Professional Documents
Culture Documents
Advanced Yaw Control of Four-Wheeled Vehicles Via Rear Active Differential Braking
Advanced Yaw Control of Four-Wheeled Vehicles Via Rear Active Differential Braking
ThBIn6.4
boniolo,savaresi]@elet.polimi.it
5176
ThBIn6.4
neglected, as the forces they generate are smaller than those
generated by the brakes in the braking maneuvers of interest.
(A.4) the braking torque Tb at each wheel is considered
proportional to the local brake pressure pb , i.e., Tb = Kb pb .
Fig. 1 schematically represents the main parameters and
conventions adopted in the analytical derivation of the considered vehicle model. i j , i = { f , r}, j = {l, r} are the wheel
Wheels
(4)
(5)
45
asphalt,dry
40
35
pressure [bar]
Fig. 1.
friction coefficient
1.2
concrete, dry
0.8
0.6
asphalt,wet
0.4
(1)
(2)
Lateral
Yaw
r = 1/J((Fyrl + Fyrl )b
+ (Fxrl + Fxrl )w
+ (Fy f l + Fy f r ) cos( )a (Fx f l + Fx f r ) sin( )a
+ (Fx f l + Fx f r ) cos( )w + (Fy f l + Fy f r ) sin( )w)
(3)
25
20
identified response
15
0.2
18 bar
30
10
0
0.2
0.4
0.6
0.8
25 ms
0
0.05
0.1
0.15 0.2
time [s]
0.25
0.3
(6)
5177
ThBIn6.4
A. Parameters Identification and Model Validation
The proposed dynamic vehicle model depends on several
parameters, which can be grouped into two categories: easily
measurable parameters (such as lengths, widths and masses)
and unmeasurable ones. The second set of parameters must
be identified from data. The identified parameters are the
cornering stiffness C (0) and the yaw inertia J. An accurate
model of the tire longitudinal characteristic is available for
the test vehicle, and thus its parameters do not need to be
identified.
The identification of the two remaining parameters, i.e., J
and C (0) has been performed on runs executed at constant
longitudinal velocity by applying a square wave input pressure excitation at the rear brakes. Note that, even though the
driver was requested to keep the steering wheel straight, i.e.,
with = 0 , it could not be kept always constant. As such,
the resulting identification problem is not genuinely a SISO
one. In fact, two inputs act on the system: a commanded
input (the pressure gradient at the rear wheels) and a non
controllable input (the steering angle ). Hence, in the
identification procedure the measured steering wheel angle
is treated as a measurable disturbance and the vehicle
model using such a as an input. By numerically solving
the yaw rate simulation error minimization on three test runs
carried out at different longitudinal velocities (80, 100 and
120 km/h), a yaw rate moment of inertia J = 1125 kg m2 and
a front and rear cornering stiffness values C f (0) = 12 rad1
and Cr (0) = 15.9 rad1 , respectively, have been found.
The effectiveness of the identification procedure can be
appreciated by inspecting Fig. 3 which shows the results
obtained in two validation tests at constant velocity. It should
simulated yaw rate
P
steer
0.5
1
time [s]
1.5
steer
0.5
1.5
2.5
time [s]
120 km/h
P set point
be noted that in the identification procedure the actual measured pressure at the brake caliper has been employed. This
measurement is available only for identification purposes,
as in the final production vehicle pressure sensors are not
available. This, as can be seen in the right plot Fig. 2 which
shows the measured pressure at the rear brake caliper in
response to a step variation of the set-point, introduces an
uncertainty in the actuator model. Static errors up to 50%
are to be expected. It should be also noted that the actuated
pressure is always lower than the set-point. The experimental
data also show that the actuator can be modeled by a second
Fig. 3. Model Validation. Measured and simulated yaw rate, steering wheel
position and pressure gradient request at constant velocity: 100 km/h (left
plot) and 120 km/h (right plot).
v = 80 km/h
5.5
6.5
time [s]
7.5
Fig. 5. Open loop responses to a step variation of the left rear braking
pressure at a deceleration of 0.4 g.
5178
60
70
magnitude [dB]
magnitude [dB]
ThBIn6.4
70
80
80
increasing velocity
increasing deceleration
90
90
100
100
110
0
10
10
10
120
2
10
0
10
increasing velocity
10
10
10
increasing deceleration
phase []
10
phase []
110
2
10
0
50
50
100
100
150
150
2
10
10
10
frequency [Hz]
10
10
10
10
10
frequency [Hz]
10
10
Fig. 4. Bode diagrams of the transfer function from rear axle pressure difference to yaw rate at constant velocity from 50 to 120 km/h (left), and at
constant acceleration from 0 to -0.8g at 80 km/h.
Fig. 6.
1/(15 2 )s + 1
,
1/(300 2 )s + 1
5179
ThBIn6.4
uncontrolled system
yaw rate
fixed controller
0
scheduled controller
14
15
16
17
time [s]
18
fixed controller
19
20
scheduled controller
rear left
rear left
front left
pressure
front left
pressure
rear right
rear right
front right
front right
0
0
14
15
16
17
18
19
14
20
15
16
17
18
19
20
Fig. 7. Closed loop simulation performed at 120 km/h; yaw rate (upper
plot), fixed structure controller brake pressures (lower right) and gainscheduled controller brake pressures (lower left).
yaw rate
fixed controller
scheduled controller
15
15.5
16
16.5
17
time [s]
17.5
fixed controller
18
18.5
rear left
rear left
front left
pressure
rear right
front left
rear right
front right
0
front right
0
15
16
17
time [s]
19
scheduled controller
pressure
Finally, the following rate bounds on velocity and acceleration have been employed: = [9.8m/s2 , 9m/s3 ]. These
rate bounds have been derived from analysis of experimental
data considering a large number of different braking maneuvers.
Linear analysis of the loop transfer functions at the grid
vertexes confirms that performance and stability are guaranteed pointwise with a bandwidth of 0.5Hz and a phase margin
of 85 meaning that the design goals are successfully met
in the linear domain.
In order to assess the advantages of the LPV controller
, a H fixed structure controller has been also devised
using the same weighting functions. In order to guarantee
stability in all working conditions without loosing too much
performance, the H controller has been designed for the
nominal condition v = 120 km/h and a = -0.4 g. This is
not the worst case condition, but it has been selected to
represent a trade off between robustness and performance.
Furthermore, it represents a realistic condition as even in the
case of a hard braking, a few seconds are needed before
reaching high decelerations and during that span of time the
velocity drops restoring a more favorable vehicle dynamic
behavior.
To validate the proposed control laws a series of closedloop simulations have been carried out and analyzed. The
simulations have been run on the full nonlinear system
derived and validated in Section II augmented with the
full actuator dynamics. In order to simulate realistic driving
conditions, the initial velocity and brake pressures recorded
from track tests are used as inputs for the nonlinear simulator.
The steering action is kept at 0 , as this allows to decouple
the effect of the driver action on the steering wheel and
the yaw control. In the simulations, the reference yaw rate
is kept at 0 and a disturbance is applied to the system
in the form of a pressure offset on the front left wheel
braking pressure. Fig. 7 shows a plot of the yaw rate and
brake pressures for a maneuver performed at the initial speed
of 120 km/h and with a maximum deceleration of -0.8 g.
Inspecting Fig. 7, some remarks are due. (1) The difference
18
19
15
16
17
time [s]
18
19
Fig. 8. Closed loop simulation performed at 80 km/h; yaw rate (upper plot),
fixed structure controller brake pressures (lower right) and gain-scheduled
controller brake pressures (lower left).
5180
ThBIn6.4
The results of Fig. 8 confirm that in this case the maximum yaw rate obtained with the uncontrolled system is
0.55 /s while the fixed controller and the gain-scheduled
controller guarantee a maximum yaw rate deviation of 0.3 /s
and 0.14 /s, respectively. As expected, the gain-scheduled
controller outperforms the fixed structure controller in this
condition which is far from the nominal one used in the
fixed structure controller design.
Finally, Fig. 9 plots the comparison between the controlled
and uncontrolled system on experimental data obtained in a
hard braking maneuver at 120 km/h with a highly unbalanced
vehicle when the driver action is considered. High speed
tests are more meaningful because in this condition the
lateral drift due to the unbalanced load is more evident.
Analyzing Fig. 9, it can be appreciated that the control
system successfully controls the lateral dynamics of the
vehicle, thus counteracting the effects of unbalanced load.
The effectiveness of the control system is also quantified by
the tracking error variance shown in Fig. 9: as can be seen,
a reduction of approximately 90% has been achieved with
respect to the open loop system behavior.
1
controlled
system
0.5
0.5
uncontrolled
system
normalized tracking
error variance
0.5
1.5
2.5
time [s]
3.5
4.5
1
0.8
0.6
0.4
0.2
0
uncontrolled
controlled
5181