Sie sind auf Seite 1von 164

Enhanced friction modeling for steady-state rolling tires

Citation for published version (APA):


Steen, van der, R. (2010). Enhanced friction modeling for steady-state rolling tires Eindhoven: Technische
Universiteit Eindhoven DOI: 10.6100/IR692262

DOI:
10.6100/IR692262

Document status and date:


Published: 01/01/2010

Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne

Take down policy


If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.

Download date: 13. May. 2019


Enhanced friction modeling for
steady-state rolling tires
The research reported in this thesis is supported by the CCAR project ‘FEM Tyre
Modelling’, in cooperation with TNO Automotive, Helmond, the Netherlands and
Apollo Vredestein B.V., Enschede, the Netherlands.

René van der Steen (2010). Enhanced friction modeling for steady-state rolling tires.
Ph.D. thesis, Eindhoven University of Technology, Eindhoven, the Netherlands.

A catalogue record is available from the Eindhoven University of Technology Library.


ISBN: 978-90-386-2390-0

Cover design: Oranje Vormgevers, Eindhoven, the Netherlands.


Reproduction: Ipskamp Drukkers B.V., Enschede, the Netherlands.

c 2010 by René van der Steen. All rights reserved.


Copyright
Enhanced friction modeling for
steady-state rolling tires

PROEFSCHRIFT

ter verkrijging van de graad van doctor aan de


Technische Universiteit Eindhoven, op gezag van de
rector magnificus, prof.dr.ir. C.J. van Duijn, voor een
commissie aangewezen door het College voor
Promoties in het openbaar te verdedigen
op donderdag 9 december 2010 om 16.00 uur

door

René van der Steen

geboren te Borsele
Dit proefschrift is goedgekeurd door de promotor:

prof.dr. H. Nijmeijer

Copromotor:
dr.ir. I. Lopez
Summary
Enhanced friction modeling for steady-state rolling tires

Tire modeling is nowadays a necessary tool in the tire industry. Car manufacturers, gov-
ernments and consumers demand better traction under all circumstances, less wear and
more recently less noise and a lower rolling resistance. Therefore finite element analysis
is adopted in the design process of new tires to cope with these, often conflicting, de-
mands. Finite element tire modeling can increase the insight on specific properties of a
tire, decrease the development time and reduce development costs of new tires. However
in practice most finite element models are still not able to match outdoor experiments.
Both the static deformation and the dynamic response of the tire rolling on the road
should be accurately predicted. The cornering, braking and traction of a tire depend on
the generated friction forces. Friction depends not only on the tread properties of the
tire, but also on the road surface and environmental conditions. The main goal of this
thesis is to develop a robust and numerically efficient friction model for finite element
tire simulations and to create a framework for the identification and implementation of
friction related parameters.
The numerical modeling of a tire in combination with its environment is a challenging
task, since different physical phenomena play a role. Typically the mechanical, thermal
and fluid domains contribute to the tire response. This research is restricted to the me-
chanical domain, where a numerical modeling framework for steady-state rolling tire
simulations is defined. In future developments of the model other effects can be in-
cluded using this framework as a base. One of the objectives of this thesis is to develop
and validate a tire friction model for finite element analysis, which captures observed ef-
fects of dry friction on the handling characteristics of rolling tires. Friction by itself is a
highly complex interaction phenomenon between contacting materials and can be mod-
eled on many different length scales, applying different numerical techniques. This can
however lead to an enormous computational burden and as a result it can be impractical

v
vi S UMMARY

for an industrial application. To provide a numerically feasible and relatively fast solution
a phenomenological friction model is chosen, where the parameters are identified using
a two step experimental / numerical approach.
Firstly, friction experiments are performed on a laboratory abrasion and skid tester to
investigate the influence of contact pressure on the frictional force. In this experimental
setup a small solid tire, with adjustable side slip angle, is pressed on an abrasive disk. The
friction present between the abrasive disk and solid tire drives the tire and the resulting
forces are measured with a force sensor. Several experiments under different normal
loads and side slip angles of the tire are conducted. These measurements, under low
rolling velocity, are used to identify contact pressure dependent friction parameters. The
relevant parts of this setup are modeled in the commercial finite element package ABAQUS
and the steady-state performance of the small tire under different slip angles is evaluated
and compared with experiments. It is shown that the present turn slip, which has great
impact on the slip velocity field at the trailing edge of the contact area, is captured well
with the model. Furthermore, the calculated cornering stiffness is in good agreement
with the experiments.
Secondly, outdoor braking experiments at different velocities with a full scale tire are
conducted to obtain a velocity dependent parameter set for the tire friction model. The
derived friction model is then coupled to a finite element model of this full scale tire,
which is also constructed in the software package ABAQUS. The finite element model is
validated statically using measurements of the contact pressure distribution, contact area
and of the radial and axial stiffness of the tire. The steady-state transport approach in
ABAQUS is used to efficiently compute steady-state solutions at different forward velocities
as used in the outdoor experiments.
Finally, the predictive capability of the finite element tire model in combination with the
proposed friction model is assessed. The basic handling characteristics, such as pure
braking, pure cornering, and combined slip under different loads, inflation pressures
and velocities are evaluated and validated with experiments. Based on this comparison, it
can be concluded that all three basic handling characteristics are adequately predicted.
Contents

Summary v

Nomenclature xi

1 Introduction 1
1.1 General introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Tire performance and modeling . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Challenges in FE tire modeling . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Objectives and research approach . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.6 Outline of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Modeling framework for steady-state rolling tires with friction 9


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Overview of friction models . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Amontons-Coulomb . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Rubber friction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Tire research using FE Models . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Implicit analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Explicit analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Modeling framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.1 Numerical method . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4.2 Friction model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Steady-state transport analysis . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.1 Contact conditions for steady-state rolling . . . . . . . . . . . . . . 25
2.5.2 Frictional stress for steady-state rolling . . . . . . . . . . . . . . . . 26
2.6 Implementation of a friction law in steady-state rolling . . . . . . . . . . . 28
2.7 Validation of the implemented friction law . . . . . . . . . . . . . . . . . . 30
2.7.1 Cylindrical model . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.7.2 Comparison with standard Coulomb friction law . . . . . . . . . . 30

vii
viii CONTENTS

2.7.3 Validation of friction law for varying friction coefficient . . . . . . 32


2.8 Steady-state free-rolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 Simulation procedure to compute handling characteristics with an FE model 37


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Approach to identify parameters of the friction model . . . . . . . . . . . . 39
3.3 Design of the test tire used for experimental validation of the FE model . . 40
3.4 FE tire model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.1 2D tire cross section . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.2 Rim mounting and inflation . . . . . . . . . . . . . . . . . . . . . 43
3.4.3 3D tire model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.4 Mesh effect on lateral force in 3D model . . . . . . . . . . . . . . . 45
3.4.5 Static loading of the tire . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Simulation procedure to compute the braking characteristic of a rolling tire 47
3.5.1 Mesh effect on the force equilibrium in vertical direction . . . . . 48
3.5.2 Effect of penalty parameter in the friction model on the longitudi-
nal force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 Simulation procedure to compute the cornering and combined slip char-
acteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4 Friction parameter identification using a Laboratory Abrasion and skid Tester 55


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Identification using lab scale experiments . . . . . . . . . . . . . . . . . . 56
4.3 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4 FE model of the setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.1 Material model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.2 2D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.4.3 3D model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.5 Parameter identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 Comparison of numerical and experimental results . . . . . . . . . . . . . 67
4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

5 Friction parameter identification using longitudinal slip characteristics 77


5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Identification using full scale experiments . . . . . . . . . . . . . . . . . . 78
5.3 Tire force and moment measurements . . . . . . . . . . . . . . . . . . . . 79
5.4 Friction parameter identification . . . . . . . . . . . . . . . . . . . . . . . 80
CONTENTS ix

5.5 Free-rolling rotational velocity: Comparison of FEM prediction and exper-


iments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.6 Comparison between FEM and MF predictions . . . . . . . . . . . . . . . 86
5.7 Effect of inflation pressure on the longitudinal force . . . . . . . . . . . . 88
5.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

6 Predictive capability of the FE tire model 93


6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2 Prediction of the handling characteristics . . . . . . . . . . . . . . . . . . 94
6.2.1 Pure cornering characteristic . . . . . . . . . . . . . . . . . . . . . 94
6.2.2 Combined slip characteristic . . . . . . . . . . . . . . . . . . . . . 100
6.2.3 Friction power distribution in the footprint . . . . . . . . . . . . . 105
6.3 Force and moment measurements and Magic Formula . . . . . . . . . . . 107
6.4 Comparison of the FE model and the Magic Formula . . . . . . . . . . . . 111
6.4.1 Pure cornering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.4.2 Combined slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

7 Conclusions and recommendations 121


7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

References 125

A Friction model implementation 133


A.1 Stick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.2 Slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

B Mesh effect on the force equilibrium in vertical direction 137

C Computation of slip velocity field for the LAT 100 setup 141

Samenvatting 143

Dankwoord 145

Curriculum Vitae 147


x
Nomenclature

Acronyms and abbreviations

acronym description acronym description


ALE arbitrary Lagrangian-Eulerian MF Magic Formula
DoE design of experiments NASA national aeronautics and space
FE(M) finite element (method) administration
FEA finite element analysis R radial
ISO international organisation for TNO Netherlands organization for applied
standardization scientific research
LAT laboratory abrasion and skid tester WLF Williams-Landel-Ferry equation
L.I. load (capacity) index

Operations and notation

symbol description symbol description


a, A scalar x̂, X̂ skew-symmetric matrix,
a, A vector or matrix associated with x, X
|·| absolute value × outer product
X Lagrangian coordinate · inner product
x Eulerian coordinate xT , A T vector or matrix transpose
ẋ, (ẍ) (double) time derivative || · || magnitude
∂ partial derivative

xi
xii N OMENCLATURE

Roman symbols and letters

symbol description unit


A0 cross sectional area m2
AExp experimental contact area m2
AMod calculated contact area m2
a line length m
tuning parameter
a Eulerian acceleration vector m/s2
aT shift factor
b line length m
tuning parameter
C10 hyperelastic material coefficient Pa
CF α cornering stiffness N/deg
CM z aligning stiffness Nm/deg
c line length m
tuning parameter
d tuning parameter
D1 hyperelastic material parameter Pa
E Young’s modulus Pa
E0 elastic storage modulus Pa
E 00 elastic loss modulus Pa
errorrel relative error
Ff ric frictional force N
Fx longitudinal force N
Fy lateral force N
Fz normal (vertical) force (load) N
f frequency rad/s
G0 shear storage modulus Pa
G00 shear loss modulus Pa
h tuning parameter friction model
penetration m
I identity matrix
J Jacobian matrix
k tuning parameter friction model
ks slope
L1,2,3 length m
L initial length m
l current length m
Mx overturning moment Nm
My moment or driving torque Nm
Mz self-aligning moment Nm
Mzr residual torque Nm
N OMENCLATURE xiii

symbol description unit


n cornering axis in normal direction
p contact pressure Pa
p0 tuning parameter friction model Pa
p1,2 tuning parameter
plimit lower bound on contact pressure Pa
Rα side slip rotation matrix
Rβ rotation matrix
Rc cornering rotation matrix
Rs spinning rotation matrix
R radius m
parameter friction model
Rinner inner tire radius m
Router outer tire radius m
r radius m
re effective rolling radius m
rl loaded rolling radius m
ru unloaded rolling radius m
T rigid axle position
T torque Nm
t time s
pneumatic trail m
ti orthogonal unit vector, i ∈ {1, 2}
uc longitudinal carcass deflection m
V velocity m/s
Vmax tuning parameter friction model m/s
Vsx longitudinal slip velocity m/s
Vsy lateral slip velocity m/s
Vx longitudinal (forward) velocity m/s
Vy lateral velocity m/s
v (longitudinal) slip velocity m/s
vc lateral carcass deflection m
vo lateral resultant force offset m
vs slip velocity m/s
v Eulerian velocity vector m/s
Width tire width m
X Lagrangian coordinate
x Eulerian coordinate
x x-position m
Y Lagrangian coordinate
y Eulerian coordinate
y y-position m
xiv N OMENCLATURE

Greek symbols and letters


symbol description unit
α side slip angle deg
β angle deg
χ deformation map
∆ increment in position m
δ phase lag rad
e nominal strain
γ̇, γ̇ slip velocity (vector) m/s
γ̇crit critical slip velocity m/s
κ longitudinal slip
κmax tuning parameter friction model
λi wavelength i m
µ friction coefficient
µk kinetic friction coefficient
µlock tuning parameter friction model
µM F Magic Formula dependent friction coefficient
µm tuning parameter friction model
µs tuning parameter friction model
static friction coefficient
Ω rotational velocity rad/s
ω rotational velocity rad/s
ωα free-rolling rotational velocity for nonzero rad/s
side slip angles
ωf ree free-rolling rotational velocity rad/s
ρ density kg/m3
σe nominal stress Pa
τ, τ frictional stress (vector) Pa
τcrit critical frictional stress Pa

Subscripts and indices


symbol description symbol description
0 center max maximum
c cornering n discrete time step
D deformable body p contact pressure dependent
disk abrasion disk R rigid foundation
eqv equivalent r reference frame
i index, i ∈ {1, 2} slip slip
j node j v slip velocity dependent
max maximum wheel wheel
n discrete time step
N OMENCLATURE xv

ISO sign conventions

Tire force and moment

x z

V α Fx Fz

Fy Fy
y y
Mz Mx

Top view tire Rear view tire

Tire velocity and slip velocity

Vy
V α Vx longitudinal slip
κ = − VVsxx
Vs Vsx
side slip
Vsy
tan α = Vx
Vsy

Top view tire


xvi
C HAPTER ONE

Introduction

Abstract / In this chapter a general introduction of tire modeling is presented. The current chal-
lenges in tire modeling using finite element methods are discussed, which form the motivation of
the research objectives. Based on these objectives, a research approach is presented and the main
contributions of this thesis are stated. The chapter ends with the outline of the thesis.

1.1 General introduction


The development of pneumatic tires started with the patent by John Boyd Dunlop in 1888
(Dunlop, 2010)1 and is still going on today. The first pneumatic tires had small cross sec-
tions and high inflation pressures, mainly for bicycle applications. From the 1920s, larger
tires were introduced for the upcoming vehicle industry. Two major evolutions took place
in the 1960s, the tubeless tire was introduced and bias ply tires were replaced with radial
ply tires, which improved the wear and handling properties significantly. The main dif-
ference between the bias and radial ply tire is the orientation of the plies. In bias ply tires,
the body ply cords are laid at angles substantially less than 90◦ to the tread centerline,
extending from bead to bead. In radial tires, the body ply cords are laid radially from
bead to bead, at 90◦ to the centerline of the tread. Two or more belts are laid diagonally
in the tread region to obtain the required strength and stability (Gent and Walter, 2005,
chapter 1). In Figure 1.1, a typical layout of a radial tire, which is now the standard for
passenger car tires, is shown.
The tire construction, such as aspect ratio and belt construction, depends on the size of
the tire and the target market. This information is printed on the sidewall of every tire,
e.g. 215/55 R16 97 H. The first number (215) is the nominal section width in mm, the

1
The idea of a pneumatic tire was already patented by Thomson (1847). Dunlop’s patent was later
declared invalid on the basis of this patent, but is generally considered as the first practical pneumatic tire.

1
2 1 I NTRODUCTION

Figure 1.1 / Cross section of a radial tire with the main components, Ghoreishy (2008)

second number (55) is the percentage of the height/width ratio of the cross section. The
R (radial) stands for the tire construction code. The rim diameter (16) is given in inches
and 97 is the load capacity index (L.I.), which is a reference to the maximum load capac-
ity. The last symbol (H) is the speed symbol, which corresponds to a maximum allowable
speed (European Tyre and Rim Technical Organisation, 2010).
More recent developments in the tire industry are the run-flat technology, which en-
ables the vehicle to continue at reduced speeds after deflation of a tire, and the so-called
ultralow-aspect tires, which have very short sidewalls (Rodgers, 2001).

1.2 Tire performance and modeling


All tires must meet the following fundamental set of performance factors (Rodgers,
2001):

• Provide load carrying capacity.

• Provide cushioning, damping and minimum noise and vibration.

• Transmit forces and moments.

• Resist abrasion.

• Have a low rolling resistance.

• Be durable and safe throughout the expected lifespan.

The different components of the tire determine the tire overall characteristics in response
to the application of load, torque or steering input, resulting in the generation of forces
1.2 T IRE PERFORMANCE AND MODELING 3

and deflection of the tire. The mechanical properties are often interrelated, which means
that a design change for one of the performance factors, can affect the other factors both
positively or negatively (Rodgers, 2001). Besides the engineering aspects, also economi-
cal factors, (limitations of) the manufacturing process (Walter, 2007b) and government
regulations (Walter, 2007a) have to be taken into account. All these aspects are captured
in specific performance criteria, which are visualized in a performance chart in Figure 1.2.
In this figure, a new tire design is compared with an existing reference tire for six perfor-
mance criteria. It should be noted that these criteria are not defined unambiguously in
literature, e.g. (Gent and Walter, 2005, chapter 1).

wet traction rolling resistance

110%
105%
100%
95%

noise and vibration ice and snow traction

dry traction wear resistance

Figure 1.2 / Example of a new tire design (dashed line) related to a reference tire for sev-
eral performance functions. Based on the tread performance chart (Mundl et al., 2008).

Dry, wet, and, ice and snow traction are directly related to the handling properties of the
tire and as such to safety issues, e.g. the braking distance at 100 km/h. Obviously, wear
resistance is related to the tire’s lifespan, while rolling resistance has direct influence on
vehicle fuel consumption. A tire rolling over a road generates undesired noise both at the
surroundings (exterior noise) and inside the vehicle itself (interior noise).
An accurate model of the tire behavior enables the engineer to optimize the overall per-
formance, while taking into account the different performance criteria. For this purpose,
several modeling techniques have been developed during the last decades.
In the case of handling models, a distinction can be made between empirical models and
(simple) physical models. Each type is developed for a specific purpose (Pacejka, 2006).
The (semi)-empirical models are based on experimental data. These models have a spe-
cific structure and the parameters are usually obtained by regression techniques. One of
the most well-known and widely used tire handling models is the Magic Formula model
of Pacejka (Bakker et al., 1987). This model is very useful to reproduce and interpolate
4 1 I NTRODUCTION

tire properties. The great advantage is of this model is the low computational cost. This
is a very important requirement for tire models that are used in vehicle dynamics, where
the tire is just one part of the total vehicle model. An overview of specific tire simulation
challenges in vehicle dynamics is presented by Rauh and Mössner-Beigel (2008). The
simple physical models use analytical expressions to describe the forces and moments
and can produce realistic results, e.g. the stretched string model (Pacejka, 2006), if the
parameters are assigned appropriate values. Usually the application field of these kind of
models is limited.
The main drawback of these two types of models is that the parameters are experimen-
tally determined from full scale tire tests and as such these models can not be used to
predict the influence of tire construction design changes. For detailed analysis of a tire
itself, the Finite Element Method (FEM) can be used. Such finite element models are
complex, but allow to investigate the effects of tire design parameters on the generated
forces and moments. FE models are nowadays a standard tool in the tire industry, and
their use opens the possibility of tire virtual prototyping.

1.3 Challenges in FE tire modeling


Finite element tire modeling can increase the insight on the relative influence of specific
tire properties on the tire behavior, decrease the development time and eventually reduce
development costs of new tires. However in practice most finite element models are still
not able to accurately match outdoor experiments.

INTERACTION MODEL
FE Model

INPUTS OUTPUTS
contact model

road

environment

Figure 1.3 / Schematic overview of the necessary components for every Finite Element
model of a rolling tire interacting with the environment.
1.3 C HALLENGES IN FE TIRE MODELING 5

The numerical modeling of a tire in combination with its environment, shown in Figure
1.3, is a difficult task, since different physical phenomena play a role. Typically the me-
chanical, thermal and fluid domains contribute to the tire response. These interrelated
domains and the wide range of operating conditions provide several challenges in the
modeling process:

• Material modeling. A correct description of the different rubber and cord-rubber


plies is required. Large deformations and large strains, as well as incompressibility
of rubber compounds should be accounted for.

• Contact modeling. Contact models in normal and tangential direction need to be


defined. Especially, friction models for tangential contact depend on environmental
conditions.

• Geometric modeling. Modeling of geometric shapes is nowadays not an issue any-


more, but in many cases the creation of valid elements for detailed tread patterns is
difficult.

• Temperature modeling. The temperature of the tire changes during operation,


which affects the mechanical properties of the tire and hence its behavior.

• Steady-steady versus transient modeling. Different numerical algorithms are re-


quired for steady-state rolling tires and the modeling of transient effects e.g. when
impacts with obstacles occur.

• Measurements for validation purposes. This is not directly related to FE tire mod-
eling, however necessary to validate FE models. Obtaining good measurements
of the forces and moments or other system quantities acting on rolling tires is a
difficult task.

The latest overview of the current state of the art of FE modeling of rolling tires is given
in the paper of Ghoreishy (2008). In the summary, it is stated that despite the substantial
progress, achieved during the last decades, the analysis of the complicated tire structure
is still a formidable task. Sofar, none of the current published works is capable of giving
a full analysis of the tire under different loading conditions. Instead, each work tried to
focus on some critical aspect of the tire and investigated this specific feature as deep as
possible.
Furthermore, the majority of tire related research in literature does not provide detailed
information about the used models and parameter values are often lacking, which makes
interpretation of the results difficult. Finally, a validation of the used FE models for rolling
situations with full scale tire experiments is often missing.
6 1 I NTRODUCTION

1.4 Objectives and research approach


This thesis focuses on the FE prediction of the steady-state handling characteristics of
rolling tires on dry roads. Although handling on wet roads is also important, the pres-
ence of water or snow add significant complexity to the tire/road interface. Additionally,
obtaining consistent measurements under controlled wet conditions is very challenging
in practice. Therefore, wet roads are not considered in this research.
The handling characteristics considered in this thesis are pure braking, pure cornering
and combined slip (Pacejka, 2006) and are defined as follows. In case of pure braking,
the tire is braked from free-rolling up to wheel lock. For pure cornering, the free-rolling
tire is steered up to ±12◦ side slip angle and in case of combined slip, the tire is braked
up to wheel lock when rolling at a constant slip angle. These conditions cover driving
situations from normal driving to extreme manoeuvres.
To evaluate the handling performance of a tire, both the static deformation and the
dynamic response of the tire rolling on the road should be accurately predicted. The
cornering, braking and traction of a tire depend on the generated friction forces. Friction
depends not only on the tread properties of the tire, but also on the road surface and en-
vironmental conditions. Frictional behavior for model parameterization can be acquired
by extensively testing tires under different conditions. Experimental characterization
of frictional properties of rubber compounds is cumbersome, since environmental
conditions influence these measurements. As a result, Coulomb’s friction law, with a
constant friction coefficient, is still often used in finite element simulations to predict
handling characteristics. It is however clear from experiments with elastomers that
rubber friction depends on various parameters like contact pressure, sliding velocity,
temperature and surface roughness. Because of these dependencies Coulomb’s law is
not sufficient to accurately predict the handling characteristics over the desired range.
Furthermore, numerical problems occur during steering at large side slip angles when
Coulomb’s law with realistic friction coefficients is used in finite element simulations.
To overcome these limitations a different strategy is needed to capture observed effects
of dry friction on the handling characteristics of rolling tires. Therefore the essential goal
of this thesis is:

The development of a robust and numerically efficient friction model for finite element tire
simulations and to create a framework for the identification and implementation of friction
related parameters.

Furthermore, this friction model should capture observed effects of dry friction and it
should be compatible with commercial FE codes. As mentioned above, rubber friction
depends on several parameters. This thesis focuses on developing a friction model,
1.4 O BJECTIVES AND RESEARCH APPROACH 7

which includes the influence of contact pressure and sliding velocity.


To achieve the above objective, the currently available state-of-the-art numerical methods
in the commercial finite element package ABAQUS are used. Emphasis is placed here on
the modeling of frictional contact of the tire with the road, the FE model of the tire itself
is provided by tire manufacturer Vredestein.

Inputs Lab scale experiments


3D Forces Experiments

Friction
FEM 3D Forces Simulations
model Lab scale setup

3D Forces Experiments
Full scale experiments

FE Tire
3D Forces Simulations
Inputs
Virtual prototyping

Figure 1.4 / Schematic overview of the two step experimental / numerical approach to
obtain friction information using both small scale and full scale experiments.

For the parameter identification of the friction model several measurements have been
carried out on two experimental setups on different scales, as shown in Figure 1.4. First,
friction experiments are performed on a commercially available small scale lab setup
(LAT 100). On this setup controlled experiments on a small solid tire, under low rolling
velocities, are performed. These measurements are used to identify the parameters of the
friction model, which are contact pressure dependent. The friction model is coupled to
an FE model of this lab setup to simulate the hub forces. These forces are compared to
the experimentally found hub forces and used to validate the parameters of the friction
model. With the small scale setup it is only possible to conduct experiments at low sliding
velocities, since excessive wear of the small tire occurs at higher velocities. Therefore
outdoor braking experiments with a full scale tire at different velocities are conducted to
obtain a slip velocity dependent parameter set for the friction model. These experiments
are carried out with the TNO Tyre Test Trailer. Finally, the completely identified friction
model is validated by comparing simulated and measured cornering and combined slip
characteristics.
8 1 I NTRODUCTION

1.5 Contributions
The main contributions presented in this thesis are:

• The definition and implementation of a complex friction model in the commercial


package ABAQUS, which is suitable for tire handling simulations in the full operating
range.

• A procedure for the identification of the parameters of the friction model using
a two step experimental / numerical approach, combining small scale lab experi-
ments with full scale outdoor experiments.

• Validation of the predicted handling characteristics with full scale outdoor experi-
ments, covering the entire operating range.

1.6 Outline of this thesis


In Chapter 2, the modeling framework to obtain the handling characteristics of rolling
tires, with friction, is described. The choice for the numerical approach and the friction
model, based on a literature overview, is made. The background of the numerical method
is reviewed and the implementation of the chosen friction model is described.
After that, the strategy for identification of the parameters of the chosen friction model is
explained in more detail in Chapter 3. Furthermore, the design of the used test tire and
the corresponding layout of the FE model are presented. The different simulation steps
required to compute the pure braking, pure cornering and combined slip characteristics
are discussed.
In Chapter 4, the identification of the contact pressure dependent part of the tire friction
model is given. Friction data is obtained using a commercial laboratory abrasion and
skid tester. From this data the contact pressure related parameters are derived and imple-
mented in an FE model for the tire/disk contact.
Next, the identification of the slip velocity dependent part, using measured axle forces, is
presented in Chapter 5. The complete identified friction model is then coupled to the FE
model of the tire under testing. The computed steady-state longitudinal slip characteris-
tics are compared with the full scale outdoor experiments and a discussion of the results
is given.
In chapter 6, the fully identified friction model is used to compute the pure cornering
and combined slip characteristics. The predicted characteristics are compared with ex-
periments and it is shown that the handling performance of the tire can be adequately
predicted with the identified friction model.
Finally, in Chapter 7, the main conclusions are summarized and recommendations for
future work are given.
C HAPTER TWO

Modeling framework for steady-state


rolling tires with friction

Abstract / A numerical modeling framework to obtain the handling characteristics of rolling tires,
with friction, is defined in this chapter. Based on a literature overview of friction models and tire
simulation methods the choice for the numerical approach and the friction model is motivated.
The background of the numerical method is presented together with the implementation of the
chosen friction model. The validation of this implementation is presented at the end of the chap-
ter.

2.1 Introduction
In this chapter the numerical approach to obtain the handling characteristics of rolling
tires including friction is described. First a short literature review of friction is presented
in Section 2.2, which shows the difference of rubber friction compared to other solids and
in Section 2.3 Finite Element Analysis (FEA) related to tires is discussed. Based on these
findings a numerical method is chosen, which is able to efficiently compute the handling
characteristics of steady-state rolling tires. Furthermore, the choice of the friction model
is motivated in Section 2.4. The background of the numerical method is reviewed in Sec-
tion 2.5 and the implementation of the friction model is discussed in Section 2.6. This
implementation is validated using different test models, which are presented in Section
2.7. In Section 2.8 a description of an algorithm to obtain the solution, which corre-
sponds to a so-called free-rolling tire is discussed. The chapter ends with conclusions in
Section 2.9.

9
10 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

2.2 Overview of friction models

2.2.1 Amontons-Coulomb

Although Leonardo da Vinci (1452-1519) is generally credited as being the first to develop
the basic concepts of friction (Bowden and Tabor, 1964), Amontons formulated the first
two laws of friction and Coulomb the last two. These classic laws are summarized by
Moore (1972) as follows:

• The friction force is proportional to load.

• The coefficient of friction is independent of the apparent contact area.

• The static coefficient is greater than the kinetic coefficient.

• The coefficient of friction is independent of the sliding velocity.

This can be formulated as

Ff ric = µFz , (2.1)

where Ff ric is the frictional force, µ the constant friction coefficient and Fz the applied
load and (2.1) is usually referred to as the Coulomb friction model.

Deviations from Coulomb friction

Experiments often show deviations from the basic Coulomb friction model. Variations of
the basic model are the difference between a static µs and a kinetic µk friction coefficient,
see Figure 2.1 for some examples. This transition can be discontinuous or continuous,
e.g. using an exponential decaying function between the static and kinetic coefficient.
In general, the friction coefficient varies for increasing sliding velocity especially in the

µ µ µ µ
friction coefficient

µs µk µs µs µs
µk µk
µk

v v v v
sliding velocity

Figure 2.1 / Coulomb friction model and possible variations.


2.2 O VERVIEW OF FRICTION MODELS 11

presence of lubricants, e.g. the Stribeck curve (Stribeck, 1902). First the coefficient de-
creases to a minimum value and after that it increases with higher sliding velocity. The
review paper of Olsson et al. (1998) describes several static friction models, which focus
only on sliding velocity, and dynamic friction models. The dynamic models also describe
the stick-slip transitions around zero sliding velocity. Examples are the Dahl model and
two variants of this model, the LuGre model and the model of Bliman and Sorine. These
models are often used in control systems, in such a way that they can be used for friction
compensation in mechanical systems.

2.2.2 Rubber friction

Rubber friction differs from friction of most other solids. According to Moore (1972) the
second friction law appears to be valid only for materials possessing a definite yield point
and it is does not apply to elastic or viscoelastic materials. The third law is not obeyed by
any viscoelastic material and the fourth law is not valid for any material (Moore, 1972).
It is also clear from experiments with elastomers that rubber friction depends on various
parameters like contact pressure, sliding velocity, temperature and surface roughness.
Because of these dependencies Coulomb’s law is not sufficient to model the frictional
response of an elastomer.
The friction force between rubber and a rough surface has two contributions commonly
described as the adhesion and hysteretic components (Moore, 1972). The hysteretic com-
ponent results from internal friction of the rubber; during sliding asperities of a rough
surface exert oscillating forces on the rubber surface. This leads to cyclic deformations of
the rubber and to energy dissipation caused by the internal damping of the rubber (Pers-
son, 2001). The adhesion component is caused by the intermolecular attractive forces
between the contacting bodies (Wriggers and Reinelt, 2009).
The article of Grosch (1963) is one of the early reports, which describes the analogy be-
tween the friction coefficient as function of sliding velocity v and the energy dissipated
per cycle (tan δ) as function of frequency f , see Figure 2.2. The phase angle is given
by tan δ = E 00 /E 0 , with δ the phase lag between stress and strain; E 0 and E 00 are the
storage and loss modulus respectively. In this work experimental results of different vul-
canized rubbers are presented. The experiments are conducted at different temperatures
and shifted, with shift factor aT , to a master curve using the WLF equation, developed by
Williams et al. (1955).
Grosch indicates that friction is due to energy dissipated when rubber is compressed
and released by asperities. Friction for dry sliding on a smooth surface is due to energy
dissipated as rubber sticks and slips on a molecular scale. In the case of sliding on a rough
dry surface the dissipative process appears at different speeds, corresponding to different
length scales: asperities (hysteretic component) and molecules (adhesion component).
12 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

Several contributions to the friction coefficient µ can arise at the same sliding speed from
stick-slip processes occurring at different length scales, see Figure 2.3.
An overview of models to describe the frictional interaction between tire and road is given
next.

Figure 2.2 / Grosch’s interpretation of the equivalence between master curves of rubber
friction versus sliding velocity and tan δ versus frequency f , Gent (2007).

Figure 2.3 / Combined effect of sliding on a rough and dry surface, Gent (2007).

Savkoor proposed a model, based on results of Grosch, where he described the shape of
an isothermal master curve with an empirical relation (Savkoor, 1966, 1987)
  
2 2 v
µ(v) = µs + (µm − µs ) exp −h log , (2.2)
Vmax

where µs is a static coefficient of friction, µm the peak value of the function (which occurs
at |v| = Vmax ) and h is a dimensionless parameter reflecting the width of the speed range
in which friction varies, as shown in Figure 2.4. According to Savkoor (1966), the values
of Vmax and µm depend on the viscoelastic properties of the rubber. At higher tempera-
2.2 O VERVIEW OF FRICTION MODELS 13

tures Vmax increases and the friction curve is shifted significantly towards higher speeds.
The advantage of this function is that the parameters are related directly to the shape of
the friction curve, which makes a physical interpretation possible. The disadvantage of
this model is that only the influence of sliding velocity on the friction coefficient is con-
sidered, at constant temperature, and that the influence of contact pressure is not taken
into account.

µm
µ

µs h

Vmax v

Figure 2.4 / Model proposed by Savkoor to describe an isothermal master curve.

Based on the original Schallamach 2D model (Schallamach, 1971) a generalized theory


that predicts the buckling effects, known as Schallamach waves (Schallamach, 1952; Bar-
quins, 1992) which occur at very smooth surfaces, is presented by Berger and Heinrich
(2000). In contrast to the original Schallamach model, where Coulomb friction is used,
this generalized theory considers a friction coefficient that depends on the local normal
pressure over the contact line,
 p n
µ(p) = µ0 (2.3)
E
with E the linear elastic modulus and µ0 and n fitted parameters. A similar pressure
dependent model is used in the work of Trinko (2007), where µ0 and E are replaced with
a single constant.

The phenomenological models proposed by Dorsch et al. (2002) are fully empirically,
where a velocity v and pressure dependency p for the friction coefficient is assumed.

µ(v, p) = c1 pc2 v c3 (2.4)


µ(v, p) = c1 p + c2 p2 + c3 v + c4 v 2 + c5 pv (2.5)

The friction coefficient is described by a power law or a linear approximation, in Dorsch


et al. (2002) a so-called full quadratic model is presented. Although these are very simple
models, several experiments are required to identify the coefficients ci . Design of Experi-
ments (DoE) techniques can be used to determine the number and range of pressure and
14 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

sliding velocity values needed (Montgomery and Runger, 1999).


A velocity and pressure dependent model (Wriggers, 2006), which is proposed by Nack-
enhorst (2000) is given by
 c 3
p v v
µ(v, p) = c1 + c4 ln − c6 ln , (2.6)
c2 c5 c7

where all seven parameters must be determined using experiments. In this model the
influence of contact pressure and sliding velocity is decoupled and summed to obtain the
total friction coefficient, which indicates that this model also originates from statistical
techniques. All these three models are able to accurately fit experimental data, but the
obtained parameters of these models have no straightforward physical interpretation.

The model proposed by Huemer et al. (2001a) is developed for sliding rubber blocks on
ice and concrete. The phenomenological friction law is given by

α|p|n−1 + β
µ(v, p) = b c
(2.7)
a + |v|1/m + |v|2/m

and is developed for a macroscopic model. In this approach the coefficient of friction de-
pends on normal pressure p, sliding velocity v and temperature. The friction coefficient
itself (2.7) is only a function of normal pressure and sliding velocity. Temperature effects
are incorporated using the WLF transformation, i.e., if the current temperature is differ-
ent from a reference temperature, an equivalent new sliding velocity for the reference
temperature is calculated.
The friction law in (2.7) requires seven parameters (a, b, c, n, m, α, β), which must be
identified using experimental data. They account for the dependence of the friction coef-
ficient on friction surface, rubber compound, and dimensions and geometric shape of the
rubber block. The identification is based on a least square error method and performed
iteratively. First the coefficients a, b and c related to the sliding velocity are fitted and then
α and β related to the contact pressure, this process is repeated until a defined error cri-
terium is reached. The whole process is done for every value of n and m. Furthermore, in
the identification procedure the contact pressure is replaced with the averaged pressure
on the contact surface. An evaluation of the model is shown in Figure 2.5.
This work has been continued by Hofstetter et al. (2003), where a thermo-mechanical
coupling has been introduced. Energy dissipation during sliding is converted into
heat and this heat flux causes a temperature rise of the rubber and road. Simulations
of abrasion of the rubber block are considered in Hofstetter et al. (2006), where the
obtained numerical results are also compared with experimental data. Material loss
occurs only at the front edge and is captured qualitatively with the proposed friction and
abrasion models. However the model also predicts material loss in the middle of the
2.2 O VERVIEW OF FRICTION MODELS 15

0.9

0.85

0.8
µ [−]

0.75

0.7

0.65
0 0.5 1 1.5 2
log(v) [mm/s]

Figure 2.5 / Friction coefficient as function of sliding velocity using (2.7), the parameters
are extracted from Hofstetter et al. (2006) and a chosen contact pressure of 0.5 MPa.

bottom surface, but this is not observed experimentally.

A different approach is used by Persson, who has published many papers (Persson, 1993,
1995, 1998, 1999, 2002; Persson and Tosatti, 2000; Persson and Volokitin, 2000, 2002;
Persson et al., 2002) on the subject of rubber friction and the role of the surface, which
is in contact with the rubber. This theory is a continuation of the early studies of Grosch.
Persson states that the friction force is related to the internal friction of the rubber, which
is a bulk property of the material. The hysteretic friction component is determined by
sliding of the rubber over asperities of a rough surface. These oscillating forces lead to
energy dissipation. The contribution of a every asperity size can be described with a
fractal description of the rough surface. Every length scale λ, up to the largest particles
of asphalt, can be related to a frequency: f ∼ v/λ. The friction coefficient is based on
analytical expressions, which limits the rubber models to linear elastic theory.
A similar method is described by Klüppel and Heinrich (2000), but this theory is based
on a cylindrical rubber block undergoing only a one-dimensional deformation during
sliding contact with a rough surface. A comparison of experiments and the predictive
capabilities of the physical theories of Klüppel and Heinrich (2000) and Persson (2001)
is presented by Westermann et al. (2004). It is concluded that each theory has one open
parameter, which needs to be fitted on experimental data to obtain a good agreement
between theory and data. However this match only holds for a sliding speed interval up
to a few cm/s. Towards higher sliding speeds, systematic deviations appear, which are
related to flash temperature effects in the contact patch.
This flash temperature effect is described in a recent article of Persson (2006), where
16 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

he also takes the local heating of the rubber into account. At very low sliding velocities
the temperature increases are negligible because of heat diffusion, but already for
velocities in the order of 0.01 m/s local heating plays a role. He shows that in a typical
case the temperature increase results in a decrease in rubber friction with increasing
sliding velocity, if the sliding velocity is above 0.01 m/s. The advanced model, with flash
temperature effects included, is able to predict the friction coefficient for slightly higher
sliding speeds compared to the model without flash temperature. However validated
models for the range of sliding velocities that appear in tire/road contact are not available
yet.

An FE multi-scale approach for frictional contact is proposed by Wriggers and Reinelt


(2009). The proposed model is based on the analytical models of Klüppel and Heinrich
(2000) and Persson (2001), however it is well-known that large deformations occur when
a tire is in contact with a rough road surface and therefore the numerical simulations
are based on finite deformation models instead of linear elastic theory. The numerical
calculation of a rough surface demands a very detailed model to include all length scales
and a model of a tire tread block also requires a very fine mesh to accurately describe
the contact surface. This is not possible with current computer resources and therefore
the rough surface is modeled with only a few superimposed harmonic functions and the
tread is replaced with a small rubber block. On each length scale several simulations for
varying sliding velocity are carried out, where the following function is fitted though the
obtained data points
 c
2vv
µ(v, p) = µmax , (2.8)
v2 + v2

in which v denotes the sliding velocity, where the maximum point of the friction curve
µmax is reached. The dependency of the applied normal pressure is included by the func-
tions

v = ap, (2.9)
b
µmax = arctan(dp), (2.10)
p

which describe the effect that for increasing global pressure the maximum friction value
decreases and is shifted to larger velocities. The friction law on micro-scale requires
a fit of four parameters a, b, c and d, which are obtained by a nonlinear least-square
method. The obtained homogenized friction law is then applied within each so-called
representative contact element at the next larger scale. This method is able to predict the
qualitative frictional behavior, but is not suitable for complete tires due to computational
limitations.
2.3 T IRE RESEARCH USING FE M ODELS 17

2.3 Tire research using FE Models


The use of FEM in tire development started together with the development of nonlinear
FEM. All four types of nonlinearities (Kouznetsova, 2006) in solid mechanics are present
in pneumatic tires. Geometric nonlinearity occurs due to the large displacements and
rotations involved, while material nonlinearity is present for the almost incompressible
(visco)elastic rubbers. Force boundary conditions nonlinearities arise as pressure load
inside the tire and displacement boundary conditions nonlinearities are present due to
contact with a foundation. This creates changes in the boundary conditions during a
simulation.
There is a vast literature, sometimes related to tires, on computational developments for
contact problems in FE and quite some work is also incorporated into ABAQUS. Examples
are the works of Oden on friction and rolling contact (Oden and Pires, 1983, 1984; Oden
and Martins, 1985; Oden and Lin, 1986; Oden et al., 1988; Faria et al., 1989), Laursen and
Simo in the field of contact problems with friction (Simo and Laursen, 1992; Laursen and
Simo, 1993b,a) and Padovan on rolling viscoelastic cylinders (Zeid and Padovan, 1981;
Padovan, 1987; Kennedy and Padovan, 1987; Nakajima and Padovan, 1987; Padovan et
al., 1992). More recent are the works of Wriggers (Wriggers et al., 1990; Zavarise et
al., 1992; Haraldsson and Wriggers, 2000; Bandeira et al., 2004; Wriggers, 2006) on
constitutive interface laws with friction.
Literature specifically related to tires does usually not provide detailed information. The
main reason is that most tire manufacturers use own (in-house) finite element codes and
specific implementations are kept confidential. However the literature provides insight in
the trends and developments of tire modeling throughout the past decades. Besides the
journal of Tire Science and Technology the reader is referred to the two overview papers
of Mackerle (1998, 2004) about rubber and rubber-like materials, finite element analyses
and simulations for an extensive reference list.
The following overview focuses on methods to obtain the cornering and braking forces
acting on a rolling tire with friction. A distinction is made between implicit and explicit
methods.

2.3.1 Implicit analysis

Static and quasi-static analysis

Static analyses are used when inertia effects can be neglected and time-dependent mate-
rial effects are not included. In this case the time increments are then simply fractions
of the total period of the step, which are used as increments in the analysis. If time-
dependent material effects are taken into account, such as viscoelastic materials, the ap-
proach is called quasi-static. An implicit analysis is solved using an incremental-iterative
18 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

procedure, which requires matrix inversion (SIMULIA, 2009c).


One of the first overview papers, where FE models and the contact problem for tires are
described, is the paper of Noor and Tanner (1985). In this article, for a NASA research
program for the space shuttle tires, the current status and developments of computational
models for tires are summarized. This review has been made again by Danielson et al.
(1996). Basically all the tire models are axisymmetric, with no tread pattern or only cir-
cumferential groves due to the limits of computational resources. The effect of friction,
using the Coulomb model, is investigated on the vertical load versus vertical deflection
curve.
Another popular way to reduce the number of degrees of freedom is the application of
the global-local analysis. In this approach, an analysis of the complete structure is first
performed with a coarse mesh. After that a part of the structure is meshed finer and in-
terpolated displacements are applied at the boundaries of this region. Such an approach
can be used to compute the forces in the contact area of a deflected tire (Gall et al., 1995;
Meschke et al., 1997). Although a local model is able to give detailed numerical results
corresponding to a tread block, the accuracy is strongly influenced by the simple global
model.
With the ever increasingly computational power it is nowadays possible to mesh a part
or even the whole tread of the tire with a detailed pattern, e.g. Cho et al. (2004), and
perform very detailed static analyses including footprint shapes as function of axle load
and inflation pressure.
There are three possibilities to obtain the cornering and braking forces acting on a rolling
tire, which follow on a (quasi)static analysis. The first possibility is an implicit dynam-
ical analysis, the second one is the arbitrary Lagrangian-Eulerian method and the last
possibility is an explicit analysis.

Dynamic analysis

An implicit dynamical analysis is not often used for rolling tires, since it is well-known
that this type of analysis is not efficient in solving changing contact conditions. The
nonlinear equation solving process is expensive due to the Newton iterations, and if the
equations are very nonlinear, as in the case of changing contact, it may be difficult to
obtain a solution (SIMULIA, 2009c).

Arbitrary Lagrangian-Eulerian method

The arbitrary Lagrangian-Eulerian (ALE) method is developed for numerical analysis of


rolling contact problems, see the articles of Nackenhorst (2004); Ziefle and Nackenhorst
(2005) and Laursen and Stanciulescu (2006). This method converts the steady state mov-
ing contact problem into a pure spatially dependent simulation, where the mesh is fixed
2.3 T IRE RESEARCH USING FE M ODELS 19

in space and the material flows trough the mesh. Thus the mesh needs to be refined only
in the contact region, which leads to a computational time reduction.
This type of modeling, for simulating tire spindle force and moment response under side
slip angles is described by Darnell et al. (2002). The simulation model is composed of
shell elements, which model the tread deformation, coupled to special purpose finite ele-
ments that model the deformation of the sidewall and contact between the tread and the
ground, where Coulomb friction is used. Despite the seemingly simple model the results
correspond quite well with experiments.
FE simulations with ABAQUS, in which the effect of tire design parameters on lateral
forces and moments is studied, are presented in a paper of Olatunbosun and Bolarinwa
(2004). Parametric studies are performed on a simplified tire (no tread, negligible rim
compliance and viscoelastic properties, shear forces modeled with Coulomb friction with
constant µ) and a comparison with literature is made to show the computational time re-
duction of the method compared to explicit analyses. The effects of variations in stiffness
and geometry on the nonuniformity of tires is investigated by Jeong et al. (2007) using
ABAQUS , where also a Coulomb friction model is used for the tire-road interaction.

2.3.2 Explicit analysis

In an explicit analysis the dynamic response problems are solved using an explicit direct
integration procedure. The displacements and velocities are calculated in terms of quan-
tities that are known at the beginning of an increment and no iterations and no tangent
stiffness matrix are required, which is an advantage compared to the implicit method.
However due to the explicit time integration a very small time-step, which depends on
the highest frequency present in the model, is usually required. Therefore this approach
is ideal to simulate transient behavior in a short time span, such as impact of a tire with a
cleat. It can also be used to compute handling characteristics, but longer time spans are
required to reach the steady-state situation. Furthermore for these longer time spans the
risk of error accumulation is present as shown by Tönük and Ünlüsoy (2001).
Explicit simulations to predict tire cornering forces with a maximum side slip angle of
three degrees, using the package PAM -SHOCK, are presented by Koishi et al. (1998). Be-
sides a comparison with experiments, parametric studies on the effect of inflation pres-
sure, belt angle and rubber modulus are performed. Koishi et al. (1998) use the Coulomb
friction model, with coefficient equal to one.
A prediction of tire cornering forces on a drum is given in a paper of Tönük and Ünlü-
soy (2001), where the FE package MARC is used. A comparison with experiments is also
presented. With the model a maximum side slip angle of five degrees is obtained, higher
angles created problems caused by the used Coulomb friction model and error accumu-
lation dominates the model results before a steady-state is reached, which occurs even
earlier with higher normal loads. In the region below five degrees, when the cornering
20 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

force is almost linear for increasing side slip angle, the results show a good comparison
with experimental data.
A cleat test, with a treadless tire model, can be found in a paper of Olatunbosun and
Burke (2002), where the package NASTRAN has been used. A comparison with experi-
ments shows deviations as the traverse speed increases above 20 km/h.
Rao et al. (2003) simulated the dynamic behavior of a pneumatic tire using
ABAQUS / EXPLICIT to replace the extensive measurement program test to fit Magic For-
mula parameters. A study on a passenger car radial tire to simulate cornering behavior,
braking behavior, and combined cornering & braking behavior is presented. Furthermore
the effect of camber angle and grooved tread on tire cornering behavior is studied. To re-
duce computational load two tire models are used, one treadless tire and a tire with five
circumferential groves. Again the Coulomb friction model with coefficient equal to one
is used. The simulations are however unstable for side slip angles above five degrees
and longitudinal slip larger than 12%, which is explained due to the use of an inadequate
friction model.
In an article of Cho et al. (2005) the dynamic response of a fully patterned tire rolling
over a cleat is presented, using ABAQUS / EXPLICIT. A constant friction coefficient of one
is used for the tire-cleat contact. To decrease the computational load the fiber-reinforced
rubber is modeled with composite shell elements and mass lumping is used to decrease
computation time (SIMULIA, 2009c). Kerchman (2008) conducted an analysis for ride
and harshness analysis with a detailed tire-wheel model coupled with a suspension and
attached to a simplified vehicle model, again with the Coulomb friction model.

2.4 Modeling framework

2.4.1 Numerical method

Based on the literature overview it is clear that there are only two methods suitable to ob-
tain the cornering and braking forces of a rolling tire: the ALE method or the dynamical
explicit approach. The explicit method does not require the inversion of the global mass
and stiffness matrices, which is an advantage. However due to the explicit time integra-
tion a very small time-step is required to obtain a stable solution. Therefore it is ideal
to simulate transient behavior in a very short time span, such as impact of a tire with a
cleat. It is however not ideal to compute handling characteristics, since longer time spans
are required and this results in unacceptable computation times. To overcome this the
ALE method can be used, which is also available in ABAQUS, and is developed to com-
pute the steady-state response of rolling structures. This is an effective method to obtain
the global force and moment characteristics of a tire under different driving conditions,
such as braking or side slip and camber angles (SIMULIA, 2000). This so-called steady-
2.4 M ODELING FRAMEWORK 21

state transport analysis (SIMULIA, 2009a, chapter 6.4) uses a moving reference frame
in which rigid body rotation is described in an Eulerian manner and the deformation is
described in a Lagrangian manner (the ALE-formulation). Furthermore frictional effects,
inertia effects, and history effects in the material can all be accounted for.
In an article of Kabe and Koishi (2000) a comparison between the steady-state transport
method and ABAQUS / EXPLICIT and experiments for steady-state cornering tires is made.
Although the results of both methods are closer to each other than to the experiments,
the steady-state transport method is significantly faster, 6 hours and 8 days respectively.
This, together with the possibility to reduce the number of elements outside the contact
area, clearly shows the benefits of the steady-state transport analysis and therefore this
method is chosen to compute the steady-state characteristics of a rolling tire.

2.4.2 Friction model

From the literature review it follows that Coulomb friction is still often used in numeri-
cal simulations. Deviations from the experiments are often attributed to this model and
this indicates that a different model should be used to improve the results. One of the
restrictions of the steady-state transport approach is that the underlying surface must be
(rigid) flat, convex or concave (SIMULIA, 2009a, chapter 6.4). This means that it is not
possible to incorporate a rough road surface as used by Wriggers and Reinelt (2009).
So the possible surface effects should be captured in the parameters of a friction model,
since a flat surface can not exert oscillating forces on the rubber surface. Although the
models of Klüppel and Heinrich (2000) and Persson (2001) incorporate the effect of sur-
face roughness directly in their model description, the models are not suitable for high
sliding velocities yet, which occur when tires are tested at high velocities.
If the sliding velocity influence in the friction models (2.2), (2.4), (2.5), (2.6) and (2.7) is
compared, it follows that all models are able describe the shape as observed by Grosch.
The models given by (2.4), (2.5) and (2.6) provide however no direct insight in the fric-
tional properties and are therefore not preferred. The model of Huemer et al. (2001a) is
developed for a rubber block with a specific geometry and dimension, which means that
(2.7) describes the combined effect of contact and block geometry on the friction force.
Therefore the model of Savkoor is chosen to describe the sliding velocity dependence,
since the parameters in this model are directly related to the shape of the friction curve.
It is shown by Lupker et al. (2004) that this model can also be used in situations with
very high sliding velocities.
Furthermore the pressure dependence in most models show a decreasing friction coeffi-
cient for increasing pressure, which is not included in the original model of Savkoor. To
incorporate this effect (2.2) is extended with a pressure term (Lupker et al., 2004), to give
22 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

a friction law, which depends on the contact pressure and the slip velocity as follows
 −k    
p 2 2 v
µ(p, v) = µs + (µm − µs ) exp −h log , (2.11)
p0 Vmax
where the parameters p0 and k are related to the contact pressure and the parameters
µs , µm , h and Vmax are related to the sliding velocity. This model can also be found in the
work of Smith et al. (2008), where the steady-state approach of ABAQUS is used to predict
the wear of a tread profile.

2.5 Steady-state transport analysis


In this section the relevant parts of the ALE approach are presented, which are based
on the documentation of SIMULIA (2009c). The kinematics of the rolling problem are
described in terms of a coordinate frame that moves along with the ground motion of
the body. In this moving frame the rigid body rotation is described in a spatial or Eule-
rian manner and the deformation in a material of Lagrangian manner. This kinematic
description converts the steady-state moving contact field problem into a purely spatially
dependent simulation. In the following derivation Lagrangian coordinates are denoted in
uppercase (e.g. X) and the Eulerian coordinates are denoted in lowercase (e.g. x).
A deformable body is rotating with a constant angular rolling velocity ω around a rigid
axle T at X0 , which in turn rotates with a constant angular velocity Ω around the fixed
cornering axis n, which is normal to the rigid surface, through point Xc , see Figure 2.6.
Hence, the motion of a particle X at time t consists of a rigid rolling rotation to position

T X0

n

Xc

Figure 2.6 / Constant cornering motion in the ALE approach. Figure reproduced from
(SIMULIA, 2009c).
2.5 S TEADY-STATE TRANSPORT ANALYSIS 23

Y, described by

Y = Rs · (X − X0 ) + X0 , (2.12)

where the spinning rotation matrix Rs is defined as

Rs = exp(ω̂t), (2.13)

with ω̂ the skew-symmetric matrix associated with the rotation vector ω = ωT. This rigid
rolling rotation is followed by a deformation to a point x, and a subsequent cornering
rotation around n to position y so that

y = Rc · (x − Xc ) + Xc , (2.14)

where Rc is the cornering rotation given by

Rc = exp(Ω̂t), (2.15)

and Ω̂ is the skew-symmetric matrix associated with the rotation vector Ω = Ωn. The
velocity of the particle then becomes

v = ẏ = Ṙc · (x − Xc ) + Rc · ẋ. (2.16)

To describe the deformation of the body a map χ(Y, t), shown in Figure 2.7, is intro-
duced, which gives the position of a point x at time t as a function of its location Y at
time t so that

x = χ(Y, t). (2.17)

Initial configuration x
X

Current
configuration
Rs

Reference configuration
χ
Y

Figure 2.7 / ALE decomposition using a rigid rotation and deformation step.
24 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

The time derivative of (2.17) is given by


∂χ ∂Y ∂χ
ẋ = · + , (2.18)
∂Y ∂t ∂t
where
∂Y
= Ṙs · (X − X0 ) = ωT × (Y − X0 ). (2.19)
∂t
Noting that Ṙs = ω̂ · Rs = ω T̂ · Rs , and introducing the circumferential direction
Y − X0
S=T× , (2.20)
R
where R = |Y − X0 | is the radius of a point on the reference body, the velocity of the
reference body can now be written as
∂Y
= ωRS (2.21)
∂t
and (2.18) can be written as
∂Y ∂χ ∂χ ∂χ
ẋ = ωR ·S+ = ωR + , (2.22)
∂t ∂t ∂S ∂t
where S is the distance-measuring coordinate along the streamline. The derivative of
(2.15) is given by

Ṙc = Ω̂ · Rc = Ωn̂ · Rc . (2.23)

The velocity of the particle can be written as


∂χ ∂χ
v = Ωn × (x − Xc ) + ωRRc · + Rc · . (2.24)
∂S ∂t
The acceleration is obtained by differentiation of (2.24)

∂χ ∂χ
a = Ω2 (nn − I) · (x − Xc ) + 2ωΩRn × Rc · + 2Ωn × Rc · (2.25)
∂S ∂t
2 2 2
∂ χ ∂ χ ∂ χ
+ω 2 R2 Rc · 2
+ 2ωRRc · + Rc · 2 .
∂S ∂S∂t ∂t

To obtain expressions for the velocity and acceleration in the reference frame tied to the
body, the following transformations are used

vr = Rc T · v, ar = Rc T · a, (2.26)

such that
∂χ ∂χ
vr = Ωn × (x − Xc ) + ωR + (2.27)
∂S ∂t
2.5 S TEADY-STATE TRANSPORT ANALYSIS 25

and
∂χ ∂χ
ar = Ω2 (nn − I) · (x − Xc ) + 2ωΩRn × + 2Ωn × (2.28)
∂S ∂t
∂ 2χ ∂ 2χ ∂ 2χ
+ω 2 R2 + 2ωR + .
∂S 2 ∂S∂t ∂t2
∂χ
For steady-state conditions it holds that ∂t
= 0 and these expressions reduce to
∂χ
vr = Ωn × (x − Xc ) + ωR (2.29)
∂S
and
2
2 ∂χ 2 2∂ χ
ar = Ω (nn − I) · (x − Xc ) + 2ωΩRn × +ω R . (2.30)
∂S ∂S 2
The first term of (2.30) can be seen as the acceleration that gives rise to centrifugal forces
resulting from rotation about n. The second term can be identified as the acceleration
that gives rise to Coriolis forces. The last term combines the acceleration that give rise to
Coriolis and centrifugal forces resulting from rotation about T. When the deformation
is uniform along the circumferential direction, both Coriolis effects vanishes so that the
acceleration gives rise to centrifugal forces only.
The velocity of the center of the body X0 is given by
v0 = Ωn × (X0 − Xc ) (2.31)
since the motions due to rolling and deformation vanish on the axis.
In the case of straight line rolling Ω → 0 (2.29) and (2.30) reduce to
∂χ
v = v0 + vr = v0 + ωR (2.32)
∂S
and
∂ 2χ
a = ar = ω 2 R2 . (2.33)
∂S 2

2.5.1 Contact conditions for steady-state rolling

Given two points on the surfaces of two bodies in contact, the relative velocity can be
expressed as
v = v D − vR , (2.34)
where vD is the velocity of a point on the deformable body, see (2.27), and vR the veloc-
ity of a point on the rigid foundation. This can be split into the normal and tangential
components. The rate of penetration is
∂χ ∂χ
ḣ = −n · v = n · vR − ωRn · −n· . (2.35)
∂S ∂t
26 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

For any point in contact it follows that n · ∂χ/∂S = 0, therefore


∂χ
ḣ = n · vR − n · , (2.36)
∂t
which in incremental form reduces to the standard contact condition

∆h = n · (∆xR − ∆xD ), (2.37)

with ∆xR and ∆xD the position increment of points on the rigid foundation and the
deformable body respectively. For steady-state conditions it follows that n · ∆xR = 0 and
∆xD = 0.
The rate of slip is given by

γ̇i = ti · v, (2.38)

where ti , i = 1, 2 are two orthogonal unit vectors tangent to the contact surface such that
n = t1 × t2 . For steady-state conditions it holds that ∂χ/∂t = 0, which gives the final
expression for the slip velocity
∂χ
γ̇i = Ωti · (n × (x − Xc )) + ωRti · − ti · v R . (2.39)
∂S
A non-slip or stick condition is obtained when (2.39) is equal to zero. In a steady-state
rolling situation no relative tangential motion indicates that locally the rotating object is
rolling and not sliding on the surface. Sliding of the object will occur if the tangential
forces exceed a certain limit.

2.5.2 Frictional stress for steady-state rolling

Stick

The stick condition for the classical Coulomb law, with constant friction coefficient µ (in
2D) is shown in Figure 2.8. This condition states that no relative motion occurs if

τ ≤ τcrit , (2.40)

where τcrit = µp, with p the (normal) contact pressure. This constraint can be handled
in ABAQUS in two different ways, using a Lagrange multiplier or approximating this no
relative motion constraint using a regularization step.
In case of the Lagrange multiplier method the constraint is obeyed exactly, however the
additional multipliers increase the cost of the analysis and convergence problems may
occur in areas where the contact conditions change (SIMULIA, 2009c).
When a regularization step is applied the non-differentiability of Coulomb’s law at zero
2.5 S TEADY-STATE TRANSPORT ANALYSIS 27

slip velocity is avoided. The regularization function is chosen such that the transition
from stick to slip is smooth. Some examples are shown in Figure 2.9. These functions
have the drawback that the transition from stick to slip is approximated and if the penalty
parameter of the regularization function is too large the model is not able to provide
realistic predictions of stick-slip motion. The great advantage of the differentiability is
however a simpler and more robust numerical algorithm, as shown by Wriggers (2006),
and this is also the default implementation in ABAQUS.

τcrit

γ̇

Figure 2.8 / Classic Coulomb’s friction law.

τcrit

γ̇

Figure 2.9 / Examples of regularization functions for Coulomb’s friction law.

Slip

If the frictional stress is larger than τcrit , sliding occurs and the frictional stress is limited
by

τ = τcrit . (2.41)

For the 3D situation, it is assumed that rubber friction is isotropic, such that the frictional
response is the same in both directions. Sliding occurs if the equivalent frictional stress
28 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

is larger than τcrit , where τeqv is defined as


q
τeqv = kτ k = τ12 + τ22 . (2.42)
Similarly the equivalent slip rate is defined as
q
γ̇eqv = kγ̇k = γ̇12 + γ̇22 . (2.43)
Furthermore for isotropic friction the direction of the slip and the frictional stress coin-
cide, therefore it holds that
τi γ̇i
= (2.44)
τeqv γ̇eqv
and the frictional stress during sliding in direction i is given by
γ̇i
τi = µp . (2.45)
γ̇eqv

2.6 Implementation of a friction law in steady-state rolling


For the implementation of friction laws other than the classical Coulomb model the sub-
routine FRIC of ABAQUS can be used to prescribe the frictional stresses. The main inputs
of this routine are the contact pressure and the slip velocity, given by (2.39). Besides the
frictional stress also the derivatives of the frictional stress with respect to the slip velocity
and the contact pressure should be provided. This is different compared to a static analy-
sis in ABAQUS, where the derivative of the frictional stress with respect to the slip distance
is required.
The friction law is implemented locally on each node j in contact, so (2.11) becomes
 −k 
kγ̇kj
  
pj 2 2
µj (pj , γ̇ j ) = µs + (µm − µs ) exp −h log , (2.46)
p0 Vmax
where pj and kγ̇kj are the normal contact pressure and equivalent slip velocity at node
j respectively. Furthermore, a threshold parameter plimit is used to prevent the pressure
term from going to infinity if the contact pressure approaches zero. If the nodal con-
tact pressure is lower than this threshold, the contact pressure pj is substituted with the
threshold value plimit . The subscript j is omitted in the remainder of the text for read-
ability, however all equations are evaluated for each node in contact.
To handle the non-differentiability at zero slip velocity a regularization step is applied. A
piecewise linear regularization function is chosen, with a user-defined critical slip velocity
γ̇crit as penalty parameter. An illustration of the regularization function is shown in
Figure 2.10. This parameter is used to calculate a slope ks , which is given by
µ(p, γ̇)p
ks = . (2.47)
γ̇crit
2.6 I MPLEMENTATION OF A FRICTION LAW IN STEADY-STATE ROLLING 29

Now the frictional stress in stick can be formulated as

τi = ks γ̇i , (2.48)

while the frictional stress for slip follows from (2.45)


γ̇i
τi = µ(p, γ̇)p . (2.49)
γ̇eqv

γ̇crit γ̇

−1

Figure 2.10 / Piecewise linear regularization function with user-defined critical slip ve-
locity γ̇crit .

In the subroutine additional outputs at every node, such as the friction coefficient µ, the
equivalent slip velocity γ̇eqv and the frictional power τeqv γ̇eqv are written to the ABAQUS
output database. For assembly of the global stiffness matrix the following matrix is also
required
 
∂∆τ1 ∂∆τ1 ∂∆τ1
∂∆γ̇1 ∂∆γ̇2 ∂∆p
J= , (2.50)
 
∂∆τ2 ∂∆τ2 ∂∆τ2
∂∆γ̇1 ∂∆γ̇2 ∂∆p

where the elements for both the stick and slip situation, using friction law (2.46), are
provided in Appendix A. A schematic overview of the in- and outputs of the subroutine is
shown in Figure 2.11.
τi
γ̇i
γ̇eqv
µ
p fric
τeqv γ̇eqv

Figure 2.11 / Interface of the subroutine FRIC.


30 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

2.7 Validation of the implemented friction law

2.7.1 Cylindrical model

To validate the implementation of the friction law several tests are performed with a sim-
ple cylindrical tire-like model. This model consists of a rubber cylinder, the relevant
parameters are shown in Table 2.1, which is first pressed against a flat rigid surface and
secondly a constant forward velocity is prescribed, while the rotational velocity is varied.
For this illustrative example, the rubber material model of the example tire of ABAQUS
(SIMULIA, 2009b) is used. The initial shear modulus of this model is given by 2C10 and
the parameter D1 governs the compressibility of the material, in this particular case the
rubber is modeled as fully incompressible.

Table 2.1 / Numerical values of the used parameters in the ABAQUS cylindrical test model.

Parameter Value Unit


Geometry Rinner 400 [mm]
Router 500 [mm]
Width 235 [mm]

Material model ρ 1 · 10−9 [tonnes/mm3 ]


Hyperelastic, C10 1.0 [N/mm2 ]
reduced polynomial D1 0.0 [N/mm2 ]

Load Fz −10000 [N]

Forward velocity Vx 13889 [mm/s]

Rotational velocity ω [1 50] [rad/s]

2.7.2 Comparison with standard Coulomb friction law

To make a comparison with the standard Coulomb friction model in ABAQUS two cylinders
are used, see also Figure 2.12, one where the default Coulomb model of ABAQUS is used
and one where (2.46) is applied in the contact interface. By choosing the parameters k
equal to zero and µs equal to µm , a constant friction coefficient is obtained. For both
models a constant friction coefficient of µ = 0.5 is used and γ̇crit is set to 50 mm/s.
The forward velocity is kept constant at 50 km/h and the rotational velocity is increased
from 1 to 50 rad/s. This guarantees that a braking, free-rolling and traction situation is
2.7 VALIDATION OF THE IMPLEMENTED FRICTION LAW 31

obtained, since ωf ree ≈ 27.5 rad/s. The computed frictional stresses of two nodes in the
contact area are shown in Figure 2.13 together with the error between these nodes and the
corresponding nodes on the other cylinder. It can be seen that the error is zero for both
nodes and this also holds for all other nodes, which shows that the standard available
Coulomb model can be reproduced using (2.46).

X Y

Figure 2.12 / Cylindrical test model used for comparison with the standard Coulomb
friction model of ABAQUS.
Frictional shear stress [MPa]

0.4
node 1
0.2
node 2
0

−0.2

−0.4
0 5 10 15 20 25 30 35 40 45 50
−6
x 10
1
Error [MPa]

0.5

−0.5

−1
0 5 10 15 20 25 30 35 40 45 50
ω [rad/s]

Figure 2.13 / Frictional shear stresses as a function of rotational velocity for two nodes on
the cylinder with friction model (2.46) and the error between the corresponding nodes
on the cylinder with the default Coulomb model of ABAQUS.
32 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

2.7.3 Validation of friction law for varying friction coefficient

To validate the implemented friction law for varying friction coefficients the same simula-
tion is done with only one cylindrical model. However now the parameters as presented
in Table 2.2 are used in the friction law. The effect of the threshold parameter plimit is
shown in Figure 2.14. The contact pressures and slip velocities of this simulation are

Table 2.2 / Numerical values of parameters for (2.46).

Parameter p0 k µ s µm h Vmax plimit


Value 0.5 0.5 0.2 1.1 0.5 695 0.2
Unit [MPa] [-] [-] [-] [-] [mm/s] [MPa]

logged and used to off-line compute µ(p, γ̇), which can then be compared with logged
values of µ. In Figure 2.15 the logged values for µ are shown for two contact nodes to-
gether with the error between these values and the recomputed values in MATLAB based
on the contact pressures and slip velocities of these nodes. These error signals confirm
that the implementation of the friction law is correct.

1.6

1.5

1.4

1.3

1.2
µp [−]

1.1

0.9

0.8

0.7
0 0.2 0.4 0.6 0.8 1
Pressure [MPa]

Figure 2.14 / Visualization of the effect of parameter plimit at 0.2 MPa, which limits the
pressure term in (2.46) to prevent numerical problems at low contact pressures.
2.8 S TEADY-STATE FREE - ROLLING 33

1.5

1
µ [−]

0.5 node 1
node 2
0
0 5 10 15 20 25 30 35 40 45 50
−5
x 10
1

0.5
Error [−]

−0.5

−1
0 5 10 15 20 25 30 35 40 45 50
ω [rad/s]

Figure 2.15 / Friction coefficient as function of the rotational velocity for two nodes on
the contact surface and the error between ABAQUS and the off-line recomputed µ, based
on logged contact pressure and slip velocity.

2.8 Steady-state free-rolling


The rotational velocity of a free-rolling tire is related to the forward velocity Vx as follows
Vx
ωf ree = , (2.51)
re
where re is the effective rolling radius defined at zero torque instead of zero longitudi-
nal force (Pacejka, 2006). The unloaded radius, ru , of the tire can be used to make an
estimation of the free-rolling rotational velocity. A braking or traction situation occurs if
the rotational velocity is not equal to the free-rolling rotational velocity at the same for-
ward velocity. Since the steady-state transport analysis requires that both the rotational
velocity and the forward velocity are prescribed, the free-rolling solution is not known a
priori. However the drive (or brake) torque T is defined as zero in case of free-rolling
and this can be used to iteratively find the rotational velocity, which corresponds to free-
rolling. While the forward velocity is kept constant during a steady-state rolling step, the
rotational velocity is incrementally updated based on previous values of the torque. The
update is based on the Newton-Raphson method
f (xn )
xn+1 = xn − . (2.52)
f 0 (xn )
In this case the goal is to find zero torque as function of ω, so (2.52) can be rewritten as
Tn ωn−1 − Tn−1 ωn
ωn+1 = , (2.53)
Tn − Tn−1
34 2 M ODELING FRAMEWORK FOR STEADY-STATE ROLLING TIRES WITH FRICTION

where the derivative is approximated using the finite difference method and n is an in-
crement in the steady-state rolling step.
For the implementation a bound on the maximum update for ω is used to prevent conver-
gence problems within an increment, which can occur if the difference in ω between two
increments is too large. This method works well if ω is initialized close to the free-rolling
rotational velocity. In Figure 2.16 a representative curve of the brake torque for increasing

Positive update ω
ωf ree

Figure 2.16 / Representative curve of the brake torque T as function of rotational velocity.
In the dashed region a fixed update for ω is used and in the vicinity of ωf ree the Newton-
Raphson method is applied.

ω is shown. It can be seen that (2.53) will not work if ω is initialized in the dashed re-
gion before the torque reaches its maximum. To overcome this, a positive update for ω is
made if the initial rotational velocity is far away from the free-rolling situation, e.g. a fully
locked wheel ω = 0. Once the solution is near the vicinity of free-rolling (2.53) is used
again. This combination makes it possible to compute the entire braking characteristic
of a tire.
The implementation of the algorithm to find the free-rolling steady-state solution is vali-
dated using the example tire of ABAQUS (SIMULIA, 2009b). In this example a 175 SR14
tire is used. A forward velocity of 10 km/h is prescribed and the obtained ωf ree of 9.027
rad/s corresponds very well with the 9.026 rad/s given in the documentation.

2.9 Conclusions
In this chapter the choice of the numerical modeling framework for steady-state rolling
tires with friction is motivated and described. Based on a literature overview to obtain
the handling characteristics of rolling tires, a choice is made for the steady-state transport
analysis of ABAQUS to compute the forces acting on the tire under different driving con-
ditions. This is a computationally efficient method to obtain the global steady-state force
2.9 CONCLUSIONS 35

and moment characteristics of a tire under different driving conditions. Additionally,


frictional effects, inertia effects, and history effects in the material can all be accounted
for. However, this method has also some limitations. The underlying surface must be
(rigid) flat, convex or concave, which means that it is not possible to model a rough road
surface. Therefore possible surface effects should be captured in the parameters of a fric-
tion model. Furthermore, the current implementation in ABAQUS is only suitable for the
mechanical domain, a fully coupled thermo-mechanical analysis can not be performed.
Most work found in literature still uses the Coulomb friction model for the tire-road in-
teraction problem, while it is clear from experiments with elastomers that rubber friction
depends on various parameters, such as sliding velocity, contact pressure, surface rough-
ness and temperature. Because of these dependencies the Coulomb friction model is not
sufficient to model the tire-road interaction. The goal of this thesis is to include some of
these dependencies in an enhanced friction model for FE tire simulations.
The choice of an improved friction model is based on the chosen numerical method to
obtain the global steady-state force and moment characteristics and an overview of fric-
tion models to describe the frictional response of rubber. The chosen friction model is
directly dependent on contact pressure and sliding velocity, while the global effect of tem-
perature and surface roughness is captured in the parameters of the friction model.
The mathematical formulation of the steady-state transport method is reviewed and the
implementation of the friction model in ABAQUS is derived. Practical issues related to the
numerical implementation are identified and discussed. Furthermore, the implementa-
tion of the friction model is validated using different test conditions. Finally, an algorithm
is presented to obtain the steady-state rotational velocity of a free-rolling tire, which is not
known in advance. This algorithm has been adapted such that it can be used to compute
the entire braking characteristic of a tire, even when the brake torque is not constant for
increasing rotational velocities.
36
C HAPTER THREE

Simulation procedure to compute


handling characteristics with an FE model

Abstract / In this chapter the strategy for identification of the parameters of the friction model is
explained. The parameters are identified sequentially using a two step experimental / numerical
approach, which is based on lab and full scale experiments. Furthermore details of the used test
tire and the corresponding FE model are provided. The different simulation steps required to
compute the three basic handling characteristics (pure braking, pure cornering and combined
slip) are discussed, while for each step the choice of boundary conditions is motivated.

3.1 Introduction
In this chapter the identification procedure to obtain the parameters of the friction model
is explained and the different simulation steps to compute the three basic handling char-
acteristics with an FE tire model are discussed.
The FE tire model has to fulfill several requirements. Information about the local friction
coefficient, slip velocity and frictional power is made available as output of the friction
model. The forces and moments acting on the axle should also be available as output to
make a comparison with experiments possible. Additionally several inputs, such as ge-
ometry and material properties, are required to construct the FE model. Different types
of simulations can be performed by prescribing specific boundary conditions on the FE
model.
Furthermore, the use of the steady-state transport algorithm introduces a constraint on
the underlying surface, which must be flat. In Figure 3.1 these requirements are incor-
porated into the different blocks, as shown in Figure 1.3, and the rough road surface is
replaced with a flat surface.

37
38 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

INTERACTION MODEL

FE Model
INPUTS OUTPUTS
Geometry 3D Force & Moment
Material data Footprint
Boundary conditions Friction coefficient
Flat road Frictional power
Slip velocity

Figure 3.1 / Schematic overview of the interaction model in the steady-state numerical
modeling framework.

This chapter focuses on the description and the effect of the boundary conditions applied
at each simulation step. Obviously, the inputs to the FE tire model, such as geometry
and material properties, determine the actual structural response of the tire. However,
the simulation steps remain the same, regardless of the type of tire investigated. There-
fore, the modeling approach can be discussed and evaluated without detailed information
about the specific tire considered. The exact description of the geometry and the material
properties of the tire under investigation is proprietary information of Vredestein and is
therefore not discussed in detail.
The chapter is organized as follows. In Section 3.2 the strategy to identify the parameters
of the friction model is discussed. The design of the test tire, which is used for experi-
mental validation, is presented in Section 3.3. Next, the different steps and corresponding
boundary conditions to create the 3D FE tire model, are given in Section 3.4. The simu-
lation procedure to compute the braking characteristic is described in Section 3.5. After
that, the simulation steps to obtain the cornering characteristic and the combination of
cornering and braking are explained in Section 3.6. A summary of the followed steps is
given in Section 3.7.
3.2 A PPROACH TO IDENTIFY PARAMETERS OF THE FRICTION MODEL 39

3.2 Approach to identify parameters of the friction model


Due to the choice of a phenomenological friction model several experiments need to be
carried out to gather enough data to extract the parameters of the model. Since the pro-
posed friction law (2.11) depends on a contact pressure related part and a slip velocity
related part, a two step experimental / numerical approach is developed and shown in
Figure 3.2. In this approach the parameters of the friction model are identified sequen-
tially using two different experimental setups.

LAT
Inputs
3D Forces Experiments

Lab scale
Friction
FEM 3D Forces Simulations
model LAT wheel

3D Forces Experiments
TNO Tyre Test Trailer
Full scale
FE Tire
3D Forces Simulations
Inputs
Virtual prototyping

Figure 3.2 / Schematic overview of the two step experimental / numerical approach to
obtain friction information using both lab and full scale experiments.

First, friction experiments are performed on a laboratory abrasion and skid tester (LAT) to
investigate the influence of contact pressure on the frictional force. These measurements,
which are conducted with very low rolling velocities and under free-rolling conditions
to minimize slip velocity, are used to identify the contact pressure dependent friction
parameters. A numerical model of this lab setup is used to validate the contact pressure
part of the friction model by comparing simulated forces with measured forces. This is
described in Chapter 4.
Secondly, braking experiments at different velocities with the test tire are conducted to
obtain slip velocity data for the friction model. The friction model is then coupled to the
FE model of the test tire and the remaining parameters are identified and validated by
comparing simulated axle forces with measured forces. This is the subject of Chapter 5.
Finally, the fully identified friction model is used to investigate the predictive capability
of the friction model by comparing simulated axle forces due to cornering and combined
slip with experiments. The discussion of these results is presented in Chapter 6.
40 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

In this research only dry friction is considered, which means that all experiments are
performed on dry surfaces and it is assumed that no external lubricants are present. The
presented two step experimental / numerical approach can also be applied under wet and
icy conditions, although performing reproducible experiments under these conditions
will be challenging.

3.3 Design of the test tire used for experimental validation


of the FE model
The passenger tire under investigation in this research has been specifically built for this
project, which allowed for some freedom in the design of the tread. With the current
implementation of the steady-state transport capability of ABAQUS it is nowadays possible
to include the full geometry of treaded tires. A limitation of this procedure is however
that solution accuracy can degrade if a treaded tire is used. Particularly the accuracy of
the longitudinal stress, when the angular extent of the so-called base pitch sector is too
large, is affected as stated by Qi et al. (2007). To avoid this undesirable numerical error
on the FE results, a tire with three longitudinal groves has been designed. This design
guarantees continuous streamlines along circumferential direction.
The longitudinal groves are placed slightly asymmetrically with respect to the center of
the tire to create a tire with a narrow and a wide shoulder. Due to this asymmetry two
different footprint sizes will result during steering to the left and right side. This allows
to investigate the influence of contact area and contact pressure on the generated forces
during steering both in experiments and in the FE model.
The radial tire, which has been manufactured by Vredestein, is shown in Figure 3.3. The
groves and the inner tread blocks have length L2, while the shoulders have length L1 and
L3 respectively.

Figure 3.3 / The test tire with three longitudinal groves, placed asymmetric with respect
to the center of the tire. The tire has been manufactured by Vredestein.
3.4 FE TIRE MODEL 41

3.4 FE tire model


The structural FE tire model is constructed in several steps, shown in Figure 3.4. First
a 2D cross section of the tire is created. This section is mounted on the rim and then
inflated to the desired inflation pressure. Secondly a 3D model is created by revolving
this 2D cross section around the axle. The 3D tire is loaded to make contact with the road
and finally a steady-state rolling step is conducted.
The structural response of the finite element model has been validated statically by Vre-
destein. This has been done using measurements of the contact pressure distribution and
contact area. Furthermore the radial and axial stiffness of the tire has been validated with
experiments on a flat plank tire tester, which have been conducted by Vermond (2008).

2D Axisymmetric 3D 3D
Rim mounting and inflation Footprint analysis Steady-state rolling

Figure 3.4 / Modeling steps for 3D steady-state rolling tire simulations.

3.4.1 2D tire cross section

The 2D cross section contains all the geometric and material information of the tire. A
radial tire consists of different rubber compounds and several reinforcement materials,
which are described in detail by Gent and Walter (2005) in chapter 1. These reinforce-
ment layers are the dominant load carrying parts of the cord-rubber composite and are
indicated in Figure 3.5 by the lines inside the contour of the cross section. The remain-
ing part of the cross section consists of the different compounds, for example the tread,
sidewalls and innerliner.
The rubber compounds are modeled as slightly compressible hyperelastic materials. Gen-
eralized 3- and 4-node linear, hybrid with constant pressure, axisymmetric elements with
twist are used to mesh the rubber compounds. These elements have three active degrees
of freedom, two displacements and one rotation, which corresponds to the twist angle.
Large deformations, which occur during steering, can be described more accurately with
quadratic elements, however for problems involving contact, linear elements are recom-
mended (SIMULIA, 2009c). The reinforcement belts and plies are modeled with so-
called rebar elements. These elements are used to define layers of uniaxial reinforcement
in membrane, shell or surface elements and can be used to model the cord properties of
the reinforcements under a specific angular orientation. The advantage of the rebar ele-
ments is that these layers are treated as a smeared layer with a constant thickness equal
42 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

to the area of each reinforcing bar divided by the reinforcing bar spacing, without the
need to model each cord separately (SIMULIA, 2009c, chapter 2.2). The rebar elements
used are generalized 2-node linear, with twist, axisymmetric surface elements. These el-
ements have the same three active degrees of freedom as mentioned above. The final
mesh is shown in Figure 3.61 , where it can be seen that the corners of the tread blocks
are rounded to reduce large gradients and improve convergence.

Figure 3.5 / Geometrical shape of the cross section, including reinforcement layers. The
aspect ratio is distorted for confidentiality reasons.

Figure 3.6 / Final mesh of the cross section, with the reinforcements layers as embedded
rebar elements.

1
Mesh layout provided by Vredestein.
3.4 FE TIRE MODEL 43

3.4.2 Rim mounting and inflation

The rim has to be modeled as well to transmit road induced forces to the axle. The
rim itself is modeled with two separate analytical rigid bodies as shown in Figure 3.7.
These two bodies are connected to the fixed world with two reference nodes, located on
the z-axis. Frictionless contact surfaces are defined on the rim and part of the tire to
establish contact between the tire and the rim. As first step the tire is mounted onto
the rim. This is done by translating the rigid bodies along the z-axis using displacement
boundary conditions, shown in Figure 3.8, such that the distance between the rim and
the tire complies with the standards specified by the European Tyre and Rim Technical
Organisation (2010). Finally the tire is inflated using a distributed surface load, which
is applied to the inner surface of the cross section. This pressure load is equal to the
desired inflation pressure and forces the tire into the final position, which can be seen in
Figure 3.9.

Figure 3.7 / Initial positions of the tire and the rim.

3.4.3 3D tire model

The 3D tire model is created by revolving the axisymmetric 2D cross section of the tire,
where a non-uniform discretization in circumferential direction is used to discretize the
3D model into seventy two sectors. The used symmetric model generation procedure
(SIMULIA, 2009c, chapter 10.4) converts the axisymmetric elements, used for the com-
pounds, into general 3D, 6- and 8-node linear, hybrid, elements. The axisymmetric rebar
elements are converted to 4-node quadrilateral, reduced integration elements. All ele-
44 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

Figure 3.8 / Mounting of the tire onto the rim, using displacement boundary conditions
on the rim.

Figure 3.9 / Final positions of the tire and the rim after inflation of the tire.

ment types have three active translational degrees of freedom.


An additional translational joint is added to connect the two separate rigid bodies together
and prevent relative rotation of the two rigid bodies. The joint is linked to a new reference
node, which represents the axle of the wheel. The reaction forces acting on this node will
be used to compare with measurements.
The final circumferential discretization is obtained by a number of mesh refinements
near the contact zone until the reaction forces acting on the wheel converge to a steady
3.4 FE TIRE MODEL 45

result. Each mesh refinement does however increase computation time, therefore the
possibility of using cylindrical elements has been investigated. These elements are avail-
able for precise modeling of regions in a structure with circular geometry and can span
large angles, which could reduce the number of used sectors significantly. However the
cylindrical elements in contact with the rim introduced convergence problems and are
therefore not used.
Furthermore a flat rigid body, which represents the road, is added. This surface is linked
to a rigid body reference node. The motion of the entire road can now be prescribed by
applying boundary conditions at this rigid body reference node, which in this case is con-
strained in all six degrees of freedom. The complete 3D model is shown in Figure 3.10.

Figure 3.10 / Non-uniform discretization of the FE tire including the rim and road sur-
face.

3.4.4 Mesh effect on lateral force in 3D model

When the 3D model is generated the stresses and strains of the 2D model are transferred
to the three-dimensional model and a static equilibrium step is performed. The element
formulations for the two-dimensional and three-dimensional elements are not identical
and as a result, there is a slight difference between the equilibrium solutions generated
by the two- and three-dimensional model (SIMULIA, 2009b, chapter 3.1).
In this equilibrium step the contact conditions between the rim and tire are changed
to prevent the tire from slipping along the rim in circumferential direction. When slip
occurs during the steady-state transport analysis, the solution obtained is no longer the
correct steady-state solution, because convective effects are ignored. It is assumed that
46 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

no convective effects are present between surfaces during steady-state transport analysis,
which means that two points on the contacting surfaces are still in the same contact
position after a rigid spinning motion. If however slip occurs, these points would not
be in the same position after a rigid spinning motion. This is not taken in account and
leads therefore to an incorrect solution. To ensure that no slip takes place the friction
properties are changed to ‘rough’ friction, which means that no slip is allowed. This is
used in combination with a so-called no separation condition in normal direction, which
prevents separation of two surfaces once contact has been established. Both constraints
are enforced by Lagrange multipliers.
Boundary conditions and loads are not transferred and are therefore redefined in the new
analysis to match the loads and boundary conditions, which were used in the 2D model,
e.g. the inflation pressure. The axle reference node is constrained, using boundary
conditions, for all six degrees of freedom in this equilibrium step.

When there is no contact with the road surface, the force on the axle in all three directions
after this static equilibrium step should be zero, since the reference point on the axle is
now the only connection to the fixed world. This is however not the case if the mesh of
the 2D cross section is not chosen carefully. If for instance the mesh, shown in Figure
3.11 is used, a lateral force of 14.8 N is obtained in the 3D model. This is due to the
asymmetrical mesh in combination with default contact and solver tolerances, which lead
to a small difference in the contact pressure distribution on both sides, when mounted
onto the rim. This in turn leads to a small force difference on the two rigid bodies. When
the 3D model is generated this small force error is amplified along the circumferential
direction and creates a significantly nonzero force in lateral direction.
This can be prevented by using the mesh as shown in Figure 3.6, which is fully symmetric
in the part that makes contact with the rim. The discretization error, which is still present,
is now equal at both sides and therefore cancels out. As a result zero lateral force in the
3D model is obtained after the equilibrium step.

Figure 3.11 / Close-up of a mesh, which leads to a nonzero lateral force in the 3D model.
3.5 S IMULATION PROCEDURE TO COMPUTE THE BRAKING CHARACTERISTIC OF A ROLLING TIRE 47

3.4.5 Static loading of the tire

After the equilibrium step the tire is loaded in two subsequent static analysis steps. The
first of these static steps establishes the initial contact between the road and the tire by
prescribing a vertical displacement on the reference node of the wheel. Since this is a
static analysis, it is recommended that contact is established with a prescribed displace-
ment, as opposed to a prescribed load, to avoid convergence difficulties that can arise due
to unbalanced forces (SIMULIA, 2009b). The prescribed boundary condition is removed
in the second static step, and replaced by a desired vertical load on the reference node of
the wheel.
Additionally, the contact between the tire and the road has been made frictionless, since
the frictional behavior in a steady-state transport step is different from the frictional mod-
els used in a static analysis as already discussed in Section 2.6. This can create disconti-
nuities between the solutions of a steady-state transport analysis and a static analysis. If
a zero coefficient of friction is used in all analysis steps prior to a steady-state analysis a
smooth transition is ensured (SIMULIA, 2009c, chapter 6.4).

3.5 Simulation procedure to compute the braking character-


istic of a rolling tire
Before an actual braking characteristic of a rolling tire can be computed using the method
presented in Section 2.8, friction between tire and road must be activated. Therefore an
additional steady-state step is introduced in which the friction model is activated. In
this step the friction coefficient is increased linearly over the step to prevent convergence
problems. To this end the local friction coefficient (2.46) is pre-multiplied with a param-
eter R (see also Appendix A), which varies linearly between zero and one during the first
steady-state step.
In this first steady-state transport step the tire is brought up to speed by setting the trans-
lational velocity of the tire to the desired forward velocity Vx , also referred to as longitu-
dinal velocity, while initializing the rotational velocity at 1 rad/s. This corresponds to an
almost locked wheel. During the last simulation step the rotational velocity of the tire is
increased every increment and the corresponding steady-state solution is computed until
the steady-state free-rolling solution is obtained. The complete simulation procedure is
summarized in Figure 3.12.
The obtained longitudinal force characteristic is referred to as the pure braking charac-
teristic in case no (global) lateral slip is present (Pacejka, 2006).
48 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

STATIC 2D model
inflation
and rim mounting

3D model Friction model

equilibrium step
2D-3D

Frictionless
axle displacement

axle force

Steady State Transport

ω=1 Enable friction

ω = ωf ree Friction

Free-rolling
algorithm

Figure 3.12 / Schematic overview of the simulation procedure to obtain the braking char-
acteristic of a rolling tire.

3.5.1 Mesh effect on the force equilibrium in vertical direction

When inertia effects are taken into account during a steady-state rolling step Coriolis and
centrifugal forces are generated. These forces are caused by deformation and spinning
of the tire and the centrifugal force will increase with increasing rotational velocity. How-
ever, when a non-uniform mesh in circumferential direction is used, a small numerically
induced force in the direction of the denser mesh, see Appendix B, is generated when in-
ertia effects are taken into account. This force is small with respect to the applied load on
the tire and as such does not affect the computed longitudinal and lateral forces. There-
fore, inertia effects are included, since the generated internal forces lead to increased
stresses in circumferential direction and stiffening of the tire for increasing rotational
velocity. Furthermore, the wheel is loaded (and not the road) in the FE model to keep the
normal force on the axle constant, because the load on the wheel is also kept constant
during experiments.
3.5 S IMULATION PROCEDURE TO COMPUTE THE BRAKING CHARACTERISTIC OF A ROLLING TIRE 49

3.5.2 Effect of penalty parameter in the friction model on the longitudi-


nal force

For the implementation of the friction model, a user-defined critical slip velocity γ̇crit
has been introduced in Section 2.6. Physically this penalty parameter can be interpreted
as the value of the sliding velocity below which sticking occurs. Simulation results are
more accurate for smaller values of the penalty, since the approximation of the transition
between stick and slip improves for smaller values as indicated in Figure 2.10. However
solving the contact conditions of the FE model can become more difficult and as a result
computation time increases. Hence using a larger penalty makes convergence of the
solution more rapid at the expense of solution accuracy. So a trade-off between accuracy
and computation time has to be made.
To investigate the influence of the value of the penalty parameter on the accuracy of the
solution, several simulations, using different values for γ̇crit have been performed. In
these simulations the braking characteristic of the tire is computed for a forward velocity
of 100 km/h and a load of 5000 N is applied on the axle. Furthermore a constant friction
coefficient of 0.8 is used.
The built-up of the longitudinal force around free-rolling is dominated by the longitudinal
slip stiffness of the tire. The value of the penalty parameter has however a direct influence
on the longitudinal force, since this force is generated during the transition from stick
to full slip. If a too large penalty value is chosen the gradient of the longitudinal force
around free-rolling decreases and an underestimation of the longitudinal slip stiffness is
observed in the numerical results.
The longitudinal force obtained with the default penalty value of ABAQUS is used as refer-
ence solution. The default value of ABAQUS is a function of the rotational velocity and the
radius of the rolling body, given by 0.01ωR, which provides an accurate solution (SIMU-
LIA, 2009c). This value is compared to the penalty value 0.04ωR used in the example
tire of ABAQUS (SIMULIA, 2009b), which solves a comparable problem, and a critical slip
velocity as function of the maximum applied sliding velocity. Typically, values between
1% to 5% give realistic results (van Breemen, 2009).
In the subroutine FRIC, where the penalty is used, the current rotational velocity of the
tire is not available. Therefore the critical slip velocity is specified directly as a percentage
of the forward velocity of the tire, which is equal to the maximal possible sliding veloc-
ity. Several simulations with percentages up to 5% of the maximum sliding velocity have
been performed to find the value that matches the reference solution.
In Figure 3.13 the longitudinal forces around free-rolling for three different penalty val-
ues are shown. It can be seen that the value taken from the example tire has a significant
influence on the longitudinal force, while the difference between the other two is negligi-
ble. The relative computation times are 1.0 for the default value, 0.88 for the value taken
from the example tire and 0.99 for the sliding velocity respectively. Although it is numer-
50 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

ically slightly more expensive than the value of the example tire, which compares to 4%
of the maximum sliding velocity, a more accurate representation of the longitudinal slip
stiffness of the tire is obtained. Therefore a penalty value of 1% of the maximum sliding
velocity, as shown in Figure 3.13, is used in further simulations.

0
Sliding velocity
−500 Example tire
Abaqus default
−1000
Longitunidal force [N]

−1500

−2000

−2500

−3000

−3500

−4000
70 75 80 85
ω [rad/s]

Figure 3.13 / Effect of penalty parameter in the friction model on the longitudinal force
around free-rolling at 100 km/h.

3.6 Simulation procedure to compute the cornering and


combined slip characteristics
Once the rotational velocity, which corresponds to the free-rolling situation, has been ob-
tained, for a given forward velocity, the cornering characteristic of a rolling tire can be
computed. This lateral force characteristic is referred to as the pure cornering character-
istic in case no (global) longitudinal slip is present (Pacejka, 2006).
The lateral force characteristic is a function of the side slip angle α, which is given by
 
Vsy
α = arctan , (3.1)
|Vx |

with Vsy the lateral slip speed, which is equal to the lateral velocity Vy on a flat road.
To induce a side slip angle on the tire in the FE model another steady-state rolling step
is performed. In this step the slip angle is gradually increased from zero degrees at the
beginning of the step to a desired slip angle at the end of the step. This can easily be
3.6 S IMULATION PROCEDURE TO COMPUTE THE CORNERING AND COMBINED SLIP CHARACTERISTICS 51

done by decomposing the velocity V into a lateral and longitudinal velocity component as
shown in Figure 3.14. The decomposition is a function of the slip angle and is given by

Vy = sin(α)V (3.2)
Vx = cos(α)V (3.3)

These two components are prescribed in the translational reference frame of the steady-
state step. The increment size within the steady-state step can be controlled with a limit
on the maximum allowable increment update. In this way a series of steady-state solu-
tions at each increment, which corresponds to slip angles between zero and the desired
slip angle, is obtained within one steady-state step.

Vx = V
V
x
Vx
α

Vy

Top view tire

Figure 3.14 / Introduction of a side slip angle α from straight line rolling (left side) by
decomposition of the translational velocity vector into a longitudinal and lateral velocity
component (right side).

When a nonzero side slip angle is present, the previously computed free-rolling rotational
velocity is not valid anymore. It follows from (3.3) that the longitudinal velocity reduces
for increasing slip angle. This indicates that the rotational velocity of the tire should also
decrease to prevent longitudinal slip.
It is possible to use the free-rolling algorithm to find the corresponding free-rolling so-
lution for every possible slip angle, but this is cumbersome to do. This would require
that after every increment several additional increments are necessary to update the ro-
tational velocity. Therefore it is chosen to approximate the free-rolling rotational velocity
for nonzero slip angles using

ωα = cos(α)ωf ree , (3.4)


52 3 S IMULATION PROCEDURE TO COMPUTE HANDLING CHARACTERISTICS WITH AN FE MODEL

which takes the effect of nonzero slip angles on the rotational velocity into account,
thereby reducing longitudinal slip.
Finally the combined slip characteristic can be computed, which is a combination of pure
longitudinal and pure lateral slip. For this last handling characteristic, the tire is rolling
freely at a nonzero slip angle and is then braked until wheel lock.
In the simulation model this can be achieved with yet another steady-state step, which is
done after the steady-state cornering step. In this step the translational velocities given by
(3.2) and (3.3) are kept constant, while the rotational velocity ωα is reduced to zero.

Table 3.1 / Settings of the rotational and translational velocities in the two reference
frames during a steady-state rolling step to compute the three basic steady-state handling
characteristics of a rolling tire.

Pure Braking Pure Cornering Combined slip


Rotational Translational Rotational Translational Rotational Translational
velocity velocity velocity velocity velocity velocity

Start SSR step Start SSR step Start SSR step


ω=1 Vx = V ω = ωf ree Vx = V ω = ωα Vx = cos(α)V
Vy = 0 Vy = 0 Vy = sin(α)V

End SSR step End SSR step End SSR step


ω = ωf ree Vx = V ω = ωα Vx = cos(α)V ω=0 Vx = cos(α)V
Vy = 0 Vy = sin(α)V Vy = sin(α)V

In Table 3.1 an overview of the settings in the two reference frames during the steady-state
rolling steps is presented, which are used to obtain the three basic handling characteris-
tics of a rolling tire.
It is of course also possible to start directly with a steady-state cornering (or a combined
slip) step after the static steps, as shown in figure 3.12, if the free-rolling rotational ve-
locity is already known. In that case one should initialize the rotational and translational
velocities to the values presented in Table 3.1 in the initial steady-state rolling step, where
the friction model is enabled.
3.7 CONCLUSIONS 53

3.7 Conclusions
In this chapter the simulation procedure to compute the steady-state handling character-
istics with a given FE tire model has been described. Additionally, the strategy for the
identification of the parameters of the friction model has been introduced. The parame-
ters will be identified sequentially using a two step experimental / numerical approach,
which is subject of the following chapters.
The design of the test tire, which is used for both experimental validation of the FE model
and identification of the friction model, has been presented together with a short descrip-
tion of the FE model. In order to develop an accurate tire-road interaction model a good
FE tire model is a necessity. The different components required to construct the FE tire
model contribute significantly to the overall accuracy of the computed steady-state han-
dling characteristics. The development of the FE tire model itself is however beyond the
scope of this thesis and a tire model provided by Vredestein is used. Therefore only the
effect of boundary conditions on the FE tire model has been discussed for each simula-
tion step.
Before an actual braking characteristic of a rolling tire, using a steady-state transport step,
can be computed, friction between the tire and the road must be activated. This has been
accomplished by the introduction of an additional steady-state transport step in which the
friction model is activated. In this step the friction coefficient is increased linearly over
the step to ensure a smooth transition between the frictionless and frictional situation.
Finally, the approach to compute the pure cornering and combined slip characteristics,
once the rotational velocity for free-rolling is known, has been been presented.
54
C HAPTER FOUR

Friction parameter identification using a


Laboratory Abrasion and skid Tester1

Abstract / In this chapter the influence of contact pressure on the frictional force between a small
rubber wheel and a rotating disk is investigated and parameters of the tire friction model are
identified. Friction data is obtained using a laboratory abrasion and skid tester. From this data
the parameters of the tire friction model are identified and implemented in an FE model for the
tire/disk contact. The different steps followed to build this FE model are also described. Sub-
sequently, results from simulations are compared with experiments for a variety of operating
conditions. The chapter is ended with a discussion of the results.

4.1 Introduction
Indoor tire testing can provide a good alternative for several outdoor tire tests. The
main advantage of indoor testing are the more controllable environmental conditions.
Information about the tire frictional behavior for model parameterization could be
acquired by extensively testing tires under different conditions, but these results are still
influenced by the structure of the tire itself. It is therefore advantageous to use small-
scale testing under controlled conditions. However, the experimental characterization
of frictional properties of rubber compounds is cumbersome due to the necessity of
complex measurement systems as shown by Huemer et al. (2001a); Gäbel and Kröger
(2006); Blume et al. (2003) and Garro et al. (1999).
The Laboratory Abrasion and skid Tester 100 (LAT 100), manufactured by VMI Holland
BV (2009), is one of the few commercially available experimental setups. This machine

1
Parts of this chapter have been presented in van der Steen et al. (2010b).

55
56 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

was specially developed for compound testing in the tire industry. It can measure the
skid, traction and wear of small rolling tires, diameter 80 mm, with a special disk,
simulating road conditions. In this way the complexity of a complete tire is reduced to a
solid tire made out of one compound.

In this chapter the influence of contact pressure on the frictional force between a small
rubber wheel and a disk is investigated and parameters of the tire friction model are
identified. Several different loading and velocity conditions are used in the LAT 100 to
obtain the frictional and stiffness characteristics of one compound. An FE model of the
LAT 100 is constructed in ABAQUS, where the framework derived in Chapter 2 is used to
efficiently compute steady-state free-rolling solutions. The obtained numerical results
are compared with measured data for a variety of operating conditions.

This chapter is organized as follows. In Section 4.2 the followed strategy, using lab scale
experiments, is presented. In Section 4.3, a description of the experimental setup is given.
Next, the different steps in the modeling approach are motivated in Section 4.4. The
identification of the velocity independent parameter set of the friction model is presented
in Section 4.5. After that, the results of both the model and the experiments are shown
in Section 4.6, while in Section 4.7 a discussion of the results is presented. Finally,
conclusions are drawn in Section 4.8.

4.2 Identification using lab scale experiments


The chosen friction model (Section 2.4.2) can be decomposed into the product of a contact
pressure dependent part and slip velocity dependent part as

µ(p, v) = µp (p)µv (v). (4.1)

This chapter focuses on the identification of the contact pressure part, which is given by
 −k
p
µp (p) = . (4.2)
p0

An overview of the modeling approach using the LAT 100 setup is shown in Figure 4.1.
Several measurements, under very low rolling velocities and free-rolling conditions, are
used as input for the friction model. Once the parameters of (4.2) are identified the
friction model is coupled to an FE model of this small tire and the FE model is validated
by comparing the computed and experimentally found hub forces. In the next chapter
the friction model is extended with the slip velocity term using measurements of the test
tire.
4.3 E XPERIMENTAL SETUP 57

LAT
Inputs
3D Forces Experiments

Lab scale
Friction
FEM 3D Forces Simulations
model LAT wheel

3D Forces Experiments
TNO Tyre Test Trailer
Full scale
FE Tire
3D Forces Simulations
Inputs
Virtual prototyping

Figure 4.1 / Schematic overview of the modeling approach to obtain friction information
from lab scale experiments.

4.3 Experimental setup


In the experimental setup, shown in Figure 4.2, a small solid tire is pressed on a driven
abrasive corundum disk. The normal force, Fz , on the tire can be preset between 40
and 140 N and a servo-controlled load cylinder is used to keep the normal force constant
during measurements. The disk can only rotate in a clockwise direction and the velocity
can be set between 0.002 and 100 km/h. It is possible to place the wheel under different
side slip angles, α, which is also shown in Figure 4.2, by moving the complete module,
with the force sensor and the wheel, over a guidance rail. The sample tire, which is made
from the tread rubber compound, is solid and without a profile. Furthermore the wheel
is fixed in all directions except for the rolling direction. The fixation of the tire is force
based using two metal plates, which prevent the tire from slipping between the plates.
Besides these two metal clamping plates, a spacer ring is inserted into the center hole of
the wheel, before the wheel is placed onto the measurement hub.
The friction present between the abrasive disk and small tire drives the tire and the result-
ing forces are measured with a triaxial force sensor. Besides abrasion experiments also
the slip characteristics, such as pure cornering, of a compound on different surfaces can
be measured at the lower speed range. For small side slip angles the lateral force, Fy , is
dominated by the dynamic stiffness of the compound, while for large side slip angles the
frictional force dominates (VMI Holland BV, 2009). An extensive overview of the setup
and the different measurement capabilities can be found in the work of Broeze (2009).
58 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

Triaxial force sensor

W Fx

Fy
a

Side slip angle


Abrasion disk Fz

(a) (b)

Figure 4.2 / a) The Laboratory Abrasion and skid Tester 100. b) Schematic top and side
view of the setup.

4.4 FE model of the setup


In this section the FE model, including the abrasion disk, is described. The derivation of
the used material model is discussed first. After that the structural model is described;
the modeling procedure is simplified by exploiting the axisymmetric properties of the
solid tire. A 2D cross section of the wheel is constructed first and later used to generate a
full 3D model by revolving this cross section.

4.4.1 Material model

One of the components of every FE model is a description of the materials under study.
To characterize a specific material in ABAQUS several standard material models are avail-
able, the parameters of these models can be specified directly or fitted on experimental
data. The rubber compounds in tires can be defined as the formulation of a mixture of
rubber and additives, which meets the needs of the tire component application. Usually
a compound consists of one or more polymers, vulcanizing agents, accelerators, rein-
forcing fillers, antidegradents, plasticizers, softeners and tackifiers. All these additives in
the compound contribute to a more complex stress-strain response compared to a pure
elastomer.
4.4 FE MODEL OF THE SETUP 59

Experimental data

The compound of the used sample tire has been tested at the Deutsches Institut für
Kautschuktechnologie e.V. to characterize the material response. The result of one of
the quasi-static uniaxial compression and tension experiments is shown in Figure 4.3.
In this experiment a dumbbell shaped sample is compressed and stretched to −30 and
+50% strain respectively in three consecutive cycles. Two softening effects occur during
this experiment, the so-called Payne effect (Payne, 1963) and Mullins effect (Mullins and
Tobin, 1957). The Payne effect is a softening effect for small strains (0.1%), attributed to
breaking apart aggregates of filler particles. The Mullins effect is a substantial softening
effect at higher strains, attributed to progressive detachment of rubber molecules from
filler particles. Both these softening effects can be seen in Figure 4.3. Furthermore it
can be seen that for the third curve the stress converges towards the stress of the second
cycle, which indicates that three cycles are sufficient to induce damage in the test sample.
Additionally, hysteresis is present in all of the cycles.
Normalized engineering stress [−]

1
cycle 1
0.5 cycle 2
cycle 3
0

−0.5

−1

−1.5

−2
−30 −20 −10 0 10 20 30 40 50
Strain percentage [%]

Figure 4.3 / Normalized engineering stress-strain response of a quasi-static uniaxial com-


pression and tension test.

Choice of material models

Only the available material models in ABAQUS are considered, since developing new nu-
merical material models is beyond the scope of this research. Therefore several assump-
tions are made to choose a material model. It is assumed that:

• The expected strains on the sample tire range from −10% to 20%.
60 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

• After several rotations of the tire, the material is in a converged damaged cycle and
Payne and Mullins effects can be ignored.

• The material is nearly incompressible.

• Hysteresis is also present in the rolling tire.

Following the guidelines of the ABAQUS manual (SIMULIA, 2009a) a hyperelastic model
is chosen for the long-term elastic response and a viscoelastic model is chosen for the
rate-dependent material behavior. For the determination of hyperelastic material param-
eters a fit procedure is available. Experimental test data can be provided directly and the
appropriate values of the coefficients for all available hyperelastic models are determined,
after which the best model can be selected.
To provide stress-strain data for the determination of the long-term coefficients it is as-
sumed that the long-term hyperelastic response can be approximated with the averaged
stress of the loading and unloading stresses of the third cycle. Furthermore emphasis
is placed on the strains ranging from −10% to +20% by providing more data points in
this region. The fit error between all strain energy potentials and the provided data is
the smallest for the Van der Waals strain energy potential (SIMULIA, 2009a, chapter
18.5) and is therefore chosen. The provided data points and the Van der Waals model are
shown in Figure 4.4.
Normalized engineering stress [−]

1
Data points
0.5 Van der Waals hyperelastic model

−0.5

−1

−1.5
−30 −20 −10 0 10 20 30 40 50
Strain percentage [%]

Figure 4.4 / Approximated long-term normalized stress-strain response data points for
the fitting procedure and the evaluated Van der Waals strain energy potential.

The hysteresis is included by adding a linear viscoelastic part to the long-term hyperelastic
part. The implementation in ABAQUS is done by means of a Prony series (SIMULIA,
2009a, chapter 18.7). This series is fitted on experimental frequency data of the shear
4.4 FE MODEL OF THE SETUP 61

storage, G0 , and loss modulus, G00 , measured in-house by Vredestein2 .

Validation of the material models

As a validation step for the material models the dumbbell experiment is reproduced with
an FE model of the dumbbell. In the numerical model one clamp is fixed and on the other
clamp a time dependent force is prescribed. This force is calculated using the nominal
stress σe and the undeformed cross sectional area A0 ,

F = A0 σe . (4.3)

The nominal strain


l−L
εe = , (4.4)
L
in the experiment is obtained by tracking the current length l of the dumbbell bar and the
initial length L. In the simulation model the current distance between two nodes on the
dumbbell surface and their initial distance are used to compare with the measured nomi-
nal strain, shown in Figure 4.5. The axisymmetric section is modeled with linear, hybrid,
4-node, axisymmetric elements3 . In Figure 4.6 the calculated (normalized) stress-strain
response is compared to the experimental response. It can be seen that the response of
the model is slightly softer than the actual material, since the dumbbell in the numerical
model is compressed and stretched slightly further than in the experiment. Furthermore
there is a small deviation in the hysteresis curve, which indicates that the energy dissipa-
tion in the experiment is larger. However the overall shape of the stress-strain response is
captured qualitatively well considering the made assumptions and the available material
models.

Figure 4.5 / Dumbbell axisymmetric cross section and location of nodes for determina-
tion of the nominal strain.

2
These parameters are company confidential.
3
Mesh layout provided by Vredestein.
62 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER
Normalized engineering stress [−]

1
Experiment
0.5 Simulation

−0.5

−1

−1.5

−2
−30 −20 −10 0 10 20 30 40 50
Strain percentage [%]

Figure 4.6 / Comparison between measured and calculated stress-strain response with
the hyperelastic and viscoelastic material models.

4.4.2 2D model

The 2D axisymmetric model in ABAQUS is defined in a cylindrical coordinate system,


where the 2D cross section is placed at a reference plane θ = 0◦ . A close look at Table
4.1 shows that the outer radius of the spacer ring is slightly larger than the inner radius
of the tire. This, together with the metal plates, already prestresses the tire. The ax-
ial deformation of the wheel due to clamping of the side disks has been experimentally
determined as 0.6 mm. Furthermore the tire has been designed with rounded corners,
which have a radius of 1 mm. This small curvature improved the overall reproducibility
of measurements on the setup as shown by Broeze (2009) and it also ensures that the
outward normal on the surface is continuous in the numerical model.

Table 4.1 / Geometric data of the parts for the 2D models.

Part Inner radius [mm] Outer radius [mm] Height [mm]


Tire 16.5 39 18
Ring 15 17.5 15
Side disk 8 30 4

A numerical efficient approach is to exclude all the details of the fixation and only model
the cross section of the tire. This is only valid if no slip occurs as contact between tire
and metal plates is established. In that case all resultant forces due to contact with the
abrasion disk are transferred to the measurement hub. Slip has not been observed in any
of the experiments, so it is assumed that there is no slip between the tire and the fixation
4.4 FE MODEL OF THE SETUP 63

parts. The no slip constraint can be achieved by first using displacement boundary con-
ditions to prestress the wheel and afterwards replacing these boundary conditions with a
kinematic coupling constraint for all degrees of freedom to a reference node. This cou-
pling constraint is chosen such that the part of the tire that is normally in contact with
the ring and side disks is now fully constrained and the no slip condition is obeyed.
To show the validity of this approach the stress distribution is compared to a model with
the three fixation components included. In this model the spacer ring and side disks are
modeled as rigid bodies, since these are much stiffer than the rubber compound. The
tire section, for both models, consists of linear, hybrid, 4-node, axisymmetric elements.
All parts are positioned such that there is an initial penetration of the spacer ring and the
side disks in the rubber tire. A special feature of ABAQUS (SIMULIA, 2009a) is available
to solve these ‘overclosures’ before other simulation steps are carried out. Once the over-
closure has been solved, the friction properties are changed such that slip is prevented
and the two models can be compared with each other.
Figure 4.7 shows the 2D mesh layout and the final stress distribution of the two mod-
els. It can be seen that the simplified model is a good approximation, since the stress
distribution outside the fixation area is equal. Therefore the simplified model is chosen,
because it saves computation time in the 3D model, since all 3D contact of the fixation is
avoided.

Viewport: 1 ODB: /home/rsteen/tmp/Groschwheel_572elem.odb

Y
Z X

Viewport: 2 ODB: /home/rsteen/tmp/Groschwh...erm_kinematic_restart.odb


S, Mises
Viewport: 2 ODB: /home/rsteen/tmp/Groschwheel_572elem.odb (Avg: 75%)
+2.700e−01
+2.529e−01
+2.358e−01
+2.187e−01
+2.017e−01
+1.846e−01
+1.675e−01
+1.504e−01
+1.333e−01
+1.162e−01
+9.917e−02
+8.208e−02
+6.500e−02

Y Y
Y

ODB: Groschwheel_572elem.odb Abaqus/Standard Version 6.8−2 Mon Jun 08 11:59:04 CEST 2009 Z X Z X ODB: Groschwheel_VDW_longterm_kinematic_restart.odb Abaqus/Standard Version 6.8−2 Mon Jun 08 11:59:52 CEST 2009
Z X Y
Y Step: Step−1
Step: Shrinkfitstep, Shrinkfit to push wheel inside the ring and side−disks
Z X Increment 6: Step Time = 1.000 Z X Increment 1: Step Time = 1.000
Primary Var: S, Mises
Deformed Var: U Deformation Scale Factor: +1.000e+00

(a) (b) (c)

Figure 4.7 / a) The mesh layout of the simplified model and difference in stress distribu-
tions of the b) simplified model and c) detailed model.

4.4.3 3D model

A non-uniform discretization in circumferential direction is used to discretize the 3D


model. Seventy sectors of general 3D, 8-node linear, hybrid, brick elements are used,
each node having three active translational degrees of freedom. The sector angle differs
from 0.75◦ in the contact area up to 10.5◦ at the top of the tire, shown in Figure 4.8a. The
64 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

computed stresses and strains of the 2D model are transferred to all these sectors and
equilibrium for this situation is calculated.
The abrasion disk of the experimental setup is included as a flat analytical rigid body.
The origin of the global coordinate system is set at the center of the wheel, such that the
X-axis corresponds to the longitudinal direction and the Y -axis to the lateral direction on
the tire. The center of the abrasion disk as a function of the side slip angle can now be
calculated as

xdisk = − sin(α)Rdisk , (4.5)


ydisk = cos(α)Rdisk , (4.6)

where Rdisk is the constant (2D) distance of 165 mm between the center of the disk and
the center of the tire, which is shown in Figure 4.8b. The coordinates of the disk are used
to prescribe the rotational velocity of the disk.

V
α
X
Rdisk Y

Abrasion disk

(a) Side view tire model. (b) Top view entire model.

Figure 4.8 / a) Non-uniform discretization of the small tire. b) Position of the origin of
the coordinate system.

Contact between abrasion disk and tire

As a first validation step the size of the contact patch is compared with ink measurements
under a static load. For this purpose the abrasion disk on the setup is replaced with a flat
and smooth glass disk on which paper is taped. For the numerical model a time period
of one second is used to apply the load, the results are shown in Figure 4.9. The relative
error of the contact area is determined by
s P
n
i=1 (AModi − AExp i )
2

errorrel = Pn 2
, (4.7)
i=1 (AExp i )
4.4 FE MODEL OF THE SETUP 65

where AExp and AMod are the experimentally obtained and computed contact areas respec-
tively, and n is equal to the number of experiments. The relative error in this case is 6.1%,
from which it can be concluded that there is a good correlation with the measurements.

260
Experiment
240 Model

220
Contact area [mm2]

200

180

160

140

120
40 60 80 100 120
Fz [N]

Figure 4.9 / Measured and simulated area of the contact patch under static load.

Steady-state free-rolling solution

After the equilibrium and the initial loading step a steady-state solution for the rolling
situation has to be found. Only the rotational velocities of the disk and tire need to be
specified to compute a steady-state solution. On the experimental setup it is not possible
to accelerate or decelerate the tire. The tire is only able to rotate freely, which means that
the moment My (or driving torque) is zero. Furthermore, a constant velocity profile of
the abrasion disk has to be applied on the setup, which corresponds to the velocity at the
center contact point of the tire. The rotational velocity of the disk is straightforwardly
calculated by

Vdisk
Ωdisk = , (4.8)
Rdisk
where Vdisk is the prescribed forward velocity. The rotational velocity of the tire which
results in free-rolling is not known a priori. Therefore an initial rotational velocity is
given and the corresponding steady-state solution is computed. Based on the value of the
driving torque the rotational velocity of the tire is adjusted, by means of the algorithm
described in Section 2.8, and a new steady-state solution is computed until the torque is
66 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

within a specified tolerance of 1 Nmm. The initial rotational velocity can be estimated us-
ing the fact that at free-rolling the velocity of the wheel approaches the prescribed forward
velocity of the disk.

Vdisk cos(α)
ωwheel = , (4.9)
r
where r is the radius of the tire and cos(α) takes the orientation of the wheel into account.

4.5 Parameter identification


To acquire frictional information, experiments on rolling tires with large side slip angles
of ±40◦ are conducted under different normal forces at 0.1 km/h. At this low velocity
it is assumed that the sliding velocity does not contribute significantly to the friction.
Furthermore wear at lower velocities is greatly reduced by rolling measurements instead
of measurements with blocked wheels. For these large side slip angles the limit of the
maximal achievable lateral force is reached. If the lateral force is divided by the normal
force, a decreasing global friction coefficient for increasing load is found. This agrees with
results found in literature, see e.g. Dorsch et al. (2002); Huemer et al. (2001a); Blume
et al. (2003) and Lindner (2005). The friction coefficient is approximated by regression
through the data points, using the model

µ(Fz ) = p1 Fzp2 , (4.10)

where Fz is the applied normal force and p1 and p2 are empirically determined coeffi-
cients. The obtained coefficient p2 of −0.254 in this power law is in good agreement with
−0.33, which is found for the load dependence of rubber friction based on hemispheri-
cal surface asperities with elastic Hertzian contact as shown by Schallamach (1952). The
measurements and the approximation using (4.10) are shown in Figure 4.10.

For the implementation of (4.2) the friction coefficient must be expressed as a function
of the contact pressure. This can be accomplished by substitution of the applied load by
the average contact pressure, as presented by Dorsch et al. (2002). The measured contact
area, as shown in Figure 4.9, is approximated with a linear least square error fit, and
dividing the applied load by the measured contact area the average contact pressure is
calculated. As expected, it follows that the average contact pressure rises for increasing
load, see Figure 4.11.
The calculated average contact pressure is substituted for the applied load in the experi-
ments to obtain a pressure dependent relation for the measurements. Now the parame-
ters p0 and k can be identified. The evaluation of (4.2) together with the experiments can
be seen in Figure 4.12.
4.6 COMPARISON OF NUMERICAL AND EXPERIMENTAL RESULTS 67

1.5
Experiments
1.4 p Fp2
1 z

1.3

1.2
µp [−]

1.1

0.9

0.8
40 50 60 70 80 90 100 110 120 130
Fz [N]

Figure 4.10 / Measured friction coefficient as function of normal force at 0.1 km/h and
the approximation using equation (4.10).

0.6

0.55

0.5
Pressure [MPa]

0.45

0.4

0.35

0.3

0.25

0.2
40 60 80 100 120
Fz [N]

Figure 4.11 / Average contact pressure as function of normal force based on the mea-
sured contact area.

4.6 Comparison of numerical and experimental results


Simulations, using (4.2) to calculate the friction coefficient for the tire/disk contact,
are carried out for a forward velocity of the disk of 0.1 km/h. The side slip angle is
varied from −40◦ to +40◦ and a load of 40, 75 or 120 N is applied. Furthermore, four
measurements are made for each side slip angle and load.
68 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

1.5
Experiments
1.4 (p/p )−k
0

1.3

1.2
µp [−]

1.1

0.9

0.8
0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65
Pressure [MPa]

Figure 4.12 / Measured friction coefficient as function of average contact pressure and
the approximation using equation (4.2).

In Figure 4.13 the mean longitudinal forces, and the highest and lowest measured value,
of four measurements are shown together with the results of the numerical model for
the entire range of side slip angles. It can be seen that for zero degree slip angle the
computed longitudinal forces are nonzero due to the viscoelastic material model. If
viscoelastic material is included, the point where the resultant normal force acts, moves
towards the leading edge. This creates an additional torque, which results in a rolling
resistance force at zero degree slip angle in the free-rolling situation (Gent and Walter,
2005). If these numerical results are compared with measurements, it can be seen that
there is a good qualitative match. For both the model and the experiments the rolling
resistance force increases for higher loads. Furthermore it can be seen that overall
the numerical model matches the experiments qualitatively and the non-symmetrical
triangular shape around the zero degree slip angle is correctly predicted. It should be
noted that the force sensor is calibrated in the range −5 to −100 N. Therefore care
should be taken when interpreting the actual values of these measurements, since these
are based on extrapolation of the sensor calibration data.
The lateral forces are shown in Figure 4.14, where it can be seen that there is a good
qualitative match for all loads and side slip angles. In the region |α| > 10◦ the lateral
force is dominated by the generated frictional force, which achieves its maximum around
|α| ∼
= 20◦ . Quantitatively, the model predicts however a lower lateral force than observed
in experiments for all normal loads. This is probably due to discrepancies in the size of
the contact area, as will be discussed further in Section 4.7.
4.6 COMPARISON OF NUMERICAL AND EXPERIMENTAL RESULTS 69

In the region for |α| < 5◦ the change in lateral force is governed by the cornering stiffness


∂Fy
CF α = , (4.11)
∂α α=0

which can be approximated by a linear least square error fit through data points for Fy at
−5◦ , 0◦ and 5◦ . The results of this approximation are shown in Figure 4.15, where it can
be seen that the cornering stiffness increases for increasing load. Although quantitatively
the cornering stiffness of the model is higher than observed in the experiments, the
increase of the cornering stiffness with increasing load is correctly predicted by the
model.

The nonzero lateral forces at zero degree side slip angle are due to turn slip (Pacejka,
2006), which occurs due to the path curvature. Turn slip is present since the radius of
the abrasion disk is too small to consider it as driving on a straight road. This effect is
well captured in the model and can be shown when the slip velocity field is visualized.
In Figure 4.16 the slip velocity field for the free-rolling tire at zero degree side slip angle
is shown, with an observer placed on the wheel for the direction of the slip vectors. The
x-y nodal position is given in the reference ISO system, the derivation of the slip velocity
field is given in Appendix C. As expected at free-rolling slip develops at the trailing edge of
the contact patch. The orientation of the slip vectors clearly indicates driving on a curved
path. The rotational and translational velocities of the wheel are constant. The rotational
velocity of the disk is also constant, but the tangential velocity of the disk increases with
the radius of the disk. As a result the direction of the slip vectors changes over the width
of the tire.
70 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

−0.2

−0.4
Fx [N]

−0.6

−0.8 40 [N] Experiment


40 [N] Model
−1
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(a)

−0.5
Fx [N]

−1

−1.5
75 [N] Experiment
75 [N] Model
−2
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(b)

−1
Fx [N]

−2

−3
120 [N] Experiment
120 [N] Model
−4
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(c)

Figure 4.13 / Longitudinal force versus side slip angle for experiments and numerical
model at 0.1 km/h under a) 40 N, b) 75 N and c) 120 N normal load. The bar around the
experimental data points indicates the highest and lowest measured value.
4.6 COMPARISON OF NUMERICAL AND EXPERIMENTAL RESULTS 71

50
40 [N] Experiment
40 [N] Model
Fy [N]

−50
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(a)

100
75 [N] Experiment
75 [N] Model
50
Fy [N]

−50

−100
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(b)

150
120 [N] Experiment
100 120 [N] Model

50
Fy [N]

−50

−100

−150
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

(c)

Figure 4.14 / Lateral force versus side slip angle for experiments and numerical model
at 0.1 km/h under a) 40 N, b) 75 N and c) 120 N normal load. The bar around the
experimental data points indicates the highest and lowest measured value.
72 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

9
Experiment
Cornering stiffness [N/deg]

8
Model
7

2
30 40 50 60 70 80 90 100 110 120 130
Fz [N]

Figure 4.15 / Cornering stiffness at different loads.

0
x

−2

−4

−6

−8

8 6 4 2 0 −2 −4 −6 −8
y

Figure 4.16 / Slip velocity field for the free-rolling tire at α = 0◦ and Fz = 120 N.

4.7 Discussion
The presented model is used to investigate the mismatch with the observed overall lower
cornering stiffness of the experiments. A sensitivity analysis with respect to side slip an-
gle, geometry and velocity is performed for a load of 40 N around zero degree side slip
angle. The results of this analysis are presented in Table 4.2. It can be seen that the side
slip angle has a large influence, while variations of the radius of the disk have little effect
4.7 D ISCUSSION 73

Table 4.2 / Lateral force for varying parameters.

α◦ Rdisk [mm] velocity [km/h]


−1 0 1 148.5 165 181.5 0.09 0.10 1.10
Fy [N] −1.52 −6.00 −11.62 −6.29 −6.00 −5.55 −5.68 −6.00 −6.04

on the lateral force. Furthermore, a 10% velocity difference has also limited influence.
This again confirms that the lateral force at small slip angles is dominated by the corner-
ing stiffness. As shown in Table 4.2, a misalignment of 1◦ in the test setup would cause
an error of 5 N in Fy . After close examination of the setup a misalignment of about −0.5◦
of the side slip angle has been discovered, which partly explains the discrepancy between
experiments and simulations at α = 0◦ . However, this difference is not enough to ex-
plain the lower cornering stiffness and the presented experimental results are therefore
not corrected for this misalignment.
Another discrepancy could be due to uncertainties and inaccuracies in the material
model. The overall lower cornering stiffness found in the experiments indicates that
the material behavior is too stiff. After several simulations it appeared that the material
response in the ALE steps is stiffer than in quasi-static steps. If for instance the foot-
print size at zero degree side slip angle for a rolling wheel at a load of 120 N is compared
with the contact area shown in Figure 4.9, a reduction of twelve percent is found. If
the viscoelastic model is turned off and only the hyperelastic model is used, there is no
significant difference between the rolling and non rolling footprint size. This might be
due to discrepancies between the viscoelastic material model and the actual viscoelastic
behavior of the rubber wheel.
The difference in contact area also causes the lower computed lateral forces in the region
|α| > 20◦ in Figure 4.14. If the reduction of the contact areas is calculated between zero
and forty degree side slip angle, which is shown in Table 4.3, it follows that there is a large
difference between zero and forty degree. This means that the average contact pressure
at ±40 degree side slip angle is much higher than shown in Figure 4.12.

Table 4.3 / Contact areas for zero and forty degree side slip angle and the corresponding
reduction for viscoelastic and only hyperelastic material model.

Material α = 0◦ [mm2 ] α = 40◦ [mm2 ] Reduction [%]


Viscoelastic 202 155 23
Hyperelastic 291 186 36
74 4 F RICTION PARAMETER IDENTIFICATION USING A L ABORATORY A BRASION AND SKID T ESTER

As presented in Table 4.3, the reduction in the contact area at ±40 degree side slip angle
varies between 23% and 36%. Therefore, an average reduction of 30% with respect to
zero degree side slip angle is assumed and a correction of 30% is applied to the measured
contact area at zero degree side slip angle. If this corrected smaller contact area is used
for substitution of the applied load, the average contact pressure is increased as shown in
Figure 4.17. It can be seen that the coefficient p0 in (4.2) increases, while the exponent k
remains the same.
In Figure 4.18 the lateral force for a load of 120 N is shown when this correction is applied
and only the hyperelastic material model is used. Now the lateral forces are much closer
to the experimental values, but there is still a difference with the experiments. This could
be due the sensitivity of the force sensor to moments, which introduces an additional
offset in the lateral force. This in turn could explain the substantial difference between
the absolute values of the measured lateral forces at 40◦ and −40◦ as observed in Figures
4.14 and 4.18. These nonequal limits are not expected for an isotropic compound and
the relatively low velocity used, since the lateral force reaches a maximum if the friction
limit is reached. This has also been confirmed by Grosch (2009), who developed the
experimental setup. Therefore the absolute values of the lateral forces should be equal
and constant for |α| ≥ 30◦ , which is the case for the numerical model and therefore show
the validity of the applied modeling approach.

1.5
Experiments
1.4 (p/p )−k
0

1.3

1.2
µp [−]

1.1

0.9

0.8
0.4 0.5 0.6 0.7 0.8
Pressure [MPa]

Figure 4.17 / Measured friction coefficient as function of corrected average contact pres-
sure and the approximation using equation (4.2).

From an experimental point of view it would be interesting to further investigate the


influence of different surface roughness profiles of the abrasion disk and temperature
4.8 CONCLUSIONS 75

150
120 [N] Experiment
100 120 [N] Corrected model

50
Fy [N]

−50

−100

−150
−50 −40 −30 −20 −10 0 10 20 30 40 50
α [°]

Figure 4.18 / Lateral force versus side slip angle for experiments and corrected numerical
model at 0.1 km/h under 120 N normal load.

effects of the rubber material, which become important when higher velocities than 0.1
km/h are applied. However as already indicated by Broeze (2009) the influence of the
mechanical design on the results should be clarified first.

4.8 Conclusions
In this chapter the pressure dependent part of the tire friction model has been identified.
To this end a numerical and experimental analysis of the Laboratory Abrasion and skid
Tester 100 has been carried out. The numerical model is implemented in ABAQUS, where
a 2D cross section of the tire is created to generate the 3D model. This model is statically
validated with footprint measurements. A pressure dependent friction model has been
derived from the identified friction characteristics. This friction model has been imple-
mented in the FE model for the tire/disk contact. The steady-state numerical modeling
framework as discussed in chapter 2 has been used to efficiently compute the steady-state
free-rolling solution under different loads and side slip angles. It has been shown that
the rolling resistance in the longitudinal direction is captured as observed in the experi-
ments. The computed steady-state forces show a good qualitative match with experiments
for all side slip angles and loads. Moreover the present turn slip is also captured well with
the model. Furthermore, the influence on the lateral force of the friction model is only
visible for large side slip angles, while for small slip angles the friction coefficient has
a very limited influence and the dynamic stiffness of the rubber compound dominates
the response. An even better quantitative result is obtained when the calculated average
contact pressure is corrected using the results of the numerical model.
76
C HAPTER FIVE

Friction parameter identification using


longitudinal slip characteristics1

Abstract / The proposed friction model is decomposed into the product of a contact pressure de-
pendent part, which has been identified in Chapter 4, and a slip velocity dependent part. In this
chapter the identification of the slip velocity dependent part, using measured axle forces, is pre-
sented. The complete identified friction model is coupled to the FE model of the tire under testing.
The steady-state transport approach is used to efficiently compute the steady-state longitudinal slip
characteristics, which show a good quantitative agreement with the experiments.

5.1 Introduction
The strategy developed to capture observed effects of dry friction on the characteristics
of rolling tires is presented in this chapter. Braking experiments at different velocities,
with the tire as discussed in Chapter 3, have been conducted to obtain a slip velocity
dependent parameter set for the tire friction model. The parameters are identified using
measured axle forces, but to exclude small variations in vertical load, which occur during
experiments, the semi-empirical Magic Formula (MF) model (Pacejka, 2006) is used to
evaluate the longitudinal slip characteristics at a constant vertical load.
For the identification procedure measurements at 0.8 times the load index (L.I.) are used,
while measurements at 0.4 and 1.2 times the load index are used for validation of the
tire friction model. With the complete identified parameter set of the friction model the
steady-state longitudinal slip characteristics are computed using the procedure described
in Chapter 3. The obtained numerical results are compared with the Magic Formula.

1
Parts of this chapter have been presented in van der Steen et al. (2010a).

77
78 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

The chapter is organized as follows. In Section 5.2 a description of the followed strat-
egy, using full scale experiments, is given. The conducted tire measurements used for
the Magic Formula are provided in Section 5.3. After that, the identification process is
explained in Section 5.4. With the identified tire friction model the steady-state longitudi-
nal slip characteristics are computed. The free-rolling rotational velocities are compared
with measurement data in Section 5.5. Next the results of both the FE model and the MF
are compared in Sections 5.6 and 5.7, while in Section 5.8 a discussion of the results is
presented. Finally, conclusions are drawn in Section 5.9.

5.2 Identification using full scale experiments


The approach followed to capture observed effects of dry friction on the handling charac-
teristics of rolling tires is described in this section. The friction model (Section 2.4.2) has
been decomposed into the product of a contact pressure dependent part and slip velocity
dependent part as

µ(p, v) = µp (p)µv (v). (5.1)

A method to obtain the two parameters of the pressure related part (p0 and k) of the
friction model has been presented in Chapter 4 using the Laboratory Abrasion and skid
Tester. This chapter focuses on the identification of the remaining velocity related part,
which is given by
   
2 2 v
µv (v) = µs + (µm − µs ) exp −h log . (5.2)
Vmax
Most experimentally determined friction laws in literature do not provide information of
the sliding velocities that occur during real operating conditions. Experimental data is
usually available in the range of low sliding velocities only, e.g. Huemer et al. (2001a);
Dorsch et al. (2002); Blume et al. (2003), while in terms of handling performance much
higher sliding velocities are experienced. Therefore braking experiments at different
velocities with the test tire will be used to obtain the velocity dependent data set for
identification of the friction model. The friction model is validated by comparing
simulated axle forces with measured forces, which is shown in Figure 5.1.

The drawback of using high velocity tire testing is that no measurements can be made
in the contact area and only axle measurements are thus available for identification pur-
poses. Therefore this ‘global’ axle data is used to identify the parameters of the friction
model, which is implemented locally in the contact area of the FE model. Furthermore
the parameters will be made dependent on the forward velocity, such that the FE model
5.3 T IRE FORCE AND MOMENT MEASUREMENTS 79

can also be used for other speeds than used in experiments, which broadens the applica-
bility of this method.

LAT
Inputs
3D Forces Experiments

Lab scale
Friction
FEM 3D Forces Simulations
model LAT wheel

3D Forces Experiments
TNO Tyre Test Trailer
Full scale
FE Tire
3D Forces Simulations
Inputs
Virtual prototyping

Figure 5.1 / Schematic overview of the modeling approach to obtain friction information
from full scale experiments.

5.3 Tire force and moment measurements

Figure 5.2 / TNO Tyre Test Trailer.

To assess the real tire behavior, experiments have been carried out with the TNO Tyre
Test Trailer, see Figure 5.2, on a proving ground. This trailer is typically used to measure
the steady-state force and moment slip characteristics of a tire on a real road. For this
research, experiments have been carried out to measure the longitudinal force, Fx , as
function of the longitudinal slip, κ, during straight line braking up to wheel lock. Vari-
80 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

ous operating conditions have been considered by changing the vertical load, Fz , trailer
forward velocity, Vx , and tire inflation pressure. The tire behavior is measured for three
vertical loads (0.4, 0.8 and 1.2 times the load index), three inflation pressures (nomi-
nal pressure and nominal pressure ±0.5 bar) and five forward velocities (from 20 to 100
km/h).
Although the vertical load is controlled to the set value, small variations in vertical load
occur in the experiments. These variations are caused by road unevenness, trailer motion
and the relatively fast tire deformation that occurs during braking. For a fair comparison
with the FE model, it is desirable to have the longitudinal slip characteristics available at a
precise vertical load. To achieve this, the semi-empirical Magic Formula model has been
used. For each velocity a Magic Formula dataset has been identified on the brake mea-
surements at the three loads with the parameter identification software MF-Tool (TNO,
2010). Then the interpolation capabilities of the Magic Formula are used to obtain the
measured slip characteristics at the exact vertical load.

5.4 Friction parameter identification


The parameter identification consists of several steps, which are explained in this section.
A schematic overview of these steps is shown in Figure 5.3. First the longitudinal braking

Magic Formula µM F (κ)


µM F (κ) = Fx
Fz
µM F v (κ) = µp
κ − Fx

µv Vmax = κmax Vx
Fit parameters Implement local:
µlocal (p, v) = µp (p)µv (v)
|κ|

Figure 5.3 / Schematic overview of the identification procedure.

force at 0.8 times the load index, using the Magic Formula, is divided by the applied
vertical load to obtain a global friction coefficient µM F as function of κ. By introducing
the global longitudinal slip as
v
κ=− , (5.3)
Vx
with v the sliding velocity given by
v = Vx − ωre , (5.4)
5.4 F RICTION PARAMETER IDENTIFICATION 81

where ω is the rotational velocity and re the effective rolling radius, it follows that κ varies
from −1 at wheel lock to zero at free-rolling. Since both κ and Fx are negative during
braking, the absolute values are taken to obtain a positive friction coefficient µM F for |κ|
in the interval [0 1].
Secondly the global friction coefficient is divided by µp . This value is computed as Fx /Fz
at wheel lock, obtained by an FE simulation. In this simulation the friction model derived
in Chapter 4 is implemented, which depends on contact pressure only. This leads to a
velocity dependent part of the global friction coefficient as function of κ
   
2 2 κ
µv (κ) = µs + (µm − µs ) exp −h log , (5.5)
κmax
with
Vmax
κmax = . (5.6)
Vx
The result of these steps is shown in Figure 5.4 for all tested velocities. From this figure it
can be seen that the overall response of the tested tire is similar for all velocities. Around
10 − 15% slip the maximum of the friction coefficient is reached and for higher slip ratios
the friction coefficient drops, this behavior is well-known from experiments and also doc-
umented in literature (Yamazaki et al., 2000; Garro et al., 1999; Gent and Walter, 2005;
Pacejka, 2006).
Furthermore, it can be seen that there is however a difference in the peak values of the
friction coefficient. This can be explained by the hysteretic friction component, which
results from internal friction of the rubber. During sliding asperities of a rough surface
exert oscillating forces on the rubber surface. This leads to cyclic deformations of the
rubber and to energy dissipation caused by the internal damping of the rubber (Pers-
son, 2001). Since all experiments have been performed on the same proving ground,
the wavelengths (λi ) remain equal but the excitation frequencies change with different
forward velocity Vx (fi ∼ Vx /λi ). This causes different losses in the rubber material for
varying forward velocity (Moore, 1972).
Additionally, the frictional energy increases with increasing velocity, which leads to a tem-
perature increase in the contact area. The temperature also affects the viscoelastic prop-
erties of the rubber, which in turn change the frictional behavior.
Both the effects on the rubber material of the surface roughness and the temperature
increase are not explicitly incorporated in the proposed friction model, therefore these
effects are captured in the velocity part of the friction model. To identify the parameters
in (5.5) the following assumptions are made to simplify the identification procedure:

• The point where the peak in the friction model is located (κmax ) can be chosen the
same for all velocities.

• The value for µM F v for a fully locked tire (|κ| = 1) can be chosen the same for all
velocities.
82 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

1.2

[−] 0.8

0.6
MFv

0.4
µ

20 km/h
0.2 40 km/h
60 km/h
0 80 km/h
100 kmh
−0.2
0 0.2 0.4 0.6 0.8 1
|κ| [−]

Figure 5.4 / Velocity part of the global friction coefficient as function of |κ| for five forward
velocities.

The global longitudinal force equals the integral of the shear stress over the contact sur-
face. The slip velocity increases however over the contact length from the leading edge
towards the trailing edge, when the tire starts to slip. Therefore the shear stresses near
the leading edge are higher than the shear stresses at the trailing edge. As a result the
maximum local peak friction coefficient must be higher than the global friction coeffi-
cient. Therefore the value of κmax is chosen at κ = 0.05, such that the maximum friction
coefficient µm will be higher than the friction coefficient shown in Figure 5.4. In this
manner the influence of the velocity part of the friction model on the longitudinal slip
stiffness is small. In other words, the structural response of the tire dominates the longi-
tudinal slip stiffness.
If κ = 1 is substituted in (5.5) the following expression for µlock is obtained
   
2 2 1
µlock = µs + (µm − µs ) exp −h log . (5.7)
κmax

By fixing the value of µlock , as the average value of µM F v at κ = 1, an explicit expression


for the parameter h2 as function of µs and µm is given by
 
ln µµlock
m −µs
−µs
h2 = 2 . (5.8)
log (κmax )
5.4 F RICTION PARAMETER IDENTIFICATION 83

The remaining parameters µs and µm of (5.5) are determined separately for all five veloc-
ities by minimizing a least-square error
m
X
min 1
2
(µv ([µs µm ], κi ) − µM F v (κi ))2 , [µs µm ] ∈ R2 . (5.9)
[µs µm ]
i=1
subject to:
−µs < 0
µs − µlock < 0
µlock − µm < 0

The interval for κ is chosen from the maximum of µM F v to κ = 1 with an equidistant


distribution for κ to put an equal weight on the used data points, this is illustrated in
Figure 5.5 for one velocity. In this way the longitudinal slip stiffness of the tire is not
included in the identification process. The obtained parameters for µm as function of

0.9

0.8
µMFv [−]

0.7

0.6

0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
|κ| [−]

Figure 5.5 / Data points of one velocity used for the identification of µs and µm .

forward velocity are shown in Figure 5.6. These five parameters values are approximated
using a linear least-square error fit as function of the forward velocity

µm = a1 Vx + a2 . (5.10)

Using (5.10) for the value of µm a second optimization is done for every velocity for µs
m
X
min 1
2
(µv (µs , κi ) − µM F v (κi ))2 , µs ∈ R. (5.11)
µs
i=1
subject to:
−µs < 0
µs − µlock < 0
84 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

The obtained parameters for µs as function of forward velocity are shown in Figure 5.7,

1.1

1.05
µm [−]

0.95

0.9
20 30 40 50 60 70 80 90 100
Vx [km/h]

Figure 5.6 / Obtained parameter values of µm as function of forward velocity.

the five parameters values are approximated using a second order linear least-square error
fit as function of the forward velocity
µs = a3 Vx2 + a4 Vx + a5 . (5.12)

0.7

0.6

0.5

0.4
µs [−]

0.3

0.2

0.1

0
20 30 40 50 60 70 80 90 100
Vx [km/h]

Figure 5.7 / Obtained parameter values of µs as function of forward velocity.

Now (5.5) can be evaluated in the entire interval of forward velocities between 20 and 100
km/h using (5.8), (5.10) and (5.12) for the values of h2 , µm and µs . The last step in the
procedure, see Figure 5.3, consists of switching from (5.5) back to (5.2) by
Vmax = κmax Vx . (5.13)
5.5 F REE - ROLLING ROTATIONAL VELOCITY : COMPARISON OF FEM PREDICTION AND EXPERIMENTS 85

5.5 Free-rolling rotational velocity: Comparison of FEM pre-


diction and experiments
The procedure described in Section 3.5 is followed to compute the entire braking char-
acteristic of the tire. The tire is brought up to speed at the desired forward velocity in
the first steady-state transport step, while initializing the rotational velocity at 1 rad/s. In
the second steady-state transport step the rotational velocity of the tire is increased every
increment until the free-rolling situation is reached. Simulations are carried out for all
five forward velocities and the three load conditions. The computed free-rolling rotational
velocity is compared with the, filtered and averaged, measured rotational velocity just be-
fore a brake test starts. The results are shown in Table 5.1, where it can be seen that there
is a good match for all loads and velocities.

Table 5.1 / Free-rolling rotational velocity (in rad/s) for five forward velocities and three
load conditions of experiments and FE model.

Velocity 0.4 L.I. [N] 0.8 L.I. [N] 1.2 L.I. [N]
20 Experiment 18.1 18.7 18.0
[km/h] FE Model 17.6 17.7 17.7

40 Experiment 36.3 36.2 35.1


[km/h] FE Model 35.2 35.4 35.4

60 Experiment 54.1 53.7 54.2


[km/h] FE Model 52.7 53.0 53.1

80 Experiment 71.1 71.5 72.3


[km/h] FE Model 70.3 70.7 70.8

100 Experiment 88.4 89.4 90.1


[km/h] FE Model 87.8 88.3 88.5

Furthermore it can be seen that for the FE model the free-rolling rotational velocity in-
creases with increasing load. This is expected, since the rotational velocity of a free-rolling
tire is directly related to the forward velocity as
Vx
ωf ree = , (5.14)
re
where re is the effective rolling radius, which decreases when the load increases. The
86 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

value of the free-rolling rotational velocity can be used to determine the longitudinal slip
in a more practical way as follows
ω − ωf ree
κ= . (5.15)
ωf ree

This formulation is used to compare the simulated steady-state braking characteristics


with the Magic Formula.

5.6 Comparison between FEM and MF predictions


The longitudinal forces as function of the longitudinal slip are shown in Figures 5.8,
where the three load conditions are grouped for each velocity. In these figures the Magic
Formula is used to describe the measured slip characteristics at the desired vertical loads.
In the MF-Tool software the parameters of the MF are chosen such that the error between
the MF and experiments is minimized. As a result, it can happen that the longitudinal
force at κ = 0 is positive. However, the measured longitudinal force at free-rolling is
always negative due to a rolling resistance force.
The longitudinal forces for the lowest and highest load conditions are computed using
the same parameters in the friction model as for the middle loads, which have been used
in the identification process. Therefore these results illustrate the predictive capability of
the friction model.
It can be seen that the axle force of the FEM curves at 0.8 times the load index, which
are used in the identification process, match the Magic Formula both qualitatively and
quantitatively. This shows that it is possible to fit the parameters of the velocity part of
the friction model using measured axle forces and implement one parameter set of the
friction model locally at each node.
Although some deviations can be seen for the higher load index, especially at 20 km/h, it
can be concluded that there is also a good quantitative agreement between the FE predic-
tions and the MF for 0.4 and 1.2 times the load index. Since the only difference in these
simulations is the applied load, it follows that the decoupling of the friction model into a
velocity and a contact pressure part is a valid assumption.
The deviation with the MF is most obvious at the lowest velocity of 20 km/h. This is due
to the choice of one value for µlock for all velocities. It can be seen in Figure 5.4 that µlock
at 20 km/h is significantly higher than at the other velocities, which leads to an underes-
timation of the friction coefficient for this velocity at increasing slip ratios. This effect is
amplified by increasing the load on the tire, which can be seen in Figure 5.8a, since the
velocity part of the friction model is proportional to the applied load.
5.6 COMPARISON BETWEEN FEM AND MF PREDICTIONS 87

−2000

−4000
Fx [N]

−6000

−8000
MF
−10000 FEM

−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0


κ [−]

(a) 20 km/h.

−2000

−4000
Fx [N]

−6000

−8000 MF
FEM
−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

(b) 40 km/h.

−2000

−4000
Fx [N]

−6000

−8000 MF
FEM
−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

(c) 60 km/h.
88 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

−2000

−4000
Fx [N]

−6000

−8000 MF
FEM
−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

(d) 80 km/h.

−2000

−4000
Fx [N]

−6000

−8000 MF
FEM
−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

(e) 100 km/h.

Figure 5.8 / Longitudinal force as function of longitudinal slip for both Magic Formula
(MF) and FE Model at 0.4, 0.8 and 1.2 times the load index.

5.7 Effect of inflation pressure on the longitudinal force


Besides the load and forward velocity also the inflation pressure in the experiments has
been varied. When the inflation pressure is decreased, at a fixed load, the contact area
increases, which leads to a lower contact pressure. This normally results in higher lon-
gitudinal forces, which is confirmed by the Magic Formula as shown in Figure 5.9. The
effect of a lower contact pressure in the contact area on the friction model is that the fric-
tion coefficient increases and a higher longitudinal force can be generated. It can be seen
in Figure 5.9 that the increase in longitudinal force is overestimated. Furthermore the
deformations for the simulation at the highest load are so large, due to the under-inflated
and overloaded tire, that convergence is not obtained for the lower range of longitudinal
slip.
5.7 E FFECT OF INFLATION PRESSURE ON THE LONGITUDINAL FORCE 89

In contrast to an under-inflated tire, an over-inflated tire has a higher contact pressure


and the tire is riding more on the center of the tread rather than on the shoulders of the
tire. The higher contact pressure leads to a lower friction coefficient and this results in
a lower longitudinal force, shown in Figure 5.10. Now it can be seen that the FE pre-
dictions slightly underestimate the longitudinal force, compared to the Magic Formula
results. An explanation for deviations of the longitudinal force with respect to the Magic
Formula could be due the change in contact pressure distribution. If the contact pressure
distribution in the contact area is inspected, shown in Figure 5.11, it follows that the con-
tact pressure is outside the measured range on the LAT 100 for a large part of the contact
area (Figure 4.17). This indicates that an accurate prediction of the friction coefficient can
not be guaranteed, since (4.2) is extrapolated outside the measurement range.

0
MF
−2000 FEM

−4000
Fx [N]

−6000

−8000

−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

Figure 5.9 / Longitudinal force as function of longitudinal slip for MF and FE Model at
0.4, 0.8 and 1.2 times the load index at 0.5 bar under nominal pressure at 60 km/h.

−2000

−4000
Fx [N]

−6000

−8000 MF
FEM
−10000
−1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0
κ [−]

Figure 5.10 / Longitudinal force as function of longitudinal slip for MF and FE Model at
0.4, 0.8 and 1.2 times the load index at 0.5 bar above nominal pressure at 60 km/h.
90 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

250

Number of nodes in contact


200

150

100

50

0
0 0.2 0.4 0.6 0.8 1

250
Number of nodes in contact

200

150

100

50

0
0 0.2 0.4 0.6 0.8 1

250
Number of nodes in contact

200

150

100

50

0
0 0.2 0.4 0.6 0.8 1
Contact pressure [MPa]

Figure 5.11 / Contact pressure distribution for the under-inflated (top), nominal inflated
(middle) and over-inflated (bottom) tire at 0.8 L.I. for peak longitudinal force.
5.8 D ISCUSSION 91

5.8 Discussion
The approach proposed in this thesis allows to compute steady-state braking character-
istics under different loads and velocities with a good accuracy compared to the Magic
Formula. The interpolation of the parameters µm and µs , using (5.10) and (5.12), allows
to use the friction model at other velocities than the five velocities used in experiments.
This broadens the applicability of the friction model, but at the five test velocities, the
deviation with the MF increases. It is shown in Figure 5.6 that the largest deviation with
respect to the fit occurs at 40 km/h and this is directly reflected in the results, shown
in Figure 5.8b, where the peak force is overestimated. If only the tested velocities are
of interest, the match with the MF at these five velocities can be enhanced by using the
optimized parameters µs and µm for each separate velocity.
The derived parameter set is not unique, but depends on the choices of µlock and κmax .
The effect of the choice for one value of µlock has already been discussed, but it was also
assumed that the location of κmax can be chosen the same for all velocities. Additional
simulations are performed, which confirm that κmax at 0.05 is indeed reasonable. In one
model the peak value is shifted to κmax = 0.01 and a second model the peak is shifted
to κmax = 0.10, which is around the point where the maximum of the longitudinal force
is located. The results of these simulations are shown in Figure 5.12. It can be seen that

0 0
MF MF
−1000 −1000
FEM κmax 0.05 FEM κmax 0.05
−2000 −2000
FEM κmax 0.10 FEM κmax 0.01
Fx [N]

Fx [N]

−3000 −3000

−4000 −4000

−5000 −5000

−6000 −6000

−7000 −7000
−1 −0.8 −0.6 −0.4 −0.2 0 −1 −0.8 −0.6 −0.4 −0.2 0
κ [−] κ [−]

(a) 60 km/h. (b) 100 km/h.

Figure 5.12 / Longitudinal force as function of longitudinal slip for both Magic Formula
(MF) and FE Model for two values of κmax at 0.8 times the load index.

if the peak location is shifted to 0.10 the tire starts to slide slightly earlier and an under-
estimation of the longitudinal force is observed. Furthermore it turns out, also for other
velocities, that the peak longitudinal force can not be reached when κmax = 0.10. How-
ever if a value of 0.01 is used, the results are more or less the same. This indicates that
92 5 F RICTION PARAMETER IDENTIFICATION USING LONGITUDINAL SLIP CHARACTERISTICS

if the value of κmax is chosen before the point where the peak longitudinal force occurs,
the presented approach leads to accurate predictions of the tire response.

5.9 Conclusions
In this chapter the parameter identification of the sliding velocity part of the tire friction
model has been presented. The described identification procedure is based on measure-
ments of the longitudinal force as function of the longitudinal slip during straight line
braking under different driving velocities. With the obtained parameters of the sliding ve-
locity part, the entire phenomenological friction model has been identified. The derived
model captures observed effects of dry friction on the longitudinal slip characteristics of
a rolling tire.
The steady-state numerical modeling framework as discussed in chapter 2 has been used
to efficiently compute the steady-state braking solution under different loads and veloc-
ities. It has been shown that the computed steady-state forces are in good quantitative
agreement with experiments for all loads. This supports the validity of fitting the param-
eters of the velocity part of the friction model using measured axle forces and implement-
ing one parameter set of the friction model locally at each node in contact. Furthermore,
it has been shown that the decoupling of the friction model into a velocity and a pressure
part is a valid assumption. However the results at different inflation pressures suggest
that extrapolation of the contact pressure dependent part of the friction model leads to
deviations with the Magic Formula.
In the next chapter the predictive capability of the derived friction model is investigated
by computing the forces and moments that occur during steady-state cornering and com-
bined slip conditions.
C HAPTER SIX

Predictive capability of the FE tire model

Abstract / The fully identified friction model is used to compute the pure cornering and combined
slip characteristics. The obtained results are discussed and the FE model is used to demonstrate
possible extensions in future applications. Inevitable variations on the forces and moments, which
occur during measurements, are provided to put the predicted forces and moments into the right
perspective. Finally, the predicted characteristics are compared with the Magic Formula and it is
shown that the handling performance of the tire can be adequately predicted with the identified
friction model.

6.1 Introduction
The final goal of the FE model is to use it in an early design stage and predict the tire
behavior, such as handling, rolling resistance and wear, by creating virtual prototypes
instead of real prototypes. In this chapter the fully identified friction model is used to
investigate the predictive capability of the friction model by comparing simulated axle
forces of cornering and combined slip characteristics with Magic Formula evaluations.
Besides the validation of the steady-state handling characteristics under dry conditions,
the FE model is also used to illustrate possible future extensions, such as wear.
Furthermore details about the tire experiments are provided to show the inevitable vari-
ations during measurements on supposedly identical tires. These uncertainties in the
characteristics arise from changing chemical and viscoelastic properties of tire materials
combined with effects of wear and environmental conditions. These variations can be
used to put the predicted cornering and combined slip characteristics of the FE model
into a better perspective.
This chapter is organized as follows. In Section 6.2, the predicted handling characteristics
for pure cornering and combined slip conditions are given and discussed. After that,

93
94 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

observations about the experimental data are presented in Section 6.3, while in Section
6.4 a comparison of the FE model with the Magic Formula is made and the results are
discussed. Finally, conclusions are summarized in Section 6.5.

6.2 Prediction of the handling characteristics

6.2.1 Pure cornering characteristic

The pure cornering characteristics have been measured for side slip angles ranging from
−12◦ to 12◦ . This range is also used in the simulations, where the procedure presented
in Section 3.6 is applied. As starting point for the rotational velocities in the steady-state
rolling step, the free-rolling rotational velocities for a zero degree slip angle as given in
Table 5.1 are used. The calculation of the cornering characteristic is therefore split into
two separate simulations, one for positive and one for negative slip angles, which both
start at zero degree slip angle.
The forward velocity in the simulations is 60 km/h, which is equal to the standard test
velocity during experiments. During a cornering manoeuvre not only a lateral force is
generated, but also a self-aligning moment Mz and an overturning moment Mx .

Lateral force

The computed lateral forces for 0.4, 0.8 and 1.2 times the load index are shown in Figure
6.11 . In the region for |α| < 2.5◦ the change in lateral force is governed by the cornering
stiffness CF α , and the built-up of the lateral force is linear. It can be seen that the cor-
nering stiffness depends on the applied load. The cornering stiffness typically increases
up to a fraction of the load index and then falls off in magnitude as the load increases
further. It is desirable that the location of this peak is above the tire load index for good
handling (Gent and Walter, 2005, chapter 8).
As the slip angle becomes larger, more and more of the available contact area starts to
slide and a maximum amount of lateral force will be generated at a certain slip angle,
in this case around ±10.5◦ . Although not visible in Figure 6.1, beyond this peak value,
increasing the slip angle decreases the lateral force due to increasing slip velocity. This
is analogous to the response of the tire subjected to pure braking, where the longitudinal
force for increasing longitudinal slip decreases.
It can also be seen that the magnitude of the lateral force at positive slip angles is slightly
larger than for negative slip angles. This can be explained by the non-symmetric tire tread

The deformations at the highest load and positive slip angle are so large, that for slip angles above 9.2◦
1

no solution is obtained with the current model.


6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 95

8000
0.4 L.I.
6000 0.8 L.I.
1.2 L.I.
4000

2000
Fy [N]

−2000

−4000

−6000

−8000
−15 −10 −5 0 5 10 15
α [°]

Figure 6.1 / Simulated lateral force as function of side slip angle at 0.4, 0.8 and 1.2 times
the load index, at a forward velocity of 60 km/h.

design, see Section 3.3, in combination with the used friction model. When a slip angle
is present, the footprint is reshaped into a trapezoid. A right turn (positive slip angle)
leads to a long left shoulder and a short right shoulder, whereas a left turn leads to a long
right shoulder and a short left shoulder. The long shoulder is also exposed to a higher
magnitude of normal stress than the more lightly loaded short shoulder, which means
that the lateral force is mainly due to the integral of the lateral shear stresses on the long
shoulder. With the asymmetric test tire, two different footprint sizes are obtained during
steering to the left and right. The larger surface area on the wide left shoulder, as shown
in Figure 6.2b for a positive slip angle, leads to a lower average contact pressure, which
in turn leads to a higher lateral force. In contrast, the smaller surface area on the right
shoulder, as shown in 6.2a leads to a higher average contact pressure, which in turn leads
to a lower lateral force with respect to the positive slip angle. The black areas in both fig-
ures, which indicate the highest local friction coefficients, are due to the low magnitude
of the normal stress and their contribution to the overall lateral force is small.
The calculated sliding velocities during steering are more or less the same for steering
to the right and left, which means that the slip velocity dependent friction coefficient
distribution is similar for left and right turns as is shown in Figure 6.3. Here the highest
local friction coefficients are found at the leading edge of the contact patch, where the
slip velocity is the lowest. As a result the lateral force acting on the wheel is higher for
positive slip angles than for negative slip angles due to the difference in footprint size.
96 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

(a) Negative slip angle of 12◦ . (b) Positive slip angle of 12◦ .

Figure 6.2 / Distribution of local pressure dependent friction coefficient, where the scale
varies from 0 to 2 and darker colors indicate higher coefficients. The bottom of the figure
corresponds to the leading edge of the contact patch.

(a) Negative slip angle of 12◦ . (b) Positive slip angle of 12◦ .

Figure 6.3 / Distribution of local slip velocity dependent friction coefficient, where the
scale varies from 0 to 2 and darker colors indicates higher coefficients. The bottom of the
figure corresponds to the leading edge of the contact patch.
6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 97

Self-aligning moment

The location of the lateral force resultant is a function of the slip angle and gives rise to
a moment around the z-axis, which is known as the self-aligning moment and can be
described by

Mz = −tFy + Mzr , (6.1)

with t the pneumatic trail and Mzr a small residual torque (Pacejka, 2006). The self-
aligning moments are shown in Figure 6.4, where it can be seen that the moments first
increase and after a few degrees slip angle decrease again. The location of the lateral force
resultant moves towards the origin of the reference coordinate system for increasing slip
angle (Pacejka, 2006) and the pneumatic trail thus decays with increasing slip angle. The
pneumatic trail can even move ahead of the origin at very large slip angles, such that the
self-aligning moment changes sign.

200

150

100

50
Mz [Nm]

−50

−100
0.4 L.I.
−150 0.8 L.I.
1.2 L.I.
−200
−15 −10 −5 0 5 10 15
α [°]

Figure 6.4 / Simulated self-aligning moment as function of side slip angle at 0.4, 0.8 and
1.2 times the load index, at a forward velocity of 60 km/h.

Plysteer and conicity

In Figure 6.5 a close-up of Figures 6.1 and 6.4 is shown, where it can be seen that both
the lateral forces and self-aligning moments at zero degree slip angle are nonzero. This is
caused by two effects known as plysteer and conicity. Plysteer effects are due to structural
tire design, such as the distance between the belts, tread pattern, as well as the frictional
98 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

2000 100
0.4 L.I.
1000
0.8 L.I.
1.2 L.I. 50

Mz [Nm]
Fy [N]

0
0

−1000
−50 0.4 L.I.
0.8 L.I.
−2000 1.2 L.I.
−100
−1 −0.5 0 0.5 1 −1 −0.5 0 0.5 1
α [°] α [°]

(a) Lateral force. (b) Self-aligning moment.

Figure 6.5 / Close-up of the lateral forces and self-aligning moments around zero degree
slip angle.

dynamics of rolling tires, which is discussed in detail by Ohishi et al. (2002). The gen-
erated lateral force due to plysteer does change sign, if the rolling direction of the tire is
reversed and plysteer is therefore also known as pseudo side slip.
Conicity can occur if the tire belt is located slightly off-center, which is the result of manu-
facturing variances (Gent and Walter, 2005, chapter 8) and is not present in the FE model.
Conicity has the effect that the rolling radius varies from one side to the other and the tire
acts as if it has a conical cross section. This also results in the development of a lateral
force and aligning torque. The lateral force does not change sign if the rolling direction
is reversed, which is similar to a cambered wheel and this effect is also known as pseudo
camber.
The residual cornering force is the cornering force at zero self-aligning torque, similarly
residual aligning torque, Mzr , is the aligning torque at zero cornering force. Both are
directly related to vehicle handling, since under free control (zero aligning torque) the
nonzero residual lateral force pushes the vehicle to the side. To maintain a straight line
on the road, a driver needs to counterbalance the residual aligning torque to have zero
lateral force.
Automobile manufactures often demand a specific range of plysteer residual aligning
torque, which is taken into account in the overall vehicle design to reduce tire-induced
pull effects. This can lead to different tire designs for a nominally identical tire and as
a result pull problems can occur if the original equipment tires are replaced (Gent and
Walter, 2005, chapter 8).
6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 99

Overturning moment and loaded radius

The overturning moment Mx at the wheel axis is a function of the lateral force and the
loaded radius of the tire and is shown in Figure 6.6.

2500
0.4 L.I.
2000
0.8 L.I.
1500 1.2 L.I.
1000

500
Mx [Nm]

−500

−1000

−1500

−2000

−2500
−15 −10 −5 0 5 10 15
α [°]

Figure 6.6 / Simulated overturning moment as function of side slip angle at 0.4, 0.8 and
1.2 times the load index, at a forward velocity of 60 km/h.

0.97

0.96
0.4 L.I.
0.95
0.8 L.I.
0.94 1.2 L.I.
rl / ru [−]

0.93

0.92

0.91

0.9

0.89

0.88
−15 −10 −5 0 5 10 15
α [°]

Figure 6.7 / Normalized loaded radius as function of side slip angle at 0.4, 0.8 and 1.2
times the load index, at a forward velocity of 60 km/h.
100 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

The loaded radius rl of the FE model is approximated as the distance from the wheel
center to a single node in the middle of the contact area. This node has also a lateral
and longitudinal displacement during steering, so only the vertical displacement is taken
into account to calculate the loaded radius of the tire. The loaded radius, normalized
with the unloaded radius ru , as function of applied load is shown in Figure 6.7, where
it can be seen that indeed a bell-shape, as discussed by Pottinger in (Gent and Walter,
2005, chapter 8), is obtained. Furthermore the decrease of the loaded radius at large slip
angles increases for increasing load. These results are in good agreement with similar
data presented in Pacejka (2006).

6.2.2 Combined slip characteristic

Combined slip is an important characteristic, since during normal vehicle operation cor-
nering is often combined with a torque, driving or braking, action. With the FE model
combined slip characteristics, under several slip angles, have been computed at a con-
stant forward velocity of 60 km/h. These simulations are started at the free-rolling sit-
uation and then the rotational velocity is reduced to zero. In Figures 6.8 and 6.9 the
longitudinal forces as function of longitudinal slip are shown for 0.4 and 0.8 times the
load index respectively.
It can be seen that for increasing slip angle the peak longitudinal force decreases. The
rate at which the force decreases for increasing α is directly related to the isotropic friction
model, where the friction coefficient is assumed to be independent of the direction of the

0

−500 +3°
+5°
−1000 +8°
−3°
−1500 −5°
Fx [N]

−8°
−2000

−2500

−3000

−3500
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.8 / Longitudinal force as function of longitudinal slip for 7 slip angles at 0.4
times the load index, at a forward velocity of 60 km/h.
6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 101

1000

0 +3°
+5°
−1000 +8°
−3°
−2000
−5°
Fx [N]

−3000 −8°

−4000

−5000

−6000

−7000
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.9 / Longitudinal force as function of longitudinal slip for 7 slip angles at 0.8
times the load index, at a forward velocity of 60 km/h.

γ̇2
τ1
γ̇eqv τeqv τ2
γ̇1

Figure 6.10 / Representation of the slip circle with slip velocity vector and corresponding
frictional shear stress vector.

slip velocity vector. This can be visualized with a so-called slip circle as shown in Figure
6.10. Since the generated frictional shear stress is oriented with the slip velocity vector,
the shear stress in longitudinal direction (τ1 ) becomes smaller if side slip is present.
For a locked wheel, under a slip angle α, all local slip velocity vectors are aligned with the
forward velocity, V , and the resulting slip velocity equals the forward velocity. Therefore
the lateral and longitudinal slip vectors are the projections of the forward velocity vector
on the lateral and longitudinal directions, given by sin(α)V and cos(α)V , respectively.
Due to the relatively small slip angles it follows that the longitudinal slip velocity is much
higher than the lateral slip velocity and the longitudinal force approaches the solution
for zero slip angle at wheel lock. This also implies that the lateral force for increasing
longitudinal slip should decrease, which is confirmed in Figures 6.11 and 6.12, where the
102 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

lateral force as function of longitudinal slip is shown.2

3000

+3°
2000
+5°
+8°
1000 −3°
−5°
Fy [N]

0 −8°

−1000

−2000

−3000
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.11 / Lateral force as function of longitudinal slip for 7 slip angles at 0.4 times
the load index, at a forward velocity of 60 km/h.

6000

+3°
4000
+5°
+8°
2000 −3°
−5°
Fy [N]

0 −8°

−2000

−4000

−6000
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.12 / Lateral force as function of longitudinal slip for 7 slip angles at 0.8 times
the load index, at a forward velocity of 60 km/h.

2
For 3◦ slip angle at 0.8 L.I. no solution is obtained for longitudinal slip ratios above 71% with the
current model.
6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 103

The shape of the self-aligning moment curve at combined slip is a complex combination
of several effects. At free-rolling, the self-aligning moment is given by the pure cornering
characteristic as shown in Figure 6.4. When a brake torque is applied, a longitudinal
force is generated and the tire deforms in longitudinal direction due to flexibility of the
carcass. This results in additional contributions to the self-aligning moment. In Figure
6.13 a brush model with flexible carcass is shown, where it can be seen that the deflection
of the carcass results in a deflection uc in longitudinal direction and a deflection vc in
lateral direction. Furthermore the location of the resultant of the longitudinal force has
also an offset vo with respect to the x-axis due to the asymmetric tread of the tire. There-
fore the total self-aligning moment originates from both the magnitude and the resultant
force location of the lateral and longitudinal force.
In Figures 6.14 and 6.15 the self-aligning moment as function of longitudinal slip is
shown. It can be seen that for small values of κ the additional deflection uc causes an
increase of the self-aligning moment. Once the lateral force starts to decrease, the lon-
gitudinal force increases and the location of the resultant of the longitudinal force gen-
erates an additional moment. At κ = −1, Mz is the sum of the lateral offset caused by
the asymmetric tread of the tire times the longitudinal force, the lateral carcass deflection
is reduced if the tire slips completely, and a small lateral force times the longitudinal de-
flection of the carcass. This last term changes sign for positive and negative slip angles,
because the direction of the lateral force changes for positive and negative slip angles.

y
V
α
Wheel plane x
Trailing edge vo Leading edge
Carcass
vc
Fx
uc t
F Fy
Wheel spin axis

Figure 6.13 / Extended brush model with flexible carcass under combined slip situation.
104 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

60

+3°
40
+5°
+8°
20 −3°
−5°
Mz [Nm]

0 −8°

−20

−40

−60
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.14 / Self-aligning moment as function of longitudinal slip for 7 slip angles at
0.4 times the load index, at a forward velocity of 60 km/h.

150

+3°
100 +5°
+8°
−3°
50 −5°
Mz [Nm]

−8°

−50

−100
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.15 / Self-aligning moment as function of longitudinal slip for 7 slip angles at
0.8 times the load index, at a forward velocity of 60 km/h.
6.2 P REDICTION OF THE HANDLING CHARACTERISTICS 105

6.2.3 Friction power distribution in the footprint

The frictional power per unit area, τeqv γ̇eqv , is calculated in order to illustrate other poten-
tial applications of the model. The integral over time is equal to the energy per unit area
and known as the shear energy intensity (Gent and Walter, 2005, chapter 7). This shear
energy intensity can be used as a predictor of tire profile wear as shown by Pottinger and
McIntyre (1999). As an illustration the shear power intensity is shown in Figure 6.16
for four distinct driving conditions. Note that, under straight line driving with constant
velocity (free-rolling), the shear power intensity is not visible on the used scale.

(a) Free-rolling, α = 0◦ , κ = 0. (b) Pure braking, α = 0◦ , κ = −1.

(c) Pure cornering, α = 8◦ , κ = 0. (d) Combined slip, α = 8◦ , κ = −1.

Figure 6.16 / Distribution of frictional power for different driving conditions at 0.8 times
the load index, at a forward velocity of 60 km/h. Darker colors indicate higher intensity.
The bottom of the figure corresponds to the leading edge of the contact patch.
106 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

This is expected, since under free-rolling the sliding velocity is very low. At wheel lock,
the sliding velocity is maximal and the frictional power intensity increases. For the pure
cornering situation the distribution originates from the lateral shear stresses at a rela-
tively low slip velocity. Finally the frictional power intensity for combined slip at wheel
lock is shown, where the highest shear stresses and slip velocities are present.
A large part of the frictional power is converted into heat, which results in increased sur-
face temperature. In Figure 6.17 snapshots, corresponding to the four driving conditions,
of the surface temperature are shown. The surface temperature has been measured with
an infrared camera mounted onto the fixed frame of Tyre Test Trailer, behind the rotating
measurement tower.

(a) Free-rolling, α = 0◦ , κ = 0. (b) Pure braking, α = 0◦ , κ = −1.

(c) Pure cornering, α = 8◦ , κ = 0. (d) Combined slip, α = 8◦ , κ = −1.

Figure 6.17 / Measured surface temperature distribution for different driving conditions
at 0.8 times the load index, at a forward velocity of 60 km/h. Darker colors indicate
higher temperature. The bottom of the figure corresponds to the leading edge of the
contact patch.
6.3 F ORCE AND MOMENT MEASUREMENTS AND M AGIC F ORMULA 107

This means that the history of points traveling through the entire contact zone is captured
and these images thus show a ‘smeared’ temperature distribution. Nevertheless, there is
a clear resemblance with the calculated shear power intensity.
Therefore an useful extension could be to incorporate temperature effects in the FE
model based on the shear power intensity. Different strategies can be followed here.
Giessler et al. (2010) used the friction heat to predict tread temperature increase and the
effect on tire traction under ice and snow conditions. The friction model of Huemer et
al. (2001a) incorporated temperature effects in the material properties using the WLF
transformation. Hofstetter et al. (2006) used a fully coupled thermo-mechanical FE
model to convert a part of frictional power into heat, which increases the tread temper-
ature. This temperature increase also results in changing viscoelastic material properties.

As mentioned, the shear energy intensity is used as a predictor of wear. After pure brak-
ing experiments on the proving ground, flat spots on the tires have been observed. Ad-
ditionally, for the combined slip experiments uneven wear of the inner tread blocks has
been noticed, which indicate that shear energy intensity is a contributing factor in the
complex wear process. Another possible future extension of the FE model could there-
fore be to use this frictional power per unit area to predict wear of the tread profile. This
is done by e.g. Smith et al. (2008), where the steady-state approach of ABAQUS is used to
predict wear of a tread profile under different driving conditions. It is even possible to
adapt the mesh to the worn configuration with the current version of ABAQUS (SIMULIA,
2009b).

6.3 Force and moment measurements and Magic Formula


In this section some observations about the experimental data are discussed before a
comparison of the FE model with the Magic Formula is made. As presented by Pottinger
(Gent and Walter, 2005, chapter 8), an individual tire does not have a single well-defined
set of force and moment characteristics. This uncertainty in the characteristics arises
from the changing chemical and viscoelastic properties of the tire materials combined
with effects of wear.
As already mentioned in Chapter 5, experiments have been carried out with the TNO Tyre
Test Trailer. Besides the straight line braking experiments at different velocities, the stan-
dard measurement program to obtain a Magic Formula dataset has been executed (TNO,
2008). This dataset is constructed from pure braking, pure cornering and combined slip
measurements under different operation conditions and using several tires. All these
measurements are combined in the parameter identification software MF-Tool to identify
one single Magic Formula parameter set. With this single so-called tire property file all
steady-state handling characteristics can be evaluated. Since the MF is a semi-empirical
108 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

approach, with a high fitting accuracy, this tire file provides a good description of mea-
sured handling characteristics at that moment in time.
To investigate external influences on the tire behavior, this complete dataset has been
measured four times (see Table 6.1). The first set has been measured at proving ground
one, while the second set has been measured several months later at proving ground two.
This proving ground has a different road surface, but tires from the same batch have been
used for these measurements. The third and fourth dataset have been measured again
at proving ground one, but one year later than the first measurements. For the third set
a newly manufactured batch of tires has been used, while the fourth set has also been
measured with tires from the first batch.
The four different Magic Formula datasets represent all the same tire, which therefore can
provide information about the sensitivity to external influences on the measured forces
and moments, such as environmental conditions, road surface or aging.

Table 6.1 / Specifications of the different sets used to obtain the Magic Formula datasets.
Tires from batch 1 have been stored at room temperate in between measurements.

Set nr. Proving ground Date Batch


1 1 Nov. 2007 1
2 2 Sep. 2008 1
3 1 Nov. 2008 2
4 1 Nov. 2008 1

In the following figures only the results for 0.8 times the load index are discussed. The
results for the other two loads tested show similar trends.
In Figure 6.18 and Figure 6.19 the pure cornering characteristics are shown. It can
be seen that the lateral force is similar for all sets under all slip angles. The present
differences are however reflected in the self-aligning moment, where the moment in
set 2 declines faster than the other sets. This can be the result of a different road
surface. Furthermore the peak value at negative slip angle of set 3 is 15% higher than the
other three, which could be due to the manufacturing variations between the two batches.

The combined slip characteristics are only shown for three different slip angles for im-
proved visibility. When the longitudinal force as function of longitudinal slip is inspected,
some remarkable differences can be seen in Figure 6.20. The longitudinal force at wheel
lock is significantly lower for sets 2 and 4, which is probably an aging effect. As indicated
by Pottinger (Gent and Walter, 2005, chapter 8), additional crosslinking of the rubber
takes place over time, which affects the viscoelastic properties. It can be seen that at set
6.3 F ORCE AND MOMENT MEASUREMENTS AND M AGIC F ORMULA 109

2 the lowest longitudinal forces have been measured, which suggest a road surface ef-
fect as well. These figures indicate that both the longitudinal and lateral force can show
deviations up to 15% between the different sets.

6000

4000

2000
Fy [N]

−2000
Set 1
Set 2
−4000
Set 3
Set 4
−6000
−15 −10 −5 0 5 10 15
α [°]

Figure 6.18 / Lateral force as function of side slip angle at 0.8 times the load index, at a
forward velocity of 60 km/h.

100
Set 1
80 Set 2
60
Set 3
Set 4
40
20
Mz [Nm]

0
−20
−40
−60
−80
−100
−15 −10 −5 0 5 10 15
α [°]

Figure 6.19 / Self-aligning moment as function of side slip angle at 0.8 times the load
index, at a forward velocity of 60 km/h.
110 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

0
Set 1
−1000 Set 2
Set 3
Set 4
−2000
Fx [N]

−3000

−4000
°
+8
−5000
−5°
−6000 °
0
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.20 / Longitudinal force as function of longitudinal slip for 3 slip angles at 0.8
times the load index, at a forward velocity of 60 km/h.

5000

4000 Set 1
Set 2
3000 Set 3 −5°
2000 Set 4

1000

Fy [N]

−1000

−2000

−3000

−4000 +8°
−5000
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.21 / Lateral force as function of longitudinal slip for 3 slip angles at 0.8 times
the load index, at a forward velocity of 60 km/h.
6.4 COMPARISON OF THE FE MODEL AND THE M AGIC F ORMULA 111

6.4 Comparison of the FE model and the Magic Formula

6.4.1 Pure cornering

The comparison of the lateral force for pure side slip at three load conditions is shown
in Figure 6.22, where a comparison is made with set 1. It can be seen that the overall
prediction of the FE model, for all three loads, is good.
Three observations can be made from this figure. First, the built-up of the lateral force for
the 0.8 and 1.2 load conditions in the FE model is faster than for the MF. This indicates
that the cornering stiffness of the FE model at these loads is higher than the MF. Second,
for high positive slip angles the predicted force of the FE model is at most 3% lower than
the MF. Third, the predicted force at high negative slip angles is at most 10% lower than
the MF, while the difference between positive and negative slip angles is less obvious
for the MF. This is again a strong indication that the influence of contact pressure in
the friction model is slightly overestimated and the friction coefficient drops too fast for
increasing contact pressure, see Section 5.7. An overestimation of the influence of contact
pressure results in an overall lower lateral force than the MF and at negative slip angles
this effect is amplified further by the smaller size of the footprint with respect to a positive
slip angle. Nevertheless, the correlation is good for the entire range of slip angles and all
loads.
It can be seen in Figure 6.23 that also the self-aligning moment is accurately predicted, es-
pecially considering the variations between different datasets. The self-aligning moment,

8000
MF
6000 FEM

4000

2000
Fy [N]

−2000 0.4
−4000
0.8
−6000
1.2
−8000
−15 −10 −5 0 5 10 15
α [°]

Figure 6.22 / Lateral force as function of side slip angle at 0.4, 0.8 and 1.2 times the load
index, at a forward velocity of 60 km/h.
112 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

200

150 1.2

100 0.8

50
0.4
Mz [Nm]

−50

−100

−150 MF
FEM
−200
−15 −10 −5 0 5 10 15
α [°]

Figure 6.23 / Self-aligning moment as function of side slip angle at 0.4, 0.8 and 1.2 times
the load index, at a forward velocity of 60 km/h.

as given by (6.1), depends on the length of the pneumatic trial and the magnitude of the
lateral force. In the FE model the lateral force for negative slip angles is already lower
than the MF, which explains the faster decay at negative slip angles. The pneumatic trail
as function of slip angle is shown in Figure 6.24, where it can be seen that the pneumatic
trail of the FE model is just a few millimeters shorter than the smooth pneumatic trail
given by the MF, which lead to a smaller peak value of Mz in the FE model. Additionally,
when the lateral force is close to zero and changes sign in the FE model, the location of
the lateral force resultant is not well-defined. This causes the observed jump at −0.1◦ slip
angle.
The small differences in the pneumatic trail could be a result of the local friction model.
However conicity, which is not present in the FE model, has an effect on the measure-
ments. Furthermore, in the experiments a rolling resistance force is always present due
to the viscoelastic materials in the tire. Rolling resistance is not present in the current
model, which also leads to a different footprint with respect to footprints during actual
experiments.

Cornering and aligning stiffness

In the region for |α| < 2.5◦ the change in lateral force is governed by the cornering
stiffness

∂Fy
CF α = , (6.2)
∂α α=0
6.4 COMPARISON OF THE FE MODEL AND THE M AGIC F ORMULA 113

60

50

40
Pneumatic trail [mm]
30

20

10

−10

−20
MF
FEM
−30
−15 −10 −5 0 5 10 15
α [°]

Figure 6.24 / Pneumatic trail as function of side slip angle at 0.8 times the load index, at
a forward velocity of 60 km/h.

which is approximated by a linear least square error fit through data points for Fy between
−2.5◦ and 2.5◦ slip angle. Similarly, the change in self-aligning moment is governed by
the aligning stiffness

∂Mz
CM z = , (6.3)
∂α α=0

which is approximated by a linear least square error fit through data points for Mz be-
tween −2.5◦ and 2.5◦ slip angle. These stiffnesses are important parameters in determin-
ing the linear range behavior of vehicles. The computed cornering and aligning stiffness
together with the MF are shown in Figure 6.25. It can be seen that the peak of the cor-
nering stiffness for increasing load is not reached and therefore lies well above the tires
load index, which is good for handling. Furthermore the computed stiffness at 0.4 and
0.8 are in good agreement with the MF, the stiffness at 1.2 is 15% higher than the MF.
This indicates that the structural response of the FE model for this high load deviates
more with respect to the Magic Formula. Additionally the cornering stiffness of set 2
deviates from the other three, which could originate from different environmental condi-
tions, since this set has been measured at proving ground 2. The aligning stiffnesses of
the FE model are slightly underestimated with respect to the MF, which is to be expected
since the pneumatic trail in the FE model is shorter.
114 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

2000 120
Set 1

Aligning stiffness [Nm/deg]


Cornering stiffness [N/deg]

100 Set 2
1500 Set 3
80
Set 4
FE
1000 60
Set 1
Set 2 40
500 Set 3
Set 4 20
FE
0 0
0 0.5 1 1.5 0 0.5 1 1.5
Fz/L.I. [−] Fz/L.I. [−]

(a) (b)

Figure 6.25 / a) Cornering stiffness and b) aligning stiffness as function of normalized


load.

Forward velocity

The Magic Formula does not consider the forward velocity as input parameter, which
means that velocity induced effects are not taken into account. As already mentioned, the
magnitude of the lateral force eventually peaks. Beyond this peak value, increasing the
slip angle decreases the lateral force due to increasing slip velocity. This indicates that
there should be a velocity effect present and the lateral force at high slip angles should
be higher for lower velocities. The sliding velocities during cornering are however much
lower compared to braking, which suggests that the effect might not be as clear as in a
braking experiment.
To investigate this, a simulation and a measurement at proving ground 2, have been
performed with a forward velocity of 40 km/h. For this single measurement and also
a single measurement at 60 km/h, two MF datasets have been generated. These sets
should be used with care, since possible disturbances are also fitted, and are therefore
only used for illustrative purposes. It can be seen in Figure 6.26 that there is a small
effect visible both for the FE model and the MF. However, as expected, it is a very small
effect, which is only visible for very high slip angles. It is however captured in the FE
model. Furthermore this shows that, for pure cornering, tire testing at different velocities
might not be necessary, since the velocity influence is negligible compared to the force
deviations observed for the different datasets.
6.4 COMPARISON OF THE FE MODEL AND THE M AGIC F ORMULA 115

−3000 −3000
FE 60 km/h MF 60 km/h
−3500 FE 40 km/h −3500 MF 40 km/h
−4000 −4000
Fy [N]

Fy [N]
−4500 −4500

−5000 −5000

−5500 −5500

−6000 −6000
0 5 10 15 0 5 10 15
α [°] α [°]

(a) (b)

Figure 6.26 / Close-up of lateral force as function of slip angle at two forward velocities
for a) the FE model and b) the MF.

6.4.2 Combined slip

The validation for combined slip characteristics is presented for 4 slip angles for the sake
of visibility. Furthermore it is chosen to compare the FE model with sets 1 and 3, since
the braking experiments used for identification in Chapter 5 have also been performed
with new tires. The validation of the longitudinal forces under combined slip for two load
conditions are shown in Figures 6.27 and 6.28 respectively. It follows that the trend of
the FE model is a bit closer to set 1, but for both load conditions and all slip angles the
FE predictions represent the tire behavior really well. If the lateral forces are inspected,
see Figures 6.29 and 6.30, all FE predictions are within the bounds given by the two MF
sets. This shows that the FE model adequately predicts the forces under the complex
combined slip situation.

Self-aligning moment

The measurements of the self-aligning moment during combined slip fluctuate severely
and are less reliable. This is a well-known issue in tire testing and as a result it is hard
to fit the Magic Formula coefficients. Therefore the MF under combined slip is adapted
based on physical insights (Pacejka, 2006), but the self-aligning moment can still deviate
with respect to the different datasets due to these fluctuations in the measurements. As
a result the obtained self-aligning moments with the MF might not be representative of
the tested tire.
In Figure 6.31 and 6.32 the self-aligning moments for two of the MF datasets are com-
pared to the FE prediction. There are large differences between the two MF datasets,
which makes model validation, based on these results, difficult. Further investigation of
116 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

the exact behavior of the self-aligning moment under combined slip is therefore neces-
sary to judge to predicted moments of the FE model.

0
Set 1
−500 Set 2
FE
−1000

−1500
Fx [N]

−2000

+8°
−2500
°
−5
°
−3000 +3

−3500
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.27 / Longitudinal force as function of longitudinal slip for 4 slip angles at 0.4
times the load index, at a forward velocity of 60 km/h.

0
Set 1
−1000 Set 2
FE
−2000
Fx [N]

−3000

−4000
+8°
−5000
−5°
°
+3
−6000

−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.28 / Longitudinal force as function of longitudinal slip for 4 slip angles at 0.8
times the load index, at a forward velocity of 60 km/h.
6.4 COMPARISON OF THE FE MODEL AND THE M AGIC F ORMULA 117

3000
Set 1 −5°
Set 2
2000
FE

1000
°
0
Fy [N]

°
−1000 +3

−2000

+8°
−3000
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.29 / Lateral force as function of longitudinal slip for 4 slip angles at 0.4 times
the load index, at a forward velocity of 60 km/h.

6000
Set 1 −5°
Set 2
4000
FE

2000


Fy [N]

−2000 +3°

−4000

°
−6000
+8
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.30 / Lateral force as function of longitudinal slip for 4 slip angles at 0.8 times
the load index, at a forward velocity of 60 km/h.
118 6 P REDICTIVE CAPABILITY OF THE FE TIRE MODEL

60

°
40 +8
°
+3
20

Mz [Nm]

°
−20 −5
Set 1
−40
Set 2
FE
−60
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.31 / Self-aligning moment as function of longitudinal slip for 4 slip angles at
0.4 times the load index, at a forward velocity of 60 km/h.

150
+3°
100
+8°

50
Mz [Nm]


0

−50 Set 1
Set 2 −5°
FE
−100
−1 −0.8 −0.6 −0.4 −0.2 0
κ [−]

Figure 6.32 / Self-aligning moment as function of longitudinal slip for 4 slip angles at
0.8 times the load index, at a forward velocity of 60 km/h.
6.5 CONCLUSIONS 119

6.5 Conclusions
In this chapter the predictive capability of the FE model has been presented and a com-
parison with the Magic Formula has been made.
It has been shown that, for a pure cornering simulation, the magnitude of the lateral force
at positive slip angles is slightly larger than for negative slip angles. This is the result of
the non-symmetric tire tread in combination with the used friction model.
The effect of plysteer, which leads to offsets in the computed lateral force and self-aligning
moment characteristics, has been discussed. The overturning moment and approximated
loaded radius as function of slip angle are in good agreement with similar data presented
in literature. The combined slip characteristics have been presented for several slip angles
and the observed shape of the longitudinal and lateral forces as function of longitudinal
slip has been explained. It has been shown that the self-aligning moment under com-
bined slip is a combination of the location of both the longitudinal and the lateral force
resultant. These locations depend not only on the local friction model, but also on the
flexibility of the carcass.
The resemblance between computed shear power distribution, under different driving
conditions, and measured surface temperature clearly illustrate the potential of the FE
model. The shear power distribution can be used in future developments of the model to
incorporate temperature or wear effects.
It has been shown that an individual tire does not have a single well-defined set of force
and moment characteristics. Therefore four datasets have been measured and the devia-
tions between the four Magic Formula evaluations have been discussed.
A comparison has been made between predicted handling characteristics of the FE model
and the Magic Formula. For both the pure cornering and the complex combined slip situ-
ation, the forces and moments are accurately predicted with the current model. Observed
deviations with respect to the cornering characteristics, for negative slip angles, suggest
that the influence of contact pressure in the friction model is slightly overestimated.
Finally, it can be concluded that all three basic handling characteristics are adequately
predicted by using the identified friction model on a full scale tire.
120
C HAPTER SEVEN

Conclusions and recommendations

Abstract / In this chapter, the main conclusions of this thesis are summarized and recommenda-
tions for future work are presented.

7.1 Conclusions
In this thesis the finite element (FE) computation of steady-state handling characteristics
of rolling tires, under driving situations varying from normal driving to extreme manoeu-
vres, has been presented. The steady-state handling characteristics considered are pure
braking, pure cornering and combined slip and have been defined as follows. In case of
pure braking, the tire is braked from free-rolling up to wheel lock. For pure cornering,
the free-rolling tire is steered up to ±12◦ side slip angle and in case of combined slip, the
tire is braked up to wheel lock when rolling at a constant slip angle.
To compute these handling characteristics the Coulomb friction model is often used to
model the tire-road interaction problem, while it is clear from experiments that tire-road
interaction can not be captured accurately with this friction model. Additionally Coulomb
friction is not a feasible choice for the simulation of extreme manoeuvres. To overcome
these limitations an enhanced friction model for the tire-road interaction problem is pro-
posed. This leads to the main objective of this thesis:
To develop a robust and numerically efficient friction model for finite element tire sim-
ulations and to create a framework for the identification and implementation of friction
related parameters. The friction model should capture observed effects of dry friction and
it should be compatible with commercial FE codes.
The proposed friction model is directly dependent on contact pressure and sliding ve-
locity, while the global effect of temperature and surface roughness is captured in the
parameters of the friction model.

121
122 7 CONCLUSIONS AND RECOMMENDATIONS

A framework for the identification of the unknown parameters of this friction model has
been developed. This framework consists of a two step experimental / numerical ap-
proach in which the parameters of the contact pressure and the sliding velocity part are
treated separately and identified sequentially by combining small scale lab experiments
with full scale outdoor experiments.
Validation results of computed steady-state handling characteristics, which cover the en-
tire operating range, confirm the effectiveness of the developed friction model.
For the FE analysis a currently available state-of-the-art numerical method of the commer-
cial finite element package ABAQUS is used. The steady-state transport analysis of ABAQUS
is a computationally efficient method to obtain the global steady-state force and moment
characteristics of a tire under different driving conditions. However, this method has also
some limitations. The underlying road must be flat and fully coupled thermo-mechanical
simulations are not possible in the current implementation. The choice for the improved
friction model is based on an overview of friction models to describe the frictional re-
sponse of rubber, while taking the limitations of the numerical method into account.
A numerical and experimental analysis of the commercially available Laboratory Abra-
sion and skid Tester 100 has been presented, where measured hub forces have been used
to identify the local contact pressure dependent parameters of the tire friction model.
A specially developed test tire has been used for the identification of the slip velocity de-
pendent parameters. This tire has an asymmetric tread profile in order to study the effect
of the contact pressure distribution on the friction force. Furthermore, the corresponding
FE tire model has been used for the experimental validation.
The slip velocity dependent parameters have been identified using measurements of the
longitudinal force at the axle under different driving velocities. To exclude small vari-
ations in vertical load during experiments, the semi-empirical Magic Formula model,
fitted on the experimental data, is used to evaluate the longitudinal slip characteristics at
a constant vertical load. It has been shown that the computed steady-state longitudinal
forces are in good quantitative agreement with experiments for all loads.
Based on these results, it is concluded that the proposed identification framework, where
the parameters are identified using measured axle forces and implemented locally at each
node in contact, is a suitable approach to obtain the parameters of the friction model.
The predictive capability of the tire model in combination with the fully identified fric-
tion model has been assessed. A comparison has been made between predicted handling
characteristics of the finite element model and experimental data for the pure cornering
and combined slip range. For both the pure cornering and the complex combined slip
situation, the forces and moments are accurately predicted with the current model. For
negative slip angles, observed deviations with respect to the cornering characteristics sug-
gest that the influence of contact pressure in the friction model is slightly overestimated.
Nevertheless, the presented comparison shows that all three basic handling characteris-
tics are well predicted by using the identified friction model on a full scale tire.
7.2 R ECOMMENDATIONS 123

7.2 Recommendations
In this final section, directions that can be pursued in future work are discussed.
The observed deviations with respect to the cornering characteristics for negative slip an-
gles and the simulations with different inflation pressure, are most likely caused by an
overestimation of the influence of contact pressure in the friction model, where the con-
tact pressure is extrapolated outside the measured range. Therefore, it is recommended
that the followed identification method using the Laboratory Abrasion and skid Tester
100 is investigated further. Alternatively, specially designed lab scale setups, such as
used by Huemer et al. (2001b) to obtain a broader range of contact pressures could also
be developed.
Within the presented numerical modeling framework, several extensions can be made.
The current friction model does not depend on temperature. The temperature during
frictional sliding of rubber is not constant. Therefore, it is recommended to incorporate
temperature effects in the FE model based on the shear power intensity. Different strate-
gies can be followed here. Giessler et al. (2010) uses the friction heat only to predict tread
temperature increase and the effect on tire traction. The friction model of Huemer et al.
(2001a) incorporates temperature effects in the material properties using the WLF trans-
formation. Hofstetter et al. (2006) uses a fully coupled thermo-mechanical FE model to
convert a part of frictional power into heat, which increases the tread temperature. This
temperature increase also results in changing viscoelastic material properties. This last
method seems the most promising, although both methods require an accurate descrip-
tion of the viscoelastic material properties of the tread compound.
The handling characteristics considered in this thesis are based on dry roads, but it is
not clear what the exact contribution of surface texture on the frictional force is. A first
step in this direction in the context of the current project has been made by Hunnekes
(2008), but further research in this area is necessary. Furthermore, the application of the
presented two step experimental / numerical approach under wet conditions should be
assessed.
In this thesis, the friction model is combined with an existing FE tire model. Other
components of the FE model, such as material behavior and mesh design, contribute to
the structural response of the model as well. Hence, these components should also be
improved in order to further enhance the prediction capability of the entire model.
Real virtual prototyping has not been achieved yet. A full scale tire has been manufac-
tured to identify the parameters of the friction model. A first step towards actual virtual
prototyping should be to use the fully identified friction model on a different type of tire,
e.g. with other size, aspect ratio or structural properties, with the same tread compound.
In this manner the predictive capability of the friction model can be further assessed. In
a second step, a more complicated tread pattern can be used, although the accuracy of the
124 7 CONCLUSIONS AND RECOMMENDATIONS

solution of the ALE method should be carefully checked (Qi et al., 2007).
Once these steps have been successfully completed, the possibility to use FE tire mod-
els in vehicle dynamic simulations should be investigated. Although the computational
effort is too large for vehicle dynamic simulations, it might be possible to fit a Magic For-
mula model on a combination of predicted handling characteristics and less experimental
data. In this way the required low computational cost for a full vehicle simulation is still
maintained.
As discussed by Pottinger (Gent and Walter, 2005, chapter 8), an individual tire does not
have a single well-defined set of force and moment characteristics. This uncertainty in the
characteristics arises from the changing chemical and viscoelastic properties of the tire
materials combined with effects of wear. Furthermore, due to manufacturing variations
other effects, such as conicity, are introduced in real tires. This is one of the reasons why
in this thesis the different Magic Formula sets obtained on tires from one batch deviate
with respect to each other, even when the environmental conditions are similar. A well-
defined upper and lower bound around a MF dataset should be developed. On the FE
model side, parametric studies with respect to manufacturing variations should be carried
out to obtain an upper and lower bound around the predicted handling characteristics.
Furthermore, in the MF-Tool software the parameters of the MF are chosen such that the
error between the MF and experiments is minimized. As a result, it can happen that the
longitudinal force at κ = 0 is positive. This makes a quantitative comparison with the FE
model prediction very difficult, since a small difference in longitudinal slip around zero
results in a large longitudinal force difference. A well-defined error criterium should be
developed.
Currently, measurements for validation purposes of FE models are still necessary. Ob-
taining good measurements of the forces and moments or other system quantities acting
on rolling tires is a difficult task. Direct measurements of the stress distribution in the
contact area, with a high spatial resolution, during driving are preferred for validation of
the local friction model. However, this is not yet possible. Therefore, the forces and mo-
ments acting on the hub are used for validation in this thesis, although the measurements
of moments, during combined slip situations, fluctuate severely and are less reliable. For
validation of a local friction model, accurate measurements of the moments are neces-
sary. Even if the force magnitude is correctly predicted, the local stress distribution in the
contact area determines the resultant force location and hence the magnitude of the mo-
ment. Therefore, further research into more advanced measuring techniques, preferably
directly in the contact area, is still required.
References

B AKKER , E., N YBORG , L., and P ACEJKA , H.B. (1987). Tyre modelling for use in vehicle dynamics studies.
SAE paper 870421.

B ANDEIRA , A.A., W RIGGERS , P., and DE M ATTOS P IMENTA , P. (2004). Numerical derivation of contact
mechanics interface laws using a finite element approach for large 3D deformation. International Journal
for numerical methods in Engineering, 59, 173–195.

B ARQUINS , M. (1992). Adherence, friction and wear of rubber-like materials. Wear, 158, 87–117.

B ERGER , H.R. and H EINRICH , G. (2000). Friction Effects in the Contact Area of Sliding Rubber: a
Generalized Schallamach Model. Kautschuk Gummi Kunststoffe, 53, 200–205.

B LUME , H., H EIMANN , B., L INDNER , M., and V OLK , H. (2003). Friction measurement on road surfaces.
Kautschuk Gummi Kunststoffe, 56, 677–681.

B OWDEN , F.P. and TABOR , D. (1964). The friction and lubrication of solids, part II. Oxford University Press,
Oxford Classics Series edition.

VAN B REEMEN , L.C.A. (2009). Contact mechanics in glassy polymers. Ph.D. thesis, Eindhoven University of
Technology.

B ROEZE , J. (2009). A semi-empirical approach on the contact dynamics of a small solid wheel. Master’s thesis,
Eindhoven University of Technology, Department of Mechanical Engineering, Dynamics and Control
group, DCT 2009.46.

C HO , J.R., K IM , K.W., Y OO , W.S., and W ONG , S.I. (2004). Mesh generation considering detailed tread
blocks for reliable 3D tire analysis. Advances in Engineering Software, 35, 105–113.

C HO , J.R., K IM , K.W., J EON , D.H., and Y OO , W.S. (2005). Transient dynamic response analysis of 3-D
patterned tire rolling over cleat. European Journal of Mechanics A/Solids, 24, 519–531.

D ANIELSON , K.T., N OOR , A.K., and G REEN , J.S. (1996). Computational strategies for tire modeling and
analysis. Computers & Structures, 61(4), 673–693.

D ARNELL , I., M OUSSEAU , R., and H ULBERT, G. (2002). Analysis of Tire Force and Moment Response
During Side Slip Using an Efficient Finite Element Model. Tire Science and Technology, 30(2), 66–82.

D ORSCH , V., B ECKER , A., and V OSSEN , L. (2002). Enhanced rubber friction model for finite element
simulations of rolling tyres. Plastics, Rubber and Composites, 31(10), 458–464.

125
126 R EFERENCES

D UNLOP (2010). http://dunloptires.com/about.

E UROPEAN T YRE AND R IM T ECHNICAL O RGANISATION (2010). http://www.etrto.org.

F ARIA , L.O., B ASS , J.M., O DEN , J.T., and B ECKER , E.B. (1989). A Three-Dimensional Rolling Contact
Model for a Reinforced Rubber Tire. Tire Science and Technology, 17(3), 217–233.

G ÄBEL , G. and K RÖGER , M. (2006). Non-linear contact stiffness in tyre-road interaction. Tampere, Fin-
land. Euronoise.

G ALL , R., TABADDOR , F., R OBBINS , D., M AJORS , P., S HEPERD , W., and J OHNSON , S. (1995). Some
Notes on the Finite Element Analysis of Tires. Tire Science and Technology, 23(3), 175–188.

G ARRO , L., G URNARI , G., N ICOLETTO , G., and S ERRA , A. (1999). A Laboratory Device to Measure
the Interfacial Phenomena between Rubber and Rough Surfaces. Tire Science and Technology, 27(4),
206–226.

G ENT, A.N. (2007). Mechanical Properties of Rubber. Presentation, 32th Tire Mechanics Short Course,
Köln.

G ENT, A.N. and W ALTER , J.D. (2005). The Pneumatic Tire. National Highway Traffic Safety Administra-
tion.

G HOREISHY, M.H.R. (2008). A state of the art review of the finite element modelling of rolling tyres.
Iranian Polymer Journal, 17(8), 571–597.

G IESSLER , M., G AUTERIN , F., W IESE , K., and W IES , B. (2010). Influence of Friction Heat on Tire
Traction on Ice and Snow. Tire Science and Technology, 38(1), 4–23.

G ROSCH , K.A. (1963). The relation between the friction and visco-elastic properties of rubber. Proceedings
of the Royal Society of London, Series A, 274(1356), 21–39.

G ROSCH , K.A. (2009). Private communication.

H ARALDSSON , A. and W RIGGERS , P. (2000). A strategy for numerical testing of frictional laws with
application to contact between soil and concrete. Computer Methods in Applied Mechanics and Engineering,
190, 963–977.

H OFSTETTER , K., E BERHARDSTEINER , J., and M ANG , H.A. (2003). A Thermo-Mechanical Formulation
Describing the Frictional Behavior of Rubber. Proceedings in Applied Mathematics and Mechanics, 2, 238–
239.

H OFSTETTER , K., G ROHS , C H ., E BERHARDSTEINER , J., and M ANG , H.A. (2006). Sliding behaviour of
simplified tire tread patterns investigated by means of FEM. Computers & Structures, 84, 1151–1163.

H UEMER , T., L IU , W.N., E BERHARDSTEINER , J., and M ANG , H.A. (2001a). A 3D finite element formula-
tion describing the frictional behavior of rubber on ice and concrete surfaces. Engineering Computations,
18(3/4), 417–436.

H UEMER , T., L IU , W.N., E BERHARDSTEINER , J., M ANG , H.A., and M ESCHKE , G. (2001b). Sliding
Behavior of Rubber on Snow and Concrete Surfaces. Kautschuk Gummi Kunststoffe, 54, 458–462.

H UNNEKES , B.G.B. (2008). Tire road frictional interaction, physical approach. Technical Report, Eind-
hoven University of Technology, Department of Mechanical Engineering, Dynamics and Control group,
DCT 2008.088.
R EFERENCES 127

J EONG , K.M., K IM , K.W., B EOM , H.G., and P ARK , J.U. (2007). Finite Element Analysis of Nonunifor-
mity of Tires with Imperfections. Tire Science and Technology, 35(3), 226–238.

K ABE , K. and K OISHI , M. (2000). Tire Cornering Simulation Using Finite Element Analysis. Journal of
Applied Polymer Science, 78, 1566–1572.

K ENNEDY, R. and P ADOVAN , J. (1987). Finite element analysis of steady and transiently moving/rolling
nonlinear viscoelastic structure-ii. Shell and three-dimensional simulations. Computers & Structures,
27(2), 259–273.

K ERCHMAN , V. (2008). Tire-Suspension-Chassis Dynamics in Rolling over Obstacles for Ride and Harsh-
ness Analysis. Tire Science and Technology, 36(3), 158–191.

K LÜPPEL , M. and H EINRICH , G. (2000). Rubber friction on self-affine road tracks. Rubber chemistry and
technology, 73, 578–606.

K OISHI , M., K ABE , K., and S HIRATORI , M. (1998). Tire Cornering Simulation Using an Explicit Finite
Element Analysis Code. Tire Science and Technology, 26(2), 109–119.

K OUZNETSOVA , V. (2006). Non-linear Finite Element Method for Solids. Lecture notes, Eindhoven Uni-
versity of Technology.

L AURSEN , T.A. and S IMO , J.C. (1993a). Algorithmic symmetrization of Coulomb frictional problems
using augmented Lagrangians. Computer Methods in Applied Mechanics and Engineering, 108, 133–146.

L AURSEN , T.A. and S IMO , J.C. (1993b). A continuum-based finite element formulation for the implicit
solution of multibody, large deformation frictional contact problems. International Journal for numerical
methods in Engineering, 36, 3451–3485.

L AURSEN , T.A. and S TANCIULESCU , I. (2006). An algorithm for incorporation of frictional sliding condi-
tions within a steady state rolling framework. Communications in numerical methods in engineering, 22,
301–318.

L INDNER , M. (2005). Experimentelle und theoretische Untersuchungen zur Gummireibung an Profilklötzen


und Dichtungen. Ph.D. thesis, Universität Hannover.

L UPKER , H., C HELI , F., B RAGHIN , F., G ELOSA , E., and K ECKMAN , A. (2004). Numerical Prediction of
Car Tire Wear. Tire Science and Technology, 32(3), 164–186.

M ACKERLE , J. (1998). Rubber and rubber-like materials, finite-element analyses and simulations: a bibli-
ography (1976–1997). Modeling and Simulation in Materials Science and Engineering, 6, 171–198.

M ACKERLE , J. (2004). Rubber and rubber-like materials, finite-element analyses and simulations, an
addendum: a bibliography (1997–2003). Modeling and Simulation in Materials Science and Engineering,
12, 1031–1053.

M ESCHKE , G., P AYER , H.J., and M ANG , H.A. (1997). 3D Simulations of Automobile Tires: Material
Modeling, Mesh Generation, and Solution Strategies. Tire Science and Technology, 25(3), 154–176.

M ONTGOMERY, D.C. and R UNGER , G.C. (1999). Applied statistics and probability for engineers. John Wiley
& Sons.

M OORE , D.F. (1972). The friction and lubrication of elastomers. International series of monographs on materials
science and technology. Pergamon Press.
128 R EFERENCES

M ULLINS , L. and T OBIN , N.R. (1957). Theoretical model for the elastic behavior of filler-reinforced vul-
canized rubbers. Rubber Chemistry and Technology, 30, 551–571.

M UNDL , R., F ISCHER , M., S TRACHE , W., W IESE , K., W IES , B., and Z INKEN , K.H. (2008). Virtual
Pattern Optimization Based on Performance Prediction Tools. Tire Science and Technology, 36(3), 192–
210.

N ACKENHORST, U. (2000). Rollkontaktdynamik - Numerische Analyse der Dynamik rollender Körper mit
der Finite Elemente Methode. Technical Report, Institut für Mechanik, Universität der Bundeswehr,
Hamburg.

N ACKENHORST, U. (2004). The ALE-formulation of bodies in rolling contact theoretical foundations and
finite element approach. Computer Methods in Applied Mechanics and Engineering, 193, 4299–4322.

N AKAJIMA , Y. and P ADOVAN , J. (1987). Finite element analysis of steady and transiently moving/rolling
nonlinear viscoelastic structure-iii. Impact/contact simulations. Computers & Structures, 27(2), 275–286.

N OOR , A.K. and TANNER , J.A. (1985). Advances and trends in the development of computational models
for tires. Computers & Structures, 20(1/3), 517–533.

O DEN , J.T. and L IN , T.L. (1986). On the general rolling contact problem for finite deformations of a
viscoelastic cylinder. Computer Methods in Applied Mechanics and Engineering, 57, 297–367.

O DEN , J.T. and M ARTINS , J.A.C. (1985). Models and computational methods for dynamic friction phe-
nomena. Computer Methods in Applied Mechanics and Engineering, 52, 527–634.

O DEN , J.T. and P IRES , E.B. (1983). Numerical analysis of certain contact problems in elasticity with non-
classical friction laws. Computers & Structures, 16(1/4), 481–485.

O DEN , J.T. and P IRES , E.B. (1984). Algorithms and numerical element approximations of results for finite
contact problems with non-classical friction laws. Computers & Structures, 19(1/2), 137–147.

O DEN , J.T., L IN , T.L., and B ASS , J.M. (1988). A Finite Element Analysis of the General Rolling Contact
Problem for a Viscoelastic Rubber Cylinder. Tire Science and Technology, 16(1), 18–43.

O HISHI , K., S UITA , H., and I SHIHARA , K. (2002). The Finite Element Approach to Predict the Plysteer
Residual Cornering Force of Tires. Tire Science and Technology, 30(2), 122–133.

O LATUNBOSUN , O.A. and B OLARINWA , O. (2004). FE Simulation of the Effect of Tire Design Parameters
on Lateral Forces and Moments. Tire Science and Technology, 32(3), 146–163.

O LATUNBOSUN , O.A. and B URKE , A.M. (2002). Finite Element Modelling of Rotating Tires in the Time
Domain. Tire Science and Technology, 30(1), 19–33.

O LSSON , H., Å STRÖM , K.J., C ANUDAS DE W IT , C., G ÄFVERT, M., and L ISCHINSKY, P. (1998). Friction
Models and Friction Compensation. European Journal of Control, 11, 16–52.

P ACEJKA , H.B. (2006). Tyre and vehicle dynamics. Butterworth-Heinemann, 2nd edition.

P ADOVAN , J. (1987). Finite element analysis of steady and transiently moving/rolling nonlinear viscoelastic
structure-i. Theory. Computers & Structures, 27(2), 249–257.

P ADOVAN , J., K AZEMPOUR , A., TABADDOR , F., and B ROCKMAN , B. (1992). Alternative formulations of
rolling contact problems. Finite Elements in Analysis and Design, 11, 275–284.
R EFERENCES 129

P AYNE , A.R. (1963). Dynamic properties of heat-treated butyl vulcanizates. Journal of Applied Polymer
Science, 7(3), 873–885.

P ERSSON , B.N.J. (1993). Theory and simulation of sliding friction. Physical Review Letters, 71(8), 1212–1215.

P ERSSON , B.N.J. (1995). Theory of friction: Stress domains, relaxation, and creep. Physical Review B,
51(19), 568–585.

P ERSSON , B.N.J. (1998). On the theory of rubber friction. Surface Science, 401, 445–454.

P ERSSON , B.N.J. (1999). Sliding friction. Surface Science Reports, 33, 83–119.

P ERSSON , B.N.J. (2001). Theory of rubber friction and contact mechanics. Journal of Chemical Physics,
115(8), 3840–3861.

P ERSSON , B.N.J. (2002). Adhesion between Elastic Bodies with Randomly Rough Surfaces. Physical Review
Letters, 89(24).

P ERSSON , B.N.J. (2006). Rubber friction: role of the flash temperature. Journal of Physics: Condensed
matter, 18, 7789–7823.

P ERSSON , B.N.J. and T OSATTI , E. (2000). Qualitative theory of rubber friction and wear. Journal of
Chemical Physics, 112(4), 2021–2029.

P ERSSON , B.N.J. and V OLOKITIN , A.I. (2000). Dynamical interactions in sliding friction. Surface Science,
457, 345–356.

P ERSSON , B.N.J. and V OLOKITIN , A.I. (2002). Theory of rubber friction: Nonstationary sliding. Physical
Review B, 65.

P ERSSON , B.N.J., B UCHER , F., and C HIAIA , B. (2002). Elastic contact between randomly rough surfaces:
Comparison of theory with numerical results. Physical Review B, 65.

P OTTINGER , M.G. and M C I NTYRE , J.E. (1999). Effect of Suspension Alignment and Modest Cornering
on the Footprint Behavior of Performance Tires and Heavy Duty Radial Tires. Tire Science and Technology,
27(3), 128–160.

Q I , J., H ERRON , J.R., S ANSALONE , K.H., M ARS , W.V., D U , Z.Z., S NYMAN , M., and S URENDRANATH ,
H. (2007). Validation of a Steady-State Transport Analysis for Rolling Treaded Tires. Tire Science and
Technology, 35(3), 183–208.

R AO , K., K UMAR , R., and B OHARA , P. (2003). Transient Finite Element Analysis of Tire Dynamic
Behavior. Tire Science and Technology, 31(2), 104–127.

R AUH , J. and M ÖSSNER -B EIGEL , M. (2008). Tyre simulation challenges. Vehicle System Dynamics, 46(1),
49–62.

R ODGERS , M. B. (2001). Rubber Tires. In Encyclopedia of Materials: Science and Technology, 8237 – 8242.
Elsevier, Oxford.

SIMULIA (2000). Tire modeling using abaqus.

SIMULIA (2009a). Abaqus Analysis user’s manual, version 6.9.

SIMULIA (2009b). Abaqus Example problems manual, version 6.9.


130 R EFERENCES

SIMULIA (2009c). Abaqus Theory manual, version 6.9.

S AVKOOR , A.R. (1966). Some aspects of friction and wear of tyres arising from deformations, slip and
stresses at the ground contact. Wear, 9, 66–78.

S AVKOOR , A.R. (1987). Dry adhesive friction of elastomers. Ph.D. thesis, Technische Universiteit Delft.

S CHALLAMACH , A. (1952). The load dependence of rubber friction. The proceedings of the physical society,
65(9), 657–661.

S CHALLAMACH , A. (1971). How does rubber slide? Wear, 17, 301–312.

S IMO , J.C. and L AURSEN , T.A. (1992). An augmented Lagrangian treatment of contact problems involving
friction. Computers & Structures, 42(1), 97–116.

S MITH , K.R., K ENNEDY, R.H., and K NISLEY, S.B. (2008). Prediction of Tire Profile Wear by Steady-State
FEM. Tire Science and Technology, 36(4), 290–303.

VAN DER S TEEN , R., L OPEZ , I., S CHMEITZ , A.J.C., DE B RUIJN , B., and N IJMEIJER , H. (2010a). Ex-
perimental and numerical study of friction and braking characteristics of rolling tires. Tire Science and
Technology, Submitted for publication.

VAN DER S TEEN , R., L OPEZ , I., and N IJMEIJER , H. (2010b). Experimental and numerical study of friction
and stiffness characteristics of small rolling tires. Tire Science and Technology, Accepted for publication.

S TRIBECK , R. (1902). Die wesentlischen Eigenschaften der Gleit- und Rollenlager. Zeitschrift des Vereines
Deutscher Ingenieure, 46(37/38/39), 1341–1348 (pt I), 1432–438 (pt II), 1463–1470 (pt III).

TNO (2008). Measurement requirements and TYDEX file generation for MF-Tyre/MF-Swift 6.1.

TNO (2010). MF-Tool.

T HOMSON , R.W. (1847). Improvement in carriage wheels, U.S. Patent No. 5,104.

T ÖNÜK , E. and Ü NLÜSOY, Y.S. (2001). Prediction of automobile tire cornering force characteristics by
finite element modeling and analysis. Computers & Structures, 79, 1219–1232.

T RINKO , M. (2007). Tire stresses and deformation analysis. Presentation, 32th Tire Mechanics Short
Course, Köln.

VMI H OLLAND BV (2009). LAT100, www.vmi-group.com/tire/compound-testing.

V ERMOND , J.G. (2008). Experimental determination of the CCAR tyre characteristics. Technical Report,
Eindhoven University of Technology, Department of Mechanical Engineering, Dynamics and Control
group, DCT 2008.113.

W ALTER , J.D. (2007a). Rules and regulations governing tires. Presentation, 32th Tire Mechanics Short
Course, Köln.

W ALTER , J.D. (2007b). Tire materials and manufacturing. Presentation, 32th Tire Mechanics Short
Course, Köln.

W ESTERMANN , S., P ETRY, F., B OES , R., and T HIELEN , G. (2004). Experimental Investigations Into the
Predictive Capabilities of Current Physical Rubber Friction Theories. Kautschuk Gummi Kunststoffe, 57,
645–650.
R EFERENCES 131

W ILLIAMS , M.L., L ANDEL , R.F., and F ERRY, J.D. (1955). The temperature dependence of relaxation
mechanisms in Amorphous polymers and other Glass-forming liquids. Journal of the American Chemical
Society, 77, 3701–3707.

W RIGGERS , P. (2006). Computational contact mechanics. Springer, Second edition.

W RIGGERS , P. and R EINELT, J. (2009). Multi-scale approach for frictional contact of elastomers on rough
rigid surfaces. Computer Methods in Applied Mechanics and Engineering, 198, 1996–2008.

W RIGGERS , P., VAN , T. V U, and S TEIN , E. (1990). Finite element formulation of large deformation
impact-contact problems with friction. Computers & Structures, 37(3), 319–331.

YAMAZAKI , S., YAMAGUCHI , M., H IROKI , E., and S UZUKI , T. (2000). Effects of the Number of Sip-
ing Edges in a Tire Tread Block on Friction Property and Contact With an Icy Road. Tire Science and
Technology, 28(1), 58–69.

Z AVARISE , G., W RIGGERS , P., S TEIN , E., and S CHREFLER , B.A. (1992). Real contact mechanisms and
finite element formulation – A coupled thermomechanical approach. International Journal for numerical
methods in Engineering, 35, 767–785.

Z EID , I. and P ADOVAN , J. (1981). Finite element modeling of rolling contact. Computers & Structures,
14(1/2), 163–170.

Z IEFLE , M. and N ACKENHORST, U. (2005). A new update procedure for internal variables in an ALE-
description of rolling contact. Proceedings in Applied Mathematics and Mechanics, 5, 71–74.
132
A PPENDIX A

Friction model implementation

In this appendix the full set of equations of the friction model is summarized.
The friction law is described by
 −k 
kγ̇kj
  
pj 2 2
µj (pj , γ̇ j ) = µs + (µm − µs ) exp −h log , (A.1)
p0 Vmax
where pj and kγ̇kj are the normal contact pressure and equivalent slip velocity at node
j respectively. The parameters p0 and k are related to the contact pressure and the
parameters µs , µm , h and Vmax are related to the sliding velocity.

The parameter R, in the interval [0 1], is used to model frictionless contact (R = 0)


during the static steps and frictional contact (R = 1) during the steady-state rolling steps.
An intermediate steady-state rolling step is used to activate the friction model. In this
step the parameter varies linearly from zero to one, which prevent convergence problems
during the transition from frictionless to frictional contact.

With the user-defined critical slip velocity γ̇crit the slope


µ(p, γ̇)p
ks = . (A.2)
γ̇crit
is calculated.

A.1 Stick
The frictional stress in stick is given by
τi = ks γ̇i . (A.3)

133
134 A F RICTION MODEL IMPLEMENTATION

The partial derivatives for the linearized incremental friction shear stress and contact
pressure are derived using Maple, version 12.0 and are given by

∂∆τ1
=
∂∆γ̇1
 −k ! ( !)
γ̇12 + γ̇22 γ̇12 + γ̇22
p p
−2h2 Rpγ̇12 p
(µm − µs ) log exp −h2 log2
γ̇12 + γ̇22 γ̇crit p0

Vmax Vmax
 −k " ( p
2 + γ̇ 2
!)#
Rp p γ̇
+ µs + (µm − µs ) exp −h2 log2 1 2
(A.4)
γ̇crit p0 Vmax

∂∆τ1
=
∂∆γ̇2
−k ! ( !)
γ̇12 + γ̇22 γ̇12 + γ̇22
p p
−2h2 Rpγ̇1 γ̇2

p
(µm − µs ) log exp −h2 log2 (A.5)
γ̇12 + γ̇22 γ̇crit

p0 Vmax Vmax

∂∆τ1
=
∂∆p
−k " ( !)#
γ̇12 + γ̇22
 p
p γ̇1
(1 − k)R µs + (µm − µs ) exp −h log 2 2
(A.6)
p0 Vmax γ̇crit

∂∆τ2
=
∂∆γ̇1
−k ! ( !)
γ̇12 + γ̇22 γ̇12 + γ̇22
p p
−2h2 Rpγ̇1 γ̇2

p
(µm − µs ) log exp −h2 log2 (A.7)
γ̇12 + γ̇22 γ̇crit

p0 Vmax Vmax

∂∆τ2
=
∂∆γ̇2
 −k ! ( !)
γ̇12 + γ̇22 γ̇12 + γ̇22
p p
−2h2 Rpγ̇22 p
(µm − µs ) log exp −h2 log2
γ̇12 + γ̇22 γ̇crit p0

Vmax Vmax
 −k " ( !)#
γ̇12 + γ̇22
p
Rp p
+ µs + (µm − µs ) exp −h log
2 2
(A.8)
γ̇crit p0 Vmax

∂∆τ2
=
∂∆p
−k " ( !)#
γ̇12 + γ̇22
 p
p γ̇2
(1 − k)R µs + (µm − µs ) exp −h2 log2 (A.9)
p0 Vmax γ̇crit
A.2 S LIP 135

A.2 Slip
The frictional stress for slip is given by
γ̇i
τi = µ(p, γ̇)p . (A.10)
γ̇eqv
The partial derivatives for the linearized incremental friction shear stress and contact
pressure in slip are given by

∂∆τ1
=
∂∆γ̇1
 −k " ( p
2 + γ̇ 2
!)#
Rp p γ̇
µs + (µm − µs ) exp −h2 log2 1 2
γ̇12 + γ̇22
p
p0 Vmax
 −k " ( !)#
γ̇12 + γ̇22
p
Rpγ̇12 p
− 3/2 µs + (µm − µs ) exp −h log 2 2
(A.11)
γ̇12 + γ̇22 p0 Vmax
 −k ! ( !)
2h2 Rpγ̇12 γ̇12 + γ̇22 2 + γ̇ 2
p p
p γ̇
− 3/2 (µm − µs ) log exp −h2 log2 1 2
γ̇12 + γ̇22 p0 Vmax Vmax

∂∆τ1
=
∂∆γ̇2
−k " ( !)#
γ̇12 + γ̇22
p
−Rpγ̇1 γ̇2

p
3/2 µs + (µm − µs ) exp −h2 log2 (A.12)
γ̇12 + γ̇22 p0 Vmax
−k ! ( !)
2h2 Rpγ̇1 γ̇2 γ̇12 + γ̇22 2 + γ̇ 2
 p p
p γ̇
− 3/2 (µm − µs ) log exp −h2 log2 1 2
γ̇12 + γ̇22 p0 Vmax Vmax

∂∆τ1
=
∂∆p
−k " ( !)#
γ̇12 + γ̇22
 p
p γ̇
(1 − k)R µs + (µm − µs ) exp −h log 2 2
p 1 (A.13)
p0 Vmax γ̇1 + γ̇22
2

∂∆τ2
=
∂∆γ̇1
−k " ( !)#
γ̇12 + γ̇22
p
−Rpγ̇1 γ̇2

p
3/2 µs + (µm − µs ) exp −h2 log2 (A.14)
γ̇12 + γ̇22 p0 Vmax
−k ! ( !)
2h2 Rpγ̇1 γ̇2 γ̇12 + γ̇22 γ̇12 + γ̇22
 p p
p
− 3/2 (µm − µs ) log exp −h log
2 2
γ̇12 + γ̇22 p0 Vmax Vmax
136 A F RICTION MODEL IMPLEMENTATION

∂∆τ2
=
∂∆γ̇2
 −k " ( p
2 + γ̇ 2
!)#
Rp p γ̇
µs + (µm − µs ) exp −h2 log2 1 2
γ̇12 + γ̇22
p
p0 Vmax
 −k " ( p
2 + γ̇ 2
!)#
Rpγ̇22 p γ̇
− 3/2 µs + (µm − µs ) exp −h2 log2 1 2
(A.15)
γ̇12 + γ̇22 p0 Vmax
 −k ! ( !)
2h2 Rpγ̇22 γ̇12 + γ̇22 γ̇12 + γ̇22
p p
p
− 3/2 (µm − µs ) log exp −h log 2 2
γ̇12 + γ̇22 p0 Vmax Vmax

∂∆τ2
=
∂∆p
−k " ( !)#
γ̇12 + γ̇22
 p
p γ̇
(1 − k)R µs + (µm − µs ) exp −h2 log2 p 2 (A.16)
p0 Vmax γ̇12 + γ̇22
A PPENDIX B

Mesh effect on the force equilibrium in


vertical direction

In a steady-state rolling step, the centrifugal force creates an additional, radially oriented,
force when inertia effects are taken into account. This is an internal force and the netto
effect should be zero. However, when a non-uniform mesh in circumferential direction
is used the center of mass does not coincide with the rotation axis of the tire. Because the
mesh is fixed in space in the ALE method, this offset creates a small numerically induced
force in the direction of the denser mesh when the tire spins with rotational velocity ω.
To illustrate this effect two simulations are carried out. In these simulations only a rota-
tional velocity profile is prescribed, while the axle of the tire is fixed, as shown in Figure
B.1. In the first simulation a uniform mesh is used and in the second simulation the

z
x ω

Figure B.1 / Exaggerated view of the influence of centrifugal force on the tire, when a
rotational velocity profile is prescribed and the axle of the tire is fixed.

mesh as shown in Figure 3.10 is used. The reaction forces in vertical direction of both
simulations are shown in Figure B.2. The reaction force in vertical direction is zero for

137
138 B M ESH EFFECT ON THE FORCE EQUILIBRIUM IN VERTICAL DIRECTION

the uniform mesh, while for the nonuniform mesh a force proportional to ω 2 in the neg-
ative z direction is observed. This demonstrates that a pure rotational velocity on the tire,
without a forward velocity, generates a small numerically induced force in the direction
of the denser mesh.
Reaction force axle [N] Reaction force axle [N]

0.5

−0.5

−1
0 10 20 30 40 50 60 70 80 90

150

100

50

0
0 10 20 30 40 50 60 70 80 90
ω [rad/s]

Figure B.2 / Reaction forces on the axle for uniform mesh (top) and reaction forces on
the axle for non-uniform mesh (bottom) as function of rotational velocity.

When the road is added and contact is present between the road and the tire, the gener-
ated numerically induced force contributes to the overall force balance between the wheel
and road reference node. In an FE analysis it is equivalent to prescribe the vertical force in
upward direction on the reference node of the road and fixate the axle reference node or to
prescribe the vertical force in downward direction on the reference node of the wheel and
fixate the reference node of the road. Two additional simulations with the non-uniform
mesh are carried out, in which the road is added and the braking characteristic of the tire
is computed for a forward velocity of 100 km/h. In the first simulation the road is loaded
with Fz = 5000 N and in the second simulation the wheel is loaded on the axle with a
force Fz of −5000 N. In Figure B.3 the reaction forces of both simulations are shown,
which confirms the decrease in reaction force on the axle and the increase of reaction
force on the road.
To avoid this force a uniform mesh should be used. However, then the ALE method loses
its main advantage. Furthermore, to obtain a sufficient dense mesh in the contact area, a
very large number of sectors is required which is computationally not feasible. It can be
seen in Figure B.3 that this force is small with respect to the applied load on the tire and
as such hardly affects the computed longitudinal and lateral forces. Therefore, inertia
139
Reaction force road [N] Reaction force axle [N]

−4850

−4900

−4950

−5000
0 10 20 30 40 50 60 70 80 90

5150

5100

5050

5000
0 10 20 30 40 50 60 70 80 90
ω [rad/s]

Figure B.3 / Close-up of the longitudinal force on the axle as function of rotational velocity
for both the road load and the axle load at a constant forward velocity of 100 km/h.

effects are included and since the load on the wheel is kept constant during experiments,
for all forward velocities, the wheel is loaded in the FE model to keep the force on the axle
constant.
140
A PPENDIX C

Computation of slip velocity field for the


LAT 100 setup

The slip velocity field is not always oriented with the global axes system in ABAQUS. This
depends on the used description for the rigid surface. In this case an analytical rigid
surface (SIMULIA, 2009a) is used to model the disk. When an analytical rigid surface
is defined in ABAQUS the orientation of the slip vectors is determined by the sequence
of the line segments to construct the surface. The first slip direction is always oriented
along the direction of the line segments forming the surface. The second slip direction
is defined such that the outward surface normal and the two surface tangents form a
right handed orthogonal system. In the case of a rigid circular disk this gives a radial
and a tangential oriented slip vector. Since the global coordinate system is Cartesian
based, with its center in the middle of the tire, it is not straightforward to interpret the
default slip velocity output. For every point on the tire, that is in contact, a different local
coordinate system exists. By post-processing the slip velocity data it is possible to align
all the local orientations with the global axes system. The alignment of the slip vectors
with the global axes system is done in two subsequent steps, see also Figure C.1. The
origin of the global axes system is located at the center of the wheel and the center of the
disk is positioned such that a specified slip angle is achieved. As a result of this, a fixed
reference line, constructed through the center of the disk and the origin projected on the
global X − Y plane, can be created. This is the line, with length b, in Figure C.1. Now it
is possible to project vslip of every arbitrary node to a reference system C by rotating over
an angle β. This angle is calculated using
 2
aj − b2 − c2j

βj = arccos , (C.1)
−2bcj

141
142 C COMPUTATION OF SLIP VELOCITY FIELD FOR THE LAT 100 SETUP

where the lengths of the line segments aj and cj follows from the current nodal position
of node j. The rotation matrix is given by
 
cos(β) sin(β)
Rβ = . (C.2)
− sin(β) cos(β)

Secondly an additional rotation over the side slip angle α,


 
− sin(α) − cos(α)
Rα = , (C.3)
cos(α) − sin(α)

is required to project the reference system onto the global axes system. The components
of the slip velocity vector in the global ABAQUS axes system can now be calculated as
   
vslipX vslip1
= Rα Rβ . (C.4)
vslipY vslip2

vslip22
vslip12

c2 a2

C2
−β2
vslip21
b C1
β1 α
a1
c1
Y
vslip11

Figure C.1 / Overview of rotation procedure to project local slip vectors, vslipj , onto the
global ABAQUS coordinate system, X-Y, for a positive side slip angle α.
Samenvatting

Bandmodellering is tegenwoordig een noodzakelijk onderdeel van het ontwerpproces van


nieuwe banden. De automobielindustrie, overheden en consumenten eisen betere grip
onder alle weersomstandigheden, minder slijtage en meer recent ook minder geluid en
een lagere rolweerstand. Eindige elementen analyses worden gebruikt in het ontwerppro-
ces van nieuwe banden, zodat rekening gehouden kan worden met deze, conflicterende,
eisen. Het modelleren van banden met behulp van de eindige elementen methode kan
het inzicht in specifieke eigenschappen van een band verhogen, de ontwerptijd verkorten
en de ontwerpkosten van nieuwe banden verlagen. Met de meeste eindige elementen
modellen is het nog steeds niet mogelijk experimenten op de weg goed te reproduce-
ren. Naast de statische deformatie moet ook de dynamische responsie van een rollende
band op de weg nauwkeurig voorspeld worden. Het sturen, remmen en optrekken hangt
af van de gegenereerde wrijvingskrachten. Wrijving hangt niet alleen af van de (ma-
teriaal)eigenschappen van het loopvlak van een band, maar ook van het wegdek en de
weersomstandigheden. Het hoofddoel van dit proefschrift is het ontwikkelen van een
robuust en numeriek efficient wrijvingsmodel geschikt voor banden simulaties met be-
hulp van eindige elementen en het creëeren van een raamwerk voor de identificatie en
implementatie van wrijving gerelateerde parameters.
Het modelleren van een band in combinatie met de omgeving is een uitdaging, omdat
verschillende fysische verschijnselen een rol spelen. In het algemeen zullen zowel de
mechanische, thermische en fluïdische effecten bijdragen aan de responsie van de band.
Dit onderzoek spitst zich toe op het mechanische domein, waarvoor een numeriek raam-
werk is gedefinieerd om simulaties met rollende banden met constante snelheid uit te
voeren. Dit raamwerk kan als uitgangspunt worden gebruikt in de verdere ontwikkeling
van het totale simulatiemodel. Eén van de doelstellingen van dit proefschrift is het ont-
wikkelen en valideren van een wrijvingsmodel dat geschikt is voor eindige elementen
analyses en effecten van droge wrijving op de wegligging van rollende banden kan be-
schrijven. Wrijving is een complexe interactie tussen twee materialen die met elkaar in
contact zijn. Wrijving kan gemodelleerd worden op verschillende lengteschalen en er kan

143
144 S AMENVATTING

gebruik gemaakt worden van verschillende numerieke methoden. Dit kan echter leiden
tot lange rekentijden, wat niet praktisch is voor gebruik in een industriële applicatie. Om
een numeriek haalbare en relatief snelle oplossing te bieden, is er gekozen voor een fe-
nomenologisch wrijvingsmodel, waar de parameters geïdentificeerd worden door middel
van een twee-staps experimenteel / numerieke aanpak.
Allereerst zijn er wrijvingsexperimenten uitgevoerd op een laboratoriumopstelling, ge-
schikt voor slijtage- en wrijvingsmetingen, om de invloed van de contactdruk op de wrij-
vingskracht te onderzoeken. In deze experimentele opstelling wordt een klein rubber
wiel, met een variërende drifthoek, op een schijf gedrukt. De aanwezige wrijving tussen
de schijf en het wiel zorgt ervoor dat het wiel wordt aangedreven door de roterende schijf.
De resulterende krachten op het wiel worden gemeten met een krachtsensor. Meerdere
metingen onder verschillende belastingen en drifthoeken zijn uitgevoerd. Deze metin-
gen, met lage rotatiesnelheid, zijn gebruikt om contactdrukgerelateerde parameters van
het wrijvingsmodel te identificeren. De relevante onderdelen van deze opstelling zijn ge-
modelleerd in het commerciële eindige elementen pakket ABAQUS. De prestaties onder
constante snelheid zijn geëvalueerd voor verschillende drifthoeken en vergeleken met de
metingen. Simulaties tonen aan dat de aanwezige ‘turn slip’, die grote invloed heeft op
het slipsnelheidsveld aan de achterkant van het contactvlak, goed wordt beschreven met
het model. Daarnaast komt de berekende spoorstijfheid goed overeen met de metingen.
In de tweede plaats zijn er met een autoband remmetingen op de weg verricht waarbij
verschillende voorwaartse snelheden zijn gebruikt om een snelheidsafhankelijke para-
meterset voor het wrijvingsmodel te verkrijgen. Het zo verkregen wrijvingsmodel is ver-
volgens gekoppeld aan een eindige elementen model van de band. Dit model is eveneens
in het eindige elementen pakket ABAQUS geconstrueerd. Het eindige elementen model
is statisch gevalideerd door middel van metingen van de contactdrukverdeling, de con-
tactoppervlakte en de radiale en axiale stijfheid van de band. De ‘steady-state transport’
methode van ABAQUS is vervolgens gebruikt om efficient de evenwichtsoplossingen on-
der verschillende voorwaartse snelheden, zoals gebruikt in experimenten op de weg, te
berekenen.
Tenslotte is de voorspellende waarde van het eindige elementen bandmodel in combi-
natie met het voorgestelde wrijvingsmodel geëvalueerd. Het weggedrag tijdens rechtuit
remmen, enkel sturen en een combinatie van remmen en sturen onder verschillende
belastingen, bandenspanningen en rijsnelheden is geëvalueerd en gevalideerd met expe-
rimenten. Gebaseerd op deze vergelijking is geconcludeerd dat het weggedrag adequaat
voorspeld wordt.
Dankwoord

Onderzoek is eigenlijk nooit klaar en een proefschrift kan altijd nog net iets beter. Tijdens
het schrijven van deze laatste pagina’s begin ik langzaamaan te beseffen dat dit toch echt
het einde inluidt van vier jaar onderzoek en dit proefschrift. Vanzelfsprekend heeft de
afgelegde weg tijdens dit promotietraject de nodige hobbels gekend. Gelukkig heb ik
deze weg niet alleen gevolgd, maar heb ik dit pad mogen bewandelen met meerdere
personen zonder wie dit proefschrift er in de huidige vorm niet zou zijn. Daarvoor wil ik
iedereen bedanken.
Henk, als eerste wil ik jou bedanken voor de mogelijkheid dit onderzoek uit te voeren.
Daarnaast waardeer ik je directe communicatie, vertrouwen en steun enorm. Je hebt me,
al sinds mijn afstuderen, altijd de vrijheid gegeven om mezelf steeds verder te verdiepen
in de materie. Tegelijkertijd stuurde je tijdig bij als ik de grote lijnen uit het oog dreigde
te verliezen. Ook daarvoor wil ik je bedanken.
Ines, zonder jouw enthousiasme en betrokkenheid was dit project heel anders verlopen!
Ik wil je ontzettend bedanken voor de fijne samenwerking, onze inhoudelijke discussies
en misschien nog wel meer voor alle andere niet werkgerelateerde gesprekken.
I would like to thank the members of the reading committee, Matthias Kröger, Daniel
Rixen and Marc Geers, for their careful reading and valuable feedback. Furthermore,
I would like to thank Wim Desmet and Bart de Bruijn for taking part in the defense
committee.
Bart, jouw enthousiasme tijdens ons eerste gesprek was een van de redenen om dit pro-
ject uit te voeren. De afstand Eindhoven-Enschede is groot, desondanks was onze sa-
menwerking goed. Ik wil je bedanken voor alle hulp en de kennis over banden en FEM
die je in korte tijd hebt overgedragen. Daarnaast was het fijn om van tijd tot tijd ABAQUS
frustraties te kunnen delen! Antoine, vanuit TNO was jij direct betrokken bij dit project.
Bedankt voor alle inhoudelijke discussies en het nauwgezet commentaar op de papers
en mijn proefschrift. Het uitvoeren van experimenten met ‘mijn’ banden was voor mij
een van de hoogtepunten in dit project. Willem, Bauke en Ton, bedankt voor jullie in-

145
146 D ANKWOORD

zet en zeker ook voor de gezellige tijd na de lange werkdagen. Verder wil ik iedereen
bij Vredestein en TNO bedanken die direct of indirect een bijdrage aan dit project heeft
geleverd.
Ik wil Jarno bedanken voor de experimentele bijdrage aan hoofdstuk vier. Al ben je mijn
enige afstudeerder geweest, jouw inspanningen om de LAT 100 te doorgronden hebben
geweldig geholpen. Daarnaast wil ik Igo, Erwin, Peter, Ruud, Martijn, Michel, Roel, Je-
roen en Bram bedanken voor hun bijdragen. Leo en Patrick wil ik bedanken voor de
ondersteuning bij alle hard- and software gerelateerde problemen tijdens dit project.
Verder wil ik alle (oud)collega’s van de DCT groep bedanken voor de uitzonderlijk goede
sfeer tijdens en vooral ook na het werk. Petra en Lia, bedankt voor alle gezelligheid op het
secretariaat en alle andere zaken die het leven van een promovendus een stuk prettiger
maken. In het bijzonder wil ik mijn roomies Gerrit, Francois en Benjamin noemen.
Bedankt voor de gezelligheid, het systeem, de vele discussies en de ontspanning in de
vorm van koffiepauzes doordeweeks en het af en toe brouwen (en drinken) van bier in
het weekend. Buiten de werkkring wil ik familie en vrienden bedanken voor de geboden
afleiding en de getoonde interesse in mijn werk.
Erik en Frank, bedankt dat jullie me terzijde willen staan. Het voelt goed om dit
promotietraject samen af te sluiten. Tenslotte zijn er twee mensen die, meer dan ze zelf
beseffen, hebben bijgedragen aan dit proefschrift. Pap en mam. Bedankt voor jullie
onvoorwaardelijke steun en het vertrouwen tijdens alle goede en slechte momenten.
Jullie hebben niet alleen mij, maar ons alle drie altijd gestimuleerd onze eigen weg te
gaan. Dank daarvoor.

René van der Steen


oktober 2010
Curriculum Vitae

René van der Steen was born on May 1st, 1981 in Borsele, the Netherlands. After finishing
his secondary education at the Rijksscholengemeenschap Goese Lyceum in Goes in 1999,
he started his study Mechanical Engineering at the Eindhoven University of Technology
in Eindhoven, the Netherlands. He carried out his international internship at Queen’s
University, Kingston, Ontario, in Canada, where he worked on ‘Adaptive extremum-
seeking control applied to bioreactors’. He received his Master’s degree cum laude in
February 2006 on the thesis entitled ‘Numerical and experimental analysis of multiple
Chua circuits’. This work was performed in the Dynamics and Control group at the
department of Mechanical Engineering. After completion of his Master’s thesis, René
worked for four months as a visiting scientist in the Dynamics and Control group.
In July 2006, he started as a Ph.D. student in the same group on the topic of tire modeling
using finite elements, with special attention to robust friction models. The project was
performed in cooperation with Apollo Vredestein B.V., Enschede, the Netherlands and
TNO Automotive, Helmond, the Netherlands. The main results of his Ph.D. research are
presented in this thesis.

147

Das könnte Ihnen auch gefallen