Calculating Longitudinal Wheel Slip and Tire Parameters Using GPS Velocity
Calculating Longitudinal Wheel Slip and Tire Parameters Using GPS Velocity
Calculating Longitudinal Wheel Slip and Tire Parameters Using GPS Velocity
Calculating Longitudinal Wheel Slip and Tire Parameters Using GPS Velocity
Shannon L. Miller Brett Youngberg Alex Millie Patrick Schweizer J. Christian Gerdes
Department of Mechanical Engineering
Stanford University
ABSTRACT
The use of GPS velocity information has an even greater This is admittedly a simplification and better results could be
benefit beyond the generation of an accurate slip obtained by including estimates of the road grade, mass and
measurement. By comparing the wheel slip to estimates of other road loads obtained online. Grade can be estimated from
the forces acting on the vehicle, the tire force versus slip the GPS receiver by examining the ratio of vertical velocity to
characteristics can be obtained. These, in turn, can be used horizontal velocity after which the mass and road loads can be
to feed model-based controllers for ABS or ESP systems estimated if a value of engine torque is available [ 13.
or more advanced driver assistance systems for
lanekeeping or collision avoidance [9]. They could also be The remaining step is to apportion the acceleration between
used to provide more accurate observers for periods of the front and rear axles and - if desired - the left and right
time when GPS information is not available. Several tires. With a four wheel drive vehicle, curves for each wheel
researchers [6,7,10,12,13] have also suggested that by can be derived during periods of acceleration alone by
fitting the low slip region of the force-slip curve to a assuming some split of torque across the four wheels. For
parameterized model - ranging in complexity from the vehicles with only two driven wheels, the force-slip
form of Equation 2 to dynamic friction models - the peak characteristics of the driving wheels can be identified while
friction point can be determined. This application accelerating, but the undriven wheels can only be identified
represents a further use for the information that can be during periods of braking. This, in turn, requires knowledge
generated from GPS-based slip measurement, although of the brake distribution among the different wheels. The
preliminary results achieved with the system demonstrate experimental results and the derivation presented here assume
some care in interpretation is necessary for friction that the vehicle has two driven wheels, the curves are
detection. identified under acceleration alone and an effective
longitudinal stiffness for the two wheels combined is desired.
This paper demonstrates how tire force-slip curves - and in
particular the linear region of these curves - can be Assuming that the tires are in the linear region of the force
determined using GPS velocity measurements and wheel versus slip curve, the longitudinal equation is given by:
speed sensors. The GPS velocity measurement is
differenced to obtain absolute vehicle acceleration, which
is multiplied by the vehicle mass to calculate the
longitudinal force on the tires. The accuracy of the GPS
1801
cx= - c - M
Combining this with Equation 4 gives simply: (13)
1802
interfacing. As demonstrated in the experimental results, Clearly, the GPS velocity and wheel speed signals are close to
the low frequency GPS velocity measurement is the identical when plotted in the same graph. The differences can
limiting factor in setting the accuracy of the slip be seen in the lower plot once these values have been
measurement, not the wheel speed. Hence, this resolution converted to slip. Periods of acceleration result in positive
could be greatly reduced without altering the basic values of slip, while periods of deceleration produce negative
character of the results presented in this paper. values. During the experiments, slip was measured to be
between 3% and -3%, typical of normal driving conditions
The data acquisition system includes a VersaLogic single- and well within the linear region of the force-slip curves.
board target computer (SBC) and a notebook host
computer connected via an Ethernet link. The target The apparent noise in the slip curve is actually a saw-tooth
computer operates under the xPC target environment of pattern caused by the lOHz update rate of the GPS velocity
MATLAB's Real-time Workshop package, providing a information. Since the wheel speed is available at a higher
consistent Ims sample time. The GPS receiver is update rate than the GPS velocity, periods of acceleration and
connected to the SBC through a serial port while the wheel deceleration produce this pattern when viewed at the higher
speed and synchronization pulses are obtained through the update rate. This does not alter the force-slip curve generation
SBC's digital and analog inputs, respectively. Data post- since the data is downsampled to match the lOHz update rate
processing and storage are handled exclusively by the of the GPS receiver prior to the use of the least-squares
notebook. Since the complete system is very portable, regression.
tests were performed on several passenger vehicles.
Figure 3 displays the raw acceleration and speed ratio data
used in the least-squares algorithm together with the best-fit
EXPERIMENTAL RESULTS Iine through the data. No filtering was performed on this data
other than the moving average filter for the wheel speeds and
A series of short test runs (30-60 seconds) were performed the elimination of a few outliers caused by occasional loss of
in two test vehicles, maintaining normal driving speeds
' GPS satellite visibility. The experimental data points clearly
within a limited range on open roads with minimal grade. support the linear tire modeling assumption at low slip.
In keeping with the assumptions made for this study, only
periods of acceleration and engine braking were used in AcceleraUonvs Speed Ratio
3r
the experimental runs, ensuring that all driving and braking
forces acted through the driven wheels. Only the driven 25-
maneuver
2- 0.5 -
m
P O-
-1.5 -
-2 I
3.04 3.06 3.08 3.1 3.12 3.14 3.16 I8
Speed Ratio
0 I
5 10 15
Timo(s=)
20 25 30
Figure 3 -Raw Acceleration Data versus Speed Ratio
Peicenf Sllp vs. Tim
After running the least-squares routine to obtain the radius,
this curve can be rearranged into the more familiar units of
force versus slip, presented in Figure 4.
5 10 15 20 25 30
Time@=)
1803
variability of tire longitudinal stiffness with road surface
Force vs. Slip condition or inflation pressure.
4000 I .
1.51
3000
1-
-
0.5 -
-0
m
5 0-
B
-0.5 -
4 0 0 0 ~
-1.5 -1 4.5 0 0.5 1
Percent Slip
-1.51
Figure 4 - Experimental Force versus Slip Curve 3.04 3.06 3.08 3.1 3.12 3 14 16
Speed Ralio
While identical in content to Figure 3, this plot has a bit Figure 5 - Comparison at Two Inflation Pressures
more physical meaning since the slope of the best fit line
through these points is the effective longitudinal stiffness,
Cx,of the tires.
CONCLUSIONS
Effective radius was found to be between .3222m and
.3246m, depending upon inflation pressure. Higher The data shows that GPS velocity information can be
inflation pressures tended to increase the effective radius; combined with wheel speed information to measure tire slip
the average value at 25psi inflation was .3229m, while the and estimate longitudinal stiffness and effective radius. The
average value at 45psi was .3244m. Standard deviation of data gathered are consistent with the assumption of a linear
the radius obtained from multiple tests at the same relationship between force and slip at low levels of slip as
inflation pressure was remarkably low, falling well below predicted by classical tire models. Radius estimation using
lmm. More tests are required to statistically establish the this method exhibited considerable precision and accuracy
precision of this measurement, but the precision is within the difference between the undeformed and static
qualitatively quite high. Furihermore, the values obtained loaded tire radii. In preliminary testing, increased inflation
matched physical intuition by consistently falling between pressure appeared to systematically lower the longitudinal
the tire’s unloaded radius and static loaded radius for each stiffness. Future work will concentrate on increasing the
tire pressure. amount of collected data and refining data processing to
establish more definitive statistical information regarding the
Stiffness values experienced more deviation than did effectiveness and sensitivity of this measurement system.
effective radius values. A typical value for longitudinal
stiffness was 1.5 x 10A6 N/m. Variations in inflation
pressure had a noticeable effect upon stiffness, though in ACKNOWLEDGEMENTS
an initially counter-intuitive way: lower pressures tended
to increase stiffness measurements. While more rigorous The authors would like to thank Chris Wilson and the
testing is necessary to define this relationship, a physical Research and Technology Center of DaimlerChrysler
argument for this result is that the increased size of the Research and Technology North America for providing
contact patch increases the tractive force for a given level funding and a test vehicle for this work. Partial funding for
of slip. As illustrated in Figure 5 , the slope of the best-fit the first four authors was provided through the School of
line through the run taken at 45 psi is less than the slope of Engineering’s Summer Undergraduate Research Fellowship
the run taken at 25 psi. (SURF) hogram and the Mechanical Engineering
Department’s Summer Undergraduate Research Program at
The results presented here are only preliminary and based Stanford University. Thanks also go out to Christopher
upon a limited number of data sets. Further testing and Carlson, David Bevly, Robert Sheridan and Eric Rossetter for
refinement of the slip measurement is required before their assistance in this work.
drawing any conclusive statements regarding the
1804
REFERENCES [ 131 Ray, Laura; “Nonlinear Tire Force Estimation and Road
Friction Identification: Simulation and Experiments,” ’
[l] Bae, Hong S., Ryu, Jihan and Gerdes, J. Christian, Automatica, vo1.33, no.10, October 1997, pp1819-33.
“Parameter Estimation with GPS Road Grade
Measurements for Automated Highways,” Submitted to [ 141 van Zanten, A., et al., “Vehicle Stabilization by the
the 41hIntemational IEEE Conference on Intelligent Vehicle Dynamics Control System ESP,” Proceedings of the
Transportation Systems, 2001. 1” IFAC Conference on Mechatronic Systems, Darmstadt,
Germany, 2000, pp. 95-102.
[2] Bakker, E., Nyborg, L. and Pacejka, H.B., “Tyre
Modelling for Use in Vehicle Dynamics Studies,” SAE
Paper 870421, 1987.
1805