Thesis by Carlos Alberto Vivas Lopez 2016

Instituto Tecnológico y de Estudios Superiores de Monterrey
Campus Monterrey
School of Engineering and Sciences
Global Chassis Control System for Passenger Vehicles

to Improve Safety and Comfort
A dissertation presented by
Carlos Alberto Vivas López
Submitted to the
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
In
Major in Mechatronics and Control
Monterrey, Nuevo León, April 1st 2016

Instituto Tecnológico y de Estudios Superiores de Monterrey
Campus Monterrey
The committee members, hereby, certify that have read the dissertation presented by Carlos Alberto
Vivas López and that it is fully adequate in scope and quality as a partial requirement for the degree
of Doctor of Philosophy in Sciences of Engineering, with a major in Mechatronics and Control.
Dr. Rubén Morales Menéndez

Tecnológico de Monterrey
Principal Advisor
Dr. Olivier Sename

Grenobe INP - GIPSA-lab
Committee Member
Dr. Luc Dugard

Grenobe INP - GIPSA-lab
Committee Member
Dr. Ricardo Ambrosio Ramı́rez Mendoza

Committee Member
Dr. Horacio Ahuett Garza

Committee Member
Dr. Jorge Welti Chanes

Associate Dean of Graduate Studies
ii
Declaration of Authorship
I, Carlos Alberto Vivas López, declare that this dissertation titled Global Chassis Control System
for Passenger Vehicles to Improve Safety and Comfort and the work presented in it are my own. I
confirm that:
• This work was done wholly or mainly while in candidature for a research degree at this
University.
• Where any part of this dissertation has previously been submitted for a degree or any other
qualification at this University or any other institution, this has been clearly stated.
• Where I have consulted the published work of others, this is always clearly attributed.
• Where I have quoted from the work of others, the source is always given. With the exception
of such quotations, this dissertation is entirely my own work.
• I have acknowledged all main sources of help.
• Where the dissertation is based on work done by myself jointly with others, I have made
clear exactly what was done by others and what I have contributed myself.

@2016 by Carlos Alberto Vivas López

All rights reserved
iii
Dedication
A Carmen.
El amor de mi vida. Gracias por estar a mi lado en cada paso, desde las largas
noches hasta las inmensas alegrı́as, tú has sido mi más grande inspiración.
A ti hijo.
Quien con tu bendición llegaste y te convertiste en mi más grande motivación
para culminar esta etapa.
A mis padres.
Conchi y Rafael, quienes me han enseñado el valor de la perseverancia y la
responsabilidad, y me han alentado a lo largo de este camino a ver mas allá.
A mi familia.
Mi hermana Alexandra, mi mamá tere, mis tı́os Claudia, Rosi y Ulises, mis
suegros Norma y Efraı́n, quienes han sido mi soporte todo este tiempo, sin su
incondicional apoyo nada de esto habrı́a sido posible.
iv
Acknowledgements
I would like to thank the Tecnológico de Monterrey for granting me a schol-

arship during my PhD studies, in particular to the chair of product development
for emerging markets and the chair of automotive engineering for their financial
support.
I thank to CONACyT for giving me the support grant along my PhD studies
and for the additional support during my international research stays.
I express my gratitude to my advisor Dr. Rubén Morales, for guiding and sup-
porting me over the years. I appreciate the continuous motivation and knowledge
he transmitted to me in every stage of my research.
Also, I would like to thank my thesis committee members: Prof. Olivier

Sename, Prof. Luc Dugard, Dr. Ricardo Ramı́rez and Dr. Horacio Ahuett, for
all of their guidance through this process; your discussion, ideas, and feedback
have been absolutely invaluable.
My sincere thanks to Jorge Lozoya, Juan Carlos Tudón, Diana Hernández, Alan
Cabello, Alejandro Ortı́z, Gerardo Villarreal and all my colleagues who shared
with me, not only their knowledge but their experiences. Thanks for all your help
and contributions.
Last but not the least, I would like to thank all my friends specially Rodrigo
Vargas, Fernando Irazoqui and Juan Carlos Velazco for their encouragement and
unconditional support through this endeavor.
v
Global Chassis Control System for Passenger
Vehicles to Improve Safety and Comfort
By
Abstract
The automotive vehicle has been experiencing rapid changes during the past few years regard-
ing the onboard control systems. That evolution is driven mainly by customer and governmental
demands of safer, more efficient, and comfortable vehicles. These changes have brought on a new
generation of Vehicle Dynamic Control (VDC) systems to cope with those demands.
The introduction of this wide diversity of VDCs with a unique control goal has arisen two main
issues: first, the vehicle infrastructure is getting more complex with more sensors and electronic
control units for each different VDC. Second, the increasing control goals of those diverse VDC
systems may lead to conflicts in their control actions, causing a deterioration in the overall vehicle
performance.
Based on those issues, the concept of Global Chassis Control (GCC) system emerges as a
suitable solution for the integration of coexisting vehicle subsystems to pursue a common perfor-
mance goal. The GCC systems are guided by two main objectives: 1) to integrate in a coordinated
way subsystems for the sake of a common goal and 2) to give the system the possibility of shar-
ing sensors and actuators information and functions. In the literature many approaches have been
proposed, but most of them failed to cope with those objectives, while some are conceived as func-
tional integration, integrating subsystems just to meet a single function, and others use the same
control framework regardless of the driving situation.
To handle these issues, the dissertation proposes a control system that is capable of coordinating
the actions and modes of operations of three subsystems: Semi-Active suspension system, 4-Wheel
Independent braking system, and Active Front Steering. Such a coordination scheme is handled
by a set of driving situations for which the system then determines when and how to apply a
specific control mode of operation to each subsystem. This coordination is managed according
to a hierarchical group of control objectives: 1) stability (security), 2) handling (security), and 3)
vibration reduction (comfort).
To finalize this research, several conclusions regarding the obtained results are discussed. Also,
the author introduces some interesting future investigations to continue the research in the field of
GCC systems. These points include topics such as: evaluation requirements and possible issues in
the physical implementation of the proposed system in a real vehicle, as well as the study of this
type of GCC system over a communication protocol like a Controller Area Network (CAN), etc.
vi
Contents
1 Introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Problem Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 State-of-the-Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.8 Proposal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.9 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.10 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.11 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Global Chassis Control System Using Suspension, Steering, and Braking Subsystems 17
2.1 Summary of the Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Full Article . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Further evaluation and justification of the GCC system . . . . . . . . . . . . . . . 37
2.3.1 Decision layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.2 Control layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3.3 Evaluation conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3 Method for Modeling Electrorheological Dampers Using Its Dynamic Characteristics 55

3.2 Full Article . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4 Influence of MR Damper Modeling on Vehicle Dynamics 73

4.2 Full Article . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
vii
5 Conclusions 89
5.1 Classification of driving situations on road vehicles . . . . . . . . . . . . . . . . . 90
5.2 Coordination of interacting subsystems . . . . . . . . . . . . . . . . . . . . . . . . 91
5.3 Global Chassis Control systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.5 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.6 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Bibliography 97
A Acronyms and Variables Descriptions 103
Curriculim Vitaes 109
viii
Chapter 1
Introduction
1.1 Introduction
Currently, road vehicles are equipped with a wide variety of on-board Vehicle Control Systems
(VCS). Each VCS handles different objectives and coordinating this information in the vehicle
becomes a difficult task. To overcome this issue, VCS are grouped in different categories regarding
their functionality: the VCS that refer to the control of the dynamical behavior of the car are
called Vehicle Dynamics Controls (VDC), e.g. Anti-lock Braking System (ABS), Electronic Stability
Control (ESC), etc.
Those VDC have a wide variety of control goals, three being the most important:
1. Security: When a vehicle has crossed its handling limits (i.e. when the tires have lost their
grip), the vehicle becomes unstable and difficult to control by the action of the driver only.
The main objective is to recover the driver’s control for safety reasons.
2. Handling: It refers to how well a vehicle can go around corners and react to cornering at
higher speeds. This condition is linked to the security characteristic of the car.
3. Vibration mitigation: It refers to the comfort that the passengers experience during riding.
This consists in the limitation of the unpleasant vibrations and movements caused by road
irregularities.
The vehicle dynamics theory defines various effects in the automobile that can be classified
in several different directions: Vertical (along the Z axis and around the X and Y axes), Lateral
(along the Y axis and around the Z axis) and Longitudinal (along the X axis). Depending on what
direction we are focusing on, some variables become relevant, see Fig. 1.1:
• Vertical dynamics refers to the movements that affect mainly the comfort of the passengers,
the important variables being: pitch (turn around the Y axis, ϕ), roll (turn around the X axis,
θ), and vertical acceleration (z̈).
1
2 CHAPTER 1. INTRODUCTION
Z
ψ (yaw)
φ (pitch)
Y
β υ
ω,λ X
θ (roll)
Figure 1.1: Vehicle reference system and important dynamical variables.
• Lateral dynamics refers to the stability and handling (safety). The important variables are:
lateral displacement (y), side slip angle (β), and yaw (turn around Z axis, ψ).
• Longitudinal dynamics also refers to stability (safety). The important variables are: longitu-
dinal velocity (υ), wheel rotational velocity (ω), and tire slip ratio (λ).
To control these variables, three vehicle subsystems have been mainly considered: steering,
braking, and suspension, see Fig. 1.2. Each subsystem demands a VDC.
a) Steering b) Brakes c) Suspension
Figure 1.2: The most studied vehicles subsystems
• Steering (Fig. 1.2a). This subsystem modifies the lateral dynamics. It can be controlled by
modifying the driver steering angle command, e.g. Active Front Steering (AFS).
• Braking (Fig. 1.2b). It acts on the longitudinal dynamics. A controller can brake or release
each tire individually, e.g. ABS or ESC.
• Suspension (Fig. 1.2c). This subsystem is able to modify the vertical behavior of the vehicle.
This is usually done by using a controllable damper (active or semi-active) and changing its
damping coefficient, e.g. Continuous Damping Control (CDC).
Higher performances for road vehicles need the collaboration of several vehicle subsystems.
Normally a VDC is designed to operate for a specific goal, but when these subsystems are si-
multaneously working, the result could be a degradation of the general performance due to their
interactions.
1.1. INTRODUCTION 3
A concept called Global Chassis Control (GCC), also known as Integrated Vehicle Dynamics
Control (IVDC), is a suitable solution to manage the interactions among multiple VDC simultane-
ously operating in a vehicle. The GCC strategy proposes the integration of different vehicle control
systems in order to meet a common goal, rather than to put them together and let them follow their
own independent goals. This offers the possibility of sharing sensors information and actuators
functions, [He, 2005].
To achieve such integration a GCC system needs to interact with the existing controllers in a
structured way. Over time, many control structures have been detected in the literature; according
to [Yu et al., 2008] three topologies are the most important, see Fig. 1.3:
Supervisory Controller
Centralized Controller
System 1 System 2 System 3
Controller Controller Controller
Sensors
Controller Controller Controller
Sensors
Sensors
Actuator Actuator Actuator System 1 System 2 System 3
Actuator Actuator Actuator System 1 System 2 System 3
Actuator Actuator Actuator
Vehicle Vehicle Vehicle
(a) Centralized (b) Decentralized (c) Multi-layer
Figure 1.3: Control system topologies.
Centralized: The structure in Fig. 1.3a, consists of a single controller that generates all control
signals for the subsystems’ actuators. This kind of structure increases the computational load
in the central ECU demanding more powerful units. Also these structures are rigid, which
limits its reconfiguration capabilities for integrating new elements into the control systems.
If there is a new sub-system for the control loop the only solution is to redesign the full
control system.
Decentralized In Fig. 1.3b, every subsystem has its own independent controller and control goals
for any particular actuator. The interaction among different control loops is limited to shared
information obtained from a communication bus. This type of control architecture was used
in the early chassis control integration. In this scheme the integration lies on the OEM’s,
while the supplier provides its systems with interconnection options. An example of decen-
tralized control architecture is a vehicle equipped with ABS and SAS, in which each control
system is seeking its own goal ignoring the action of the other system.
Multi-Layer In this scheme three layers can be identified: the coordination layer, the controller
layer, and the actuation layer, Fig. 1.3c. According to [Gordon et al., 2003] the coordina-
tion layer has two main functions: the first is based on the dynamic state of the vehicle it
determines as desired set-points for the controller layer and the second is to set a specific
mode of operation into the actuators. The second layer receives the control signals from the
coordination one and it selects and controls its own control subsystem in the mode indicated
by the upper layer. For instance, depending on the overall vehicle situation, the suspension
system can operate in comfort mode or in road-holding mode.
Each one presents its benefits and limitations to achieve the GCC goal. The Multi-layer is the
more suitable option for this purpose, since the supervisory layer can coordinate the actions of the
different subsystems.
In literature, the concept of GCC is usually presented from two points of view:
• The integration of different subsystems for a single objective, e.g. using suspension and
braking to avoid rollover.
• The application of different VDC to the same vehicle subsystem with different objectives,
e.g. ABS and ESP controllers using the braking system, but one prevents the tires from
locking during braking and the other produces a stabilizing momentum by braking the tires.
These points of views can be combined to create a global control strategy to achieve multiple
control objectives, improving each subsystem’s functionality.
The GCC is a control system that integrates different vehicle subsystems and VDC in a coordi-
nated way to improve vehicle performance and functionality.
1.2 Motivation
Recent studies in transportation global trends aim towards achieving “intelligent” vehicles in the
next 10 years, [TechCast, 2012]. Nonetheless, these vehicles are required to meet the upcoming
state regulations of more efficient, ecological, secure, and comfortable vehicles, [WHO, 2013]. In
terms of road safety, the World Health Organization (WHO) estimates that around 1.24 million
deaths are related to road incidents, which is approximately 2.2 % of the global mortality. Specif-
ically in America, 42 % of road related deaths are car occupants and 23 % are pedestrians. These
numbers indicate that preserving the integrity of vehicle occupants must be a major concern for ve-
hicle manufacturers and researchers. These challenges require more sophisticated control systems
to allow the vehicle to increase its security capabilities and performance.
Due to these challenges, many research works have been developed in the field of VCS, es-
pecially for dynamics control (VDC). Most of these VDC are designed to perform depending on
the driving situation to fulfill a specific goal. This single-purpose orientation could cause unex-
pected performances in other dynamic planes, [Sato et al., 1992], e.g. the suspension system can
be oriented towards comfort or road-holding (but not both at the same time), but if the suspen-
sion is oriented towards comfort, the vehicle loses tire grip capabilities which then makes it more
1.3. PROBLEM DESCRIPTION 5
susceptible to skidding or to a roll over condition. This situation is worsened when more driving
conditions and subsystems are considered into the control scheme, each VDC pursuing its own
goal. Moreover, when more than one VDC acts in the same actuator, [El-Sheikh, 2011], the situa-
tion may be even worse: for instance, the ABS tends to release the brake pressure from a tire that
has been locked, but at the same time the ESC tries to brake the wheel to generate a stabilizing
moment.
All of these different control objectives need to be coordinated in an efficient and cooperative
way, i.e., depending on the driving situation some performances need to be degraded in order to
achieve a better overall performance. For example, in the case of emergency braking, the suspen-
sion goal of giving comfort to the passengers must be overwritten in order to cooperate with the
braking system to reduce the braking distance by changing its goal to improve the tire-road contact,
[Valasek et al., 2004].
To achieve such efficient coordination schemes while coping with the current objectives of
more comfortable and safer vehicles, the implementation of a Global Chassis Control (GCC) sys-
tem rises as a feasible solution. Such a GCC system should be able to be implemented in real time,
by deciding online the proper coordination scheme of the available subsystems according to the
classified driving situation. Also, the GCC system mainly prioritizes its control goals as: (1) stabil-
ity, (2) handling, and (3) comfort. This control system should use classical vehicle instrumentation
and actuators.
1.3 Problem Description

Since the addition of the first VDC to road vehicles in the mid 80’s, [Shibahata, 2005], the trend, to
face more dynamics performance objectives, has been added to more independent control systems.
The coexistence of different VDCs will eventually lead to conflicts among their control goals due
to functions overlapping. In addition, having different dedicated infrastructures for each VDC
increases the complexity of the vehicle. Considering these issues, the main problems to be solved
by the GCC system, [Yu et al., 2008], are:
1. To avoid possible conflicts or interventions among different subsystems’ controllers and

functions.
2. To gain potential advantages of each subsystem by communicating and coordinating their

actions and signals, since different subsystems have different action domains.
If these problems are not solved, they can lead to: increased production and maintenance costs,
requirements for more complex control systems, more powerful vehicle electronics to handle the
augmented amount of required information, etc.
Regarding the reported solutions, most of the GCC strategies in literature have at least one of
the following areas of opportunity:
â The integration should be functional, i.e. it is done to meet a specific objective. For example:
objectives such as reducing braking distance, preventing roll over, etc, which are essential to
reach higher dynamical performance.
â The GCC strategy does not cover the three dynamic planes (vertical, longitudinal, and lat-
eral).
â The GCC strategy treats all of the different driving conditions within the same framework,
i.e. regardless the driving condition, the controller operates based on a set of parame-
ters/indices (maneuverability, [Cho et al., 2012]) or regions (phase β − β̇ plane, [He, 2005]).
â The control topology is centralized, i.e the controller design is computationally heavy and
difficult to implement.
Based on the previous issues and areas of opportunity, the GCC system should provide an
efficient coordination of the available subsystems to provide better performance. This coordination
must be driven by a hierarchical structure of control objectives, such as: 1) Stability (security), 2)
Handling (security), and 3) Vibration reduction (comfort).
1.4 Objectives
This research is focused on the development of a method to generate a GCC system able to manage
different critical situations that may occur during normal vehicle driving. To achieve such a goal,
the main objective of this research is:
• To develop a GCC system for full vehicle dynamics (vertical, lateral and longitudinal), co-
ordinated by the current driving situation.
To support the main objective, the following specific objectives are considered:
1. To design a flexible structure that allows the coordination of the subsystems individually to
enhance the overall system performance.
2. To classify different driving situations based on a minimal set of vehicle measurements.
3. To develop a coordination strategy that selects the control modes of the subsystems for the
current driving situation.
1.5. STATE-OF-THE-ART 7
1.5 State-of-the-Art
Since the 80’s, vehicle manufacturers have been developing a wide variety of VCS to improve
the performance of road vehicles. These control systems have evolved, from secondary control
systems helping the drivers, to intelligent systems capable of managing some specific situations
without any human intervention, [Shibahata, 2005].
These VCS gained such an importance in the automotive manufacturing industry at that time
that several ideas of globally controlled vehicles developed. The ideal of Global Chassis Control
can be defined as a fully controlled vehicle coordinating all of its systems to achieve the best
possible performance in every aspect. The best example of this Global Control is the Project Trilby,
initiated by GM in the early 80’s [Schilke et al., 1988]. This project proposes the integration of
existing and new VCS on the vehicle with the intention of controlling everything that happens in
the vehicle, from simples tasks like headlights position to more complicated ones such as motor
efficiency. The project was so ambitious that it was limited by the technology of that time, but
many innovations and developments arose from it, e.g. the concept of integrated brake and steer
subsystems using a three layer control to estimate, coordinate, and control the vehicle subsystems.
Later, other car manufacturers started to develop control systems with less ambitious scopes. These
control systems integrate multiple vehicle subsystems to achieve a common goal; for instance,
Nissan with an integrated 4WS and a differential slip control system, using a decentralized control
topology, [Imaseki and Kobari, 1990], and Toyota with its integrated control system in which they
integrate an active suspension, 4WS and ABS to achieve total performance of vehicle dynamics,
[Yokoya et al., 1990].
Since those days, recent advances in computational technologies and electronics have been de-
veloping more powerful Electronic Control Units (ECU), allowing the design and implementation
of more complex VCS. These new developments open the possibility of the integration of more
subsystems.
From recent works reported in literature, a division among different GCC integration ap-
proaches can be defined as:
• Integration of multiple vehicle subsystems to improve only one direction (lateral, longitudi-
nal, or vertical) of dynamical performance, of Table 1.1 which presents the most representa-
tive works in literature.
• Control in more than one direction, of Table 1.2 which presents the most representative
works for this approach.
• Fully directional vehicle dynamics control, of Table 1.3 which presents the most important
works on GCC.
Table 1.1: Single purpose integration approaches

Author Subsystems Control structure Objective Drawbacks
[Valasek et al., Suspension, Decentralized Reduction of the braking distance No coordination among the sub-
2004] braking systems is considered.
[Wei et al., 2006] Steering, braking Multilayer topology with DYC using fuzzy logic Only focused on lateral dynamics.
coordination based on the
Yaw moment
[Gáspár et al., Suspension, Multilayer topology with Prevent a rollover during corner- The integration is highly dedi-
2007a], [Gáspár braking coordination driven by a ing cated to roll over.
et al., 2007b] weighting function related
to the rollover
[Canale and Fa- Steering, Braking Centralized topology Stability control through DYC A single control algorithm needs
giano, 2008] to generate the commands for
the subsystems demanding more
computational effort.
[Hwang et al., Steering, Braking Multilayer and centralized Lateral dynamics control using Needs a precise model of the sys-
2008] topology comparison AFS and ESC coordination tem and is highly specialized for a
single purpose control objective.
[Poussot-Vassal Steering, Braking Multilayer topology with Stability control with AFS and Is intended for lateral dynamics
et al., 2008], LPV gain scheduling EMB control but only reacts to the yaw
[Poussot-Vassal angle and not other situations.
et al., 2011b]
[Lee et al., 2009] Suspension, Multilayer topology Lateral stability control with SH The control strategy just consid-
Braking driven by the Yaw rate and ESC ers a SH controller for the suspen-
deviation sion instead of more suitable tech-
niques for road holding.
[Tjonnas and Jo- Steering, Braking Multilayer with coordina- Yaw Stability control using a dy- The optimization used to allocate
hansen, 2010] tion based on the yaw rate namic allocation scheme the control actions is based on
the ability to properly estimate the
tire-road friction.
[Yi and Yi, 2011], Steering, Brak- Multilayer topology with Improve maneuverability through It considers the integrated control
[Yim et al., 2012] ing, Anti-roll bar, Weighted Least Square to Active Roll Control System as a support for the anti-roll con-
4WD distribute the desire yaw troller, also uses 4WD to generate
moment the desired yaw moment.
[Di Cairano et al., Steering, Braking Centralized control based Yaw stability control Here the MPC schemes are based
2013] on hibrid-MPC and on the correct estimation of tire
switched-MPC slip angles but this system is very
sensible to variations in the plant.
[Doumiati et al., Steering, Braking Centralized controller Stability and control of lateral dy- The centralized structure restricts
2013] based on LPV frame- namics using AFS and EMB the flexibility in the system and
work to determine the fixes the performance of the con-
subsystem involved trol system.
[Aripin et al., Steering, Braking Centralize control Stability through AFS + DYC The proposed control scheme has
2014] Slide Mode Control with a nonlin- not been proven.
ear surface using Composite Non-
linear Feedback (CNF) Algorithm
This research focuses only on the approaches that fulfill the GCC concept, i.e. those which
control the vehicle dynamics in the three dynamic planes: vertical, lateral, and longitudinal, as
shown in Table 1.3.
It can be concluded that the multilayer topology is the best option. This topology allows divid-
ing the computational effort into different control levels, which represents an advantage in terms of
1.5. STATE-OF-THE-ART 9
Table 1.2: Multi purpose integration approaches

[Chou and Suspension, Decentralized control with Reduce vertical motions and It uses Active suspension and
D’andréa-Novel, Braking interaction maintain lateral stability. when maintaining lateral stability
2005] it uses torque distribution.
[Andreasson and Steering, Braking Centralized controller Inverse model dynamic control al- It relies on a precise model of the
Bünte, 2006] based on inverse model location for lateral and longitudi- system for the inverse dynamics
dynamics nal dynamics and based on that, it calculates
the control inputs. It is designed
just for one purpose and operation
condition.
[Mokhiamar and Steering, Braking Multilayer topology Horizontal dynamics control Requires 4WS and does not con-
Abe, 2006] driven by the side slip based on 4WS and independent sider other operation conditions.
angle braking to generate optimum
force distribution of the tires to
maintain stability and handling
[Falcone et al., Steering, Braking Centralized topology Horizontal dynamic control based It focuses on following the de-
2007] on MPC to maintain a reference sired yaw moment bu, does t not
yaw rate consider different operation con-
ditions.
[Miura et al., Driveline, Brak- Multilayer topology Horizontal dynamics control cal- It considers only the action of
2008] ing driven by the Yaw moment culating a desired yaw moment braking and acceleration to gener-
tracking error and coordinating each wheel as a ate a yaw moment but its just fo-
control input whereas it is braking cused on stability.
or accelerating.
[Xiao et al., Suspension, Multilayer topology Control of lateral and vertical dy- It uses Active suspension and re-
2009] Braking namics using Active suspension lies on the ability of the suspen-
and ESC focusing on stability sion to generate a desired moment
which may not be achieved at cer-
tain conditions.
[Yoon et al., Steering, Braking Multilayer topology Maintain lateral stability and pre- The algorithm does not consider a
2010], [Cho et driven by a Rollover Index vent rollover categorizing 3 op- condition of normal riding.
al., 2012] and the side slip angle eration conditions: maneuver-
ability, stability and imminent
rollover
[Poussot-Vassal Suspension, Multilayer topology with a Yaw control using braking system The monitor scheduling param-
et al., 2011a] Braking supervisory monitor to al- and vertical dynamics control us- eter is the longitudinal slip ra-
locate control actions ing active suspension tio which is not representative for
vertical dynamics.
[Ali et al., 2013] Steering, Braking Multilayer topology Prevention of control loss in an The strategy selects among three
accident avoidance framework states and reacts according to it,
but it relies on highway infras-
tructure.
processing demand, compared with centralized or decentralized topologies, [Yu et al., 2008]. The
multilayer topology offers the possibility to divide the different control tasks into multiple ECUs
while ensuring communication and information sharing among all of them [[Gordon et al., 2003;
Chen et al., 2011]].
In terms of the coordination strategies, evidence suggests that a Case-Structure [[Poussot-
Vassal, 2008; Poussot-Vassal et al., 2011a; Lu et al., 2011; Fergani, 2014]] offers good coor-
dination capabilities when multiple driving conditions are considered. In particular, when each
Table 1.3: Full dynamics integration approaches

[Lu et al., 2010], Steering, Suspen- Multilayer topology Full vehicle dynamics control The strategy does not consider the
[Lu et al., 2011] sion, Braking based on different operation states transient behavior of the system
between mode of operation tran-
sitions.
[Poussot-Vassal, Suspension, Multilayer topology based Full dynamics vehicle control The author treats the problem as
2008] Steering, Braking on LPV gain scheduling based in LP V /H∞ two separate problems: vertical
and horizontal, each controlled by
different methodologies.
[Fergani et al., Suspension, Multilayer topology Full dynamics control. It features It does not consider different op-
2012], [Fergani, Steering, Braking a monitor that decides when and eration conditions besides normal
2014], [Fergani et how to use each subsystem riding and critical (to maintain
al., 2015] stability).
driving condition has a specific mode of operation for each subsystem that is desired; e.g. the
suspension system can have a comfort mode of operation or a road-holding oriented one, and it
is the responsibility of the coordination scheme to select the more suitable one for each driving
condition.
1.6 Hypothesis
Based on the presented state-of-the-art review, the following hypotheses are proposed:
i. A GCC system that manages the vehicle subsystems according to the driving situations can
improve the performance of the vehicle.
ii. Different driving situations can be classified using a reduced set of vehicle measurements.
iii. Using steering, braking, and suspension subsystems, a control strategy covers the three dy-
namic planes efficiently.
iv. For a specific driving condition, there is a proper combination of operating modes and actu-
ators that improves the performance of the vehicle.
v. The overall dynamic performance of the vehicle can be improved by coordinating the oper-
ation modes of the available subsystems using a single decision system.
1.7 Scope
The expected outcome is a Global Chassis Control system that efficiently coordinates the involved
subsystems based on a classification of the driving situation.
1.8. PROPOSAL 11
The scope of this work is to test the proposed strategy only to a simulation level, in order to
demonstrate the effectiveness of a coordination strategy driven by the classified driving situations.
1.8 Proposal
To overcome the aforementioned areas of opportunity and to cope with the research objectives, the
next requirements for a GCC system are proposed:
• A multi-layer Global Chassis Control system which uses the current driving situations of the
vehicle to adapt the mode of operation of the subsystems’ controllers. This proposed control
system structure, Fig. 1.4, is described as follows:
1. Decision layer: It identifies the current driving situation and decides how to coordinate
the subsystems’ actions and their operating mode.
1.a Classification algorithm: It takes the information from the sensors and classifies
the current driving condition using a clustering technique.
1.b Decision logic: The information from the current driving condition is used to de-
cide when to apply the best mode of operation for each individual actuator con-
troller. For the coordination, each situation requires a specific mode of operation
of the controllers in the control allocation layer.
2. Control layer: It receives the control goal from the decision layer, and determines the
proper orientation for each of the local controllers.
2.a Control allocation: The coordination scheme is received and adapts its control
outputs according to the selected mode of operation of the decision logic module.
Such systems are based on two approaches: 1) Heuristic data driven and 2) Fuzzy
Logic control frameworks.
2.b Local controllers: These are the single loop controllers in charge of guaranteeing
that the actuators follow the control allocation layer commands.
3. Physical layer: It involves the actuators and sensors from the vehicle. It receives the
control output from the control layer and feedbacks the vehicle measurements.
3.a Subsystems: The selected actuation subsystems are 4-Wheel Independent Braking
and Active Steering systems for the longitudinal and lateral dynamics, and Semi-
Active Suspension for the vertical dynamic.
3.b Sensors: These are the available sensors and the communication infrastructure of
the vehicle, which measure the important vehicle variables and vehicle status, and
also share it with the rest of the GCC system.
Decision
a Clasification Algorithm
Layer
Driving Situation
b Decision Logic
Control Objective
Information Bus
2 a Control Allocation
Control
Control Command
Layer
(Outer Loop)
Local Controllers
b
FCS ABS SCS
Control Command
(Inner Loop)
3 a Sub-Systems
Physical
Layer
Sensors
b
Figure 1.4: Proposal of the GCC system.
• A discrimination algorithm is used to classify the driving situation based on the measure-
ments of the system. This algorithm is based on k-NN method, simple enough to be im-
plemented in real-time and must use data processing tools (PCA and contribution plots) to
reduce the amount of used variables’ measurements.
• A coordination scheme that modifies the operation modes of the subsystems’ controllers to
obtain the best performance of the subsystems’ interactions. The proposed scheme considers
the subsystems’ capabilities to decide when to use or to ignore the controllers commands for
a specific subsystem.
1.9 Methodology
The methods and procedures used to achieve the research objectives are the following:
Classifier design
6 Driving situations: Since the goal of the GCC system is to improve the performance of
the vehicle during different driving situations, it is necessary to select the situations that
are likely to appear. The classification algorithm must distinguish among the following
situations: 1) ride, 2) road irregularity, 3) acceleration/braking movement, 4) emergency
braking, 5) cornering, 6) rapid steering, and 7) loss of vehicle control.
1.9. METHODOLOGY 13
6 Discrimination Algorithm: Using the vehicle dynamics theory, a set of required vehicle
measurements is established to characterize the behavior of the vehicle in the different driv-
ing situations. Since a high number of vehicle measurements are needed, a dimension re-
duction is also necessary to facilitate the classification task by avoiding the effects of the
curse of dimensionality, [Beyer et al., 1999]. For this purpose, two data processing tools are
used. First, the initial measurement set is analyzed by the Principal Components Analysis
(PCA). Even when the PCA manages to reduce the amount of needed variables, these new
variables are linear combinations of the initial set. This issue is coped with using the second
data processing tool, the contribution plots, [Isermann, 2006], which is used to extract only
the important variables.
6 Classification algorithm: One of the main parts of this work is the ability of the GCC sys-
tem to classify the driving situations. For this purpose, a clustering technique is exploited
as a classifier. Because of its easy computation, fast clustering response, and good perfor-
mance, the k-Nearest Neighbor (k-NN) algorithm is considered. The k-NN approach is one
of the simplest machine learning algorithms and is based on the minimum distance criterion
where the Euclidean distance metric is used. The performance of the classifier is evaluated
with the Receiver Operating Characteristic (ROC) curve.
Control design
6 Vehicle subsystems: To improve the vehicle performance in the three dynamic planes, three
subsystems are used. The considered subsystems are: 1) braking, 2) steering, and 3) suspen-
sion. Each of these subsystems is locally controlled with a VDC under the coordination of a
supervisory level regarding the driving condition.
6 Vehicle dynamics controls: The design of a new VDC system approach was not within the
scope of this work, as the used VDC systems were selected from literature. Like VDCs, some
low-complexity control approaches (e.g. Heuristic, Fuzzy-Logic) have proven their effective-
ness in improving the dynamic performance of road vehicles, with the advantages of being
computationally cheap and easily implemented when compared with other approaches (e.g.
model-based) and also maintaining similar performances. For vertical dynamics, Heuristic
On/Off techniques (Sky-Hook and Ground-Hook, [Poussot-Vassal et al., 2012]) are consid-
ered; for horizontal dynamics, Fuzzy Logic control techniques are considered; [Boada et al.,
2005] for braking and [Krishna et al., 2014] for steering the steering system.
6 Control allocation: The control logic uses a case-based structure [Lu et al., 2011], to decide
among multiple states of operation and the appropriate control objectives to be used. This
structure includes all of the conditions required for each operation state to be set in the local
controllers. During transition between control strategies for each driving situation, the sta-
bility for the system must be guaranteed. According to [Tavasoli et al., 2012], using smooth
transitions between controllers ensures the stability of the system; these transitions must
be designed linearly and sufficiently slow to dissipate the transient effects of the switching,
[Liberzon, 2003].
Assessment of the proposal
6 Validation tools: The validity of the proposal was proven using simulation software tools.
The vehicle model was created using CarSim R , where a sedan D type was considered as
the evaluation vehicle. The control systems were programmed in Matlab/Simulink R . Both
softwares were coupled in a Software-in-the-Loop (SiL) configuration, with CarSim as the
slave system.
6 Comparison cases: The vehicle performance was compared in 3 different configurations:

1) the passive case (without any VDC system), 2) the GCC case, in which the proposed GCC
system is included, and 3) the decentralized configuration case, where the vehicle includes
the same VDCs as the GCC case but without coordination.
6 Evaluation tests: The assessment of the proposed GCC system was carried out using a
series of standard tests: 1) Double line change, 2) Fishhook, 3) Brake distance test, 4) Split-
µ braking test, and 5) Road profile with a bump.
6 Performance indices: The performance of the proposed GCC system was compared in the
time domain using the RMS index of the important variables for each test. The reported per-
formance is presented as a percentage of improvement with the passive case as a reference.
1.10 Contributions
A GCC system comes with the following contributions:
I. Classification of vehicle driving situations. A new approach to classify the experienced

driving situation by a road vehicle has been proposed with validation in a professional simu-
lation software. Specific evaluation tools for classification systems prove the accuracy of the
method in this domain.
- Specific contribution: A discrimination method for the classification of road vehicle driving
situations.
II. Coordination of control systems. The coordination of several VDC simultaneously inter-
acting has been tackled, in which the coordinated system considers the specific actuation
1.11. OUTLINE 15
capabilities of the involved subsystems and decides the proper mode of operation. It also
decides when to change the mode of operation.
- Specific contribution: A control coordination scheme that uses the current driving situation
to select the proper mode of operation for each subsystem.
1.11 Outline
This thesis is structured as follows:
Chapter 2 corresponds to the publication [Vivas-Lopez et al., 2015], as it introduces the proposed
GCC system. This system is based on a multi-layer architecture with three levels: decision,
control and physical levels. This system incorporates a data-based classification and a coor-
dination algorithm in a single control structure. Based on a clustering technique, the GCC
system identifies the actual driving situation, and based on that situation, it coordinates the
actions and modes of operation of the vehicle subsystems (steering, braking and suspension).
This GCC system focuses its coordination scheme to maintain the control objectives of 1)
stability, 2) handling, and 3) vibration reduction, in a hierarchical order.
Chapter 3 presents the article [Vivas-Lopez et al., 2014], presenting a damper modeling method
based on the dynamic characteristics of the semi-active damper. This method involves two
steps: characterization and model customization. These steps are based on experimental
data of the damper behavior obtained through a specific design of experiments. This method
has the advantage of not being dependent on physical information from the damper; only
experimental data is required.
Chapter 4 corresponds to the publication [Lozoya-Santos et al., 2013], which is devoted to the
evaluation of the implications and effects of the modeling of the semi-active damper on
vehicle dynamics using a control system. In this chapter, a model-free control system is
used to guarantee that the only effect is the one that is caused by the modeling of the damper.
This evaluation is done using CarSIM and two damper models.
Chapter 5 presents the conclusions derived from this research, highlights the contributions ob-
tained and introduces some future works and research lines as a continuation of this thesis.
Chapter 2
Global Chassis Control System Using

Suspension, Steering, and Braking
Subsystems
2.1 Summary of the Chapter

This chapter presents in section 2.2 a research journal paper in which a novel Global Chassis Con-
trol (GCC) system is proposed. The proposed GCC system is based on a multi-layer architecture
with three levels: top: decision layer, middle: control layer, and bottom: physical layer. The main
contribution of this work is the development of a data-based classification and coordination algo-
rithm into a single control problem. Based on a clustering technique, the decision layer classifies
the current driving condition. Afterwards, heuristic rules are used to coordinate the performance
of the considered vehicle subsystems (suspension, steering, and braking) using local controllers
hosted in the control layer. The control allocation system uses fuzzy logic controllers.
The performance of the proposed GCC system was evaluated using different standard tests.
Simulation results illustrate the effectiveness of the proposed system compared to an uncontrolled
vehicle and a vehicle with a non-coordinated control. The proposed system decreases the braking
distance by 14 % in the hard braking test with respect to the uncontrolled vehicle, the roll and yaw
movements are reduced by 10 % and 12 % respectively in the double line change test, and the
oscillations caused by load transfer are reduced by 7 % in a cornering situation.
In section 2.3, a further description and evaluation of the proposed GCC system is carried out.
In this section, the evaluation was especially focused to the decision layer.
17
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2015, Article ID 263424, 18 pages
http://dx.doi.org/10.1155/2015/263424
Research Article
Global Chassis Control System Using Suspension,
Steering, and Braking Subsystems
Carlos A. Vivas-Lopez,1 Juan C. Tudon-Martinez,2

Diana Hernandez-Alcantara,1 and Ruben Morales-Menendez1
1
Tecnológico de Monterrey, School of Engineering and Sciences, Avenida E. Garza Sada No. 2501, 64849 Monterrey, NL, Mexico
2
Universidad de Monterrey, Avenida Ignacio Morones Prieto No. 4500, 66238 San Pedro Garza Garcı́a, NL, Mexico
Correspondence should be addressed to Carlos A. Vivas-Lopez; a00794204@itesm.mx
Received 4 August 2015; Accepted 21 October 2015
Academic Editor: Xinggang Yan
Copyright © 2015 Carlos A. Vivas-Lopez et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
A novel Global Chassis Control (GCC) system based on a multilayer architecture with three levels: top: decision layer, middle: control
layer, and bottom: system layer is presented. The main contribution of this work is the development of a data-based classification
and coordination algorithm, into a single control problem. Based on a clustering technique, the decision layer classifies the
current driving condition. Afterwards, heuristic rules are used to coordinate the performance of the considered vehicle subsystems
(suspension, steering, and braking) using local controllers hosted in the control layer. The control allocation system uses fuzzy
logic controllers. The performance of the proposed GCC system was evaluated under different standard tests. Simulation results
illustrate the effectiveness of the proposed system compared to an uncontrolled vehicle and a vehicle with a noncoordinated control.
The proposed system decreases by 14% the braking distance in the hard braking test with respect to the uncontrolled vehicle, the
roll and yaw movements are reduced by 10% and 12%, respectively, in the Double Line Change test, and the oscillations caused by
load transfer are reduced by 7% in a cornering situation.
1. Introduction Stability Control (ESC) which creates a turning moment

using the brakes to prevent loss of control. Both control
A road vehicle has a conjunction of interconnected subsys- systems have conflicting principles, the ABS releases the
tems, such as brakes, steering, suspension, engine, and tires, brake pressure whereas the ESC generates an additional one.
which is difficult to be controlled [1]. These subsystems inter- Table 1 summarizes the used acronyms in this document.
act among them modifying their individual behavior and Studies in global trends aim towards achieving intelligent
consequently the overall performance of the vehicle; these vehicles in upcoming years [3]. These vehicles are required to
interactions could cause negative effects. For example, the meet present and future state regulations related to efficiency,
way the suspension system is tuned can dramatically affect autonomy, ecology, safety, and comfort [4]. Particularly, the
the performance of the steering system; a soft suspension global automotive industry is paying special attention to
affects the tire grip decreasing the ability of the vehicle to safety systems that guarantee the integrity of occupants,
steer; if the suspension is too stiff, tires will hop causing loss of pedestrians, and/or other drivers when an accident occurs.
control. Additionally, all vehicle subsystems have a nonlinear Even the newest safety features (Collision Mitigation or Line
behavior, making their coordination a complex task. Keeping Systems) are oriented to prevent possible dangerous
Normally, each vehicle subsystem has an independent situations; the main topic of research refers to the systems that
control system to accomplish a specific objective. For exam- react against a dangerous situation [5].
ple, the brake system has the Antilock Braking System (ABS) These opportunities demand more of the actual Vehicle
to prevent tires from locking during hard braking avoiding Control Systems (VCS). The standard solutions was to inde-
skidding and loss of control [2]. Also, it has the Electronic pendently treat any new objective by adding a new VCS.
2 Mathematical Problems in Engineering
Table 1: Acronyms definition. This integration offers (1) the ability to simultaneously control
various subsystems, (2) the ability to coordinate the actions
Acronyms Description of individual subsystems regarding a general goal [8], and (3)
ABS Antilock Braking System the ability to share information from sensors and actuators
AS Active Steering (also functions) [9].
AFS Active Front Steering This concept has been investigated in the literature.
CDC Continuous Damping Controller The approaches can be differentiated regarding the number
DC Decentralized Controllers of Degrees of Freedom (DoF) in which they act (vertical,
DL Decision Logic longitudinal, or lateral dynamics). Some cases have been
DLC Double Line Change studied as functional integrations (specific objective or single
DoF Degree of Freedom
DoF). In [10], an integration of the braking and Semiactive
(SA) suspension system is proposed to decrease braking
DBC Data-Based Controller
distance in an emergency maneuver; but, in this case, there
EMB Electromechanical Braking is no coordination. An integration of active suspension and
ESC Electronic Stability Control braking is proposed in [11] to prevent a rollover situation
ECU Electronic Control Unit using a Linear Parameter-Varying (LPV) controller driven by
FH Fish Hook a scheduling parameter related to load transfer; the controller
FCS Force control system is highly model dependent.
FL Fuzzy logic Most of the research of this topic lays in the case of
FIS Fuzzy Inference System multipurpose integration (i.e., multiple DoF). An integration
GCC Global Chassis Control of steering and braking to enhance horizontal dynamics
IVDC Integrated Vehicle Dynamics Control using robust control is proposed in [12]; however, the coor-
dination algorithm and controller synthesis are not clear.
𝑘-NN 𝑘-Nearest Neighbor
In [13] the authors proposed a controller based on inverse-
LPV Linear parameter varying
model dynamics to generate the desired control commands
MBC Model-Based Controller for braking and steering subsystems; the method is highly
MF Membership Function dependent in an accurate model of the system and sensitive
PC Principal Component to modeled dynamics. In [14] an integration of steering and
PCA Principal Component Analysis braking systems is proposed. The goal is to maintain an
RMS Root Mean Square optimum tire forces using Sliding Mode Control (SMC). It
ROC Receiver operating characteristic uses an optimization procedure to compute the optimum
SA Semiactive force distribution among the tires. In [15] a control strategy
SAS Semiactive suspension to improve the horizontal dynamics involving the braking
SAP Suspension Adjustment Plane
and steering subsystems is proposed; it uses an optimization
procedure to allocate the desired yaw moment; but, the cost
SCS Steering control system
functions have to be modified depending on the situation.
SISO Single-Input Single-Output In [16] a loss prevention control system is introduced using
SMC Sliding Mode Control the braking and steering subsystems; this strategy relies on
TRC Traction Control physical infrastructure in the road to calculate the reference
UC Uncontrolled signals, which nowadays is not available in most roads.
VCS Vehicle Control System Only few works have been published on full dynamics
VDC Vehicle Dynamics Control integration (able to act in the vertical, longitudinal, and lateral
4WIB Four-Wheel Independent Braking dynamics). In [17], an integration of an active suspension, a
4WS Four-Wheel Steering 4-Wheel Steering (4WS), and a driving/braking force control
using Traction Control (TRC) and ABS is proposed, but the
coordination algorithm is not clear and the use of active
suspension and 4WS is expensive. In [1], a strategy using
This parallel architecture can lead to some drawbacks, for the differential braking for horizontal dynamics and the
example, a VCS is normally designed to seek a specific goal, active suspension for vertical is proposed; this method has
but when it interacts with other VCS the result could degrade to calculate the control command of each sampling time by
the global performance, overruling the original objectives of optimization; this demands a lot of computational resources.
the individual controllers due to inherent coupling effects. The authors in [18] divide the problem; they treated vertical
Since those controllers have to work simultaneously, they and horizontal dynamics separately. For vertical dynamics, it
have their own information system and Electronic Control determines the desired force for each damper based on the
Unit (ECU) demanding more costs and space; that is, vehicle LPV framework; but, its performance is fixed by design. For
infrastructure and complexity increase [6]. horizontal dynamics, it uses a gain scheduling parameter to
The concept of GCC, also called Integrated Vehicle decide when to apply braking control or a combination of
Dynamics Control (IVDC), proposes the coordinated integra- steering with braking control. It defines some driving con-
tion of those different VCS to pursuit a common goal [7]. ditions (i.e., stable or unstable) but they are limited. The work
Mathematical Problems in Engineering 3
of [19] proposed a control architecture capable of identifying Table 2: Modeling: variables description.
the actual driving situation; based on the estimated situation,
the control mode of each subsystem is changed. Even when Variable Description Units
this method is capable of reacting against different situations, 𝑎1 , 𝑎2 Hysteresis coefficients related to s/m, 1/m
it does not consider the switching implications in the control displacement and velocity
system, which could lead to unexpected behaviors. In [20], a 𝑐𝑝 Viscous damping coefficient Ns/m
control system based on LPV is proposed; in this case all three 𝑓brake , 𝑓steer Cut-off frequency of the actuator Hz
directions are simultaneously controlled using two varying dynamics
parameters, one for suspension-steering and another for 𝑓𝑐 Force coefficient due to N/V
braking; however, the controller operates under predefined manipulation
conditions without depending on the driving situation. 𝐹𝐷 Damper force N
Although there are interesting results, they do not include 𝐹SA Semiactive damper force N
the full dynamics integration; others are strongly model- 𝑘𝑝 Stiffness coefficient N/m
dependent or they are robust to some uncertainty which
𝑙 Vehicle wheel base m
could generate conservatism issues. Furthermore, some of
them are not easy to implement in on-board Vehicle Dynam- 𝑚𝑠 , 𝑚us Sprung/unsprung mass kg
ics Controllers. 𝑇𝑏+ , 𝛿+ Actuator output MPa, deg.
The main contribution of this proposal lays in a new 𝑇𝑏∗ , 𝛿∗ Actuator controller output MPa, deg.
GCC system based on a discrimination of the operation 𝑉𝑥 , V𝑥𝑖,𝑗 Vehicle/tire longitudinal vel m/s
conditions of the vehicle. The strategy architecture is a Longitudinal/lateral/vertical
supervisory decentralized control to improve flexibility and 𝑥, 𝑦, 𝑧 m
displacement
modularity [21]. The goal is to identify the current driving
𝑥,̇ 𝑦,̇ 𝑧̇ Longitudinal/lateral/vertical m/s
situation, based on vehicle measurements using clustering velocity
methods [22]; then, the control modes of each subsystem are
𝑥,̈ 𝑦,̈ 𝑧̈ Longitudinal/lateral/vertical m/s2
coordinated to ensure the best global performance. The coor- acceleration
dination includes three subsystems: SAS, AFS, and 4-Wheel
𝑧def , 𝑧̇ def Damper deflection and velocity m, m/s
Independent Braking (4WIB), a full dynamics integration. The
effectiveness of the new strategy was validated in CarSim. 𝑧𝑟 , 𝑧𝑠 , 𝑧us Road/sprung mass/unsprung m
mass vertical position
The paper is organized as follows. Section 2 describes
the problem. Section 3 presents the GCC architecture and 𝛽 Vehicle side slip angle deg.
modules in detail. Section 4 discusses the results, based on 𝛿 Steering wheel angle deg.
a case study. Section 5 concludes the research and proposes 𝛿driver Driver’s steering command deg.
future work. Tables 2 and 3 summarize all the used variables 𝜆 Tire slip ratio —
in this work. 𝜐 Damper manipulation V
𝜙, 𝜃, 𝜓 Pitch/roll/yaw angle deg.
2. Problem Description 𝜙,̇ 𝜃,̇ 𝜓̇ Pitch/roll/yaw rate deg./s
Passenger vehicles are equipped with a wide diversity of on- 𝜓̇ 𝑑 Desired yaw rate deg./s
board VCS, and those focused on the management of the 𝜔 Rotational speed of the tires rad/s
dynamical behavior of the vehicle are called Vehicle Dynamics
Controllers (VDC), for example, Continuous Damping Con-
troller (CDC) and Electronic Stability Control (ESC). Usually A VDC can have one or more of these three control goals:
each VDC operates in a set of DoF defined by the vehicle stability, handling, or vibration mitigation:
reference system, Figure 1; the most important variables for (1) Stability: when a vehicle has crossed its handling
each vehicle motion are as follows: limits and tires lost their grip, it is said that the vehicle
(i) Vertical dynamics refers to the movements that affect has become unstable and cannot be controlled by the
mainly the comfort of the passengers (vibration action of the driver only; here, the main objective is
reductions); the important variables are pitch (turn to recover the driver’s control to guarantee passengers
around 𝑦-axis, 𝜙), roll (turn around 𝑥-axis, 𝜃), and safety.
vertical acceleration (𝑧)̈ in the Chassis and wheels. (2) Handling: it refers to how well a vehicle can achieve
(ii) Lateral dynamics refers mainly to the stability and cornering at high speeds; this condition is related to
handling (safety) of the vehicle; the important vari- the safety characteristic of the vehicle.
ables are lateral displacement (𝑦), side slip angle (𝛽), (3) Vibration mitigation: it refers to the comfort that the
and yaw (turn around 𝑧-axis, 𝜓). passengers will experience during riding; this consists
(iii) Longitudinal dynamics refers to vehicle stability in the limitation of the unpleasant vibrations and
(safety) and its performance (power train); the impor- movements caused by road irregularities; also, road-
tant variables are longitudinal velocity (𝑉𝑥 ), wheel holding is considered, that is, vibration reduction in
rotational velocity (𝜔), and tire slip ratio (𝜆). the unsprung mass at high vehicle velocities.
Table 3: Algorithms: variables description. Table 3: Continued.
Variable Description Variable Description

𝑎𝑚 Actuation vector 𝜆 crit Critical tire slip ratio
Gain of the allocation 𝜎𝑥𝑖,𝑘 Variance of the 𝑖th variable in a 𝑘
𝑎𝑢
coordination driving situation
Suspension/braking/steering R𝑘 Covariance matrix
𝑎𝑢susp , 𝑎𝑢braking , 𝑎𝑢steer
coordination gain X𝑘 Data matrix of 𝑛 variables
Softest/hardest damping
𝑐min , 𝑐max ̂ 𝑘, P
P ̃𝑘 Principal/residuals components
coefficient matrix
𝐶𝑖 Driving class T ̃𝑘
̂ 𝑘, T Principal/residual scores matrix
Contribution index of the 𝑖th Principal/residual transformed
Con𝑖,𝑘 X ̃𝑘
̂𝑘 , X
variable in a 𝑘 driving situation variables
Minimal set of important 𝑖 Variable number
CS𝑘
variables for a driving situation
𝑘 Driving situation
𝑑(𝑥, 𝑦𝑗 ) Euclidian distance
𝑗 Number of neighborhoods
𝐷𝐶 Driving situation
𝑒(𝛽), 𝑒(𝜓)̇ Slip angle/yaw rate error
Z
𝐺ABS ABS braking gain
IS Initial set of vehicle variables ψ
𝑘𝑖 Number of nearest neighbors
𝑙𝑘 First 𝑙 principal components
𝑚 Number of measurements
Minimal set of important 𝜙
MS
variables for all driving situations
𝑀𝑧 Corrective yaw moment Y
𝑛, 𝑛 ∗ Original/reduced number of β υ
variables
𝑟 Number of driving conditions
Driving situation critical ω, λ X
𝑠𝑠 θ
condition
𝑠𝑖 Driving situation importance Figure 1: Vehicle reference system and important dynamical vari-
𝑡crit Situation changing waiting time ables.
𝑡con Variable contribution threshold
𝑇driver Driver braking torque
𝑇ESC𝑟 , 𝑇ESC𝑙
Right/left corrective braking To develop a full dynamics integration in the vehicle it
pressure is necessary to control/coordinate the different subsystems
𝑇𝐺 Pressure gain simultaneously [17]. Usually the selected subsystems are (1)
𝑡𝑛𝑦 Noise variance threshold Active Front Steering, (2) Independent Braking, and (3)
Allocated/coordinated controller SAS. Another requirement is the use of different controllers
𝑢𝑐 , 𝑢𝑐∗ acting in those subsystems depending on the current driving
output
Full suspension situation [23]; but, this feature leads to switching those
𝑈|rh , 𝑈|conf road-holding/comfort controllers, which could cause unstability [24].
suspension command In road vehicles, the inherent coupling effects among
∗ Single corner/full suspension their subsystems impact the overall performance of the
𝑢susp𝑖,𝑗 , 𝑈susp
control command vehicle. The coupled effects plus the issues that come with
𝑥 Data point the interactions of different independent control systems
Passive/controlled 𝑖th increase the complexity of controlling the vehicle dynamics.
𝑋𝑖passive , 𝑋𝑖controlled The above conditions can be handled with integrated
performance variable
Residual data point of the 𝑖th control strategies, where subsystems and control actions are
̃ 𝑖,𝑘
𝑥 efficiently coordinated. Such strategies avoid contradictions
measure in a 𝑘 driving situation
Desired vehicle slip angle/yaw in the control goals of the subsystems to obtain a better global
𝛽𝑑 , 𝜓̇ 𝑑 performance from them at each situation [25]. Model-based
rate
𝛿AFS Corrective steering angle control approaches represent a complex and unpractical
solution, considering the vehicle as a highly nonlinear system.
∗ ∗ Steering wheel/wheels
𝛿steering wheel , 𝛿wheels On the other hand, Data-Based Controllers could represent
directional angle
a reliable option in practice, assuming that the vehicle
0.8 0.8
MBC
Vertical acceleration (G’s)
DBC DBC
0.4
Pitch angle (deg.)

0.4
0 0
−0.4 −0.4
Passive
MBC Passive
−0.8 −0.8
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Time (s) Time (s)
(a) Vertical acceleration (𝑧̈ 𝑠 ) (b) Pitch angle (𝜙)
Figure 2: Comparison of MBC and DBC approaches.
dynamics is monitored by some sensors. Figure 2 compares

the two approaches under the road profile with bump test. This Classification algorithm
Decision
layer
test is used to evaluate the road isolation characteristics of Driving situation
the control systems. Here 3 cases are compared: (1) Passive: Decision logic
a vehicle with nominal shock absorbers, (2) a Model-Based
Controller (MBC): a full suspension 𝐻∞ controller designed Control objective
to minimize the vertical acceleration (𝑧̈ 𝑠 ) and the pitch
Information bus
Control allocation
(𝜙) movements of the sprung mass, and (3) a Data-Based
Control command
Controller (DBC): a controller that considers the classical Sky-
Control
(outer loop)
layer
Hook and Ground-Hook algorithms.

Local controllers
From Figure 2 it can be seen that the two approaches
have a very similar performance. Quantitatively the MBC FCS ABS SCS
improved the vertical acceleration in 14.5% and the pitch
Control command
movements in 9.74% with respect to the Passive case, whereas (inner loop)
the DBC achieved an improvement of 9.1% and 17.8%,
respectively. Both approaches result in similar performances, Subsystems
but the DBC is simpler and faster to implement, compared
to the MBC which has large dimension matrices inside the
Physical
Sensors
layer
controller.
..
Based on the results of the literature review, VDC systems .
that achieve full dynamics integration are still a research
topic.
3. Global Chassis Control Figure 3: GCC architecture, hierarchical approach.
The architecture of the GCC system is divided into three main

layers, Figure 3:
a modular scheme capability [26]. The information bus con-
(1) Decision layer: it identifies the current driving situa- tains sensors measurements and actuators functions. Since
tion and its stability; it decides how to coordinate the the methods to observe or estimate the considered variables
subsystems actions and their operating mode. are not within the scope of this work, they are assumed to be
(2) Control layer: it receives the control goal from the available. Figure 4 shows a full scheme of the proposed GCC
decision layer and determines the proper orientation system.
for each of the local controllers.
(3) Physical layer: it comprehends the actuators and 3.1. Decision Layer. This layer has two main tasks: (1) classify
sensors from the vehicle; it receives the control output the current driving situation and (2) coordinate the control
from the control layer and sends the process variables strategy for the classified situation.
from the sensors.
This multilayer and hierarchical architecture has some 3.1.1. Classification Algorithm. The first step of the algorithm
advantages: (1) it divides the computational load into several is to decide, from an Initial Set (IS) of variables, which is the
ECUs, (2) it allows information sharing, (3) it improves Minimal Set (MS) to classify the current driving situation.
system flexibility (reconfigurability), and (4) it introduces The elements of IS must be variables that describe the global
Decision layer
SAP
Control layer
Control allocation
aususp Local controllers
k-means
.. algorithm
.
Decision Voltage control Physical layer
logic
Susp ⌉ cmin ⌉ ⌉ 10% ⌉
󳨀→ 𝜐 = Actuators
control ⌋ cmax ⌋ ⌋ 90% ⌋
austeer SA suspension
AFS Angle control
Sensors
∗ 28.74 ∗ Active front
𝛿steering wheel = 𝛿
1.18 tires steering
𝛿driver
aubraking Braking system
ESC
Brake control
∗
Tb∗ = GABS · max(Tdriver , TESC )
Tdriver
Sensors/actuators signals
Figure 4: Control scheme of the GCC system.
behavior of the vehicle, as well as variables that are commonly situation. Then, the scores matrices T ̃𝑘 ∈
̂ 𝑘 ∈ R𝑚×𝑙𝑘 and T
𝑚×(𝑛−𝑙𝑘 )
used in VCS: R are obtained:
MS ⊆ IS : |MS| ≤ |IS| . (1) T ̂ 𝑘,

̂ 𝑘 = X𝑘 P
(2)
̃ 𝑘 = X𝑘 P
T ̃ 𝑘.
Some vehicle variables are correlated; for example, pitch
is correlated to deflections of the suspension. Besides, not all
variables have the same importance in all driving situations; ̂𝑘 + X
By decomposing X𝑘 = X ̃𝑘 with
for example, in a riding situation the vertical acceleration will
gain more importance, but during cornering yaw moment
could be the most important. To determine the most rep- ̂𝑘 = T
X ̂ 𝑇,
̂ 𝑘P
𝑘
resentative vehicle variables during a driving situation, a (3)
Principal Component Analysis (PCA) is carried out. ̃𝑘 = T
X ̃ 𝑇,
̃ 𝑘P
k
Measurements of the variables in the IS set under different
driving situations are used in this step, where 𝑟 is the the modeled data is obtained with principal and residual
number of driving situations to be considered, X𝑘 , with 𝑘 = components, respectively.
{1, 2, . . . , 𝑟}, is the data matrix of 𝑛 variables, and |IS| = 𝑛, According to [28], using the residual components, X ̃𝑘 ,
and 𝑚 measurements, with zero mean and unit variance after the contribution of each variable 𝑖 can be obtained in each
a scaling process. To neglect the noise effect in the scaling 𝑘 driving situation as
process, variables with small variance are not considered;
𝜎𝑥𝑖,𝑘 < 𝑡𝑛𝑦 , where 𝑡𝑛𝑦 > 0 is a defined threshold.
∑𝑚 ̃ 2𝑖,𝑘 (𝑗)
𝑗=1 𝑥
Based on [27], P ̂ 𝑘 ∈ R𝑛×𝑙𝑘 is obtained; it contains the Con𝑖,𝑘 = ∀𝑖 = {1, 2, . . . , 𝑛} . (4)
first 𝑙𝑘 eigenvectors from the covariance matrix R𝑘 and P ̃𝑘 ∈ ∑𝑛𝑖=1 ∑𝑚 ̃ 2𝑖,𝑘 (𝑗)
𝑗=1 𝑥
𝑛×(𝑛−𝑙𝑘 )
R which contains the last 𝑙𝑘 − 𝑛 eigenvectors. The first
𝑙𝑘 eigenvectors associated with the PC have to explain at least Once the contributions are obtained, if Con𝑖,𝑘 > 𝑡con ,
90% of the total variance of the original data for each driving then Con𝑖,𝑘 󳨃→ CS𝑘 , where CS𝑘 is the set of the variables
DC𝑖 [si ] < DC𝑗 [si ]
··· DC𝑖 DCj ···
DC𝑖 [si ] < DC𝑗 [si ] DC𝑖 [si ] < DC𝑗 [si ]
⌈ ⌉⌈ ⌉
DC𝑖 [si ] < DC𝑗 [si ] ∧ ∧
⌈ [ ⌉ [[ [
∧ ∨[
[DC𝑖 [ss ] ≤ DC𝑗 [ss ] =
[ [ D [s ] ≤ D [s ] =
1 [ [ C𝑖 s C𝑗 s 1[
[
[ [[ [
⌊D [s ] = D [s ] = 0 ⌋ [ [[ [
C𝑖 s C𝑗 s ∧ ∧
⌊ ⌋⌊ ⌋
t > tcrit t < tcrit
Figure 5: Graph for the DL module.
that contributes in each 𝑘 driving situation and 𝑡con > 0 is First, the structure for a driving situation, 𝐷𝐶, must be
a defined threshold. Finally, the MS set is formed by defined. For this purpose, the DL module depends on the
𝑟 number of subsystems to use and the amount of driving
MS = ⋃ CS𝑘 ∀𝑘 = {1, 2, . . . , 𝑟} , (5) situations, 𝑟, to consider. The definition of a 𝐷𝐶 is as follows:
𝑘=1 (i) Each driving situation is defined with a vector in R3
with the resultant number variables equal to |MS| = 𝑛 . ∗ in the form: 𝐷𝐶 := [𝑠𝑠 , 𝑠𝑖 , 𝑎𝑚 ].
After defining the MS set that contains the most represen- (ii) 𝑠𝑠 ∈ 0, 1 refers to normal (𝑠𝑠 = 0) or critical (𝑠𝑠 = 1)
tative variables for all studied driving situations, a clustering situation.
technique is exploited as classifier. Because of its easy com-
(iii) 𝑠𝑖 ∈ 1, . . . , 𝑞 indicates the critical level of the driving
putation, fast clustering response, and good performance, the
situation: 𝑠𝑖 → 1 is a stable situation with negligible
𝑘-Nearest Neighbor (𝑘-NN) algorithm was used.
danger while 𝑠𝑖 → 𝑞 corresponds to the most
The 𝑘-NN approach is one of the simplest machine
dangerous condition.
learning algorithms, which it is based on the minimum
distance criterion. A data set of reference patterns in a (iv) 𝑎𝑚 ∈ R𝑢 is the actuation vector for the 𝑢 subsystems
multidimensional feature space is required to establish the that must be accomplished according to the detected
learning of this classifier, where the Euclidian distance is driving situation; that is, 𝑎𝑚 = {𝑎1 , . . . , 𝑎𝑢 }. Consid-
the most common metric. In this case, MS set contains the ering an on-off control law for each subsystem, the
reference patterns of the studied driving situations. control output is dichotomic 𝑎𝑖 ∈ [0, 1].
Given a new data vector 𝑥1 , 𝑥2 , . . . , 𝑥𝑛 of 𝑛 variables, the Then, a set of rules that manage the transition between
𝑘-NN algorithm assigns each new observation into a class of driving situations is defined. Figure 5 shows the graph dia-
driving situation 𝐶𝑖 according to gram with the transition rules; the DL rules are carried as
𝑘 follows:
1 𝑖
𝑥 ∈ 𝐶𝑖 when min ∑𝑑 (𝑥, 𝑦𝑗 ) , (6) (1) A 𝐷𝐶 with bigger 𝑠𝑖 overrules a 𝐷𝐶 with a smaller 𝑠𝑖
𝐶 𝑘𝑖 𝑗=1
index, and 𝑎𝑚 is updated.
where 𝑘𝑖 is the number of nearest neighbors used in the (2) At any safe driving situation, independently of the 𝑠𝑖
classification associated with the driving situation 𝑖 and index, the current 𝐷𝐶 updates the actuation vector 𝑎𝑚 .
𝑑(𝑥, 𝑦𝑗 ) is the Euclidian distance between a new observation
(3) In an unsafe driving situation, the current 𝐷𝐶 keeps
𝑥 and the nearest neighbors of reference defined by 𝑦. A large
the actuation vector in safe mode until the 𝑠𝑖 index
value of 𝑘 reduces the effect of noise in the classification;
decreases through the time (𝑡 > 𝑡crit ) to ensure
cross-validation can be used to define this parameter. For the
completely the vehicle safety and avoid any false
vehicle, the sensor measurements contained in the MS set are
alarms.
used to construct the features space. The online result given
by this clustering technique is sent to the Decision Logic (DL) Finally, the DL layer sends the mode of operation of the
module to classify the situation. subsystems controllers in the control layer.
3.1.2. Decision Logic Module. Based on a set of heuristic rules, 3.2. Control Layer. This layer comprehends the control
the DL module is in charge to decide (1) whether a driving actions (allocation, manipulation) to be taken through the
situation is critical or normal, (2) if the control actions should physical layer. The desired control allocation is determined
change, and (3) when to change from one situation to another based on the information received from the previous layer,
based on the importance of the previous situation. and then the desired control manipulations are computed.
3.2.1. Control Allocation. Based on the driving situation, the Table 4: Linguistic terms.
desired control action for each subsystem is defined in this
NB Negative big
sublayer. The control mode acts as a gain in the form of
NMH Negative medium high
𝑢𝑐∗ = 𝑎𝑢 ⋅ 𝑢𝑐 , (7) NM Negative medium
NMS Negative medium small
where 𝑢𝑐∗ is the controller output which depends directly on NS Negative small
the gain of the coordinated allocation 𝑎𝑢 ; that is, 𝑎𝑢 = 1 means Z Zero
that the GCC demands the actuation on the subsystem, PS Positive small
and vice versa (i.e., when 𝑎𝑢 = 0). Thus, 𝑢𝑐 is the local
PMS Positive medium small
controller output of the subsystem obtained from any control
PM Positive medium
law, but before the full dynamics integration. The allocation
controllers for each subsystems are defined as follows. PMH Positive medium high
PB Positive big
(i) SAS. The SA dampers must always receive a manipulation;
the decision is whether to select a manipulation oriented to Table 5: FL inference rules for the braking system.
comfort (𝑈|comf ) or to road-holding (𝑈|rh ) for each corner of
the vehicle: 𝑒(𝜓)̇
𝑒(𝛽)
NB NS Z PS PB
∗
𝑈susp = (1 − 𝑎𝑢susp ) ⋅ 𝑈|comf + 𝑎𝑢susp ⋅ 𝑈|rh , (8) NB PB PB NS NB NB
NS PB PM NS NM NB
∗
where 𝑈susp = 𝑢susp𝐹,𝐿 , 𝑢susp𝐹,𝑅 , 𝑢susp𝑅,𝐿 , 𝑢susp𝑅,𝑅 . Note that, Z PM PS Z NS NM
according to the driving situation, the GCC can orient the PS PB PM PS NM NB
suspension to comfort or to road-holding using the weighting PB PB PS PS NS NB
parameter 𝑎𝑢susp . At each corner, the SAS controller output can
be oriented to comfort (𝑢𝑖,𝑗 |comf ) or to road-holding (𝑢𝑖,𝑗 |rh )
inspired in the classical Sky-Hook and Ground-Hook control the rules for the proposed FL controller. This FL controller
strategies: uses a Mamdani Fuzzy Inference System (FIS).
To allocate the desired output for the braking local
󵄨 {𝑐min if − 𝑧̇ 𝑢𝑠 ⋅ 𝑧̇ def ≤ 0 controllers, 𝑀𝑧 is transformed in terms of ESCs as
𝑢𝑖,𝑗 󵄨󵄨󵄨󵄨rh = {
𝑐 if − 𝑧̇ 𝑢𝑠 ⋅ 𝑧̇ def > 0 𝑀𝑧 > 0 󳨀→ Brake rear left wheel:
{ max
(9)
if 𝑧̇ 𝑠 ⋅ 𝑧̇ def ≤ 0 𝑇ESC𝑟 = 0, 𝑇ESC𝑙 = 𝑇𝐺 ⋅ 𝑀𝑧 ;
󵄨 {𝑐min
𝑢𝑖,𝑗 󵄨󵄨󵄨󵄨comf = {
𝑐 if 𝑧̇ 𝑠 ⋅ 𝑧̇ def > 0, 𝑀𝑧 = 0 󳨀→ No added braking:
{ max (11)
𝑇ESC𝑟 = 0, 𝑇ESC𝑙 = 0;
where 𝑐min = 0 and 𝑐max = 1 represent the softest and hardest
damping coefficient, respectively. 𝑀𝑧 < 0 󳨀→ Brake rear right wheel:
(ii) Braking System. The braking action is computed by the 𝑇ESC𝑟 = −𝑇𝐺 ⋅ 𝑀𝑧 , 𝑇ESC𝑙 = 0,
fuzzy logic (FL) controller [29]. It uses two inputs: (1) side slip
angle error (𝑒(𝛽) = 𝛽 − 𝛽𝑑 ∈ [−10, 10]) and (2) yaw rate error where 𝑇𝐺 is a parameter that relates the corrective yaw
(𝑒(𝜓)̇ = 𝜓̇ − 𝜓̇ 𝑑 ∈ [−10, 10]) and one output: (1) corrective moment (𝑀𝑧 ) and the brake pressure to be applied by the
yaw moment (𝑀𝑧 ∈ [−1, 1]). The control goal is to reduce braking system.
the errors to zero; for this purpose the reference signals are Additionally, the coordinated allocation affects the action
defined as 𝛽𝑑 = 0, since the goal is to have 𝛽 as close to zero of the braking FL controller, including or ignoring it, using
as possible and the value of 𝑎𝑢braking as a gain:
𝑉𝑥 ∗
𝑇ESC = 𝑎𝑢braking ⋅ 𝑇ESC .
𝜓̇ 𝑑 = 𝛿 , (10) (12)
𝑙 driver
where 𝑉𝑥 is the longitudinal velocity of the vehicle, 𝑙 is the (iii) AFS System. For this subsystem the allocation decides to
wheel base, and 𝛿driver is the drivers steering angle command. introduce or not the AFS control action, as
For the two input variables, five fuzzy sets with triangular
Membership Functions (MFs) were used for each variable: 𝛿∗ = 𝛿driver + 𝑎𝑢steer ⋅ 𝛿AFS , (13)
̇ = {NB, NS, Z, PS, PB}, whereas, for the output
{𝑒(𝛽), 𝑒(𝜓)}
variable, seven fuzzy sets also with triangular MFs were used: where 𝛿∗ is the desired steering wheel angle, 𝛿driver is
{𝑀𝑧 } = {NB, NM, NS, Z, PS, PM, PB}. Table 4 describes the the driver’s command, and 𝛿AFS is the compensation angle
meanings of the used linguistic terms and Table 5 shows calculated by the AFS system. From (13), it can be seen
Table 6: FL inference rules for the steering system. (iii) Braking: the local controller is an ABS; it modifies the
desired command as
𝑒(𝜓)̇
𝛽 𝛿driver
NB NS Z PS PB 𝑇𝑏∗ = 𝐺ABS ⋅ max (𝑇driver , 𝑇ESC
∗
), (16)
NB NS NS Z PB PB
NS NMS NMS Z PMH PMH where 𝑇driver is the driver braking command, 𝑇ESC
Low Z NM NM Z PM PM is the braking command from the allocation system,
PS NMH NMH Z PMS PMS and 𝐺ABS is the gain of the ABS that releases the
PB NH NH Z PS PS tire when it is locked. This control law selects the
NB NH NH Z PS PS
maximum between the command coming from the
allocation system and the command from the driver.
NS NMH NMH Z PMS PMS
The 𝐺ABS gain is obtained by the function
High Z NM NM NS PMS PMS
PS NMS NMS NS NS NS
{0 if 𝜆 𝑖,𝑗 ≥ 𝜆 crit
PB NS NS NS NS NS 𝐺ABS𝑖,𝑗 = { (17)
1 if 𝜆 𝑖,𝑗 < 𝜆 crit
{
that the driver’s command is always considered, but the with
compensation angle is considered if 𝑎𝑢steer = 1 or not if 𝑎𝑢steer =
0. 𝑉𝑥 − V𝑥𝑖,𝑗
As for the braking system, the steering action is computed 𝜆 𝑖,𝑗 = , (18)
𝑉𝑥
using a FL controller [30]. Three input variables were selected:
(1) side slip angle (𝛽 ∈ [−10, 10]), (2) yaw rate error where 𝜆 𝑖,𝑗 is the slip ratio for each wheel. The
(𝑒(𝜓)̇ = 𝜓̇ − 𝜓̇ 𝑑 ∈ [−10, 10]), and (3) steering angle input operation of this ABS controller is guided by 𝜆; if 𝜆
(𝛿driver ∈ [−10, 10]), and one output: (1) steering correction grows beyond the admissible range (𝜆 crit = 0.1), the
angle (𝛿AFS ∈ [−5, 5]). The FL controller is oriented to create braking system releases the tire until it recovers grip
a steering wheel angle correction that minimizes the yaw rate and the slip ratio decreases to the admissible range,
error; the desired yaw rate is calculated using (10). where the braking torque is again applied to the tire.
For input 𝛽 two fuzzy sets with sigmoid MFs were used:
𝛽 = {Low, High}. For the other two input variables, five
fuzzy sets with triangular MFs were used: {𝑒(𝜓), ̇ 𝛿driver } = 3.3. Physical Layer. The physical layer is integrated by the
{NB, NS, Z, PS, PB}, whereas, for the output variable, eleven sensors and actuators of the SAS, AFS system, and 4WIB
fuzzy sets, also with triangular standard MFs, were used system.
{𝛿AFS } = {NB, NMH, NM, NMS, NS, Z, PS, PMS, PM, PMH,
PB}. Table 6 shows the rules for the proposed FL controller, 3.3.1. Suspension System. The suspension system is composed
whose linguistic definitions are given in Table 4. of a linear spring and a SA shock absorber. The key element is
the set of SA shock absorber, one at each corner, which needs
3.2.2. Local Controllers. This sublayer contains the local to be modeled as a function of a manipulation signal (𝜐). The
controllers for each subsystem. These controllers are Single- damper force (𝐹𝐷) is modeled as a function of the damper
Input Single-Output (SISO) and only interact with their deflection (𝑧def ), deflection velocity (𝑧̇ def ), and manipulation
particular subsystem. These controllers receive the desired signal [31]:
set-point 𝑢𝑐∗ from the control allocation sublayer and are 𝐹𝐷 = 𝑐𝑝 (𝑧̇ def ) + 𝑘𝑝 (𝑧def ) + 𝐹SA , (19)
in charge of executing it. Also, these controllers have to be
selected regarding the subsystem to control. The subsystem where 𝐹SA = 𝜐 ⋅ 𝑓𝑐 ⋅ tanh(𝑎1 ⋅ 𝑧̇ def + 𝑎2 ⋅ 𝑧def ) is the SA force
controllers for this strategy are the following: due to 𝜐, 𝑐𝑝 is a viscous damping coefficient, 𝑘𝑝 is a stiffness
(i) SAS: because the control command coming from the coefficient, and 𝑎1 and 𝑎2 are hysteresis coefficients due to
allocation step is binary, the force control system is velocity and displacement, respectively. The SA damping
defined as follows: force at each damper is continuous from [−10000, 6000] N.
𝑐min 0 10% 3.3.2. Steering System. A steer-by-wire Active Steering (AS)

[ ] = [ ] 󳨃󳨀→ 𝜐 = [ ], (14)
𝑐max 1 90% system is used to provide an additional steering angle for
corrective purposes; the actuator model is given by [32]
where 𝜐 is the manipulation delivered to the SA +
damper, (i.e., electric current, voltage, and duty cycle). 𝛿̇ = 2𝜋𝑓steer (𝛿∗ − 𝛿+ ) , (20)
(ii) Steering: the command, in terms of tire angles, to where 𝑓steer = 10 Hz is the cut-off frequency of the actuator
transform it to a single gain controller is dynamics, 𝛿∗ is the steering controller output, and 𝛿+ is
∗ 28.74 ∗ the actuator output; the bounded limits of actuation are
𝛿steering wheel = 𝛿 . (15) [−5∘ , +5∘ ].
1.18 tires
10
zr (mm)
−10
0 50 100 150 200
Station (m)
(a) Road profile with bump (b) Brake distance test (hard braking)
(c) DLC test (rapid steering) (d) FH test (cornering)
(e) Split 𝜇-surface braking (loss of vehicle control)
Figure 6: Implemented tests in CarSim.
3.3.3. Braking System. The corrective front and rear braking of the implemented tests to represent the driving situations;
torques are provided by brake-by-wire Electromechanical these tests are as follows:
Braking (EMB) actuators; the model of these EMB actuators
(a) Road profile with bump test: this test is intended
is given by [32]
to evaluate the road isolation characteristics of the
+ control strategy; it consists in a rough road with a
𝑇̇ 𝑏 = 2𝜋𝑓brake (𝑇𝑏∗ − 𝑇𝑏+ ) , (21)
sharp bump of 35 mm height and 400 mm length.
where 𝑓brake = 10 Hz is the cut-off frequency of the actuator (b) Brake distance test: it consists in a hard braking action
dynamics, 𝑇𝑏∗ is the braking controller output, and 𝑇𝑏+ is the by the driver to measure the distance that takes the
actuator output. The bounded limits of actuation for each vehicle when it goes from 100 km/h to a full stop.
braking actuator are [0, 15] MPa. (c) Double Line Change (DLC) maneuver: it consists in
a change of driving line to simulate an obstacle
4. GCC System Evaluation avoidance maneuver or an overtaking action at high
speed (120 km/h); a rapid steering maneuver is taken
The results of this proposal are presented based on a case by the driver to change from the original line, and
study. then another rapid steering action to turn back to the
original line.
4.1. Case Study. CarSim was used to generate an accurate (d) Fish Hook (FH) maneuver: this maneuver consists in
vehicle model whose VDC were hosted in Matlab/Simulink. a wide, but constant steering action; first a movement
A 𝐷-class sedan was the selected vehicle. Seven driving of 270∘ of the steering wheel is taken to one side
situations were considered (𝑟 = 7). Figure 6 illustrates some at a constant turning rate; then a turn of 540∘ to
Table 7: IS set. Variables correlations matrix 1

ID Variable Description 20
0.8
1–4 𝑧def𝑖,𝑗 Damper deflection
0.6
5–8 𝑧̇ def𝑖,𝑗 Damper deflection rate
15 0.4
9 𝜙 Pitch
10 𝜙̇ Pitch rate 0.2
Variables
11 𝜃 Roll 0
12 𝜃̇ Roll rate 10
−0.2
13 𝜓 Yaw
−0.4
14 𝜓̇ Yaw rate
15 𝑥̈ Longitudinal acceleration 5 −0.6
16 𝑦̈ Lateral acceleration −0.8
17 𝑧̈ Vertical acceleration
0 −1
18 𝛽 Vehicle slip angle 0 5 10 15 20
19 𝛽̇ Vehicle slip angle rate Variables
20 𝛿 Steering wheel angle
Figure 7: Correlation matrix of the IS set for a rapid steering driving
21 𝛿̇ Steering wheel angle rate situation.
the opposite side is executed transferring the vehicle 8 100

load from one side to another; then the steering angle
Cumulative variance (%)

is held to create a cornering situation which maintains 6
the load transfer at a dangerous limit. 90% of the
Eigenvalue
variance
(e) Split 𝜇-surface braking test: the intention of this test 4 50
is to evaluate how well a vehicle can keep its line
during a braking action while riding with different 2 4 components
friction coefficients in each side of the vehicle; this
test is intended to cause loss of vehicle control from 0 0
the driver. 1 5 10 15 21
Number of components
4.2. Decision Layer. For space limitations, the results of the Figure 8: PCA result for the rapid steering situation.
DL are presented just for the rapid steering driving situation.
4.2.1. Classification Algorithm. The IS set has a cardinality of

Contribution to data variance (%)
25
|IS| = 21 elements (variables). Table 7 describes the consid- .
𝛿
𝜓
ered variable set. These variables were selected because they 20
..
describe completely the behavior of the vehicle. x
.
15 𝜙
Figure 7 shows the correlation level among the elements
in the IS set, for the DLC test. Each square represents the 10
level of correlation between two variables; for example, in
the grid the principal diagonal shows a correlation of 1 (dark 5
red) because it is the correlation of the variable with itself. 0
A correlation of −1 (dark blue) indicates that the variable 0 5 10 15 20
has an inverse correlation with the variable in question; for Variables
example, variable #1 (𝑧def 𝐹𝐿 ) has a correlation of 1 with variable Figure 9: Contribution plot of the variables in the IS set.
#2 (𝑧def 𝑅𝐿 ) and a correlation of −1 with variable #3 (𝑧def 𝐹𝑅 ).
The above example can be interpreted as follows: during a
roll situation, caused by the steering maneuver, the dampers
of the same side (left) have the same compression whereas the test. Figure 8 demonstrates that when the cumulative vari-
the dampers of the other side have the same movement, ance marker (green line) surpasses the 90% mark (horizontal
but in the opposite direction. It is notable that there is an red dotted line), the number of components is 4.
inherent correlation among some variables, some of them can It is important to relate the result with the IS set, such
be neglected. that the contribution of each variable to explain the total
To find the MS set for this situation, the PCA algorithm data variance can be studied in the residual space. Figure 9
was performed, Figure 8. These results point out that only shows the contribution of all variables for the DLC test. Those
4 PC could explain more than 90% of the total variance of variables that overshoot the threshold 𝑡con = 10% (red dotted
Table 8: PCA results.

ID Test # of PC % of explained variance Variables in MS set
1 Ride 7 92.7 8, 12, 14, 17, 21
2 Road irregularity 4 90.2 9, 10, 17
3 Acceleration/braking 4 97.8 5, 6, 7, 8, 10
4 Hard braking 5 97.5 5, 6, 7, 8, 10
5 Cornering 7 93.1 10, 13, 18, 20
6 Rapid steering 4 93.7 10, 13, 15, 21
7 Loss of control 7 92.6 13, 14, 18, 19
15
Target
DC = 6
10 DC = 5
5
Roll rate (deg./s)
Output
0
DC = 1
−5
0 5 10 15 20 25
−10 Time (s)
Figure 11: Validation test for the 𝑘-NN classifier.

−15
0.2 0.4 0.6 0.8 1
Pitch (deg.)
the two conditions (𝐷𝐶 = 5 and 𝐷𝐶 = 6) are really
Cornering Cornering similar in essence: both include steering action, but the
Loss of control Rapid steering
Rapid steering Riding
difference lays in the velocity of the steering movement.
Riding Centroids From Figure 11, it is clear that rapid steering action is well
detected by the classifier, but cornering has some drawbacks.
Figure 10: Clustering distribution for 2 dimensions. During the rapid steering situation, some variables related
to the vertical and lateral dynamics are highly sensitive, but
during cornering the steady behavior of some movements is
misjudged as riding conditions.
line) are considered the most representative variables in this Figure 12 presents the ROC of the classifier results for the
driving situation; they integrate the MS set. test in Figure 11. The riding situation is well classified, with a
Table 8 presents the PCA results for the seven driving false alarm rate of 20%. For the other two situations, the false
situations. It can be concluded that the minimum set of alarm is zero but the detection error is 10% because of the
variables for identifying any driving situation in MS = rapid steering situation, while the cornering condition has the
{5–10, 12–15, 17–21}. Once MS set is defined, the clustering worst identification performance (error of 50%).
step runs. From the previous analysis it can be concluded that
A 𝑘-NN classifier with 𝑘 = 7 was trained to classify 4 the situations that involve roll and yaw movements can be
major driving situations (riding, cornering, rapid steering, well classified, but other driving situations (road irregularity,
and loss of control), where the first four tests in Table 8 acceleration/braking, and hard braking) which are mostly
are tagged as riding situation. Figure 10 shows the clusters pitch related can not be classified since all other driving
distribution in 2D (pitch angle versus roll rate); note that situations also involve pitch movement.
clusters are differentiated by color. More than one cluster To assist the 𝑘-NN classifier, a Suspension Adjustment
determine one of the four driving conditions: a cornering Plane (SAP) was designed. SAP studies the vertical load trans-
situation uses two clusters (red and cyan cluster). fer caused by vehicle movements, to adjust the SAS system.
To evaluate the classifier performance, a test, with a series Its objective is to select the suspension settings depending on
of different driving situations, was designed. The test begins which area of the plane the vehicle is in. For example, if the
with the vehicle being driven in a rough road (𝐷𝐶 = 1); vehicle is in a frontal pitch situation (braking) the system sets
afterwards it has a rapid steering maneuver (𝐷𝐶 = 6); the two front dampers in road-holding mode, since most of
once the DLC maneuver passes and the vehicle is stabilized the load is transferred to those tires. The same happens with
again, the test finishes with a cornering action (𝐷𝐶 = 5, a roll movement where the tires of the same side receive most
FH maneuver). Figure 11 shows the target as well as the of the load. The center area (between the red dotted lines) of
classification result for this test. This test was selected because the SAP is concerned with a normal riding situation. This area
1 Table 9: 𝐷𝐶 vectors for different driving conditions.
𝐷𝐶
Driving situation
DC = 6 𝑠𝑖 𝑠𝑠 𝑎𝑢susp 𝑎𝑢steer 𝑎𝑢braking
0.8
Ride 1 0 [0, 0, 0, 0] 0 0
Road irregularity 2 0 SAP 0 0
DC = 5 Acceleration/braking 3 0 SAP 0 0
True positive rate
0.6
Hard braking 4 0 SAP 0 0
Cornering 5 0 SAP 1 0
Rapid steering 6 1 SAP 1 1
0.4
DC = 1 Loss of control 7 1 [1, 1, 1, 1] 1 1
0.2 whereas the horizontal central zone refers to primarily roll

movements like cornering.
4.2.2. Decision Logic. Table 9 presents the proposed values

0
0 0.2 0.4 0.6 0.8 1 for vector 𝐷𝐶 at each driving situation.
False positive rate
4.3. Control Layer. The performance of the proposed GCC
Figure 12: ROC for the 𝑘-NN classifier.
system is evaluated and compared with two additional control
cases. The first case corresponds to an uncontrolled (UC) sys-
tem; that is, it is a vehicle which lacks any control system and
aususp = [1, 1, 0, 0] is equipped with standard actuators and a passive suspension
1.4 aususp = [1, 1, 1, 0]
system. The second case corresponds to a vehicle with a set
aususp = [1, 1, 0, 1] of dynamics controllers without a coordination strategy in
1 Decentralized Controllers (DC); that is, a set of controllers are
aususp = [1, 0, 1, 0] acting simultaneously seeking their own control goals.
Pitch (deg.)
To evaluate quantitatively the performance at each driv-

0.6 ing test, the Root Mean Square (RMS) value of the signals
is used. To have a point of comparison the RMS values of
the controlled cases (DC and GCC) are normalized with that
0.2 obtained by the UC case using
aususp = [0, 0, 1, 1] aususp = [0, 1, 0, 1]
% of Improvement
−0.2
−3 −2 −1 0 1 2 3
RMS (𝑋𝑖Passive ) − RMS (𝑋𝑖Controlled ) (22)
Roll (deg.) = .
RMS (𝑋𝑖Passive )
Riding Road irregularity
Hard braking Cornering
Acceleration/braking Rapid steering 4.3.1. Road Profile with Bump Test. For this test, two variables
Loss of control are important: (1) vertical acceleration of the sprung mass
Figure 13: Suspension Adjustment Plane (SAP). (𝑧̈ 𝑠 ), Figure 14(a), and (2) pitch angle (𝜙), Figure 14(b). For the
vertical acceleration, Figure 14(a), the GCC system achieves
9.6% of improvement in vibration mitigation when compared
with the reference case (UC); on the other hand the DC
was defined by considering different riding conditions (road case achieves 3.11%. During this test the bump it intended to
roughness and velocities). Figure 13 shows the SAP with excite the pitch motion of the vehicle, which is also related
different driving situations and also the proper suspension to the comfort of the passengers, Figure 14(b); the GCC
combinations are marked for each region. With this plane it is system manages to reduce the pitch motion of the vehicle
possible to classify which riding situation occurs (riding per by 11.7% whereas the DC case achieves 10.8%, both results
se, a road irregularity, an acceleration/braking situation, or a when compared with the UC case. The main improvement
hard braking). of the control strategy can be seen after the bump (1 second
In Figure 13, the dotted lines delimit the threshold for mark); during that transient response the vehicle reaches a
pitch and roll movements caused by road roughness; outside stable behavior faster than the reference case.
those limits it is considered that the experienced movements
are caused by other driving situations. The central verti- 4.3.2. Braking Distance Test. Two variables are the most
cal zone includes primarily pitch situations, like braking, important for this test: (1) brake distance, Figure 15(a), and
Vertical acceleration (G’s) 0.8 1.2

UC
UC
Pitch angle (deg.)

0.4 0.8
0 0.4
−0.4 GCC 0
DC DC GCC
−0.8 −0.4
0 0.5 1 1.5 0 0.5 1 1.5
Time (s) Time (s)
(a) Vertical acceleration (𝑧̈ 𝑠 ) (b) Pitch angle (𝜙)
Figure 14: Road profile with bump test.
70
60 1.6
DC
Brake distance (m)
50
1.2
Pitch (deg.)
40
30 UC (69 m) 0.8
20 DC (63 m) GCC
0.4
10 UC
GCC (59 m)
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time (s) Time (s)
(a) Brake distance (b) Pitch angle (𝜙)
Figure 15: Brake distance test.
(2) pitch angle 𝜙, Figure 15(b). Regarding brake distance, the 4

reference case (UC) takes 69 m to fully stop, whereas the 3 UC
controlled cases have improvements of 8.7% for the DC and
y-coordinate (m)
GCC
14.5% for the GCC, meaning a stop distance of 10 m less than 2
that of the UC case. For the pitch angle, the improvement 1
of the DC against the UC was 6.2%; whereas GCC has an
0
improvement of 10.3%. Also, in Figure 15(b) it can be seen
that the GCC has less overshoot and shorter stabilization −1 DC
time. Those oscillations cause loss of grip of the tires and −2
produce uncomfortable movements for the passengers. Even 0 50 100 150 200 250
when the improvement indices for both controlled cases are x-coordinate (m)
marginal for the pitch angle, the uncomfortable behaviors are
Figure 16: Vehicle trajectory for a DLC test.
successfully attenuated.
4.3.3. DLC Test. Figure 16 shows the vehicle trajectory for

a rapid stabilization. For the DC the roll movement is abruptly
each case. It can be observed that the GCC has the smallest
caused by the lack of coordination of the steering and braking;
𝑦-coordinate deviation, and also its finish is straight, whereas
this causes those oscillations and a deterioration of 8.9%
for the UC the 𝑦-coordinate deviation is bigger and at the end
compared to the UC case. In Figure 17(b) also the GCC
its path is diverging. The DC has a performance similar to the
system has less yaw motion, 12.5% less than the UC, while
UC, but at the end the 𝑦-coordinate deviation is bigger (more
the DC has a deterioration of 4% compared to the UC case.
negative) than the other cases.
After assessing the vehicle trajectory, it is important to
analyze the behavior of the related variables to the vehicle 4.3.4. FH Test. In this test, the evaluation starts with the
stability, such as roll angle (𝜃), Figure 17(a), and yaw angle assessment of the vehicle trajectory. Figure 18 shows the
(𝜓), Figure 17(b). Figure 17(a) illustrates that the GCC system trajectory for the three cases. As it can be seen, the GCC
has less roll movement than the other cases, that is, 10% less system has a smaller turn radius than the other two cases,
than UC; also the roll movement has less duration having especially when compared with UC; this indicates more
2 6
GCC
1 GCC 2
Roll (deg.)
Yaw (deg.)
0 −2
−1 DC −6 DC
UC
UC
−2 −10
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time (s) Time (s)
(a) Roll angle (𝜃) (b) Yaw angle (𝜓)
Figure 17: DLC test.
25
20 DC
y-coordinate (m)
15
10
UC
5
0 GCC
−5
0 20 40 60 80 100
x-coordinate (m)
Figure 18: Vehicle trajectory for FH test.
4
0.8
GCC
2
Lateral acc. (g)
GCC 0.4
UC
Roll (deg.)
DC
0 0
−0.4
−2
UC
−0.8
DC
−4
0 1 2 3 4 5 0 1 2 3 4 5
Time (s) Time (s)
(a) Roll angle (𝜃) (b) Lateral acceleration (𝑦)̈
Figure 19: FH test.
handling capabilities in GCC. Also, the proposed GCC has load takes place, even when the improvement is marginal: 2%
a shorter trajectory, which indicates less skidding than other with respect to UC. There is no significant improvement with
cases. respect to DC.
The significant variables are (1) roll angle (𝜃), Figure 19(a),
and (2) lateral acceleration (𝑦), ̈ Figure 19(b). Since the
induced cornering situation by the steering movement is 4.3.5. Split 𝜇-Surface Test. This test is designed to evaluate
really similar regardless of the control system, the important the ability of a vehicle to remain in control against a critical
features to study are the transient responses of the variables situation. It is important to verify how well the vehicle
at the time when the cornering starts and ends. From remains in its original path after the braking action starts.
Figure 19(a) it is evident that in GCC the induced oscillations Figure 20 shows the trajectory of the three Vehicle Control
by the change of load transfer are better absorbed by the Systems. The controlled vehicle with GCC remains in straight
suspension system, that is, 7% less oscillations than in the UC line during the whole test, whereas UC moves away from the
case and 3% less than in DC. Regarding lateral acceleration, original trajectory. On the other hand, the vehicle with DC
Figure 19(b), GCC has less oscillations when the change of also manages to remain close to the straight path.
UC
y-coordinate (m)
1
GCC
−1 DC
0 10 20 30 40 50 60
x-coordinate (m)
Figure 20: Vehicle trajectory for Split 𝜇-surface braking test.
800
DC
0.2 UC
Steering wheel angle (deg.)
Lateral acceleration (g)

600
UC
400 0
DC
200
−0.2
GCC
0 GCC
−0.4
0 1 2 3 4 5 0 1 2 3 4 5
Time (s) Time (s)
(a) Steering wheel angle (𝛿) (b) Lateral acceleration (𝑦)̈
20
0
Yaw rate (deg./s)
−20 DC
GCC
−40
UC
−60
−80
−100
0 1 2 3 4 5
Time (s)
(c) Yaw rate (𝜓)̇
Figure 21: Split 𝜇-surface braking test.
The important variables for this test are (1) steering the steering action, similar results are obtained, Figure 21(b).
wheel angle (𝛿), Figure 21(a), (2) lateral acceleration (𝑦), ̈ GCC has 88.7% less lateral acceleration than UC and 10.1%
Figure 21(b), and (3) yaw rate (𝜓),̇ Figure 21(c). Figure 21(a) less acceleration than DC.
shows that the driver does not affect significantly the steering Finally, to evaluate the rotational movements of the
wheel angle, that is, the GCC and DC approaches can stabilize vehicle in the road, the yaw rate is analyzed. Figure 21(c)
the vehicle. On the other hand, the UC vehicle begins to lose shows that in UC the driver loses completely the vehicle
control and the driver has to make a huge adjustment to try control causing too much yaw motion. GCC has 98% less yaw
to keep the stability of the vehicle. It can be noticed that the rate than UC and 96% less than DC.
steering allocation system takes action increasing the steering
maneuver for DC; this extra movements cause the vehicle to 5. Conclusions
diverge slightly from the original path. The steering action
for GCC is 87.9% less than UC vehicle, while DC has 80.6% A novel Global Chassis Control system was proposed. It
less movement. Because the lateral acceleration is induced by combines fuzzy logic inference systems, data-based logic and
heuristic techniques as classifiers, allocation systems, and International Congress on Transportation Electronics, pp. 97–106,
model-free controllers; also, it takes the advantages of a priori Dearborn, Mich, USA, October 1988.
knowledge of a vehicle driving to develop an efficient control [8] T. El-Sheikh, Concept of global chassis control: using model
strategy. Also, it has the advantage of being modular and com- reference and virtual actuators [Ph.D. thesis], University of
putationally light; based on Principal Components Analysis California, Oakland, Calif, USA, 2011.
data reduction was applied. Most of the computing is done [9] J. He, Integrated vehicle dynamics control using active steering,
offline; the onboard systems only execute the classification of driveline and braking [Ph.D. thesis], University of Leeds, Leeds,
the driving situations and calculate the allocation and control UK, 2005.
procedures. [10] M. Valasek, O. Vaculin, and J. Kejval, “Global chassis control:
The evaluation of the strategy was performed with several integration synergy of brake and suspension control for active
well-known tests using CarSim. Simulation results showed safety,” in Proceedings of the 7th International Symposium on
that the performance of the Global Chassis Control system Advanced Vehicle Control, pp. 495–500, Arnhem, The Nether-
lands, 2004.
outperforms the uncontrolled vehicle case and the Decen-
tralized Controllers scheme that do not have coordination [11] P. Gáspár, Z. Szabó, J. Bokor, C. Poussot-Vassal, O. Sename,
features. Qualitative and quantitative results of the evaluation and L. Dugard, “Global chassis control using braking and
suspension systems,” in Proceedings of the 20th Symposium of
demonstrate the importance of a coordination level when
the International Association for Vehicle System Dynamics, pp.
different controllers coexist and interact in an integrated 13–17, Berkeley, Calif, USA, August 2007.
system as a vehicle. Such coordination level consists in a
[12] R. D. Fruechte, A. M. Karmel, J. H. Rillings, N. A. Schilke, N.
classification algorithm, based on the 𝑘-Nearest Neighbor M. Boustany, and B. S. Repa, “Integrated vehicle control,” in
clustering technique, supported by a Suspension Adjustment Proceedings of the 39th IEEE Vehicular Technology Conference:
Plane to identify different driving situations. Gateway to New Concepts in Vehicular Technology, vol. 2, pp.
868–877, San Francisco, Calif, USA, May 1989.
Conflict of Interests [13] J. Andreasson and T. Bunte, “Global chassis control based on
inverse vehicle dynamics models,” Vehicle System Dynamics, vol.
The authors declare that there is no conflict of interests 44, supplement 1, pp. 321–328, 2006.
regarding the publication of this paper. [14] O. Mokhiamar and M. Abe, “How the four wheels should
share forces in an optimum cooperative chassis control,” Control
Engineering Practice, vol. 14, no. 3, pp. 295–304, 2006.
Acknowledgments [15] W. Cho, J. Choi, C. Kim, S. Choi, and K. Yi, “Unified chassis
The authors thank Tecnológico de Monterrey and CONACyT control for the improvement of agility, maneuverability, and
lateral stability,” IEEE Transactions on Vehicular Technology, vol.
for their partial support through the Automotive Consortium
61, no. 3, pp. 1008–1020, 2012.
research chair and PCP 03/10 and 06/13 bilateral (México-
France) projects. [16] M. Ali, P. Falcone, C. Olsson, and J. Sjöberg, “Predictive preven-
tion of loss of vehicle control for roadway departure avoidance,”
IEEE Transactions on Intelligent Transportation Systems, vol. 14,
References no. 1, pp. 56–68, 2013.
[17] S. Sato, H. Inoue, M. Tabata, and S. Inagaki, “Integrated chassis
[1] H. Chou and B. D’Andréa-Novel, “Global vehicle control using control system for improved vehicle dynamics,” in Proceedings
differential braking torques and active suspension forces,” of the International Symposium on Advanced Vehicle Control
Vehicle System Dynamics, vol. 43, no. 4, pp. 261–284, 2005. (AVEC ’92), pp. 413–418, Yokohama, Japan, 1992.
[2] S. Drakunov, U. Ozguner, P. Dix, and B. Ashrafi, “ABS control [18] C. Poussot-Vassal, Robust multivariable linear parameter vary-
using optimum search via sliding modes,” IEEE Transactions on ing automotive global chassis control [Ph.D. thesis], Institut
Control Systems Technology, vol. 3, no. 1, pp. 79–85, 1995. Polytechnique de Grenoble, Grenoble, France, 2008.
[3] TechCast, “Bubble charts of TechCast’s latest results on TRANS- [19] S.-B. Lu, Y.-N. Li, S.-B. Choi, L. Zheng, and M.-S. Seong,
PORTATIONs,” Tech. Rep., TechCast-LLC, 2012. “Integrated control on MR vehicle suspension system associated
[4] WHO, “Global status report on road safety 2013: supporting a with braking and steering control,” Vehicle System Dynamics,
decade of action,” Technical Report, World Health Organiza- vol. 49, no. 1-2, pp. 361–380, 2011.
tion, 2013. [20] S. Fergani, O. Sename, and L. Dugard, “Performances improve-
[5] C. J. Kahane, “Lives Saved by vehicle safety technologies and ment through an 𝐿𝑃𝑉 = 𝐻∞ control coordination strategy
associated federal motor vehicle safety standards, 1960 to 2012 involving braking, semi-active suspension and steering Sys-
passenger cars and LTVs with reviews of 26 FMVSS and the tems,” in Proceedings of the 51st IEEE Conference on Decision
effectiveness of their associated safety technologies in reducing and Control (CDC ’12), pp. 4384–4389, Maui, Hawaii, USA,
fatalities, injuries, and crashes,” Tech. Rep., National Highway December 2012.
Traffic Safety Administration, Washington, DC, USA, 2015. [21] T. Gordon, M. Howell, and F. Brandao, “Integrated control
[6] F. Yu, D.-F. Li, and D. A. Crolla, “Integrated vehicle dynamics methodologies for road vehicles,” Vehicle System Dynamics, vol.
control-state-of-the art review,” in Proceedings of the IEEE 40, no. 1–3, pp. 157–190, 2003.
Vehicle Power and Propulsion Conference (VPPC ’08), pp. 1–6, [22] J. Nayak, B. Naik, and H. Behera, “Fuzzy C-means (FCM)
Harbin, China, September 2008. clustering algorithm: a decade review from 2000 to 2014,” in
[7] N. Schilke, R. Fruechte, N. Boustany, A. Karmel, B. Repa, and Computational Intelligence in Data Mining—Volume 2, L. C.
J. Rillings, “Integrated vehicle control,” in Proceedings of the Jain, H. S. Behera, J. K. Mandal, and D. P. Mohapatra, Eds., vol.
32 of Smart Innovation, Systems and Technologies, pp. 133–149,

Springer, 2015.
[23] P. Gáspár, Z. Szabó, J. Bokor, C. Poussot-Vassal, O. Sename, and
L. Dugard, “Towards global chassis control by integrating the
brake and suspension systems,” in Proceedings of the 5th IFAC
Symposium on Advances in Automotive Control, pp. 563–570,
Pajaro Dunes, Calif, USA, August 2007.
[24] A. Tavasoli, M. Naraghi, and H. Shakeri, “Optimized coordina-
tion of brakes and active steering for a 4WS passenger car,” ISA
Transactions, vol. 51, no. 5, pp. 573–583, 2012.
[25] T. Raste, R. Bauer, and P. Rieth, “Global chassis control:
challenges and benefits within the networked chassis,” in Pro-
ceedings of the FISITA World Automotive Congress, Munich,
Germany, 2008.
[26] W. Chen, H. Xiao, L. Liu, J. W. Zu, and H. Zhou, “Integrated
control of vehicle system dynamics: theory and experiment,”
in Advances in Mechatronics, chapter 1, InTech, Rijeka, Croatia,
2011.
[27] R. P. Good, D. Kost, and G. A. Cherry, “Introducing a unified
PCA algorithm for model size reduction,” IEEE Transactions on
Semiconductor Manufacturing, vol. 23, no. 2, pp. 201–209, 2010.
[28] R. Isermann, Fault-Diagnosis Systems, Springer, Berlin, Ger-
many, 1st edition, 2006.
[29] B. L. Boada, M. J. L. Boada, and V. Dı́az, “Fuzzy-logic applied
to yaw moment control for vehicle stability,” Vehicle System
Dynamics, vol. 43, no. 10, pp. 753–770, 2005.
[30] S. Krishna, S. Narayanan, and S. D. Ashok, “Control of yaw
disturbance using fuzzy logic based yaw stability controller,”
International Journal of Vehicular Technology, vol. 2014, Article
ID 754218, 10 pages, 2014.
[31] S. Guo, S. Yang, and C. Pan, “Dynamic modeling of magne-
torheological damper behaviors,” Journal of Intelligent Material
Systems and Structures, vol. 17, no. 1, pp. 3–14, 2006.
[32] M. Doumiati, O. Sename, L. Dugard, J.-J. Martinez-Molina,
P. Gaspar, and Z. Szabo, “Integrated vehicle dynamics control
via coordination of active front steering and rear braking,”
European Journal of Control, vol. 19, no. 2, pp. 121–143, 2013.
2.3. FURTHER EVALUATION AND JUSTIFICATION OF THE GCC SYSTEM 37
2.3 Further evaluation and justification of the GCC system

This section is an addendum of the publication presented in Chapter 2.2. Here, a further evaluation
of the proposed GCC system is carried out. I]n particular, a deeper assessment of the decision
and control layers is done. Those evaluations are oriented to validate some features of the system
such as the discrimination algorithm, the design flexibility characteristic, etc. The evaluations are
divided in terms of the layers and are done individually.
2.3.1 Decision layer

This section presents more results regarding the classification algorithm as well as an extensive
justification of the decision logic module coordination scheme.
Classification algorithm
First, further results of the discrimination algorithm are shown; then, a more detailed evaluation of
the classification algorithm is carried out.
Discrimination algorithm - Offline analysis
First, the proposed discrimination procedure is recalled by presenting briefly the algorithm
below. This algorithm was fully presented in section 2.2:
Discrimination Algorithm: from an Initial Set (IS) of variables, obtain the the Minimal Set
(M S) to classify the current driving situation. where M S ⊆ IS : |M S| ≤ |IS|
Requirement: The seven IS sets obtained from the driving situations.
1: To neglect the noise effect in the scaling process, variables with small variance are not
considered, σxi,k < tny , where tny > 0 is a defined threshold.
2: Determine the most significant components, X̂k , for each of the seven situations by
doing a PCA with a cumulative variance limit of 90 %.
3: Using the residual components, X̃k , the contribution of each variable i can be obtained
in each k driving situation as:
Pm 2
j=1 x̃i,k (j)
Coni,k = Pn Pm 2 ∀i = {1, 2, ..., n}
i=1 j=1 x̃i,k (j)
4: Once the contributions are obtained, if Coni,k > tcon then Coni,k 7→ CSk , where CSk
is the set of variables that contributes to each k driving situation and tcon > 0 is a
defined threshold.
38 CHAPTER 2. GLOBAL CHASSIS CONTROL SYSTEM
5: Obtain the M S set as:

r
[
MS = CSk ∀k = {1, 2, ..., r}
k=1
The aforementioned discrimination algorithm has two main functions:
• Reduce the size of the required variable set to be used in the k-NN algorithm.
• Identify the key variables that allow the correct classification of the current driving situation,
by means of a k-NN algorithm.
In this work, for the Initial Set (IS) 21 variables were selected. Table 2.1 shows these variables.
Table 2.1: IS set

ID Variable Description ID Variable Description
1-4 zdefi,j Damper compression 15 ẍ Longitudinal acceleration
5-8 żdefi,j Damper compression rate 16 ÿ Lateral acceleration
9 φ Pitch 17 z̈ Vertical acceleration
10 φ̇ Pitch rate 18 β Vehicle slip angle
11 θ Roll 19 β̇ Vehicle slip angle rate
12 θ̇ Roll rate 20 δ Steering wheel angle
13 ψ Yaw 21 δ̇ Steering wheel angle rate
14 ψ̇ Yaw rate
In addition, 7 driving conditions (DC ) were considered. These driving conditions are some of
the most common driving situations experienced. The seven driving situations are presented in a
Pitch-Roll plane, as shown in Fig 2.1. This plane is used to put the behavior of a driving situation
in perspective.
[DC = 1] Ride: This situation considers a vehicle being driven at a constant speed on different
types of road profiles (smooth to rough) in a straight line.
[DC = 2] Road irregularity: These irregularities include bumps and potholes. It is characterized
by having abrupt damper deflections with a short duration.
[DC = 3] Acceleration/Braking: Here the vehicle experiences back/frontal pitch movements caused
by the acceleration/braking action of the driver.
1.4 Riding
Road Irregularity
1
Pitch [deg]
Acceleration/Braking
0.6 Hard Braking
Cornering
0.2 Rapid Steering
−0.2 Loss of Control
−3 −2 −1 0 1 2 3
Roll [deg]
Figure 2.1: Pitch-Roll plane, showing the behavior of the seven driving situations.
[DC = 4] Hard braking: This is an emergency braking maneuver when the driver suddenly fully
brakes. This action generates a large and sustained frontal pitch movement, but also can
cause the tires to lose friction and slip.
[DC = 5] Cornering: This situation considers a vehicle going around a corner at a constant speed
with a constant or slow steering angle movement (≈ δ̇ < 20 deg/s).
[DC = 6] Rapid steering: Here the vehicle suddenly makes a steering movement (≈ δ̇ > 20
deg/s). This action usually causes under/oversteering.
[DC = 7] Loss of control: This situation is characterized by having large values of yaw angle rate
(≈ ψ̇ > 30 deg/s), and such value can be interpreted as the vehicle is spinning.
In the following discussion the results of the discrimination algorithm for the seven driving
conditions are presented. These results are shown in terms of the correlation matrices of the IS set
and the resulting variables of the CS set, after performing the PCA and the contribution procedure.
In the correlation matrices, each square represents the level of correlation of the intersecting vari-
ables, in which this level is described by a color range which goes from red (1), meaning a strong
positive correlation, to blue (-1) meaning a strong negative correlation. If the color of the square is
light green (0), it means that the variables are not correlated. A correlation level is considered as
significant if it is bigger than 0.5 or smaller than -0.5. From this discrimination algorithm, what we
are looking for is a correlation level in the range of [-0.5,0.5], which would mean that the reduced
set has no redundant information from the significant variables.
1. Ride - for this driving condition, the resulting set of important variables CS is: 8, 12, 14,
17, and 21. These variables make sense since the riding condition considers a road with
roughness, which causes various movements. Thus, to keep the trajectory, the driver has to
make small but fast steering movements. When comparing Figs. 2.2a and 2.2b, it can be seen
that the highly correlated variables were excluded in CS, 2.2b, and the remaining variables
are in the acceptable region (between ±0.4). As a result, we avoid processing redundant data
that does not provide any additional information regarding the situation.
a) Variables in the IS set b) Variables of the CS set

Variables Correlations Matrix Variables Correlations Matrix
1 1
20 0.8 0.8
0.6 21 0.6
15 0.4 0.4
Variables
Variables
0.2 17 0.2
0 0
10 −0.2 14 −0.2
−0.4 −0.4
5 −0.6 12 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 8 12 14 17 21
Variables Variables
Figure 2.2: Correlation matrices for the IS and CS sets under riding condition.
2. Road irregularity - for this situation, the resulting CS set is: 9, 10, and 17. In Fig. 2.3b it is
clear that the important variables have no significant correlation among them and that those
are the ones that describe this type of driving condition.

1 1
20 0.8 0.8
0.6 0.6
15 0.4 17 0.4
Variables
Variables
0.2 0.2
0 0
10 −0.2 −0.2
−0.4 10 −0.4
5 −0.6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 9 10 17
Variables Variables
Figure 2.3: Correlation matrices for the IS and CS sets experiencing a road irregularity condition.
3. Acceleration/Braking - here, the resulting CS set is: 5, 6, 7, 8, and 10. In this case, as
shown in Fig. 2.4b, the variables 5 and 7 are strongly correlated since they are the com-
pression rates of the front corners; the same occurs with the variables 6 and 8 which are the
rear corners. The variable 10 also has a strong correlation with the front corners’ compres-
sion rates since the convention is the same for those variables. For this driving situation, the
movements generated by an acceleration/braking maneuver are pure pitch, so there are no
other important variables. Therefore the correlation matrix for the IS set, Fig. 2.4a, does
not indicate significant correlations among the remaining variables.

1 1
20 0.8 0.8
0.6 10 0.6
15 0.4 0.4
Variables
Variables
0.2 8 0.2
0 0
10 −0.2 7 −0.2
−0.4 −0.4
5 −0.6 6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 5 6 7 8 10
Variables Variables
Figure 2.4: Correlation matrices for the IS and CS sets for acceleration and braking maneuvers.
4. Hard braking - this driving situation is similar to that of the Acceleration/Braking condition.
The CS set is: 5, 6, 7, 8 and 10, Fig. 2.5b.

1 1
20 0.8 0.8
0.6 10 0.6
15 0.4 0.4
Variables
Variables
0.2 8 0.2
0 0
10 −0.2 7 −0.2
−0.4 −0.4
5 −0.6 6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 5 6 7 8 10
Variables Variables
Figure 2.5: Correlation matrices for the IS and CS sets under a hard braking condition.
5. Cornering - for this driving situation, the resulting CS set, (see Fig. 2.6b), is: 10, 13, 18,
and 20. The cornering situation is intended to represent a constant steering maneuver going
at a constant speed. From the vehicle dynamics theory, it is clear that yaw and side slip
angle variables are important because the vehicle has no neutral steer characteristics. Also,
the pitch angle rate is important because there is a load transfer to the front of the vehicle.
In this situation, the correlation levels are in the acceptable range and it can be said that the
variables contribute essential information.
1 1
20 0.8 0.8
0.6 20 0.6
15 0.4 0.4
Variables
Variables
0.2 0.2
0 18 0
10 −0.2 −0.2
−0.4 13 −0.4
5 −0.6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 10 13 18 20
Variables Variables
Figure 2.6: Correlation matrices for the IS and CS sets under a cornering situation.
6. Rapid Steering - in this situation, the important variables are: 10, 13, 15, and 21, Fig. 2.7b.
The main difference between this driving condition and that of cornering is that here, the
important variable is the velocity of the driver action on the steering wheel. In a cornering
situation, the movement can be wide but not precisely fast. In contrast, in the rapid steering
situation the movement can be smaller and it is faster.
1 1
20 0.8 0.8
0.6 21 0.6
15 0.4 0.4
Variables
Variables
0.2 0.2
0 15 0
10 −0.2 −0.2
−0.4 13 −0.4
5 −0.6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 10 13 15 21
Variables Variables
Figure 2.7: Correlation matrices for the IS and CS sets in a rapid steering situation.
7. Loss of control - in this driving situation, the vehicle is spinning and cannot be stabilized
by the driver’s actions only. Here, the important variables are: 13, 14, 18, and 19. In this
driving condition, it is clear that the variables that mainly contribute to its classification are
the ones oriented toward the horizontal plane and toward a spinning movement.

1 1
20 0.8 0.8
0.6 20 0.6
15 0.4 0.4
Variables
Variables
0.2 0.2
0 18 0
10 −0.2 −0.2
−0.4 14 −0.4
5 −0.6 −0.6
−0.8 −0.8
−1 −1
0 5 10 15 20 13 14 18 19
Variables Variables
Figure 2.8: Correlation matrices for the IS and CS sets under riding condition.
The results obtained for the acceleration/braking and hard braking conditions are not conclu-
sive. Therefore, an additional procedure has to be done to clearly classify these situations. The
aforementioned procedure is presented in the next section.
Classification algorithm - Online analysis
After defining the M S set that contains the most representative variables for all studied driving
situations, a clustering technique is exploited as a classifier. Because of its easy computation, fast
clustering response, and good performance, the k-Nearest Neighbor (k-NN) algorithm was used.
The k-NN approach is one of the simplest machine learning algorithms which is based on the
minimum distance criterion. A data set of reference patterns in a multidimensional feature space
is required to establish the learning of this classifier, where the Euclidian distance is the most
common metric. In this case, the M S set contains the reference patterns of the studied driving
situations.
Given a new data vector x1 , x2 , . . .,xn of n variables, the k-NN algorithm assigns each new
observation to a class of driving situations Ci according to :
(2.1)
1
Pki
x ∈ Ci when minC ki j=1 d(x, yj )
where, ki is the number of nearest neighbors used in the associated classification to the driving
situation i and d(x, yj ) is the Euclidian distance between a new observation x and the nearest
neighbors of reference defined by y.
To complement the k-NN algorithm, another procedure, which only analyzes the pitch move-
ment of the pitch predominant situations, is proposed. This procedure is done using the auxiliary
classification plane presented in Fig. 2.9. In other words, the classification of the predominant
pitch movements is done by this procedure while the other situations use the k-NN algorithm.
Therefore, the seven driving situations can be classified.
This auxiliary plane is expressed in terms of the pitch movements of the vehicle under differ-
ent driving situations, and the horizontal bands mark the movement threshold of the considered
driving situations. Such thresholds were obtained by evaluating the vehicle in those situations un-
der different circumstances (velocities, road profiles, etc.). The vertical lines represent the normal
ranges of movement for those situations. At the center, in gray, the riding movements are pre-
sented; in green, the movements caused by a road irregularity are presented. The ones caused by
acceleration/braking are in yellow and those caused by the hard braking condition are in cyan.
Hard braking Emergency braking

maneuver
Acceleration and braking
Pitch [deg]
Braking
movements
Road Iregularity Bump
Riding Riding movements
Road Irregularity
Acceleration
Figure 2.9: Auxiliary classification plane for situations with predominant pitch movement.
The results of the classification algorithm during the proposed tests are presented below. Each
figure presents the target (actual driving situation) with a dashed blue line and the output of the
algorithm with a solid red line. The target was selected as a priory by the author based on the
known situations that are presented for each test.
Road profile with bump: during this test, the classifier is able to detect the ride situation, but the
irregularity in the road (bump) is not well classified. This is due to the auxiliary plane that
uses only the pitch angle to help the main classifier. After the vehicle passes the bump, the
oscillations generate some false classifications.
Target
Dc=2
Dc=1
Output
0 0.4 0.8 1.2 1.6
Time [s]
Figure 2.10: Classification algorithm result for the road profile with bump test.
Brake distance: in this test, the classifier performs a good classification of the driving situation. In
this case, the hard braking action (Dc =4) is considered only at the beginning of the braking
action, and then it becomes a normal braking (Dc =3). Thus a confusion between these
driving situations, for this particular system, is not critical since the control action for both
situations is the same.
Dc=4
Dc=3
Target Output
Dc=1
0 1 2 3 4 5
Time [s]
Figure 2.11: Classification algorithm result for the brake distance test.
DLC test: this test is used to emulate a situation in which the driver has to make an abrupt line
change and then return back. Here the classifier obtains minimum errors.
Dc=6
Target
Output
Dc=1
0 2 4 6
Time [s]
Figure 2.12: Classification algorithm result for the DLC test.
FH test: in this test, the driver executes a sustained cornering maneuver at a constant speed. Here
the classifier also performs well, with only a few missed classifications.
Dc=5
Target
Output
Dc=1
0 2 4 6 8 10
Time [s]
Figure 2.13: Classification algorithm result for the FH test.

Split µ-surface braking: this test has two options for target situations. First, if the vehicle has
any VDC, the target becomes a loss of control situation, Fig. 2.14; second, if the vehicle is
equipped with braking controls, the target situation becomes a braking situation, Fig. 2.15,
since the braking control maintains the stability of the vehicle. For the first case, the classifier
succeeds to classify a loss of control (Dc=7) situation but with many false classifications of
a riding situation (Dc=1). For the second case, the classification shows a good performance,
identifying the braking situation (Dc =3).
Dc=7 Target Output
Dc=1
0 1 2 3 4 5 6 7
Time [s]
Figure 2.14: Classification algorithm result for the split µ-surface braking test for the uncontrolled
case.
Dc=3
Target Output
Dc=1
0 1 2 3 4 5 6 7
Time [s]
Figure 2.15: Classification algorithm result for the split µ-surface braking test for the controlled
case.
To summarize the results, the ROC curve for the classifier algorithm is presented in Fig. 2.16.
This curve shows the probability of classification in the vertical axis and the probability of false
alarms in the horizontal axis. The area above the main diagonal (gray) line indicates the acceptable
classification region. The area from the main diagonal line and below indicates no discrimination
skill from the classifier. Ideally, a perfect classifier is located at the top left corner with a probability
of 1 for detecting a true positive, and with a probability of 0 for giving a false alarm. This plot
presents a curve for each driving situation considering the classifier.
The proposed classifier has a good performance for most of the driving conditions, but not for
the loss of control condition.
ROC Curve
1
Dc=1
True Positive Rate

0.8 Dc=2
0.6 Dc=3
Dc=4
0.4 Dc=5
Dc=6
0.2
Dc=7
0
0 0.2 0.4 0.6 0.8 1
False Positive Rate
Figure 2.16: ROC curve for the classifier algorithm.
Decision logic
In this section, a more precise justification of the coordination scheme is given.

The decision logic module uses the Dc vector to execute the coordination of the subsystems.
This vector is formed by three elements [si , ss , au ]:
• si indicates the critical level of the situation. This element depends on the number of driving
situations considered in the GCC system, and the value assigned to each situation depends
on the user criteria.
• ss refers to the characteristic of the situation; that is, whether the situation is critical or not.
In this case, there are only two situations that are considered as critical (rapid steering and
loss of control), and the restriction is grouping the normal and critical situations together.
• au is the actuation vector. This vector contains the mode for operations of each subsystem.
The values of au are based on the following modes of operation:
– Suspension: The coordination of the suspension system is carried out by means of the
Suspension Adjustment Plane (SAP). This plane was designed to improve the road hold-
ing characteristics of the tires, where most of the load is transfered according to each
movement, [Tudon-Martinez et al., 2013]. To include these effects, two movements of
the vehicle are considered: pitch and roll. Fig. 2.17 shows the different combinations
of controllers’ orientations according to these movements. There are two special cases
in which the SAP plane is not considered: 1) in riding condition, the objective is to
preserve the comfort, and thus all corners of the suspension system are commanded to
use the comfort-oriented controller (Sky-Hook), and 2) in a loss of control situation,
the main objective is to allow the driver to still control the vehicle. In this case, the
suspension is set to a full road-holding mode to help the other systems.
aususp=[1,1,1,0] aususp=[1,1,0,1]
aususp=[1,1,0,0]
aususp=[1,0,1,0] aususp=[0,1,0,1]
Ride
aususp=[0,0,1,1] Limits
Figure 2.17: Suspension Adjustment Plane (SAP), showing the orientation of each suspension
corner where “0” means comfort orientation and “1” means road-holding orientation.
– Steering: The implemented AFS controller is oriented to reduce the yaw angle rate,
[Krishna et al., 2014]. This characteristic is only desired in situations where the vehicle
is subjected to steering wheel commands (cornering and rapid steering). Thus, the
action of this controller is penalized in driving situations where there is no steering
action or where this action is oriented to keep the vehicle in line, [Cho et al., 2012].
– Braking: This ESC controller, [Boada et al., 2005], aims to reduce the yaw angle rate
and the side slip angle to be as close to zero as possible. In other words, the objective
of this controller is to bring the vehicle back to a situation in which the driver can
still control it. These types of controllers are based on the braking system, but in this
implementation only the rear tires are active. Indeed, by braking only the rear tires, the
velocity of the vehicle is less affected and it still generates a good stabilizing moment,
[Doumiati et al., 2013].
The values of the terms of Dc , with the exception of ss , were proposed specifically for each
situation regardless of the number of situations considered. This characteristic allows for the modi-
fication of the number of considered driving situations to be carried out without major implications.
Based on the aforementioned mode of operations for each subsystem, Table 2.2 was proposed.
This table presents the proposed values for vector DC during each driving situation.
2.3.2 Control layer

In terms of the control characteristics, two features must be discussed: the design modularity and
the control flexibility.
Table 2.2: DC vectors for different driving conditions.

DC
Driving situation si ss aususp austeer aubraking
Ride 1 0 [0,0,0,0] 0 0
Road irregularity 2 0 SAP 0 0
Acceleration/Braking 3 0 SAP 0 0
Hard Braking 4 0 SAP 0 0
Cornering 5 0 SAP 1 0
Rapid Steering 6 1 SAP 1 1
Loss of control 7 1 [1,1,1,1] 1 1
Design modularity evaluation
This evaluation consists of comparing the current proposal with a variety of configurations chang-
ing the subsystems involved in the coordination scheme. The purpose of this evaluation is, on the
one hand, to demonstrate the effects of different configurations of subsystems, and on the other
hand to validate the modularity characteristic of the GCC system. This means that the GCC sys-
tem can be easily modified if the subsystems change. For this evaluation, the considered options
are:
1. Full GCC system: This is the original proposal which considers the three subsystems SAS,
AFS, and 4-WIB.
2. O1 (SAS + AFS) : This option considers only a vehicle with the SAS and AFS systems.
3. O2 (SAS): This configuration only considers the SAS subsystem. This can be considered as
a vehicle without ESC and AFS control systems.
4. O3 (SAS + 4-WIB): This option represents a vehicle equipped with the SAS and ESC sys-
tems.
5. O4 (AFS + 4-WIB): This configuration can be considered as a vehicle featuring AFS and
ESC control systems.
The most suitable test to evaluate the behavior of the different configurations is the Double
Line Change maneuver because in this test, the three subsystems contribute and any change can
affect the performance significantly. The first thing to evaluate in this test is the covered trajectory
by the vehicle. The objective is that the vehicle makes a change and goes back; as a reference, the
objective trajectory (dashed gray line) is shown in Fig. 2.18.
From Fig. 2.18, the first thing that can be noticed is that two groups appear: the first one
includes the GCC system and options O1 and O4 , and in the other group are O2 and O3 ; the
common denominator in these groups is the presence or absence of the AFS subsystem. This
difference is due to the fact that most of the control contribution is found in the steering system.
At first glance, it appears that the group without the AFS tends to follow the desired path better,
but it actually goes out of the trajectory; at the end of the test, the ending path appears as a loss of
control.
y-coordinate [m]
Reference O2
3 O3
path
2
1
O1 O4
0
GCC
−1
0 50 100 150 200
x-coordinate [m]
Figure 2.18: Vehicle trajectories for different configurations of the coordination scheme under a
DLC test.
Then, two variables must be analyzed: the roll motion, Fig. 2.19a, and the yaw angle, Fig.
2.19b. These variables describe the movements of the vehicle in terms of the vertical and lateral
dynamic planes. Analyzing the roll movement of the vehicle during the test, in Fig. 2.19a, one
sees that the configuration O1 is less affected in terms of the roll movement; it has a performance
decrement of 3 %, compared to other configurations O3 and O4 where the effect is 13.1 % and 13.9
% of reduction respectively, whereas the configuration O2 presents the worst performance at 25 %
less than the original GCC system. Due to the load transfer caused by the steering movements,
the roll angle can be affected dramatically even with a SAS system; in this case, the AFS system
compensates the aggressive steering maneuvers reducing the roll movements.
Regarding the yaw angle, Fig. 2.19b, the configuration with the worst performance is again
the O2 with a reduction of 30.5 %. By means of only the suspension, the reduction that can be
experienced is marginal. The configurations O1 and O4 present a reduction of 13.47 % and 5.9
% respectively. In this case, the effectiveness of the AFS is demonstrated again by reducing the
undesired movements through aggressive steering actions. The configuration O3 has a loss in
performance of 19.3 %, also caused by the lack of the AFS control system.
Control flexibility evaluation
For the control flexibility evaluation, the proposed GCC system is compared with a variation. It
consists of changing the suspension controller from being heuristic to be a H∞ controller. This
controller was designed to minimize the vertical acceleration as well as the pitch and roll move-
ments of the sprung mass. Figure 2.20 shows the controller structure.
a) Roll angle (θ) b) Yaw angle (ψ)

1 O1 4 O3 GCC
Roll [deg]
Yaw [deg]
0 0
GCC
−1 O3 O1
−4
−2 O2 O4 O2 O4
−8
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time [s] Time [s]
Figure 2.19: Results for different configurations of the coordination scheme under a DLC test.
P
Wzs z1
Wzr Wθ z2
 zrfl zrfr zrrl zrrr 
fl
Wzr  zs θ φ zɺs θɺ φɺ  Wφ z3
fr
Wzr Σ Wzɺs z4
rl
Wzr Wθɺ z5
rr
Wφɺ z6
K susp
[ufl ufr url urr ]  ɺɺ
zsfl ɺɺ
zsfr ɺɺ
zsrl ɺɺ
zsrr ɺɺ
zusfl ɺɺ
zusfr ɺɺ zusrr 
zusfl ɺɺ
Figure 2.20: H∞ controller structure.
This is based on the 7 DoF model of the vertical dynamics of a full vehicle presented in [Zin
et al., 2009]. The state-space representation of the augmented system is:

 ẋ(t) = Ax(t) + Bu u(t) + Br w(t)

P : z(t) = C1 x(t) + Du11 u(t) + Dr12 w(t) (2.2)

y(t) = C2 x(t) + Du21 u(t) + Dr22 w(t)

where the states vector is defined as:

h iT
x = zs , zusfl , zusfr , zusrl , zusrr , θ, φ, żs , żusfl , żusfr , żusrl , żusrr , θ̇, φ̇
the vector of measured variables is defined as:
y = [z̈sfl , z̈sfr , z̈srl , z̈srr , z̈usfl , z̈usfr , z̈usrl , z̈usrr ]T
the vector of controlled outputs is:

h iT
z = Ws · zs , Wθ · θ, Wφ · φ, Wżs · żs , Wθ̇ · θ̇, Wφ̇ · φ̇
the road profile input vector is:

T
w = Wzrfl · zrfl , Wzrfr · zrfr , Wzrrl · zrrl , Wzrrr · zrrr
and the controller manipulations vector is:
u = [ufl , ufr , url , urr ]T
The weighting functions for the inputs are:

w zr s
Wzri,j = ∀ i ∈ [f, r], j ∈ [l, r]
s+1
and the whetting functions for the controlled outputs are:
2 2 2
mzs s + wzs mθ s + wθ mφ s + wφ
Wzs = , Wθ = , Wφ =
s + wzs · zs s + wθ · θ s + wφ · φ
2 2 2
mzs1 s + wzs1 mθ1 s + wθ1 mφ1 s + wφ1
Wżs1 = , Wθ̇ = , Wφ̇ =
s + wzs1 · zs1 s + wθ1 · θ1 s + wφ1 · φ1
The H∞ controller is synthesized using the Riccati solution with a performance criteria of:
||Tew (s)||∞ ≤ γ, where γ is considered as 0.01.
For this evaluation, two tests are used: 1) road profile with bump, to evaluate the comfort
capabilities of the controllers, and 2) DLC test to evaluate the road holding characteristics of the
GCC system with the two suspension controllers: 1) heuristic and 2) H∞ .
• Road profile with bump test: This test was designed to evaluate the different suspension
controllers approaches. In this case, the GCC (H∞ ) obtained an improvement in the vertical
acceleration of 5.95 % when compared to the GCC (Heuristic), see Fig. 2.21a. In terms of
the pitch movement, the GCC (H∞ ) has 9.8 % more oscillations than the GCC (Heuristic)
system, Fig. 2.21b.
a) Vertical acceleration (z̈s ) b) Pitch angle (φ)

0.8
Pitch Angle [deg]
0.8
Acceleration [G’s]
GCC (H∞)
0.4
0.4
Vertical
0 0
−0.4 −0.4
GCC (Heuristic) GCC (Heuristic)
GCC (H∞) −0.8
−0.8
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Time [s] Time [s]
Figure 2.21: Results for the GCC(Heuristic) and GCC(H∞ ) for the road profile with bump test.
• DLC test: The DLC test evaluates how well a vehicle can react to a vigorous steering action.
The first variable of interest is the trajectory of the vehicle. In this case, the control system
has two goals: first, to maintain control of the vehicle and, second, to improve the vehicle
response (maneuverability). This means that the control system should improve the handling.
In Fig. 2.22 the trajectory of the proposed GCC system is close to the reference path, whereas
the GCC(H∞ ) configuration is not bad but its trajectory is not as close as the proposed
control system.
y-coordinate [m]
3 Reference GCC (H∞)
2 path
1
0
GCC (Heuristic)
−1
0 50 100 150 200
x-coordinate [m]
Figure 2.22: Vehicle trajectory for a DLC test.
The other important variables for this test are the roll and the yaw rates. These two variables
are indicators of how much vertical and horizontal movements the vehicle experienced. For
the roll angle, in Fig. 2.23a, the proposed GCC system has smoother behavior, whereas the
GCC(H∞ ) approach has rougher behavior; these differences are translated into a deteriora-
tion of 2 % with respect to the proposed GCC system. For the yaw angle, in Fig. 2.23b,
the behavior is more or less the same in the two cases, but those differences cause 2 % of
deterioration in the performance of the GCC(H∞ ) approach compared to the one obtained
with the GCC(Heuristic).
a) Roll angle (θ) b) Yaw angle (ψ)

1 GCC (Heuristic) 4 GCC (Heuristic)
Yaw [deg]
Roll [deg]
0 0
−1 −4
−2 GCC (H∞) GCC (H∞)
−8
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time [s] Time [s]
Figure 2.23: Results for the GCC(Heuristic) and GCC(H∞ ) for the DLC test.
2.3.3 Evaluation conclusions

Decision layer
A large part of the contribution of this thesis lays in this layer, specifically on the novel approach to
identify driving situations. From this part, two contributions were obtained based on the previous
evaluation:
• Discrimination algorithm (offline analysis): The approach to characterize a driving situation,

based on data-processing tools such as PCA and contribution plots, not only proves its effec-
tiveness to detect the important variables for each driving situation, but also the algorithm
succeeds to obtain a set of variables that offers important information about the situation
without selecting redundant variables.
• Classification algorithm (online module): The advantage of selecting a simple classifica-

tion algorithm (in terms of computational complexity) gives the opportunity of being imple-
mented in a common ECU. Also, the limitations that come along with the selected system (a
road vehicle) can be easily tackled by adding auxiliary tools in a straightforward algorithm
that is able to detect seven driving situations with good accuracy.
Control layer
Regarding the control layer, two main conclusions were derived from this evaluation:
• Modularity: This characteristic offers the possibility of modifying the considered subsys-
tems according to the type of vehicle and the available infrastructure. The proposed GCC
system considers three subsystems, but the same concepts and structure can be used for more
subsystems and, as demonstrated previously, with fewer subsystems. Additionally, the re-
configuration of the control system needs to be executed in the coordination scheme. It is
important to take into account that the coordination scheme and modes of operation must
be proposed individually for each subsystem, but also consider the interactions with other
subsystems.
• Flexibility: For this proposal, flexibility presents a main advantage. Regarding the type of
actuators used, the control technique and framework can be modified without significantly
changing the overall structure of the GCC system. Also the coordination scheme was pro-
posed, considering three elements (ss , si , and au ) that can be used as scheduling parameters,
for gain scheduling, or LPV approaches.
Chapter 3
Method for Modeling Electrorheological

Dampers Using Its Dynamic Characteristics

Among the physical components that affect the dynamic behavior of the vehicle, the shock ab-
sorbers are one of the most difficult to model. This chapter presents an article which deals with
this issue, proposing a modeling method for Electro-Rheological (ER)dampers.
The modeling method includes two simple steps: characterization and model customization.
These steps are based on experimental data about damper behavior. Experiments were designed
to explore the nonlinear behavior of the damper at different frequencies and actuation signals (i.e.
automotive domain). The resulting model has low computational complexity. To evaluate the
method a commercial ER damper was used. As performance index, the Error to Signal Ratio
(ESR) was used to evaluate the model accuracy.
The results were quantitatively compared with two well-known ER damper models: the Choi
parametric model and the Eyring-plastic model. The new proposed model has a 44 % better ESR
index than the Choi parametric model and 28 % for the Eyring-plastic model. A qualitative com-
parison based on density plots highlights the advantages of this proposal.
55
Hindawi Publishing Corporation
Mathematical Problems in Engineering
Volume 2015, Article ID 905731, 15 pages
http://dx.doi.org/10.1155/2015/905731
Research Article
Method for Modeling Electrorheological Dampers Using
Its Dynamic Characteristics
Carlos A. Vivas-Lopez, Diana Hernández-Alcantara, Ruben Morales-Menendez,

Ricardo A. Ramírez-Mendoza, and Horacio Ahuett-Garza
ITESM Campus Monterrey, Avenue Eugenio Garza Sada 2501, Col. Tecnológico, 64849 Monterrey, NL, Mexico
Correspondence should be addressed to Carlos A. Vivas-Lopez; a00794204@itesm.mx
Received 6 June 2014; Accepted 30 August 2014
Academic Editor: Xingsheng Gu
Copyright © 2015 Carlos A. Vivas-Lopez et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.
A method for modeling an Electrorheological (ER) damper is proposed. The modeling method comprehends two simple steps:
characterization and model customization. These steps are based on the experimental data of the damper behavior. Experiments
were designed to explore the nonlinear behavior of the damper at different frequencies and actuation signals (i.e., automotive
domain). The resulting model has low computational complexity. The method was experimentally validated with a commercial
damper. The error-to-signal Ratio (ESR) performance index was used to evaluate the model accuracy. The results were quantitatively
compared with two well-known ER damper models: the Choi parametric model and the Eyring-plastic model. The new proposed
model has a 44% better ESR index than the Choi parametric model and 28% for the Eyring-plastic model. A qualitative comparison
based on density plots highlights the advantages of this proposal.
1. Introduction An accurate mathematical model to predict the nonlinear

dynamic behavior of the ER damper is needed in order to
In an automotive suspension system the shock absorber get a better control of the SA suspension system. There are
has the purpose of dissipating the energy of the motion of several contributions in this topic [2, 3]. However, most of
the vehicle caused by the road disturbances. This energy them are highly dependent on internal physical properties of
dissipation allows the suspension to achieve two important the damper (usually confidential information), demand too
objectives: decrease the vertical acceleration and maintain much computational effort, or fail to capture the nonlinear
the tires in contact with the ground. Passive suspension behavior of the ER damper.
systems are tailored to achieve a tradeoff of these objectives A new method to model an ER damper is proposed. The
[1]. Semiactive (SA) suspension systems use a particular type method comprehends two main steps: (1) a characterization
of shock absorber which is capable of online modifying procedure where the dynamical response of the damper is
the amount of energy that can dissipate. This change on analyzed and (2) a model customization where a general
the damper needs to be controlled, to achieve the desired model is tailored. This method requires experimental data
objectives. of the ER damper. The resulting model is light enough to
The electrorheological (ER) damper is a hydraulic device, be implemented in an embedded system. The method is
which is filled with a mixture of low viscosity oil and particles validated with intensive experimental data and compared to
that are sensitive to an electric field. The ER fluid, when others published.
exposed to the electric field, behaves as a viscoelastic material, This paper is organized as follows: in Section 2 a bib-
known as a Bingham plastic. This means that ideally it behaves liographic review of ER damper modeling is presented.
as a solid at low stress efforts, but it flows as a viscous fluid In Section 3 the experimental system and the Design of
when this force reaches its yield stress. Furthermore, the yield Experiments (DoE) are shown. Section 4 describes the pro-
stress is field dependent; it increases as the electric field does. posed method. Section 5 presents the modeling procedure.
Section 6 shows the results and evaluates the performance of but it does not seem to be very accurate; also it needs a set
the customized model. Finally, Section 7 concludes the paper. of constants for every field manipulation interval. Another
approach is the Eyring model [12] which uses an Arcsinh
function with shape parameters that depends on the electric
2. ER Damper Models: State of the Art field intensity and the frequency. This model can represent the
There are many mathematical models to reproduce the behavior in both the preyield and the postyield zone but needs
characteristic behavior of the ER damper. The major efforts the identification of every parameter in each combination
have been aimed at parametric models. [4] proposes a model of frequency and field intensity; the accuracy of the model
based on the viscoelastic characteristics of the ER damper; depends on how small are the considered intervals of the
this model contemplates a linear passive damping force that variables, but when changing the between this levels the
depends on the piston velocity and a lineal SA effect that only model does not consider a transient response of the force.
depends on the electric field applied to the ER fluid, this is [13] introduces a neurofuzzy training algorithm to model the
a simple model that does not contemplate hysteresis and the force of the ER damper, using the values of the acceleration
damping force is completely linear. and velocity of the damper; this model captures the nonlinear
A model based on the pressure drop in the ER channel behavior of the ER damper with high level of accuracy, but the
is presented by [5]; this pressure drop considers an effect evaluation was done under very limited conditions.
that depends on the damper velocity and others that just Most of the models are dependent on internal physical
depend on the electric field; the coefficients for this model properties of the damper, ER fluid, and its design; this makes
are based on physical dimensions of the damper and physical the implementation of these models very restricted (i.e.,
properties of the ER fluid. [6] proposes a model using a confidential information). Our proposal considers general
bond graph to model de governing equation of motion of model that is customized based only on experimental data
the damper, but the physical dimensions and properties of of the ER damper. The experimental system and Design of
the damper and ER fluid are needed. Later [7] shows two Experiments are shown in the next section.
different types of ER damper configurations. The first one is
the most common, the cylindrical type, in which the ER fluid
flows through an annular channel where the electric field is 3. Design of Experiments
applied. The second one is the orifice type; this type has a
A commercial ER damper was used, Figure 1(a). The damper
mechanism located inside the piston of the damper, which
has a stroke of ±150 mm and a force range of [−2500, 4500] N.
regulates the flow of the ER fluid through its chambers; two
The damper is actuated by a 2 module which is controlled by
models were proposed, one for each type of damper, but the
a 25 kHz pulse-width modulated (PWM) signal. The PWM
ones of physical parameters are needed.
Following the same line in terms of parametric models, duty cycle range was 10–80%. Since the ER damper needs
[8] describes a hydromechanical based model. This model to be operated with a voltage signal of 0–5 kV the 2 module
divides the damper in different zones where the pressure proportionally transforms the duty cycle of PWM signal to
drop is calculated and the damper force depends on those voltage. Figure 1(b) shows the characteristic force-velocity
pressure drops. This model captures the damper behavior in (FV) diagrams at different PWM duty cycles.
the preyield zone, in terms of the hysteresis. The authors do The experimental setup, Figure 2(a), consists of three
not evaluate the effects of the frequency in the model and the modules: (1) the acquisition system, which registers the dis-
transient behavior of the force during changes in intensity of placement, velocity, damper force, and PWM signals using a
the electric field, which is important for control purposes. [9] National Instruments (NI) cDAC; (2) the hydraulic actuation
presents a parametric dynamic model in which the pressure system which consists of a piston that is actuated by a MTS
drops in the annular duct are calculated with respect to 407 controller; and (3) the control system, which is used as
time. This model represents the hysteretic behavior of the an interface (control panel programmed in NI LabView) to
ER damper in postyield zone and its increment due to the operate the system.
frequency, but the assessment of the model is done with con- A series of displacement sequences and actuation signals
stant conditions of frequency displacement and electric field. were used to capture the static and dynamic relations between
Another model is based on a lumped parameter method, in velocity, displacement, actuation signal, and the damper
which the sections of the ER damper (upper chamber, lower force [14]. These sequences ensure the ER damper will be
chamber, annular duct, and connecting pipe) are divided tested in the automotive domain. The sequences used for the
into lumps and modeled with differential equations. This displacement of the piston are road profile (RP), Figure 2(b),
model predicts the nonlinear behavior of the ER damper in [15], and decreasing-amplitude stepped frequency sinusoidal
the preyield and postyield zones but depends on physical (DSFS) signal, Figure 2(c), [16].
properties of the damper and it is sensitive to the initial The RP sequence is used to test the ER damper under
conditions, [10]. standard automotive conditions and represents the motion
Regarding the nonparametric models, [11] presents a in a vehicle suspension when the car is driven through a
polynomial equation with only three constants that can be specific surface. The RP smooth highway is the most common
fitted by least square estimation (LSE) methods; the force road for commercial vehicles. On the other hand, the DSFS
of the damper only depends on the velocity of the piston. signal is used to analyze the transient response and the hys-
The advantage of this model is the few number of constants teresis loops when changes in magnitude and frequency are
4000
Damper force (N)

2000
−2000
0.8
PW 0.6 00.11
M
du 0.4 0
ty
cyc ( m /s)
le 0.2 −0.1 city
Velo
(a) (b) Characteristic FV diagrams
SOBEN
ER
damper
Figure 1: ER damper and its force-velocity diagrams.
MTS 407 Position

controller sensor Hydraulic
Force piston
Displacement
Acquisition Displacement command

system ER
damper
PWM duty cycle Velocity 2
Displ (mm)
Velocity sensor 0
AI Voltage −2
Operation/ module PWM Carcon 2 0 10 20 30
monitor module module Time (s)
(a) Experimental system (b) RP
2
Displ. (mm)
−2
0 5 10 15 20 25
Time (s)
(c) DSFS
Figure 2: Experimental system and displacement sequences (RP and DSFS).
present. The bandwidth of the DSFS displacement sequence analyze the damper transient response under actuation signal
is [0.5, 14.5] Hz wherein lie comfort and road holding speci- variations. The ICPS is a signal with random amplitude varia-
fications. In addition, DSFS sequence has a similar frequency tions, whereas the PRBS is a signal whose amplitude switches
spectrum experienced by automotive suspension systems. between two constant values with a random frequency.
For the PWM duty cycle, the Stepped inCrements (SC) The DoE consists of a combination of displacement
signal, Figure 3(a), is used to study the effect of the actuation and actuation sequences (i.e., 5 experiments). Experiment 1
signal under different displacements sequences. Increased (E1 ) for characterization purposes uses the DSFS sequence
clock period signal (ICPS), Figure 3(b), and pseudorandom repeated at each of the steps of the SC signal so that the
binary signal (PRBS), Figure 3(c), sequences are used to damper could be excited in a wide range of displacements
Table 1: Design of experiments.
Displacement
Experiment Actuation signal Purpose
Sequence Amp. [mm] Frequency [Hz]
E1 DSFS ±1–±8 [0.5–14.5] SC Characterization
E2 RP ±8 [0–3] PRBS Identification
E3 RP ±8 [0–3] ICPS Identification
E4 DSFS ±1–±8 [0.5–14.5] ICPS Identification
E5 DSFS ±1–±8 [0.5–14.5] PRBS Identification
1 1
0.8 0.8
PWM duty cycle
PWM duty cycle

0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 50 100 150 200 250 0 5 10 15
Time (s) Time (s)
(a) SC (b) ICPS
1
0.8
PWM duty cycle
0.6
0.4
0.2
0
0 5 10 15 20 25 30
Time (s)
(c) PRBS
Figure 3: Actuation signals.
and forces. Experiments 2–5 (E2 , . . . , E5 ) for identification the damper, Figure 4. In the DoE step, the experiments
purposes are a combination of RP-DSFS sequences and are defined on the automotive range of operation. In the
PRBS-ICPS signals; while the DSFS explores a wide range characterization step, the dynamical characteristics of the
of movement the RP only explores a limited one. For the damper behavior are analyzed based on the characteristic
actuation signal the use of a PRBS signal shows how the force-displacement (FD) and force-velocity (FV) diagrams.
damper behaves when operated at its limit conditions; for the Then the general model is customized. In the identification
case of ICPS the full range of force was shown. Three replicas step the model fits the 60% of the experimental data. In the
were implemented at a sampling frequency of 512 Hz, Table 1. validation step, the model is tested with the remaining 40%
of the experimental data using the error-to-signal ratio (ESR)
performance index.
4. Proposed Method
4.1. Characterization. The ER damper force can be repre-
The proposed method does not need a priori knowledge sented by two components: a passive component, which is
(i.e., physical properties, dimensional information, etc.) of present for all voltage input values, and a SA component,
Start End
Yes
1 2
Yes
DoE Characterization Customization Identification Convergence Validation Convergence
criteria criteria
No No
Still no
Figure 4: Proposed modeling method.
Viscous damping
600 600
400 400 Internal stiffness

Friction
Friction force behavior
Force (N)
Force (N)
200 200
Friction
0 0 displacement
Friction
−200 −200
velocity
−400 −400
−1000 −500 0 500 1000 −20 −15 −10 −5 0
Velocity (mm/s) Displacement (mm)
(a) Passive FV diagram (b) Passive FD diagram
600
400 Postyield region

(velocity independent)
200 Field increment
Force (N)
Preyield region effect

(velocity dependent)
0
−200 Yield
velocity
−400
−1000 −500 0 500 1000
Velocity (mm/s)
(c) SA FV diagram
Figure 5: Characteristics diagrams of a semiactive damper.
which depends on the actuation input [5]. The SA force and 5(b), are analyzed and the following characteristics can
component depends on the actuation signal as be identified:
𝐹SA (𝑉) = 𝐹𝐷 (𝑉) − 𝐹𝑃 , (1) (i) hysteresis: the observed hysteresis is considered sig-
nificant, based on the dispersion of the force measure-
where 𝐹SA is the added force due to the manipulation signal ments (wider plot forms);
𝑉 to de damper force 𝐹𝐷 if a voltage 𝑉 is applied and 𝐹𝑃 is the (ii) static friction: the passive force presents a drastic
measured damper force in an experiment with zero or min- change of force at low velocities
imum manipulation. The FV and FD diagrams are analyzed
to find the characteristics that define the damper, Figure 5. (iii) viscous damping: it is the constant force gain that
The passive FV and FD experimental diagrams, Figures 5(a) depends of the velocity
1000 Stick-slip 1000 Internal Abnormal

stiffness stick-slip
Viscous
500 Friction 500 Friction
damping
Force (N)
Force (N)
0 Viscous Hysteresis 0
damping
Frequency
Stick-slip
−500 −500 increment effect Stiffness
−0.1 0 0.1 −25 −20 −15 −10 −5 0
Velocity (m/s) Displacement (m) ×10−3
(a) Passive FD diagram (b) Passive FV diagram
Figure 6: Characteristic diagrams of the ER damper passive behavior.
Table 2: Model terms used to represent the ER damper characteris- applied field, the sequence at 10% of duty cycle PWM will be
tics. considered passive.
Characteristic Observed diagram Model term
Passive Behavior. Figure 6 shows the significant effects that
Viscous damping Passive FV 𝑐𝑝 ż
are present in this ER damper operating in passive mode.
Stiffness Passive FD 𝑘𝑝 z
From the FV diagram, Figure 6(a), it can be seen that this
Friction Passive FV 𝑓fr ER damper is asymmetrical; the maximum force in extension
Hysteresis loop Passive FV 𝑓ℎ,𝑧 (positive velocity) is greater than the force generated in
Frequency dependent hysteresis Passive FV 𝑓ℎ,𝑧̇ , 𝑚𝐷 compression (negative velocity). The force has a velocity
Preyield zone SA FV 𝑓pre-𝑦,𝑧̇ dependent component, which is also different in extension
Gain in force due to manipulation SA FV 𝑐SA and compression. This damper presents significant hysteresis
in all its operational range, being more notorious at low speed
and in positive velocity. At low speed a friction component
(iv) stiffness: this characteristic is related to how much (∼700 N) can be observed. This ER damper is subjected to
the gas in the accumulator can be compressed after the stick-slip phenomenon, especially in positive velocity;
defriction force has yielded. according to [5] this phenomenon appears in the ER damper
as a force overshot when the flow changes its direction in the
Afterwards the SA diagram, Figure 5(c), is analyzed
annular duct. In the FD diagram, Figure 6(b), an abnormal
using (1). The SA phenomena include preyield and postyield
stick-slip appears as a peak, as well as effect of the frequency
regions and hysteresis. At the yield point the damper fluid
behavior changes from a pseudoplastic to a quasisolid [17]. In in the damper stiffness.
the FV diagram the yield point is a Cartesian point where the
Semiactive Behavior. The behavior of the SA component
damping force becomes independent of the velocity. The yield
of the force is presented in Figure 7. The relation between
point defines where the SA damper operates: in preyield zone
or in postyield zone. Also, at the postyield zone, an average the SA force and the PWM duty cycle becomes evident;
force gain (FM) is obtained, based on the average value in this relationship is asymmetrical, Figures 7(a) and 7(b). In
which the yield of the force occurs at each manipulation the postyield region the force is almost independent of
value. the piston velocity, but in the preyield zone the force is
The characterization procedure was applied to the velocity dependent. Also at low speed the hysteresis loop
damper, using experiment E1 , Table 2. E1 was selected in SA force is not significant, but as the velocity and the
because the DSFS displacement excitation signal explores all PWM duty cycle rise, the hysteresis rises too, Figure 7(a).
the realistic operational range of an automotive damper, in In Figure 7(b) the stiffness of the damper is affected when
terms of frequency and displacement, and the SC actuation the frequency is incremented; also it is notorious how the
signal causes the damper to show its entire force range. stick-slip phenomenon became greater as the manipulation
E1 allows observing the same dynamical effects of the ER increases. The average FM diagram, Figure 7(c), shows that
damper caused by the excitation signal at different levels of the average force gain for this particular ER damper has a
actuation. Since this damper cannot be operated with no linear behavior.
4000 4000
3000 3000
2000 2000
Force (N)
Force (N)
1000 1000
0 0
−1000 −1000
−2000 −2000
−3000 −3000
−0.1 0 0.1 −0.02 −0.01 0
Velocity (m/s) Displacement (m)
(a) SA FD diagram (b) SA FV diagram
3500
3000
2500
Force (N)
2000
1500
1000
500
0
0 0.2 0.4 0.6 0.8
PWM duty cycle
(c) SA average FM diagram
Figure 7: Characteristic diagrams of the ER damper SA behavior.
4.2. Customization. Equations (2), (3), and (4) are a general Vℎ,𝑧̈ 𝑧̇ + 𝑥ℎ,𝑧̈ sign (𝑧)̈
𝑓ℎ,𝑧̈ = 𝑓ℎ,𝑧̈ ( 󵄨 󵄨) ,
SA model which includes almost all the phenomena observed 1 + 󵄨󵄨󵄨Vℎ,𝑧̈ 𝑧̇ + 𝑥ℎ,𝑧̈ sign (𝑧)̈ 󵄨󵄨󵄨
in SA dampers. Consider
Vpre-𝑦,𝑧,𝐼̇ 𝑧𝑉 ̇
𝐹𝐷 (𝑉) = 𝐹𝑃 + 𝐹SA (𝑉) , ̇ =(
𝑓pre-𝑦,𝑧,𝑉 󵄨󵄨 󵄨) ,
̇ 󵄨󵄨󵄨󵄨
(2)
1 + 󵄨󵄨󵄨Vpre-𝑦,𝑧,𝐼̇ 𝑧𝑉
where
𝑥pre-𝑦,𝑧 𝑧
𝑓pre-𝑦,𝑧 = ( 󵄨 󵄨) .
𝐹𝑃 = 𝑓0 + 𝑐𝑝 𝑧̇ + 𝑘𝑝 𝑧 + 𝑚𝐷𝑧̈ + 𝑓fr + 𝑓ℎ,𝑧 + 𝑓ℎ,𝑧̈ , (3) 1 + 󵄨󵄨󵄨󵄨𝑥pre-𝑦,𝑧 𝑧󵄨󵄨󵄨󵄨
𝐹SA (𝑉) = 𝑉𝑐SA [𝑓pre-𝑦,𝑧,𝑉 (5)

̇ + 𝑓pre-𝑦,𝑧 ] (4)
Equation (3) describes the passive force (𝐹𝑃 ). The passive
with force component 𝑓0 is an initial compensation force gener-
V𝑓 𝑧̇ + 𝑥𝑓 𝑧 ated by the accumulator, 𝑐𝑝 is the viscous damping coefficient
𝑓fr = 𝑓𝑓 ( 󵄨 󵄨) , which describes the linear viscous damping of the Newtonian
1 + 󵄨󵄨󵄨󵄨V𝑓 𝑧̇ + 𝑥𝑓 𝑧󵄨󵄨󵄨󵄨 fluids, 𝑘𝑝 is the stiffness coefficient which is the characteristic
of linear elastomers, 𝑚𝐷 is the virtual damper mass, 𝑓fr is
Vℎ,𝑧 𝑧̇ + 𝑥ℎ,𝑧 sign (𝑧) the damping force due to friction, and 𝑓ℎ,𝑧 , 𝑓ℎ,𝑧̇ model the
𝑓ℎ,𝑧 = 𝑓ℎ,𝑧 ( 󵄨 󵄨) ,
1 + 󵄨󵄨󵄨Vℎ,𝑧 𝑧̇ + 𝑥ℎ,𝑧 sign (𝑧)󵄨󵄨󵄨 hysteresis [18–20]. Equation (4) represents the SA force 𝐹SA ,
3000 3000
2000 2000
1000 1000
Force (N)
Force (N)
0 0
−1000 −1000
−2000 −2000
−0.2 −0.1 0 0.1 0.2 −0.01 −0.005 0 0.005 0.01
Velocity (m/s) Displacement (m)
(a) FV diagram (b) FD diagram
3000 3000
2000 2000
1000 1000
Force (N)
Force (N)
0 0
−1000 −1000
−2000 −2000
0.1 0.3 0.5 0.7 0.9 10 20 30
Actuation signal (%) Time (s)
(c) FI diagram (d) FT diagram
Figure 8: Comparison of estimated (green) and experimental (black) data of based on E2 .
where 𝑉 is the manipulation applied to the damper, 𝑐SA is the Based on those observations, (3), (4) were customized for
force gain due to manipulation, and 𝑓pre-𝑦,𝑧 , 𝑓pre-𝑦,𝑧,𝐼̇ describe this ER damper; the following asymmetric model for the ER
the behavior of the damper in the preyield zone. If the SA damper is proposed:
damper has an asymmetric behavior the model needs to have
different coefficients for positive and negative velocities. 𝐹𝑃 = 𝑓0 + 𝑐𝑝 𝑧̇ + 𝑘𝑝 𝑧 + 𝑚𝐷𝑧̈ + 𝑓fr , (6)
In these equations, the use of the tanh function is 𝐹SA = 𝑉𝑐SA [𝑓pre-𝑦,𝑧,𝑉
̇ + 𝑓pre-𝑦,𝑧 ] . (7)
replaced with the so-called squash function: 𝑥/(1 + |𝑥|). This
function has the advantage of been lighter for computing
while reproducing almost the same pattern as the tanh. 5. Modelling
Depending on the physical characteristics of the ER Based on the previous analysis, the identification and valida-
damper, (3), (4) are customized so that only the required tion steps are relatively standard.
terms are considered to preserve an accurate representation
while simplifying the model structure, Table 2. For the passive 5.1. Identification. The model parameters were fitted using
force, Figure 6, friction, stiffness, and viscous damping were the LSE method. Since the ER damper behavior is asymmet-
observed. Frequency dependent hysteresis was also present ric, the model parameters have different values for positive
in the ER damper response. The SA damper force, Figure 7, and negative velocities. Three replicas of each experiment
presents a sigmoid behavior without significant hysteresis. were used to evaluate the performance of the customized
The preyield and postyield zones depend on the actuation model. Figures 8 and 9 show the experimental and estimated
signal but only the preyield zone depends on the damper FV, FD, FE, and FT diagrams generated from experiments E2
velocity. and E3 , respectively.
2000 2000
1000 1000
Force (N)
Force (N)
0 0
−1000 −1000
−0.1 0 0.1 −6 −4 −2 0 2 4 6
Velocity (m/s) Displacement (m) ×10−3
(a) FV diagram (b) FD diagram
2000 2000
1000 1000
Force (N)
Force (N)
0 0
−1000 −1000
0 0.2 0.4 0.6 0.8 Time (s)

Actuation signal (%)
(c) FI diagram (d) FT diagram
Figure 9: Comparison of estimated (green) and experimental (black) data based on E3 .
Based on Figures 8 and 9 the customized model correctly Table 3: ESR index for the full and customized models.
describes the nonlinear behavior of the ER damper and the Experiment Replica 1 Replica 2 Replica 3
effect of the actuation signal, and estimated data looks similar
Customized model
to the real data. However, this model was unable to describe
the stick-slip phenomenon, Figures 8(a) and 9(a), in the FV E2 0.0741 0.0749 0.0716
diagrams; they are the force peaks around 0.04 and −0.05 m/s E3 0.062 0.0627 0.0654
that the model cannot mimic. In Figures 8(b) and 9(b) it E4 0.1284 0.1337 0.1315
can be seen that the model can represent the rigidity of the E5 0.1558 0.1494 0.1416
damper, but in the same way as in Figures 9(a) and 10(a) Full model
the stick-slip phenomenon appears again. This force peaks E2 0.0730 0.0739 0.0719
appear also in Figures 8(c) and 8(d) and Figures 9(c) and 9(d), E3 0.0674 0.0661 0.0681
where the real data differ from the estimated data in the top E4 0.1258 0.1353 0.1330
and bottom zones of the diagrams.
E5 0.0797 0.0762 0.0744
For quantitative validation purposes the error-to-signal
ratio (ESR) performance index was selected. It represents the
ratio between the variance of the estimation error and the
variance of the experimental damper force [21]. If the value of observed that the values of the ESR are consistent in the
the ESR is 0, it indicates that the model estimates exactly the three replicas, with an average difference of 3%; therefore,
damper force; however, a value of 1 indicates that the model for the following analysis only the first replica will be
only predicts the mean value of the damper force. considered. It can be observed that in almost all experiments
The performance indexes for all the experiments (cus- the customized model shows same results as the full model,
tomized and full models) are shown in Table 3. It can be with the exception of E5 . This is because E5 uses the DSFS
Table 4: Performance indices for different datasets.

Experiment used for identification
Experiment used for validation Force variance 𝜎2 × 105
E2 E3 E4 E5
E2 8.39 0.0741 0.0811 0.1988 0.1065
E3 6.50 0.0676 0.0620 0.1762 0.1154
E4 5.79 0.3041 0.0614 0.1258 0.1664
E5 6.13 0.2546 0.3032 0.1727 0.1558
Average 0.1751 0.1269 0.1684 0.1360
displacement sequence with the PRBS control signal. This model can extrapolate the results for different experimental
combination, at high frequencies, introduces high variabil- data.
ity in the force; variability induces more hysteresis in the
measured force. Since the terms related with the hysteresis 6. Results
have been taken out from the model, when the hysteresis is
predominant, the customized model is not able to reproduce These results were also validated with two-dimensional den-
the force as correct as the full model. sity plots. The density plots are scatter plots that use different
colors to indicate the density of incidences in different
zones of the diagram; blue color indicates a lower number
5.2. Validation. The first step of the validation process is of occurrences (i.e. data samples), whereas red indicates a
to prove that the terms discarded have little influence in higher number. The FD, FV, and FM density plots with
the modeling performance; this is done by comparing the the customized model and experimental data are compared,
performance indexes obtained with the full model, (3), (4), Figure 10.
versus the ones obtained with the customized model, (6), (7). According to [22], since the experiment is a RP the zones
Table 3 shows that the performance indices of the cus- with higher density of occurrences should be at low velocities
tomized model are very similar to the full model; for both for the FV diagram; in the case of the FD diagrams these
models the ESR indices are low. Therefore, the discarded zones should be in the small displacement range; on the other
terms in the customized model has little effect in the response hand this experiment has a PRBS actuation signal; therefore
of this damper; this is consistent with the results obtained the higher density zones must be in the ends of the control
in the characterization step. Experiments E2 and E5 have a signal (0.1 and 0.8). The FV diagram of the estimated data,
greater ESR index when compared with the ones achieved in Figure 10(b), is similar to the one obtained with experimental
experiments E3 and E4 , respectively. This is because the small data, Figure 10(a), but because the stick-slip phenomenon is
changes in the ICPS manipulation signal have little effect not considered by the model, the estimated force does not
on the variability of the force; on the other hand the PRBS present the peaks around 0.04 and −0.05 m/s observed in
manipulation signals, which are steps of 10% to 80% and experimental data. For the FD diagrams, the friction zone is
vice versa, cause the force to experiment greater variability well defined in the model, Figure 10(d); this means, that in
that increments the effects of some phenomena as stick- the zone of the 𝑦-axis between ±0.2, there should be little
slip and hysteresis. Since in the model customization step incidences. Also, the two levels of force, caused by the PRBS
those terms were excluded, the model was less effective in signal, should be clearly defined as well. In the case of the FM
capturing those highly hysteric behaviors. This explains why diagrams, Figures 10(e) and 10(f), only four points should be
in the experiment E5 the ESR is almost the double of the one well define, since only two levels of current were used.
achieved with the full model. In order to analyze the effectiveness of the customized
A second validation procedure was to prove the extrap- model, a comparative analysis with other two well-known
olation ability. This is realized with a cross-validation of a models was carried out: the Choi parametric model [23] and
model with other datasets; the results are shown in Table 4. the Eyring-plastic model [12]. The Choi parametric model is
It was observed that the customized model can be based on the physical characteristics of the ER damper. It is
extrapolated to other signals different from those used in defined as follows:
the identification stage. The best average performance was
obtained by the experiment E5 (even though the ESR for each 𝐹𝐷 = 𝑘𝑒 𝑧 + 𝑐𝑒 𝑧̇ + 𝐹ER sign (𝑧)̇ (8)
experiment is not the smallest). This is result of the actuation with
and manipulation signals used in that experiment, because
the DSFS signal captures best the dynamical behavior of the 𝐹ER = (𝐴 𝑝 − 𝐴 𝑟 ) 𝑃ER , 𝑃ER = 𝑑𝛼𝐸𝛽 ,
damper in its whole range of operation while the RP only
explores a limited zone. The ICPS covers the whole force 𝐴2𝑟 2
(9)
range of the shock absorber while the PRBS only captures the 𝑘𝑒 = , 𝑐𝑒 = (𝐴 𝑝 − 𝐴 𝑟 ) Re,
𝐶𝑔
limits of the force range.
The two validation procedures show that the customized where 𝐴 𝑝 and 𝐴 𝑟 are the piston and rod areas, respectively,
model performance is as the full model. Also, the customized 𝐶𝑔 is the gas compliance, 𝑅𝑐 is the flow resistance due to
1 1
0.8 60 0.8 60
0.6 0.6
50 50
0.4 0.4
0.2 40 0.2 40
Force
Force
0 0
30 30
−0.2 −0.2
−0.4 20 −0.4 20
−0.6 −0.6
10 10
−0.8 −0.8
−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Velocity Velocity
(a) Experimental FV diagram (b) Estimated FV diagram
1 1
0.8 60 0.8 60
0.6 0.6
50 50
0.4 0.4
0.2 40 0.2 40
Force
Force
0 0
30 30
−0.2 −0.2
−0.4 20 −0.4 20
−0.6 −0.6
10 10
−0.8 −0.8
−1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Displacement Displacement
(c) Experimental FD diagram (d) Estimated FD diagram
1 1
0.8 60 0.8 60
0.6 0.6
50 50
0.4 0.4
0.2 40 0.2 40
Force
Force
0 0
30 30
−0.2 −0.2
−0.4 20 −0.4 20
−0.6 −0.6
10 10
−0.8 −0.8
−1 −1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Manipulation Manipulation
(e) Experimental FM diagram (f) Estimated FM diagram
Figure 10: Density plots for experimental and estimated data.

Table 5: Comparison of ER damper models.
Model Choi parametric model Eyring-plastic model Customized model

Parameters 4 7 20
Inputs 𝑧, 𝑧̇ 𝑧, 𝑧̇ 𝑧, 𝑧,̇ 𝑧̈
Actuation signal as input Yes No Yes
Hysteresis No Yes Yes
Choi parametric model Eyring-plastic model Customized model

3000 3000 3000
2000 2000 2000
Force (N)
Force (N)
Force (N)
1000 1000 1000
0 0 0
−1000 −1000 −1000

−2000 −2000 −2000
−0.2 −0.1 0 0.1 0.2 −0.2 −0.1 0 0.1 0.2 −0.2 −0.1 0 0.1 0.2
Velocity (m/s) Velocity (m/s) Velocity (m/s)
(a) Experiment E 2 (c) Experiment E 2 (e) Experiment E 2
2000 2000 2000
1000 1000 1000

Force (N)
Force (N)
Force (N)
0 0 0
−1000 −1000 −1000
−0.1 0 0.1 −0.1 0 0.1 −0.1 0 0.1

Velocity (m/s) Velocity (m/s) Velocity (m/s)
(b) Experiment E 3 (d) Experiment E 3 (f) Experiment E 3
Figure 11: Comparison of models performance based on FV diagrams.
the velocity of the ER fluid, 𝑃ER is the pressure drop due Table 6: Performance indices of the ER damper models.
to the field-dependent yield stress, 𝐸 is the applied electric
Experiment
field, 𝑑 is a constant related to the geometry of the electrode, Model
and 𝛼 and 𝛽 are experimentally determined constants which E2 E3 E4
characterize the yield stress of the ER fluid. Choi parametric model 0.1337 0.1164 0.2222
In the Eyring-plastic model the force is considered a Eyring-plastic model 0.0996 0.0816 0.1996
nonlinear function of the velocity: Customized model 0.0714 0.0642 0.1284
̇
𝐹𝐷 = 𝐹𝛼 [arcsinh (𝜆 1 𝑧̇ − 𝜆 2 𝑧)] (1 + 𝛽1 𝑒−𝛽2 |𝑧| )
(10)
+ 𝑐 − 1𝑧̇ + 𝑐3 𝑧̇ 3 , include the actuation signal in the model structure whereas
in the Eyring-plastic model the parameters are undefined
where 𝜆 1 defines the slope of the response in the preyield functions of the actuation signal.
region, 𝜆 2 defines the preyield hysteresis loop, 𝐹𝛼 is related These models were identified using the same data and
to the yield force amplitude, 𝛽1 , 𝛽2 are yield force correction algorithms, Table 6. Analyzing the ESR index, the customized
factors, and 𝑐1 , 𝑐3 model the damping in the postyield model had the best modeling performance for all experi-
region. The seven parameters are functions of the excitation ments, followed by the Eyring-plastic model.
frequency and electric field. Table 5 summarizes the different Figure 11 compares the FV diagrams for each model in
features of the models. experiments E2 and E3 . The Choi parametric model, Figures
All the analyzed models are nonlinear and depend on 11(a) and 11(b), does not estimate correctly the hysteresis
the damper displacement (𝑧) and velocity (𝑧). ̇ Only the and nonlinearities of the damping force, but the levels of
customized model includes the acceleration 𝑧̈ as input. The force caused by the changes in the manipulation signal are
Choi parametric model and the customized model explicitly notorious. The Eyring-plastic model, Figures 11(c) and 11(d),
Experimental data Choi model Eyring-plastic model Customized model

1 1 1 1
60
0.5 0.5 0.5 0.5 50
40
Force
Force
Force
Force
0 0 0 0 30
−0.5 −0.5 −0.5 −0.5 20
10
−1 −1 −1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Velocity Velocity Velocity Velocity
(a) FV diagram (b) FV diagram (c) FV diagram (d) FV diagram
1 1 1 1
60
0.5 0.5 0.5 0.5 50
40
Force
Force
Force
Force
0 0 0 0 30
−0.5 −0.5 −0.5 −0.5 20
10
−1 −1 −1 −1
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
Displacement Displacement Displacement Displacement
(e) FD diagram (f) FD diagram (g) FD diagram (h) FD diagram
1 1 1 1
60
0.5 0.5 0.5 0.5 50
40
Force
Force
Force
Force
0 0 0 0 30
−0.5 −0.5 −0.5 −0.5 20
10
−1 −1 −1 −1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Manipulation Manipulation Manipulation Manipulation
(i) FM diagram (j) FM diagram (k) FM diagram (l) FM diagram
Figure 12: Density plots of experimental and estimated data for different models (experiment E2 ).
has acceptable results at high velocities, but at low velocities Figure 12(e). The Choi model presents higher density in
(±0.02 m/s) it does not capture the hysteresis effect correctly. extension forces with zero displacement; thus it generates
The customized model, Figures 11(e) and 11(f), shows the best larger forces with small displacements, Figure 12(f). This
modeling performance since the nonlinearities added by the model also produces larger compression forces with large
manipulation signal are well described and the low and high displacements. In contrast with the experimental data, in
damping forces are correctly identified. None of the analyzed the Eyring-plastic model the higher density appears with
models consider the stick-slip effect so the force peaks around large forces and exhibits a saturation, Figure 12(g); hence
0.04 and −0.05 m/s are not reproduced by any of them. the Eyring-plastic model produces smaller forces with large
The ER damper models are also qualitatively compared displacements than the real damper. Finally, the customized
using density plots in order to identify if these models predict model, Figure 12(h), produces slightly higher forces at low
correctly the distribution of the experimental data. Figure 12 frequencies and a density distribution similar to the exper-
presents a comparison of the density plots of experiment imental data.
E2 . In the experimental FV diagram, Figure 12(a), the higher The FM diagram is important for control systems pur-
density of data appears with small compression forces while poses. A model with the same shape and density distribution
in the Choi model, Figure 12(b), the higher density appears to the experimental data is required in order to compute a
with larger forces; hence, this model represents a stiffer damp- right manipulation to achieve a desired force. Since in exper-
ing force than the real damper at low velocities. In the Eyring- iment E2 a PRBS actuation signal was used, the FM diagram
plastic model FV diagram, Figure 12(c), the higher density mostly exhibits two manipulation values, Figure 12(i). All the
appears with zero force; therefore the model generates smaller models generate smaller forces with a manipulation of 90%
forces than the real damper with low velocities. Finally, the where the stick-slip effect is more evident. Nonetheless, the
customized model, Figure 12(d), generates a similar density FM diagram obtained with the customized model resembles
of experimental data for extension forces and slightly larger the experimental data the most. The Choi and Eyring-plastic
compression forces. models present smaller forces than the customized model.
In the FD diagram the experimental data presents The Choi model is not able to generate small forces due to the
higher density with small forces, especially in compression, use of a discontinuous function; this explains why this model
Table 7: Execution times of the ER damper models. Conflict of Interests

Model Time [s] The authors declare that there is no conflict of interests
Choi parametric model 0.1500 regarding the publication of this paper.
Eyring-plastic model 0.1970
Customized model (squash) 0.3093 Acknowledgments
Customized model (tanh) 0.3600
Full model 0.3950 Authors thank Tecnológico de Monterrey and national sci-
ence and technology council of México through the bilateral
Projects (no. 142183 México-Spain and PCP-03/10 México-
could not predict the small forces present on the experimental France) because of their partial support.
data.
Since the model should be well suited for real time References
implementations, there was another comparative test done to
all the models. This test consists in measuring the time that [1] T. D. Gillespie, Fundamentals of Vehicle Dynamics, SAE, War-
the model takes to compute a vector of data points; in this case rendale, Pa, USA, 1992.
the selected vector contains 58,123 data points. The results of [2] B. F. Spencer Jr., S. Dyke, M. Sain, and J. Carlson, “Phenomeno-
this test are presented in Table 7, and each point is the average logical model for magneto-rheological dampers,” Journal of
of 3 replicas of the test. For comparison purposes there are Engineering Mechanics, vol. 123, no. 3, pp. 230–238, 1997.
included the full model and the customized model using the [3] T. Butz and O. von Stryk, “Modelling and simulation of electro
tanh function instead of the squash function. and magneto-rheological fluid dampers,” Journal of Applied
Mathematics and Mechanics, vol. 82, no. 1, pp. 3–20, 2002.
The results show, as expected, that the Choimodel spends
less than half the time (0.15 s) than the customized model [4] R. Stanway, J. L. Sproston, and A. K. El-Wahed, “Applications of
electro-rheological fluids in vibration control: a survey,” Smart
(0.3093 s), but its ESR index is almost twice the index of
Materials and Structures, vol. 5, no. 4, pp. 464–482, 1996.
the customized model. The Eyring-plasticmodel spends 33%
less time (0.197 s) than the customized model, but its ESR [5] J. Dixon, The Shock Absorber Handbook, John Wiley & Sons,
New York, NY, USA, 2007.
index is more than 30% bigger. Even if the customized model
lasts longer to be computed its precision is much better. [6] S. B. Choi, Y. T. Choi, E. G. Chang, S. J. Han, and C. S. Kim,
“Control characteristics of a continuously variable ER damper,”
Comparing the customized model using the tanh function its
Mechatronics, vol. 8, no. 2, pp. 143–161, 1998.
time is 20% less, and as expected the time to compute the full
[7] S. B. Choi, H. K. Lee, and E. G. Chang, “Field test results of
model is 30% more.
a semi-active ER suspension system associated with skyhook
controller,” Mechatronics, vol. 11, no. 3, pp. 345–353, 2001.
7. Conclusions [8] S. R. Hong, S. B. Choi, Y. T. Choi, and N. M. Wereley, “A hydro-
mechanical model for hysteretic damping force prediction of
A method for modeling ER dampers was proposed. The ER damper: experimental verification,” Journal of Sound and
proposal does not need prior knowledge of the damper Vibration, vol. 285, no. 4-5, pp. 1180–1188, 2005.
such as physical properties, dimensions, and so forth, just [9] Q.-H. Nguyen and S.-B. Choi, “Dynamic modeling of an
experimental data. The main contribution of this method is electrorheological damper considering the unsteady behavior
the simplicity that by analyzing the experimental data (i.e., in of electrorheological fluid flow,” Smart Materials and Structures,
the form of characteristic diagrams) the ER damper can be vol. 18, no. 5, Article ID 055016, 2009.
modelled with a general equation; then it can be customized [10] Q.-H. Nguyen and S.-B. Choi, “A new approach for dynamic
to obtain a simple model that captures the real behavior of a modeling of an electrorheological damper using a lumped
damper. parameter method,” Smart Materials and Structures, vol. 18, no.
Based on design of experiments (DoE), the representative 11, Article ID 115020, 2009.
behavior of the damper into the automotive domain can be [11] S.-M. Chen and C.-G. Wei, “Experimental study of the rheolog-
obtained. The resultant model proves its accuracy with an ical behavior of electrorheological fluids,” Smart Materials and
error-to-signal ratio (ESR) of 15.5% in the worst case and an Structures, vol. 15, no. 2, pp. 371–377, 2006.
average of 12.7% when extrapolating other experiments. Also, [12] L. Bitman, Y. T. Choi, S. B. Choi, and N. M. Wereley, “Electrorhe-
when compared with well-known models, the results have ological damper analysis using an Eyring-plastic model,” Smart
better performance, an average of 44.5% less than the Choi Materials and Structures, vol. 14, no. 1, pp. 237–246, 2005.
parametric model and 28.4% less than the Eyring-plastic [13] S. D. Nguyen and S.-B. Choi, “A new neuro-fuzzy training
model. Additionally, the density plots allow a qualitative algorithm for identifying dynamic characteristics of smart
dampers,” Smart Materials and Structures, vol. 21, no. 8, Article
comparison of the results, giving same conclusions.
ID 085021, 2012.
The customized model ends with a short equation with
[14] J. de-J. Lozoya-Santos, R. Morales-Menendez, R. Ramirez-
high performance. Compared with well-known approaches, Mendoza, J. C. Tudón-Martinez, O. Sename, and L. Dugard,
the simplicity of the method that does not demand a special- “Magnetorheological damper—an experimental study,” Journal
ized background of design and modeling of electrorheologi- of Intelligent Material Systems and Structures, vol. 23, no. 11, pp.
cal dampers is the main important contribution. 1213–1232, 2012.
[15] D. Kowalski, M. Rao, J. Blough, and S. Gruenberg, “The effects

of different input excitation on the dynamic characterization of
an automotive shock absorber,” in Proceedings of the SAE Noise
and Vibration Conference and Exhibition, Traverse City, Mich,
USA, 2001, SAE paper no. 2001-01-1442.
[16] C. Boggs, L. Borg, J. Ostanek, and M. Ahmadian, “Efficient
test procedures for characterizing MR dampers,” in Proceedings
of the ASME International Mechanical Engineering Congress
and Exposition (IMECE ’06), pp. 173–178, Chicago, Ill, USA,
November 2006.
[17] F. Irgens, Continuum Mechanics, Springer, Berlin, Germany,
2008.
[18] S. Guo, S. Yang, and C. Pan, “Dynamic modeling of magneto-
rheological damper behaviors,” Journal of Intelligent Material
Systems and Structures, vol. 17, no. 1, pp. 3–14, 2006.
[19] N. M. Kwok, Q. P. Ha, T. H. Nguyen, J. Li, and B. Samali, “A novel
hysteretic model for magnetorheological fluid dampers and
parameter identification using particle swarm optimization,”
Sensors and Actuators A: Physical, vol. 132, no. 2, pp. 441–451,
2006.
[20] S. Cesmeci and T. Engin, “Modeling and testing of a field-
controllable magnetor-rheological fluid damper,” International
Journal of Mechanical Sciences, vol. 52, no. 8, pp. 1036–1046,
2010.
[21] S. M. Savaresi, S. Bittanti, and M. Montiglio, “Identification
of semi-physical and black-box non-linear models: the case of
MR-dampers for vehicles control,” Automatica, vol. 41, no. 1, pp.
113–127, 2005.
[22] N. Fukushima, K. Hidaka, and K. Iwata, “Optimum charac-
teristics of automotive shock absorbers under various driving
conditions and road surfaces,” JSAE Review, pp. 62–69, 1983.
[23] S.-B. Choi, Y.-M. Han, and K.-G. Sung, “Vibration control of
vehicle suspension system featuring ER shock absorber,” Inter-
national Journal of Applied Electromagnetics and Mechanics, vol.
27, no. 3, pp. 189–204, 2008.
Chapter 4
Influence of MR Damper Modeling on

Vehicle Dynamics

As mentioned in the previous chapter, it is important to have an accurate model of the shock
absorbers, especially when assessing the performance of control systems using simulation. In this
chapter, a research article which details the influence of magneto-rheological damper modeling in
vehicle dynamics analysis is presented.
The method consists of using several tests using CarSimTM compare a four-corner controlled
semi-active suspension for two different magneto-rheological damper models. The magneto-
rheological damper characteristics were identified using experimental data. AS a control system
a model-free controller was used to discard its influence in the system and emphasizes the com-
pliance of the magneto-rheological damper model; the characteristics of the vehicle index perfor-
mance that were considered were: comfort, road holding, handling, roll, and suspension deflection.
The comparison of magneto-rheological damper dynamics and semi-active suspension covers the
automotive bandwidth.
The results of this work show that high precision of a magneto-rheological damper model as an
isolated feature is not enough. The magneto-rheological damper model, as a component of vehicle
suspension, needs to simulate passive precision and variable damping forces. The findings exhibit
the requisite of accurate models for evaluation of semi-active control systems in classic tests. The
lack of the friction component in a magneto-rheological damper model leads to an overestimation
of handling and stability.
73
IOP PUBLISHING SMART MATERIALS AND STRUCTURES
Smart Mater. Struct. 22 (2013) 125031 (13pp) doi:10.1088/0964-1726/22/12/125031
Influence of MR damper modeling on

vehicle dynamics
Jorge de-J Lozoya-Santos, Ruben Morales-Menendez,
Ricardo A Ramirez-Mendoza and Carlos A Vivas-Lopez
Tecnológico de Monterrey, Campus Monterrey, Avenida E Garza Sada 2501, Col. Tecnológico,
Monterrey, Nuevo León, 64849, Mexico
E-mail: jorge.lozoya@itesm.mx, rmm@itesm.mx, ricardo.ramirez@itesm.mx

and a00794204@itesm.mx
Received 3 July 2013, in final form 21 October 2013

Published 22 November 2013
Online at stacks.iop.org/SMS/22/125031
Abstract
The influence of magneto-rheological damper modeling in vehicle dynamics analysis is
studied. Several tests using CarSimTM compare a four-corner controlled semi-active
suspension for two different magneto-rheological damper models. The magneto-rheological
damper characteristics were identified from experimental data. A model-free controller
discards the influence of control and emphasizes the compliance of the magneto-rheological
damper model; the characteristics of the vehicle index performance considered were comfort,
road holding, handling, roll and suspension deflection. The comparison for
magneto-rheological damper dynamics and semi-active suspension covers the automotive
bandwidth. The results show that high precision of a magneto-rheological damper model as an
isolated feature is not enough. The magneto-rheological damper model, as a component of a
vehicle suspension, needs to simulate with passive precision and variable damping forces. The
findings exhibit the requisite of accurate models for evaluation of semi-active control systems
in classic tests. The lack of the friction component in a magneto-rheological damper model
leads to an overestimation in handling and stability.
(Some figures may appear in colour only in the online journal)
1. Introduction tires also affect vehicle handling performance as well as ride

quality and noise and vibration [22]. Road holding refers to
The automotive industry is very competitive field; research the tire condition of the vehicle; tires must always remain in
centers exploit several tools such as simulation to reduce contact with the road surface. Ride comfort demands a perfect
effort in product design and development. The impressive isolation of the body from the road profiles. Ride comfort
development of computer software and hardware allows for needs soft damping in damper compression and high damping
a sophisticated simulation of vehicle dynamics. in extension; while road holding uses a hard damping.
The shock absorber is one of the most important
For passive damper suspension, the challenge is to find
components in a vehicle suspension system. Furthermore,
a balanced damping element with both conditions. For a
for simulation it is one of the most complex elements
semi-active damper suspension using a control system both
of the vehicle suspension system for simulation, because
of its nonlinear and time-variant behavior. There are three ride comfort and road holding can be adapted on demand.
main types of damper: passive, semi-active and active. Research project results show that a semi-active suspension
Magneto-rheological (MR) dampers are relatively low-cost, can decrease vertical sprung mass acceleration and vertical
high-value semi-active dampers; and these MR dampers use sprung mass displacement by up to 52% and 20% [1].
less energy than active ones. The damper capabilities are described by the force–
A key issue in the design of an automobile suspension velocity (F–V) diagram (figure 1(A)), where hysteresis and
is the balance between road holding and ride comfort goals; bilinear relations in the F–V diagram are shown. The MR fluid
0964-1726/13/125031+13$33.00 1 c 2013 IOP Publishing Ltd Printed in the UK & the USA
Smart Mater. Struct. 22 (2013) 125031 J de-J Lozoya-Santos et al
Figure 1. Force–velocity diagram. (A) Typical F–V diagram. (B) MR F–V diagram.
varies the damping coefficient according to an electric current In semi-active dampers, [31] has shown that the
(figure 1(B)). asymmetry and hysteresis phenomena of an MR damper can
This research work studies the influence of adaptive produce a drift in sprung mass displacement; however, the
damping in vehicle dynamics performance such as comfort, analysis uses only linear damping coefficients. Reference [18]
handling, and roll. It presents an alternative methodology to studied the effect of the design parameters of a semi-active
analyze semi-active damper suspensions using MR dampers. damper based on a physical damper model. Reference [16]
The analysis focuses on the effect of using accurate MR reported that the magnitude of vertical acceleration in the
damper models. sprung mass can increase due to the nonlinear stiffness
This paper is organized as follows. A literature review produced by an MR damper. Reference [30] analyzed the
is presented in section 2. Section 3 presents the experimental vehicle response with optimal MR damper parameters leaving
MR damper models. Section 4 describes the control system. out the damper characteristic analysis. All these results are
The results and discussion are shown in section 5. Section 6 limited to the vertical dynamics and the modeling approach,
concludes the paper. and the influence of damper characteristics on vehicle
performance has not been explored.
Therefore, several areas of opportunity in the discussion
2. Literature review
of results have been found in the aforementioned works.
Regarding vehicle dynamics, it is necessary to discuss (a) the
The influence of dampers in vehicle dynamics has been
influence of the modeling of the damper characteristics,
explored in several works, for a quarter of vehicle model (b) the precision of a semi-active damper model in key
dynamics with passive dampers [26, 36] and with semi-active vehicle performances, and (c) the use of better models based
dampers [31, 18, 16, 30] and for a full vehicle with passive on the observation of rich experimental data. The influence
dampers [5, 43]. of damper friction and damping changes due to operational
A phenomenological model is better than a lookup table range and manipulation has not been widely studied in full
containing several slopes [36]. However, the use of lookup vehicle simulations. These issues are important since the use
tables with different damping for low/high velocities for of this kind of suspension demands an integrated chassis
simulation of passive dampers shows the importance of static control algorithm with a supervisor, control algorithms, and
friction in vertical behavior [26]. Nonlinear models of passive a coordinator [19].
dampers including friction, hysteresis and linear stiffness, The main modeling approaches for semi-active dampers
and damping are compared through lookup tables and used are parametric [10, 17, 38] and non-parametric [4, 7, 33]. The
in full vehicle simulation. Using complex models improves parametric approach includes the phenomenological (P) and
simulation; however, it uses only a pothole test and limits the semi-phenomenological (SP) frameworks. The phenomeno-
results again to vertical phenomena [43]. A full analysis [5] logical approach considers parameters that describe the phys-
observes the lateral, longitudinal and vertical behavior of a ical dimensions of the damper. The semi-phenomenological
full vehicle suspension when using the Bouc–Wen model, (SP) approach [10, 17, 42] allows the simulation of dampers
an asymmetric lookup table and a linear model for the without detailed modeling issues. Also, this type of approach
passive damper. A lookup table is found to be sufficient for is good for controller synthesis; examples of SP approaches
damper simulation; however, the tests used do not explore the are the modified Bouc–Wen model [38] and the hyperbolic
automotive bandwidth ranging from 0 to 25 Hz, nor other tangent-based model [17].
variables of interest such as roll. All aforementioned works Non-parametric modeling approaches show good results
have been developed for passive dampers. without extrapolation capabilities; the parameters have no
2
Figure 2. Schemes of MR damper models. (A) Full modified SP model. (B) Two-zone model.
physical meaning. Some examples are ones based on artificial where cMRpre−yield is the joint damping coefficient due to the
neural networks (ANNs) [33] and ones based on polynomial electric current and the velocity of the piston, and vyield is
approaches [7]. the velocity where the MR damping force yields and the fluid
Two parametric models have been chosen for this study: changes from viscous to non-Newtonian. The model scheme
(a) a full modified SP model [23] and (b) a modified two-zone consist of a viscous linear damper, a spring, a linear variable
model, derived from [26], where the electric current is an viscous damper dependent on electric current and velocity in
input. the pre-yield regime, and a gain of damping depending only
The full modified SP model [23] relies on the precision on the electric current for the post-yield regime, figure 2(B).
of the force estimation when manipulation is 0 A, reproducing
There are two types of semi-active control system.
viscous damping, stiffness and nonlinearities such as friction
The first is continuously variable semi-active (CSA) control
and hysteresis. When manipulation is different (from
[13, 14, 20, 25, 27, 37]; this type of control requires a variable
0 A), the sigmoidal F–V diagram is well shaped by the
actuator interface to modulate the damping coefficient. The
hyperbolic tangent and the proportional effect of manipulation
(figure 2(A)). second is on–off semi-active (OSA) control [3, 27, 32, 34, 35];
this type of control is achieved using only two values of
FD = FP (z, ż, z̈) + FMR (z, ż, I) (1) damping coefficient, hard for road holding or soft for ride
FP = kp z + cp ż + ff tanh (vfriction ż + xfriction z) (2) comfort.
FMR = cMRpost−yield I tanh Normally, the OSA control strategies are model free. The
performance of model-free controllers depends only on their
× vpre−yield ż + xpre−yield z (3)

logic. Model-free controllers are the most used in commercial
where FD is the MR damping force, FP is the damping vehicles; examples of this approach are the Ground-Hook,
force when manipulation is zero, FMR is the damping force the modified Sky-Hook (SH) and the Mix-1-Sensor (M1S)
due to the MR fluid changes, kp is the stiffness coefficient, [39, 20, 35]. The characteristics of model-based controllers
cp is the damping coefficient, ff is the friction coefficient, are the fast model predictive control, H∞ , and linear
vfriction and xfriction define the sigmoidal behavior of the parameter varying/H∞ control [6, 8, 12].
friction component, cMRpost−yield is the damping coefficient This work uses independent corner control systems for a
due to electric current, and vpre−yield and xpre−yield define the semi-active suspension in a full vehicle, which are focused on
sigmoidal force component due to the MR fluid changes. the control of vertical dynamics. A corner of a vehicle can be
A two-zone model for an MR damper is proposed as represented as a lumped parameter quarter of vehicle (QoV)
a modification of the passive damper model shown in [26].
model (figure 3). It consists of the sprung mass ms (weight
Experimental results support the modification since the MR
of cabin, power engine, etc), unsprung mass mus (weight of
damper force FD depends on both the electric current and the
tire, wheel, arms), the suspension strut, and the stiffness of
velocity of the piston in the pre-yield zone. In the pre-yield
zone, the gain of damping force due to electric current is the tire kt . A suspension strut is composed of a spring ks ,
independent of the velocity of the piston: a damping element (one mechanical device) with a passive
 damping component cp , and a variable damping component
 cp ż + kp z + cMRpre−yield Iż for |ż| < vyield cv . An external power source modifies a physical condition
FD = ż in the damper to change the damping. The road disturbances,
cp ż + kp z + cMRpost−yield I for |ż| > vyield
|ż| zr , excite the QoV model. The vertical displacements of the
(4) sprung and unsprung masses, zs and zus , result from excitation.
3
Table 1. Design of experiments.

Duration Range Range
DoE Displacement Current (s) (m) (A)
E1 SFS SC 280 ±0.009 0–2.5
E2 RP SC 100 ±0.03 0–2.5
where fms and fmus are the resonance frequencies of the sprung
and unsprung masses, kr = (kw × kt )/(kw + kt ) is the ride rate,
Figure 3. Quarter of vehicle model. kw = ks × µ2 is the wheel rate, and µ is the motion ratio of
the suspension of the vehicle for the damper.
The second step is the definition of frequency intervals
The QoV model suspension tuning looks for good based on fms and fmus . High electric currents maintain comfort,
comfort, road holding and safety suspension deflections. low vertical accelerations of the sprung mass. Road holding of
These goals have the following mathematical definitions [29]. low variations in vertical tire deflection, figure 4 is observed in
• Vertical chassis acceleration (z̈s ) response to road the frequency of resonance of the sprung mass and unsprung
disturbances (zr ) between 0 and 20 Hz represents mass, respectively.
the acceleration felt by the driver, i.e. ride comfort Hence, the limits of each interval of the equation can be
specification. defined around resonance frequencies:
• Vertical wheel deflection (zus − zr ) response to road fms fm
disturbances (zr ) between 0 and 30 Hz represents the ability f1 = fms + f2 = fmus − us
2 4 (7)
of the wheel to stay in contact with the road, i.e. the road fm
f3 = fmus + us .
holding specification. 4
Although a hard damping suspension helps road holding, it Finally, the control system that meets both comfort and
deteriorates ride comfort. Hence the goals are summarized as road holding requirements is chosen: hard damping (I = 5 A)
the minimization of gain of vertical acceleration (z̈s ) in the in bandwidth intervals that contain the resonance frequencies,
resonance frequency of the sprung mass, and the minimization and soft damping (I = 0 A) in the remaining intervals
of tire hop phenomena, i.e. minimum tire deflection (zus − zr ), (figure 4). Estimation of the actual frequency for choosing the
in the resonance frequency of tire deflection. right interval is not trivial [9].
A model-free controller based on the OSA control
strategy is proposed to avoid the influence of the control 3. MR damper models
system on damper model performance. The frequency-
estimation-based (FEB) controller principle [24, 28] is the The used shock absorber was a commercial MR damper
selection of high/low damping at each sampling time to manufactured by DelphiTM . The experiments were imple-
achieve a comfortable ride or a good road holding according to mented in an industrial laboratory with MTSTM equipment
the frequency of the road profile. The FEB control algorithm that includes a 3000 PSI actuator with a 15 Hz bandwidth, and
is a 407 controller hardware unit. The load capacity is 25 kN

I1 , 0 < f̂ < f1 at 3000 psi. The actuator stroke is 150 mm; the span of
the piston deflection is ±12.5 mm. The span of the electric


I , f1 < f̂ < f2

2
I= current is from 0 to 2.5 A. A LabVIEWTM interface operates
I3 ,
 f2 < f̂ < f3 the experimental setup using an NI-DAQ 9172 chassis.

I4 , f3 < f̂ < f4 , This system commands piston displacement and electric

current through the damper coil and records damper force,
f̂ = (żrms )/(2 · π · zrms ). (5)
displacement, velocity, acceleration, and electric current.
The design of the FEB controller consists of three The design of experiments (DoE) considers two
steps: (a) computation of the resonance frequencies of the patterns of displacement under different electric currents:
QoV masses according to the lumped parameter model; stepped frequency sinusoidal (SFS) and road profile (RP)
(b) definition of the limits of the frequency intervals according (figure 5). The SFS sequence is the standard test for damper
to the control goals; (c) definition of the electric current characterization. A total of 29 frequencies were explored, ω =
values for each frequency interval. This controller uses the {0.5, 1, . . . , 15}. Each frequency duration was five sinusoidal
parameters ms , mus , kt , and ks for tuning. The first step is to periods. The RP pattern was computed according to [11],
compute the resonance frequencies of the QoV model masses keeping the ideal bandwidth at high frequencies to avoid
according to unreal behavior of the MR damper [2]. The electric current
s s sequence is a stepped current (SC), which explores several
1 kr 1 kr + kt damper curves. The sequence of constant currents was I =
fms = fmus = (6)
2π ms 2π mus {0, 0.25, . . . , 2.5} A. Table 1 summarizes the DoE.
4
Figure 4. Electric currents for comfort and road holding. The high electric current, 5 A, is for 0–f1 and f2 –f3 frequency bandwidth; the
remainder must be 0 A.
Figure 5. Road profile for experiments. (A) Road profile bandwidth. (B) Road profile displacement. (C) Road profile velocity.
Figure 6. MR damper characteristics for different electric currents. (A) Behavior under null electric current (0 A). (B) Gain of damping
force under changing electric current. (C) Gain of damping force versus electric current.
The F–V and force–displacement (F–D) diagrams, as pre-yield zone, the gain of damping force depends mainly on
well as the gain of damping force due to the electric current, both electric current and velocity, while in the post-yield zone
are shown in figure 6. These diagrams were obtained using the gain of damping force is strongly dependent on electric
the DoE E1 . For null electric current (passive), I = 0 A, current. In both zones it is almost independent of the velocity
figure 6(A) presents the damping force stiffness, friction, figure 6(C); this gain is proportional to the electric current in
and viscous damping, the gain of the damping force due the post-yield zone.
to different electric currents exhibited in the pre-yield and Parameters of both MR damper models were identified
post-yield zones [41]; moreover, figure 6(B) shows the with the least squares method (LSM). The full modified SP
proportional relation of the force to electric current. In the model uses one set of parameters for positive and another for
5
Figure 7. Voronoi diagrams of the experimental and estimated MR damper force. (A) Experimental force. (B) Estimation of full modified
SP model. (C) Estimation of two-zone model.
Table 2. Identified parameters for both MR damper models.

Full modified SP asymmetric model
Two-zone model Positive velocity, ż > 0 Negative velocity, ż < 0
cMRpost−yield (N A )−1
799 cMRpost−yield (N A )
−1
873.9 cMRpost−yield (N A−1 ) 828.4
— — vpre−yield (s m−1 ) 7.7 vpre−yield (s m−1 ) 9.6
— — xpre−yield (m−1 ) 12.2 xpre−yield (m−1 ) 15.3
cp (N s m−1 ) 1017 cp (N s m−1 ) 530.0 cp (N s m−1 ) 925.2
kp (N m−1 ) −3460 kp (N m−1 ) −7361.6 kp (N m−1 ) −8941.1
— — ff (N) 115.3 ff (N) 31.7
— — vfriction (s m−1 ) 32.5 vfriction (s m−1 ) 208.4
— — xfriction (m−1 ) 66.6 xfriction (m−1 ) 609.9
cMRpre−yield (N A s m−1 ) 6383 — — — —
negative velocity. The identification of the MR damper was a grid, and in each cell of the grid, the density is computed;
done with the DoE E2 . Table 2 shows the identified parameters each color represents a level of density. The more similar the
for both models. diagram, the better the model. All the signals in the diagrams
The error-to-signal ratio (ESR) index [33] quantifies each have been normalized using a 99% percentile in order to get
model precision. It represents the ratio between the variance a uniform gridding. The Voronoi diagrams of the F–V, F–D,
of the estimation error and the variance of the real MR damper and force–electric current (F–I) of the experimental force and
force: force estimated by the SP and two-zone models are shown in
PK 2 figure 7.
i=1 (FDi − F̂Di )
ESR = PK × 100% (8) Figure 7 shows the high precision of the full modified
j=1 FDj 2
PK
(F
i=1 Di − K ) SP model to simulate the force of the MR damper. The
where K is the number of samples and FDi is the real force two-zone model lacks precision at low velocities and small
in the ith sample. F̂Di is the estimated force. The ESR index displacements, as is shown in the F–V and F–D diagrams.
for the two-zone model was 2.22% and the LSM converged in However, it simulates the stiffness in the whole plot, F–D
9.69 s; the full modified SP model has an ESR of 1.62% but diagram. The lack of precision at low velocities is due to the
took 131.84 s. A lower ESR means better precision. lack of mathematical terms to describe nonlinearities when
Also, the validation of the models was done qualitatively the electric current is null (I = 0). F–I diagrams for the two
using Voronoi diagrams [40]. These diagrams are a special models are almost the same, although the two-zone model
type of decomposition of a metric space; they show the density simulates the proportional relation between force and electric
of points of two variables in a plane. The plane is divided by current with more dispersion.
6
Figure 8. An F-class vehicle and the semi-active suspension control system. (A) F-class vehicle dimensions. (B) F-class vehicle
semi-active suspension.
4. Control system grip the road surface, (d) handling is the percentage of the
available friction in tires or the maximum achievable lateral
An F-class vehicle (figure 8(A)) is a car with high quality acceleration utilized by the vehicle–driver combination and
equipment, excellent performance in handling and comfort, (e) roll is presented in the cornering and braking maneuvers
detailed construction, and innovative technology. This vehicle and it can occur on perfectly smooth road surfaces [27].
possesses an independent corner suspension configuration, For comparison of the performance, the full modified SP
where each corner suspension uses an MR damper and a model is considered as the reference; each variable is obtained
deflection sensor. The MR damper receives the electric current from a normalized root mean square (RMS) index:
computed by the FEB controller, figure 8(B).
Each MR damper is simulated by a nonlinear model. variable2Z,RMS
xvariable = (9)
Several tests evaluate and compare two different four-corner variableSP,RMS
semi-active suspensions. The MATLAB/Simulink R software
where variable is the RMS value of the signal during any test,
simulates the control strategy and the semi-active suspension.
The full vehicle model was a commercial automobile 2Z refers to the suspension using the two-zone MR damper
developed by CarSim R . The simulation was performed in model, and SP refers to the full modified SP MR damper
real time. The vehicle measurements were received from model. The response time of the MR damper was chosen as
CarSim where Simulink computes the control algorithm and 25 ms [21]. A first order filter with this response time was
the semi-active forces for each damper. These forces were added between each controller output and the MR damper
sent to the vehicle model in CarSim. The vehicle tests were model input.
specified in CarSim (figure 9) with a sampling frequency of Four tests were implemented to evaluate the full vehicle
512 Hz. The axis system is the vehicle-body-fixed system, model in CarSimTM (figure 10). The tests are the bounce sine
SAE Standard J670e [15]. sweep test (BSST), which is focused on human comfort and
The performance to consider was as follows: (a) human ride comfort, the double lane change (DLC) test for handling,
comfort is the human response to vibration, (b) ride comfort the fishhook (FSH) test, which is focused on handling and
requires a perfect isolation of the body from the road profiles, stability, and the bump test for a behavioral comparison of
(c) road holding performance is the ability of a car to suspensions.
7
Figure 9. Integration of CarSim with MATLAB/Simulink.
Figure 10. Implemented tests in CarSimTM . (A) Bounce sine sweep test (BSST). (B) Double lane change (DLC). (C) Fishhook (FSH).
(D) Bump.
(i) BSST tests the vehicle suspension to measure the the suspension system and allows the behavior of the
comfort over a range of frequencies, typically 0–20 Hz suspension system to be compared (figure 10(D)).
(figure 10(A)).
All the tests are standard, pre-programmed and validated by
(ii) DLC. The test driver steers the vehicle through the
CarSimTM . At the end of the tests, the vehicle is in motion, so
entrance lane, turns left to avoid the single cone in the
the variables are not around zero but have a transient value.
second lane, turns right to return to the original lane, and
Table 3 shows the performances, the variables and the vehicle
then straightens the vehicle to leave the course via the
tests.
exit lane cones (figure 10(B)).
The lumped parameters of front and rear QoV models are
(iii) FSH. This produces two-wheel lift by imparting to the needed for the controller design. These were obtained from
vehicle a rapid steering reversal while the vehicle is at the vehicle parameters (table 4).
maximum lateral acceleration in one direction due to the Thus, the sprung masses corresponding to the front and
initial steer (figure 10(C)). This rapid steering reversal rear corner vehicle chassis and components supported by the
quickly changes the vehicle’s lateral acceleration to the suspension, and the unsprung mass (mus ) corresponding to
maximum value in the other direction. each wheel component link and tire, are obtained from the
(iv) Bump. This profile is a standard to simulate vertical vehicle parameters. They are computed using the ratio of
obstacles which are used to test and set the behavior of the body weight from the front end to the rear end, which
8
Table 3. Performances to be evaluated in full vehicle tests. 5. Results

Physical
Performance measurements Test Speed Duration Figure 11 summarizes the transient response of the variables
for each test. Human and ride comfort performance
Human Vertical body BSST 40 20
comfort acceleration
(figures 11(A) and (B)) show discrepancies in resonance
Ride comfort Pitch angle BSST 40 20 frequencies when different MR damper models are used.
Road holding Lateral FSH 80 10 Road holding (lateral acceleration; figure 11(C)) does not
acceleration present significant differences, but handling (yaw rate;
Handling Yaw rate FSH 80 10 figure 11(D)) shows a significative deviation at the end of a
Roll Roll DLC 120 6.3
Suspension Vertical body Bump 30 3
fishhook maneuver. Roll (figure 11(E)) is the most sensitive
displacement performance to the MR damper model since the magnitude of
roll is bigger with the two-zone MR damper model, as well as
in suspension performance (figure 11(F)), where performance
is degraded.
Table 4. Parameters of an F-class vehicle (CarSimTM ). Figure 12 presents the transient response of the estimated
Variable Description Value Unit frequency in each left corner using both MR damper models
mbs Sprung mass or body 1823 kg for all tests. The performance in the DLC test is different since
msfront Front QoV sprung mass 546.5 kg it exhibits a different estimated frequency in rear suspension
msrear Rear QoV sprung mass 364.5 kg deflection. The BSST, DLC and FSH tests exhibit differences
mus QoV unsprung mass 50 kg in the rear corner, but the DLC test shows the highest
ksf Front spring stiffness 83 N mm−1 difference.
ksr Rear spring stiffness 44 N mm−1
Figure 13 shows the F–V and F–D diagrams for the
[zf , zf ] Suspension front stroke +80/−60 mm
(jounce/rebound) left side of the vehicle. In the F–V diagrams the differences
[zr , zr ] Suspension rear stroke +70/−50 mm between suspensions are the hysteretic behavior and friction at
(jounce/rebound) low velocities, whereas in the F–D diagrams the amplitude of
kt All tire stiffness 230 N mm−1 displacement for the full modified SP model is lower than for
Whbase Wheel base 3165 mm the two-zone model. In the front corner, the full modified SP
CG Center of gravity 1265 mm
(horizontal)
model generates a higher force than the two-zone model at low
η Motion ratio 0.614 — velocities. Figure 14 presents the F–V diagram and transient
response of the BSST for the left side of the vehicle, where
the effect of sweep frequency is shown.
Table 6 summarizes the normalized RMS indices. The
Table 5. FEB controller design. main difference occurs in handling, roll and suspension
QoV fms fmus 0–f1 f1 –f2 f2 –f3 f3 –f4 performance. The two-zone model exhibits low performance
Front 1.2 11.4 0–1.8 1.8–8.5 8.5–14.3 14.3–20 for this type of test.
Rear 0.98 11.13 0–1.5 1.5–8.3 8.3–13.9 13.9–20 The vehicle performance indices related to comfort,
roll and suspension are the most sensitive to the MR
Electric current (A) 5 0 5 0
Goal Comfort Road holding
damper models. The main difference between the models
is the representation of passive behavior, where friction
and valve system cut-off forces are well represented in the
full modified SP model as parameter cf and hyperbolic
is computed using the dimensions of figure 8(B), given by tangent function, respectively. This difference does not affect
1 − [CG/Wheelbase ] = 1 − 1265/3165 = 0.6. Hence, the precision comparing MR damper models.
front end has an approximate weight of mf = 0.6 × 1823 = However, these phenomena, friction and valve system
cut-off forces, can be of importance when the MR damper
1093.8 kg, and the rear end mf = 0.4 × 1823 = 729.2 kg. The
model is part of a semi-active suspension. Some findings show
lumped parameters of the QoV model are computed assuming the influence of lack of precision in MR damper simulation
a uniform distribution of the weight, two front passengers and low velocities.
mid-level fuel tank. The parameters of the QoV model follow The transient response of comfort performance shows
a distribution of the body weight of 60% in front and 40% in that the control system is better when the two-zone model
the rear. Hence, a single front corner of the vehicle has 30% is used (±3%; figures 11(A) and (B) and table 6). However,
of the total body weight. the manipulation of the FEB controller was the same for the
different semi-active suspensions (±4%; figures 14(C) and
With the QoV model front and rear parameters, the (D)). Hence, the difference is in the use of the full modified
resonance frequencies of the masses were computed using SP model, which simulates friction and hysteretic forces, and
equations (6) (table 5). Equation (7) computes the intervals the asymmetry in the F–V diagrams (figures 14(A) and (B)).
of frequencies for the FEB controller in front and rear The accurate simulation of low magnitude forces leads to an
semi-active suspensions (figure 4). accurate assessment of the control system.
9
Figure 11. Performances of the F-class vehicle. The dark line is the four-corner semi-active suspension using the full modified SP model;
the green dashed line corresponds to the two-zone model for the MR damper. (A) Human comfort. (B) Ride comfort. (C) Road holding.
(D) Handling. (E) Roll. (F) Suspension.
Figure 12. Estimated frequencies for left corners. The dark line is for the full modified SP model, and the green dashed line the two-zone
model. (A) BSST. (B) DLC. (C) FSH. (D) Bump.
Table 6. RMS comparison of full vehicle performance. When more different from unity, the two-zone model is worse.
Human Ride Road
comfort comfort holding Handling Roll Suspension
2Z,RMS
SP,RMS 0.97 1.02 1.00 1.08 1.33 1.06
10
Figure 13. F–V and F–D diagrams for DLC test. The dark line is the full modified SP model; the green dashed line is the two-zone model.
(A) F–V diagram in DLC test. (B) F–D diagram in DLC test.
Figure 14. F–V diagram and transient response of controller output of BSST test. The dark line is the semi-active suspension with full
modified SP model, and the green dashed line is the two-zone model.
The discrepancies in road holding (0%) and handling suspension deflection need MR damper models with friction
(+8%) performance when using a fishhook test are and valve system parameters.
not important when using different MR damper models The roll performance shows the highest degradation of all
(figures 11(C) and (D), and table 6). In steering, the the performance (+33%) (figure 11(E) and table 6). The roll
performances of the two suspensions are quite similar. The is more sensitive to damping estimation errors. The two-zone
difference is at the end of the FSH test, where the yaw rate model allows the highest vertical velocities and suspension
varies. The cause is the low amplitude of vertical velocities deflections (figure 13).
(very small frequencies) in vehicle corners (figure 12(C)). Suspension performance is similar for medium to large
Since the FEB controller uses hard damping according to velocities (figure 11(F)), but when the vertical velocity starts
controller design, the difference in damping force at low to decrease, at time t = 1.5 s (figure 12(D)), the differences
velocities is caused by the lack of friction. Since the two-zone in force estimation make a difference, making the suspension
model computes forces based on damping coefficients at low appear as a hard damped suspension while the reference
velocities disregarding friction (offset force), the computed suspension exhibits a smooth damped suspension.
forces will be zero if vertical velocity is zero (frequency zero); It is important to notice that the RMS ratio of the
see figure 12(C) from time t = 5 s to the end. Summarizing, manipulations of the FEB controller has an average of 99%
simulations of maneuvers that present very small velocities in with a standard deviation of 2%, while the RMS ratio of the
11
damping force at each corner has an average of 94% with a [9] Chu W, Cao K, Li S, Luo Y and Li K 2013 Vehicle mass
standard deviation of 12% in force. The reference semi-active estimation based on high-frequency-information extraction
suspension, the suspension with the full modified SP model, Preprints of the 7th IFAC-Symp. on Advances in Automotive
Control (Japan, Sept.) pp 72–6
generates more damping force in each vehicle corner with the [10] Şahin I, Engin T and Çeşmeci Ş 2010 Comparison of some
same manipulation as the other one. existing parametric models for magnetoreological fluid
dampers Smart Mater. Struct. 19 1–11
[11] da Silva J G S 2004 Dynamical performance of highway
6. Conclusion bridge decks with irregular pavement surface Comput.
Struct. 82 871–81
Two MR damper models have been compared as an element [12] Do A L, Sename O and Dugard L 2010 An LPV control
of a semi-active automotive suspension controlled by a model- approach for semi-active suspension control with actuator
free controller. One model emphasizes the nonlinearities due constraints American Control Conf. (Baltimore, MD)
to the mechanical design of the damper, while the other [13] Do A L, Sename O, Dugard L and Ramirez-Mendoza R A
2010 An LPV approach for semi-active suspension control
assumes a linear viscous damping and stiffness. MR damper 11th Pan-American Congr. of Applied Mechanics (Brazil)
models were compared to each other as isolated features [14] Dong X M, Yu M, Liao C R and Chen W M 2010
with excellent results for force and the qualitative simulation Comparative research on semi-active control strategies for
of characteristic diagrams; the result was that the models magneto-rheological suspension Nonlinear Dyn. 59 433–53
have the same performance. The MR damper models were [15] Gillespie T D 1992 Fundamentals of Vehicle Dynamics
(Warrendale, PA: SAE Int.)
evaluated as a component of a semi-active suspension in a full [16] Guo D and Hu H 2005 Nonlinear stiffness of a
vehicle. The vehicle performance indices related to comfort, magneto-rheological damper Nonlinear Dynam. 40 241–9
roll and suspension are the most sensitive to the MR damper [17] Guo S, Yang S and Pan C 2006 Dynamical modeling of
model. These are influenced by friction and the valve system magneto-rheological damper behaviors J. Intell. Mater.
cut-off force simulation, which appear to be key components Syst. Struct. 17 3–14
[18] Heo S J, Park K and Son S H 2003 Modelling of continuously
in an MR damper model when it is to be used as a part variable damper for design of semiactive suspension
of a semi-active suspension in a full vehicle. Simulations of systems Int. J. Veh. Des. 31 41–57
maneuvers that present very small velocities in suspension [19] Her H, Yi K, Suh J and Kim C 2013 Development of
deflection need MR damper models with friction and valve integrated control of electronic stability control, continuous
system parameters. damping control and active anti-roll bar for vehicle yaw
stability Preprints of the 7th IFAC-Symp. on Advances in
Automotive Control (Japan, Sept.) pp 83–8
Acknowledgments [20] Hong K S, Sohn H C and Hedrick J K 2002 Modified Skyhook
control of semi-active suspensions: a new model, gain
The authors thank CONACyT (Proyecto Bilateral México- scheduling, and hardware-in-the-loop tuning J. Dyn. Syst.
Meas. Control 124 158–67
España 142183) and Tecnológico de Monterrey (autotronics [21] Koo J H, Goncalves F D and Ahmadian M 2006 A
research chair) for partial support; also, especially METALSA comprehensive analysis of the response time of MR
SA de CV for the experimental setup for this research. dampers Smart Mater. Struct. 15 351–8
[22] Kusaka K and Suzuki T 2013 The statistical tire model and its
application to vehicle dynamics design Preprints of the 7th
References IFAC-Symp. on Advances in Automotive Control (Japan,
Sept.) pp 77–82
[1] Barak P 1992 Passive versus active and semi-active suspension [23] Lozoya-Santos J, Aubouet S, Morales-Menendez R,
from theory to application in North American industry SAE Sename O, Ramirez-Mendoza R A and Dugard L 2010
Tech. Pap. 922140 Hysteresis modelling for a MR damper Proc. of EUROSIM
[2] Boggs C, Borg L and Ostanek J 2006 Efficient test procedures (Czech Republic, Sept.)
for characterizing MR dampers ASME Int. Mechanical Eng [24] Lozoya-Santos J, Morales-Menendez R, Sename O,
Congr. and Exposition IMECE2006-13275 Ramirez R and Dugard L 2011 Control strategies for an
[3] Bolandhemmat H, Clark C M and Golnaraghi F 2009 automotive suspension with a MR damper IFAC World
Development of a systematic and practical methodology for Conf. (Italy) pp 1820–5
the design of vehicles semi-active suspension control [25] Luo N, Rodellar J, Vehi J and De la Sen M 2001 Composite
system Veh. Syst. Dyn. 48 567–85 semiactive control of a class of seismically excited
[4] Burton S A, Makris N, Konstantopoulos I and Antsaklis P J structures J. Franklin Inst. 338 225–40
1996 Modeling the response of ER damper: [26] Maher D and Young P 2011 An insight into linear quarter car
phenomenology and emulation J. Eng. Mech. 122 897–906 model accuracy Veh. Syst. Dyn. 49 463–80
[5] Calvo J A, Lopez-Boada B, San Roman J L and Gauchia A [27] Miller L R 1988 Tuning passive, semi-active and fully active
2009 Influence of a shock absorber model on vehicle suspension systems Proc. of the 27th Conf. on Decision and
dynamic simulation Proc. Inst. Mech. Eng. D 223 189–202 Control pp 2047–53
[6] Canale M, Milanese M and Novara C 2006 Semi-active [28] Pletschen N, Spirk S and Lohmann B 2013 Frequency-
suspension control using fast model-predictive techniques selective adaptive control of a hybrid suspension system
IEEE Trans. Control Syst. Technol. 14 1034–46 Preprints of the 7th IFAC-Symp. on Advances in Automotive
[7] Choi S B, Lee S K and Park Y P 2001 A hysteresis model for Control (Japan, Sept.) pp 237–42
field-dependent damping force of a magneto-rheological [29] Poussot-Vassal C, Spelta C, Sename O, Savaresi S M and
damper J. Sound Vib. 245 375–83 Dugard L 2012 Survey and performance evaluation on
[8] Choi S B and Sung K G 2008 Vibration control of magneto- some automotive semi-active suspension control methods: a
rheological damper system subjected to parameter comparative study on a single-corner model Annu. Rev.
variations Int. J. Veh. Des. 46 94–110 Control 36 148–60
12
[30] Prabakar R S, Sujatha C and Narayanan S 2013 Response of a [37] Spelta C, Previdi F, Savaresi S M, Fraternale G and
quarter car model with optimal magneto-rheological Gaudiano N 2009 Control of magnetorheological dampers
damper parameters J. Sound Vib. 332 2191–206 for vibration reduction in a washing machine Mechatronics
[31] Rajalingham C and Rakheja S 2003 Influence of suspension 19 410–21
damper asymmetry on vehicle vibration response to ground [38] Spencer B F Jr, Dyke S J, Sain M K and Carlson J D 1997
excitation J. Sound Vib. 266 1117–29 Phenomenological model of a MR damper J. Eng. Mech.
[32] Sankaranarayanan V, Emekli M E, Güvenc B A, Güvenc L, 123 230–8
Serdar Öztürk E, Ersolmaz S S, Eyol I E and Sinal M 2008 [39] Valasek M, Novak M, Sika Z and Vaculin O 1997 Extended
Semiactive suspension control of a light commercial ground-hook—new concept of semi-active control of truck
vehicle IEEE/ASME Trans. Mechatronics 13 598–604 suspension Veh. Syst. Dyn. 29 289–303
[40] Voronoi G 1908 Nouvelles applications des parametres
[33] Savaresi S M, Bittanti S and Montiglio M 2005 Identification
continus a la theorie des formes quadratiques J. Reine
of semi-physical and black-box non-linear models: the case Angew. Math. 133 97–178
of MR-dampers for vehicles control Automatica 41 113–27 [41] Wang D H and Liao W H 2011 Magneto-rheological fluid
[34] Savaresi S M and Spelta C 2007 Mixed Sky-Hook and ADD: dampers: a review of parametric modelling Smart Mater.
approaching the filtering limits of a semi-active suspension Struct. 20 023001
J. Dyn. Syst. Meas. Control 129 382–92 [42] Worden K, Hickey D, Haroon M and Adams D E 2009
[35] Savaresi S M and Spelta C 2009 A single-sensor control Nonlinear system identification of automotive dampers: a
strategy for semi-active suspensions IEEE Trans. Control time and frequency-domain analysis Mech. Syst. Signal
Syst. Technol. 17 143–52 Process. 23 104–26
[36] Simms A and Crolla D 2002 The influence of damper [43] Zach C, Mack W, Fruhmann G and Tieber W 2007 On the
properties on vehicle dynamic behavior SAE Tech. Pap. performance of rheological shock absorber models in full
2002-01-0319 vehicle simulation Veh. Syst. Dyn. 45 981–99
13
Chapter 5
Conclusions
The development of VDC has been a subject of study during the past few decades. The increasing
demand for safer, more efficient, and more comfortable vehicles has sparked the development of a
greater range of more specialized VDC, causing many of them in the vehicle to be included to cover
more performance goals and functions. Such an increment of isolated control systems may cause
undesired interactions among them. This causes conflicts with their control goals and requires
more complex infrastructure to manage the increased communication and required measurements.
To tackle those demands in the most effective way, the concept of Global Chassis Control
(GCC) offers a feasible solution with the advantages of: 1) avoiding possible conflicts or interven-
tions among different VDC that coexist in the vehicle, and 2) communicating and coordinating the
subsystems’ actions and signals. However, there are some issues with the current approaches:
1. The integration is oriented to be functional, i.e. the integration is done to meet a specific
objective. For example: reduce braking distance, prevent roll over, etc, which is essential to
reach higher dynamical performance.
2. Many of the existing GCC strategies do not cover the three dynamic planes (vertical, longi-
tudinal, and lateral).
3. The GCC strategy treats all of the different driving conditions with the same control frame-
work.
4. The control topology is centralized, i.e the controller design is computationally heavy and
difficult to implement.
To overcome these issues, five contributions were proposed in this research project:
I. A control strategy that can improve the dynamic performance of a road vehicle in the three
dynamic planes by using steering, braking and suspension subsystems.
89
90 CHAPTER 5. CONCLUSIONS
II. There is a reduced set of vehicle measurements that can be used to classify a group of driving
situations.
III. For a specific driving condition, there is a proper combination of operation modes and ac-
tuators that improves the performance of a vehicle when comparing it to an uncoordinated
system.
IV. The overall dynamic performance of the vehicle can be improved by independently coordi-
nating the available subsystems using a single decision system.
V. A GCC system can improve the dynamic performance of the vehicle if it reacts according to
the experienced driving situation in the vehicle.
Based on the aforementioned theses, the conclusions of the main work of this research can be
grouped into five sections. The first section presents the concluding remarks regarding the efforts
to classify the driving situation experienced by the vehicle during common driving. The second
section shows the conclusions of the realized work on coordination schemes for subsystems’ con-
trollers. The third concludes the proposed GCC system by considering three vehicle subsystems
and a coordination scheme driven by the current driving situation. The fourth frames the contribu-
tions resulting from this research and, finally, the fifth section presents possible continuations and
future works to be done in the field of GCC.
5.1 Classification of driving situations on road vehicles

A discrimination algorithm to classify different driving situations has been proposed. Since most
of the GCC approaches in literature are designed to perform for a specific driving situation, e.g. in
emergency braking or rollover situations, it is important to know which driving situation the vehicle
is experiencing in order to react properly to that situation. In this research, the main objective was
to propose a control system that can react to many driving situations; thus, it was necessary to
know the current driving situation.
Such a classification algorithm was designed to be light in terms of computational effort while
also maintaining the same infrastructure. This was done by not increasing the complexity of the
GCC system which demands a big amount of extra signals (measurements). For that reason, this
algorithm was proposed in two steps. The first was an offline data analysis of different vehicle
variables in various driving situations to determine the minimal set of variables required to clas-
sify all driving situations. In this step, the combined use of data processing tools like PCA and
contribution plots was able to find a reduced set of variables from which a driving situation can be
differentiated.
5.2. COORDINATION OF INTERACTING SUBSYSTEMS 91
The second step is executed online and is based on the k-NN algorithm, which is one of the sim-
plest machine-learning approaches. It is light enough to be implemented in real time and accurate
enough to correctly classify the considered driving situations.
5.2 Coordination of interacting subsystems

Two premises were the basis for the conception of a coordination scheme which is driven by
various driving situations.
• Vehicle subsystems that impact the vehicle dynamics have strong interdependence. This
impact can be beneficial for the dynamic objective of the vehicle or it can degrade the ef-
fectiveness of the control system. For example, in an emergency braking situation, the most
important subsystem is the braking one, but the suspension can serve as a support by main-
taining the tire-road contact through a pitch-oriented action. On the other hand, the steering
system could degrade the performance by changing the direction of the tires affecting the
maximum longitudinal force that the tires could deliver. Therefore, the proper coordination
of such subsystems improves the performance of the vehicle.
• According to the driving situation experienced by the vehicle, the subsystems have to react
with a proper mode of operation regarding the situation. For example, the suspension sys-
tem can be oriented to improve the passengers’ comfort or to maintain the road-holding in
situations where maintaining the stability of the vehicle is the priority, even if it affects the
comfort.
Using those premises, a coordination scheme for the three considered subsystems was pro-
posed. In this scheme, the mode of operation of the subsystems and their participation is coor-
dinated using the driving situation. This coordination used the classified driving situation as an
input and, as outputs, sent the subsystems’ controllers a single parameter, to change the mode of
operation for each.
5.3 Global Chassis Control systems

In the literature, different approaches of GCC have been presented, each managing to improve
the original system by considering the global structure of the vehicle in its design rather than
just including isolated control systems in the vehicle. The proposed GCC system in this research
includes new characteristics compared to existing solutions:
• Driving situation driven coordination: It is clear that not all of the subsystems have to be
used in all situations and, not in the same way either. This is why the proposed GCC system
was designed to be coordinated based on the current driving situation, changing its mode of
operation according to it. This characteristic gave the system the advantage of introducing
or reducing the number of considered driving situations and the advantage of modifying the
coordination scheme for each driving situation as well.
• Modularity: Since the design of the GCC system used a multi-layer topology with a hierar-
chical structure, each layer has specific tasks to perform. This dispatches the computational
effort in different ECUs allowing the system to be decentralized and makes it easier for the
system to implement standard electronics.
• Design flexibility: Each of the subsystems is coordinated according to its characteristics

and its contributions to the driving situations in the automotive vehicle: i.e. the Semi-Active
suspension system coordination and modes of operation are designed to contribute the most
against each of the driving situations; in the case where the comfort is required (e.g. riding),
the suspension is adjusted to preserve that requirement, but when the situation changes and
merits another operation mode like road-holding (e.g. rapid steering), the suspension is
adjusted by monitoring the movements of the vehicle with the SAP. This property gives the
designer the possibility of modifying the subsystems involved. Also the considered modes
of operation, included in the coordination scheme, must be designed for each subsystem
based on the number of driving situations. In the case of a reconfiguration of the system (a
different type of subsystems or number) it is just done in the coordination scheme leaving
the remaining subsystem controllers unmodified.
• Control flexibility: This characteristic refers to the ability of the system to change the
control framework used in the design. This was achieved through two characteristics: 1)
modularity of the proposal: in the control layer, each subsystem must have its own control
allocation module and local controller, which allows the selection of the control framework
without affecting the other subsystems. Also, the coordination is based on the driving situa-
tions; this characteristic can be used as a scheduling parameters for model-based approaches
as LPV systems. 2) Design flexibility; each subsystem must have its own control system,
which gives the possibility of modifying the type or number without affecting the function-
ality of the others.
These characteristics make this proposal suitable to be adapted to different vehicles and phys-
ical setups, although not all the subsystems are the same and the technology of actuation may be
different.
5.4. CONTRIBUTIONS 93
5.4 Contributions
During the course of this research work, a GCC system was proposed. In this control system, two
main contributions were obtained:
1. A classification algorithm for vehicle driving situations. This algorithm introduces a new
approach to how a driving situation is classified. This approach uses data processing tools,
like PCA and contribution plots, to analyze and classify driving situations experienced by a
road vehicle in normal driving. The proposed classifier algorithm consists of two steps: the
first is done offline and analyzes the influential variables in each driving situation, selecting
the minimal amount needed. The second step is performed online and is based in a k-NN
algorithm that determines the driving situation based on the current measurements. The
selected classification algorithm is light and simple enough to be implemented in real-time.
- Specific contribution: A discrimination method, based on data processing and machine
learning approaches, for classification of road vehicle driving situations.
2. A coordination structure for control systems. The structure takes into account the proper
moment when a specific subsystem has to act (driving situation) and how the subsystem has
to actuate (mode of operation). Those decisions for each subsystem are oriented to select
the interaction that performs better and contributes more to the overall performance of the
vehicle. Based on these considerations, the system coordinates the participation and mode
of operation for each subsystem individually during the specific driving situation. Also, this
approach considers two characteristics: 1) subsystem modularity and 2) control framework
flexibility. These characteristics are oriented to give the GCC system the capabilities of
being reconfigured without major effort in changing the type and quantity of subsystems and
to modify the control method selected.
- Specific contribution: A control coordination scheme which uses the current driving situa-
tion to choose the more efficient mode of operation for each subsystem.
5.5 Publications
Indexed journal papers
J1 C.A. Vivas-Lopez, D. Hernandez-Alcantara, R. Morales-Menendez, R.A. Ramirez-Mendoza,

and H. Ahuett-Garza, “Method for Modeling Electrorheological Dampers Using Its Dynamic
Characteristics”, Mathematical Problems in Engineering, 2014.
J2 C.A. Vivas-Lopez, J.C. Tudon-Martinez, D. Hernandez-Alcantara, and R. Morales-Menendez,

“Global Chassis Control System using Suspension, Steering and Braking Subsystems”, Math-
ematical Problems in Engineering, 2015.
International conference papers with proceedings
C1 C.A. Vivas-Lopez, R. Morales-Menendez, J.C. Tudon-Martinez, and D. Hernandez-Alcantara,

“Review on Global Chassis Control”, in proceedings of the 5th Symposium on System
Structure and Control (SSSC), pp: 875-80, Grenoble, France. February 2013.
C2 C.A. Vivas-Lopez, D. Hernandez-Alcantara, R. Morales-Menendez, and A. Martinez Vargas,

“Modeling Method for Electro-Rheological Dampers”, in proceedings of the 19th IFAC
World Congress, pp: 5005-10, Cape Town, South Africa. August, 2014.
C3 C.A. Vivas-Lopez, J.C. Tudon-Martinez, D. Hernandez-Alcantara, R. Morales-Menendez, and

R.A. Ramirez-Mendoza, “On/Off Controllers Design: Experimental Results for Automotive
Suspension Systems”, in proceedings of the 12th International Symposium On Advanced
Vehicle Control (AVEC) 2014, pp: 558-63, Tokyo, Japan. September 2014.
C4 C.A. Vivas-Lopez, D. Hernandez-Alcantara, M.Q. Nguyen, R. Morales-Menendez, and O.

Sename, “Force Control System for an Automotive Semi-active Suspension”, in proceedings
of the 1th IFAC Workshop on LPV Systems (LPVS), Grenoble, France. September 2015.
C5 C.A. Vivas-Lopez, R. Morales-Menendez, and R.A. Ramirez-Mendoza, “Chassis Control

based in Fuzzy Logic”, to be appear in 2016 IEEE International Conference on Fuzzy
Systems (FUZZ-IEEE) part of the IEEE WCCI 16, Vancouver, Canada. July 2016.
Participation in indexed journal papers
J3 J.de-J. Lozoya-Santos, R. Morales-Menendez, R.A. Ramirez-Mendoza and C.A. Vivas-Lopez,

“Influence of MR damper modeling on vehicle dynamics”, Smart Mater. Struct. 22(12),
2013.
J4 D. Hernandez-Alcantara, J.C. Tudon-Martinez, L. Amezquita-Brooks, C.A. Vivas-Lopez, and

R. Morales-Menendez, “Modeling, Diagnosis and Estimation of Actuator Faults in Vehicle
Suspensions”, Control Eng. Practice. In Press, 2015.
Participation in int. conference papers with proceedings
C6 J.C. Tudon-Martinez, J.de-J. Lozoya-Santos, C.A. Vivas-Lopez, R. Morales-Menendez and

R.A. Ramirez-Mendoza, “Model-Free Controller for a Pick-Up Semi-Active Suspension Sys-
tem”, in proceedings of the ASME 2012 Int Mechanical Engineering Congress and Ex-
position Vol 11: Transportation Systems, pp: 67-76, USA. Nov 2012.
5.6. FUTURE WORK 95
C7 S. Varrier, C.A. Vivas-Lopez, J.de-J. Lozoya-Santos, J.C. Tudon-Martinez, D. Koenig, J.J.

Martinez, and R. Morales-Menendez, “Applicative Fault Tolerant Control for semi-active
suspension system: Preliminary results”, in proceedings of the European Control Confer-
ence (ECC) 2013, pp: 3803-8, Zurich, Switzerland. July 2013.
C8 D. Hernandez-Alcantara, L. Amezquita-Brooks, C.A. Vivas-Lopez, R. Morales-Menendez,

and R.A. Ramirez-Mendoza, “Fault detection for automotive semi-active dampers”, in pro-
ceedings of the Conference on Control and Fault-Tolerant Systems (SysTol) 2013, pp:
625-30, Nice, France. October 2013.
C9 D. Hernandez-Alcantara, J.C. Tudon-Martinez, L. Amezquita-Brooks, C.A. Vivas-Lopez, and

R. Morales-Menendez, “State Observers for Semi-Active Suspensions: Experimental Re-
sults”, in proceedings of the IEEE Multi-Conference on Systems and Control (MSC)
2014, Antibes, France. October 2014.
5.6 Future work

Considering this research as a starting point, there are some open issues in the field of GCC to be
studied which may lead to new research lines in this topic. The following are still open problems
in the field of GCC:
ê The automotive vehicle experiences a wide variety of driving situations during normal driv-
ing, but also others that occur less frequently, e.g. rollover, strong wind action, slalom, climb
slopes, etc. Those situations are just as important and can compromise the integrity of the
occupants, making the identification/classification of such situations mandatory to: 1) de-
cide how to react when they occur and 2) extend the classification algorithm proposed in this
work to include these situations.
ê An experimental validation of the proposed GCC system is a natural progression of this

research work. Such an evaluation could be done in two parts: 1) a HiL evaluation of the
feasibility of implementation in a ECU, and 2) implementation of the proposal in a real
vehicle.
ê Commercial vehicles can be divided regarding the target market for which they are intended.
This orientation creates a large range of available configurations, making an area of oppor-
tunity for the performance evaluation of the proposed control system against the different
configurations of subsystems and types of technologies.
ê Most research is in the field of model-based control approaches, and therefore a comparison
of the current proposal with a GCC system designed using only model-based approaches
could be interesting to do. Specifically, using the LPV approach where the result of the
decision layer can be used as a scheduling parameter to coordinate the actions of the different
controllers and subsystems would be interesting to examine.
ê According to the current automotive standards, most of the information running in the ve-
hicle flows through a Controller Area Network (CAN) communication system. Therefore, it
is essential to evaluate the performance of the proposed control system under this protocol
regarding: 1) discretization of the required information, 2) reliability against saturation of
the network and loss of packages, 3) feasibility of its physical implementation under this
protocol, and 4) effects on the performance of the overall system.
ê Since the automotive system is prone to faults, the study and development of a fault-tolerant
version of this proposal is still an open area of study.
ê Regarding the actuator modeling, especially in the field of Semi-Active dampers, the evalu-
ation of the effects of the model on physical systems in the controller design process is still
open.
ê The current trend in the automotive industry is the introduction of electric vehicles for the
massive market; this fact opens up an area to explore the implications of such technologies
when designing a GCC system. Some of the variants to consider here are: 1) new actuators
(in-wheel motors), 2) effects of different mass distributions (extra masses in the unsprung
mass adding motors, or in the sprung mass with the battery packs), and 3) different vibration
limits for the new elements (motors, batteries, etc).
Bibliography
[Ali et al., 2013] Mohammad Ali, Paolo Falcone, Claes Olsson, and Jonas Sjöberg. Predictive
Prevention of Loss of Vehicle Control for Roadway Departure Avoidance. IEEE Trans on
Intelligent Transportation Sys, 14(1):56–68, Mar 2013.
[Andreasson and Bünte, 2006] J. Andreasson and T. Bünte. Global Chassis Control based on In-
verse Vehicle Dynamics Models. Vehicle System Dynamics, 44(sup1):321–328, 2006.
[Aripin et al., 2014] M.K. Aripin, S. Yahaya, K.A. Danapalasingam, K. Peng, N. Hamzah, and
M.F. Ismail. A Review of Active Yaw Control System for Vehicle Handling and Stability En-
hancement. Int J of Vehicular Technology, 2014:1–15, 2014.
[Beyer et al., 1999] Kevin Beyer, Jonathan Goldstein, Raghu Ramakrishnan, and Uri Shaft. When
is nearest neighbor meaningful? In Catriel Beeri and Peter Buneman, editors, Database Theory
ICDT99, volume 1540 of Lecture Notes in Computer Science, pages 217–235. Springer Berlin
Heidelberg, 1999.
[Boada et al., 2005] B. L. Boada, M.J. Boada, and V. Dı́az. Fuzzy-logic Applied to Yaw Moment
Control for Vehicle Stability. Vehicle System Dynamics, 43(10):753–770, 2005.
[Canale and Fagiano, 2008] M. Canale and L. Fagiano. Vehicle Yaw Control using a fast NMPC
Approach. In 47th IEEE Conf on Decision and Control, pages 5360–5365, Cancun, México,
dec. 2008.
[Chen et al., 2011] Wuwei Chen, Hansong Xiao, Liqiang Liu, Jean W. Zu, and HuiHui Zhou.
Advances in Mechatronics. Eng. InTech, 2011.
[Cho et al., 2012] Wanki Cho, Jaewoong Choi, Chongkap Kim, Seibum Choi, and Kyongsu Yi.
Unified Chassis Control for the Improvement of Agility, Maneuverability, and Lateral Stability.
IEEE Trans on Vehicular Technology, 61(3):1008–1020, 2012.
[Chou and D’andréa-Novel, 2005] H. Chou and B. D’andréa-Novel. Global Vehicle Control us-
ing Differential Braking Torques and Active Suspension Forces. Vehicle System Dynamics,
43(4):261–284, 2005.
97
98 BIBLIOGRAPHY
[Di Cairano et al., 2013] S. Di Cairano, H.E. Tseng, D. Bernardini, and A. Bemporad. Vehicle
Yaw Stability Control by Coordinated Active Front Steering and Differential Braking in the
Tire Sideslip Angles Domain. Control Systems Technology, IEEE Transactions on, 21(4):1236–
1248, July 2013.
[Doumiati et al., 2013] M. Doumiati, O. Sename, L. Dugard, J.-J. Martinez-Molina, P. Gaspar,

and Z. Szabo. Integrated Vehicle Dynamics Control via Coordination of Active Front Steering
and Rear Braking. European J of Control, 19(3):121–143, 2013.
[El-Sheikh, 2011] T. El-Sheikh. Concept of Global Chassis Control: using Model Reference and
Virtual Actuators. PhD thesis, U of California, Davis, USA, 2011.
[Falcone et al., 2007] P. Falcone, F. Borrelli, J. Asgari, H.E. Tseng, and D. Hrovat. A Model
Predictive Control Approach for Combined Braking and Steering in Autonomous Vehicles. In
Mediterranean Conf. on Control Automation, pages 1–6, Athens, Greece, Jun 2007.
[Fergani et al., 2012] S. Fergani, O. Sename, and L. Dugard. Performances Improvement Through
an LP V /H∞ Control Coordination Strategy Involving Braking, Semi-Active Suspension and
Steering Systems. In 51st IEEE Conf on Decision and Control, pages 4384–4389, Maui, Hawaii,
Dec 2012.
[Fergani et al., 2015] S. Fergani, O. Sename, and L. Dugard. An LP V /H∞ integrated Vehicle
Dynamic Controller. IEEE Transactions on Vehicular Technology, PP(99):1–1, 2015.
[Fergani, 2014] S. Fergani. Commande Robuste LP V /H∞ Multivariable pour la Dynamique

Véhicule. PhD thesis, Université de Grenoble, France, 2014.
[Gáspár et al., 2007a] P. Gáspár, Z. Szabó, J. Bokor, C. Poussot-Vassal, O. Sename, and

L. Dugard. Global Chassis Control using Braking and Suspension Systems. In 20th Symp
of the Int Association for Vehicle System Dynamics, pages 13–17, Berkeley, USA, 2007.
[Gáspár et al., 2007b] P. Gáspár, Z. Szabó, J. Bokor, C. Poussot-Vassal, O. Sename, and

L. Dugard. Towards Global Chassis Control by Integrating the Brake and Suspension Systems.
In 5th IFAC Symp on Advances in Automotive Control, pages 563–570, USA, 2007.
[Gordon et al., 2003] T. Gordon, M. Howell, and F. Branda. Integrated Control Methodologies
for Road Vehicles. Vehicle System Dynamics, 40(1-3):157–190, 2003.
[He, 2005] J He. Integrated Vehicle Dynamics Control Using Active Steering, Driveline and Brak-
ing. PhD thesis, University of Leeds, UK, 2005.
[Hwang et al., 2008] T. Hwang, K. Park, S. Heo, S. Lee, and J. Lee. Design of Integrated Chassis
Control logics for AFS and ESP. Int J of Automotive Technology, 9:17–27, 2008.
BIBLIOGRAPHY 99
[Imaseki and Kobari, 1990] T. Imaseki and Y. Kobari. Integrated Four-Wheel Steer and Differen-
tial Slip Control System. In US Patent 4,941,095, Jul 10 1990.
[Isermann, 2006] R. Isermann. Fault-Diagnosis Systems. Springer, Germany, 1st edition, 2006.
[Krishna et al., 2014] S. Krishna, S. Narayanan, and S. Denis Ashok. Control of Yaw Disturbance
using Fuzzy Logic based Yaw Stability Controller. Int. J. of Vehicular Technology, 2014:1–10,
2014.
[Lee et al., 2009] Kangwon Lee, Youngwoo Kim, and Jinhee Jang. A Study of Integrated Chassis
Control Algorithm with Brake Control and Suspension Control Systems for Vehicle Stability
and Handling Performance. In ICCAS-SICE, pages 4053–4057, Tokio, Japan, Aug 2009.
[Liberzon, 2003] Daniel Liberzon. Switching in Systems and Control. Systems and Control.
Birkhäuser Boston, 2003.
[Lozoya-Santos et al., 2013] J.J. Lozoya-Santos, R. Morales-Menendez, R.A. Ramirez-Mendoza,

and C.A. Vivas-Lopez. Influence of MR damper modeling on vehicle dynamics. Smart Materi-
als and Structures, 22(12):125031, 2013.
[Lu et al., 2010] S.B. Lu, S.B. Choi, Y.N. Li, M.S. Seong, and J.S. Han. Global Integrated Control
of Vehicle Suspension and Chassis Key Subsystems. Proc of the Institution of Mechanical Eng,
224:423–441, 2010.
[Lu et al., 2011] S.B. Lu, Y.N., S.B. Choi, L. Zheng, and M.S. Seong. Integrated Control on
MR Vehicle Suspension System Associated with Braking and Steering Control. Vehicle System
Dynamics, 49(1-2):361–380, 2011.
[Miura et al., 2008] T. Miura, Y. Ushiroda, K. Sawase, N. Takahashi, and K. Hayashikawa. Devel-
opment of Integrated Vehicle Dynamics Control System S-AWC, volume 20 of Technical Review.
Mitsubishi Motors Co, 2008.
[Mokhiamar and Abe, 2006] O. Mokhiamar and M. Abe. How the Four Wheels should share
Forces in an Optimum Cooperative Chassis Control. Control Eng Practice, 14(3):295–304,
2006.
[Poussot-Vassal et al., 2008] C. Poussot-Vassal, O. Sename, and L. Dugard. A LPV/H∞ Global

Chassis Controller for Handling Improvements involving Braking and Steering Systems. In 47th
IEEE Conf on Decision and Control, pages 5366–5371, Cancun, México, Dec 2008.
[Poussot-Vassal et al., 2011a] C. Poussot-Vassal, O. Sename, L. Dugard, P. Gáspár, Z. Szabó,

and J. Bokor. Attitude and Handling Improvements through Gain-Scheduled Suspensions and
Brakes Control. Control Eng Practice, 19(3):252–263, 2011.
100 BIBLIOGRAPHY
[Poussot-Vassal et al., 2011b] C. Poussot-Vassal, O. Sename, L. Dugard, and S. M. Savaresi. Ve-

hicle Dynamic Stability Improvements through Gain-Scheduled Steering and Braking Control.
Vehicle System Dynamics, 49(10):1597–1621, 2011.
[Poussot-Vassal et al., 2012] C. Poussot-Vassal, C. Spelta, O. Sename, S.M. Savaresi, and

L. Dugard. Survey and Performance Evaluation on some Automotive Semi-Active Suspen-
sion Control Methods: A Comparative Study on a Single-Corner Model. Annual Reviews in
Control, 36(1):148–160, 2012.
[Poussot-Vassal, 2008] C. Poussot-Vassal. Robust Multivariable Linear Parameter Varying Auto-

motive Global Chassis Control. PhD thesis, Grenoble Institut Polytechnique, France, 2008.
[Sato et al., 1992] S. Sato, H. Inoue, M. Tabata, and S.Inagaki. Integrated Chassis Control System
for Improved Vehicle Dynamics. In Proc. AVEC, pages 413–418, 1992.
[Schilke et al., 1988] N.A. Schilke, R.D. Fruechte, N.M. Boustany, A.M. Karmel, B.S. Repa, and
J.H. Rillings. Integrated Vehicle Control. In Int Congress on Transportation Electronics, pages
97–106, Dearborn, USA, 1988.
[Shibahata, 2005] Y. Shibahata. Progress and Future Direction of Chassis Control Technology.
Annual Reviews in Control, 29(1):151–158, 2005.
[Tavasoli et al., 2012] A. Tavasoli, M. Naraghi, and H. Shakeri. Optimized Coordination of

Brakes and Active Steering for a 4WS Passenger Car. ISA Trans, 51(5):573 – 583, 2012.
[TechCast, 2012] TechCast. Bubble Charts Of TechCast’s Latest Results on TRANSPORTA-

TIONs. Technical report, TechCast-LLC, Jun 2012.
[Tjonnas and Johansen, 2010] J. Tjonnas and T.A. Johansen. Stabilization of Automotive Vehicles
Using Active Steering and Adaptive Brake Control Allocation. IEEE Transactions on Control
Systems Technology, 18(3):545–558, May 2010.
[Tudon-Martinez et al., 2013] J. C. Tudon-Martinez, S. Varrier, R. Morales-Menendez, and

R. Ramirez-Mendoza. Suspension Control Strategy for a Fully Electrified Vehicle. In Euro-
pean Control Conference, pages 1469–1474, Zürich, Switzerland, Jul 2013.
[Valasek et al., 2004] M. Valasek, O. Vaculin, and J. Kejval. Global Chassis Control: Integration
Synergy of Brake and Suspension Control for Active Safety. In 7th Int Symp on Advanced
Vehicle Control, pages 495–500, Netherlands, 2004.
[Vivas-Lopez et al., 2014] C. A. Vivas-Lopez, D. Hernandez-Alcantara, R. Morales-Menendez,

R. A. Ramirez-Mendoza, and H. Ahuett-Garza. Method for modeling Electrorheological
BIBLIOGRAPHY 101
Dampers Using Its Dynamic Characteristics. Mathematical Problems in Engineering, 2014:1–

15, 2014.
[Vivas-Lopez et al., 2015] C. A. Vivas-Lopez, J. C. Tudon-Martinez, D. Hernandez-Alcantara,

and R. Morales-Menendez. Global Chassis Control System Using Susension, Steering, and
Braking Subsystems. Mathematical Problems in Engineering, 2015:1–18, 2015.
[Wei et al., 2006] J. Wei, Y. Zhuoping, and Z. Lijun. Integrated Chassis Control System for Im-
proving Vehicle Stability. In IEEE Int Conf on Vehicular Electronics and Safety, pages 295–298,
Beijing, China, Dec 2006.
[WHO, 2013] WHO. Global Status Report on Road Safety 2013: Supporting a Decade of Action.
Technical report, World Health Organization, 2013.
[Xiao et al., 2009] H. Xiao, W. Chen, H. Zhou, and J.W. Zu. Integrated Vehicle Dynamics Control
through Coordinating Electronic Stability Program and Active Suspension System. In Int Conf
on Mechatronics and Automation, pages 1150–1155, Changchun, China, Aug 2009.
[Yi and Yi, 2011] Seongjin Yi and Kyongsu Yi. Design of Active Roll Control System and Inte-
grated Chassis Control for Hybrid 4WD Vehicles. In 14th Int IEEE Conf on Intelligent Trans-
portation Systems, pages 1193–1198, Washington, USA, Oct 2011.
[Yim et al., 2012] Seongjin Yim, Jaewoong Choi, and Kyongsu Yi. Coordinated Control of Hy-
brid 4WD Vehicles for Enhanced Maneuverability and Lateral Stability. IEEE Trans on Vehic-
ular Technology, 61(4):1946–1950, May 2012.
[Yokoya et al., 1990] Y. Yokoya, R. Kizu, H. Kawaguchi, K. Ohashi, and H. Ohno. Integrated
Control System Between Active Control Suspension and Four Wheel Steering for the 1989
CELICA. In SAE Technical Paper 901748, 1990.
[Yoon et al., 2010] Jangyeol Yoon, Wanki Cho, Juyong Kang, Bongyeong Koo, and Kyongsu Yi.
Design and Evaluation of a Unified Chassis Control System for Rollover Prevention and Vehicle
Stability Improvement on a Virtual Test Track. Control Eng Practice, 18:585–597, 2010.
[Yu et al., 2008] F. Yu, D.F. Li, and D.A. Crolla. Integrated Vehicle Dynamics Control - State-of-
the-Art Review. In Vehicle Power and Propulsion Conf, pages 1–6, Harbin, China, Sep 2008.
[Zin et al., 2009] Alessandro Zin, Olivier Sename, and Luc Dugard. A feedback-feedforward sus-
pension control strategy for global chassis control through anti-roll distribution. International
Journal of Vehicle Autonomous Systems, 7(3-4):201–220, 2009.
Appendix A
Acronyms and Variables Descriptions
Table A.1: Acronyms Definitions
Acronyms Description
ABS Anti-Lock Braking System
AS Active Steering
AFS Active Front Steering
CDC Continuous Damping Controller
DC Decentralized Controllers
DL Decision Logic
DLC Double Line Change
DoF Degree of Freedom
DBC Data-Based Controller
EMB Electro Mechanical Braking
ESC Electronic Stability Control
ECU Electronic Control Unit
FH Fish Hook
FCS Force Control System
FL Fuzzy Logic
FIS Fuzzy Inference System
GCC Global Chassis Control
IVDC Integrated Vehicle Dynamics Control
k-NN k Nearest Neighbor
LPV Linear Parameter Varying
Continued. . .
103
104 APPENDIX A. ACRONYMS AND VARIABLES DESCRIPTIONS
Table A.1: Acronyms Definitions (Continued)
Acronyms Description
MBC Model-Based Controller
MF Membership Function
PC Principal Component
PCA Principal Component Analysis
RMS Root Mean Square
ROC Receiver Operating Characteristic
SA Semi-Active
SAS Semi-Active Suspension
SAP Suspension Adjustment Plane
SCS Steering Control System
SISO Single-Input Single-Output
SMC Sliding Mode Control
TRC Traction Control
UC UnControlled
VCS Vehicle Control System
VDC Vehicle Dynamics Control
4WIB Four Wheel Independent Braking
4WS Four Wheel Steering
Table A.2: Modeling. Variables Descriptions
Variable Description Units

a1 ,a2 Hysteresis coefficients related to displacement and velocity s/m,1/m
cp Viscous damping coefficient Ns/m
fbrake ,fsteer Cut-off frequency of the actuator dynamics Hz
fc Force coefficient due to manipulation N/V
FD Damper force N
FSA Semi-Active damper force N
kp Stiffness coefficient N/m
l Vehicle wheel base m
ms , mus Sprung/Unsprung mass kg
Tb+ ,δ + Actuator output MPa,deg
Continued. . .
105
Table A.2: Modeling. Variables Descriptions (Continued)
Variable Description Units

Tb∗ ,δ ∗ Actuator controller output MPa,deg
Vx , vxi,j Vehicle/Tire longitudinal vel m/s
x,y,z Longitudinal/Lateral/Vertical Displacement m
ẋ,ẏ,ż Longitudinal/Lateral/Vertical Velocity m/s
ẍ,ÿ,z̈ Longitudinal/Lateral/Vertical Acceleration m/s2
zdef ,żdef Damper deflection and velocity m,m/s
zr , zs , zus Road/Sprung mass/Unsprung mass vertical position m
β Vehicle side slip angle deg
δ Steering wheel angle deg
δdriver Driver’s steering command deg
λ Tire slip ratio –
υ Damper manipulation V
φ, θ, ψ Pitch/Roll/Yaw angle deg
φ̇, θ̇, ψ̇ Pitch/Roll/Yaw rate deg/s
ψ̇d Desired yaw rate deg/s
ω Rotational speed of the tires rad/s
Table A.3: Algorithms. Variables Descriptions
Variable Description
am Actuation vector
au Gain of the allocation coordination
aususp , aubraking , austeer Suspension/Braking/Steering coordination gain
cmin , cmax Softest/Hardest damping coefficient
Ci Driving class
Coni,k Contribution index of the ith variable in a k driving situation
CSk Minimal set of important variables for a driving situation
d(x, yj ) Euclidian distance
DC Driving Situation
e(β),e(ψ̇) Slip angle/yaw rate error
GABS ABS braking gain
IS Initial set of vehicle variables
Continued. . .
106 APPENDIX A. ACRONYMS AND VARIABLES DESCRIPTIONS
Table A.3: Algorithms. Variables Descriptions (Continued)
ki Number of nearest neighbors
lk First l principal components
m Number of measurements
MS Minimal set of important variables for all driving situations
Mz Corrective yaw moment
n,n∗ Original/Reduced number of variables
r Number of driving conditions
ss Driving situation critical condition
si Driving situation importance
tcrit Situation changing waiting time
tcon Variable contribution threshold
Tdriver Driver braking torque
TESCr , TESCl Right/Left corrective braking pressure
TG Pressure gain
tny , Noise variance threshold
uc ,u∗c Allocated/Coordinated controller output
U |rh , U |conf Full suspension road-holding/comfort suspension command
ususpi,j ,Ususp
∗
Single corner/Full suspension control command
x Data point
Xipassive ,Xicontrolled Passive/Controlled ith performance variable
x̃i,k Residual data point of the ith measure in a k driving situation
βd ,ψ̇d Desired vehicle slip angle/yaw rate
δAF S Corrective steering angle
δsteering wheel ,δwheels
∗ ∗
Steering wheel/wheels directional angle
λcrit Critical tire slip ratio
σxi,k Variance of the ith variable in a k driving situation
Rk Covariance matrix
Xk Data matrix of n variables
P̂k , P̃k Principal/Residuals components matrix
T̂k , T̃k Principal/Residual scores matrix
X̂k , X̃k Principal /Residual transformed variables
i Variable number
Continued. . .
107
Table A.3: Algorithms. Variables Descriptions (Continued)
k Driving situation
j Number of neighborhood
Curriculum Vitae
Carlos Alberto Vivas López was born in Villahermosa, Tabasco, México on April 29th , 1987.
He received the degree of Bachelor’s of Science in Mechatronics from the Instituto Tecnolgico
y de Estudios Superiores de Monterrey, Campus Monterrey in May 2010. He was accepted to
the graduate program in Sciences and Engineering in August 2010 where he is currently a Ph.D.
candidate at ITESM’s Autotronics Research Chair. His research interests include advance control
systems for vehicles and robust control systems.
This document was typed using LATEX by Carlos Alberto Vivas López
109

Thesis by Carlos Alberto Vivas Lopez 2016

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis by Carlos Alberto Vivas Lopez 2016

Uploaded by

Copyright:

Available Formats

Instituto Tecnológico y de Estudios Superiores de Monterrey

School of Engineering and Sciences

Global Chassis Control System for Passenger Vehicles

Carlos Alberto Vivas López

Major in Mechatronics and Control

Monterrey, Nuevo León, April 1st 2016

School of Engineering and Sciences

Dr. Rubén Morales Menéndez

Dr. Olivier Sename

Dr. Luc Dugard

Dr. Ricardo Ambrosio Ramı́rez Mendoza

Dr. Horacio Ahuett Garza

Dr. Jorge Welti Chanes

Monterrey, Nuevo León, April 1st 2016

• I have acknowledged all main sources of help.

Carlos Alberto Vivas López

@2016 by Carlos Alberto Vivas López

I would like to thank the Tecnológico de Monterrey for granting me a schol-

Also, I would like to thank my thesis committee members: Prof. Olivier

3 Method for Modeling Electrorheological Dampers Using Its Dynamic Characteristics 55

4 Influence of MR Damper Modeling on Vehicle Dynamics 73

A Acronyms and Variables Descriptions 103

Curriculim Vitaes 109

Figure 1.1: Vehicle reference system and important dynamical variables.

a) Steering b) Brakes c) Suspension

Figure 1.2: The most studied vehicles subsystems

Vehicle Vehicle Vehicle

(a) Centralized (b) Decentralized (c) Multi-layer

Figure 1.3: Control system topologies.

1.3 Problem Description

1. To avoid possible conflicts or interventions among different subsystems’ controllers and

2. To gain potential advantages of each subsystem by communicating and coordinating their

2. To classify different driving situations based on a minimal set of vehicle measurements.

Table 1.1: Single purpose integration approaches

Table 1.2: Multi purpose integration approaches

Table 1.3: Full dynamics integration approaches

Figure 1.4: Proposal of the GCC system.

Assessment of the proposal

6 Comparison cases: The vehicle performance was compared in 3 different configurations:

I. Classification of vehicle driving situations. A new approach to classify the experienced

Global Chassis Control System Using

2.1 Summary of the Chapter

Carlos A. Vivas-Lopez,1 Juan C. Tudon-Martinez,2

Correspondence should be addressed to Carlos A. Vivas-Lopez; a00794204@itesm.mx

Received 4 August 2015; Accepted 21 October 2015

Academic Editor: Xinggang Yan

1. Introduction Stability Control (ESC) which creates a turning moment

Table 3: Algorithms: variables description. Table 3: Continued.

Variable Description Variable Description

Pitch angle (deg.)

Figure 2: Comparison of MBC and DBC approaches.

dynamics is monitored by some sensors. Figure 2 compares

Hook and Ground-Hook algorithms.

3. Global Chassis Control Figure 3: GCC architecture, hierarchical approach.

The architecture of the GCC system is divided into three main

Figure 4: Control scheme of the GCC system.

MS ⊆ IS : |MS| ≤ |IS| . (1) T ̂ 𝑘,

DC𝑖 [si ] < DC𝑗 [si ]

··· DC𝑖 DCj ···

Figure 5: Graph for the DL module.

𝑐min 0 10% 3.3.2. Steering System. A steer-by-wire Active Steering (AS)

(c) DLC test (rapid steering) (d) FH test (cornering)

(e) Split 𝜇-surface braking (loss of vehicle control)