Professional Documents
Culture Documents
Discrete-Time For Traction: Robust
Discrete-Time For Traction: Robust
Introduction
Vehicle traction control, which includes
antiskid braking and antispin acceleration,
can enhance vehicle performance and handling. The objective of this control is to
maximize tire traction by preventing the
wheels from locking during braking and from
spinning during acceleration. At the same
time, vehicle traction control helps to maintain adequate vehicle stability and steerability. In general, there are two major difficulties involved in the design of a practical
traction control algorithm: (1) the vehicle/
brake system is highly nonlinear with timevarying parameters and uncertainties; (2) the
performance depends strongly on the knowledge of the tire/road surface condition.
Wheel slip, the difference between the vehicle speed and the wheel speed (normalized
by the vehicle speed for braking and the
wheel speed for acceleration), is chosen as
the controlled variable for most of the traction control algorithm because of its strong
influence on the tractive force between the
tire and the road. Typical relationships between the longitudinal adhesion coefficient
(the ratio between the longitudinal tractive
Presented at the 1989 American Control Conference, Pittsburgh, Pennsylvania, June 21-23, 1989.
This work was conducted while both authors were
at the Department of Mechanical Engineering,
University of California, Berkeley, CA 94720.
Han-Shue Tan is now with the Missile System
Group, Hughes Aircraft Company, Canoga Park,
CA 91304.
VI:
(1) Design of a traction controller that can
regulate the wheel slip at any desired
value.
Apni 1990
I f
Dly Pavement
.Unpacked Snow
1 .o
1990 IkEE
107
System Dynamics
A simplified vehicle model appropriate for
both vehicle acceleration and deceleration is
described in this section (see Fig. 2). This
model contains one-wheel rotational dynamics and linear vehicle dynamics, as well as
the interaction between them [ 7 ] , [8].
The wheel dynamics is determined by
Newton's law as in Eq. (1). The angular
acceleration of the wheel (&) equals the total torque to the wheel divided by the moment of inertia of the wheel ( J J . The torque
consists of shaft torque from the engine (TJ,
brake torque (TJ,tire tractive force (F,)
timed by the wheel radius (I?,,), and wheel
viscous friction [F,.(w,)], where the overdot
denotes time differentiation, and the variables T, and J,. of the driving wheel change
in accordance with the transmission gear
shifts.
Gl(.
= [(T, - Tb - R,,.F, - F,AU,~.)I/J,,.
Ni,~(h)
(4)
(3)
(2)
x-2 = -&(XI)
b2Nm
T , - Th
A = (x2 - xJ/x
Deceleration Case We start with the formulation of the antiskid problem (deceleration, h < 0). The corresponding system
equation with respect to h is presented by
Eq. (6).
h =
=
[ji2
- (I
{[-h(xz)
+
-
X)i,]/Xl
h V d N + 47-1
Wheel Speed
Vehicle Speed
'
= W(W
+ h ) f i ( x , )-&(xz)
> V/R,
= V / R ( W < V/R,
Fig. 2.
108
1,
li+
I
[B?(T)
+ b3Te(7)1/Xi(7) d7
li 4 I
b3/~1(7)
d7
Tb(fd
(8)
VfJ + At([&(fd
M = [
= w(fr-l)
-
Aprii 1990
- [A(?,)
k sgn
X(f,-~)l/Af
lW,) - X&)l
k,AfP)lz
+(I -P)=O
(12)
The linear system ( 1 1 ) is asymptotically stable if, and only if, the eigenvalues of Eq.
(12) are inside the unit circle. The permissible ranges of k,, and P for Eq. (11) to be
asymptotically stable are as follows:
O < P < 2
0 < k,At
(1 3 4
< [2(2
P)]/P
+ l ) +~ 1
1I
DEN
(1 1 )
This system looks as if it were a linear timeinvariant system with exogenous inputs X T
and D2 = (B2
b3T,)/xl.To guarantee a
(13b)
At(z - 1)
(loa)
-[ X I ( f k ) / ~ 3 1 w ( ~ L )
z2 - [(2 - P)
k,(z - 1) -(k,At
k, ArP
b3 Te(ta)l/xl (2,)
Tdtk)
Vh+l) - h(rJ
=
1 - 1 1
VehicleiBrake Dynamics
IW - b l
5 , = o I~lmp"lse(~,)lDb
k:
(2 - 0)/(0Ar)
(15)
IW,)
- XTI
and
tk
fi)
for 0 5
IArD,/(l
> ro
( 16a)
Ih(fk) - XI,
5
0) for 0 >
ArDbl(2 and
t,
>
to
( 16b)
= [B,
-
+ (I
- h)b,T,]/x,
[(I - Vb,/x,lT,
(17)
Bl =fi(xl)
(1
(1
NbZNAN
TAfk) = -[x,(rk)/(l -
RtL))&IW(fL)
(1 8 4
M'(fL)
W(fk-1)
- kv[h(tk) -
- [ X ( t k ) - X(fk-l)]/Af
170
(18b)
It has the same feedback structure as the deceleration case (Fig. 3) except for the linearization factor shown in Eq. (18a). In the
acceleration case, the input ratio 0is defined
by [b,X(t,)]/[&,fi(t,)], where is the estimate
with
Eqs. (12)of X. By replacing
(16) are also valid for the acceleration case
provided that the sampling interval is sufficiently small.
o2
oI,
Acruaror Dynamics
Thus far we have not yet considered the
actuator dynamics. The actuator dynamics
has been implicitly assumed to be much
faster than that of the vehicle and wheel,
and, therefore, has been neglected in the
course of model development. However, if
not so, the actuator dynamics needs to be
included together with the other major dynamics mentioned earlier. For such a case,
let G,(s) be the transfer function of the actuator whose input is Tb,.The actuator dynamics can be represented as shown, where
s is the differential operator.
Th(d = GJs) Th,(O
(19)
. z ( L - '(G,(s)/s2))]- I
(20)
-[x*(rk)/(l - Wk))&I
. G;'(z)
W(tk)
(2 Ib)
Cf.
Comments
This control scheme is robust, well-performing, and easy to design and modify. The
robustness of the scheme can be explained
by Eq. (13a). This condition implies that no
matter how overestimated b, is (which may
also include the gain of the actuator), the
system remains stable. And it is unstable only
when b, is underestimated by 50 percent.
Furthermore, it is observed from Eqs. (16)
that the resultant wheel slip will be smooth
(so as the brake torque) when the uncertainties change insignificantly during sampling
intervals and the steady state error is zero.
This indicates good performance. The system has been linearized, and linear control
theory has been adopted. In addition, cancellation of the actuator dynamics makes the
scheme adaptive to different actuators.
The scheme has an extra advantage. Since
the system is sampled every At, the influences of the uncertainties ( B , or B,) are accumulated over the sampling period. Only
the integrated values of B , and B2 can be
detected from the sampled output. Therefore, the high-frequency components of those
uncertainties are spontaneously filtered out.
Since we are not concerned with the reconstruction of output X ( t ) , the longer the sampling period, the fewer frequency components of the uncertainties (e.g., any
frequency below the Nyquist frequency
1/2Ar) are of consequence. The trade-off ap-
pears only when the control chattering increases as a result of the lengthened sampling period.
Experiment
Experiments conducted in the dynamometer test cell as well as in the vehicle at General Motors Research Laboratories are discussed in this section. Because of the limited
time available for building experimental
hardware, only antiskid brake tests were performed. The results prove successful.
Experimental Setup
The proposed vehicle traction control algorithms are first evaluated in a dynamometer test cell. Mechanically, the test cell contains a large aluminum drum driven by a
dynamometer (see Fig. 4). The inertia of the
drum set is approximately half the inertia of
a midsize car. Pressed against the drum are
two production wheel assemblies, one front
and one rear. The two wheels are thus driven
by the drum, which simulates a road surface.
Each wheel is equipped with production
brake hardware and a toothed wheel for
sensing the wheel speed. A brake pedal, depressed by the person acting as the test
driver, activates the brakes. The drum can
be water-sprayed to simulate a road surface
with a low friction coefficient.
The actuators are solenoid valves, which
control the opening and closing of the brake
orifices (with bandwidth smaller than 50 Hz).
They are pulse width modulated to simulate
the proportional action. When the sampling
period is small (e.g., > 10 msec), the actuator behaves like a slow first-order system
with its gain depending on both the states
of the valves (apply or release) and the pressure inside the hydraulic circuit. Typical values of this gain (go) range from 4 to 24
N-m/msec. G&) was obtained by approximating the actuator as a pure integrator with
g, = 20 N-m/msec. Notice that fi = (b,g,)/
(b,g,) for this case. fi could vary from 0.2
to 1.2 if b, = b3 (6, = 1.4 for the experiments). Although Eq. (15) suggests k: =
0 . 6 7 / A t , experimental results indicated that
k,. could be made as large as 1.25IAt. Therefore, we select & = 20 N-m/msec, b, =
1.4, and k,. = 1.25/At for all the tests documented in this section.
The proposed vehicle traction controller is
then implemented in a Buick C-car. The
brake assemblies and the electronic control
hardware are virtually the same as those used
in the test cell. The difference is that the
accelerometer data are used as an indirect
vehicle speed reference.
Apni 1990
Rear
Pmponioning
valve
6
-
I1
Front
Fmnt
Wheel-slip
controller
t 4
Time
Time
o2
40
I
r_
0
._a
zi
P4
Time
Fig. 6. Test cell result (wet drum surface,
5-msec sampling period).
Conclusion
This paper has proposed a discrete-time
robust vehicle traction control algorithm that
includes both antiskid and antispin actions.
The basic idea of the design methodology is
summarized as follows: group the uncertainties of the vehiclelbrake system together;
compensate this grouped uncertainty by proportional plus integral feedback; locally linearize the system subsequently; and determine the control parameters based on the
linear control theory. The initial test results
of this control algorithm show good performance regardless of vehicle environment. A
disadvantage of the proposed scheme is the
requirement of vehicle speed measurement.
Since this measurement is expensive at
present, antiskid brake system control cannot
be implemented directly (although this paper
demonstrates that good performance can be
achieved by simple feedback if the vehicle
speed is accessible). However, the employment of this scheme in the traction control
problem (i.e., the acceleration case) is fea-
;b,
OS L F W H E E L
0
36
0 6 R F WHEEL
l 0 8 R.R W H,E E L
,
35
' ,
r-.
PJ
=-I
Time
Time
00.0
TE7.r
07.9
Acknowledgment
This work was partially supported by the
Electrical Engineering and Electronics Department of the General Motors Research
Laboratories.
References
H. S . Tan, Adaptive and Robust Controls
with Application to Vehicle Traction Control, Ph.D. Dissertation, Univ. of California, Berkeley, 1988.
H. Leiber and A. Czinczel, Four Years of
Experience with 4-Wheel Antiskid Brake
(ABS), SAE 830481, 1983.
J . Y. Wong, Theory of Ground Vehicles,
New York: Wiley, 1978.
H. S. Tan and Y. K. Chin, VariableStructure Vehicle Traction Control, ASME
Winter Annual Meeting, 1988.
J . W. Zellner, An Analytical Approach to
Antilock Brake System Design, SAE
840249, 1984.
H. Schurr and A. Dittner, A New AntiSkid-Brake System for Disc and Drum
Brakes, Braking: Recent Developments,
SAE 840486, May 1984.
H. Leiber et al., Anti-skid System (ABS)
for Passenger Cars, Bosch Technical and
Scientific Report, Feb. 1982.
J . L. Hamed et al., Measurement of Tire
Brake Force Characteristics as Related to
and Electronics Engineering Department at General Motors Research Laboratories for three summers between 1984 and 1987. He presently works
at Missile System Group, Hughes Aircraft Company. Dr. Tans current interests include control
theory research and application for various automotive and manufacturing systems.
Out of Control
No, sire...not SPR as in strictly positive real. Theyve brought a petition from SPCR,
the society for prevention of cruelty to researchers!
April 1990
173