Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

SAE TECHNICAL
PAPER SERIES 2006-01-0499

A Method for Overcoming Limitations of Tire

Models for Vehicle Level Virtual Testing
Nicolas Gandoin
Renault S.A.

Shawn You
MTS Systems Corporation

Reprinted From: Load Simulation & Analysis in Automotive Engineering

(SP-2038)

2006 SAE World Congress

Detroit, Michigan
April 3-6, 2006

400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-5760 Web: www.sae.org
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

The Engineering Meetings Board has approved this paper for publication. It has successfully completed
SAE's peer review process under the supervision of the session organizer. This process requires a
minimum of three (3) reviews by industry experts.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise,
without the prior written permission of SAE.

For permission and licensing requests contact:

SAE Permissions
400 Commonwealth Drive
Warrendale, PA 15096-0001-USA
Email: [email protected]
Tel: 724-772-4028
Fax: 724-776-3036

For multiple print copies contact:

SAE Customer Service

Tel: 877-606-7323 (inside USA and Canada)
Tel: 724-776-4970 (outside USA)
Fax: 724-776-0790
Email: [email protected]

ISSN 0148-7191
Copyright © 2006 SAE International
Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE.
The author is solely responsible for the content of the paper. A process is available by which discussions
will be printed with the paper if it is published in SAE Transactions.

Persons wishing to submit papers to be considered for presentation or publication by SAE should send the
manuscript or a 300 word abstract to Secretary, Engineering Meetings Board, SAE.

Printed in USA
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

2006-01-0499

A Method for Overcoming Limitations of Tire Models for Vehicle

Level Virtual Testing
Nicolas Gandoin
Renault S.A.

Shawn You
MTS Systems Corporation
Copyright © 2006 SAE International

ABSTRACT INTRODUCTION

The intention of this work is to illustrate a method used to In recent years, neural network methods have been
overcome limitations of tire models developed during an investigated as a way to model nonlinear, frequency
evaluation study of an Empirical Dynamic™ (ED) damper dependent components. Neural networks are
model. A quarter vehicle test system was built to support computational structures that can ‘learn’ a complex
the evaluation, and a model of the test system was also relation or pattern, much like biological neural systems.
developed in ADAMSTM. In the model, the damper was They’re used extensively in modern pattern recognition
represented by a polynomial spline function and by an ED algorithms, with applications such as handwriting and
model separately. Vehicle level comparisons between speech recognition. With proper interfacing, they can
the physical measurements and the model predictions also be used as black box models for dynamic systems.
were conducted. The actuator displacement signal from In this application, the models are generated solely from
the physical test was used to drive the virtual test system. measured input and output signals, without a direct
Spindle acceleration, spindle force, and other signals requirement for information about the component
were collected for comparison. geometry or physical parameters. The principal benefit of
neural network modeling is to enable accurate predictions
The tire model was identified as a significant source of of component behavior when significant nonlinearity and
error and as a result, the direct vehicle level correlation hysteresis are present.
study did not illustrate any advantage of the ED damper
model over a spline damper model. Previously, work has been done to compare the accuracy
of ED damper models and spline damper models in
Subsequently the Remote Parameter Control™ (RPC®) component level experiments [1]. It has been shown that
process was used to reproduce lab measured spindle at the component level, ED damper models are more
acceleration and body to wheel displacement signals and accurate in simulating high load (velocity) conditions.
thus to minimize the influence of the tire model. A fair However, the impact of this accuracy improvement on the
comparison base was established for vehicle signals vehicle level simulation has not been thoroughly
above the tire. evaluated. For this purpose, a quarter vehicle test
system was built (see Figure 1 and 2). Three events
After RPC™ iterations, a correlation study was (Three Bumps, Belgian Block, and Step Event) were
conducted on shock tower load and acceleration signals. tested using the system. Actuator displacement, spindle
It was found that the ED damper is better than the spline acceleration, spindle wheel force transducer loads, body
damper in high load (velocity) conditions. For low damper to wheel displacement, shock tower load, and shock
force conditions, the model with an ED damper predicts tower acceleration signals were collected for vehicle level
similar results as compared to the model with a spline evaluation.
damper. This is understandable since for conditions of
low damper force, the damper exhibits a more linear Two ADAMS models of this quarter vehicle test system
behavior. In this case, both the spline damper model and were also created (see Figure 3). The only difference
the ED damper model were adequate. between the two models is that one has a spline damper
model and the other has an ED damper model.

The ED damper model was created by exciting the

damper with a shaped random noise signal while
recording damper load and displacement. Component
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

EDM (cEDM) software was then used to build the ED

damper model based upon the damper load and
displacement signals.

Vehicle system is often very complicated and difficult to

model. Some elements, such as the tire, are extremely
nonlinear. It is very important to model these elements
accurately in order to achieve meaningful comparison
results between measurements and simulations.
Likewise it is also important to accurately model such
elements to compare ED damper and spline damper
model results. Otherwise, the modeling error associated
with elements other than the damper may be much larger
than the impact of damper model improvement and thus
lead to incorrect conclusions.
Figure 3, ADAMS model of the quarter vehicle test
system

DIRECT VEHICLE LEVEL CORRELATION

Actuator displacement signals collected in the test were

used as the actuator motion command in the virtual
models. Spindle loads, spindle acceleration, body to
wheel displacement, shock tower load, and shock tower
acceleration signals were collected for comparison with
the tests.

RPC Pro™ software, an application software widely used

in testing labs, was employed as the main analysis and
control software. RPC Pro™ was configured such that it
could open the ADAMS model, run the simulation and
Figure 1, Quarter vehicle test system collect response signals in the RPC™ data format
automatically. In this way, the simulation process was
simplified significantly.
Body accel sensor

Az_c Figures 4 to 8 show comparisons of spindle vertical load,

Ms Entry Body load sensor body to wheel displacement, spindle acceleration, shock
Supplementary mass Fz_c tower load, and shock tower acceleration signals of
models with ED and spline dampers and of test
measurements for the Belgian Block event.
Body/wheel displacement sensor
Dz_tc
From the plots, it is easy to see that spindle vertical force
Dz_rc (Fz) has poor correlation between the test measurements
and model predictions. The predicted spindle Fz signals
load sensor:
Body/rod displacement sensor
Dynamometric wheel were much lower than the measured signal. Spindle
acceleration, body to wheel displacement and shock
Driving : road profile Az_pf Fz_rd tower load signals had better correlation. However,
clearly there was phase error between the virtual and
Spindle acceleration sensor physical results. The phase error contributed to high
levels of RMS errors.

Poor virtual to physical correlation results indicated that

Figure 2, Schematic diagram of the quarter vehicle test
the ADAMS models (with both the ED damper and spline
system
damper models) were quite different from the physical
testing system. Since the spindle forces had poor
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

correlation and the tire was modeled as a simple impact

80
function in the ADAMS model, it was suspected that the Test Result
tire modeling method was a significant source of the 60 ED Model
error. The tire modeling error would certainly propagate Spline Model
40
to other parts of the model and negatively impact

Displacement [mm]
correlation of all signals. 20

0
Among the correlation signals, the spindle force had
generally poor correlation when compared to other -20
signals. This may be due to the fact that spindle load is -40
more sensitive to the tire modeling error. The behaviors
such as tire rotation, rubbing, and squeezing were not -60

modeled in the ADAMS model. These factors would have -80

a direct impact on spindle load. However, spindle motion
may not be as sensitive to these modeling errors. This 4.9 5.0 5.1 5.2 5.3 5.4 5.5
theory supports the observations of poor spindle load Time [s]

correlation while spindle acceleration and body to wheel Figure 5, Comparison of body to wheel displacement
displacement correlation is relatively good. Likewise the (mm) signal for the Belgian Block event
effect of tire modeling errors may be damped out by the
20
damper, therefore, supporting the observation of better Test Result
body load and body acceleration signal correlation. ED Model
10 Spline Model
Figure 4 to 8 show that signals predicted by models with Acceleration [g]
the ED damper and spline damper are similar. This 0
means that by implementing the ED damper, the overall
accuracy of the vehicle model did not significantly
-10
change. The effect of the ED damper may be
overwhelmed by the tire modeling error.
-20

-30
7000
Test Result
ED Model 4.9 5.0 5.1 5.2 5.3 5.4 5.5
5000
Spline Model Time [s]

3000
Figure 6, Comparison of spindle acceleration (g) for
Force [N]

1000 the Belgian Block event

5000
-1000 Test Result
4000 ED Model
-3000 Spline Model
3000

-5000 2000
Force [N]

1000
4.9 5.0 5.1 5.2 5.3 5.4 5.5
Time [s] 0
Figure 4, Comparison of spindle load Fz (N) for the
-1000
Belgian Block event
-2000

-3000

4.9 5.0 5.1 5.2 5.3 5.4 5.5

Time [s]

Figure 7, Comparison of shock tower load (N) for the

Belgian Block event
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

and body to wheel displacement were reduced to 15.3%

and 12.3 % respectively for the Belgian Block event with
1.2
Test Result an ADAMS model using the ED damper. The RMS error
1.0 ED Model was relatively low. Figure 10 shows the comparison of
0.8 Spline Model measured spindle acceleration vs. achieved signals after
0.6 the RPC™ iterations. Figure 11 shows the same
Acceleration [g]

0.4 comparison for body to wheel displacement signals. The

0.2 figures show that the achieved signals matched very well
with lab measured signals.
0.0
-0.2
The same RPC™ iterations were conducted for the
-0.4 Belgian Block event with the ADAMS model using a
-0.6 spline damper. Similar results were obtained.
-0.8
For the Step event RPC™ iterations were able to drive
4.9 5.0 5.1 5.2 5.3 5.4 5.5 the RMS error to low levels similar to the Belgian Block
Time [s]
event. As a result, the achieved signals after RPC™
Figure 8, Comparison of body acceleration (g) for the
iterations were very close to the lab measured signals.
Belgian Block event
For the Three Bumps event, jounce bumper impact
For the Step event and Three Bumps event, similar
occurred. In this case, a small change in drive file
correlation results were achieved.
displacement would cause large difference in spindle
acceleration. Since the impact was not modeled to a
RPC™ SIMULATION PROCESS sufficient level of accuracy, the RMS error was still above
40% after RPC™ iterations were conducted.
Remote Parameter Control (RPC™) is an advanced
simulation technique used to repeatedly replicate and
analyze “in service” vibrations and motions of a specimen
using a dynamic mechanical system in a controlled
laboratory environment or a virtual model [2].

In this work, the RPC™ process was used to make sure

that spindle acceleration and body to wheel displacement
of the virtual models matched the lab measurements.
Subsequently shock tower load and acceleration signals
could be compared to determine whether the ED damper
model was more accurate than the spline damper model.
This method minimizes the requirement for an accurate
tire model and establishes a fair comparison base for ED Figure 9, RMS error plot for the Belgian Block event with
and spline damper model results. ED damper model
Lower line: spindle acceleration; Upper line: wheel
In the RPC™ iteration process, the actuator displacement signals
displacement signal was the drive channel. The spindle
acceleration and body to wheel displacement signals
were used as response channels. The control band was
from 0.25 Hz to 80 Hz. The reason for this frequency
range selection was that the testing system was an
inertial reacted system. It is difficult to achieve low
frequency control without large actuator displacements.
On the other hand, beyond 80 Hz, there was very little
road induced damaging content. Therefore, there was no
need to reproduce the content beyond 80 Hz.

During the iteration process, gains were adjusted to

achieve the fastest convergence. Figure 9 shows that
after six iterations, the RMS errors of spindle acceleration
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

model with an ED damper predicted more accurate loads

30
Lab Measurement than the model with a spline damper did. At low load
Achieved after RPC Iterations situations (usually corresponding to low damper velocity),
20 the model with an ED damper predicted similar results as
compared to the model with a spline damper. This is
Acceleration [g]

10 likely due to the fact that for low damper velocities, the
damper is in the linear operation range. Both ED damper
0 and spline damper models are accurate in this range.
The advantage of the ED damper is in the high velocity,
non-linear range.
-10

Figure 13 shows shock tower acceleration comparisons

-20
of the test measurement with the ED damper model and
spline damper model predictions. The figure shows that
9.0 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 10.0
Time [s] correlation between measurement and prediction was
poor because the measured signal was noisy. However,
Figure 10, Spindle accelerations for the Belgian Block the trends of the signals matched well.
event with ED damper model

5000
Test Result
70 4000 ED Model
Lab Measurement
60 Spline Model
Achieved after RPC Iterations 3000
50
40 2000
Displacement [mm]

Force [N]

30
20 1000
10 0
0
-10 -1000
-20 -2000
-30
-40 -3000
-50
3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8
Time [s]
9.0 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 10.0
Time [s] Figure 12, Comparison of body load signal (N) for the
Belgian Block event
Figure 11, Body to wheel displacements for the Belgian
Block event with ED damper model 1.0
Test Result
0.8 ED Model
0.6 Spline Model
VEHICLE LEVEL CORRELATION AFTER
0.4
Acceleration [g]

RPC™ ITERATIONS
0.2
After RPC™ iterations, spindle acceleration and body to 0.0
wheel displacement signals were accurately reproduced
-0.2
for two events. The influence of the simple tire model was
minimized. A fair comparison base was established for -0.4
vehicle models with an ED damper and a spline damper -0.6
model. -0.8

Figures 12 to 13 show comparisons of shock tower load 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8
signals from models with an ED damper and a spline Time [s]
damper and from lab measurements for the Belgian Block Figure 13, Comparison of body acceleration signal (g) for
event. It is clear that at most of the high load peaks Belgian Block event
(usually corresponding to peak damper velocity), the
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

Figure 14 shows the comparison of shock tower load Figure 15, Comparison of shock tower acceleration signal
signal from test measurement, and from predictions of (g) for the Step event
the models with an ED damper, and a spline damper for
Step event. It is clear that at the high peak, the model Figure 16 and 17 show the comparison of shock tower
with an ED damper predicted more accurate loads than load and acceleration signals from models with an ED
the model with a spline damper did. This shows that the damper and a spline damper and from lab measurements
ED model is more accurate in the high load case. At low on the three bumps event. In this event, jounce bumper
load cases, the figure shows that the model with an ED impact occurred. The result is sensitive to small details in
damper predicted similar results as the model with a modeling the impact. After approximately 12 RPC™
spline damper. Again, this is due to the fact that for low iterations, the RMS error of spindle acceleration was still
damper velocity situations, the damper is in the linear about 40%. This means the jounce bumper impact was
operating range. Both the ED damper model and the not modeled properly in the ADAMS model. As a result,
spline model are accurate. the lab measured spindle acceleration and body to wheel
displacement could not be accurately reproduced. The
Figure 15 shows same comparison for shock tower jounce bumper impact error in the models may
acceleration signal. The figure shows that correlation is overwhelm the accuracy improvement of the ED damper.
not very good because the measured signal is noisy. Therefore, the ADAMS models with ED and spline
However, at the peak, which corresponding to an impact, dampers show similar results.
ED modal predicted more accurate shock tower
acceleration.
10000
8000
6000
14000
Test Result 4000
12000 ED Model
2000
Force [g]

10000 Spline Model

0
8000
-2000
Force [N]

6000
-4000
4000 -6000 Test Result
2000 -8000 ED Model
Spline Model
0 -10000
-2000
4.3 4.5 4.7 4.9 5.1 5.3 5.5 5.7 5.9 6.1 6.3 6.5 6.7
-4000 Time [s]
Figure 16, Comparison of shock tower load signal (N) for
0.20 0.25 0.30 0.35 0.40 0.45 the Three Bumps event
Time [s]
Figure 14, Comparison of body load signal (N) for the
3
Step event

2
4
Test Result
Acceleration [g]

ED Model 1
3
Spline Model

2
Acceleration [g]

1
-1 Test Result
0 ED Model
Spline Model
-2
-1
4.3 4.5 4.7 4.9 5.1 5.3 5.5 5.7 5.9 6.1 6.3 6.5 6.7
-2 Time [s]
Figure 17, Comparison of shock tower acceleration signal
0.20 0.25 0.30 0.35 0.40 0.45 (g) for the Three Bumps event
Time [s]
 Downloaded from SAE International by Steven Sullivan, Wednesday, November 28, 2018

CONCLUSION REFERENCES
[1] A.J. Barber, Accurate Models for Bushings and
To adequately evaluate the accuracy of an ED damper Dampers using the Empirical Dynamics Method,
model at the vehicle level, an accurate vehicle model is Proceedings of the International ADAMS Users’
required. In this case, since the tire in the vehicle model Conference, Mechanical Dynamics GmbH, 1999.
was modeled by a simple impact function, high levels of [2] Grote, P., Grenier, G. “Taking the Test Track to the
error were induced. This error may overwhelm the Lab”, Automotive Engineering, June 1987, Volume 95,
influence of an ED damper model. As a result, the direct Number 6.
vehicle level correlation study did not illustrate the
advantage of an ED damper model over a spline damper
model for all three events.

The RPC™ process was used to reproduce lab

measured spindle acceleration and body to wheel
displacement signals. In this way, the influence of the
simple tire modeling method was minimized. A fair
comparison base was established even though the
vehicle model was not perfect. The RPC Pro™ Virtual
Test software was used for the RPC™ iteration process.
This software feature enables RPC Pro™ to send a drive
file to the ADAMS model, run the ADAMS model and then
collect desired response signals automatically. A large
amount of time is saved by this easy-to-use process.
RPC™ iterations were able to reproduce spindle
acceleration and body to wheel displacement signals for
the Belgian Block and Step events. However, for the
Three Bumps event, RPC™ iterations could only reduce
spindle acceleration RMS errors to about 40%. This is
attributed to insufficient accuracy of the jounce bumper
impact modeling.

After the RPC™ iterations, a correlation study was

conducted on shock tower load and acceleration signals.
For the Belgian Block and Step events, it was found that
for most of the high damper load conditions (usually
corresponding to high damper velocity situations), the
vehicle model with an ED damper predict more accurate
results. Therefore, it is fair to conclude that the ED
damper is better than the spline damper in high load
(velocity) conditions. For low damper force conditions
(usually corresponding to low damper velocity situations),
the model with an ED damper predicts similar results as
compared to the model with a spline damper. This is
understandable since for low damper force situations, the
damper exhibits a more linear behavior. In this case,
both the spline damper model and the ED damper model
are accurate.

ACKNOWLEDGEMENTS

The authors are grateful for the assistance given by

Darragh Murphy, Christoph Leser, Juan Garcia, Marv
Westermann, and Mike Englerth at MTS Systems
Corporation and Eric Vaillant, Huu-Khoa Nguyen,
François Parolini and Daniel Corneille at Renault.