Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

SAE TECHNICAL
PAPER SERIES 2002-01-0302

A Predictive Control Algorithm for an

Anti-Lock Braking System
Sohel Anwar and Behrouz Ashrafi
Visteon Corporation

SAE 2002 World Congress

Detroit, Michigan
March 4-7, 2002

400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-5760
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

The appearance of this ISSN code at the bottom of this page indicates SAE’s consent that copies of the
paper may be made for personal or internal use of specific clients. This consent is given on the condition,
however, that the copier pay a per article copy fee through the Copyright Clearance Center, Inc. Operations
Center, 222 Rosewood Drive, Danvers, MA 01923 for copying beyond that permitted by Sections 107 or
108 of the U.S. Copyright Law. This consent does not extend to other kinds of copying such as copying for
general distribution, for advertising or promotional purposes, for creating new collective works, or for
resale.

Quantity reprint rates can be obtained from the Customer Sales and Satisfaction Department.

To request permission to reprint a technical paper or permission to use copyrighted SAE publications in
other works, contact the SAE Publications Group.

All SAE papers, standards, and selected

books are abstracted and indexed in the
Global Mobility Database

No part of this publication may be reproduced in any form, in an electronic retrieval system or otherwise, without the prior written
permission of the publisher.

ISSN 0148-7191
Copyright © 2002 Society of Automotive Engineers, Inc.

Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE. The author is solely
responsible for the content of the paper. A process is available by which discussions will be printed with the paper if it is published in
SAE Transactions. For permission to publish this paper in full or in part, contact the SAE Publications Group.

Persons wishing to submit papers to be considered for presentation or publication through SAE should send the manuscript or a 300
word abstract of a proposed manuscript to: Secretary, Engineering Meetings Board, SAE.

Printed in USA
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

2002-01-0302

A Predictive Control Algorithm for an

Anti-Lock Braking System
Sohel Anwar and Behrouz Ashrafi
Visteon Corporation

Copyright © 2002 Society of Automotive Engineers, Inc.

ABSTRACT coefficient. Tan and Chin [3] discussed an anti-lock

braking algorithm based on sliding mode control theory.
Generalized predictive control (GPC) is a discrete time Based on a longitudinal one-wheel vehicle model,
control strategy proposed by Clark et al [1]. The controller sufficient conditions for applying sliding mode control to
tries to predict the future output of a system or plant and vehicle traction were derived via Lyapunov stability theory.
then takes control action at present time based on future Athan and Papalambros [4] presented a multi-criteria
output error. Such a predictive control algorithm is quasi-Monte Carlo method to optimize and compare three
presented in this paper for deceleration slip regulation in ABS nonlinear control algorithms. An adaptive fuzzy logic
an automobile. Most of the existing literature on the anti- controller for an anti-lock braking system was discussed
lock brake control systems lacks the effectiveness of the by Kokes and Singh [5]. Their controller initially employed
wheel lockup prevention when the automobile is in a skid a priori training data to control the braking system, but
condition (in a low friction coefficient surface with panic continued to train on-line while continuously updating
braking situation). Simulation results show that the confidence parameters and placement of fuzzy sets by
predictive feature of the proposed controller provides an employing optimization algorithms. Suh et al [6]
effective way to prevent wheel lock-up in a braking event. presented a real time simulator for an anti-lock brake
system based on methodology of hardware-in-the-loop
INTRODUCTION simulation using a personal computer. They also provided
an analysis and validation of the control logic that used
Anti-locking brake systems (ABS) are well established in commercial hardware. A genetic algorithm based fuzzy
the automotive industry as a safety feature. ABS generally logic controller for ABS systems was the subject of the
offers superior vehicle safety by limiting the longitudinal paper written by Chen and Liao [7]. The controller was
wheel slip in a braking event with deep slip condition. based on a nonlinear feedback linearization scheme and
Drivers would have better directional control of the vehicle fuzzy logic strategy. Given the desired wheel slip-ratio,
equipped with ABS. Most of the ABS systems in the the feedback linearizing controller cancels all nonlinear
published literature utilize the wheel slip estimate to dynamics and imposes an appropriate linear behavior on
control the wheel cylinder pressure through a set of valves the wheel slip-ratio. Huang and Wang [9] discussed a
and a pump in order to regulate the wheel braking torque. mixed fuzzy controller to navigate escaping motions of
These control systems are built around the actuators’ wheeled vehicles under the assumption of Coulomb’s
dynamic characteristics. With new generation of actuators viscous friction. The paper focused on modeling, analysis,
on the horizon (e.g. electromagnetic, electromechanical, and control issues of anti-lock braking systems under the
etc.), it may be necessary to design a supervisory control assumption of Coulomb’s viscous friction while applying
system that would issue torque command as control input the mixed fuzzy controller.
based on vehicle dynamics. The slave controller would
then close the loop on the actuators based on actuator A predictive control algorithm for an anti-lock braking
dynamics. application is presented in this paper. The objective of the
present paper is to investigate the proposed control
Watanabe et al [2] presented an ABS algorithm that used strategy for improving vehicle braking performance on a
a vehicle deceleration threshold to activate the brake low friction coefficient surface. The predictive nature of the
pressure reduction algorithm and the duration of the control algorithm is expected to provide an insight to slip
pressure pulse was determined by the road friction growth that can be leveraged to provide better wheel
adhesion with the road surface.
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

The present work is based on generalized predictive

control (GPC) theory. First a simplified linearized vehicle
∑M y = Tbi sgn( −ωi ) + Fxi R − Frri R − Tdi = I wi ω
&i
model is derived. This linear vehicle model is then used to (2)
derive a control algorithm based on GPC theory. where,
Simulation results are presented next based on the Tbi = Brake torque at i-th wheel
derived controller. ωi = Angular speed of i-th wheel
Fxi = Longitudinal friction force at i-th tire contact patch
LINEARIZED VEHICLE MODEL R = Effective wheel rolling radius
Frri = Rolling Resistance at i-th tire contact patch
Tdi = Drive torque at i-th wheel
In order to project the future output for the vehicle and
Iwi = i-th wheel rotational inertia
wheel speed, it is necessary to obtain the dynamic
equations for the vehicle motion. A simplified vehicle ω & i = Angular acceleration of i-th wheel
model obtained first; then the model is discretized using a
bilinear transformation. The vehicle motion in the The pitch dynamics of the vehicle in the first equation is
longitudinal direction on the road plane is described by the assumed to have negligible effect on the wheel braking
following equation [8]. forces. For the sake of simplicity, the effect of terrain
forces arising out of road slopes and grades are also
∑F xr = Fxsumr + Ftxr − Faxr = M (U& r − Vr rr ) + ms Z& sr q r neglected, The drive torque (in a braking situation) is
assumed to be insignificant in the second equation.
(1) Further simplification is made by assuming that the steer
where wheel angle is zero resulting in zero lateral motion. Now
the following relationships are defined :

Fxi = µi Fzi ; Frri = ηFzi

where
R µι(κ) = Friction Coefficient, and
η = Rolling Resistance Coefficient;
Ur
Since a simple model is desired for the proposed
controller development, the effect of aerodynamic drag and
Tbi ωi
rolling resistance on the above equation are neglected.
The above assumption is justified based on the fact that
the rolling resistance is insignificant compared to the
braking force in a braking event. Also, the aerodynamic
drag is small for the normal city driving speeds. Since this
controller is a closed loop system, these effects can be
Fxi compensated through the feedback information. The
Fzi
Figure 1 Wheel forces and moments in a braking following equations are obtained :
event. Fxsumr = − ∑ µi (κi ) Fzi ;
Fxsumr = sum of road forces in the x-direction at the road The simplified equations of motion are then given by :
tire interfaces − ∑ µi (κi ) Fzi = MU& r
Ftxr = Terrain forces at the c.g. arising out of road slopes (3)
and grades
Tbi sgn( −ωi ) − µi (κi ) Fzi R = I wi ω
&i (4)
Faxr = Aerodynamic drag forces on the vehicle
M = Total vehicle mass
U r = Vehicle longitudinal velocity Based on the above equations, the plant model for
designing a controller is obtained as follows:
Vr = Vehicle lateral velocity
rr = Vehicle yaw velocity
1
ms = Sprung mass of the vehicle U& r = − ∑ µi (κi ) Fzi
Z& sr = Sprung mass velocity in the M (5)
1
qr = Pitch velocity of the sprung mass
ωi = − (Tbi − µi (κi ) Fzi R )
I wi (6)
The wheel rotational dynamics is given by the following
equation [8]:
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

Now tire slip is obtained from the following definition : nonlinearity to equation (8). Since all of the curves in
κi (t ) = U r − Rωi (7) figure 2 exhibit linear relationship with slip below the peak
of the curve, this relationship between the coefficient of
friction and the slip can be approximated with a piecewise
where,
linear function. This concept is illustrated in figure 3. The
R= Effective rolling radius for the tire
friction curves are approximated by a straight line with a
ωi = Wheel rotational speed for i-th tire
slope of αsi and a slip threshold of κth. While the peak of
Ur = Vehicle longitudinal speed in road co-ordinate
these friction curves varies over a slip range, a slip
system
threshold and initial slope can be established for sub-
optimal performance. The piecewise linear friction
Therefore,
• • •
coefficient-slip relationship can be described as follows.
R
κ i = (U r − R ω i ) = (Tbi − µ (κ i ) Fzi R ) µi (κi ) = αsi * κi if κi ≤ κth
I wi (10)
(8) = αsi * κth if κi ≥ κth
1
−
M
∑ µ (κ i i ) Fzi
Therefore, equation (9) can be rewritten as,
• R 1
Assuming the vehicle is moving forward, equation (8) can
be further simplified as follows.
κi =
I wi
(Tbi − αsi κi Fzi R) −
M
∑α κ F
si i zi
(11)
• R 1 for κi ≤ κth
κi =
I wi
(Tbi − µi (κi ) Fzi R ) −
M
∑ µ (κ )F i i zi (9)
Now the last term in the above equation represent the
total friction force on the vehicle. The variation in this term
due to variation in the friction coefficient from one wheel to
1 the other is not going to affect the overall equation
Asphalt
0.9 Snow significantly. Therefore, it is assumed that the friction
Ice
coefficient in the last term in equation (11) has the same
0.8
value as that of the particular wheel. With this
0.7
assumption, equation (11) can be further simplified as
Friction Coefficient

0.6 follows:
0.5

•
R 1
κi = (Tbi − αsi κi Fzi R) − αsiκi ∑ Fzi
0.4

0.3 I wi M
0.2
(12)
•
R
0.1 κi = (T − αsi κi Fzi R) − αsi κi g
0
I wi bi
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Longitudinal Slip Ratio

Figure 2 Friction Coefficient vs. Slip Curve for different The above is the linearized equation for the braking
surfaces. dynamics. A Laplace transform of the above yields :

κi ( s ) R / I wi
Figure 2 shows the friction coefficient curves for a number = (13)
Tbi ( s) R 2 Fziαsi
of road-tire interfaces. It is evident that the peak of the s + (αsi g + )
friction coefficient curve varies significantly depending on I wi
the road condition. The slip value at the peak friction Or,
coefficient also varies between 0.1 to 0.2. It is clear that κi (s ) B
the friction coefficient relationship with slip adds =
Tbi ( s) s + A
where
R R 2 Fzi αsi
B= ; A = (αsi g + ) (14)
I wi I wi
µi
α σι

κι

Figure 3 Simplified Friction Coefficient vs. Slip Curve.

 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

E j ( z −1 )( c + dz −1 ) ∆ + z − j F j ( z −1 ) = 1 (18)
where,
PREDICTIVE CONTROL LAW
Ej(z -1) = A polynomial in z-1 with order (j-1)
Fj(z -1) = A polynomial in z-1 of degree 1.
Anti-Lock Braking System (ABS) has been around in the
cars and trucks for many years. The effectiveness of
these systems varies widely depending on the system Multiplying both sides of equation (18) by κi ( t + j ) and
design, road conditions and driver's response. Most of rearranging,
these systems are based on empirical data and heavily κi ( t + j ) = F j κi (t ) + E j e (1 + z −1 ) ∆Tbi (t + j − 1)
dependent on testing. In the current paper, a more
systematic approach is taken to develop an ABS system (19)
based on a linearized vehicle model and a predictive The objective function can now be rewritten in matrix
control algorithm. format as,
J = [ K iDes − K i ] T [ K iDes − K i ] (20)
Like most of the ABS control algorithm, the current where,
controller also requires the knowledge of wheel slip. The K iDes = [κiDes ( t + 1)κiDes (t + 2)........κiDes (t + N )]
objective of the controller is to keep the wheel slip at a
value that would maximize the tire-road adhesion (or
minimize the tire slip). K i = [ K i (t + 1) K i (t + 2)........ K i (t + N )] (21)
where
K i (t + 1) = F1κi (t ) + G1 ∆Tbi (t )
N
J i = ∑ [κides (t + j ) − κi (t + j )] 2 (15)
j= 0 K i (t + 2) = F2 κi ( t ) + G2 ∆Tbi (t + 1)
.
where
Ji = Slip performance index for i-th tire .
N = Prediction horizon
K i (t + N ) = FN κi (t ) + GN ∆Tbi (t + N − 1)
κides (t + j ) = Desired slip for i-th tire at time t+j =
where
Percent Desired Slip for i-th tire * Vehicle Speed
G j ( z −1 ) = E j ( z −1 )e(1 + z −1 )
κi ( t + j ) = Estimated slip for i-th tire at time t+j
The predicted slip equations can be re-written in a matrix
Generalized predictive control (GPC) utilizes Diophantine
format as follows :
type discrete mathematical identities to obtain predicted
plant output in the future. In addition to its predictive
K i = G *U + f
capabilities, GPC has been shown to be robust against where
modeling errors and external disturbances [1].
 g0 0 . . 0
In the following section, a discrete version of the GPC  g g 0 . . 0 
(Generalized Predictive Control) is derived. A bilinear  1
transformation of the above equation yields G= . . . . . 
 
 . . . . . 
κi ( z ) e(1 + z −1 )  g N −1 g N −2 . . g 0 
= (16)
Tbi ( z ) c + dz −1
U = [∆Tbi (t )∆Tbi (t + 1)....∆Tbi (t + N − 1)]T
where,
e = BT f = [ f (t + 1) f (t + 2).... f (t + N )]T
c = ( AT + 2 ) f (t + 1) = [G1 ( z −1 ) − g10 ]∆Tbi ( t ) + F1κi ( t )
d = ( AT − 2) f (t + 2) = z[G2 ( z −1 ) − z −1 g 21 − g 20 ]∆Tbi (t ) + F2 κi ( t )
The above can be rewritten as
..
( c + dz −1 )κi ( z ) = e (1 + z −1 )Tbi ( z ) (17) ..

Now the Diophantine prediction equation (j-step ahead

predictor) is given by, Gi ( z −1 ) = g i 0 + g i 1 z −1 + .....
(22)
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

The objective function can now be rewritten as follows : wheel speed information.
J = [ K iDes − f − GU ] [ K iDes − f − GU ] T
(23)
Figures 4 though 6 show the braking simulation results on
snow without any anti-lock braking control. These figures
Minimization of the Objective function yield the following
show the baseline performance of the vehicle.
predictive control law :
U = [ GT G] −1 G T ( K ides − f ) (24) It is evident from figure 6 that the front wheels reached a
lock-up condition while the rear wheels exhibits only a
In the above equation, U is a vector. To obtain the control
law at present time, only the first element of U is used.
Therefore the control law is given by, 30

∆Tbi ( t ) = ∆Tbi (t − 1) + g T ( K ides − f ) FL Wheel Speed, m/s

(25) FR Wheel Speed, m/s
RR Wheel Speed, m/s
[G T G ]−1 G T . 25
where gT is the first row of RL Wheel Speed, m/s
Vehicle Speed, m/s

20
Equation (25) is the predictive control law for the anti-lock
braking system.
15

10
SIMULATION RESULTS
5
The above control law (25) has been implemented on full
vehicle model in a simulation environment. The results for
simulation runs are discussed in this section. Since 0
11 12 13 14 15 16 17
equation (25) will provide ABS functionally based on a time, sec

predefined slip threshold value, the braking performance

may be compromised for a normal high friction coefficient Figure 5 Wheel speed and Vehicle speed in a
1
road surface. Hence, in the controller implementation, braking event on snow.
ABS mode detection is implemented based on the 0.9

impending wheel lock-up. If the wheel deceleration is 0.8

greater that certain threshold value, the controller
0.7
activates the predictive ABS part of the algorithm.
0.6
Wheel Slip

0.5

The vehicle model has 14 degrees of freedom (DOF). A 0.4

vehicle speed estimator, which is not the subject of this
0.3
paper, is utilized to obtain the vehicle speed. Wheel
speed is directly obtained for the vehicle model. In reality, 0.2

the wheel speed will be obtained from the sensors. Wheel 0.1
slip can now be computed based on equation (7) for all
0
four wheels. The wheel deceleration is estimated from 11 12 13 14 15 16 17
time, sec

30 Figure 6 Wheel slip ratio in a braking event on snow.

Vehicle Speed, m/s

partial lock-up condition.

20

10
Simulation results with the predictive ABS controller
turned on are present in this section. The following
0 controller parameters has been used in simulation :
11 12 13 14 15 16 17
time, sec

0 Prediction Horizon = 3
Vehicle Accleration, m/s 2

-2

-4
Control Horizon = 1
-6
Deceleration Threshold = 0.9g
-8

-10
11 12 13 14 15 16 17
time, sec

Figure 4 Vehicle speed and acceleration in a

 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

Speed Threshold for ABS Deactivation = 0.5 kph or less been able to prevent wheel lock-up for all four wheels.
Figure 10 shows the wheel torque command from the
controller and actual wheel torques.

Figure 7 Vehicle speed and acceleration in a Figure 9 Wheel slip ratio in a braking event on snow.
braking event on snow.
It is noted that the rear wheels of the vehicle have
4 actuators with speed dependent torque characteristics
αsi =
Ur with limited torque generation capability. As a result, no
significant slip is resulted at the rear wheels.
The normal force on each wheel is estimated from the
static weight distribution and dynamic weight transfer in
an acceleration/deceleration event.

Figure 7 shows vehicle speed and deceleration in a

braking event on snow. It is clear that the deceleration is
less than that for the non-ABS case. This is the result of
the sub-optimal performance due to variation in the peak

Figure 10 Wheel torque command and braking

torque at wheel in the braking maneuver on snow.

It is observed that the tracking of the desired slip ratio is

under-damped. This deviation from the desired slip ratio
can be traced back to the fact that the assumption of
linearity between friction coefficient and wheel slip was not
maintained in the simulation. In simulation, the simplified
slip curve coefficient αsi has been a function of vehicle
Figure 8 Wheel speed and Vehicle speed in a
speed as indicated in above. Nevertheless, the controller
braking event on snow.
was able to prevent any wheel lock-up for the vehicle on
friction coefficient with respect to slip ratio. However, the low coefficient surface.
figures 8 and 9 show that the predictive controller has
 Downloaded from SAE International by Univ of California Berkeley, Monday, August 06, 2018

CONCLUSIONS
A generalized predictive control law has been derived for a 8. Kiencke, U. and Nielsen, L., "Automotive Control
simplified linear vehicle model for an anti-lock braking System for Engine, Driveline, and Vehicle", SAE
system. The predictive nature of the controller has been International, 2000.
utilized to regulate wheel slip at a desired threshold value 9. Huang, H.-P. and Wang, C.-K., “Intelligent Control of
based on predicted slip values. Simulation results show Wheeled Vehicles with Anti-Braking Systems”,
that wheel slip can be regulated within certain range of the International Journal of Vehicle Design, Vol. 26, No. 2-
desired slip ratio. Further investigation of the controller is 3, pp. 218-238, 2001.
underway to enhance the slip tracking performance of the
proposed controller.
Sohel Anwar received his Ph.D. in Mechanical
Engineering from University of Arizona in 1995. Dr. Anwar
REFERENCES
worked as a consultant to Caterpillar, Inc. from 1995 to
1999 where he was responsible for the development of
1. Clarke, D.W, Mohtadi, C., Tuffs, P.S., “Generalized
prototype control systems for earthmoving machinery.
Predictive Control – Part I. The Basic Algorithm”,
Since 1999 he has been with the Chassis Advanced
Automatica, Vol. 23, No. 2, pp. 137-148, 1987.
Technology Department in Visteon Corporation as a
2. Watanabe, M. and Noguchi, N., “New Algorithm for
Product Design Engineer where his responsibilities
ABS to compensate for Road Disturbances”, SAE
include control system design & development, system
Transactions, Vol. 99, No. Sec. 6, pp. 271-279, 1990.
modeling & simulation, and software tool development. Dr.
3. Tan, H-S. and Chin, Y-K., “Vehicle Traction Control.
Anwar received his B.S.M.E. and M.Sc.E. from
Variable-Structure Control Approach”, J. Dynamic
Bangladesh University of Engineering & Technology
System, Measurement, & Control, Transactions of the
(B.U.E.T.) in 1986 and 1988 respectively. He taught at
ASME, Vol. 113, No. 2, pp. 23-230, 1991.
B.U.E.T. as a Lecturer in Mechanical Engineering from
4. Athan, T.W. and Papalambros, P.Y., “Multicriteria
1986 to 1988. He received his M.S.M.E. from Florida
Optimization of Anti-Lock Braking System Control
State University in 1990.
Algorithms”, Engineering Optimization, Vol. 27, No. 3,
pp. 199-227, 1996.
Behrouz Ashrafi received the B.S., M.S., and Ph.D.
5. Kokes, G. and Singh, T., “Adaptive Fuzzy Logic
degrees in Mechanical Engineering from Drexel University,
Control of an Anti-Lock Braking System”, IEEE
Philadelphia, PA, in 1983, 1985, and 1991, respectively.
Conference on Control Applications – Proceedings,
From 1992 until June 1998, Dr. Ashrafi was a Product
Vol. 1, pp. 646-651, 1999.
Design Engineer in the Electronics and Advanced Vehicle
6. Suh, M.W., Chung, J.H., Seok, C.S., and Kim, Y.J.,
Technology Divisions of Ford Motor Company, Dearborn,
“Hardware-in-the-Loop Simulation for ABS based on
MI. He is currently a Technical Specialist in the Chassis
PC”, International Journal of Vehicle Design, Vol. 24,
Advanced Technology Department of Visteon Corporation.
No. 2, pp. 157-170, 2000.
His current interests include vehicle dynamics, vehicle
7. Chen, F.W. and Liao, T.L., “Nonlinear Linearization
stability control, and application of estimation and control
Controller and Genetic Algorithm-Based Fuzzy Logic
theory to advanced automotive vehicles.
Controller for ABS Systems and Their Comparison”,
International Journal of Vehicle Design, Vol. 24, No. 4,
pp. 334-349, 2000.