IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 3, NO.

4, NOVEMBER 1995 381

A Fuzzy Logic Controller

for an ABS Braking System
Georg F. Mauer

Abstract- Anti-blocking system (ABS) brake controllers pose Commercial ABS braking systems have been on the market
unique challenges to the designer: a) For optimal performance, for more than 20 years [l]. A discussion of customary ABS
the controller must operate at an unstable equilibrium point, b) systems is found in the Bosch Automotive Handbook [2].
Depending on road conditions, the maximum braking torque may
vary over a wide range, c) The tire slippage measurement signal, A digital adaptive controller concept based on least squares
crucial for controller performance, is both highly uncertain and optimization has been developed by Landau and Dugard [3].
noisy, d) On rough roads, the tire slip ratio varies widely and Although an improvement over previous controller designs,
rapidly due to tire bouncing, and e) The braking system contains the adaptive method must sample a sufficient set of new
transportation delays which limit the control system bandwidth. data points after each change of plant parameters before
A digital controller design was chosen which combines a fuzzy
logic element and a decision logic network. The controller identi- it can compute the poles and zeros of the new controller.
fies the current road condition and generates a command braking This requirement introduces a significant additional delay of
pressure signal, based on current and past readings of the slip several sampling periods each time a new control law is
ratio and brake pressure. The controller detects wheel blockage computed. Fuzzy controllers, on the other hand, have an
immediately and avoids excessive slipping. The ABS system per- inherently parallel structure, which allows the controller to
formance is examined on a quarter vehicle model with nonlinear
elastic suspension. The parallelity of the fuzzy logic evaluation respond immediately once a new situation has been identified.
process ensures rapid computation of the controller output signal, Detailed discussions of fuzzy control theory, its advan-
requiring less time and fewer computation steps than controllers tages, and theoretical foundations is found, among others, in
with adaptive identification. The robustness of the braking system Mamdami [4], Kosko [ 5 ] , Cox [6], and Jamshidi et al [7].
is investigated on rough roads and in the presence of large Fuzzy controllers differ from model-based controllers in that
measurement noise. This paper describes design criteria, and the
decision and rule structure of the control system. The simulation they encode heuristic knowledge. When properly designed, the
results present the system’s performance on various road types performance of fuzzy controllers compare favorably with that
and under rapidly changing road conditions. of advanced model-based digital controllers ( [ 5 ] ,[7]). Fuzzy
controllers can be designed for a large range of plants, includ-
ing systems with unstable open-loop poles [5], [8]. Adaptive
I. INTRODUCTION fuzzy controllers are described by Lin-Xin [9]. Despite the

T HE goal of anti-blocking system (ABS) braking is to absence of analytical modeling information, systems governed
generate in real time the largest possible brake force while by fuzzy controllers are often highly robust. Learning fuzzy
keeping the vehicle maneuverable and avoiding excessive controllers are described by Layne and Passino [lo]. The
wheel slippage. The absence of sensors capable of detecting application of fuzzy control, including fuzzy model reference
on-line measurement variables, such as tire slip ratio and road learning, to automotive ABS braking is discussed in Layne
condition, forces the ABS system designer to seek means to et al [111. Here, the plant to be controlled is modeled as a
infer the road condition from the few available sensory data. first-order linear dynamic system without transportation delay.
The most important process parameters affecting the quality The authors show that the learning component can adjust
of control are the coefficient of friction between tire and the controller immediately to rapidly varying road conditions.
road, the tire slip ratio, and the vertical force on the wheel. A performance study of the fuzzy model reference learning
Generally, a high degree of uncertainty is associated with controller with a more realistic plant model would be desirable
the measurement of the slip ratio between wheel and road to better assess the method’s potential for ABS applications.
surface. This problem is compounded by the sometimes rapid The objective of the present paper is to investigate and analyze
variation of road conditions with its attendant large variations the benefits of designing a fuzzy logic controller for ABS
of friction coefficients, slip ratio and vertical contact force braking.
between tire and road surface. The plant to be controlled
(elastically suspended wheel, braking servo system, actuator) 11. THE PLANT
comprises significant transportation delays which limit the Scope of Discussion: The system to be controlled is repre-
frequency response of the controller. sented as the front wheel of a vehicle in straightforward motion
on a plane road. Motion dynamics arising from rotation of the
Manuscript received March 10, 1994; revised February 28, 1995. vehicle about the vertical axis, or from uneven braking forces
The author is with the Department of Mechanical Engineering, University
of Nevada, Las Vegas, NV 89154-4027 USA. on other wheels, are not considered. In straight line braking,
IEEE Log Number 9412918. lateral tire forces do not exist. The model presented below
1063-6706/95$04.00 0 1995 IEEE
382 IEEE TRANSACTIONS ON FLJZZY SYSTEMS, VOL. 3, NO. 4, NOVEMBER 1995

Vertical f F~ Honzontal Components

Components

Fig. 1. Quarter vehicle model 500

does include, however, the effects of road roughness on the

braking process.
Quarter Vehicle Model: Fig. 1 shows schematically the dy-
namic system model of the quarter vehicle. Vertical model
Computer
components on the left side of Fig, 1 are represented by the
suspended quarter vehicle mass, M , and the unsuspended - Presswe conttol
Operanon and Safety Checks
j '
mass, m, coupled by nonlinear springs and dampers K I , I

Kz, K T , B1,and Bz.Vertical system variables are the road

roughness, 20, representing an external input, and the vertical
displacements of m and M , zl,and 2 2 . In the honzon-
tal motion model, the suspended modified quarter vehicle
mass, M L , is coupled to the wheel having inertia I through Fig. 3. ABS brake schematic.
longitudinal elasticity KL. The vehicle mass is modified in
proportion to the bending moment FG h / L discussed below, braking torque beyond the optimum results in increased tire
which places a larger vertical force on the forward wheels slippage and reduced braking force. The peak braking force
and a proportionally lesser one at the rear. State variables at 8% slip ratio thus represents an unstable equilibrium point.
are the vehicle's horizontal forward travel, IG, and velocity, To maintain the slip ratio at the desired high braking force
w . The braking torque, Tb, applied to the wheel produces therefore requires continuous control of the braking torque.
a reaction force, FB, at the point of contact with the road A 100% slip ratio implies wheel locking, i.e., the vehicle is
surface. In the steady state, variables zo and 21 are zero, and no longer steerable. As is evident from the characteristics of
the nominal wheel radius is Eo. The effective radius, R(t) Fig. 2, the ABS controller must identify the road condition,
during vehicle motion is then R(t) = Ro - zo z1. Assuming + i.e., select the correct characteristic, and then maximize the
that the vehicle's center of gravity is at height h above the road braking force.
surface and that the front and rear wheel centers are distance Brake System Model: Customary ABS braking systems
L apart, the braking force produces a moment which results in employ a sensor on each wheel for detection of wheel angular
an additional vertical force at the front, FSh/L. F, represents velocity and servovalves which allow the modification of
the total vertical reaction force. The dynamic system model is the brake cylinder hydraulic pressure as determined by the
then found using standard Newtonian or Lagrange methods. central brake controller (see Fig. 3). The controller seeks to
The quarter vehicle plant was modeled in the Matrix-x [14] optimize the braking performance of each wheel, based on the
graphical user environment. estimated slip ratio. The hydraulic system (master cylinder,
Tire Model: The input variables to the tire model are the pressure modulating valves, fluid flow through pipes, and
total applied vertical force, F,, and the friction coefficient wheel brake cylinders) and the brake pad travel create a
between tire and road surface. A model for the longitudinal transportation delay or dead time, which is generally on the
slippage and breaking force is described in Pacejka and Sharp order of 20-50 milliseconds. The transportation delay in the
[12] and Bakker et al [13]. This tire model was used in the brake system implies a severe limitation of the control system
simulation. The tire slip ratio is defined as bandwidth. Current control computers operate at a sampling
rate of approximately 5 ms. ABS designs are described, among
Slip Ratio =
w - w * R ( t ) * 100 others, in Luetteke [15], Yagi et al [16], and Beyer et a1 [17].
71
III. CONTROLLER
CONCEPT
Detection of Slip Ratio and Road Condition: All existing
The variable w represents the angular velocity of the wheel. A ABS braking controllers are severely limited by the fact
zero slip ratio indicates perfect rolling without slipping. Fig. 2 that no practical means exist for the direct measurement
shows the front wheel braking force as a function of the slip of a) the current coefficient of friction between tire and
ratio for a passenger car with a total mass of approximately road surface and b) the tire slip ratio. The slip ratio is
1700 kg. At first, braking increases the slip ratio and the inferred from the comparison of the deceleration of all four
contact force between road and tire. On a dry road, braking wheels and sometimes also by comparison with an on-vehicle
is optimal at approximately 8% slip ratio. Application of a accelerometer. The road condition (coefficient of friction
MAUER: A FUZZY LOGIC CONTROLLER FOR AN ABS BRAKING SYSTEM 383

between tire and road) can be inferred by observing the Uneven Road Sampling Time = 5 ms
30
slip ratio resulting from a given braking force. If, for instance, I
the vehicle is observed to be slipping excessively at a braking
force of 1500 N under the conditions of Fig. 2, we would
conclude that the road surface is icy. 60 I I

Controller Design Considerations: The controller must be

able to vary its gain and reference slip ratio over a wide range,
with the maximum braking force on dry pavement at four times
the level permissible on ice. Due to the fundamentally parallel
structure of the fuzzy logic controller all rules can be active
simultaneously. The large range of plant parameter variations
n .9
at different road conditions makes this evaluation critical, all
the more since the rules to be fired must remain consistent,
a .6
.3
i.e., the controller must prevent antagonistic or contradictory
rules. For example, difficulties arise during braking on ice, 0 .2 .4 .6 .8 1 1.2 1.4 1.6
TIME in seconds
where the activation of even a small portion of a high-gain, FUZZY ABS BRAKING, Elastic Suspension, Quantizing Actuator
dry road rule tends to result in brake force command outputs
that generate too much tire slippage. 1 ;. 4. Fuzzy logic ABS control

Fig. 4 shows fuzzy logic ABS control in which a set of

eight rules covers braking on dry and icy roads. The two upper INPUT
plots show vehicle velocity and slip ratio, and the bottom plots VARIABLES
show Rule 1 and the aggregate fuzzy controller output, both h
Tire Slip j U1
normalized to a range between zero and one. Defuzzification is
performed by Mamdami's method as the centroid of minima.
Rule 1 initiates braking when no slippage is present and is
formulated as: If the average slip ratio is zero, and if the slip
ratio is zero, then the braking torque is large. Here, the average
slip ratio was chosen as the arithmetic mean of the past nine
samples. The averaging over nine samples smooths out the Z"
- V
inherently noisy slip ratio signal, without causing undue delays
in the event of road condition changes. The inspection of Rule
1 in Fig. 4 shows that Rule 1 is active-as it should be-when
braking starts at a low slip ratio. The center section of Fig. 4,
between 0.5 s and 1 s, shows braking on ice. As the slip
Fig. 5. Decision logic state diagram.
ratio declines to almost zero, rule 1 turns active to a small
amount. The result is a comparatively large braking torque
and excessive slippage on ice. By adding further conditions to with only a few discrete output levels. It is not useful,
Rule 1, one can reduce its output magnitude during braking on therefore, to discriminate among more than a few discrete road
ice. It was not possible, however, to eliminate the excessive conditions. The decision logic shown in Fig. 5 compares the
braking torque on ice without making Rule 1 ineffective on braking torque with the detected slip ratio. If the current slip
dry roads. ratio differs widely from the estimate, based on the current
Conceivably the problem of unwanted rule activation can assumption on road conditions and braking torque, or if wheel
be remedied by a two-stage fuzzy controller which would first blockage is detected, a new output state is selected.
generate reference values for controller gain and reference slip Wheel blockage is defined as a slip ratio exceeding a preset
ratio, and which would then apply the control law with the limit not encountered during normal operation (14% on dry
current gain and reference values. To avoid control loop insta- roads, less on wet and icy roads). Decision logic output states
bilities, the eigenfrequencies governing the rate at which gain are dry, wet, ice, and blockage, denoting three possible road
and reference are adjusted must be significantly lower than the conditions, and wheel blockage due to excessive braking force.
plant's transportation delay. Simulations with a variable gain Only one condition can be true at a time. This means that the
two-stage fuzzy ABS controller showed that the achievable fuzzy controller contains a distinct set of rules for each road
rate of gain adjustment after a change of road condition was condition and only activates the appropriate set. A transition
so slow as to make it impractical. from one state to another occurs only if wheel blockage occurs
Fuzzy Logic Controller with Discrete Road Condition Iden- or if the slip ratio has had its values out of the preselected
tification: A discrete logic element (finite state machine) can threshold levels for more than nine sampling periods.
perform the road condition identification for the fuzzy logic Slip Ratio Prediction: The braking system induces a delay
controller (see Fig. 5). The discrete identifier cannot detect of about 35 ms, or seven sampling periods at the selected
gradual changes of road condition such as a transition from dry rate of five ms. A predicted slip ratio is computed from the
to wet pavement. Customary ABS systems employ actuators numerical derivative of recent measurement samples and made
384 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 3, NO. 4, NOVEMBER 1995

1 Fuuy Output
9
.a
m .7
6 .6
L
E .5
!2 .4
1 .3
2 0.
1 Slip % 20. 10. Brake Torque
‘0 .1 .2 .3 .4 .5 6 7 .8 .9 1
Class Range
Fig. 7. Dry road fuzzy control surface: Output‘ signal as function of brake
Fig. 6. Definitions of membeship classes torque and slip ratio.

available to the fuzzy logic controller and to the decision logic

which identifies the road condition. The slip ratio predictor
can partially compensate for the effect of the aforementioned
transportation delay. In Figs. 12 and 13, for instance, we
observe after 0.5 s a braking torque of zero while the wheel
stalls on ice. The predictor causes the controller to issue an
appropriately small braking torque command signal several
sampling periods before the actual recovery from tire stall.
Fig. 8. Dry road fuzzy control surface: Output signal as function of predicted
and actual slip ratio.
IV. FUZZYLOGICCONTROLLER
DESIGN
Fuzzy Output
The fuzzy logic controller processes the four logic inputs
4.
from the road condition identifier, the current and predicted
slip ratio, and the braking torque command signal. The con-
troller evaluates a set of 11 rules. Four rules govern braking
on dry roads, four on ice, two on wet pavement, and one
governs wheel blockage. Only one set of rules is active at a
0.
time. Three input variables are continuous or “crisp,” namely
the slip ratio variable GLISS, the predicted slip ratio variable
GLPRED, and the braking torque variable couple. The range Fig. 9. Icy road fuzzy control surface: Output signal as function of brake
torque and slip ratio.
of input values for control is zero to 15% for the actual and
predicted slip ratio. Although both can vary between zero and
loo%, the upper range of 15% represents the useful upper signal on dry roads will vary between “medium” and “large,”
limit for control. The crisp output variable is named “DGL;” generating tire slip between 7.5% and approximately 10%. As
it represents the braking torque. The output of each rule is the shown in Figs. 7 and 8, the output is a maximum at low actual
minimum of the belief in the membership of each function. and predicted slip ratios. As the braking torque and slip ratio
Output defuzzification is performed by computing the centroid increase, however, the controller output signal dips to 7.5 (the
of all minima according to Mamdami’s method. Each crisp trough in Fig. 7). Similarly, large predicted slip ratio values
variable is defined in terms of six fuzzy classes ranging from reduce the output signal further (Fig. 8) to below three.
“zero” to “very large.” Fig. 6 shows the class definitions. Icy and Wet Road Rules: On ice, the optimal slip ratio is
Each membership function is defined as a sinusoid. Braking 2% (see Fig. 2). The fuzzy control surface of Fig. 9 provides
torque input and fuzzy controller output both are scaled so that for moderate braking at low slippage, with further gradual
they have a range from zero to 10, where 10 represents the reductions to 10% of maximum torque as the slip ratio
maximum available braking torque. The four logical decision increases to 4% or above. The control surface of Fig. 9
output variables (blockage, ice, wet, dry) assume values of shows an essentially flat region bounded by slip ratio values
zero or one only. A delay of one sampling period is merted ranging from 2-5%, and braking torque values from 0-3%,
between the controller output and its return to the fuzzy input, in which the output signal is slightly above one, producing a
so as to avoid an algebraic loop. The fuzzy logic rules are braking torque appropriate for ice. The road condition can
listed in the appendix. only be inferred from the relation between braking torque
Dry Road Rules: On dry roads, the braking torque is op- and slippage. While the fuzzy controller can deliver near-
timal when the slip ratio is at or near 8% (see Fig. 2). The optimum performance on ice as well as on dry surfaces, the
control surfaces of Figs. 7 and 8 show the dry road fuzzy only available means to test for a possible change of road
controller characteristic. The controller output is an aggregate condition is to apply a higher braking torque periodically and
of the two control surfaces. At low slip ratio, the braking to observe the resulting slippage. The excess braking torque
torque is high, but declines by about 20% as the slip ratio induces wheel blockage on ice, but not on wet or dry roads.
increases from 7% to 9%. Above 10% slippage, the controller As Fig. 9 indicates, this is done by increasing the fuzzy output
output is zero. As the dry rules indicate, the controller output to a value up to four as the slip ratio decreases.
MAUER A FUZZY LOGIC CONTROLLER FOR AN ABS BRAKING SYSTEM 385

Even Road, Sampling Perlod=Sms

Fig. 10. Wet road fuzzy control surface: Output signal as function of brake
torque and slip ratio.

9 5i
4
0 .2 .4 .6 .8
TIME in seconds
1 1.2 1.4

F U M ASS BRAKING, Elastic Suspension, Quantizing Actuator

1.6

Fig. 12. Fuzzy logic ABS control, smooth road, ice between 0.5 s and 1.0 s.

employ the same scenario concerning road condition: The road

output identify current m a d
Condition Based on
is initially dry for the first half second. Between 0.5 and 1.0
(Brake Torque) Data f r o m Prev~ous
seconds, the road is icy, reverting again to dry conditions
Fuzzy Controller
thereafter.
I l I Figs. 12 and 13 show simulation results on flat and rough
U roads, respectively. The first dry road segment of the simu-
Fig. 11. Closed-loop ABS braking control. lation in Fig. 12 shows near-optimal braking after an initial
small oscillation, visible in the speed record, the attendant
excessive slippage, and an oscillatory vertical force. The
The wet road rules result in an essentially proportional
controller responds to the excess slippage by briefly reducing
controller (Fig. 10).
the braking torque. Upon transition to ice conditions in the
center section, the decision logic recognizes the onset of
V. COMPLETE VEHICLEBRAKING SYSTEMMODEL blockage at once, and the braking torque reduces to zero as
Fig. 11 shows the complete simulation model. The plant rapidly as the system permits. Wheel slippage is limited to
dynamics, including the brake servoamplifier and the brake, a maximum of 50%, which compares favorably with model-
are modeled as continuous elements. The control block mainly based adaptive systems, in which full wheel blockage can
consists of the road condition identifier and the fuzzy con- occur before returning to normal operation. Note also the
troller. Customary ABS actuators act only in discrete intervals effect of the slip rate prediction: the actuator applies a low
(quantizing control). The discretization of fuzzy controller braking torque before the slip ratio has returned to zero, thus
output therefore reflects some constraints enforced by the reducing the time spent in recovery from blockage. The third
actuator design. In this application, seven quantization levels segment shows the response to the return of dry conditions.
have been used. The decision logic identifies a dry road as having a low slip
ratio in the presence of a moderate torque. As seen in Fig. 12,
VI. SIMULATION RESULTS the braking torque is rapidly restored to the previous high level.
Fig. 13 shows the braking performance of the same vehicle
Labeling of Graphs: Figs. 12, 13, and 15-17 depict the and quantizing controller on a rough road. The two bottom
performance of ABS braking systems with various configu- strips show the variation of the vertical force in Newtons and
rations. In each graph, the ordinates denote the following:
the road roughness in meters. The wheel is in physical contact
with the road at all times. Again, ice is encountered in the
Label Definition center segment. The rough road causes large variations of the
Speed Vehicle translational velocity in vertical force and therefore considerable random variations of
meters per second. the slip ratio. In Fig. 13, only the largest vertical accelerations
Slip Slip ratio in %. are seen to have an impact on the braking torque. A brief drop
Torque Braking torque in Newtonmeters. of the braking torque in the third segment at 1.15 s elapsed time
Force Vertical contact force between tire and road, is the only instance of misinterpretation of the road condition.
i.e., F, in Fig. 1, in Kilonewtons. Sensitivity to Noisy Slip Ratio Measurements: Since slip
Fifth ordinate in some graphs: ratio measurement data are notorious for uncertainty and
2-Road Vertical road roughness in meters as function noise, the fuzzy controller was tested for its robustness to
of distance traveled. noisy measurement data. A zero mean random noise sequence
All simulations start at an initial velocity of 30 m / s (106 M). was added to the slip ratio signal at the digital controller input.
The actuator braking torque range ranges from 0 to 1500 Nm, The noise signal is shown in Fig. 14. Noise signal amplitudes
with five equally spaced intervals in between. All simulations reach values up to f 2 % slippage. This noise level represents
386 IEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 3, NO. 4, NOVEMBER 1995

Uneven Road Sampling Period = 5 ms Rough Road, Sampling Period E 5ms

30

25

:1
20

0
I 2wo
E
6 5
4
1wo -1
P
01 F 0 "

0 V\n A

- 01
0 .2 .4 .6 .8 1 1.2 1.4 1.6
TIME in Seconds
FUZZY ABS BRAKING, Elastic Suspenslon, Quantizing Actuator

Fig. 13. Fuzzy logic ABS control, uneven road, ice between 0.5 s and 1.0 s.
Fig. 15. Fuzzy logic A B S control with noisy slip ratio measurement, uneven
road, ice between 0.5 s and 1.0 s.
1.5 I

Rough Road, Elastic Suspension

z

-2.5"'" " " ' ' 1

0 .2 .4 .6 .E 1
TIME in seconds
1.2 1.4 1.6
-
8 6
Fig. 14. Random noise signal. 0
$ 4
U.
7
0 10 20 30 40 50 60 70
Vehicle Travel in m
approximately 33% of the slip ratio on a dry road and is ABS BRAKING with PI Controller ice between 14and 26 m
on par with the slip ratio signal amplitudes during braking
Fig. 16. ABS with P I controller, quantizing actuator, rough road.
on ice. Fig. 15 shows the fuzzy controller's response to the
noisy input signal. The comparison with Fig. 13 shows that
the control system's performance is virtually unaffected by the being controlled is the quarter vehicle with quantizing actuator.
signal noise. The only degradation is observed during braking Fig. 16 shows braking on a rough road with PI control. From
on ice, where a negative slippage signal prompts the controller an initial forward speed of 30 d s , the vehicle comes to a
to generate an excessively large command signal for 0.02 s full stop at a braking distance of 61 meters. When using the
duration. Another form of noise is present when braking on a fuzzy-logic controller under identical operating conditions, the
rough road: the unpredictable variations of the vertical contact braking distance is shortened to 54 meters, as seen in Fig. 17.
force, F,, between tire and road create large variations of the Closer inspection of Figs. 16 and 17 shows that the PI-type
braking force, and thus large slip ratio variations. Figs. 13 ABS control tends toward more oscillations than the fuzzy
and 17 show slip ratio variations to 15 and 20%, respectively, logic based control system. This effect is particularly visible
during braking on uneven, dry road segments. In Fig. 17, the during the last third of the braking curve, where the road
controller maintains an appropriate braking torque during the roughness amplitudes and frequency are higher. It should also
first 50 meters of braking distance with only two instances (at be noted that, for equal road conditions and braking torque,
29 and 49 meters distance, respectively) in which the command the slip ratio, defined as (v - u R ) / v ,assumes larger numerical
signal is briefly reduced to zero because of excessive slippage. values as v approaches zero. Note the increased slip ratio and
Overall, the fuzzy ABS controller exhibits remarkably low vehicle speed variations in the PI controlled system at slow
sensitivity to the large signal noise levels caused by the road vehicle speeds.
roughness.
Comparison with Linear Control: The performance of the VII. CONCLUSION
fuzzy-logic controller was compared with that of a discrete A fuzzy logic controller design for ABS braking was pre-
PI controller. Gain and integrating time constant of the PI sented. A series of simulations was conducted with quantizing
controller were selected such that excessive output signal actuators. The simulations were conducted with a quarter
oscillations were avoided, while preserving rapid controller re- vehicle model which includes a nonlinear model of the vehicle
sponse to changes of road condition. In both cases, the system suspension. The simulations showed that a fuzzy controller
 ~

MAUER A FUZZY LOGIC CONTROLLER FOR AN ABS BRAKING SYSTEM 387

Rough Road, Elastic Suspension

40 L I
GLISS IS ZS AND;
20

0
100

50
1 COUPLE IS ZS;
THEN DGL IS ZS;
RULE-7;
6. RULE ICES;
IF GLISS IS ZERO AND;
0
2000 ICE IS TRUE;
1000
THEN DGL IS SMALL;
RULE-5;
0
7. RULE ICE8;
IF GLISS IS SMALL AND;
ICE IS TRUE;
0 10 20 30 40 50 60
THEN DGL IS ZERO;
Vehicle Travel in m RULES;
FUZZY Logic ABS Braking b e between 14and 26 m
8. RULE BLOCKAGE;
Fig. 17. Fuzzy logic ABS control, rough road. IF GLISS IS VLARGE AND;
GLPRED IS VLARGE;
THEN DGL IS ZERO;
in combination with a decision logic for estimation of the
road condition is a rapid and effective means to provide RULE-9;
9. RULE WET10;
braking torque control over operating conditions ranging from
dry pavement to black ice. The controller was found to be
IF WET IS TRUE AND;
GLISS IS ZS AND;
generally quite robust, and largely insensitive to internal signal
GLPRED IS NOT LARGE;
noise, as well as to large noise amplitudes induced by rough
road profiles. The fuzzy controller’s response time to sudden THEN DGL IS SMALL;
RULE- 10;
changes of road condition compares favorably with that of
model-based adaptive control schemes. 10. RULE WET1 1;
IF WET IS TRUE AND;
GLISS IS SMALL;
APPENDIX THEN DGL IS ZS;
RULE-11;
A. Fuzzy Logic Rules RULE WET12;
1. RULE DRY1; IF WET IS TRUE AND;
IF DRY IS TRUE AND; GLISS IS ZERO AND;
GLPRED IS NOT VLARGE; GLPRED IS NOT LARGE;
THEN DGL IS LARGE; THEN DGL IS SMALL;
RULE-1; RULE-12.
2. RULE DRY2;
REFERENCES
IF GLISS IS LARGE AND;
DRY IS TRUE AND; W. Douglas and T. C. Schafer, “The Chrysler sure brake: The first
COUPLE IS LARGE; production-four-wheel anti-skid system,” SAE Tech. Paper 710248,
1971.
THEN DGL IS MEDIUM; Automotive Handbook, Robert Bosch GmbH. Dusseldorf, Germany:
RULE2; VDI Verlag.
I. D. Landau and L. Dugard, Commande Adaptative: Aspects Pratiques
3. RULE DRY3; et Thiorktiques. Paris: Masson, 1986.
IF GLISS IS SMALL AND; E. H. Mamdami, “The application of fuzzy set theory to control
DRY IS TRUE AND; systems-A survey,’’ in Fuzzy Automata and Decision Processes. Am-
sterdam: North Holland, 1977, pp. 77-88.
COUPLE IS LARGE AND; Bart Kosko, “Neural networks and fuzzy systems,” in A Dynamical
GLPRED IS NOT VLARGE; Systems Approach. Englewood Cliffs, N J Prentice-Hall, 1992.
THEN DGL IS LARGE; Earl Cox, “Fuzzy fundamentals,” ZEEE Spectrum, vol. 29, no. 10, pp.
58-61, Oct. 1992.
RULE-3; M. Jamshidi, N. Vadiee, and T. J. Ross, Eds., Fuzzy Logic and Control.
4. RULE DRY4; Englewood Cliffs, NJ: Prentice-Hall, 1993.
G. F. Mauer, “computer-supported controller design with model-based
IF GLISS IS MEDIUM AND; and fuzzy-logic methods,” in Proc. 6th IAR Colloquium Fuzzy Syst. Lean
DRY IS TRUE AND; Production, Duisburg, Germany, Nov. 1992, pp. 151-162
GLPRED IS NOT VLARGE AND; Wang Lin Xin, “Stable adaptive fuzzy control of nodinear systems,”
ZEEE Trans. Fuzzy Syst., vol. 1, no. 2, May 1993.
COUPLE IS LARGE; J. R. Layne and K. M. Passino, “Fuzzy model reference learning
THEN DGL IS LARGE; control,” in Proc. 1st ZEEE Con$ Contr. Appl., Dayton, OH, Sep. 1992,
RULE-4; pp. 68-91.
J. R. Layne, K. M. Passino, and S. Yurkovich, “Fuzzy learning control
5. RULE ICE7; for antiskid braking systems,” ZEEE Trans. Contr. Syst. Tech, vol. 1, no.
IF ICE IS TRUE AND; 2, pp. 122-129, June 1993.
388 JEEE TRANSACTIONS ON FUZZY SYSTEMS, VOL. 3, NO. 4, NOVEMBER 1995

[12] H. B Pacejka and R. S. Sharp, “Shear force development by pneumatic Georg F. Mauer received the Ph.D. degree in me-
tyres in steady state conditions: A review of modeling aspects,” Vehicle chamcal engineenng from the Technical Umversity
Syst. Dynamics, vol. 20, pp. 121-176, 1991. of Berlm, West Germany, in 1977
[13] E Bakker, H. B. Pacejka, and L Lidner, “A new tire model with an He is a Professor of Mechanical Engineenng at
application in vehcle dynamics studies,” SAE Paper 890087, 1989. the Umversity of Nevada, Las Vegas, where he
[14] Integrated Systems, Inc., “Matrix-x user gmde,” Santa Clara, CA. jomed the faculty in 1986. His research interests
[15] H. Luetteke, “Anti-lock hydraulic brake system,” U S . Patent 5 282 677, mclude intelligent, computer-basedsensing and con-
1994. trol systems. He i s currently workmg on automotive
[16] E. Yagi, N.Inoue, and S Siokawa, “Brake control system for automohve measurement and control systems, on innovative
vehicles,” U S. Patent 5 273 348, 1993. methods in dynarmc systems and control instmction,
[17] C. Beyer, P. Dominke, and E. Sonntag, “Anti-lock control system,” and on sensor-based decision and control for robots,
U.S. Patent 5249852, 1993 mcludlng AI and fuzzy logic methods for object recognihon.