Fuzzy Application Library/Technical Applications/AntilockBraking System
AntilockBraking System and Vehicle Speed Estimation using Fuzzy Logic
by Ralf Klein, Corporate R&D Siemens, Munich; Armin Daiß, Institute of Industrial Information Systems, Karlsruhe; Herbert Eichfeld, Corporate R&D Siemens, Munich; email: ralf.klein@zfe.siemens.de
Citation Reference: This paper has been published with the 1st Embedded Computing Conference, October 1996, Paris and has been presented at the 6. Aachener FuzzySymposium, November 1996
A study and tests on an experimental car are presented dealing with antilockbraking system (ABS) and vehicle speed estimation using fuzzy logic. Vehicle dynamics and braking systems are complex and behave strongly nonlinear which causes difficulties in developing a classical controller for ABS. Fuzzy logic, however facilitates such system designs and improves tuning abilities.The underlying control philosophy takes into consideration wheel acceleration as well as wheel slip in order to recognize blocking tendencies. The knowledge of the actual vehicle velocity is necessary to calculate wheel slips. This is done by means of a fuzzy estimator, which weighs the inputs of a longitudinal acceleration sensor and four wheel speed sensors. If lockup tendency is detected, magnetic valves are switched to reduce brake pressure. Performance evaluation is based both on computer simulations and an experimental car. To guarantee realtime ability (one control cycle takes seven milliseconds) and to relieve the electronic control unit (ECU), all fuzzy calculations are made by the fuzzy coprocessor SAE 81C99A. Measurements in the experimental car prove the functionality of this automotive fuzzy hardware system.
Introduction
Fuzzy Control, a relatively new, intelligent, knowledge based control technique performs exceptionally well in nonlinear, complex and even in not mathematically describable systems. Thus the use of fuzzy logic for an antilockbraking system (ABS) seems to be promising.
AntilockBraking Systems
The aim of an ABS is to minimize brake distance while steerability is retained even under hard braking. To understand the underlying physical effect which leads to wheelblocking during braking, consider Figure 1a: Coefficient of friction is shown as a function of wheel slip, relating to the terms given in Figure 1b.
Figure 1: a) Friction characteristics  
b) Wheel model FZ: Wheel load R: Wheel radius w: Angular wheel frequency v: Velocity of wheel center FL: Longitudinal force 
Calculating the wheel slip by  ^{, }  the longitudinal wheel force results in  . 
At the beginning of an uncontrolled full braking, the operating point starts at s = 0, then rises steeply and reaches a peak at s = s _{max}. After that, the wheel locks within a few milliseconds because of the declining friction coefficient characteristic which acts as a positive feedback. At this moment the wheel force remains constant at the low level of sliding friction. Steering is not possible any more.
Therefore a fast and accurate control system is required to keep wheel slips within the shaded area shown in Figure 1a.
Vehicle Speed
A crucial point in the development
of wheel slip control systems is the determination of the vehicle
speed. There are several methods possible: until now the velocity
is measured with inductive sensors for the wheel rotational
speed. Especially in the case of brake slips the measured speed
does not correspond with reality. To obtain very accurate
results, optical or microwave sensors take advantage of a
correlation method. However, these sensors are very expensive and
will not be used for ABS.
Sensors and Actuators
The experimental car was fitted with sensors and actuators shown in Figure 2. Each wheel is connected to a metallic gearwheel, which induces a current within an attached sensor. The frequency of the rectangular shaped current is proportional to the angular frequency w _{i, j} and can be evaluated by a microcontroller. In addition to common ABS fitted cars, a capacitive acceleration sensor for measuring the longitudinal acceleration a_{x} is implemented.
Furthermore Figure 2 depicts the hydraulic unit including main brake cylinder, hydraulic lines and wheel brake cylinders. By means of two magnetic twoway valves each wheel, braking pressure p_{i, j} is modulated. Three discrete conditions are possible: decrease pressure, hold pressure firm and increase pressure (up to main brake pressure level only). Each valve is hydraulically connected to the main brake cylinder, to the wheel brake cylinders and to the recirculation.
Figure 2:
Sensors and actuators of the experimental car

Estimation of Vehicle Speed Using Fuzzy Logic
As described in the first chapter, the knowledge of the actual vehicle speed over ground is vital in order to calculate wheel slips correctly. Daiß and Kiencke [1] presented an estimation system based on KalmanFilter which performs well, but is not suitable because of very high performance requirements. In this approach the speed estimation uses multisensor data fusion that means several sensors measure vehicle speed independently and the estimator decides which sensor is most reliable. Figure 3 represents the schematic structure of the fuzzy estimator. The signals of the four wheel speed sensors w _{i,j }are used as well as the signal of the acceleration sensor a_{x}.
Figure 3: Estimation of car velocity (large) 
In a data preprocessing block the measured signals are filtered
by a lowpass and the inputs for the fuzzy estimator are
calculated: four wheels slip ,
and an acceleration value D v_{a}. The applied formulas
are:
and  , 
whereby a_{Offset} is a correction value consisting of an offset and a road slope part. It is derived by comparing the measured acceleration with the derivative of the vehicle speed v _{Fuz}, which is calculated with the fuzzy logic system. After this subtraction, the signal is lowpass filtered to obtain the constant component a_{Offset}^{.} v _{Fuz}(k1) is the estimated velocity of the previous cycle. A timedelay of T is expressed by the term 1/z.
The fuzzy estimator itself is divided into two parts. The first (Logic 1) determines which wheel sensor is most reliable, and the second (Logic 2) decides about the reliability of the integral of the acceleration sensor, shown in Figure 4. This cascade structure is chosen to reduce the number of rules.
Figure 4: Stucture of the fuzzy estimator (large) 
Starting at block „Logic 1" and „Logic 2" the
crisp inputs are fuzzificated. Figure 5 shows the
inputmembershipfunctions (IMF) with four linguistic values (Negative,
Zero, Positive and Very_Positiv).
Figure 5: Input membership functions (large) 
The rule base consists of 35 rules altogether. To classify the
present driving condition vehicle acceleration is taken into
consideration. This should be explained for three situations:
 D v_{a} Positive: Braking situation, all wheels are weighted low because of wheel slips appearing.
 D v_{a} Zero: If wheel speeds tend to constant driving the acceleration signal is low weighted in order to adjust the sensor.
 D v_{a} Negative: The experimental car was rearwheel driven therefore rear wheels are less weighted than front wheels.
Figure 6: Output membership functions (large) 
Figure 6 depicts the outputmembershipfunctions (OMF). Here,
three linguistic values are sufficient. The output of the
estimation is derived as a weighted sum of the wheel measurement
plus the integrated and corrected acceleration:
.
The FuzzyABS Algorithm
The FuzzyController uses two input values: the wheel slip S_{B}:
and the wheel acceleration: 
with wheel speed v_{Wheel} and vehicle speed v_{Fuz}, which is given by the FuzzyEstimator.
The input variables are transformed into fuzzy variables slip and dvwheel/dt by the fuzzification process. Both variables use seven linguistic values, the slip variable is described by the terms
slip = {zero, very small, too small, smaller than optimum, optimum, too large, very large},
and the acceleration dvwheel/dt by
dvwheel/dt = {negative large, negative medium, negative small, negative few, zero, positive small, positive large}.
As a result of two fuzzy variables, each of them having seven labels, 49 different conditions are possible. The rule base is complete that means, all 49 rules are formulated and all 49 conditions are allowed. These rules create a nonlinear characteristic surface as shown in Figure 7.
Figure 7: Fuzzy characteristic surface (large) 
Using this characteristic surface, the two fuzzy input values slip
and dvwheel/dt can be mapped to the fuzzy
output value pressure. The labels for this value are:
pressure = {positive fast, positive slow, zero, negative slow, negative fast}
The structure of the fuzzy ABS controller is shown in Figure 8.
Figure 8: Structure of the fuzzy ABS controller (large) 
The optimal breaking pressure results from the defuzzification of the linguistic variable pressure. Finally a threestep controller determines the position of the magnetic valves, wether the pressure should be increased, hold firm or decreased.
Figure 9 summarizes the total amount of fuzzy calculations. Numbers within a rectangle indicate the quantity of fuzzy rules.
Figure 9: Fuzzy calculations (large) 
It should be noted that linguistic variables and rule tables can
be designed with numerical optimization methods, for example
described in [2]. In this work they were created using expert
knowledge and analysis of measured data during ABS braking
action.
Simulation of a Full Braking
After implementation of the whole system in SIMULINK, a full braking on highm road was carried out, with and without the fuzzy ABS.
Without fuzzy ABS the braking pressure reaches a very high level and the wheels block within short. This results in an unstable behavior, the vehicle cannot be steered any more and the stopping distance increases.
With fuzzy ABS controller activated, steerability is not only retained during the whole braking maneuver, but the slowing down length was considerably shortened as well.
The following graphs show the steady decline of the vehicle speed, the fluctuating decline of the wheel speed of the left front wheel as an example and the fluctuating level of the wheel slip. The applied braking pressure is depicted in the last diagram. The other wheels behave approximately similar.
Figure 10a: Simulations of a full braking (large)  
Figure 10b: Simulations of a full braking (large)  
Figure 10c: Simulations of a full braking (large) 
Implemenation of the Fuzzy ABS Controller
The fuzzy ABS controller uses the microprocessor SAB 80C166 together with the fuzzy coprocessor SAE 81C99A [3]. Due to the implementation of Fuzzy algorithms into the hardware of the coprocessor, the calculation speed of the host processor increased significantly. While the control cycle time was set to a standard value of 7 msec, the computation time was only 0.5 msec! This offers facilities for implementation of extended vehicle dynamics control [4].
The flexibility of the coprocessor [5] is considerable, up to 64 rule bases are possible, each of them having up to 256 inputs and rules. Furthermore an interface to most commonly used microprocessors is available. Arbitrary shapes of membership functions, different defuzzification modes including „Center of Gravity", an enormous rule engine with up to 10 million rule calculations per second makes this device a very interesting product in the field of realtime fuzzy control.
Test Results
After the whole system was carefully simulated, tests on an experimental car, a BMW 328i, were carried out. Figure 11 shows a full braking with ABS on dry asphalt.
Figure 11: Results from testbrake (large) 
The first diagram displays the decreasing estimated speed of the
vehicle v_{Fuz} and the fluctuating decrease of the speed
of the left front wheel v_{l,f}. Wheel acceleration and
wheel slip are shown in the second and third graph. The slip
value is limited successfully by means of the output of the ABS
controller, which is the driver current of the magnetic valve,
presented in the next diagram. Finally the system performance is
proved by the last graph. The longitudinal acceleration a_{x}
is near the physical limit.
Conclusion
The basis of the controlling algorithm consists of a nonlinear characteristic surface, which was created by fuzzy logic. The convincing advantage of fuzzy logic is the ability to modify and tune certain parts of this characteristic surface easily and carefully. Just the linguistic rules or variables need to be varied.
This simplifies the development and shortens the development time considerable.
Implementation of the
fuzzy ABS leads to excellent results of braking behavior of the
test vehicle. The deceleration level and steerability is
comparable to commercially available systems.
Acknowledgement
The authors wish to thank Thomas
Brandmeier from SIEMENS AG, Automotive Division, for handing out
the experimental car and for advice during test drives.
References
[1]  Daiß, A. and Kiencke, U.: Estimation of Vehicle Speed  FuzzyEstimation in Comparison with KalmanFiltering, 4th IEEE CCA, New York, 1995.  
[2]  Ostertag, M.: Strukturierte Optimierung technischer Prozesse am Beispiel der KFZ Crasherkennung, Institute for Industrial Information Systems, University of Karlsruhe, Ph. D. dissertation, 1996.  
[3]  Klein, R.: Realisierung einer FuzzyABSRegelung mit dem Mikrocontroller SAB 80C166 und dem FuzzyCoprozessor SAE 81C99A, Project work at the Institute for Industrial Information Systems, University of Karlsruhe, 1995.  
[4]  Daiß, A.: Beobachtung fahrdynamischer Zustände und Verbesserung einer ABS und Fahrdynamikregelung, Institute for Industrial Information Systems, University of Karlsruhe, Ph. D. dissertation, 1996.  
[5]  Eichfeld, H., Leindl, R. and Künemund, T.: The SAE 81C99A Fuzzy Logic Coprocessor, Embedded Intelligence ´96, ed. J. Wiesböck, Design & Elektronik, Sindelfingen 1996, pp. 347353.  