low training errors). The optimal number of neurons in hidden layers is to be
determined through experimentation. It is usual to start with a single hidden layer
having three to five neurons. Training algorithm and transfer functions also need to
be chosen by a trial-and-error procedure. A backpropagation neural network having
one hidden layer comprising sufficient number of neurons can accurately approx-
imate almost any complex function [33].
Neural network training using training data can start when the network archi-
tecture for the first iteration is set. In this example, neural network will be trained
using backpropagation algorithm, a widely used algorithm for the training of
feedforward neural networks. The name of the algorithm comes from the principle
of its functioning – input signals are sent forward and afterward the errors are
propagated backward. In the first epoch (training iteration), the weights of the
connections are initiated at random, and after that they are adjusted in every epoch
to minimize the output error. Weights are adjusted based on a comparison of the
obtained and desired outputs. Network training stops when one of the following
conditions occurs:
● predefined training performance goal is reached;
● if using validation, predefined validation performance goal is reached;
● predefined maximum number of epochs is reached (too high number of train-
ing epochs also can lead to over-fitting of the training data); or
● predefined maximum amount of time is exceeded.
When training is completed, validation and testing of artificial neural network
using the validation and test data sets need to be conducted. The purpose of these
System approach to vehicle suspension system control in CAE environment 323

two steps is to evaluate the network ability to give the correct outputs as a response
to unknown inputs. Training, validation, and testing results need to be carefully
observed. If errors are acceptable, the trained network can be used for the analysis
concerned, and if it is not the case, network architecture and training parameters
need to be changed, and training, validation, and testing need to be conducted
again. This trial-and-error process needs to be repeated until acceptable errors are
attained. Neural network controller training scheme is shown in Figure 11.12.
The neural network used as an active suspension controller needs to be trained
using the experimental data or data obtained using the numerical model such as
one presented in the text before. For the set of combinations of input parameters
(x1, x_ 1 , Y1, Y2) there are values of control force Fa that gives the acceptable values
of overshoot and settling time. These data are used to train the network to respond
to input parameters in such way to provide desired compromise between the ride
comfort and the road holding.
Presented neural network also makes it possible to obtain the needed range of
active force intensity for the predicted ranges of input parameters. By knowing this
range, the constructor can optimize the actuator in terms of size, weight, energy
consumption, and price while retaining the acceptable performance of the suspen-
sion system.

11.5 Conclusions
Over the past few years, active suspension system has seen an increasing applica-
tion. The simulation of this system in the initial development phase has many
advantages:

● we gain time – by not having to wait for exploitation results;

● it is cost-effective – we do not have to design an expensive model in the lab
and perform complex laboratory examinations;
● we can simulate suspension systems of almost all categories and kinds of
motor vehicles, which could also be hardly possible in laboratory conditions.

Before the practical implementation of the active suspension system, it is

recommended and required to simulate the obtained analytical solutions on the
computer in order to examine the range of change of regulative action and
the response of the regulated variable to a typical input signal, and thus confirm
the possibility of the system physical realization, bearing in mind the actuator
limitations, as well as check whether other possible limitations within the object
of control as a whole have been reached. The above statements by no means
suggest that laboratory testing is no longer necessary, but that it can be carried out
in a shorter period of time and with reduced cost [34]. The key challenge asso-
ciated with active suspension and their actuators is size, weight, and energy
consumption required to achieve acceptable performance. For this reason, the
physical properties of the actuator are to be included in the optimization problem.
Quite often in practical experience, actuator faults are the largest source of
324 Handbook of vehicle suspension control systems

control system degradation. Suspension system modeling was performed on

quarter-car model, using the MATLAB interactive environment and by the state-
space equation. The step input is the unit step function, that is, a certain value of
road disturbance. It has been concluded that the settling time and overshoot of the
vehicle, after encountering any kind of obstacle, are too long, and that a controller
must be introduced into the suspension system. We designed a digital controller
by the pole-placement method, which makes only one of the possible solutions.
Also, the possibility to control the active suspension system using neural network
is presented.
The presented dynamic model is only a very rough representation of the true
dynamic system, which is applicable only in the early concept design phases of the
research and development process. It is expected that, in the near future, this design
and the above-mentioned modifications and the necessary improvements should
have application in designing such systems in the motor vehicle industry.

References
[1] V. Popović, B. Vasić, B. Rakićević, and G. Vorotović, ‘‘Optimization of
maintenance concept choice using risk-decision factor – a case study,’’ Inter-
national Journal of Systems Science, vol. 43, no. 10, pp. 1913–1926, 2012.
[2] W. Bolton, Mechatronics: A Multidisciplinary Approach, Fifth Edition.
Pearson/Prentice Hall, Upper Saddle River, NJ, 2011.
[3] H. Du and N. Zhang, ‘‘H? control of active vehicle suspensions with
actuator time delay,’’ Journal of Sound and Vibration, vol. 301, no. 1–2,
pp. 236–252, 2007.
[4] D. Fischer and R. Isermann, ‘‘Mechatronic semi-active and active vehicle
suspensions,’’ Control Engineering Practice, vol. 12, no. 11, pp. 1353–1367,
2004.
[5] G. P. A. Koch, ‘‘Adaptive control of mechatronic vehicle suspension systems,’’
doctoral dissertation, Technische Universität München, Germany, 2011.
[6] V. Popović, B. Vasić, M. Petrović, and S. Mitić, ‘‘System approach to
vehicle suspension system control in CAE environment,’’ Strojniški vestnik –
Journal of Mechanical Engineering, vol. 57, no. 2, pp. 100–109, 2011.
[7] V. Popović, D. Janković, and B. Vasić, ‘‘Design and simulation of active
suspension system by using MATLAB,’’ In Proceedings of of FISITA World
Automotive Congress, paper F2000G269, Seoul, South Korea, 2000.
[8] V. Popović, B. Vasić, and D. Janković, ‘‘Development of semi-active sus-
pension system in CAE environment,’’ In Proceedings of FISITA World
Automotive Congress, paper F02V062, Helsinki, Finland, 2002.
[9] M. M. Fateh and S. S. Alavi, ‘‘Impedance control of an active suspension
system,’’ Mechatronics, vol. 19, no. 1, pp. 134–140, 2009.
[10] M. Jonasson and F. Roos, ‘‘Design and evaluation of an active electromechanical
wheel suspension system,’’ Mechatronics, vol. 18, no. 4, pp. 218–230, 2008.
System approach to vehicle suspension system control in CAE environment 325

[11] G. Priyandoko, M. Mailah, and H. Jamaluddin, ‘‘Vehicle active suspension

system using skyhook adaptive neuro active force control,’’ Mechanical
Systems and Signal Processing, vol. 23, no. 3, pp. 855–868, 2009.
[12] S. J. Huang and H. Y. Chen, ‘‘Adaptive sliding controller with self-tuning
fuzzy compensation for vehicle suspension control,’’ Mechatronics, vol. 16,
no. 10, pp. 607–622, 2006.
[13] A. B. Sharkawy, ‘‘Fuzzy and adaptive fuzzy control for the automobiles’
active suspension system,’’ Vehicle System Dynamics, vol. 43, no. 11,
pp. 795–806, 2005.
[14] Y. Taskin, Y. Hacioglu, and N. Yagiz, ‘‘The use of fuzzy-logic control to
improve the ride comfort of vehicles,’’ Strojniški vestnik – Journal of
Mechanical Engineering, vol. 53, no. 4, pp. 233–240, 2007.
[15] R. Guclu, ‘‘The fuzzy-logic control of active suspensions without suspension-
gap degeneration,’’ Strojniški vestnik – Journal of Mechanical Engineering,
vol. 50, no. 10, pp. 462–468, 2004.
[16] J. Lin, R. J. Lian, C. N. Huang, and W. T. Sie, ‘‘Enhanced fuzzy sliding
mode controller for active suspension systems,’’ Mechatronics, vol. 19,
no. 7, pp. 1178–1190, 2009.
[17] İ. Eski and Ş. Yıldırım, ‘‘Vibration control of vehicle active suspension
system using a new robust neural network control system,’’ Simulation
Modelling Practice and Theory, vol. 17, no. 5, pp. 778–793, 2009.
[18] J. S. Lin and C. J. Huang, ‘‘Nonlinear backstepping active suspension design
applied to a half-car model,’’ Vehicle System Dynamics, vol. 42, no. 6,
pp. 373–393, 2004.
[19] N. Yagiz and Y. Hacioglu, ‘‘Backstepping control of a vehicle with active sus-
pensions,’’ Control Engineering Practice, vol. 16, no. 12, pp. 1457–1467, 2008.
[20] M. R. Stojić, Digital Control Systems (in Serbian), Fourth edition. Faculty of
Electrical Engineering, Belgrade, Serbia, 1998.
[21] C. B. Moler, Numerical Computing with MATLAB. SIAM, Philadelphia, PA,
2004.
[22] A. C. J. Luo and A. Rajendran, ‘‘Periodic motions and stability in a semi-
active suspension system with MR damping,’’ Journal of Vibration and
Control, vol. 13, no. 5, pp. 687–709, 2007.
[23] D. Lee, J. Allan, H. A. Thompson, and S. Bennett, ‘‘PID control for a dis-
tributed system with a smart actuator,’’ Control Engineering Practice, vol. 9,
no. 11, pp. 1235–1244, 2001.
[24] P. Cominos and N. Munro, ‘‘PID controllers: recent tuning methods and
design to specification,’’ IEE Proceedings – Control Theory and Applica-
tions, vol. 149, no. 1, pp. 46–53, 2002.
[25] K. J. A° ström and T. Hägglund, PID Controllers: Theory, Design and Tuning,
Second edition. Instrument Society of America, Research Triangle Park, NC,
1995.
[26] J. R. Rogers and K. Craig, ‘‘On-hardware optimization of stepper-motor
system dynamics,’’ Mechatronics, vol. 15, no. 3, pp. 291–316, 2005.
326 Handbook of vehicle suspension control systems

[27] M. Nagai, A. Moran, Y. Tamura, and S. Koizumi, ‘‘Identification and control

of nonlinear active pneumatic suspension for railway vehicles, using neural
networks,’’ Control Engineering Practice, vol. 5, no. 8, pp. 1137–1144, 1997.
[28] A. Moran and M. Nagai, ‘‘Optimal active control of nonlinear vehicle sus-
pensions using neural networks,’’ JSME International Journal, vol. 37-C,
no. 4, pp. 707–718, 1994.
[29] M. Zapateiro, N. Luo, H. R. Karimi, and J. Vehı́, ‘‘Vibration control of a
class of semiactive suspension system using neural network and back-
stepping techniques,’’ Mechanical Systems and Signal Processing, vol. 23,
no. 6, pp. 1946–1953, 2009.
[30] K. Spentzas and S. A. Kanarachos, ‘‘Design of a non-linear hybrid car sus-
pension system using neural networks,’’ Mathematics and Computers in
Simulation, vol. 60, no. 3–5, pp. 369–378, 2002.
[31] S. J. Huang and R. J. Lian, ‘‘A combination of fuzzy logic and neural net-
work algorithms for active vibration control,’’ Proceedings of the Institution
of Mechanical Engineers, Part I: Journal of Systems and Control Engi-
neering, vol. 210, no. 3, pp. 153–167, 1996.
[32] N. Al-Holou, T. Lahdhiri, D. S. Joo, J. Weaver, and F. Al-Abbas, ‘‘Sliding
mode neural network inference fuzzy logic control for active suspension sys-
tems,’’ IEEE Transactions on Fuzzy Systems, vol. 10, no. 2, pp. 234–246, 2002.
[33] A. T. C. Goh, ‘‘Back-propagation neural networks for modeling complex sys-
tems,’’ Artificial Intelligence in Engineering, vol. 9, no. 3, pp. 143–151, 1995.
[34] D. Stamenković, V. Popović, and D. Aleksendrić, ‘‘Fully renewing combination
free replacement and pro-rata warranty cost assessment using Monte Carlo
simulation,’’ In Proceedings of The 18th ISSAT International Conference on
Reliability and Quality in Design, pp. 315–319, Boston, MA, 2012.
 Chapter 12
Vibration control of vehicle engine-body systems
with time delay: an LMI approach
Hamid Reza Karimi

Abstract
The objective of this chapter is to study the problem of vibration control analysis
and synthesis in a vehicle engine-body vibration structure. It is assumed that the
actuator is subject to a time-varying delay for control of bounce and pitch vibra-
tions. Based on a Lyapunov–Krasovskii functional and using some free weighting
matrices, delay-dependent sufficient conditions for designing desired state- and
output-feedback controllers are given in terms of linear matrix inequalities (LMIs).
The state- and output-feedback controllers, which guarantee asymptotic stability
with a prescribed g-level L2 -gain (or H1 performance), are then developed directly
instead of coupling the second-order model to a first-order system. The controller
gains are determined by convex optimization over LMIs. Simulation results are
included to demonstrate the validity and applicability of the technique.

12.1 Introduction
Active control of sound and vibration has emerged as an important area of scientific
and technological development in recent years. Developments in active control have
allowed successful application of the concept in numerous industrial areas [1].
Recently, the noise and vibration of cars have become increasingly important. The
predominant sources of interior noise in cars are engine and wheel vibrations,
which propagate as structure-borne sound through the car body and finally radiate as
airborne sound into the cabin [1–5]. The engine mount is an efficient passive means to
isolate the car chassis structure from the engine vibration. However, the passive means
for isolation is efficient only in the high-frequency range. However, the vibration
disturbance generated by the engine occurs mainly in the low-frequency range [8].
A major comfort aspect is the transmission of engine-induced vibrations
through powertrain mounts into the chassis (see Figure 12.1). Engine and power-
train mounts are usually designed according to criteria that incorporate a trade-off
Department of Engineering, Faculty of Engineering and Science, University of Agder, 4898 Grimstad,
Norway
328 Handbook of vehicle suspension control systems

Z: Engine block
X
Chassis

Y Piston
O
Pitch

Crank

Engine
mount

Figure 12.1 Chassis excited by the engine

Disturbance source
ENGINE

Error signal

CHASSIS SUBFRAME

Actuator

u(t) Controller
Amplifier
Algorithm

Figure 12.2 Feedback control arrangement setup [8]

between the isolation of the engine from the chassis and the restriction of engine
movements. The engine mount is an efficient passive means to isolate the car
chassis structure from the engine vibration. The passive means for isolation is
efficient only in the high-frequency range. However, the vibration disturbance
generated by the engine occurs mainly in the low-frequency range (see the Refer-
ences [2], [6–9] and the references therein). These vibrations are a result of the fuel
explosion in the cylinder and the rotation of the different parts of the engine
(see Figure 12.2). The commercial use of engine and wheel mounts has been
 Vibration control of vehicle engine-body systems with time delay 329

impeded so far by technical problems. Compact and robust combinations of con-

ventional rubber mounts with electrodynamically driven hydraulics have been
constructed as active hydromounts for a wide frequency range [10], but the stroke
and power required for cars at low frequencies cannot yet be fulfilled by active
hydromounts of reasonable size [1].
In the literature, two main methods for active vibration control exist. The first
method, based on an assumption that information about the disturbance source are
available, is called feedforward control and is usually realized with the famous Fx-
LMS algorithm [6, 11]. The second method for active vibration control is the
feedback control. The main idea of feedback control is to use a sensor only to detect
the output response of the system. Therefore, the disturbance source is assumed to
be unknown. Several strategies of feedback control for vibration attenuation of
unknown disturbance exist ranging from classical methods like velocity feedback
control or acceleration feedback control to a more advanced methods like optimal
control, LQG control and neural network control [12]. A variety of control tech-
niques such as PID or Lead-Lag compensation, LQG/H2 , H1 , m-synthesis and
feedforward control have been used in active vibration systems (see for instance
References [4], [9], [13–22]). The main characteristic of feedforward control is that
information about the disturbance source is available and is usually realized with
the Fx-LMS algorithms. However, the disturbance source is assumed to be
unknown in feedback control, and then different strategies of feedback control for
vibration attenuation of unknown disturbance exist ranging from classical methods
to more advanced methods. Recently, the performance results obtained by feedback
and feedforward controllers using Fx-LMS algorithms for vehicle engine-body
vibration system were compared in References [8] and [9]. In Reference [8], the
arrangement shown in Figure 12.2 is used to control implementation.
On the other hand, in the field of dynamic systems and control, delays appear
in the state, in the control input, or in the measurements (see for instance Refer-
ences [23] and [24] and the references therein). The presence of a delay in a system
may be the result of some essential simplification of the corresponding process
model. Generally, time delay exists inevitably in control systems, which mainly
results from the following: (1) the time taken in the online data acquisition from
sensors at different locations of the system; (2) the time taken in the filtering and
processing of the sensory data for the required control force to the actuator; (3) the
time taken by the actuator to produce the required control force. Therefore, how to
analyse and synthesize dynamic systems with delayed arguments is a problem of
recurring interest, as the delay may induce complex behaviours (oscillation,
instability, bad performances) for the systems concerned (see References [25–38]
and the references therein).
It is also worth citing that some appreciable pieces of work have been per-
formed to design a delayed control for vibration control with time delay in structure
(see the References [14], [15] and [39]). However, the system performance and
stability are not investigated for a vibration structure with a time-varying actuator
delay in these works. Up to now, to the best of the author’s knowledge, no results
about a convex optimization method for delay-dependent H1 control problem of
330 Handbook of vehicle suspension control systems

vehicle engine-body vibration structures with a time-varying actuator delay are

available in the literature, which remains to be important and challenging. This
motivates the present study.
In this chapter, the author contributes to the further development of delay-
dependent state- and output-feedback aspects of H1 control for vehicle engine-
body vibration systems with a time-varying actuator delay. The feedback loop is
subjected to a time-varying delay within the actuator and the system. Then the
author presents the application of the controllers to the vibration control of the
vehicle chassis subframe. The main merit of the proposed method lies in the fact
that it provides a convex problem via introduction of additional decision variables
such that the control gains can be found from the LMI formulations without
reformulating the system equations into a standard form of a first-order neutral
system. By using a Lyapunov–Krasovskii method and some free weighting matri-
ces, new sufficient conditions are established in terms of delay-dependent LMIs
for the existence of desired state- and output-feedback controllers such that the
resulting closed-loop system is asymptotically stable and satisfies a prescribed
g-level L2 -gain (or H1 performance). A significant advantage of our result is that
the desired controllers are designed directly instead of coupling the model to a first-
order system and then applying the corresponding control designs existing in the
literature in a higher dimensional space. Therefore, our results can be implemented
in a numerically stable and efficient way for the system under consideration.
Furthermore, as pointed out in Reference [40], retaining the model in a matrix of
second-order form has many advantages such as preserving physical insight of the
original problem, preserving system matrix sparsity and structure, preserving
uncertainty structure and entailing easier implementation (feedback control can be
used directly). Finally, simulation results are given to illustrate the validity and
applicability of the technique.
Notations. The superscript ‘T ’ stands for matrix transposition; <n denotes the
n-dimensional Euclidean space; <nm is the set of all real m by n matrices. k:k
refers to the Euclidean vector norm or the induced matrix 2-norm. colf  g and
diagf  g represent, respectively, a column vector and a block diagonal matrix and
the operator symðAÞ represents A þ AT . lmin ðAÞ and lmax ðAÞ denote, respectively,
the smallest and largest eigenvalue of the square matrix A. The notation P > 0
means that P is real symmetric and positive definite; the symbol * denotes the
elements below the main diagonal of a symmetric block matrix. In addition,
L2 ½0, 1Þ is adopted for the space of all functions f : < ! < which are Lebesque
integrable in the square over ½0, 1Þ, with the standard norm k:k2 .

12.2 The vehicle engine-body system

In this section a dynamic formulation of the characteristics of the vehicle engine-

body vibration system is provided for vibration control design. A schematic dia-
gram of the vehicle engine-body vibration structure is shown in Figure 12.3, where
the engine with mass Me and inertia moment Ie is mounted in the body by the
 Vibration control of vehicle engine-body systems with time delay 331

de
Me, Ie
x1
y(t) f(t) f(t – h(t))
Controller Delay ke, ce
Mb, Ib

x2
kb, cb

2L

Figure 12.3 Sketch of engine-body vibration system (first configuration)

engine mounts ke , ce and the vehicle body with mass Mb and inertia moment Ib is
supported by front and rear tyres, each of which is modelled as a system consisting
of a spring kb and a damping device cb .
The front mount is the active mount, the output force of which can be con-
trolled by an electric signal. The active mount consists of a main chamber where an
oscillating mass (inertia mass) moves up and down. The inertia mass is driven by
an electromagnetic force generated by a magnetic coil which is controlled by the
input current (see Figure 12.4).
The following differential equations describe the motion of the inertia mass
actuator:
M€x þ cx_ þ kx ¼ k1 i
di
L þ Ri þ k2 x_ ¼ u
dt

Coil
Inertia mass
M

S N S

k c

Figure 12.4 Sketch of an active engine mount

332 Handbook of vehicle suspension control systems

In Laplace coordinate, the system above can be transformed as

ðMs2 þ cs þ kÞX ¼ k1 I
ðLs þ RÞI þ k2 sX ¼ U
or

ðMs2 þ cs þ kÞX ¼ k1 Gc ðU  k2 sX Þ
where
1
Gc ðsÞ ¼
Ls þ R
The inertia mass create a force f ¼ mp€x . Then the transfer function becomes

f mp G1 Gc s2
¼
U ðMs þ ðc þ k1 Gc k2 Þs þ kÞ
2

In general case we neglect the dynamic of the coil, and the transfer function above
can be represented as

f s2
¼ g 2
U ðs þ 2xwa s þ w2a Þ
where wa is the resonance associated with the spring-mass system; the Bode dia-
gram of the function above is shown in Figure 12.5.

ωc
g Hz

ωa

–180

Hz

Figure 12.5 Bode diagram

 Vibration control of vehicle engine-body systems with time delay 333

The critical frequency wc is the frequency above which the actuator behaves
like an ideal force actuator.
The ideal engine mount system should isolate engine vibration caused by
engine disturbance force in engine speed range and prevent engine bounce from
shock excitation. This implies that the dynamic stiffness and damping of the engine
mount should be frequency- and amplitude-dependent. The development of engine
mounting systems has mostly concentrated on improvement of frequency- and
amplitude-dependent properties. The conventional elastomeric mounts do not
meet all the requirements and can only offer a trade-off between static defection and
vibration isolation. Passive hydraulic mounts can provide a better performance than
elastomeric mounts especially in the low-frequency range. Semi-active techniques
are usually used to further improve performance of hydraulic mounts by making
them more tuneable. The active engine mounting system can be very stiff at low
frequency and be tuned to be very soft at the higher frequency range to isolate the
vibration. The active engine mounts have been considered as the next generation of
engine mounts. The optimization of engine mounting systems is quite desirable [5].
In our study, only the bounce and pitch vibrations in the engine and body are
considered. It is assumed that the actuator and sensor used to this control frame-
work are selected to be collocated, since this arrangement is ideal to ensure the
stability of the closed-loop system for a slightly damped structure.
The derivation of the dynamic equations for a four degree-of-freedom vibra-
tion suspension model shown in Figure 12.3 accordingly follows [9]:
8
> Me€x 1 ðtÞ þ 2ce x_ 1 ðtÞ þ 2ke x1 ðtÞ  2ce x_ 2 ðtÞ  2ke x2 ðtÞ
>
>
>
>  2ðL  lÞ ce x_ 4 ðtÞ  2ðL  lÞ ke x4 ðtÞ ¼ f ðt  hðtÞÞ þ de ðtÞ
>
>
>
>
>
>
>
>
> Mb €x 2 ðtÞ þ 2ðce þ cb Þx_ 2 ðtÞ þ 2ðke þ kb Þ x2 ðtÞ  2ce x_ 1 ðtÞ
>
>
>
>  2k x ðtÞ þ 2ðL  lÞ c x_ ðtÞ þ 2ðL  lÞk x ðtÞ ¼ f ðt  hðtÞÞ
>
>
> e 1 e 4 e 4
>
>
>
<
Ie €x 3 ðtÞ þ 2l2 ce x_ 3 þ 2l2 ke x3 ðtÞ  2l2 ce x_ 4 ðtÞ
>
>
>  2l2 ke x4 ðtÞ ¼ l f ðt  hðtÞÞ
>
>
>
>
>
>
>
>
>
>
> Ib €x 4 ðtÞ þ ððL2 þ ðL  2lÞ2 Þce þ 2L2 cb Þx_ 4 ðtÞ
>
>
>
> þ ððL2 þ ðL  2lÞ2 Þ ke þ 2L2 kb Þx4 ðtÞ  2l2 ce x_ 3 ðtÞ
>
>
>
>
>
>  2l2 ke x3 ðtÞ  2lce x_ 1 ðtÞ  2lke x1 ðtÞ þ 2ðL  lÞ ce x_ 2 ðtÞ
:
þ 2ðL  lÞke x2 ðtÞ ¼ L f ðt  hðtÞÞ
ð12:1adÞ

where x1 ðtÞ, x2 ðtÞ, x3 ðtÞ, x4 ðtÞ are the bounces and pitches of the engine and body
respectively. Input force f ðtÞ is used as the active force to compensate the vibration
transmitted to vehicle body (or to the chassis); engine disturbance de ðtÞ is the
excitation signal which is generated by the motion up/down of the different parts
inside the engine.
334 Handbook of vehicle suspension control systems

Assumptions 12.1. There exists an unknown and time-varying time delay hðtÞ in
_  hD , where hM and hD are the
the actuator, which satisfies 0 < hðtÞ  hM and hðtÞ
_
maximums of hðtÞ and hðtÞ, respectively.

_  hD on time-varying delay hðtÞ means that our

Remark 12.1. The condition hðtÞ
method can deal with actuators with any fast time-varying delay.

Remark 12.2 (second configuration). In the configuration shown in Figure 12.6,

the actuator force is located at the chassis of the car. The differential equations of
the system are similar to those of the first configuration in Figure 12.3, except some
changes as shown in the following system:
Me€x 1 þ 2ce x_ 1 þ 2ke x1  2ce x_ 2  2ke x2  2ðL  lÞce x_ 4  2ðL  lÞke x4 ¼ de

Mb€x 2 þ 2ðce þ cb Þx_ 2 þ 2ðke þ kb Þx2  2ce x_ 1  2ke x1 þ 2ðL  lÞce x_ 4

þ 2ðL  lÞke x4 ¼ f

Ie€x 3 þ 2l2 ce x_ 3 þ 2l2 ke x3  2l2 ce x_ 4  2l2 ke x4 ¼ 0

Ib€x 4 þ ððL2 þ ðL  2lÞ2 Þce þ 2L2 cb Þx_ 4 þ ððL2 þ ðL  2lÞ2 Þke

þ 2L2 kb Þx4  2l2 ce x_ 3  2l2 ke x3  2lce x_ 1  2lke x1
þ 2ðL  lÞce x_ 2 þ 2ðL  lÞke x2 ¼ Lf

2l Actuator force

Me, Ie de

x1

ke, ce f

Mb, Ib

x2
kb, cb

2L

Figure 12.6 Sketch of engine-body vibration system (second configuration)

 Vibration control of vehicle engine-body systems with time delay 335

12.3 Problem formulation

The state-space representation of the vehicle engine-body vibration system (12.1) is
8
> M€x ðtÞ þ AxðtÞ
_ þ BxðtÞ ¼ Bf f ðt  hðtÞÞ þ Bd de ðtÞ
>
>
>
>
< xðtÞ ¼ fðtÞ, t 2 ½hM , 0
>
xðtÞ
_ ¼ fðtÞ,_ t 2 ½hM , 0 ð12:2Þ
>
>
>
> zðtÞ ¼ C1 xðtÞ þ C2 xðtÞ
_ þ D1 f ðt  hðtÞÞ
>
>
:
yðtÞ ¼ C3 xðtÞ þ C4 xðtÞ
_
where xðtÞ ¼ ½x1 ðtÞ, x2 ðtÞ, x3 ðtÞ, x4 ðtÞ 2 <4 is the state vector; uðtÞ 2 < is the
control input; de ðtÞ 2 L2 ½0, 1Þ is the external excitation (disturbance) and
zðtÞ 2 <s is the controlled output and the time-varying vector valued initial func-
_
tions fðtÞ and fðtÞ are continuously differentiable functionals. The state-space
matrices are defined as

M ¼ diag fMe , Mb , Ie , Ib g
2 3
2ce 2ce 0 2ðL  lÞce
6 2ce 2ðce þ cb Þ 0 2ðL  lÞce 7
A¼64 0
7
5
0 2l2 ce 2l2 ce
2lce 2ðL  lÞce 2l2 ce ðL2 þ ðL  2lÞ2 Þce þ 2L2 cb
2 3
2ke 2ke 0 2ðL  lÞke
6 2ke 2ðke þ kb Þ 0 2ðL  lÞke 7
B¼6 7
4 0 0 2l2 ke 2l2 ke 5
2lke 2ðL  lÞke 2l2 ke ðL2 þ ðL  2lÞ2 Þke þ 2L2 kb
2 3
1
6 1 7
Bf ¼ 6 7
4 l 5
L
2 3
1
607
Bd ¼ 6 7
405
0

Remark 12.3. The dynamical system (12.2) arises naturally in a wide range
of applications, including control of large flexible space structures, earthquake
engineering, control of mechanical multi-body systems, stabilization of damped
gyroscopic systems, robotics control, vibration control in structural dynamics, lin-
ear stability of flows in fluid mechanics and electrical circuit simulation (see, e.g.,
References [41–43] and the many references therein).
336 Handbook of vehicle suspension control systems

The objective of this chapter is to solve the following two problems:

Problem 12.1. For system (12.2) subject to the time-varying actuator delay, if all
the states and their derivatives are measurable, design a delayed state-feedback
controller f ðtÞ ¼ K xðt  hðtÞÞ,
Problem 12.2. For system (12.2) subject to the time-varying actuator delay, if all
the states and their derivatives are not measurable, design a delayed output-
feedback controller f ðtÞ ¼ K yðt  hðtÞÞ,
where xðtÞ :¼ colðxðtÞ, xðtÞÞ
_ and K is the controller gain matrix such that
1. the resulting closed-loop system is asymptotically stable with time-varying
delays in Assumption 12.1;
2. under zero initial conditions and for all non-zero de ðtÞ 2 L2 ½0; 1Þ, the induced
L2 -norm of the operator from de ðtÞ to the controlled outputs zðtÞ is less than a
ð
1

positive scalar g (i.e. J1 ¼ ½zT ðtÞ zðtÞ  g2 deT ðtÞ de ðtÞ dt < 0).
0

In this case, the system is said to be robustly asymptotically stable with an H1

performance measure.
Remark 12.4. It is noted that the vehicle engine-body vibration system is con-
trolled by a proportional and derivative (PD) H1 control.

12.4 Main results

In this section, sufficient conditions for the solvability of Problems 12.1 and 12.2 in
section 12.3 are proposed using the Lyapunov method and an LMI approach.
Before proceeding further, we give two technical lemmas, which are useful in the
proof of our main results.
Lemma 12.1 ([44]). For any arbitrary column vectors aðsÞ, bðsÞ 2 <p , and any matrix
W 2 <pp and positive-definite matrix H 2 <pp the following inequality holds:
ðt ðt     
T aðsÞ T H HW aðsÞ
2 bðsÞ aðsÞds  ds
bðsÞ  ðHW þ IÞT H 1 ðHW þ IÞ bðsÞ
trðtÞ trðtÞ

Lemma 12.2 ([28]). For given positive-definite matrices G and Q, the nonlinear matrix
inequality K T Q1 K < G1 with W :¼ K X is satisfied if the following LMI holds:
 
2 X þ G W T
<0
 Q
Lemma 12.3 ([45]). For a given M 2 <pn with rankðMÞ ¼ p < n, assume that
b 2 <pp such that
Z 2 <nn is a symmetric matrix, then there exists a matrix Z
MZ ¼Z b M if and only if
 
Z1 0
Z¼V VT
0 Z2
b ¼UM
Z b 1 U T
b Z1 M
 Vibration control of vehicle engine-body systems with time delay 337

where Z1 2 <pp , Z2 2 <ðnpÞðnpÞ and the singular value decomposition of the

matrix M is represented as M ¼ U½ Mb 0 V T with the unitary matrices U 2 <pp ,
V 2< nn
and a diagonal matrix M b 2 <pp with positive diagonal elements in
decreasing order.

12.4.1 State-feedback control design

Substitute the state-feedback controller
f ðtÞ ¼ K xðt  hðtÞÞ
into (12.2) and consider Problem 12.1.
Theorem 12.1. For given scalars g, hM > 0, hD , there exists a delay-dependent
state-feedback H1 control in the form of f ðtÞ ¼ K xðt  hðtÞÞ such that the
resulting closed-loop system is robustly asymptotically stable and satisfies the
constraint J1 < 0, if there exist a scalar a, matrices N b 1, N
b 2, X
b 1 , X2 , X 3 , L
~ and
b
positive-definite matrices X1 , Q 1 , G, H , satisfying the following LMIs:
2       3
0 b1
hM N ~T
X1 C
S
6 11 S 12  S16 S16 hM ða þ 1ÞH 7
6 Bd 0 0 7
6  S T 0 0 0 7
6 0 b2
hM N ðD1 Xb1 Þ 7
6 
22
2
0 0 0 7
6  g I 0 0 0 0 0 7
6    hM X1 0 7<0
6 0 0 0 7
6     I 7
6 hM X1 0 0 7
6      7
6  hM G 0 7
4 5
       hM H
    
ð12:3aÞ
2  T 3
0
6 2X þ G
4
1
~ 7
L 5<0 ð12:3bÞ
 H
where
" #! " n o #
bI T ~I X1 þ ~I T X3 ~I T X2 b 1 þ sym N
Q b1 0
S11 :¼ sym þ
A X1 þ L ~ þ M X3 MX2  0
" #
Nb T2  N b1
S12 :¼
ð1  aÞ L ~ þ Bf X b1
" T
#
X1~I bI þ X3T ~I
S16 :¼ hM
X2T ~I
n o
S22 :¼  ð1  hD ÞQ b 1  sym N b2

bI :¼ ½I, 0, ~I :¼ ½0, I, A :¼ ½B, A, ~ :¼ ½C1 , C2 

C
338 Handbook of vehicle suspension control systems

The desired control gain is given by

b 1 X11 from LMIs ð12:3Þ
K¼X ð12:4Þ
Proof: First, we represent (12.2) in an equivalent descriptor model form as
(
€x ðtÞ ¼ hðtÞ,
ð12:5Þ
0 ¼ M hðtÞ þ A xðtÞ  Bf K xðt  hðtÞÞ  Bd de ðtÞ

By utilizing the Leibniz–Newton formula we have the following zero equation:

ðt
L xðtÞ  L xðt  hðtÞÞ  L _
xðsÞ ds ¼ 0 ð12:6Þ
thðtÞ

We can represent the system (12.5) as

8
>
> €x ðtÞ ¼ hðtÞ,
>
>
>
< 0 ¼ M hðtÞ þ ðA þ LÞ xðtÞ  ðL þ Bf K Þ xðt  hðtÞÞ
>
ðt ð12:7Þ
>
>
>
> L _
xðsÞ ds  Bd de ðtÞ
>
>
:
thðtÞ

Define the Lyapunov–Krasovskii functional

V ðtÞ ¼ V1 ðtÞ þ V2 ðtÞ þ V3 ðtÞ þ V4 ðtÞ ð12:8Þ

where
 
T T T xðtÞ
V1 ðtÞ ¼ xðtÞ P1 xðtÞ :¼ ½ xðtÞ hðtÞ  T P
hðtÞ
ðt
V2 ðtÞ ¼ xðsÞT Q1 xðsÞds
thðtÞ
ð0 ðt
V3 ðtÞ ¼ _ T P1 xðsÞds
xðsÞ _ dq
hM tþq
ðt  T  
_ T 0 0 _
V4 ðtÞ ¼ ðs  t þ hM ÞxðsÞ H xðsÞds
L L
thM

with
T ¼ diag f I, 0g
 Vibration control of vehicle engine-body systems with time delay 339

and
 
P1 0
P¼
P3 P2
where P1 ¼ PT1 > 0.
Noting that V ðfðtÞ, tÞ  lmin ðP1 Þ ðkfð0Þk2 þ kfð0Þk
_ 2
Þ. According to Refer-
ence [46], using the Cauchy–Schwarz inequality, we have
ð0
2 _ 2
2
kfðqÞk  2 kfð0Þk  2q kfðuÞk du
q

ð0 ð0
_ 2 _ 2 € 2
kfðqÞk dq  2h kfð0Þk þ 2h2 kfðuÞk du
h h

and
ð0 ð0
2 _2 3 2 3 € 2
kfðqÞk dq  2h kfð0Þk þ 4h kfð0Þk þ 4h kfðuÞk du
h h

From this, after some manipulations, we obtain:

ð0
_ 2 2 € 2
V ðfðtÞ, tÞ  V ðfð0Þ, 0Þ  r ½ kfð0Þk þ kfð0Þk þ kfðqÞk dq
hM

ð12:9Þ
where r :¼ max ðr1 , r2 , r3 Þ with

r1 :¼ lmax ðP1 Þ þ 2hM lmax ðQ1 Þ

  T  !
2 0 0
r2 :¼ 2hM lmax ðP1 Þ þ H þ 2hM ð1 þ 2h2M Þlmax ðQ1 Þ
L L

and
" #T " #!
0 0
r3 :¼ 2h2M ð1 þ 2hM Þlmax ðQ1 Þ þ hM ð1 þ 2h2M Þ lmax ðP1 Þ þ H
L L

Therefore, by Theorem 1.6 of Reference [47], we conclude that the system

(12.2) with de ðtÞ 0 is asymptotically stabilizable by the state-feedback H1
controller.
340 Handbook of vehicle suspension control systems

Differentiating V1 ðtÞ along the system trajectory becomes:

" #
_
xðtÞ
T _ ¼ 2 ½ xðtÞT hðtÞT  PT
V_ 1 ðtÞ ¼ 2 xðtÞ P1 xðtÞ
0
2 _ 3
xðtÞ
6 0 1 7
6 M hðtÞ þ ðA þ LÞ xðtÞ 7
6B C 7
6 B ðL þ Bf KÞ xðt  hðtÞÞ C 7
¼ 2 ½ xðtÞT hðtÞT  PT 6
6BB C 7
6B ðt C77
6B C 7
4 @ L _
xðsÞ ds  Bd de ðtÞ C
A5
thðtÞ
(" #" #
T T T
bI T ~I ~I T xðtÞ
¼ 2 ½ xðtÞ hðtÞ  P
AþL M hðtÞ

   ) 
0 0
 xðt  hðtÞÞ  d ðtÞ þ bðtÞ ð12:10Þ
L þ Bf K Bd e

where
ðt  
T T 0 _ T
bðtÞ ¼ 2 ½ xðtÞ hðtÞ  P xðsÞds
L
thðtÞ

Using Lemma 12.1 for

aðsÞ ¼ colf 0, LgxðsÞ
and
b ¼ P col fxðtÞ, hðtÞg
we obtain
 
T T T T 1
xðtÞ
bðtÞ  hM ½ xðtÞ hðtÞ  P ðW H þ IÞH ðH W þ IÞ  P
hðtÞ
 
0
þ 2 ½ xðtÞT hðtÞT  PT W T H  ðxðtÞ  xðt  hðtÞÞÞ
L
ðt  T  
T 0 0
þ xðsÞ H xðsÞds ð12:11Þ
L L
thM

Differentiating the second and third Lyapunov terms in (12.8) give respectively:
_
V_ 2 ðtÞ ¼ xðtÞT Q1 xðtÞ  ð1  hðtÞÞxT
ðt  hðtÞÞ Q1 xðt  hðtÞÞ
 xðtÞT Q1 xðtÞ  ð1  hD ÞxT ðt  hðtÞÞ Q1 xðt  hðtÞÞ ð12:12Þ
 Vibration control of vehicle engine-body systems with time delay 341

ðt
_ T Q2 xðtÞ
V_ 3 ðtÞ ¼ hM xðtÞ _  _ T Q2 xðsÞds
xðsÞ _
thM
ðt
 hM _ T Q2 xðtÞ
xðtÞ _  _ T Q2 xðsÞds
xðsÞ _ ð12:13Þ
thðtÞ

and the time derivative of the last term of V ðtÞ in (12.8) is

 T   ðt  T  
_ _ T 0 0 _ _ T 0 0 _
V 4 ðtÞ ¼ hM xðtÞ H xðtÞ  xðsÞ H xðsÞds ð12:14Þ
L L L L
thM

Again, from the Leibniz–Newton formula, the following equation holds for any
matrices N1, N2 with appropriate dimensions:
0 1
ðt
B _ C
2ðxðtÞT N1 þ xT ðt  hðtÞÞN2 Þ@xðtÞ  xðt  hðtÞÞ  xðsÞdsA¼0 ð12:15Þ
thðtÞ

Using the obtained derivative terms (12.10)–(12.14) and adding the left side of
(12.15) into the derivative of the Lyapunov–Krasovskii functional in (12.8), we
obtain the following result for V_ ðtÞ:
X
4
V_ ðtÞ ¼ V_ i ðtÞ
i¼1 (" #   
T T T bI T ~I ~I T xðtÞ 0
 2 ½ xðtÞ hðtÞ  P þ WTH xðtÞ
AþL M hðtÞ L

       )
T 0 0 0
 W H þ L þ B K xðt  hðtÞÞ  d ðtÞ
L f Bd e
 
T T T T 1 xðtÞ
þ hM ½ xðtÞ hðtÞ  P ðW H þ IÞH ðH W þ IÞP
hðtÞ

þ xðtÞT ðQ1 þ symfN1 gÞ xðtÞ

 xT ðt  hðtÞÞðð1  hD ÞQ1 þ symfN2 gÞ xðt  hðtÞÞ

 T   !
_ T 0 0 _ þ hM JT ðtÞc P1 cT JðtÞ
þ hM xðtÞ P1 þ H xðtÞ 1
L L

þ 2 xT ðtÞðN2T  N1 Þ xðt  hðtÞÞ

ðt
 ðJT ðtÞc þ x_ T ðsÞP1 Þ P1
1 ðJT ðtÞc þ x_ T ðsÞP1 ÞT ds ð12:16Þ
thðtÞ
342 Handbook of vehicle suspension control systems

where the vectors JðtÞ and c are respectively:

JðtÞ :¼ col fxðtÞ, hðtÞ, xðt  hðtÞÞ, de ðtÞg
and
c ¼ colfN1 , 0, N2 , 0, 0g
On the other hand, under asymptotically stable and zero initial condition, the
H1 performance measure can be rewritten as
ð
1

J1 ¼ ½zðtÞT zðtÞ  g2 de ðtÞT de ðtÞdt þ V ðtÞjt!1  V ðtÞjt¼0

0
ð
1

¼ ½zðtÞT zðtÞ  g2 de ðtÞT de ðtÞ þ V_ ðtÞdt ð12:17Þ

0

Substituting the terms of

~ xðtÞ þ D1 K xðt  hðtÞÞ
zðtÞ ¼ C
h i 
_ ¼ bI T ~I ~I T xðtÞ
xðtÞ
hðtÞ
and upper bound of V_ ðtÞ in (12.16) results in (12.17) being less than the integrand
JðtÞT P1 JðtÞ where the matrix P1 is given by
2    3
~ 11 S
S ~ 12 PT 0 hM N1
6 Bd 0 7
6 7
6
P1 :¼ 6  S 22~ 0 hM N2 7 ð12:18Þ
7
4   g2 I 0 5
   hM P1
where
" #! " #
bI T ~I ~I T ~TC
Q1 þ symfN1 g þ C ~ 0
~ 11 :¼ sym P
S T
þ
AþL M  0
" T # " #T " # ! " T # T
~I bI 0 0 ~I bI
þ hM P1 þ H
~I L L ~I
þ hM PT ðW T H þ IÞH 1 ðH W þ IÞP
     
0 0 T ~T
~ 12
S :¼  PT W T H þ PT þ N2  N1 þ C D1 K C
L L þ Bf K 0
~ 22 :¼  ð1  hD ÞQ1  symfN2 g þ ðD1 KÞT ðD1 KÞ
S
Now, if P1 < 0, then J1 < 0, which means that the L2 -gain from the disturbance
de ðtÞ to the controlled output zðtÞ is less than g. It is easy to see that the inequality
 Vibration control of vehicle engine-body systems with time delay 343

P1 < 0 implies S~ 11 < 0. Hence, by Proposition 4.2 in Reference [33], the matrix P
is nonsingular. Then, according to the structure of the matrix P, the matrix
X :¼ P1 has the form:
 
X1 0
X ¼ ð12:19Þ
X3 X2

i ði ¼ 1, 2Þ and X3 ¼ X2 P3 X1 .
where Xi ¼ P1
Let
 
z ¼ diag X T , X1 , I, X1

Premultiplying z and postmultiplying zT to P1 < 0 and considering HW ¼ aI to

eliminate the nonlinearities in the matrix inequality, we obtain (by Schur
complement):
2        T  3
0 b1
hM N ~T ~ b
b
6 S 11
b 12
S  X1 C hM X T I I hM ða þ 1ÞH 7
6 Bd 0 0 ~I
7
6  S22 0 b2
hM N b 1 ÞT
ðD1 X 0 0 7
6 7
6   g2 I 0 0 0 0 7
6 7<0
6    hM X1 0 0 0 7
6 7
6     I 0 0 7
6 7
4      hM X1 0 5
      hM H
ð12:20Þ
where
" #! 2 n o 3
bI T ~I X1 þ ~I T X3 ~I T X2 b 1 þ sym N
Q b1 0
b 11 :¼ sym
S þ4 5
~
A X1 þ L þ M X3 MX2  0
" T # " #T " # " T #
~I bI 0 1 0 ~I bI
T
þ hM X H X
~I L L ~I

and
" #
b 12 :¼ NbT  Nb1
S 2
~ b1
ð1  aÞ L þ Bf X

with X b 1 :¼ KX1 , L b 1 ¼ X T Q1 X1 and N

~ :¼ L X1 , H ¼ H 1 , Q b i ¼ X T Ni X1 ði ¼ 1, 2Þ,
1 1
b b
where the matrices H , Q 1 , N i are, respectively, new decision variables instead of
the matrices H, Q1 , Ni . Obviously, the matrix inequality (12.20) includes multi-
plication of control gain and the decision variable X1 . According to Lemma 12.2,
it is easily seen that the LMI (12.3b) satisfies the matrix inequality
344 Handbook of vehicle suspension control systems
 T  
0 1 0
H < G1 ð12:21Þ
L L
Now, by considering (12.21) and applying Schur complement, the matrix inequality
(12.20) is converted into a convex programming problem written in terms of LMI
(12.3a). This completes the proof. &

12.4.2 Output-feedback control design

In this section, a sufficient stability condition for system (12.2) with the output-
feedback controller stated in Problem 12.2 of the form f ðtÞ ¼ K yðt  hðtÞÞ will
be presented and methods of computing the controller gain matrix will be
proposed.
Theorem 12.2. For given scalars g, hM > 0, hD , there exists a delay-dependent H1
output-feedback control in the form f ðtÞ ¼ K yðt  hðtÞÞ such that the resulting
closed-loop system is robustly asymptotically stable and satisfies the constraint
J1 < 0 if there exist a scalar a, matrices N b 1, N
b 2 , X~ 1 , X2 , X3 and positive-definite
b 1 , G, H , satisfying the following LMIs:
matrices X11 , X22 , Q

2 " #  " # " # 3

Nb T Nb1 0 h M
b1
N X1
~T
C
6 S11 2  S16 S16 hM ðaþ1ÞH 7
6 ~
ð1aÞ LþB ~
f X 1C Bd 0 0 7
6 7
6  7
6  S22 0 hM Nb 2 D1 X~ 1C T 0 0 0 7
6 7
6   g2 I 0 0 0 0 0 7
6 7
6 7 <0
6    hM X1 0 0 0 0 7
6 7
6     I 0 0 0 7
6 7
6      hM X1 0 0 7
6 7
6 7
4       hM G 0 5
       hM H
ð12:22aÞ

2  T 3
0
4 2X1 þG L
~ 5 <0 ð12:22bÞ
 H

where C :¼ ½C3 , C4 . The desired control gain is given by

b 1 from LMIs ð12:22Þ

K ¼ X~ 1 X ð12:23Þ
1

with
 
X11 0
X1 ¼ V VT ð12:24Þ
0 X22
 Vibration control of vehicle engine-body systems with time delay 345

b1 ¼ U C
X b 1 U T
b X11 C ð12:25Þ
where X11 2 <2l2l , X22 2 <2ðnlÞ2ðnlÞ , rankðC3 Þ ¼ l < n and C ¼ U ½ C
b 0  VT
(the singular value decomposition of the matrix C), with rankðCÞ ¼ 2l,
U 2 <2l2l , V 2 <2n2n and Cb 2 <2l2l .
Proof: The proof is similar to proof of Theorem 12.1. Obviously, in this case the
matrix inequality (12.20) includes multiplication of control gain, the matrix C and
the decision variable X1 . Therefore, by introducing X~ 1 :¼ K X
b 1 as a new decision
variable instead of the matrix K in (12.20) under the equality constraint
C X1 ¼ X b 1 C, the LMIs (12.22) are easily concluded (by Schur complement and
Lemma 12.3). This completes the proof. &
Remark 12.5. If rankðCÞ ¼ l ¼ 8, the matrix C is nonsingular; it is clear that
the matrix equation CX1 ¼ Xb 1 C is solvable on X b 1 ¼ C X1 C 1 . In this case,
b 1 , i.e. X
the results of Theorem 12.2 are true for a full (non-diagonal) matrix X1 , i.e.:
 
X11 X12
X1 ¼
 X22
1
and the desired control gain in f ðtÞ ¼ Kyðt  hðtÞÞ is given by K ¼ X~ 1 C X11 C .
Remark 12.6. In the second-order linear system (12.2), taking x1 ðtÞ ¼ xðtÞ,
_ and xðtÞ ¼ col fx1 ðtÞ, x2 ðtÞg yields an augmented system model, i.e. a
x2 ðtÞ ¼ xðtÞ
first-order neutral linear system:
_ ¼ Ae xðtÞ þ Bfe f ðt  hðtÞÞ þ Bde de ðtÞ
Me xðtÞ
where
       
I 0 0 I 0 0
Me ¼ , Ae ¼ , Bfe ¼ , Bde ¼
0 M B A Bf Bd
It is easy to understand that the proposed methods in References [33] and [34] to
find a suitable robust control for the above delay system eventually involve
manipulations of 16-dimensional matrices Me , Ae , Bfe , Bde and hence will increase
the dimension and number of the LMI variables in comparison with our result in
this chapter.
Remark 12.7. It is noted that our approach is different from that in References [8],
[9], [22] and [39] in different perspectives: (a) the system structure in References
[8], [9] and [22] considers a delay-free control input as compared to our case that
considers the time-varying delay (hðtÞ) appears in the actuator force; (b) in Reference
[39] the authors design a delay-dependent H1 state-feedback controller for active
vehicle suspensions with an actuator constant time delay based on a non-convex
method as compared to our case that a convex optimization method is proposed for
state- and output-feedback control designs of vehicle engine-body vibration system
with an actuator time-varying time delay and a relaxation on the bound of the time-
_  hD .
varying delay, i.e. hðtÞ
346 Handbook of vehicle suspension control systems

12.5 Simulation results

The parameters of the vehicle engine-body vibration model, used for the design
and simulation, are given in Table 12.1 [21]. The controlled output zðtÞ can be
considered as
2 3
2ke =Mb 2ðke þ kb Þ=Mb 0 2ðL  lÞke =Mb
6 7
zðtÞ ¼ 6
4 e1 I 0 0 0 7
5
0 e2 I 0 0
2 3 2 3
0 0 0 0 1
6 7 6 7
þ6
4 2ce =Mb 2ðce þ cb Þ=Mb 0 2ðL  lÞce =Mb 7 6 7
5 þ 4 0 5f ðt  hðtÞÞ
0 0 0 0 0
ð12:26Þ

In the vibration control of a vehicle engine-body system, both performances in the

frequency domain and the time domain should be considered. The simulation
results about the frequency-response curve and the transient-response time history
under a vibration disturbance generated by the engine are mainly analysed. For a
delay-free vehicle engine-body model in (12.2), the transfer function of the system
can be generated by using the Mathematica software. For instance, the path from
the actuator to the chassis acceleration has the following form:

N1 ðsÞ
Gf €x 2 ðsÞ ¼ ð12:27Þ
D1 ðsÞ

where N1 ðsÞ is the numerator having the following form:

N1 ðsÞ ¼ 25 Me kb ke s4 þ ð25 Me ce kb þ 25 Me cb ke Þ s5
þ ð25 Me cb ce þ 50 Me Ie kb þ 2 Me Ib ke  6 Me Ie ke Þ s6
þ ð2 Me ce Ib þ 50 Me cb Ie  6 Me ce Ie Þ s7 þ 4 Ib Ie Mb Me s8

Table 12.1 Vehicle–engine body parameters

Parameters Values Parameters Values

Mb 1,000 (kg) Me 250 (kg)
Ib 810 (kg m2 ) Ie 8.10 (kgm2 )
kb 20,000 (N/m) ke 200,000 (N/m)
cb 300 (N/m/s) ce 200 (N/m/s)
Lb 2.5 (m) Le 0.5 (m)
 Vibration control of vehicle engine-body systems with time delay 347

102
Log magnitude

100

10–2 0
10 101 102 103 104
Frequency (rad/s)
0
Phase (deg)

–200

–400

–600 0 1
10 10 102 103 104
Frequency (rad/s)

Figure 12.7 Bode diagram of the system with acceleration output (first
configuration)

and the denominator is

D1 ðsÞ ¼  4 Ib Ie Mb Me s8 þ ð8 ce Ib Ie Mb  8 cb Ib Ie Me
8 ce Ib Ie Me  2 ce Ib Mb Me  50 cb Ie Mb Me
34 ce Ie Mb Me Þ s7 þ    þð100 ce Me kb ke
50 cb Me ke2  72 ce Me ke2 Þ s  50 Me kb ke2  24 Me ke3
According to the vehicle–engine body parameters in Table 12.1, the magnitude and
phase curves of the transfer function (12.27) are plotted in Figure 12.7.
Remark 12.8. In the second configuration shown in Figure 12.6, the transfer
function from the delay-free actuator force f to the displacement of the chassis x2
can be evaluated by using the data table illustrated in Table 12.1:
N1 ðsÞ
Gfx2 ðsÞ ¼
D1 ðsÞ
where

N1 ðsÞ ¼ 0:001 s6  0:0182 s5  13:96 s4  101:4 s3 2:081  104 s2

1:914  105 s  3:536  107
and
D1 ðsÞ ¼ s8 þ 21:67 s7 þ 1:692  104 s6 þ 1:851  105 s5
þ 5:72  107 s4 þ 2:546  108 s3 þ 3:905  1010 s2
þ 2:887  1010 s þ 1:414  1012
348 Handbook of vehicle suspension control systems

and the related Bode diagram is depicted in Figure 12.8. Furthermore, Figure 12.9
shows the Bode diagram of the transfer function from the force actuator f ðtÞ to the
acceleration of the chassis €x 2 ðtÞ.
In real application, an electromechanical shaker is usually used to generate a
broadband random signal considered as the excitation signal which travels into the

Bode diagram
–60
Magnitude (dB)

–80
–100
–120
–140
–160
–180
180
90
Phase (deg)

0
–90
–180
–270
–360
100 101 102 103
Frequency (rad/s)

Figure 12.8 Bode diagram of the system with chassis displacement output (second
configuration)

Bode diagram
–20

–40
Magnitude (dB)

–60

–80

–100
0
–90
Phase (deg)

–180
–270
–360
–450
–540
100 101 102 103
Frequency (rad/s)

Figure 12.9 Bode diagram of the system with acceleration output (second
configuration)
 Vibration control of vehicle engine-body systems with time delay 349

structure through the engine mount. An accelerometer is mounted on the chassis

structure to measure the signal in the feedback control loop. For simulation
purposes, it is now assumed that the low-frequency range of the model is excited by
the vibration disturbance generated by the engine of the form:

500 sin2 ð2pf tÞ, 0  t  1=f

de ðtÞ ¼ ð12:28Þ
0, otherwise
at the frequency range f 2 ½20 Hz, 40 Hz.
First, the problem is to determine the maximum allowable bound of hðtÞ for
guaranteeing the stability of the structure. Now, by solving the LMIs given in
Theorem 12.1 with Matlab LMI Control Toolbox [48] for given parameters e1 ¼ 5,
e2 ¼ 10 and g ¼ 1:1, one can obtain the maximum bound hðtÞ as hM ¼ 0:15. For
simulation purposes, when hðtÞ ¼ 0:05 sin2 ð100ptÞ, i.e. hM ¼ 0:05 and hD ¼
5p > 1, Figures 12.10–12.12 show the comparison of states x1 ðtÞ, x2 ðtÞ and
delay-dependent H1 state-feedback (SF) controller (active force) f ðtÞ found by
(12.4) where
K ¼ 105 ½8:56 72:872 0:094 9:994 0:867 0:832 0 0:024
at two low frequencies 20 Hz and 40 Hz. It is seen from Figures 12.10–12.12 that
the amplitudes of the engine–chassis deflections and of state-feedback control
effort are increasing when the frequency of an engine vibration is decreasing from
40 Hz to a lower frequency at 20 Hz.
Then, we design another H1 state-feedback controller which does not consider
the time-delay problem during the controller design process. After setting e1 ¼ 5,
e2 ¼ 10 and g ¼ 1:1, we obtain such a controller gain as
K ¼ 105 ½0:027 26:771 0:032 8:648 0:293 0:015 0 0:001
and we denote this controller as memoryless control.

0.8
Engine deflection (mm)

0.6

0.4

0.2

–0.2
0 0.5 1 1.5
t (s)

Figure 12.10 Time-response curve of the engine under a disturbance

with an excitation frequency: (a) 40 Hz (dashed line) and
(b) 20 Hz (solid line)
350 Handbook of vehicle suspension control systems

0.3

0.2
Chassis deflection (mm)

0.1

–0.1

0 0.5 1 1.5
t (s)

Figure 12.11 Time-response curve of the chassis under a disturbance with an

excitation frequency: (a) 40 Hz (dashed line) and (b) 20 Hz (solid line)

3,000

2,500

2,000

1,500
Active force (N)

1,000

500

–500

–1,000
0 0.2 0.4 0.6 0.8 1 1.2
t (s)

Figure 12.12 Time-response curve of the active force under a disturbance with an
excitation frequency: (a) 40 Hz (dashed line) and (b) 20 Hz (solid
line)
 Vibration control of vehicle engine-body systems with time delay 351

For subsequent comparison, based on Theorem 12.2 a delay-dependent H1

output-feedback (OF) controller for system (12.2) is designed by setting e1 ¼ 5,
e2 ¼ 10, g ¼ 1:1, C3 ¼ ½0, 1, 0, 0 and C4 ¼ 0 and the control signal
f ðtÞ ¼ 851:448 yðt  hðtÞÞ is obtained.
To compare the performances of three (SF, memoryless and OF) controllers
under the excitation signal (12.28) at the frequency 20 Hz, power spectral density
(PSD) of the engine deflection, chassis deflection and active force are respectively
depicted in Figures 12.13–12.15. These figures show that frequency variations are
strong at low-frequency ranges around 0–100 Hz and frequency variations are
weak at high frequency ranges around 200–500 Hz. It can also be seen from the
Figures 12.13 and 12.14 that the state-feedback controller has more attenuation in
the most significant mode located around low-frequency ranges in comparison with
other two controllers. Furthermore, Figure 12.15 shows that energy of the actuator
force based on H1 state-feedback controller has more attenuation at the high-
frequency ranges as compared to memoryless controller.

12.6 Conclusion
In this chapter, state- and output-feedback H1 controllers are designed for stabili-
zation of bounce and pitch vibrations in vehicle engine-body vibration structures
with a time-varying actuator delay. An appropriate Lyapunov–Krasovskii functional
is constructed using some free weighting matrices to characterize some delay-
dependent sufficient conditions for driving desired state- and output-feedback

–80
with delayed SF control
–90 with memoryless SF control
with delayed OF control
–100
Engine deflection PSD (dB/Hz)

–110

–120

–130

–140

–150

–160
0 0.1 0.2 0.3 0.4 0.5
Frequency (kHz)

Figure 12.13 PSD of the engine deflection

352 Handbook of vehicle suspension control systems

–70
with delayed SF control
with memoryless SF control
–80
with delayed OF control
Chassis deflection PSD (dB/Hz)

–90

–100

–110

–120

–130

–140

–150
0 0.1 0.2 0.3 0.4 0.5
Frequency (kHz)

Figure 12.14 PSD of the chassis deflection

50
with delayed SF control
45 with memoryless control
40
Active force PSD (dB/Hz)

35

30

25

20

15

10

5
0 0.1 0.2 0.3 0.4 0.5
Frequency (kHz)

Figure 12.15 PSD of the active force

controllers in terms of LMIs. State- and output-feedback stabilization schemes have

been developed directly instead of coupling the model to a first-order system such
that the closed-loop feedback enjoys the delay-dependent asymptotic stability with a
prescribed g-level L2 -gain (or H1 performance). The feedback gains are determined
by convex optimization over LMIs. Finally, simulation results have been given to
show the effectiveness of the method.
 Vibration control of vehicle engine-body systems with time delay 353

References
[1] Tokhi O. and Veres S., Active sound and vibration control: Theory and
applications, IEE Control Engineering Series, vol. 62, London, UK, 2002.
[2] Karkosch H. J., Svaricek F., Shoureshi R. and Vance J. L., ‘Automotive
applications of active vibration control’, Proceedings of the European
control conference, 2000.
[3] Krtolica R. and Hrovat D., ‘Optimal active suspension control based on
A half-car model’, Proceedings of the 29th CDC, pp. 2238–2243, 1990.
[4] Preumont A., Vibration control of active structures: An introduction, Kluwer
Academic, The Netherlands, 1997.
[5] Yu Y., Naganathan N. G. and Dukkipati R. V., ‘A literature review of
automotive vehicle engine mounting systems’, Mechanism and Machine
Theory, vol. 36, pp. 123–142, 2001.
[6] Elliott S. J. and Nelson P. A., ‘Active noise control’, IEEE Signal Processing
Magazine, vol. 10, no. 4, pp. 12–35, October 1993.
[7] McDonald A. M., Elliott S. J. and Stokes M. A., ‘Active noise and vibration
control within the automobile’, Proceedings of the active control of sound
and vibration, pp. 147–157, Tokyo 1991.
[8] Seba B., Nedeljkovic N., Paschedag J. and Lohmann B., ‘Feedback control
and FX-LMS feedforward control for car engine vibration attenuation’,
Applied Acoustics, vol. 66, pp. 277–296, 2005.
[9] Yang J., Suematsu Y. and Kang Z., ‘Two-degree-of-freedom controller to
reduce the vibration of vehicle engine-body system’, IEEE Transactions on
Control Systems Technology, vol. 9, no. 2, pp. 295–304, March 2001.
[10] Kim G. and Singh R., ‘A study of passive and adaptive hydraulic engine
mount systems with emphasis on nonlinear characteristics’, Journal of
Sound and Vibration, vol. 179, pp. 427–453, 1995.
[11] Fuller C. R. and von Flotow A. H., ‘Active control of sound and vibration’,
Control Systems Magazine, pp. 9–19, December 1995.
[12] Nagaya K. and Li L. ‘Control of sound noise radiated from a plate using
dynamic absorbers under optimization by neural network’. Journal of Sound
and Vibration, vol. 208, no. 2, pp. 289–298, 1997.
[13] Aglietti G., Stoustrup J., Rogers E., Langley R., and Gabriel S., ‘LTR control
methodologies for micro vibrations’, Proceedings of the IEEE CCA,
September 1998.
[14] Zhao Y. Y. and Xu J., ‘Effects of delayed feedback control on nonlinear
vibration absorber system’, Journal of Sound and Vibration, vol. 308,
pp. 212–230, 2007.
[15] Wang Z. H. and Hu H. Y., ‘Stabilization of vibration systems via delayed
state difference feedback’, Journal of Sound and Vibration, vol. 296,
pp. 117–129, 2006.
[16] Oglac N. and Holm-Hansen B. T., ‘A novel active vibration absorption
technique: Delayed resonator’, Journal of Sound and Vibration, vol. 176,
no. 1, pp. 93–104, 1994.
354 Handbook of vehicle suspension control systems

[17] Cavallo A., Maria G. and Setola R., ‘A sliding manifold approach for vibration
reduction of flexible systems’, Automatica, vol. 35, pp. 1689–1696, 1999.
[18] Hong J. and Bernstein D. S., ‘Bode integral constraints, collocation and spill
over in active noise and vibration control’, IEEE Transactions on Control
Systems Technology, vol. 6, no. 1, pp. 111–120, 1998.
[19] Kamman J. W. and Naghshineh K., ‘A comparison of open-loop feedforward
and closed-Loop methods for active noise control using volume velocity
minimization’, Applied Acoustics, vol. 57, pp. 29–37, 1999.
[20] Weng M., Lu X. and Tumper D., ‘Vibration control of flexible beams using
sensor averaging and actuator averaging methods’, IEEE Transactions on
Control Systems Technology, vol. 10, no. 4, pp. 568–577, July 2002.
[21] Karimi H. R., ‘Optimal vibration control of vehicle engine-body system
using Haar functions’, International Journal of Control, Automation, and
Systems, vol. 4, no. 6, pp. 714–724, 2006.
[22] Karimi H. R. and Lohmann B., ‘Haar wavelet-based robust optimal control
for vibration reduction of vehicle engine-body system’, Electrical Engi-
neering, vol. 89, no. 6, pp. 469–478, June 2007.
[23] Zhu X., Hua C. and Wang S., ‘State feedback controller design of networked
control systems with time delay in the plant’, International Journal on Inno-
vative Computing, Information and Control, vol. 4, no. 2, pp. 283–290, 2008.
[24] Basin M., Sanchez E. and Martinez-Zuniga R., ‘Optimal linear filtering for
systems with multiple state and observation delays’, International Journal
Of Innovative Computing, Information and Control, vol. 3, no. 5, pp. 1309–
1320, 2007.
[25] Niculescu S. I., Delay effects on stability: A robust control approach, Berlin:
Springer, 2001.
[26] Jiang X. and Han Q. L., ‘Delay-dependent robust stability for uncertain
linear systems with interval time-varying delay’, Automatica, vol. 42,
pp. 1059–1065, 2006.
[27] Karimi H. R., and Gao H., ‘LMI-based delay-dependent mixed H2 =H1
control of second-order neutral systems with time-varying state and input
delays’, ISA Transactions, vol. 47, no. 3, pp. 311–324, 2008.
[28] Karimi H. R., ‘Observer-based mixed H2 =H1 control design for linear sys-
tems with time-varying delays: An LMI approach’, International Journal of
Control, Automation, and Systems, vol. 6, no. 1, pp. 1–14, 2008.
[29] He Y., Wang Q. G., Lin C. and Wu M., ‘Delay-range-dependent stability
for systems with time-varying delay’, Automatica, vol. 43, pp. 371–376, 2007.
[30] Karimi H. R., Zapateiro M. and Lou N., ‘Robust mixed H2 =H1 delayed
state-feedback control of neutral delay systems with time-varying delays’,
Asian Journal of Control, vol. 10, no. 5, pp. 571–582, 2008.
[31] Lam J., Gao H. and Wang C., ‘H1 model reduction of linear systems with
distributed delay’, IEE Proceedings of the Control Theory Applications,
vol. 152, no. 6, pp. 662–674, 2005.
[32] Chen J. D., ‘LMI-based robust H1 control of uncertain neutral systems with
state and input delays’, Journal of Optimization Theory and Applications,
vol. 126, pp. 553–570, 2005.
 Vibration control of vehicle engine-body systems with time delay 355

[33] Fridman E. and Shaked U., ‘A descriptor system approach to H1 control

of linear time-delay systems’ IEEE Transactions on Automatic Control,
vol. 47, no. 2, pp. 253–270, 2002.
[34] Gao H. and Wang C., ‘Comments and further results on a descriptor system
approach to H1 control of linear time-delay systems’, IEEE Transactions on
Automatic Control, vol. 48, pp. 520–525, 2003.
[35] Gao H., Lam J. and Wang C., ‘Multi-objective control of vehicle active
suspension systems via load-dependent controllers’, Journal of Sound and
Vibration, vol. 290, pp. 654–675, 2006.
[36] Basin M., Perez J. and Calderon-Alvarez D., ‘Optimal filtering for linear
systems over polynomial observations’, International Journal of Innovative
Computing, Information and Control, vol. 4, no. 2, pp. 313–320, 2008.
[37] Wang M., Chen B. and Tong S., ‘Adaptive fuzzy tracking control for strict-
feedback nonlinear systems with unknown time delays’, International
Journal of Innovative Computing, Information and Control, vol. 4, no. 4,
pp. 829–838, 2008.
[38] Shi P., Zhang J., Qiu J. and Xing L., ‘New global asymptotic stability cri-
terion for neural networks with discrete and distributed delays’, Proceedings
of the Institution of Mechanical Engineers, Part I, Journal of Systems and
Control Engineering, vol. 221, pp. 129–135, 2007.
[39] Du H. and Zhang N., ‘H1 control of active vehicle suspensions with actuator
time delay’, Journal of Sound and Vibration, vol. 301, pp. 236–252, 2007.
[40] Henrion D., Sebek M. and Kucera V., ‘Robust pole placement for second-
order systems: An LMI approach’, http://citeseer.ist.psu.edu/623785.html,
October 2002.
[41] Balas M. J., ‘Trends in large space structure control theory: Fondest
hopes, wildest dreams’, IEEE Transactions on Automatic Control, vol. 27,
pp. 522–535, March 1982.
[42] Rincon F., ‘Feedback stabilization of second-order models’, Ph.D. disserta-
tion, Northern Illinois University, De Kalb, IL, 1992.
[43] Bhaya A. and Desoer C., ‘On the design of large flexible space structures
(LFSS)’, IEEE Transactions on Automatic Control, vol. AC-30, pp. 1118–
1120, November 1985.
[44] Park P., ‘A delay-dependent stability criterion for systems with uncertain
time-invariant delays’, IEEE Transactions on Automatic Control, vol. 44,
pp. 876–877, 1999.
[45] Ho D. W. C. and Lu G., ‘Robust stabilization for a class of discrete-time
nonlinear via output feedback: The unified LMI approach’, International
Journal of Control, vol. 76, no. 2, pp. 105–115, 2003.
[46] Xu S., Lam J., Yang C. and Verriest E. I., ‘An LMI approach to guaranteed
cost control for uncertain linear neutral delay systems’, International Jour-
nal of Robust and Nonlinear Control, vol. 13, pp. 35–53, 2003.
[47] Hale J. and Verduyn Lunel S. M., Introduction to functional differential
equations, Springer-Verlag, New York, NY, 1993.
[48] Gahinet P., Nemirovsky A., Laub A. J. and Chilali M., LMI control toolbox:
For use with Matlab, Natik, MA: The MATH Works, 1995.
 Chapter 13
Frequency domain analysis and design
of nonlinear vehicle suspension systems
Yue Chen, Xingjian Jing and Li Cheng

Abstract
In the analysis and design of vehicle suspension systems, springs and dampers,
which are usually inherently nonlinear, are the most crucial elements to improve the
ride comfort, assure the stability, and increase the longevity of suspension systems to
a large extent. Therefore, it is of great significance to determine a proper stiffness
and damping characteristics to meet various requirements in practice. In this study,
a nonlinear frequency domain analysis method is introduced for nonlinear analysis
and design of vehicle suspension systems. The explicit relationship between system
output spectrum and model parameters is derived by using the nonlinear frequency
domain analysis method, and the characteristic parameters of interest can therefore
be analyzed directly. The optimal nonlinear stiffness and damping characteristics of
vehicle suspension systems can then be achieved. Comparative studies indicate that
the optimal nonlinear damping characteristics demonstrate better dynamic perfor-
mance than the corresponding linear counterparts and several existing nonlinear
optimal damping characteristics obtained by simulations. Simulation studies based
on the full vehicle dynamic model verify the nonlinear advantages in terms of three
different vehicle evaluation standards. The study shows that the nonlinear optimal
damping characteristic obtained by using the nonlinear frequency domain analysis
method is very helpful in improvement of vehicle vibration performance and
decrease of suspension stroke. Meanwhile, the optimized nonlinear damper will not
cause any negative effect on the handling capability.

13.1 Introduction
Vehicle suspension is the combination of springs, shock absorbers and linkages. It
plays an important role in connecting vehicle chassis to its wheels. For a vehicle

Department of Mechanical Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon,
Hong Kong
358 Handbook of vehicle suspension control systems

suspension system, it has three main purposes: first, to transmit force and torque to
the chassis to make sure the vehicle works well; second, to reduce and isolate
vibration by using its springs and dampers; third, to make sure wheels to jump as a
given trajectory by using its guide components. An ideal vehicle suspension system
should be able to reduce the acceleration and displacement of the vehicle body to
meet the requirement from ride comfort. Meanwhile, the suspension system should
also meet the requirement from handling ability [1]. To achieve the desired perfor-
mance, vehicle suspension systems have been widely investigated and studied for a
long time. Previous studies in the design and analysis of vehicle suspension systems
show that a suspension system can be divided into three main types, passive sus-
pension, semi-active suspension, and active suspension according to force generation
mode [2]. Each type of suspension systems has its own advantages and dis-
advantages. Active and semi-active suspension system can show better performance
in vehicle vibration and handling performance, but passive suspension systems still
dominate the market [3]. Despite the use of numerous advanced technologies and
control strategies in active and semi-active suspension systems, it is significant to
investigate inherent nonlinear properties in springs and dampers of passive suspen-
sion systems for better performance. In this study, nonlinear stiffness and damping
characteristics will be systematically analyzed using a frequency domain method,
and the relationship with various vehicle performances will be discussed.
Previous studies often regard the suspension systems as the model which
consists of springs and dampers. An important characteristic of any spring is that it
can store energy from bumps and acceleration easily. However, these vehicle
springs, such as leaf spring, coil spring, and rubber spring, cannot release energy in
a desired way and often leads to an unstatisfactory life cycle before failure [4].
Therefore, a damper should be well designed in suspension systems as it plays an
important role in the suppression of vehicle vibration by damping out vertical
motion and thus increasing spring’s life cycle. With a proper damping level, the car
will have a good riding comfort and handling ability. How to design a proper and
realistic damper is of great significance in the vehicle suspension system, especially
the damping coefficient. In most vehicle dynamic analysis, the damping coefficient
is based on the assumption that the damping force is proportional to the velocity of
the damper piston, which mainly considers the fact of simplicity, and easiness in
implementation and computer simulation [4–8]. However, the relationship between
the damping force and piston velocity in both desired and realistic dampers is not
proportional at low and high velocity. In practice the damping characteristics for a
suspension damper can be obtained approximately as shown in Fig. 13.1.
From Fig. 13.1 it is obvious that the damping force is piecewise linear instead of
a single linear function of the damping coefficient and damper piston velocity.
Meanwhile, with the wide usage of some novel dampers, such as magnetorheological
fluid dampers, electrorheological fluid dampers, and some other smart materials, the
analysis and design of the damping characteristics cannot be simply regarded as a
linear relationship. For example, the viscoelastic damper is not only affected by
the piston velocity but also exhibits a complex nonlinear function between the
damping force and the deformation [9, 10]. The spring system also has nonlinear
Frequency domain analysis and design of nonlinear vehicle suspension systems 359

d
c

Damping force
b
Rebound
a

e Piston velocity (m/s)

f
Compression

Fig. 13.1 Characteristics curve for a suspension damper [4]

characteristics; for example, the air suspension system has the quadratic and cubic
nonlinearities after further calculation [11]. The characteristics of rubber components
which are widely used in vehicle suspension systems are also nonlinear [12]. There-
fore, in the analysis of vehicle vibration performance, nonlinear effects need to be
considered to make sure the accuracy of the suspension system.
In order to analyze nonlinear suspension systems, some methods in the literature
can be used. There are several analysis methods especially for nonlinear systems,
such as harmonic balance methods [13, 14], describing function methods [15, 16] and
averaging methods [17, 18]. The harmonic balance method is based on the assump-
tion that the solution can be represented in the form of a Fourier series. The describing
function method is based on quasi-linearization, which is the approximation of the
nonlinear system under investigation by a linear system transfer function that depends
on the amplitude of the input waveform. The averaging method is based on the
averaging principle when the exact differential equation of the motion is replaced by
an averaged equation. However, these methods cannot well reflect the relationship
between the vehicle performance and system parameters. Therefore, a systematic
frequency domain method based on Volterra series expansion for nonlinear systems,
which has been developed in recent years, was adopted in this study [19–27]. This
method is to determine the generalized frequency response functions (GFRFs) for the
nonlinear system which can be described by the Volterra series. Based on this con-
cept, many works have been carried out to analyze the dynamic characteristics of
nonlinear systems in frequency domain [22–24]. By using this method, the system
output spectrum can be derived to analyze the effect of nonlinear parameters on the
output spectrum. Numerical methods to determine the nonlinear output spectrum of a
nonlinear system are also developed using a parametric characteristic analysis
method, which can greatly facilitate the analysis and design of important physical
parameters in the nonlinear system [24, 25]. The main advantage of this numerical
determination approach is that it can explain how the system output frequency
response is affected by the system nonlinear parameters. By using the numerical
method, the system output frequency response can be derived easily and engineers
360 Handbook of vehicle suspension control systems

can analyze the system in terms of any model parameters of interest [25]. This fre-
quency domain analysis method makes the nonlinear system analysis much more
straightforward and easily understood.
The main objective of this work is to employ the nonlinear frequency domain
analysis method together with existing knowledge of vehicle suspension system to
investigate the relationship between the performance of vehicle suspension systems
and their inherent parameters or some control parameters in order to meet different
requirements. According to the previous result [27], by properly introducing the
nonlinearities and determining the model parameter of interest the system may
suppress vibration and achieve a better output. As the relationship between the
system output and system parameters can be derived, it provides a useful tool by
which the optimal value of system parameters can be designed to achieve different
performance requirements.
The rest of the chapter is organized as follows.
In section 13.2, the mathematical model of the vehicle suspension system is
provided. The theory of the nonlinear frequency domain analysis method is introduced
in detail. The system output is derived as per the knowledge of the frequency domain
analysis method. The optimal value of the nonlinear parameters is also derived.
In section 13.3, the existing optimal nonlinear damping characteristics are
adopted from [28] and the detailed function of the nonlinear damping character-
istics are obtained using the fitting method. In this study, another nonlinear optimal
damping characteristics are obtained by using the nonlinear frequency domain
analysis method. Then the comparative studies between these two nonlinear
damping characteristics are conducted to show which one will be better in the
suppression of the vibration performance. The dynamic model of the spring damper
system is built to verify the accuracy of the result.
In section 13.4, the full-vehicle dynamic model is built in Adams/View. Then
the nonlinear damping characteristics are imported to this dynamic model and the
comparative studies are made to show their effect on the vehicle ride comfort,
suspension stroke and the handling ability.
Section 13.5 provides a conclusion and discusses some further development
topics.

13.2 System model and the output frequency response

function (OFRF) method
In this section, the vehicle suspension model is presented first. Then the relation-
ship between system output and system parameters is derived by using the non-
linear frequency domain analysis method. Finally, the optimal value of the system
parameters is obtained.

13.2.1 System model

The vehicle suspension system can be investigated by using the three different types
of models, the quarter-vehicle suspension model [4, 5], half-vehicle suspension
model [29], and the full-vehicle suspension model [30]. Quarter-vehicle suspension
Frequency domain analysis and design of nonlinear vehicle suspension systems 361
x2
ms

Fc
Fk

x1

Fig. 13.2 1 DOF quarter vehicle suspension model

is the simplest one and it is mainly used for the analysis of the vehicle’s vertical
movement and vibration. For a half-vehicle suspension system, it can well reflect the
vertical vibration and slip motion. The full-vehicle suspension system can be used to
investigate the vertical vibration, slip motion, and roll behavior under road dis-
turbances. In the present study, the quarter suspension system is adopted to be the
system model and it can be seen in Fig. 13.2.
The mass ms represents the quarter of the body mass. The vertical displacement
of the sprung mass is x2 and the base excitation displacement is x1 . For the non-
linear suspension model, the damping force can be described as Fc ¼ cx_ þ c1 x_ 3 þ
c2 x_ 2 x þ c3 xx
_ 2 and the spring force can be written as Fk ¼ kx þ c4 x3 . Here, k is the
linear stiffness and c is the linear damping, c1 , c2 , c3 , c4 are the nonlinear values, x is
the relative displacement between the sprung mass and the base excitation, which can
be defined as
x ¼ x2  x 1 ð13:1Þ
Then the governing equation for this suspension system can be written as
ms€x 2 þ Fk þ Fc ¼ 0 ð13:2Þ
Substituting, Fc , Fk , Eq. (13.2) can be written as
ms€x þ kx þ cx_ þ c1 x_ 3 þ c2 x_ 2 x þ c3 xx
_ 2 þ c4 x3 ¼ ms€x 1 ð13:3Þ
Assuming that the base excitation is considered to be a sinusoidal function which
can be written as
x1 ¼ Y sinðwtÞ ð13:4Þ
where w is the frequency and Y is the magnitude of the base motion. In order to
conduct an analysis which is not specific to particular choices of system initial
parameters, such as the sprung mass and the spring stiffness, the non-dimensional
form of the governing equation can be derived as
€y ðtÞ þ yðtÞ þ xy_ ðtÞ þ x1 y_ ðtÞ3 þ x2 y_ ðtÞ2 yðtÞ þ x3 y_ ðtÞyðtÞ2 þ x4 yðtÞ3 ¼ Y sinðWtÞ
ð13:5Þ
362 Handbook of vehicle suspension control systems

where
 
pffiffiffiffiffiffiffiffiffiffi w 2 t kzðtÞ
t ¼ w0 t, w0 ¼ k=ms , W ¼ , Y ¼ ms w , zðtÞ ¼ xðtÞ ¼ x , yðtÞ ¼
w0 w0 Y
2 2 2 2
c c1 Y c2 Y c3 Y c4 Y
x ¼ pffiffiffiffiffiffiffiffi , x1 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffi , x2 ¼ 2 , x3 ¼ pffiffiffiffiffiffiffiffi , x4 ¼ 3
kms 3 k ms k 5m k
ðkm Þ s
s

In the present study, the main objective is to analyze the nonlinear effect on the
vibration performance. The evaluation standard of the vibration performance is
the mean square body acceleration. Let the output be the transmitted force F. Then
the objective is to minimize the transmitted force. As the system output spectrum
will reach the maximum value at the resonance frequency point, therefore the input
frequency is chosen to be system’s natural frequency and thus W ¼ 1. The detailed
form of the transmitted force can be measured by

F ¼ yðtÞ þ xy_ ðtÞ þ x1 y_ ðtÞ3 þ x2 y_ ðtÞ2 yðtÞ þ x3 y_ ðtÞyðtÞ2 þ x4 yðtÞ3 ð13:6Þ

Therefore, the vehicle suspension system can be described as a non-dimensional

model as
(
€y ðtÞ þ yðtÞ þ xy_ ðtÞ þ x1 y_ ðtÞ3 þ x2 y_ ðtÞ2 yðtÞ þ x3 y_ ðtÞyðtÞ2 þ x4 yðtÞ3 ¼ Y sinðWtÞ
F ¼ yðtÞ þ xy_ ðtÞ þ x1 y_ ðtÞ3 þ x2 y_ ðtÞ2 yðtÞ þ x3 y_ ðtÞyðtÞ2 þ x4 yðtÞ3
ð13:7Þ

The objective of this study is to analyze the effect of the nonlinear terms with
coefficients x1 , x2 , x3 , x4 on the system output spectrum. In order to determine the
nonlinear parameters’ effect on the output, the system output spectrum needs to be
obtained first. In section 13.2.2, two approaches based on the nonlinear frequency
domain analysis method are given to derive the system output spectrum.

13.2.2 Determination of the system OFRF

In this section, a brief review about the theory of nonlinear frequency domain
analysis method is given first and then an efficient numerical method is discussed
for the determination and optimization of nonlinear output spectrum in terms of
system parameters. By using the steepest decent method, the optimal value of the
nonlinear system parameters is obtained.

13.2.2.1 Nonlinear output spectrum: the theory

For nonlinear systems, the output f ðtÞ can be expressed by a Volterra functional
polynomial of the input uðtÞ [31] as
X
N
f ðtÞ ¼ fn ðtÞ ð13:8Þ
n¼1
Frequency domain analysis and design of nonlinear vehicle suspension systems 363

where N is the maximum order of the system nonlinearity, and then the nth-order
output of the system is given by
ð1 ð1 Y
n
fn ðtÞ ¼ ... hn ðt1 , . . . , tn Þ uðt  ti Þdti ð13:9Þ
1 1 i¼1

where hn ðt1 , . . . , tn Þ is the real function of t1 , . . . , tn and is defined as the ‘‘nth-order

kernel’’ or ‘‘nth-order impulse response’’ of the system. Then the system ‘‘nth-order
transfer function’’ (GFRF) can be derived by using the multi-dimensional Fourier
transform of the nth-order impulse response, which can be written as

ð
1 ð
1

Hn ð jw1 , . . . , jwn Þ ¼ ... hn ðt1 , . . . , tn Þexpðjðw1 t1 þ    þ wn tn ÞÞdt1 , . . . , dtn

1 1
ð13:10Þ

And when the system is subjected to an input:

X
K  
uðtÞ ¼ Ai  cosðwi þ ffAi Þ ð13:11Þ
i¼1

According to [20], the system output spectrum can be written as

XN
1 X
Fð jwÞ ¼ n Hn ðjw1 , . . . , jwn ÞAðwk1 Þ    Aðwkn Þ ð13:12Þ
n¼1
2 w1 þþw ¼w
n

where
   
Aðwki Þ ¼ Aki ejffAjki j sgnðki Þ for ki 2 1, . . . ,  K ,

1 a0  
sgnðaÞ ¼ , wki 2 w1 , . . . ,  wK
1 a < 0

Therefore, the nonlinear system output spectrum can be obtained according to

the theory above, and probing method, the following result is obtained [32]:

X
K
Ln ð jw1 þ    þ jwn ÞHn ð jw1 , . . . , jwn Þ ¼ c0, n ðk1 , . . . , kn Þð jw1 Þk1    ð jwn Þkn
k1 , kn ¼1
!
X nq
n1 X X q
L
Y
kpþi

þ cp, q k1 , . . . , kpþq jwnqþi Hnq, p jw1 , . . . , jwnq

q¼1 p¼1 k1 , kpþq ¼0 i¼1

X
n X
K

þ cp, 0 k1 , . . . , kp Hn, p ð jw1 , . . . , jwn Þ

p¼2 k1 , kp ¼0

ð13:13Þ
364 Handbook of vehicle suspension control systems

npþ1
X
Hn, p ðÞ ¼ Hi ð jw1 , . . . , jwi Þð jwiþ1 , . . . , jwn Þð jw1 þ    þ jwi Þkp ð13:14Þ
i¼1

Hn, 1 ð jw1 , . . . , jwn Þ ¼ Hn ð jw1 , . . . , jwn Þð jw1 þ    þ jwn Þk1 ð13:15Þ

PK
where Ln ð jw1 þ    þ jwn Þ ¼  k1 ¼0 c1, 0 ðk1 Þð jw1 þ    þ jwn Þk1 . From Eqs. (13.12)
and (13.13), the system output polynomial can be derived.
Single-input-two-output systems, output frequency response can be referred
to [23]. In this study, the non-dimensional system output frequency response is
calculated up to the fifth order and the detailed expression can be written as
follows:
1. The first-order frequency response is
xð jw1 Þ þ 1
H12:1 ð jw1 Þ ¼ ð13:16Þ
ð jw1 Þ2 þ xð jw1 Þ þ 1
2. The third-order frequency response is
2 3
x4 þ x3 ð jw3 Þ
H1 ð jw1 ÞH2 ð jw2 ÞH3 ð jw3 Þ4 þ x2 ð jw3 Þð jw2 Þ 5
þ x1 ð jw1 Þð jw2 Þð jw3 Þ
H31:111 ð jw1 , jw2 , jw3 Þ ¼ 
ð jw1 þ jw2 þ jw3 Þ2 þ xð jw1 þ jw2 þ jw3 Þ þ 1
ð13:17aÞ
 
H32:111 ðjw1 , jw2 , jw3 Þ ¼ ð jw1 þ jw2 þ jw3 Þ2 H31:111 ð jw1 , jw2 , jw3 Þ
ð13:17bÞ
3. The fifth-order frequency response is much more complicated and will not be
given in this section due to space constraint.
Then the system output can be derived according to Eq. (13.12), which can be
written as
X 5
1 X
Fð jwÞ ¼ n Hn2 ð jw1 , . . . , jwn ÞAðwk1 Þ . . . Aðwkn Þ
n¼1
2 w1 þþwn¼w
1 2:1 1 X
¼ H1 ð jw1 ÞAðwÞ þ 3 H 2:111 ð jw1 , . . . , jw3 ÞAðwk1 Þ . . . Aðwk3 Þ
2 2 w1 þþw3 ¼w 3
1 X
þ 5 H 2:11111 ð jw1 , . . . , jw5 ÞAðwk5 Þ . . . Aðwk5 Þ
2 w1 þþw5 ¼w 5
ð13:18Þ

13.2.2.2 Numerical determination of nonlinear output spectrum

In section 13.2.2.1, the analytical computation of the nonlinear output spectrum has
been discussed. In practice, a more efficient numerical method can be adopted,
which allows nonlinear output spectrum to be determined directly in terms of
Frequency domain analysis and design of nonlinear vehicle suspension systems 365

system physical parameters. To this end, the system output spectrum can be written
into a more explicit polynomial form as follows [24, 25, 33]:
X
N
Fð jwÞ ¼ CEðHn ð:ÞÞjn ð jwÞT ð13:19Þ
n¼1

where
ð Y
n
1
fn ð jwÞ ¼ pffiffiffiffi fn ð jw1 , . . . , jwn Þ Uð jwi Þdsw ð13:20Þ
N ð2pÞn1 i¼1
w1 þþwn ¼w

CEð:Þ is a coefficient extraction operator which has two fundamental operations

‘‘’’ and ‘‘’’. The detailed definitions are given in [33], and CEðHn ð:ÞÞ is the
parametric characteristics of the nth-order GFRF Hn ð:Þ, which can be written as
   
n1 nq n
CEðHn ðÞÞ ¼ C0,n    Cp,q CEðHnqpþ1 ð:ÞÞ   Cp,0 CEðHnpþ1 ð:ÞÞ
q¼1 p¼1 p¼2

ð13:21Þ
Obviously, Eq. (13.19) can be written as

Fð jwÞ ¼ yfð jwÞT ð13:22Þ

where
N
y¼  CEðH ð:ÞÞ,
n¼1
n fð jwÞ ¼ ½f1 ð jwÞ, f2 ð jwÞ, . . . , fn ð jwÞ ð13:23Þ

Therefore, the determination of the output spectrum by using the numerical

determination method in this study can be carried out by the following steps:
(1) Determination of the parametric characteristics of OFRF
The first step of this task is to determine the largest order N. This can be
done by evaluating the magnitude of the nth-order output frequency response
Fn ð jwÞ. If the magnitude bond of Fn ð jwÞ is less than a predefined value, then
the largest order N can be obtained. The second step is to determine the
parametric characteristics according to Eq. (13.21). For a given system
Eq. (13.5), the nonlinear parameters of interest are x1 , x2 , x3 , x4 and all the
other parameters are zero. According to [24], the parametric characteristics of
OFRF can be shown as

y ¼ ½1, x1 , x2 , x3 , x4 , x1 2 , x1 x2 , x1 x3 , x1 x4 , x2 2 , x2 x3 , x2 x4 , x3 2 , x3 x4 , x4 2 , . . .
ð13:24Þ

(2) Determination of fð jwÞ for the OFRF

The first step of this task is to construct a non-singular matrix which should
cover a large range. Then the second step is to obtain the output frequency
response Fð jwÞ of the system for different combinations of nonlinear
366 Handbook of vehicle suspension control systems

parameters x1 , x2 , x3 , x4 . This can be achieved by using fast Fourier transform

on the time domain output response. Then fð jwÞ for the OFRF can be written as

1
fð jwÞT ¼ yT y yT F ðjwÞ ð13:25Þ
Therefore, the OFRF of the system with respect to nonlinear parameters
x1 , x2 , x3 , x4 can be obtained according to Eq. (13.21).
In conclusion, the main idea of numerical determination method is that, given a
nonlinear model, CEðHn ð:ÞÞ can be derived according to Eq. (13.21) and fn ð jwÞ
can be obtained by the numerical method which will be discussed in detail in the
next section. Then the system output spectrum will be achieved according to
Eq. (13.22), and finally frequency domain analysis can be conducted.

13.2.3 Optimization and system analysis

In this section, the relationship between the system output and the system parameters
of interest is derived. Then the optimal value of the system nonlinear parameters is
obtained by using the steepest decent method. Finally, some conclusions about the
advantage of system nonlinearity on the system vibration control are provided.

13.2.3.1 Computation of the nonlinear output spectrum

With the nonlinear output spectrum of the vehicle suspension system derived
above, parameter optimization can then be conducted to find the optimal nonlinear
parameters x1 , x2 , x3 , x4 in order to achieve the best vibration suppression in the
system output which is defined proportionally to the vertical acceleration of the
vehicle. To understand the nonlinear output spectrum with respect to any nonlinear
parameters of interest in the system, consider the output spectrum with respect to
only two parameters which are defined as x1 , x2 . Following the numerical method
above, it can be eventually obtained that

Fð jWÞW¼1 ¼ ð1:5599e  001 þ 1:6160e  001iÞ þ ð2:4599e  002
þ 5:7895e  004iÞ
x1 þ ð4:8410e  002  3:1037e  002iÞ
x2
þ ð2:5315e  003  2:1792e  003iÞ
x21 þ ð1:2208e  002
þ 1:0914e  002iÞ
x1
x2 þ ð3:4105e  003  1:5597e  003iÞ
x22
þ ð1:5519e  004  9:5445e  005iÞ
x31 þ ð2:9229e  004
 6:2336e  005iÞ
x21
x2 þ ð2:0404e  003  1:8717e  003iÞ
x1
x22
þ ð5:2480e  004 þ 1:0759e  003iÞ
x32 þ ð2:4643e  005
þ 2:5870e  005iÞ
x41 þ ð4:5495e  005  6:8368e  005iÞ
x31
x2
þð1:1712e  004 þ 1:1884e  004iÞ
x21
x22 þ ð4:6285e  005
þ 4:7699e  005iÞ
x1
x32 þ ð4:5823e  005  7:5763e  005iÞ
x42
ð13:26Þ
where the frequency W can be chosen at any values, for example here W ¼ 1 rad=s.
In order to well reflect the relationship between the system output spectra the initial
damping value is z ¼ 0.01 and the magnitude of the system is Y ¼ 0:2. The result is
shown in Fig. 13.3.
Frequency domain analysis and design of nonlinear vehicle suspension systems 367

ζ3 = ζ4 = 0

0.7
0.6
2│F(ω)│

0.5
0.4
0.3
0.2
0 0
2 2
4 4
6 6
8 8
10 10 ζ1
ζ2

Fig. 13.3 Nonlinear output spectrum with respect to x1 ; x2 (the stars are obtained
by the theoretical approach)

From Fig. 13.3, it can be seen that the output spectrum is a typical nonlinear
function of nonlinear parameters x1 , x2 . The nonlinear output spectrum provides a
straightforward and powerful insight into the analytical relationship between sys-
tem output response and physical parameters.
In order to find the optimal parameter values for x1 , x2 , x3 , x4 the nonlinear
output spectrum with respect to all these four parameters can be derived by following
the procedure in the previous section. It should be noted that the range of the nonlinear
parameters should be sufficiently large in order to get a globally optimal solution.
However, when the parameters cover a large range, the matrix inverse in Eq. (13.24)
is easy to be ill-conditioned. To solve this problem, matrix y can be written as [24]

y ¼ 1, aðx1 =aÞ, aðx2 =aÞ, aðx3 =aÞ, aðx4 =aÞ, a2 ðx1 2 =a2 Þ, a2 ðx1 x2 =a2 Þ,
a2 ðx1 x3 =a2 Þ, a2 ðx1 x4 =a2 Þ, a2 ðx2 2 =a2 Þ, a2 ðx2 x3 =a2 Þ, a2 ðx2 x4 =a2 Þ,

a2 ðx3 2 Þ, a2 ðx3 x4 =a2 Þ, a2 ðx4 2 =a2 Þ ð13:27Þ

Then Eq. (13.22) can be written as

Fð jwÞ ¼yfð jwÞT ¼ 1, ðx1 =aÞ, ðx2 =aÞ, ðx3 =aÞ, ðx4 =aÞ, ðx1 2 =a2 Þ,
ðx1 x2 =a2 Þ, ðx1 x3 =a2 Þ, ðx1 x4 =a2 Þ, ðx2 2 =a2 Þ, ðx2 x3 =a2 Þ, ðx2 x4 =a2 Þ,

ðx3 2 =a2 Þ, ðx3 x4 =a2 Þ, ðx4 2 =a2 Þ f0 ð jwÞ, af1 ð jwÞ, af2 ð jwÞ, af3 ð jwÞ,
af4 ð jwÞ, a2 j5 ð jwÞ, a2 f6 ð jwÞ, a2 f7 ð jwÞ, a2 f8 ð jwÞ, a2 f9 ð jwÞ, a2 f10 ð jwÞ,
T
a2 f11 ð jwÞ, a2 f12 ð jwÞ, a2 f13 ð jwÞ, a2 f14 ð jwÞ
ð13:28Þ
Using this scaling method, the parameters could not only cover a large range,
but also ensure the non-singularity of the matrix in Eq. (13.22).
368 Handbook of vehicle suspension control systems

13.2.3.2 Parameter optimization

According to Eq. (13.22), the polynomial with respect to all the nonlinear para-
meters can be derived. To this end, the following points should be noted:
(a) Matrix y should cover a large range to make sure that the optimal results can be
found as mentioned above. However, usually when the range of the nonlinear
parameter is too large, the matrix inverse in Eq. (13.24) would be ill-conditioned.
To solve this problem, the range of the variables can also be divided into
several parts. For example, the range of ½0, 100 can be divided into ½0, 1 ,
½1, 10 , ½10, 100 . Using this method, together with the scaling method discussed
in Eq. (13.27–13.28), the matrix will be easy to be non-singular and the accuracy
of the solution can be guaranteed.
(b) The optimal solution should be the optimal one within all the sub-ranges.
(c) The nonlinear parameters x1 , x2 , x3 , x4 have their physical meaning. There-
fore, some parameters can be negative while the others cannot.
Moreover, in practice to further restrict the searching space, two new variables
a, b can be introduced, which are defined as

x1 ¼ a, x2 ¼ 3ab, x3 ¼ 3ab2 , x4 ¼ ab3 ð13:29Þ

Therefore, Eq. (13.6) can be written as

F ¼ yðtÞ þ xy_ ðtÞ þ a½y_ ðtÞ þ byðtÞ 3 ð13:30Þ

Then the system output spectrum is a function with respect to two variables
a, b and can be obtained by the frequency domain analysis method. In this study,
the range of a, b is [0, 10] and [1, 0], respectively. The system output frequency
response with respect to nonlinear variables is obtained as

F ðjWÞjW¼1 ¼ ð1:2297e  001 þ 2:6101e  001iÞ þ ð1:4043e  002

 2:1368e  002iÞ
a þ ð3:1627e  002 þ 3:7759e  003iÞ
3ab
þ ð7:5741e  003 þ 2:3285e  002iÞ
3ab2 þ ð4:8602e  003
þ 1:7763e  001iÞ
ab3 þ ð1:8905e  003 þ 1:9025e  003iÞ
a2
þ ð2:8603e  003 þ 9:4444e  004iÞ
3a2 b þ ð6:9260e  005
 8:4367e  004iÞ
3a2 b2 þ ð1:2002e  002 þ 1:3644e  002iÞ
a2 b3
þ ð1:5270e  003  9:8513e  004iÞ
9a2 b2 þ ð1:6447e  003
 2:7494e  004iÞ
9a2 b3 þ ð2:2671e  003 þ 2:4648e  002iÞ
3a2 b4
þ ð5:7254e  003  3:0388e  003iÞ
9a2 b4 þ ð3:5859e  003
þ 2:4658e  002iÞ
3a2 b5 þ ð7:3119e  002 þ 2:0120e  001iÞ
a2 b6
ð13:31Þ
Note that the parameter here is dimensionless. Therefore, a value for
qffiffiffiffiffiffiffiffiffiffiffiffiffi
a ¼ 0.1, for example, is equivalent to the value c1 ¼ a ðkms Þ3 =Y 2 ¼ 75248:3
Frequency domain analysis and design of nonlinear vehicle suspension systems 369

1.2
1
2│F(ω)│

0.8
0.6
0.4
0.2
0 8
–0.2 7
6
–0.4 min 5
–0.6 4
3
–0.8 2
1
b –1 0 a

Fig. 13.4 Nonlinear output spectrum with respect to a; b

(e.g. k ¼ 16,000, ms ¼ 240, and Y ¼ 100). Also, note that the parameter (a) takes very
small values; the scaling method and subsection method above can still be applied
similarly in order to avoid matrix singularity. Using the steepest decent method, the
optimal solution can be obtained. For the first polynomial, the optimal value is
a ¼ 3:89, b ¼ 0:77, 2jFð jwÞj ¼ 0:251 ð13:32Þ
According to Eq. (13.5), the definition of W is the ratio of the input frequency to
the system’s natural frequency. In this study, the range of W is ½0:1, 10 , and then the
realistic road input frequency is ½0:1w0 , 10w0 , which can cover the normal working
frequency range [8]. For an optimal value, it should not only suppress the vibration in
the resonance frequency, but also be helpful in other frequency range. For the present
study, system output spectrum with respect to a and b will be given in Fig. 13.4.

13.2.3.3 Comparison with linear systems

For pure linear system, the system spring force is Fk ¼ kx and the damper force is
_ Then the system transfer function can be written as
Fc ¼ cx.
xð jWÞ þ 1
H ðjWÞ ¼ ð13:33Þ
ð jWÞ2 þ xð jWÞ þ 1
From Eq. (13.33), it can be shown that when the input frequency W ¼ 1 rad/s,
the transfer function is a decreasing function and the minimum is 1 when the linear
damping z is as large as possible. However, the vehicle suspension linear damping
usually gets the value of z ¼ 0.25 [8]. In this study, the recommended linear damping
will be used to compare with the nonlinear optimal damping. The system spectrum
for the system with respect to different system values can be shown in Table 13.1.
In order to verify whether the nonlinear optimal value has a positive effect on
the system output spectrum in the whole frequency range, the system transmissibility
370 Handbook of vehicle suspension control systems

Table 13.1 System output spectrum

z a b 2|F(jW)|
0.01 0 0 20
0.25 0 0 0.825
0.01 3.89 0.77 0.251

50
Linear ζ = 0.01
40 Linear with recommended ζ = 0.25
Nonlinear with optimization
30 ζ = 0.01

20
Magnitude (dB)

10

–10

–20 ζ = 0.25

–30

–40
10–1 100 10
1

Ω (rad/s)

Fig. 13.5 Force transmissibility for different systems

is introduced in this section according to [26], as the figure of the system transmis-
sibility can well reflect how the system parameters affect the system output just as
the transfer function does in the linear theory. In this study, the system transmissi-
bility can be defined as the system output divided by input which can be written as
F
T¼ ð13:34Þ
Y
The figure of the system transmissibility can be plotted in Fig. 13.5.
By further inspection of Fig. 13.5, some conclusions are made as follows:
1. It is obvious that when the nonlinear parameters get an optimal value, the system
transmissibility at the frequency W ¼ 1 is more excellent than the initial linear
system. In Fig. 13.5, the peak at the frequency of W ¼ 1 means the maximum
transmissibility. It can be shown that the nonlinear optimal system can get the
smallest transmissibility at the frequency W ¼ 1, which means that when
the system is under the same input signal, the system output spectrum with the
nonlinear optimal value will get the lowest output spectrum. Therefore, it can be
concluded that the system vibration is well suppressed at the system resonant
frequency when the nonlinear system gets the optimal value.
Frequency domain analysis and design of nonlinear vehicle suspension systems 371

2. From Fig. 13.5, it is also clear that the optimal nonlinear value can not only
minimize the transmissibility in the resonant frequency, but also be helpful
for the system vibration suppression in other frequency ranges. In the high-
frequency range, the nonlinear optimal value can keep the transmissibility
unchanged just as the curve obtained by the initial damping value, which is
much better than the curve obtained by the recommended damping value. In
the low-frequency range, the transmissibility does not change with the change
in the linear damping value; the three curves match very well. However, at
a frequency of about W ¼ 0.5, the transmissibility obtained by the optimal
nonlinear system is bigger than the other two systems and there exists a peak.
This is because that in this study the term nonlinear stiffness is also introduced.
This will decrease the system stiffness, and then the resonant frequency will be
decreased. Therefore, the nonlinear system transmissibility will get a max-
imum value below the system’s natural frequency of W ¼ 1. This will also be
helpful as the vehicle suspension system resonance frequency will be changed
by introducing the nonlinear terms, which can be used to design the suspension
systems in order to avoid some important frequency points, such as the human
sensitive frequency or the engine resonant frequency.
In this study, the area which was combined by the three output spectrum
curves, x ¼ 0.1, x ¼ 10, and y ¼ 0 can be calculated to show the effect of optimal
nonlinearity in the whole frequency range. The detailed steps of calculating the
output spectrums can be seen in [7, 8]. Because the unit of y label is N and the unit
of x label is rad/s, in this study the unit of the area is N.rad/s. The areas of different
systems can be seen in Table 13.2.
From Table 13.2 it can be shown that the area obtained by the nonlinear system
is the smallest one in the three different systems above. Therefore, the optimal
nonlinear system is much more competitive than the linear system in the resonant
frequency range.

13.2.4 Conclusion
Based on some previous works, a brief introduction about the nonlinear frequency
domain analysis method is given in this section. The output spectrum obtained by the
analytical determination approach and numerical determination approach is used to
analyze the effect of the model parameters on the nonlinear vehicle suspension
system. From the result it can be shown that the system output spectrum varies with
respect to different system parameters. Then the output spectrum derived by the

Table 13.2 Area of different systems

System Area (N.rad/s)

Linear, z ¼ 0.01 1.3091
Linear, z ¼ 0.25 0.71116
Nonlinear optimal system 0.47281
372 Handbook of vehicle suspension control systems

numerical determination approach is used to find the optimal solution because of its
advantage in computation cost. The nonlinear optimal solution is obtained and
compared with the initial linear system and the system with recommended damping
values. The results show that when the system nonlinear parameters are set to the
optimal value, the output spectrum can be much better than the linear system at the
resonant frequency range. This is very useful in the application of the suppressing
vibration. Note that the optimal value of the system parameters presented in this
study is obtained at a given input signal. When the input signal is changed, the
corresponding optimal will be different. Therefore, when the system input is given,
the system vibration performance will be improved by properly determining the
model parameters. In the next sections, the OFRF-based analysis method will be
used to find the optimal value of a realistic suspension system. Different evaluation
standards will be adopted to analyze the effects of system nonlinearity.

13.3 Comparative studies

In this section, the nonlinear frequency domain analysis method is applied to obtain
the nonlinear optimal damping characteristics, and the comparative studies between
two nonlinear optimal damping characteristics are conducted to verify the advan-
tage of the nonlinear frequency domain analysis method. In section 13.3.1, the
existing optimal vehicle damping characteristics [28] are derived by using a curve
fitting method. Section 13.3.2 analyzes the relationship between the system output
spectrum and the system nonlinearities. In section 13.3.3, comparative studies
between these two optimal damping characteristics are conducted. Finally, a
dynamic model of the spring damper system is built to verify the effectiveness of
the result derived from the pure theoretical approaches.

13.3.1 Existing nonlinear damping characteristics

In this section, existing nonlinear optimal damping characteristics are adopted
to perform the comparative study. The authors consider the ride comfort and
safety as main performance index, and a weighted objective function is used as
follows;
Y ¼ wc Yc þ ws Ys ð13:35Þ
where wc and ws are the weight values for ride comfort and safety respectively,
while Yc and Ys are the criteria for the ride comfort and safety and Y is the overall
evaluation function. The optimal damping characteristics, as example, can be seen
in Fig. 13.6.
In Fig. 13.6, the curves show the optimal damping characteristics for a sus-
pension system. But the complete set of data, such as the detailed function of the
damping characteristics, has not been given. In order to make a comparative study,
one method is to fit the data from Fig. 13.6. By using the fitting method, the
optimal damping characteristics can be obtained and the damping force can be
described as the function with respect to the velocity. According to the definition of
Frequency domain analysis and design of nonlinear vehicle suspension systems 373

Damper force FD (N)

3000
wc = 1 ws = 0
wc = 0.8 ws = 0.2
wc = 0.5 ws = 0.5
2000 wc = 0.2 ws = 0.8
wc = 0 ws = 1

1000

–1000

–2.0 –1.0 0.0 1.0 2.0 3.0

Relative velocity vD (m/s)

Fig. 13.6 Optimal damping characteristics [28]

wc and ws , the curve obtained under the condition of wc ¼ 1, ws ¼ 0 is the objective

curve that is used to obtain the data.
There are many techniques that can be used to fit the data, such as least square
fitting method [34], interpolation method [35], and nonlinear least square fitting
method [34]. In the present study, the least square method is adopted to fulfill this
objective mainly due to the fact that the present data is relatively simple. The
method is used to find the best polynomial for a set of points by minimizing the
sum of the squares of the offsets which are obtained from the given curve [36],
which can be written as
X
min ½yi  f ðxi , a1 , a2 , . . . , an Þ 2

where yi are the points in the curve and the f ðxi , a1 , a2 , . . . , an Þ are the values of the
polynomial derived by fitting method respectively. In the present study, the fitting
process is finished by Matlab and the damping force can be described as the nth
degree polynomial:

y ¼ an xn þ an1 xn1 þ an2 xn2 þ    þ a1 x þ a0 ð13:36Þ

where an is the coefficient and y is the final polynomial. In Eq. (13.36), x is the
velocity, an is the relative damping value and y is the damping force. In this form,
the damping force is the function with respect to the velocity. Note that, when the
velocity x ¼ 0, the damping force should be 0. By choosing different orders of the
equation above, the polynomial can be achieved. The last step is to fit the poly-
nomial in Matlab GUI. The desired form of the damping function can be seen in
374 Handbook of vehicle suspension control systems

Eq. (13.36). The data of velocity which is used to fit the polynomial is
½0:5, 0:5 m=s. Then the polynomial of the damping characteristics can be
achieved readily and it can be written as

Ffit ¼ 1641x_ 3 þ 1882x_ 2 þ 937:2x_ ð13:37Þ

where Ffit is the damping x_ force and is the velocity. In Eq. (13.37), the last term is
the linear damping and the first two terms are the nonlinear damping effect.

13.3.2 Damping characteristics designed via the

OFRF-based analysis method
In this section, the nonlinear optimal damping characteristics are obtained by the
nonlinear frequency domain analysis method.

13.3.2.1 Derivation of the optimal damping characteristics

In this section, the model is the same as the model presented in section 13.2.1, and
it can be seen in Fig. 13.7.
The definitions can be seen in the previous sections, except the nonlinear
damping force Fc . In this section, the main objective is to compare the nonlinear
optimal damping characteristics obtained by the OFRF-based analysis method with
the existing nonlinear optimal damping curve. The nonlinear damping force is
Fc ¼ cx_ þ ax_ 2 þ bx_ 3 and therefore the governing equation can be written as

ms€x þ kx þ cx_ þ ax_ 2 þ bx_ 3 ¼ ms€x 1 ð13:38Þ

where a and b are the nonlinear damping parameters which need to be determined
in this study. The transmitted force can be written as

F ¼ kx þ cx_ þ ax_ 2 þ bx_ 3 ð13:39Þ

According to [28] and [37], the linear damping value is 937:2 N  s=m, and the
detailed values of the stiffness and mass are ms ¼ 290 kg, k ¼ 16812 N=m. The
input signal is assumed to be a sine wave and the amplitude is 50 mm, which can be

x2
ms

Fc
Fk

x1

Fig. 13.7 1 DOF quarter-vehicle suspension model

Frequency domain analysis and design of nonlinear vehicle suspension systems 375

approximately regarded as the real road amplitude [38]. Then the system governing
equation can be obtained as

290€x þ 16812x þ 937:2x_ þ ax_ 2 þ bx_ 3 ¼ 840:6084€x 1 ð13:40Þ

Finally, after some calculations and simplifications, the system governing the
equation and objective function can be written as

290€x þ 16812x þ 937:2x_ þ ax_ 2 þ bx_ 3 ¼ 840:6084€x 1
ð13:41Þ
F ¼ 16812x þ 937:2x_ þ ax_ 2 þ bx_ 3
In this study, the objective is to reduce the transmitted force between the
sprung mass and the base excitement. In order to realize this objective, the system
output frequency response function should be obtained. The steps to derive the
system output spectrum have been discussed in the previous sections and thus are
not given here due to space constraint. Then the system output spectrum with
respect to system parameters a, b can be seen in Fig. 13.8.
From Fig. 13.8, it is clear that when the values of a and b increase, the system
output spectrum decreases, which means the decrease in transmitted force. If the
objective is to reduce the transmitted force, the values of a and b should be large.
However, this does not mean that in order to get the minimum transmitted force,
the value of a and b should be large enough or infinity. There are two reasons. The
first reason is that when the values of a and b are very large, then the equivalent
damping becomes very large. And the second reason is that if the values of a and b
are very large, the damping force in the compression period is positive, but this is
not acceptable. In the present study, the combination of a ¼ 3800 and b ¼ 4000 is

2000
Transmitted force

1800

1600

1400

1200
0
500
1000
0
1500 500
2000 1000
1500
2500 2000
2500
3000 3000
b a

Fig. 13.8 Nonlinear output spectrum with respect to a and b

376 Handbook of vehicle suspension control systems

regarded as the optimal damping characteristics determined by the nonlinear fre-

quency domain analysis method and it can be written as
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 ð13:42Þ

13.3.3 Comparative studies

In sections 13.3.1 and 13.3.2, two optimal damping characteristics are obtained.
The first one is the damping characteristics fitted from a previous literature, and it
can be written as
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 ð13:43Þ
The second one is obtained by using the nonlinear frequency domain analysis
method and it can be written as

Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 ð13:44Þ

In this study, these two damping characteristics are used to make several
comparative studies to show their effects in suppression of the vibration. The sys-
tem performance under different input signals is discussed in this section.

13.3.3.1 Change in input frequency

In this case, the input frequency is chosen as the variables, and the system transmitted
force of the linear system and two nonlinear optimal systems is given as in Fig. 13.9.

104
Linear
a = 1882, b = 1641
a = 3800, b = 4000
Transmitted force (N)

103

2
10
101
ω (rad/s)

Fig. 13.9 Transmitted force withrespect to different systems

Frequency domain analysis and design of nonlinear vehicle suspension systems 377

In Fig. 13.9, the transmitted forces of three different systems are given. The
frequency of the input signal is chosen from 4 rad/s to 10 rad/s. It can be shown that
in the low-frequency range, the vibration performance of the nonlinear system does
not have significant improvement when compared with the linear system. In the
resonant frequency, the vibration can be suppressed significantly. Some specific
cases are given as follows.
1. The input signal is x1 ¼ 0:05 sinð7:6tÞ (see Table 13.3 and Fig. 13.10)

Table 13.3 System output when the input

frequency is 7.6 rad/s

Damping characteristics System output (N)

Fd ¼ 937:2x_ 2.1599eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 1.5364eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 1.2780eþ003

2500
Linear
2000 a = 1882,b = 1641
a = 3800,b = 4000
1500
Transmitted force (N)

1000

500

–500

–1000

–1500

–2000

–2500
188 190 192 194 196 198 200
t (s)

Fig. 13.10 Transmitted force in time domain when the input frequency is 7.6 rad/s

2. The input signal is x1 ¼ 0:05 sinð6tÞ (see Table 13.4 and Fig. 13.11)

Table 13.4 System output when the input

frequency is 6 rad/s

Damping characteristics System output (N)

Fd ¼ 937:2x_ 1.0875eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 1.0564eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 1.0277eþ003
378 Handbook of vehicle suspension control systems
1500
Linear
a= 1882,b= 1641
1000 a= 3800,b= 4000
Transmitted force (N)

500

–500

–1000

–1500
188 190 192 194 196 198 200
t (s)

Fig. 13.11 Transmitted force in the time domain when the input frequency
is 6 rad/s
3. The input signal is x1 ¼ 0:05 sinð10tÞ (see Table 13.5 and Fig. 13.12)

Table 13.5 System output when the input

frequency is 10 rad/s

Damping characteristics System output (N)

Fd ¼ 937:2x_ 1.8516eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 1.5529eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 1.3699eþ003

2000
Linear
a = 1882,b= 1641
1500
a = 3800,b= 4000

1000
Transmitted force (N)

500

–500

–1000

–1500

–2000
188 190 192 194 196 198 200
t (s)

Fig. 13.12 Transmitted force in the time domain when the input frequency is
10 rad/s
Frequency domain analysis and design of nonlinear vehicle suspension systems 379

From Figs. 13.10 to 13.12, it is obvious that in the resonant frequency range,
the optimal damping characteristics obtained by the OFRF-based analysis method
are much better than the existing nonlinear optimal damping characteristics and the
linear damping characteristics. In the low- or high-frequency range, the optimal
damping characteristics are still better than the fitted optimal damping value. But
the vibration performance will not improve too much. Therefore, it can be con-
cluded that at the resonant frequency range, the nonlinear damping characteristics
are better than the linear damping characteristics in the vibration suppression;
the nonlinear optimal damping characteristics obtained by using the nonlinear
frequency domain analysis method are better than the previous optimal nonlinear
damping characteristics.

13.3.3.2 Change in input magnitude

In this case, the input magnitude is chosen to be variables and the system transmitted
force of the linear system and two nonlinear optimal systems is given in Fig. 13.13.
In Fig. 13.13, the transmitted force of three different systems is given. The
magnitude of the input signal is chosen from 0.01 m to 0.1 m to show the effect of
nonlinearities on the system vibration suppression. It is clear that when the magnitude
is large, the vibration performance obtained by the nonlinear damping characteristics
will be more obvious. Meanwhile, the damping characteristics obtained by using the
nonlinear frequency domain analysis method are better than the existing optimal
damping characteristics. Some specific cases are given as follows.

4500
Linear
4000 a = 1882, b = 1641
a = 3800, b = 4000
3500

3000
Transmitted force (N)

2500

2000

1500

1000

500

0
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Input magnitude (m)

Fig. 13.13 Transmitted force with respect to different input magnitudes

380 Handbook of vehicle suspension control systems

1. The input signal is x1 ¼ 0:07 sinð7:6tÞ (see Table 13.6 and Fig. 13.14)

Table 13.6 System output when the input magnitude

is 0.07 m

Damping characteristics System output (N)

Fd ¼ 937:2x_ 3.0238eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 1.9315eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 1.6117eþ003

4000
Linear
a = 1882,b= 2641
3000
a = 3800,b= 4000

2000
Transmitted force (N)

1000

–1000

–2000

–3000

–4000
188 190 192 194 196 198 200
t (s)

Fig. 13.14 Transmitted force in the time domain when the input magnitude
is 0.07 m

2. The input signal is x1 ¼ 0:1 sinð7:6tÞ (see Table 13.7 and Fig. 13.15)

Table 13.7 System output when the input magnitude

is 0.01 m

Damping characteristics System output (N)

Fd ¼ 937:2x_ 4.3198eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 2.4686eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 2.1056eþ003
Frequency domain analysis and design of nonlinear vehicle suspension systems 381
5000
Linear
4000 a = 1882,b = 1641
a = 3800,b= 4000
3000
Transmitted force (N)

2000

1000

–1000

–2000

–3000

–4000

–5000
188 190 192 194 196 198 200
t (s)

Fig. 13.15 Transmitted force in the time domain when the input magnitude is
0.1 m

3. The input signal is x1 ¼ 0:03 sinð7:6tÞ (see Table 13.8 and Fig. 13.16)

Table 13.8 System output when the input

magnitude is 0.03 m

Damping characteristics System output (N)

Fd ¼ 937:2x_ 1.2959eþ003
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 1.0659eþ003
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 913.8744

1500 Linear
a = 1882,b= 1641
a = 3800,b= 4000
1000
Transmitted force (N)

500

–500

–1000

–1500
188 190 192 194 196 198 200
t (s)

Fig. 13.16 Transmitted force in the time domain when the input magnitude
is 0.03 m
382 Handbook of vehicle suspension control systems

4. The input signal is x1 ¼ 0:01 sinð7:6tÞ (see Table 13.9 and Fig. 13.17)

Table 13.9 System output when the input

magnitude is 0.01 m

Damping characteristics System output (N)

Fd ¼ 937:2x_ 431.9741
Ffit ¼ 937:2x_ þ 1882x_ 2 þ 1641x_ 3 417.3850
Fd ¼ 937:2x_ þ 3800x_ 2 þ 4000x_ 3 398.9664

500
Linear
400 a = 1882, b= 1641
a = 3800,b = 4000
300
Transmitted force (N)

200

100

–100

–200

–300

–400

–500
188 190 192 194 196 198 200
t (s)

Fig. 13.17 Transmitted force in the time domain when the input magnitude is
0.01 m

From Figs. 13.14 to 13.17, it can be seen that nonlinear damping characteristics
are better than the linear damping characteristics. To be more specific, the non-
linear optimal damping characteristics obtained by the nonlinear frequency domain
analysis method are much more competitive than the existing optimal damping
values when the signal input magnitude ranges from 0.01 m to 0.1 m. When the
input magnitude is larger, the performance of the damping characteristics obtained
by nonlinear frequency domain analysis method is better than the fitted optimal
damping characteristics. When the input magnitude is relatively small, the damping
characteristics obtained by nonlinear frequency domain analysis method are still
better than the fitted optimal damping characteristics. Therefore, it can be con-
cluded that the damping characteristics obtained by nonlinear frequency domain
analysis method are more effective than the existing optimal damping character-
istics when the input magnitude changes.
Frequency domain analysis and design of nonlinear vehicle suspension systems 383

13.3.3.3 Comparison between small damping system and

nonlinear damping system
The comparative studies between the system with small linear damping and the
system with small linear damping and nonlinear damping are conducted to show
the effect of system nonlinearities. Here, the system mass and spring stiffness are
the same as described in the previous sections and the linear damping value is
c ¼ 100 N  s=m. The nonlinear system is obtained by adding some nonlinear terms
on the linear ones. In order to show the effect of system nonlinearity on the
vibration performance, the nonlinear term values of both systems are assumed to be
100x_ 2 þ 1000x_ 3 . The system input is x1 ¼ 0:05 sinð7:6tÞ. Then the system trans-
mitted force at the resonant frequency can be obtained.
1. c ¼ 100 N  s=m

Table 13.10 System output when the linear damping

is c ¼ 100 N  s=m

Damping characteristics System output (N)

Fd ¼ 100x_ 1.8468eþ004
Ffit ¼ 100x_ þ 100x_ 2 þ 1000x_ 3 2.3437eþ003

× 104 c = 100 N . s/m

2
Linear
a=100,b =1000
1.5

1
Transmitted force (N)

0.5

–0.5

–1

–1.5

–2
0 5 10 15 20 25 30 35 40
t (s)

Fig. 13.18 Transmitted force in the time domain when the linear damping is
c ¼ 100 N  s=m

From Fig. 13.18, it can be concluded that system nonlinearities can be effective
in suppressing the system vibration for the small damping system. In addition, when
the system damping value is small, the effect of nonlinearities will be more obvious.
Meanwhile, the system nonlinearities can also be helpful in reducing transient state
and make the system become stable quickly.
384 Handbook of vehicle suspension control systems

13.3.3.4 Comparison between large damping system and

nonlinear damping system
The comparative studies between the system with large linear damping and the
system with large damping and nonlinear damping are conducted to show the
effect of system nonlinearities. Here, the system mass and spring stiffness are
the same as described in the previous sections and the linear damping value is
c ¼ 5000 N  s=m. The nonlinear system is obtained by adding some nonlinear
terms on the linear ones. In order to show the nonlinearity on the system perfor-
mance, the nonlinear term value of the systems is assumed to be 5000x_ 2 þ 8000x_ 3 .
The system input is x1 ¼ 0:05 sinð7:6tÞ. Then the system transmitted force at the
resonant frequency can be obtained.

Table 13.11 System output when the linear

damping is c ¼ 5000 N  s=m

Damping characteristics System output (N)

Fd ¼ 5000x_ 921.6483
Ffit ¼ 5000x_ þ 5000x_ 2 þ 8000x_ 3 916.5248

c = 5000 N · s/m
1000
Linear
800 a =5000,b =8000

600
Transmitted force (N)

400

200

–200

–400

–600

–800

–1000
0 5 10 15 20 25 30 35 40
t (s)

Fig. 13.19 Transmitted force in the time domain when the linear damping is
c ¼ 5000 N  s=m

From Fig. 13.19, it can be concluded that the vibration performance of the
nonlinear large damping system cannot be improved significantly when compared
with the linear system. The time for both systems from transient state to stable state
is almost the same. In this case, the system nonlinearities cannot show any
improvement as compared to the linear system.
Frequency domain analysis and design of nonlinear vehicle suspension systems 385

13.3.4 Dynamic model verification

In the earlier sections, it can be seen that the nonlinear damping characteristics
obtained by the nonlinear frequency domain analysis method are better than the
previous optimal damping characteristics and both of these two damping char-
acteristics are better than the linear one. However, these results are obtained based
on the theoretical and mathematical study, and it is necessary to conduct some
experiments to verify whether the result is accurate on the real dynamic model. In
this study, the experiment is replaced by the simulation study based on Adams. The
dynamic model of the spring damper system can be built by using Adams and it can
be seen in Fig. 13.20.
In Fig. 13.20, there are three balls and each ball has different meanings. Ball A
represents the body mass of. Ball B represents the base excitation. In the present
study, the base excitation is a sine function, and ball C is used to regard as the
reference coordinate. Note that the distance between ball A and ball B should be
reasonable; otherwise, the simulation study will fail. A simple case is that if the
magnitude of the movement of ball A is larger than the distance between A and B,
then A will move to the location lower than B and this is not reasonable. In this
case, the base excitation in the Adams model can be written as

x1 ¼ 50 sinð7:6  timeÞ mm ð13:45Þ

By using the base excitation above, the system vertical accelerations of three
different systems with linear damping, the nonlinear damping obtained in the lit-
erature and that obtained by the nonlinear frequency domain analysis method can
be achieved and is shown in Fig. 13.21.

Fig. 13.20 Dynamic model of the 1 DOF spring damper system

386 Handbook of vehicle suspension control systems

model_1
10000.0
.cm.Translational_Acceleration.Y_linear
.cm.Translational_Acceleration.Y_nonilnear fitted damping
Acceleration (mm/s2)

.cm.Translational_Acceleration.Y_nonilnear OFRF damping

5000.0

0.0

–5000.0

–10000.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Time (s)

Fig. 13.21 Vertical acceleration with respect to three different systems with linear
damping

Table 13.12 RMS and maximum of the vertical acceleration

Maximum vertical RMS of vertical

acceleration (mm/s2) acceleration (mm/s2)
Linear damping 7198 5084
Nonlinear damping 6634 3878
by fitted method
Nonlinear damping 6331 3368
by OFRF method

From Figure 13.21 and Table 13.12, it is clear that the nonlinear damping is more
effective in suppressing the vertical vibration. Meanwhile, the nonlinear damping
characteristics obtained by the nonlinear frequency domain analysis method is better
than the fitted curve, which means that the nonlinear frequency domain analysis
method is effective in the analysis and design of the damping characteristics.

13.3.5 Conclusion
In this section, the comparative results show that the nonlinear optimal damping
characteristics obtained by using the nonlinear frequency domain analysis method
are more effective than the existing nonlinear optimal damping characteristics. By
using the nonlinear frequency domain analysis method, the relationship between the
system output spectrum and system nonlinear parameters can be derived. Through
further analysis of the system output, the following conclusion can be made:
1. The system output spectrum is a decreasing function with respect to nonlinear
parameters a and b.
2. The value of the nonlinear parameters a and b need to be determined properly
in order to make sure the real damping characteristics are accurate and
reasonable.
Frequency domain analysis and design of nonlinear vehicle suspension systems 387

From the comparative studies presented above, it is clear that the nonlinear
damping characteristics obtained by using the nonlinear frequency domain analysis
method have a better performance than the existing optimal damping character-
istics. However, it is also necessary to do some studies to analyze their effect on
the vehicle dynamic models. Meanwhile, it is still unknown whether the induced
nonlinearities can have some negative effects on the other vehicle evaluation
standards, such as the handling ability and suspension stroke. In the next section, a
full-vehicle dynamic model is built to analyze the induced nonlinear effects on
these evaluation indexes.

13.4 Application on a dynamic vehicle model

It is hard to realize the proposed nonlinear optimal damping characteristics in a real

vehicle. Therefore, the simulation study is adopted and it is mainly based on the
model built in Adams/View. Note that the full-vehicle dynamic model built in this
study is mainly to verify the nonlinear effect on the vehicle vibration performance.
Meanwhile, some other performances are also obtained in this study, such as the
suspension stroke and handling ability. In this section, the detailed parameters of
the vehicle system are taken from [39], such as mass, inertia, and kinematics
parameters. A brief description of the vehicle dynamic model is provided. The
simulation studies are conducted under different conditions to show the nonlinear
effect on various vehicle evaluation standards.

13.4.1 Dynamic vehicle model

The dynamic vehicle model is divided into several subsystems, which are the
chassis system, front suspension system, steering system, rear suspension system,
tire, and road. In these subsystems, the relationship between each component can
be reflected by the graphical topology. The graphical topology of the chassis sys-
tem can be seen in Fig. 13.22.
In Fig. 13.22, it can be seen that the chassis is connected with different
components of the vehicle through some detailed joints. For example, the con-
nection style between the chassis and the low control arm is revolute joint while
the connection between the chassis and the steering wheel is cylindrical joint.
Meanwhile, a cylindrical motion is added in order to make sure the steering wheel
works well.
In this section, the details regarding steps of building the chassis system, sus-
pension system, steering system, tire, and road are not provided due to space con-
straint. The full-vehicle system model can be seen in Fig. 13.23.
In Fig. 13.23, the basic components of a vehicle dynamic model are given. The
detailed characteristics of the components such as the mass, inertia, length, and
width are described in Fig. 13.24.
In Fig. 13.24, ms , m11 , m12 , m21 , m22 are the masses for the vehicle body and the
mass of the four wheels c11 , c12 , c21 , c22 are the damping coefficients of suspension
388 Handbook of vehicle suspension control systems

Right_LCA Right_UCA

Pilman_arm Left_LCA

Idler_arm Chassis Left_UCA

Steering_– Right_RCA

Steering_– Left_RCA

Fig. 13.22 Graphical topology of chassis

Last_run time= 0.0000 Frame=01 Last_run time= 0.0000 Frame=01

Last_run time= 0.0000 Frame=01

Fig. 13.23 Full-vehicle dynamic model

systems and k11 , k12 , k21 , k22 are the spring stiffness of the suspension system. In the
real working conditions of the vehicle, zr1 , zr2 , zr3 , zr4 are the road inputs respec-
tively and zu1 , zu2 , zu3 , zu4 are the displacements of wheels respectively. In this
study, system parameters are directly obtained from the previous work [39], which
can be seen in Table 13.13.
Frequency domain analysis and design of nonlinear vehicle suspension systems 389

y
f zu3
x θ
zu4
zr3
zu1 Suspension zr4

zu2
Wheel
zr1
Tyre
zr2

Fig. 13.24 Full-vehicle suspension mathematical model [30]

Table 13.13 System parameters

ms ¼ 2010 kg kt11 ¼ 310 KN=m ks12 ¼ 129:8 kN=m cs21 ¼ 1000 N  s=m
m11 ¼ 29:2 kg kt12 ¼ 310 KN=m ks21 ¼ 129:8 kN=m cs22 ¼ 1000 N  s=m
m12 ¼ 29:2 kg kt21 ¼ 310 KN=m ks22 ¼ 129:8 kN=m
m21 ¼ 29:2 kg kt22 ¼ 310 KN=m cs11 ¼ 1000 N  s=m
m22 ¼ 29:2 kg kt11 ¼ 310 KN=m cs12 ¼ 1000 N  s=m

13.4.2 Simulation study

Several simulation studies are conducted to show the effect of nonlinear suspension
systems on the vehicle vibration performance. By using the method proposed in the
previous two sections, the optimal damping characteristics can be obtained easily.
Then the comparative studies between nonlinear suspensions systems and linear
suspension systems can be conducted. The simulation studies based on three dif-
ferent evaluation standards are proposed and each simulation study is conducted
under different working conditions. The results of the simulation studies are dis-
cussed in detail in this section.

13.4.2.1 Vehicle vibration study

As demonstrated in the previous section, the effect of system nonlinearity on the
vehicle vibration suppression is different with respect to the input frequencies. The
nonlinear suspension system is helpful in the vibration suppression while in other
cases it is not. Several case studies are presented here to evaluate the vibration
performance.
390 Handbook of vehicle suspension control systems

1. v ¼ 1 m/s
JEEP
1000.0
.cm.Translational_Acceleration.Y_linear
.cm.Translational_Acceleration.Y_nonlinear optimal
Acceleration (mm/s2)

500.0

0.0

–500.0

–1000.0
10.0 12.5 15.0 17.5 20.0
Time (s)

Fig. 13.25 Vertical vibration at the speed of 1 m/s

Table 13.14 Vertical acceleration at the

speed of 1 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 240 342
Nonlinear 240 342

2. v ¼ 5 m/s
JEEP
1000.0
.cm.Translational_Acceleration.Y_nonlinear optimal
.cm.Translational_Acceleration.Y_linear
Acceleration (mm/s2)

5000.0

0.0

–5000.0

–1000.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.26 Vertical vibration at the speed of 5 m/s

Table 13.15 Vertical acceleration at the

speed of 5 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 5671 8641
Nonlinear 5463 7203
Frequency domain analysis and design of nonlinear vehicle suspension systems 391

3. v ¼ 5.1 m/s

JEEP
1000.0
.cm.Translational_Acceleration.Y_linear
.cm.Translational_Acceleration.Y_nonlinear
Acceleration (mm/s2)

5000.0

0.0

–5000.0

–1000.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.27 Vertical vibration at the speed of 5.1 m/s

Table 13.16 Vertical acceleration at the

speed of 5.1 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 5554 8463
Nonlinear 5307 7008

4. v ¼ 5.2 m/s

JEEP
1000.0
.cm.Translational_Acceleration.Y_nonlinear
.cm.Translational_Acceleration.Y_linear
Acceleration (mm/s2)

5000.0

0.0

–5000.0

–1000.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.28 Vertical vibration at the speed of 5.2 m/s

Table 13.17 Vertical acceleration at the

speed of 5.2 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 5450 8163
Nonlinear 5243 6827
392 Handbook of vehicle suspension control systems

5. v ¼ 5.5 m/s

JEEP
1000.0
.cm.Translational_Acceleration.Y_linear
.cm.Translational_Acceleration.Y_nonlinear optimal
Acceleration (mm/s2)

5000.0

0.0

–5000.0

–1000.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.29 Vertical vibration at the speed of 5.5 m/s

Table 13.18 Vertical acceleration at the

speed of 5.5 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 5030 7371
Nonlinear 4948 6379

6. v ¼ 6 m/s

JEEP
1000.0
.cm.Translational_Acceleration.Y_nonlinear optimal
.cm.Translational_Acceleration.Y
Acceleration (mm/s2)

5000.0

0.0

–5000.0

–1000.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.30 Vertical vibration at the speed of 6 m/s

Table 13.19 Vertical acceleration at the speed

of 6 m/s

System RMS (mm/s2) Max|a| (mm/s2)

Linear 4558 6660
Nonlinear 4588 6762
Frequency domain analysis and design of nonlinear vehicle suspension systems 393

From Figs. 13.25 to 13.30, it can be noted that when the vehicle velocity is
in a specific range around resonance frequency, the vehicle vibration performance
obtained by the nonlinear optimal damping would be better than the linear sus-
pension system. However, the nonlinear suspension system can bring some bad
effects on the vibration performance when the velocity is very high. In the rela-
tively low velocities, the vibration performance of the nonlinear vehicle suspension
system will not be improved compared with the linear suspension system.

13.4.2.2 Suspension stroke

In the analysis of vehicle suspension systems, the distance of the sprung mass and the
unsprung mass should be controlled. This is mainly due to the fact that it will strike
the vehicle frame when the distance is large. In this study, the nonlinear effects on
the vehicle suspension stroke need to be considered. Some comparative studies
between linear suspension systems and nonlinear suspensions are conducted. As in
the vehicle system, there are four independent suspension systems; the suspension
stroke of the front right suspension system is given as an example. The figures of the
suspension stroke with different velocities can be shown as follows.
1. v ¼ 1 m/s

JEEP
240.0
SUBTRACT(.plot_2.curve_2)_nlinear
SUBTRACT(.plot_2.curve_2)_nonlinear
236.25
Length (mm)

232.5

228.75

225.0
5.0 8.75 12.5 16.25 20.0
Time (s)

Fig. 13.31 Suspension stroke at the velocity of 1 m/s

2. v ¼ 5 m/s

JEEP
350.0 SUBTRACT(.plot_2.curve_2)_nonlinear
SUBTRACT(.plot_2.curve_2)_linear

287.5
Length (mm)

225.0

162.5

100.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Figure 13.32 Suspension stroke at the velocity of 5 m/s

394 Handbook of vehicle suspension control systems

3. v ¼ 10 m/s
JEEP
300.0
SUBTRACT(.plot_2.curve_2)_nonlinear
SUBTRACT(.plot_2.curve_2)_linear
262.5
Length (mm)

225.0

187.5

150.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.33 Suspension stroke at the velocity of 10 m/s

4. v ¼ 15 m/s

JEEP
300.0
SUBTRACT(.plot_2.curve_2)_nonlinear
SUBTRACT(.plot_2.curve_2)_linear
262.5
Length (mm)

225.0

187.5

150.0
5.0 6.25 7.5 8.75 10.0
Time (s)

Fig. 13.34 Suspension stroke at the velocity of 15 m/s

From Figs. 13.31 to 13.34, it can be seen that in the relatively low-velocity
range, the effect of nonlinearity on the suspension stroke is not obvious. However,
in the high-velocity range, such as 15 m/s, the positive effect of system nonlinearity
on the suspension stroke is extraordinarily obvious. Therefore, it can be concluded
that the system nonlinearity can be helpful in the suppression of the relative dis-
placement between the sprung mass and the unsprung mass.

13.4.2.3 Handling ability

In the analysis of the vehicle system, the handling ability is another important issue
to consider. As discussed before, the spring and damping characteristics often reach
a compromise of the ride comfort and handling ability. In this study, the main
evaluation standard is the vehicle vibration, and the effect of the damping non-
linearity on the vibration performance has been discussed in the previous sections.
However, it is also worthwhile to evaluate the nonlinear effect on the handling
ability to make sure the results are more reasonable. Therefore, the simulation study
based on the full-vehicle dynamic model is conducted to evaluate the handling
ability.
Frequency domain analysis and design of nonlinear vehicle suspension systems 395

35.0
SUBTRACT(.plot_1.curve_1)_linear
30.0 SUBTRACT(.plot_1.curve_1)_nonlinear

25.0
Angle (deg)

20.0
15.0
10.0
5.0
0.0
0.0 10.0 20.0 30.0 40.0
Time (s)

Fig. 13.35 Lather slip angle of linear and nonlinear system

The evaluation index of the handling ability is given as

 
1 1
a1  a2 ¼ 57:3L  ð13:46Þ
R0 R i
In the equation above, a1 is the front wheel slip angle, a2 is the rear wheel slip
angle. If the value of Eq. (13.46) is larger than zero, then the system will be under-
steered and it is desired for the vehicle system; if the value of Eq. (13.46) is smaller
than zero, then the system will be over-steered and it will have a bad effect on the
handling ability; if the value of Eq. (13.46) is equal to zero, then the system will be
neutral steered.
The simulation study conducted here is based on the GB/T6323.6-94 Standard.
Two different systems are used to make the comparative study, the first one is the
vehicle suspension system with linear damping and the second one is the vehicle
suspension system with nonlinear damping. The result is shown in Fig. 13.35.
In Fig. 13.35, it is obvious that the slip angle of these two systems is almost the
same. Therefore, it can be concluded that the nonlinear damping characteristics will
not have any effect on the performance of the vehicle handling ability.

13.4.3 Summary
In this section, the full-vehicle dynamic model is built by using the Adams software
to analyze the nonlinear effect on the vehicle performance, such as the ride com-
fort, suspension stroke, and the handling ability. In the vehicle dynamic model,
some realistic parameters are taken into consideration while conducting the simu-
lation study, such as the mass, inertia, road input, and the vehicle velocity, and
therefore the simulation results are more accurate and reasonable. Three different
types of simulation studies are conducted independently to evaluate the vehicle
performance. For the vehicle vibration study, it can be shown that in the resonance
frequency range, the vehicle suspension system with nonlinear damping can
achieve a better ride comfort performance. In relatively low- and high-frequency
ranges, the system nonlinearity cannot induce some positive effects on the vibration
performance. In the analysis of the suspension stroke, the results show that the
system nonlinearity can be very helpful in the suppression of the suspension stroke.
396 Handbook of vehicle suspension control systems

Unlike the vibration performance, in the relatively high-velocity range, the sus-
pension stroke can also be suppressed with the nonlinear damping characteristics.
The final task is the verification of the handling ability. The result in this section
shows that if the damping characteristics are nonlinear, it basically has no effect on
the vehicle handling ability when compared with the linear damping characteristics.
Therefore, it can be concluded that the nonlinear damping characteristics will be
helpful in the vibration suppression in many important practical situations, the
performance of the suspension stroke will be improved and the handling ability will
not be affected by the nonlinearity.

13.5 Conclusion and future work

The nonlinear output frequency response function method is adopted in the analysis
of vehicle suspension systems. By using this method, the relationship between the
nonlinear suspension system output and system parameters is obtained. The results
obtained from the mathematical and simulation study show that the nonlinear
damping characteristics achieved by the nonlinear frequency domain analysis
method can be helpful in the design and analysis of the vehicle suspension system.
From the details provided in section 13.2, it can be shown that the system para-
meters determined by the nonlinear frequency domain analysis method is very
helpful in the suppression of vibration of vehicle systems. In order to show the
nonlinear effect, the comparative studies between the nonlinear system, the linear
system and the recommended linear system are conducted. The results show that
around the resonant frequency, the system with nonlinearities can be very helpful in
the suppression of the system vibration. In section 13.3, two different optimal
nonlinear damping characteristics are presented. The first one is borrowed from
existing results and the second one is obtained by using the nonlinear frequency
domain analysis method. Comparative studies between these two damping char-
acteristics show that compared with the existing nonlinear optimal damping char-
acteristics the nonlinear system determined by the nonlinear frequency domain
analysis method is much better. A simple spring damper dynamic model is also
built in this section. The results of the dynamic model simulation studies also
support the results obtained in Matlab. In section 13.4, the nonlinear damping
characteristics are exerted in the full-vehicle dynamic model. Note that the full
vehicle dynamic model is built based on the Adams software. In this model, the
road input, vehicle velocity and some other effects which might appear in the
real-vehicle working conditions are taken into consideration. The comparative
simulation results of two different vehicle dampers with the nonlinear damping
characteristics and the linear damping characteristics are presented. The results
show that the nonlinear damping characteristics can be very helpful in the suppres-
sion of the vehicle vibration around the resonant frequency. Other vehicle evaluation
standards, such as the suspension stroke, can also be improved with the help of the
nonlinear damping characteristics. The simulation study based on the evaluation of
Frequency domain analysis and design of nonlinear vehicle suspension systems 397

the handling ability shows that the introduced nonlinearity will not have effect on the
handling ability.
Future work will be focused on the following points. First, nonlinear stiffness
needs to be considered in order to make the optimization more realistic. It has been
shown in this chapter that the spring nonlinearity does exist in the real system and
therefore it is of significance to analyze the nonlinear spring effect on the vehicle
performance. Second, the present study focuses on the study of vibration perfor-
mance and the results are obtained based on the sine input signal. However, the
real-vehicle road working conditions vary. They contain, for example, the random
input and pulse input. Therefore, it is also necessary to study the nonlinear per-
formance under these working conditions and to obtain the optimal nonlinear sys-
tem parameters to meet different requirements.

References
[1] Eslaminasab N., ‘Development of a semi-active intelligent suspension sys-
tem for heavy vehicles, PhD thesis in Department of Mechanical and
Mechatronics Engineering’, 2008, Waterloo, Ontario, Canada: University of
Waterloo.
[2] Haiping D., S. Kam, L. James, ‘Semi-active H-infinity control of vehicle
suspension with magneto-rheological dampers’. Journal of Sound and
Vibration, 2005. 283: pp. 981–996.
[3] Chuan L., M. Liang, Y. Wang, Y. Dong, ‘Vibration suppression using
two-terminal flywheel. Part II: application to vehicle passive suspension’.
Journal of Vibration and Control, 2011. doi: 10.1177/1077546311419547.
[4] Howard G., D. Bastow, J.P. Whitehead, eds. Car Suspensions and Handling.
2004, London: Pentech Press.
[5] Rajamani R., ed. Vehicle Dynamics and Control. 2006, New York, NY:
Springer.
[6] Sun L., ‘Optimum design of ‘‘road-friendly’’ vehicle suspension systems
subjected to rough pavement surfaces’. Applied Mathematical Modelling,
2002. 26(5): pp. 635–652.
[7] Wang X., H.W., ed. Vehicle Dynamic, N.D.I. Press. 2008, Beijing.
[8] Yu Z., Vehicle Theory. 2009, China Machine Press, Beijing.
[9] Gandhi F., I. Chopra, ‘A time-domain non-linear viscoelastic damper
model’. Smart Materials & Structures, 1996. 5(5): pp. 517–528.
[10] Kamath G.M., N.M. Wereley, ‘A nonlinear viscoelastic-plastic model for elec-
trorheological fluids’. Smart Materials & Structures, 1997. 6(3): pp. 351–359.
[11] Fang Chang Z.L., Air Suspension Performance Analysis using Nonlinear
Geometrical Paramteres Model. SAE Technical Paper 2007-01-4270, 2007,
doi:10.4271/2007-01-4270, 2007.
[12] Berg M., ‘A nonlinear rubber spring model for vehicle dynamics analysis’.
Vehicle System Dynamics, 1998. 29: pp. 723–728.
398 Handbook of vehicle suspension control systems

[13] Borowiec M., G. Litak, M.I. Friswell, ‘Nonlinear response of an oscillator with
a magneto-rheological damper subjected to external forcing’. Modern Practice
in Stress and Vibration Analysis VI, Proceedings, 2006. 5-6: pp. 277–284.
[14] Jones J.C.P., ‘Automatic computation of polyharmonic balance equations
for non-linear differential systems’. International Journal of Control, 2003.
76(4): pp. 355–365.
[15] Xing X.N., D.M. Zhang, ‘Realization of nonlinear describing function
method virtual experiment system based on LabVIEW’. International Con-
ference of China Communication (Iccc 2010), 2010: pp. 211–213.
[16] Haeri M., M. Attari, M.S. Tavazoei, ‘Analysis of a fractional order Van der
Pol-like oscillator via describing function method’. Nonlinear Dynamics,
2010. 61(1–2): pp. 265–274.
[17] Elliott A.M, Bernstein M.A., Ward H.A., Lane J., Witte R.J., ‘Nonlinear
averaging reconstruction method for phase-cycle SSFP’. Magnetic Reso-
nance Imaging, 2007. 25(3): pp. 359–364.
[18] Stanzhitskii A.N., T.V. Dobrodzii, ‘Study of optimal control problems on
the half-line by the averaging method’. Differential Equations, 2011. 47(2):
pp. 264–277.
[19] Billings S.A., J.C. Peyton Jones, ‘Mapping nonlinear integro-differential
equation into the frequency domain’. International Journal of Control, 1990.
54: pp. 863–879.
[20] Lang Z.Q., S.A. Billings, ‘Output frequency characteristics of nonlinear
systems’. International Journal of Control, 1996. 64(6): pp. 1049–1067.
[21] Jing X.J., Z.Q. Lang, S.A. Billings, ‘Output frequency properties of non-
linear systems’. International Journal of Non-Linear Mechanics, 2010.
45(7): pp. 681–690.
[22] Yue R., S.A. Billings, Z.Q. Lang, ‘An investigation into the characteristics
of non-linear frequency response functions. Part 1: understanding the higher
dimensional frequency spaces’. International Journal of Control, 2005. 78:
pp. 1031–1044.
[23] Lang Z.Q., S.A. Billings, G.R. Tomlinson, R. Yue, ‘Analytical description of
the effects of system nonlinearities on output frequency responses: A case
study’. Journal of Sound and Vibration, 2006. 295: pp. 584–601.
[24] Jing X.J., Z.Q. Lang, S.A. Billings, ‘Output frequency response function-
based analysis for nonlinear Volterra systems’. Mechanical Systems and
Signal Processing, 2008. 22(1): pp. 102–120.
[25] Jing X.J., Z.Q. Lang, S.A. Billings, ‘Determination of the analytical para-
metric relationship for output spectrum of Volterra systems based on its
parametric characteristics’. Journal of Mathematical Analysis and Applica-
tions, 2009. 351(2): pp. 694–706.
[26] Lang Z.Q., Jing X.J., Billings S.A., Tomlinson G.R., Peng, Z.K., ‘Theoretical
study of the effects of nonlinear viscous damping on vibration isolation of
sdof systems’. Journal of Sound and Vibration, 2009. 323(1–2): pp. 352–365.
[27] Jing X.J., Lang Z.Q., Billings S.A., Tomlinson G.R., Peng, Z.K., ‘Frequency
domain analysis for suppression of output vibration from periodic
Frequency domain analysis and design of nonlinear vehicle suspension systems 399

disturbance using nonlinearities’. Journal of Sound and Vibration, 2008.

314(3–5): pp. 536–557.
[28] Eberhard P., U. Piram, D. Bestle, ‘Optimization of damping characteristics
in vehicle dynamics’. Engineering Optimization, 1999. 31(4): pp. 435–455.
[29] Hrovat D., ‘Survey of advanced suspension development and related optimal
control application’. Automatica, 1997. 30(10): pp. 1781–1817.
[30] Wang J., S.W. Shen, ‘Integrated vehicle ride and roll control via active
suspensions’. Vehicle System Dynamics, 2008. 46: pp. 495–508.
[31] Volterra V., ed. Theory of Functionals and of Integral and Integro-differential
Equations. 1959, New York, NY: Dover.
[32] Jing X.J., Z.Q. Lang, S.A. Billings, ‘Mapping from parametric character-
istics to generalized frequency response functions of non-linear systems’.
International Journal of Control, 2008. 81(7): pp. 1071–1088.
[33] Jing X.J, Z.Q. Lang, S.A. Billings, ‘The parametric characteristics of fre-
quency response functions for nonlinear systems’. International Journal of
Control, 2006. 79(12): pp. 1552–1564.
[34] Daniel C. and Wood, F. (1980), Fitting Equations to Data, Revised Edition,
New York: John Wiley and Sons, Inc.
[35] West P.S.H.G., ‘Interpolation methods for curve construction’. Applied
Mathematical Finance, June 2006. 13(2): pp. 89–129.
[36] http://mathworld.wolfram.com/LeastSquaresFitting.html
[37] Lin J.S., I. Kanellakopoulos, ‘Nonlinear design of active suspensions’. IEEE
Control Systems Magazine, 1997. 17(3): pp. 45–59.
[38] http://www.mscsoftware.com/Products/CAE-Tools/MD-Adams.
aspx. 2010.
[39] Li J., ed. Adams shi li jiao cheng. 2002, Beijing, China: Beijing Institute of
Technology.
 Index

acting force actuator

bump response of 297 faults 29
road test response of 299 modeling of 18
active fault-tolerant controllers 27 electromagnetic actuators 19–20
active full-car suspension model MR damper 20–2
204–8 saturation 221
active half-car suspension model modelling 223–4
200–3 saturated control and constrained
active quarter-car suspension model control 224–5
197–9 types 223
Takagi–Sugeno representation uncertain half-car suspension
209–10 systems 125
active suspension 1–2, 70, 196, 287 main results 131–4
control 304 non-fragile H? controller for 125
dynamic output feedback problem formulation 126–31
controller design 152 simulation results 134–41
entire frequency case 156–8 Adams model 385
with finite frequency approach adaptive fuzzy control, see fuzzy control
143 adaptive fuzzy sliding-mode controller
finite frequency case 153–5 (AFSMC) 55
problem formulation 144–8 adaptive methodologies, capabilities
simulation 158 of 42
using neural network 318–23 admissible control gain matrix 158
using PID controller 313–18 amplitude actuator saturations 223
joint design of system 290–3 artificial neural networks (ANNs) 41,
problem formulation 89–90 318–19, 322
simulation results 293–9
state feedback controller design body vertical acceleration 160, 166, 186
148–52, 158–66 bump response of 296
systems 248, 305, 308 road test response of 298
model of system in state-space British Standard – BS 6841 5
309 bump road disturbances 85–6
parameters of 145 for passive and active suspension
synthesis of active suspension systems 86
digital system 313 on car suspension deflections 87
active suspension control system on driver head acceleration 86
(ASCS) 40, 41 on tyre deflections 87
402 Handbook of vehicle suspension control systems

Cadillac models 2 electrohydraulic actuators 16, 70, 196

CAE tool and vehicle suspension force tracking control of 82–3
system control 303 parameter values of 84
active suspension system modeling electromagnetic motor actuators 19
308 electro-magnetic suspension (EMS)
in state-space 309–12 system 25
synthesis of 313 engine-body vibration system 329,
using PID controller 313–23 331, 345
design development process 306–8 engine-induced vibrations,
mechatronic suspension systems, transmission of 327–8
classification of 304–5 engine mounting systems 333
chassis suspension forces 70, 88 equivalent control 52, 55, 258, 264
classification of suspension systems evolutionary computing (EC) 41
305
closed-loop system block diagram 312 fault-tolerant control
complete-car model 14, 15 active 27
control input 26, 69, 197, 243, 252, design 181–4
258, 269, 270, 275, 329 methods 27–9
vector 77 passive 27
control output 54, 75–6, 108, 109, 176, problem formulation 175–81
182 simulation results 185–9
control strategies, for intelligent for uncertain vehicle suspension
suspension design 22–4 systems 173
fault-tolerant control method 27–9 feedback control 329
linear control strategies 24 strategies 24
nonlinear control strategies 24–5 feedforward control 329
time delay control methods 26–7 force transmissibility for different
uncertainty control methods 25–6 systems 370
control system diagram 289 full-car chassis suspension (7 DOF)
conventional controller 27, 174, 260 70, 72
full-car suspension model 204–8
damping force 11, 14, 22, 252, 259, parameter values of 84
272, 373 Takagi–Sugeno representation
degrees of freedom (DOF) motions 6 212–16
design development process, of full-vehicle dynamic model 388
mechatronic systems 306–8 full-vehicle suspension mathematical
digital signal processor (DSP) 25 model 389
driver body model (4 DOF) 70, 72, fuzzy control 50–1
89 GA–NN combined control 62
dynamic vehicle model genetic-based 60–1
application on 387 neuro- 59–60
handling ability 394–5 sliding-mode control (SMC) 52
simulation study 389 chattering, alleviating SMC 53–5
suspension stroke 393–4 neural network (NN) control 56–8
vehicle vibration study 389–93 nonlinear system model 55–6
 Index 403

fuzzy logic (FL) 41, 53 quadratic stabilization of 225–30

fuzzy logic controller (FLC) 50, 55–6, validation of 216–20
100, 293 half-car suspension model 200–3
fuzzy logic system (FLS) 62, 100 active control for 125
Fx-LMS algorithms 329 system parameter values for 135
Takagi–Sugeno representation
generalized frequency response 210–12
functions (GFRFs) 359 half-vehicle model 127
genetic algorithm (GA) 58–9, 287, 293 and LQ control design 46–8
genetic algorithm–neural network suspension model 102
(GA–NN) combined control handling ability 394–5
62 handling performance 39
genetic-based optimal fuzzy control harmonic balance method 359
60–1 heave accelerations 139
graphical topology of chassis 388 for passive and closed loop systems
ground disturbance 137
for front wheel 136 hybrid control technique 22, 313
for rear wheel 136
groundhook control 22 inertia mass 331, 332
input frequency, change in 376–9
H? fuzzy control 195, 230 input magnitude, change in 379–82
actuator saturation 221 intelligent control approaches for
modelling of 223–4 active suspension systems
saturated control and constrained 39–43
control 224–5 adaptive fuzzy control 50–1
types of 223 GA–NN combined control 62
constrained control 230–2 genetic-based optimal fuzzy control
optimization of attraction region 60–1
234–5 neuro-fuzzy system 59–60
quarter-car active suspension system, genetic algorithms (GAs) 58–9
control design for 235–9 half-vehicle model and LQ control
saturated control 233–4 design 46–8
suspension systems model 197 nonlinearity and unmodeling
full-car suspension model 204–8 dynamic description of
half-car suspension model 200–3 48–50
quarter-car suspension model quarter vehicle active suspension
197–9 system modeling and LQ
Takagi–Sugeno (T–S) fuzzy model control design 43–6
208 sliding-mode control (SMC) 52
of active full-car suspension adaptive neural network control
212–16 56–8
of active half-car suspension alleviating chattering 53–5
210–12 nonlinear system model 55–6
of active quarter-car suspension intelligent full-vehicle model 10–11
209–10 intelligent half-car model 8–9
404 Handbook of vehicle suspension control systems

intelligent quarter-car model 6–7 magnetorheological (MR) dampers 3

interval type-2 fuzzy system control scheme using inverse model
(IT2 FS) 100 of 254
IT2 T–S fuzzy control system damping characteristics of 253
107–9 mechanical model 10
Karnik–Mendel algorithm mechanism of 253
100, 101 model-following 255
nonlinear active suspension system simulation results 259–61
101–3 sliding mode controller 257–9
simulation examples 111–13 system model and problems
half-vehicle active suspension 255–7
system 113–20 model of 20–2
numerical examples 111–13 semi-active suspension systems,
stability analysis of 109–11 VSS observer of 272–3
T–S fuzzy model and fuzzy control design 275–8
system 104–5 numerical simulations 279–83
for vehicle active suspension plant 273–4
systems 99 problem formulation 274–5
Wu–Mendel uncertainty bound 100, semi-active suspension systems
101 control with 247, 252–5
ISO 2631 standard 5 simple physical model of 254
sliding mode controller with
Kalman filter 275, 280, 283 describing function method
Kalman–Yakubovich–Popov (KYP) 247, 261
lemma 143 integral sliding mode control
Karnopp’s law 252, 264, 271 263–5
parameter variation 269–72
large damping system and nonlinear passivity constraint 268–9
damping system 384 problem formulation 262–3
linear control strategies 24 redesign of relay input with
linear matrix inequalities (LMIs) 69, describing function method
71, 125, 144, 196, 225, 327 265–6
linear quadratic (LQ) controllers 43 simulation conditions 266–7
half-vehicle model and 46–8 structure of 253
quarter-vehicle model and 43–6 typical characteristics of 257
Linear-Quadratic-Gaussian (LQG) variable orifice damper 250–2
controller 55 mechatronic suspension systems,
linear systems, comparison 369–71 classification of 304–5
with nonlinear output spectrum M-files 311–12
369–71 modeling of vehicle suspension
Lyapunov function 71, 77, 101, 132, system 5
191, 192, 258 actuator modeling 18
Lyapunov–Krasovskii method 330 electromagnetic actuators 19–20
Lyapunov stability theory 125, 141, MR damper 20–2
174 intelligent full-vehicle model 10–11
 Index 405

intelligent half-car model 8–9 dynamic vehicle model 387

intelligent quarter-car model 6–7 application on 387
nonlinear multibody dynamic handling ability 394–5
model 14 simulation study 389
nonlinear uncertainty modeling suspension stroke 393–4
15–16 vehicle vibration study 389–93
nonlinear vehicle dynamic model frequency domain analysis and
11–14 design of 357
with fault 17–18 linear systems, comparison with
with time delay 16–17 369–71
road model 6 optimization and system analysis
multiple-input multiple-output 366
(MIMO) nonlinear model output frequency response function
103 (OFRF) method 360–2
parameter optimization 368–9
neural network (NN) system model 360–2
control systems 56–8 numerical simulations 29, 57, 83–90,
controller training scheme 322 279–83
model creation procedure 321
Nissan Motor 248 one-degree-of-freedom (1-DOF) 288
nonlinear control strategies 24–5 optimal control theory 23–4
nonlinear damping system optimal damping characteristics 372
and large damping system 384 derivation of 374–6
and small damping system and 383 output-feedback control design 344–5
nonlinearity and uncertainty 55–6 output frequency response function
nonlinearity and unmodeling dynamic (OFRF) method 360–2
description 48–50 nonlinear output spectrum 362–4
nonlinear multibody dynamic model computation of 366–7
14–15 linear systems, comparison with
nonlinear output spectrum 362–4, 369 369–71
computation of 366–8 numerical determination of 364–6
numerical determination of 364–6 parameter optimization 368–9
nonlinear uncertainty modeling 15–16
nonlinear vehicle dynamic model parallel distributed compensation
11–14 (PDC) scheme 196
with fault 17–18 particle swarm optimization (PSO)
with time delay 16–17 algorithm 115
nonlinear vehicle suspension systems passive fault-tolerant controllers 27
357 passive suspension systems 70, 195,
comparative studies 372–6 248
dynamic model verification performance output vector 129
385–6 pitch acceleration for passive and
existing nonlinear damping closed loop systems 137
characteristics 372–4 plant faults 29
OFRF-based analysis method 374 polynomial model 21–2
406 Handbook of vehicle suspension control systems

power spectral density (PSD) 115, 160 seat and driver body model, parameter
of body acceleration 164, 168, 169 values of 84
values 44 seat-driver suspension model,
process model-based fault-detection parameters of 72
and diagnosis 28 seat suspension (2 DOF) 70, 72
proportional integral (PI) control forces 88
scheme 24 and driver body 92
pseudobacterial genetic algorithms semi-active suspension system 2, 26,
(PBGAs) 60, 61 70, 195–6, 304–5
with magnetorheological dampers
quarter-car active suspension system 247, 252–5
289 model-following 255
control design for 235–9 simulation results 259–61
quarter-car model 6, 11–12, 144, 175, sliding mode controller 257–9
197–9 system model and problems 255–7
with active suspension 145 sliding mode controller with
and control scheme 256, 262 describing function method
and LQ control design 43–6 247, 261
parameters 158 integral sliding mode control
263–5
random road inputs parameter variation 269–72
front-body accelerations with 117 passivity constraint 268–9
front control force with 118 problem formulation 262–3
rear-body accelerations with 118 redesign of relay input with
rear control force with 119 describing function method
rate actuator saturations 223 265–6
recursive least squares (RLS) simulation conditions 266–7
algorithms 60 variable orifice damper 250–2
relation dynamic tire load 167 VSS observer of 272–3
ride comfort 1, 5, 40, 70, 89, 125 design 275–8
ride index calculation method 119 numerical simulations 279–83
road holding 1, 4, 5 plant 273–4
road model 6 problem formulation 274–5
robust active control of integrated sensor faults 29
suspension system 69 Six-degree-of-freedom vehicle
control system design 75 model 43
force tracking control of skyhook control 22
electrohydraulic actuators skyhook damper 58, 252
82–3 sliding-mode control (SMC) 52
objectives 75–6 adaptive neural network control
robust controller design 77–82 56–8
numerical simulations 83–90 chattering, alleviating 53–5
uncertain integrated system FL controller complementary to
modelling 71–5 SMC for system nonlinearity
root mean square (RMS) acceleration 5 and uncertainty 55–6
 Index 407

sliding mode controller 247, 257–9 nonlinear vehicle dynamic model

with describing function method 11–14
261 of road profile 6
integral sliding mode control validation method 29–31
263–5 state-space model 291, 309–12
parameter variation 269–72 static output feedback controller 77
passivity constraint 268–9 suspension deflection constraint 129
problem formulation 262–3 of front wheel 137
redesign of relay input with of rear wheel 137
describing function method suspension stroke 393–4
265–6 bump response of 297
simulation conditions 266–7 road test response of 298
slowly variable/adaptive suspension
systems 304 Takagi–Sugeno (T–S) fuzzy model 25,
small damping system vs. nonlinear 27, 174, 196, 208
damping system 383 of full-car suspension 212–16
soft computing 41 of half-car suspension 210–12
sprung-mass system 263 quadratic stabilization of 225
state-feedback control design 337–41 attraction region 227–8
state-of-the-art of intelligent convex analysis and linear matrix
suspension system 1 inequalities 225–6
control strategies 22–4 via PDC control 228–9
fault-tolerant control method stability in the sense of Lyapunov
27–9 226–7
linear 24 of quarter-car suspension
nonlinear 24–5 209–10
time delay control methods 26–7 validation of 216–20
uncertainty control methods 25–6 Team Lotus 248
evaluation criterion 4 time delay control methods 26–7
ride comfort 5 time delay of actuators 16
road holding 5 type-2 fuzzy logic system 100
modeling 5 tyre deflection constraints 138, 139,
of actuator 18–22 140
intelligent full-vehicle model
10–11 uncertain integrated system modelling
intelligent half-car model 8–9 71–5
intelligent quarter-car model 6–7 uncertainty control methods 25–6
nonlinear dynamic model with
time delay 16–17 validation method 29–31
nonlinear dynamics model with variable orifice damper 250–2
fault 17–18 variable structure system (VSS)
nonlinear multibody dynamic observer, of semi-active
model 14 suspension systems 250,
nonlinear uncertainty modeling 272–3
15–16 design 275–8
408 Handbook of vehicle suspension control systems

numerical simulations 279–83 vibration control analysis and

plant 273–4 synthesis 327
problem formulation 274–5 main results 336
VDI 2057 5 output-feedback control design
Vehicle chassis suspension 69–70 344–5
vehicle engine-body system 330–4 state-feedback control design
vehicle oscillations 308 337–41
vehicle suspension control system problem formulation 335–6
146 simulation results 346–51
vehicle vibration study 389–93 vehicle engine-body system 330–4