Download as pdf or txt
Download as pdf or txt
You are on page 1of 30

This article was downloaded by: [University of Arizona]

On: 15 November 2012, At: 00:42


Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Vehicle System Dynamics: International


Journal of Vehicle Mechanics and
Mobility
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/nvsd20

Wheel–rail contact models for vehicle


system dynamics including multi-point
contact
a b
J. Piotrowski & H. Chollet
a
Warsaw University of Technology, Institute of Vehicles, Narbutta
84, 02-524, Warszawa, Poland
b
Institut National de Recherche sur les Transport et leur Sécurité,
F94114, Arcueil, France
Version of record first published: 16 Feb 2007.

To cite this article: J. Piotrowski & H. Chollet (2005): Wheel–rail contact models for vehicle system
dynamics including multi-point contact, Vehicle System Dynamics: International Journal of Vehicle
Mechanics and Mobility, 43:6-7, 455-483

To link to this article: http://dx.doi.org/10.1080/00423110500141144

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-


conditions

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation
that the contents will be complete or accurate or up to date. The accuracy of any
instructions, formulae, and drug doses should be independently verified with primary
sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
Vehicle System Dynamics
Vol. 43, No. 6–7, June–July 2005, 455–483

Wheel–rail contact models for vehicle system dynamics


including multi-point contact
J. PIOTROWSKI*† and H. CHOLLET‡
Downloaded by [University of Arizona] at 00:42 15 November 2012

†Warsaw University of Technology, Institute of Vehicles, Narbutta 84, 02-524 Warszawa, Poland
‡Institut National de Recherche sur les Transport et leur Sécurité, F94114 Arcueil, France

Advanced modelling of rail vehicle dynamics requires realistic solutions of contact problems for
wheels and rails that are able to describe contact singularities, encountered for wheels and rails. The
basic singularities demonstrate themselves as double and multiple contact patches. The solutions of
the contact problems have to be known practically in each step of the numerical integration of the
differential equations of the model. The existing fast, approximate methods of solution to achieve this
goal have been outlined. One way to do this is to replace a multi-point contact by a set of ellipses. The
other methods are based on so-called virtual penetration. They allow calculating the non-elliptical,
multiple contact patches and creep forces online, during integration of the model. This allows nearly
real-time simulations. The methods are valid and applicable for so-called quasi-Hertzian cases, when
the contact conditions do not deviate much from the assumptions of the Hertz theory. It is believed
that it is worthwhile to use them in other cases too.

Keywords: Wheel/rail contact; Contact singularity; Online calculations; Rail vehicle dynamics

1. Introduction

The interaction between the rail vehicle and the track takes place in the contact zones between
wheels and rails and results in surface stresses distributed over the contact patches. At some
distance from the contact zone the distributed stresses have the same effect on the vehicle
motion as concentrated forces calculated as integrals of the surface stresses over the contact
regions. These forces are referred to as interaction forces and it is usual for people involved in
vehicle dynamics to think of interaction between vehicle and track in terms of the interaction
forces. The normal interaction force comes from the normal stress (contact pressure). The
tangential (creep) forces come from tangential stresses at the contact patch.
The tangential forces and components of the normal forces guide the vehicle on the track.
The tangential forces depend on the shape and size of the contact region, the distribution of the
normal pressure and friction between rail and wheel. It has been stressed in numerous literature
sources over the years that the description of rolling contact applied in models of vehicle/track
interaction strongly influences the validity and quality of simulation results. Because of this
well-understood fact, the problem of rolling contact of wheel and rail is still a topic of interest
and research.

*Corresponding author. Email: jpt@simr.pw.edu.pl

Vehicle System Dynamics


ISSN 0042-3114 print/ISSN 1744-5159 online © 2005 Taylor & Francis Group Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/00423110500141144
456 J. Piotrowski and H. Chollet

In the last 15 years two state-of-the-art articles devoted to this problem were published.
These are the paper by Elkins [3] in 1991 and the article by Knothe et al. [23] in 2001.
The recent and important state-of-the-art paper by Knothe et al. [23] describes the advanced
methods of contact mechanics applicable for wheel–rail and tyre–road contact.
For wheel–rail contact, where small deformations occur, special attention is paid to the local
effects of creep, wear, temperature development and fatigue. The paper [23] outlines advanced
boundary element approaches and finite element formulations to solve the contact problems.
The boundary element approaches are presently more efficient for the description of wheel–
rail contact in rail vehicle dynamics than the finite element formulations, but the latter are
less limited in advanced modelling of material properties and wear. Although the existing
computers are very powerful, an advanced method for online solving of the complete three-
dimensional contact problem is not yet available.
Downloaded by [University of Arizona] at 00:42 15 November 2012

The article by Elkins [3] gives an overview of the methods and techniques used until 1991 for
the prediction of wheel–rail interaction. The description of interaction forces in vehicle–track
models requires solving some geometrical as well as contact problems.
The description is based on the geometric data for profiles of the rolling surfaces of wheels
and rails. The data are collected using machines, which have been developed over the years by
several railways and small companies [6]. Such machines allow acquisition of data describing
the cross sectional geometry of both profiles (left and right profiles of wheels and rails) and
the relative position of both profiles. The paper by Elkins describes the processing of data
to obtain the geometrical functions in tabulated form. In 1991, the method by Duffek [2] for
calculating geometrical functions of yawed wheels/wheelsets was already in use.
As far as the previous topic is concerned, not much new development has taken place over
the last decade. At the present time, the calculation of the geometrical functions is a standard
step in building the model of vehicle–track dynamics.
At the time of Elkins’ article, the application of contact mechanics for the description of
contact between wheel and rail for vehicle–track models was mainly limited to elliptical
contact zones. Elliptical contact zones resulting from calculations using the theory of Hertz
[10, 11, 23] were commonly used in modelling. The solutions of the tangential contact prob-
lem, determining tangential (creep) forces for an elliptical contact zone were available from the
works by Kalker both in the framework of his ‘complete’ theory [12–14, 16] and the simplified
theory (Fastsim algorithm) [14]. The results pre-tabulated by British Rail from the complete
theory were used in the form of a book of tables [3]. Using his computer code CONTACT
[13, 14], Kalker designed a more comprehensive and advanced book of tables in 1996 [15].
At that time some so-called ‘exact’ methods for solving the non-Hertzian normal contact
problem were available, developed by Kalker [13, 14] and Hashemi and Paul [9]. Because of
the time consuming calculations those methods have not been incorporated in computer codes
for rail vehicle dynamics, where during simulation, the solutions of a great number of contact
problems are required.
At about the same time, approximate methods for solving the non-Hertzian normal contact
problem emerged. The method is called approximate if it assumes a semi-definite normal
stress distribution. For a body of revolution, such as wheel or a roller, a natural choice
is the semi-elliptical normal stress distribution in the direction of rolling. This assump-
tion follows the results of the Hertz theory. The first work on an approximate method by
Reusner [37] was devoted to the contact of rollers in roller bearings. Gostling [7], Gostling
et al. [8], Le [24], and Knothe and Le [22] developed similar methods for railway wheel
and rail.
The approximate methods mentioned earlier were not completely reliable and therefore
their use was limited to tabulation in order to properly access the Kalker book of tables,
Wheel–rail contact models 457

execute the Fastsim algorithm or apply heuristic formulae, such as Shen–Hedrick–Elkins [38],
for example.
The description of the normal contact between wheelset and track may have the form of a
multi-dimensional table [31, 35]. There are generally six independent variables in the table:
lateral shift of the wheelset, yaw angle, track gauge spreading (or narrowing), normal load,
and two angles of individual rail cross section rigid rotation. With the use of the presently
available approximate methods for the normal contact problem and modern computers, the
table may be computed in a few minutes.
The tabulation of the results of the normal contact calculations requires replacing the actual
contact patch by a regular figure. The obvious choice is the ellipse, because it is compatible
to some extent with the assumption of semi-elliptical stress distribution and is required for
accessing the methods mentioned earlier, to determine the creep forces. The resulting ellipse
Downloaded by [University of Arizona] at 00:42 15 November 2012

is called an equivalent one. In cases of multi-point contact, the calculation of the equivalent
ellipse is more complex [27, 42, 45–47].
There is a fundamental difference between the contact ellipse calculated using the Hertz
theory and the equivalent ellipse determined by a non-Hertzian contact calculation. The former
uses the radii of curvatures at the geometrical point of contact as the input data. During the
geometrical processing of the measured data for profiles, the smoothing and interpolation
strongly influence the local geometry and the calculated radii of curvatures at a point are
biased by these procedures. The smoothing and interpolation do not strongly influence the
global geometry and therefore the data required for non-Hertzian calculations in some range
near the point of contact (undeformed distance) is little affected by these procedures. In the
result both ellipses might differ drastically.
Even before the approximate calculation methods for the normal contact problem emerged,
the results of geometrical calculations frequently indicated the possibility of two- and multi-
point contact formation in profile locations where jumps in rolling radii difference and contact
angles appeared.
The existence and importance of two-point contact has been realised since the early days
of railways when nominal wheel profiles lead to double contact on the guiding wheels on
curves with one contact patch on the tread and the other on the flange. The earlier mentioned
geometrical calculations show that a similar situation is possible and frequently happens in
different locations on the wheel profile. The first visualizations of multiple contact patches of
wheel and rail were probably obtained with the use of the approximate calculation method
of Gostling [7, 8]. The work [28] was probably the first to use approximate results of nor-
mal contact calculations for the prediction of normal and tangential forces in two distant
contact patches, one on the wheel flange and the other on the tread. Other works followed
refs. [17, 30, 32, 36].
Multiple contact belongs to a category of non-elliptical contact, appearing when the geome-
try of contacting bodies does not satisfy the constant curvature assumption of Hertz theory. In
the state-of-the-art paper by Knothe et al. and numerous literature sources it has been repeat-
edly stated that a necessary condition for improving computer codes for rail vehicle dynamics
is introducing solutions for non-elliptical contact problems.
The most successful methods for solving both normal and tangential contact problems, when
the geometry of contacting bodies does not satisfy the assumption of Hertz, are by Kalker,
and are summarized in his book [14]. They are built into the computer code CONTACT
that is able to deal with majority of the important problems for vehicle system dynamics.
Frequently encountered non-elliptic and multiple contact patches may be calculated. The
stress distribution and creep forces may be calculated for steady-state and non-stationary
conditions of rolling. The solutions are available for elastic and visco-elastic bodies.
458 J. Piotrowski and H. Chollet

The advantages and shortcomings of CONTACT are outlined by Knothe et al. [23] in
the state-of-the-art paper. From the viewpoint of the present applications in vehicle system
dynamics the main shortcoming is a high computational time, which does not allow use of
CONTACT online in simulations. Because of this, attempts have been made to write simplified
methods, which allow online solution of the contact problems for wheel and rail.
The problems encountered due to multiple contact during the modelling of vehicle system
dynamics are illustrated by drawings in figure 1. The drawings show the results of geometrical
and contact calculations for a wheelset and track using measured geometrical data. The method
by Kik and Piotrowski [19], see section 3.2, was used for the calculations.
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 1. Contact patches for measured wheelset and track under normal load 105 kN.
Wheel–rail contact models 459

Each drawing shows parts of wheel and rail profiles and contact patches as they match at
a given position of the wheelset relative to the track centre line. The position is determined
by the lateral shift of the wheelset, which is given by the first number in the column between
drawings, for example −9.22 mm at the top of figure 1. The number just below the shift is the
difference of rolling radii r = rr − r1 ; (right–left). The numbers with the degree sign, are
contact angles. The maximum normal pressure is also transferred to the drawing, for example
1731 MPa.
On the left wheel–rail somewhere in the range of lateral shift −4.24 to −2.58 two-point
contact would form with about 12◦ contact angle at one zone and 0.5◦ at the other. In the lat-
eral shift range 0.73–4.05 mm, an even more complicated situation is encountered at the right
wheel–rail with transitions from one- to two- and three-point contact and back to one. The
next clearly visible singularity takes place in the lateral shift range 7.37–9.03 mm at the right
Downloaded by [University of Arizona] at 00:42 15 November 2012

wheel–rail. There the two-point contact would form with a large difference in contact angles.
Advanced modelling of vehicle system dynamics requires realistic modelling of individual,
normal and tangential forces in the multiple contact patches.
This publication focuses on the approximate methods for fast solving of the contact problems
for wheel and rail, including online methods, which are applicable for simulation models of
vehicle system dynamics. There are two kinds of methods used. One of them replaces the
non-elliptical contact zone by a set of ellipses. The first such method is by Pascal and Sauvage
[27], and the analytical one by Ayasse [42]. The other methods are based on virtual penetration
of contacting bodies and were developed by Kik and Piotrowski [19], Linder [25] (see also
[39]), and Ayasse and Chollet [1].

2. Approximation of non-elliptic contact zone by a set of ellipses

2.1 Sauvage’s method: interaction between elastic profiles

In the 1980s, the focus was on the TGV development in France. The constant conicity was used
at SNCF for this programme and for the first nonlinear VOCO code, developed by Ayasse at
INRETS for freight wagon simulations [41]. Sauvage had the idea of making a more powerful
contact model.
The calculation of the Hertzian contact ellipse is classically related to the curvatures of wheel
and rail profiles at the point of minimal distance between wheel and the rail ‘rigid’ profiles. But
Sauvage proposed an innovation to determine a secondary ellipse, then possibly a third one,
etc. – in fact, the number is limited by the user (in practice to seven in his code). Unfortunately,
Sauvage did not publish this part of the process but only gave some oral descriptions of it.
The method (figure 2) is based firstly on the interpenetration of the profiles, and secondly
on subtracting the Hertzian interpenetration section from the initial profiles. Starting from the
Hertzian curvatures A and B the interpenetration (or elastic approach) δ is calculated under
the normal load N for materials with elastic constants E and ν:
   2 1/3
1 1 1 1 3 1 − ν2
A= , B= + , δ=r N (A + B) . (1)
2Rwy 2 Rrx Rwx 2 E

The elastic approach δ is calculated and normalized by N 2/3 for storage. If δ is used directly
for calculation of the virtual region, the area of it is too large for the following step.
Recent researchers suggest to use h1 = δ/2 (figure 2); however, Sauvage takes h1 = δ
which leads to larger contact overlaps (figure 4). In both cases, if the modified profiles still
interpenetrate, it is concluded that a second ellipse is appearing with the interpenetration h2 .
460 J. Piotrowski and H. Chollet
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 2. Determination of possible secondary ellipse using the interpenetration δ.

The secondary ellipse has its own Hertzian curvatures A2 and B2 but its load is not known.
The combination of h1 and h2 , taking into account the Hertzian property δ ∼ N 2/3 allows
determination of the loads.
A similar method (figure 3) can be applied by subtracting the Boussinesq deformation
from the profile. The deformation distance is a bell-shaped curve. This shape is better than a
semi-ellipse whose shape is related to 2b (ellipse width) with the depth defined in the same
way. With the bell shape, frequent secondary contacts near the border of the first contact are
avoided.

2.2 Pascal’s method (1992): the equivalent ellipse

Sauvage worked out a wheelset dynamic model named DIPOINT [47, first part] and then
tried to model a bogie, but the simulation speed became very low and it appeared that the
full method cannot be applied directly in a commercial code, which requires almost real-time
simulations.
Pascal [47] proposed the use of wheelset model for the determination of the parameters of
the equivalent ellipse for use in the VOCO code. In this application, the increasing lateral force
is quasi-statically applied to the wheelset at the rail level. The vertical load is kept constant. The
normal loads (figure 4) are computed taking into account the equilibrium of the wheelset with
tangential forces calculated by the FASTSIM algorithm with a given value of the coefficient
of friction (0.4 in figure 4, [46]). This is important because the tangential forces are unloading
the contact during flanging.
The dynamic wheelset model led to the benchmark proposed at the IAVSD workshop in
Herbertov [47].

Figure 3. Determination of possible secondary ellipse using the Boussinesq deflection.


Wheel–rail contact models 461
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 4. Distribution of normal load between ellipses. Lateral displacement 0 to +8 mm.

For the determination of the equivalent ellipse, the resultant contact forces are stored in a
file and a combination of contact angle and curvatures is found to obtain the same forces under
the same conditions.
For example, in the case of two separate contact zones approximated by two ellipses, the
ratio of axes of the equivalent ellipse is calculated as the weighted mean of the same ratio for
two original ellipses
a      
a1 a2 1
= N1 + N2 . (2)
b equiv b1 b2 N1 + N 2
This approach has been used in the work presented in the paper [45], describing the dynamics
of TGV wheelset with S1002 wheel profiles rolling on rails with UIC60 profiles inclined by
1:40. Another paper [46] refers to the jump of the same rolling pair (figure 5) at 1 mm lateral
shift of the wheelset, as described by the equivalent ellipse method.
The method has its limitations because the tangential forces required for the equilibrium of
the wheelset, when determining the equivalent ellipse, are calculated with Fastsim for small
creepages, when the linear theory of rolling contact is valid. Then the results are implemented
for conditions beyond the validity of the linear theory, such as, for example, low normal load
or lubrication.
Of course, this method is adequate for dynamic simulations, but it is necessary to go back
to the multi-Hertzian description in order to make, for example, surface stress analysis and
wear calculations. A similar approach was proposed in [35].

2.3 Ayasse’s method: analytical formulation

In order to increase the speed of simulations, Ayasse proposed several simplifications and an
analytical formulation of the jump overlap as a function of the initial normal load, with the
aim of reducing the description of the contact to small pre-calculated tables, combined and
interpolated in the dynamic model code.
462 J. Piotrowski and H. Chollet
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 5. Equivalent ellipse for S1002/UIC60 pair at wheelset central position [46].

2.3.1 Simplifying assumptions. The question is whether to take into account the second-
order couplings between degrees of freedom of the wheelset or not. These couplings are due
to the contacting surfaces. The contact location and the corresponding parameters are affected
by the yaw and roll angles. Of course, a numerical analysis can be made but the data tables
are memory and time consuming and therefore the analytical solutions are generally faster.
A consequence of the important lift translation tz of a flanging wheel is the rotation of
the whole wheelset, and of the opposite wheel around the corresponding rail tread radius Rrx .
The variation of the contact angle is at the first-order given by tz /gauge. The variation of the
contact position has a small value λ = Rrx tz /gauge, which depends on the rail radius, see
figure 6. In the limit case, where the contact point is near the boundary radius on the rail or
on the wheel, the situation is more complex.
If the flange lift is taken into account, its effect on the other wheel depends on the gauge.
A way of taking this left–right interdependence into account was proposed by Sauvage with
a shift of the central parts of the curves, but it does not work well if there are jumps in the
middle of the curves. Sauvage, who was interested in high-speed trains running with moderate
yaw angles, did not consider the effect of yaw on the contact location.
The roll effect was also neglected for several reasons: the approximation is small for flange
contact with high contact angle; the approximation is also small on the tread; the major part of
the lateral force is not affected by a minor variation of the contact angle. But more importantly,
Wheel–rail contact models 463

Figure 6. The roll angle at flanging.


Downloaded by [University of Arizona] at 00:42 15 November 2012

this approximation removes both difficulties, so that the left and right wheel–rail pairs may
be described independently.
The roll angle being neglected, the relative movement between profiles is a translation:
t = ty + tz = ty y + tz z in the transversal plane YOZ of the track.
This allows variation of the gauge without difficulty. The contact parameters are computed
at the nominal gauge (for example 1435 mm) and stored as functions of ty . In the code of the
dynamic model they are read and interpolated with the input data: ty + gauge/2.

2.3.2 Piecewise linearized contact tables. Additionally, Ayasse proposed pre-calculation


of the contact conditions in order to accelerate the simulations. The input variable is the lateral
translation ty .
A first table is used for the description of the rail by a series of constant radii Rrxx . In the
same way the nominal profiles are dimensioned, for example the UIC 60 profile. Instead of
the wheel profile, the second table is built from the contact angle γ of wheel and rail profiles,
as a function of ty .
When this table is linearized (figure 7), only a few lines are necessary to describe the
function, each line corresponding to a particular position tylim where there is a jump between
two contact areas.

Figure 7. Rail profile data: radius of curvature and contact angle. Piecewise linearization for tables.
464 J. Piotrowski and H. Chollet
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 8. Principle of the multi-Hertzian overlap near jump.

When there is a jump, one can observe from figure 8 that the normal load varies quasi-
linearly with the lateral displacement ty = yw − yr . It is proposed that only the limits tylim are
stored and the points limiting the overlap are calculated in the code of the dynamic model,
without storing any loading/unloading values.
The wheel and rail profile are not useful inside the vehicle–track model dynamic code. By
combining the contact angles, the rail radius Rrx and the rail profile, it is possible to get back
to the active part of the wheel profile by integration of the wheel curvature Rwx
1
Rrx + Rwx = − ,
cos γ (dγ /dty )
then to all the Hertzian parameters (1) necessary to calculate different ellipses.

2.3.3 Overlap analytical determination. Two basic expressions may be proposed,


depending on whether the jump is supposed to be large or small, i.e. if the ellipses are supposed
to be distant or close.
Here again the starting parameter is the contact angle function, used in the expression of
the overlaps (3), see figure 8.
A jump of width ty is determined at each side 1 and 2 of the jump limit tylim for rigid
conditions by the contact angle function. Generally, ty1 is different from ty2 . The values
of these overlap widths are based on the interpenetration di = δi /2, δi , (i = 1, 2) being the
deformation due to normal load Ni determined by the Hertz formulae.
For large jumps
cos γ1 δ1
ty1 = . (3)
tg (γ2 − γ1 ) 2
A more general formulation than equation (3) is proposed in [42] in order to be compatible
with limit cases, when the jump is small (γ2 → γ1 ). Formula (3) shows that on the tread the
overlap is large, but when flanging, the angle difference is great and the overlap order is close
to δ/2.
An interesting aspect of this method is the load dependency. The overlap is determined at
the beginning of the jump as a function of the dynamic load on the wheel at this moment. This
determination is load dependent, but once a transfer is running, it has been observed that the
overlap description should not be changed for reasons of numerical stability.
Wheel–rail contact models 465

The method of Pascal presented in section 2.2 and the one of Ayasse have been validated
with similar results on a low speed run of the CORAIL French passenger car. The simulations
were presented at the Chengdu IAVSD Symposium [40]. At that time, the multi-body model
was limited to three DOF per body and to a standard two-bogie vehicle, able to simulate
many rail vehicles, but not all of them. When microcomputers became more powerful, it was
extended to six DOF and to a wider range of vehicle models.
The analytical multi-Hertzian method has been improved during the following years in order
to stabilise the numerical calculation. This has been obtained by the use of two serial springs
between the contact and the ground, representing the rail, ballast, pads and wheel compliance
in the Y and Z direction.
The description of the principle, the analytical expression of the jump overlap, and some
dynamic validations in comparison with full scale test has been published in [42] and later
Downloaded by [University of Arizona] at 00:42 15 November 2012

in [43]. The process has been named the ‘CAF method’, because it is based on the analysis
and storage of the main parameter – the contact angle function.

2.3.4 Extension of the method. Considering the contact angle function, Shen et al. [50]
have shown that it is possible to design a new profile for a running gear with independently
rotating wheels in a situation of excessive wear, starting from a given rail profile and a target
contact angle function, where the gravitational stiffness is increased. The real use of this wheel
profile has shown a large increase in the profile life.
A way of simulating the profile variation in a switch has even been proposed in ref. [43]. In
figure 9 the representation of the switch blade is shown with the contact angle function. This
graph is a way to see directly that the lowest contact angle on the blade is lower than 1 rad and
that derailment would be likely to occur at this point.
However, even if the method has been found to simulate variable profiles in the switch,
this leads to multiplication of artificial jumps in the data tables. For these artificial jumps, two
ellipses with the same b/a ratio are located at the same place, with the same creepages. It is

Figure 9. Variable CAF in a switch.


466 J. Piotrowski and H. Chollet

possible to demonstrate that the sum of the two artificial ellipses has the same resultant creep
forces as the real single one, but the direct use of the surfaces and pressures of these ellipses
is difficult in models composed of rigid bodies.
Originally, Ayasse’s method was not focused on the exact location of the ellipses but on
a fast and physically-based determination of the jump overlap. When there are no profile
variations along the track, the multiple ellipses can be taken as they come, and because they
are computed inside the dynamic loop, the computed forces and stresses can be used with
confidence. Instead when the profile variation does exist, as in switches and worn curves,
another method is necessary. This will be presented in section 3.3.
Downloaded by [University of Arizona] at 00:42 15 November 2012

3. Methods based on the virtual penetration

The common assumption for these methods is the semi-elliptical normal stress distribution
in the direction of rolling. The similarity of the stress distribution to that resulting from the
theory of Hertz makes it possible to use properties of the Hertz solution to solve approximately
the normal contact problem for conditions when the geometry of the contacting bodies does
not rigorously satisfy the assumption of Hertz. In the result calculated contact patches may be
non-elliptical and in many cases multiple.
The idea of the methods follows from figure 10. When the undeformed surfaces of wheel
and rail, touching in the geometrical point of contact 0, are shifted towards each other by the
distance δ, called the penetration, they penetrate and intersect on a closed line (lines), whose
projection on the 0xy plane is called the interpenetration region. In reality, the bodies cannot
penetrate, deflections occur and the contact zone comes into being.
It has been shown by Hashemi and Paul [9] that the interpenetration region encircles the
contact zone if the influence function describing normal deflection of bodies is unidirectional.
The Boussinesq influence function [11, 14] for elastic half space is indeed unidirectional.
It is used by the methods in question, implicitly in refs. [1, 25] and explicitly in [19].
On the basis of some similarity of shapes of the contact zone and interpenetration region, the
contact zone is determined by the following assumption: the surfaces of bodies are virtually
penetrated to the depth δo < δ and the resulting interpenetration region is taken for the area
of contact [19].

Figure 10. The interpenetration region and the contact zone.


Wheel–rail contact models 467

It is known that the interpenetration region and the contact zone might differ in shape,
especially when they are slender. No correction of shape is introduced in the method of
Linder [25]. The shape correction for the method of Kik and Piotrowski [19] is described
in section 3.2. The shape correction is built into the method of Ayasse and Chollet [1], see
section 3.3.
To proceed with the presentation of the methods the interpenetration function of profiles is
defined as
δ − f (y) if f (y) ≤ δo
g(y) = o , (4)
0, if f (y) > δo
where f (y) = z(y) = z1 (y) + z2 (y) is the separation of profiles.
The contact zone is determined by the x co-ordinates of its leading and trailing edges
described by the formula
Downloaded by [University of Arizona] at 00:42 15 November 2012

x1 (y) = −xt (y) ≈ 2Rg (y). (5)


Here R is the Hertzian radius in the 0xz plane at the geometrical point of contact. It is
usually assumed that this radius is constant throughout the contact zone. This is not a necessary
assumption. It is easy to account for variable radii but the effect is marginal.
Having defined the contact zone it is necessary to calculate the normal stress distribution
and the normal load resulting from the prescribed penetration. This is done in a different way
by different methods.
The creep forces are calculated on the basis of the simplified theory of rolling contact [14]
using Kalker’s Fastsim algorithm. This requires to determine the value of elasticity parameter
(or parameters) for the non-elliptical contact patch. This determination may only be carried
out by referring to exact solutions for elliptical contact patches. This may be done either by
replacing the actual contact patch by an equivalent ellipse or by assuming different equivalent
ellipses for each strip along the rail of the non-elliptical contact patch.
The other problem is the choice of the traction bound in Fastsim. The natural choice is
the semi-elliptical traction bound. But the parabolic traction bound along the rail gives better
qualitative agreement with the complete theory of Kalker than the semi-elliptical one. When
there is a similarity of the slip and adhesion regions in both theories the agreement is called
qualitative. It has not been proved that the parabolic traction bound improves quantitative
agreement, i.e. the closeness of the creep forces to those resulting from the complete theory.

3.1 The method of Linder

The value of the virtual interpenetration was taken δo = 0.55δ after [19]. This value was
confirmed as realistic by numerical experiments using the program CONTACT [19]. The
analysis of the theory of Hertz leading to this value of the virtual penetration may be also
found in [25].
The origin of the coordinate system 0xy of the contact area in figure 11 is taken at the
middle of the contact area. The contact area defined by the equation

x1 (y) = −xt (y) ≈ 2Rg (y),

is discretized by a number of strips in the direction of rolling as shown in figure 11.


It is assumed that each pair of points [x1 (yi ), yi ], [xt (yi ), yi ] belongs to an ellipse. The
semi-axis across the rail of each ellipse is assumed the same and equal to

1
be = (y1 − yr ). (6)
2
468 J. Piotrowski and H. Chollet

Figure 11. Discretization of the contact area by strips.


Downloaded by [University of Arizona] at 00:42 15 November 2012

The semi-axis along the rail of the ellipse of strip i is

be
aei = g(yi ) . (7)
be2 − yi2

It has been assumed that each ellipse results from a solution of the Hertzian normal contact
problem and that the penetration δei leading to the formation of the ellipse i with semi-axes
aei , and be may be determined from the interpenetration function as shown in figure 12.
The interpenetration function in the Hertzian case is a section of a circle with radius rei . It
is shown in figure 12 by the thin line. The following equations hold:

be2 + zmi
2
= rei2
yi2 + [zmi + g(yi )]2 = rei2

from where
1
zmi = [b2 − yi2 − g 2 (yi )]
2g(yi ) e
(8)
δei = be2 + zmi
2
− zmi + 0.45δ.

With aei , be and δei known it is possible to calculate the normal load Ni transmitted by the
given ellipse. The pressure at the centre of the strip is
 2
3Ni yi
p0i = 1− . (9)
2πai be be

Figure 12. The penetration δei leading to the formation of the Hertzian ellipse.
Wheel–rail contact models 469

The total normal load transmitted by the contact patch is calculated by integration
 
n xl (yi )
x2 1
n
N= p0i 1 − 2 dx yi = π xl (yi )p0i yi . (10)
i=1 −xl (yi ) xl (yi ) 2 i=1

Contact patches calculated by the approximate method and program CONTACT are shown
in figure 13. In cases of one-point contact the patches are very similar. In cases of multi-point
contact the similarity is less good.
There are two properties of the approximate solution that decide its usefulness in vehicle
system dynamics:
• predicted normal loads for given penetration should be close to that predicted by an
advanced, ‘exact’ method, such as the program CONTACT for example;
Downloaded by [University of Arizona] at 00:42 15 November 2012

• predicted creep forces should be close to those obtained by the advanced method.
The first property makes it possible to apply the technique of so-called flexible contact. The
second one requires among others that the method predicts the shape and size of the contact
patch well.
In the work by Linder [25] no detailed information about the performance of the method
with respect to the above properties is given.

3.1.1 Calculation of creep forces. The creep forces are calculated in the framework of
the simplified theory using the Fastsim algorithm [14]. The version of Fastsim used by Linder

Figure 13. Contact patches during negotiation of 300 m radius curve calculated by Linder [25], for worn rails and
wheels. Leading wheelset. ‘+’ indicates Linder’s method, square indicates code CONTACT.
470 J. Piotrowski and H. Chollet

involves three elasticity parameters Lx , Ly and Lz . In this version, the components of tangential
traction resulting from the linear, simplified theory are:
 
νx y·ϕ
px (x, y) = − [x − xl (y)]
Lx Lz
 
νy x·ϕ
py (x, y) = + [x − xl (y)],
Ly Lz

where νx , νy and ϕ are longitudinal, lateral and spin creepages, respectively. These equations
are used in Fastsim.
The elasticity parameters are determined by the equalization of solutions from the linear,
Downloaded by [University of Arizona] at 00:42 15 November 2012

complete theory and the simplified theory for elliptical contact area and pure longitudinal,
lateral and spin creepages. The equalization results in [14]:

8a 8a πa a
Lx = , Ly = , Lz = , (11)
3Gc11 3Gc22 4Gc23 b

where G is the modulus of elasticity and Kalker creep coefficients cij depend on a/b and
Poisson’s ratio [12].
In the calculation of creep coefficients cij the semi-axes aei , be of the virtual ellipse are taken
instead of a, b. This leads to different elasticity parameters in each strip:

8aei 8aei π aei aei
Lxi = , Lyi = , Lzi = ,
3Gc11i 3Gc22i 4Gc23i be

where c11i , c22i , c23i depend on the aei /be ratio.


The elliptical traction bound along the strip has been replaced by the parabolic one in such
a manner that the normal load on the strip remains the same. This leads to a redefinition of

p0i , which becomes p0i = 3πp0i /8. The traction bound in the strip is defined by the equation
 
3π x2
pτ i (x, yi ) = f · p0i 1 − 2 ,
8 xl (yi )

where f is the coefficient of friction.


Linder applies his method for wear simulations [25, 26]. The direct comparison of creep
forces with the program CONTACT has not been given.

3.2 The method of Kik and Piotrowski

It has been assumed that the calculated contact is quasi-Hertzian, so it possesses some prop-
erties of Hertzian contact. One specific assumption of the method is that the normal pressure
distribution is semi-elliptical in the direction of rolling and has the form

p0
p(x, y) = x 2 (y) − x 2 , (12)
xl (0) l

where p0 is the maximum normal pressure.


Wheel–rail contact models 471

The normal load is calculated by integration of the pressure


 yl  xl
p0
N= xl2 (y) − x 2 dx dy. (13)
xl (0) yr −xl

The normal deflection at the point (0,0) is



  xl2 (y) − x 2
1 − σ 2 po yl xl
w(0, 0) = wo =
dx dy. (14)
πE xl (0) yr −xl x2 + y2

Here σ is Poisson’s ratio and E Young’s modulus.


Downloaded by [University of Arizona] at 00:42 15 November 2012

The contact condition at the geometrical point of contact (0,0) reads 2wo = δ, so from
equations (13) and (14) one obtains

 yl  xl 2
πEδ yr −xl xl (y) − x 2 dx dy
N= , (15)
2(1 − σ 2 )  yl  xl x 2 (y) − x 2 
x 2 + y 2 dx dy
yr −xl l

N 2Rδo
po =   . (16)
yl x l
yr −xl x 2
l (y) − x 2 dx dy

In rail vehicle dynamics one deals with measured profiles of wheels and rails. The measured
profiles are irregular and the integrals in equations (15) and (16) have to be and are evaluated
numerically.
It is clear that the calculation reduces to the evaluation of the formulae (15) and (16). The
method is therefore fast and completely reliable. It has been applied for wear simulation by
Kik and Moelle [18, 20].
It is possible to use the method when the normal load is prescribed. In such a case the
iteration of penetration is necessary.
Tests of the method were carried out by comparing results with those obtained from CON-
TACT. They are reported in [19]. The comparison showed that the normal loads from both
methods are almost the same for a broad range of contact conditions, including multi-point
contact.
The results of the comparison are shown in figure 14. Two cases with different geometry
(undeformed distance) were calculated for prescribed penetration in the range 0–0.12 mm.
Three values of Hertzian radius of curvature along the rail were assumed: 150, 550 and
2000 mm.
The geometry of case 1 leads to the formation of double contact zones. In case 2, a single
contact patch is formed. Comparison of the normal forces is given by plots on the left hand
side of the figure. The upper diagram is for case 1, the lower for case 2. The contact patches
for Hertzian radius of 550 mm and several values of penetration are drawn on the right hand
side of the figure.
The creep forces are calculated by applying the Fastsim algorithm to the non-elliptical
contact area. After defining three elasticity parameters as in [11], a single elasticity parameter
is calculated as a weighted mean of three before-mentioned parameters. The weighing factors
are the creepages and spin.
The creep forces are calculated taking into account the fact that the contact patch is spread
on curved surfaces, so that rigid slip is defined locally [33, 51].
472 J. Piotrowski and H. Chollet
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 14. Comparison of results calculated by the method of Kik and Piotrowski [19] with those from Kalker’s
program, CONTACT.

As far as creep forces are concerned the comparison is less favourable for the method in
question. There are two reasons for this:
(1) The simplified theory always gives different results from the complete theory; only in
cases of pure translational creepage and pure spin the results may be quite close.
(2) Contact zones might differ from those calculated with the use of the complete theory,
especially when they are slender.
It is possible to improve the solution with respect to the last cause of discrepancy by
introducing a shape correction based on the theory of Hertz. The idea of the correction comes
from the fact that in the case of Hertz assumptions it is possible to calculate the ratio of
the contact ellipse semi-axes from the ratio of the semi-axes of the interpenetration region,
which is also an ellipse. It has been assumed that a similar procedure would improve the
contact patches calculated by the approximate method. The calculated contact patches are
generally non-elliptical but due to the assumptions the method is quasi-Hertzian. This means
that the correction might be effective in cases when the resulting contact zones do not differ
significantly from ellipses. It is hoped that the improvement will also be worthwhile in other
cases.

3.2.1 The shape correction. The length and width of the virtual interpenetration region
are L and W . The Hertzian curvatures at the geometrical point of contact are
1 4δ0 1 4δ0
A= = 2, B= = 2.
2Rx L 2Ry W
Wheel–rail contact models 473

In the case of Hertzian geometry A/B ratio is used to find coefficients na , nb determining
the shape of the contact ellipse. To make use of the Hertz tables of [29] the A/B ratio is
redefined in the following manner:

A (W/L)2 if (W/L) ≤ 1
=
B (L/W )2 if (W/L) > 1.

After determining na and nb , α0 = na /nb , and



α0 if (W/L) ≤ 1
α= (17)
1/α0 if (W/L) > 1.
Downloaded by [University of Arizona] at 00:42 15 November 2012

The shape correction of the virtual interpenetration region satisfying the following equations
is proposed:

Lc /Wc = α
(18)
Lc Wc = LW

where Lc and Wc are corrected length and width of the region.


From equation (18) one obtains


Wc = LW/α, Lc = αLW . (19)

To correct the width of the contact patch a ‘stretching’ or ‘contracting’ of the separation
across the rail is introduced. This is done in such a manner that the geometrical point of contact
stays at its place.
In figure 15 the contact patch corresponding to the uncorrected virtual interpenetration
region is shown. Suppose that the width W of the contact patch should be widened by
W = Wc − W . When narrowing is required W is negative.
The uncorrected separation function corresponding to the virtual penetration δ0 is described
by the function g(y), see equation (4). The stretching across the rail results in the situation
shown in figure 16.
The maximum δ0 of the function g(y) is at y0 . This is the geometrical point of contact. The
co-ordinates of the edges of the uncorrected separation function are ys and ye .

Figure 15. Uncorrected contact zone.


474 J. Piotrowski and H. Chollet

Figure 16. Correction of separation function by stretching/contracting along y-axis.

The corrected separation function is described by


Downloaded by [University of Arizona] at 00:42 15 November 2012

W
gc (y ) = g(y), y = y − (y0 − y). (20)
ye − y s
The value of the corrected function at coordinate y is the same as the original one at
coordinate y, but y is shifted corresponding to y, depending on y and the sign of W .
The corrected separation of the profiles is
W
fc (y ) = f (y), y = y − (y0 − y). (21)
ye − y s
The correction of the length of the contact patch is made by the modification of Hertzian
radius of curvature in the longitudinal direction. Suppose that the length of the contact area
should be increased by L = Lc − L. Then,
1
Rc = (L + L)2 . (22)
8δ0

The length of the uncorrected contact patch is L = 2 2Rδ0 .

3.2.2 Numerical example. In this example the approximate solutions of normal and
tangential contact problems are compared with those obtained from the program CONTACT.
The creep forces of the approximate solution are calculated by the Fastsim algorithm as
described in [19].
The basic data are the separation of profiles (undeformed distance) shown in figure 17
and the Hertzian radius of curvature along the rail R = 1400 mm. It has been assumed that

Figure 17. Original and corrected separation of profiles.


Wheel–rail contact models 475
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 18. Uncorrected contact zone with surface quantities.

the penetration is δ = 0.074 mm. The creepages are υx = υy = 0, ϕ = 0.0003 [1/mm] (pure
spin) and coefficient of friction f = 0.4.
The contact patches obtained from uncorrected and corrected separation are shown in
figures 18 and 19, where the surface quantities are also visible. The rolling direction is from left
to right. The shaded elements indicate the area of sliding. The distribution of normal pressure
is shown in figures 20a and b. The results of the calculations are listed in table 1.
The solution obtained with the use of program CONTACT is described in table 1 and the
shape of the normal pressure is presented in figure 20c.
Comparing the contact patches resulting from corrected and uncorrected solutions one may
notice that the length to width ratio has changed. The normal pressure is almost the same for
corrected and uncorrected solutions. The normal loads of the three solutions are about the
same. The change of shape after correction influences the creep forces. This is the main effect
of the correction. The maximum normal pressure from the approximate solution is lower than

Figure 19. Corrected contact zone with surface quantities.

Figure 20. The shape of normal pressure distribution. (a) Uncorrected solution, (b) corrected solution and (c) from
program CONTACT.
476 J. Piotrowski and H. Chollet

Table 1. Results of calculations.

Uncorrected Corrected Program


Quantity solution solution CONTACT

L/W 1.36 1.51 1.43


Area (mm2 ) 153.7 153.7 154.0
N (kN) 72.41/2.85† 72.99/2.89†
Resultant 75.3 75.9 76.0
pn max (MPa) 837 847 1035
pt max (MPa) 316 337 372
Fx (kN) −4.69/−1.14† −5.17/−1.18†
Resultant −5.83 −6.35 −6.17
Fy (kN) −18.55/−0.031† −20.47/−0.035†
Resultant −18.58 −20.5 −19.72
Downloaded by [University of Arizona] at 00:42 15 November 2012

†In two separate zones.

that resulting from CONTACT. This is typical for approximate methods, see also figure 22 of
section 3.3 where comparison with the FEM method is shown.
Owing to the asymmetry of the contact patch in the direction along the rail, the spin creepage
generates not only lateral but also longitudinal creep force.
Further validation of the proposed correction is necessary and would need more compre-
hensive comparisons with CONTACT.

3.3 Ayasse and Chollet method for semi-Hertzian contact

A first approach to the approximation of the non-Hertzian contact was made by Chollet in
1995–1996. Two aspects were observed during numerical tests on elliptical cases:

• an interpenetration close to δ/2 was necessary to obtain the surface area determined by
Hertz’ method;
• to obtain proper a/b ratio, a correction was necessary; it was proposed that only the wheel
rolling radius should be corrected, then modifying the longitudinal curvature parameter A.

Numerical trials were performed in order to obtain the contact patches determined by
CONTACT in the S1002/UIC60 1:40 case. Similar to Kik and Piotrowski the method was
based on longitudinal strips with Hertzian characteristics, suggesting the algorithm name:
STRIPES and the method name: semi-Hertzian.
Non-elliptical cases reveal that it was necessary to modify the lateral curvature parameter
B for two reasons:

• between two close contact patches, value of B is very often negative and this required
replacement by a small positive number;
• at each side of the rail radius boundary, the pressure shows a discontinuity which is not
observed using exact methods such as CONTACT, where the shear is taken into account
between the strips. A solution is to smooth the B parameter.

The first use of this method has been to calculate the parameters of an equivalent single
contact as in Pascal’s method, see section 2.2, in order to use them in a Simulink multibody
model.
Similarly to Linder, the determination of the ai /bi shape ratio was expressed with the help of
the ‘virtual ellipse’ as different for each strip, but this ellipse width was virtual and determined
by smoothing the local curvatures B. Despite good results on the Manchester Benchmark [44],
Wheel–rail contact models 477

the Simulink model appeared to be much slower than Ayasse’s VOCO multibody software with
the multi-Hertzian contact developed in Fortran, described in section 2.3.
In order to include the semi-Hertzian description into that faster code, a new tentative study
was led by Chollet and Ayasse in a project supported by the IDR2 association (RATP – Paris
Metro, SNCF, CORUS rail). The basis of this work has been presented in ref. [1] and is
summarised here.
The first result of this research is to propose the alternative presentation of the tables describ-
ing the traditional Hertzian method. Although a and b relate to an ellipse, the parameters n and
m are determined by the curvatures A and B. With the parameter λ = A/B, b/a is replaced
by n/m in table 2.
Downloaded by [University of Arizona] at 00:42 15 November 2012

3.3.1 Correction of the shape factor given by the crude interpenetration. With the
interpenetration method, the crude geometrical semi-axis ag and bg are related to the normal
translation h0 , see figure 21, by h0 = ag2 A = bg2 B.

This interpenetration leads to the ratio bg /ag = λ, different from the Hertzian ratio
b/a except for the case λ = A/B = 1.
To obtain the correct ratio, it is proposed to compensate the local value of the curvature A,
by describing Ac with the simple formula

Ac (b/a)2 (n/m)2
= = ,
A λ λ
giving
n2 δ
h 0 = a 2 Ac = b 2 B = . (23)
r 1+λ
In the Hertzian case, this compensated curvature must be used for the calculation of the
length a of the ellipse. The last expression can be used in semi-Hertzian cases.

3.3.2 Transformation of the expressions from Hertzian to semi-Hertzian cases. More


generally, in order to pass from the Hertzian case to a semi-Hertzian one, it is necessary to
transform the expressions using elliptical parameters such as a or b to expressions based on
the interpenetration parameters, such as hi or ai .
Starting from the Hertzian case (figure 21), the ellipse is divided into independent longitu-
dinal strips of length ai and width y with the local compensated curvature Ac and unchanged
lateral curvatures giving B. Note that

hi = ai2 Ac (24)

is used to calculate the strip length ai with a sufficient accuracy.

Table 2. Coefficients of Hertz theory for θ = 0 to 180◦ .

θ0 0 5 10 30 60 90 120 150 170 175 180

λ = A/B 0 0.0019 0.0077 0.0717 0.3333 1 3.0 13.93 130.6 524.6 ∞


n/m 0 0.0212 0.0470 0.1806 0.4826 1 2.0720 5.5380 21.26 47.20 ∞
m ∞ 11.238 6.612 2.731 1.486 1 0.7171 0.4931 0.311 0.2381 0
r 0 0.2969 0.4280 0.7263 0.9376 1 0.9376 0.7263 0.4280 0.2969 0
478 J. Piotrowski and H. Chollet
Downloaded by [University of Arizona] at 00:42 15 November 2012

Figure 21. Local parameters of the elliptical contact zone.

3.3.3 Expression for stresses and forces. The shear stresses and traction bound expres-
sions from Kalker are:
   
3 4 n ai − x
σxi (x, yi ) = Gc11 νx − Gc23 ϕyi
8 π m a
   
3 2 n ai − x
σyi (x, yi ) = Gc22 νy + Gc23 ϕ(ai + x)
8 π m a
    2   
µN  y 2  x 2 µN x ai 2
i
µσzi (x, yi ) = 2 1− − =2 1−
πab b a π ab ai a

The compliance† L1 (similarly L2 ) can be expressed using the ratio ki = ai /a, giving

8a 8ai 8ai
L1 = = = . (25)
3c11 G 3c11 G(ai /a) 3c11 Gki

The value of ki can be expressed as

 a 2  y 2 hi
i i
ki2 = =1− = . (26)
a b h0

The Hertzian theory is used to express the ratio N/b

N 2 1 + λ Eh0
= ,
b 3 n3 1 − ν 2

† The notation L1 L2 L3 is from Kalker but the parameters Lx Ly Lz in equation (11) are the same.
Wheel–rail contact models 479

and to substitute it into expression for the traction bound. After making use of equation (26)
and some manipulations one obtains
  
3 x 

σxi (x, yi ) = Gc11 νxi 1 − ki 

8 ai 

    

3 2 n x 
σyi (x, yi ) = Gc22 νy + Gc23 ϕi m(ai + x) 1− ki . (27)
8 π m ai 

  2  

4 1 + λ E hi 

x 

µσzi (x, yi ) = µ 3 1 − k i 

3π n 1 − ν ai
2 ai

The expressions are only dependent on the strip parameters ai , hi , νxi , cij and µ, the local
Downloaded by [University of Arizona] at 00:42 15 November 2012

friction coefficient. Now the method can be applied to semi-Hertzian contact, where the
maximum virtual interpenetration on the main strip is h0 .
Note that the spin term has disappeared from the expression for σxi in equation (27), because
there is no lateral deviation yi relative to the centre of the strip. The spin effect is replaced
by the local longitudinal creepage νxi a priori different on each strip. This gives a deviation
relatively to FASTSIM.
Calculations using CONTACT and different finite element models have been made in
order to compare the pressure distribution, see figure 22. The correspondence depends on
the smoothing coefficients used for B, see [1].
A difference between the Ayasse and Chollet method and that of Kik and Piotrowski
(of 1996) is that the creep coefficients vary from strip to strip and not from patch to patch.

Figure 22. Contact patch and normal pressure profile for S1002/UIC60 1:40 pair under load 78.5 kN for wheelset
central position. Results from the FEM ANSYS model and STRIPES/VOCOLIN.
480 J. Piotrowski and H. Chollet

For this, the expressions of Kalker’s coefficients cij are given as a function of n/m, while in
the initial paper by Kalker they were expressed as a function of b/a, focusing the attention on
the ellipse shape but not on the curvatures.

3.3.4 Implementation in the dynamic codes. The initial STRIPES algorithm has been
implemented in the INRETS multibody code VOCOLIN with some additional improvements:
• the tables of data, function of the lateral displacement, are initially sampled regularly every
0.1 mm, for example. When there is no change observed between contiguous positions,
which means that the same strips are considered from step to step, the tables are compressed;
• the tangential stresses calculation are based on the FASTSIM formulation translated into
equation (27), but with a specific algorithm to accelerate the calculation along the strip.
Downloaded by [University of Arizona] at 00:42 15 November 2012

The results are generally similar when compared with the previous multi-Hertzian method.
But the physical interpretation of the new formulation looks a step ahead.
Several particular applications of the method have been made, which enlarge the application
field. Quost, in his PhD thesis [48], has made a variant implemented in a specific code, with
an additional roll degree of freedom between the wheel and the rail, in order to evaluate the
risk of derailment versus the risk of rollover of TGV in strong cross winds. Sebes and Ayasse
[49] have prepared a version with variable rail profiles to simulate traversing a switch at high
speed.
At the present time, extensions of the method are planed for some further cases:
• by taking into account the influence of the yaw angle on the strip longitudinal position;
• by description of the flangeback contact for tramways and in low speed switch traversing;
• when the profiles are theoretical, and when one is not interested in contact stresses, it will
be faster to use the same data tables but to simulate with the multi-Hertzian model.

4. Final remarks

Advanced modelling of rail vehicle dynamics requires realistic solutions of the contact prob-
lems for wheels and rails. The solutions of the contact problems have to be known practically
in each step of the numerical integration of the differential equations of the model.
One way to achieve this is pre-tabulation, but then the stored data can only contain parame-
ters of regularised contact patches, usually equivalent ellipses. With such a description, some
important properties, resulting from non-elliptical shape of the contact patch or multiple
patches are lost. One important property worth mentioning is the longitudinal creep force
due to spin, which is absent for a patch with two axes of symmetry, such as an ellipse.
The pre-tabulation does not require a very fast or completely reliable method of calculation.
The pre-tabulation may apply to one wheel and rail or to the wheelset and two rails of the
track.
The geometry of the real profiles of wheels and rails frequently leads to singularities, which
require special description. A realistic solution at a singular situation cannot be obtained solely
on the basis of the contact mechanics. It requires simultaneous consideration of the dynamic
equilibrium of the wheel/wheelset, which is a part of the bigger model. The basic singularities
demonstrate themselves as double and multiple contact patches.
During the pre-tabulation a singularity leads to a jump in the geometrical parameters,
positions of the contact patches and to the appearance of multiple contact patches. The
available methods to deal with these jumps within the model of vehicle system dynamics
Wheel–rail contact models 481

have been outlined in section 2. Enough experience gained allows using these methods with
confidence.
The solutions of the contact problems, applicable in modelling vehicle system dynamics
with the aim of preserving properties resulting from actual non-elliptical or multiple contact
patches, may only be partially tabulated due to the huge volume of data to be recorded and
later accessed. They have to be obtained online during the numerical integration of the model
equations. This requires that the calculation method used is completely reliable and fast enough
to allow nearly real time simulation, which is especially important for wear simulation. With
such strong requirements the methods cannot be ‘exact’ in the sense of the complete theory of
Kalker. Indeed, the methods outlined in section 3 of this paper are approximate ones. Owing
to specific assumptions, those methods are valid for so-called quasi-Hertzian cases, when the
contact conditions do not deviate much from the assumptions of Hertz. It is assumed that it is
Downloaded by [University of Arizona] at 00:42 15 November 2012

also worth using these methods in other cases. The methods may be used for one wheel and
rail and for a wheelset and two rails of the track.
It is probable that fast, approximate methods for solving contact problems for railway
applications will be in use for some years to come. The experience in applications would be
the basis for possible improvements of these methods.
However, some directions of possible development may be foreseen, which were also
mentioned in paper [23] on advanced methods of contact mechanics.
One useful development would be the inclusion of friction dependent on the local slip
velocity in the simplified solution of the tangential contact problem, when the normal contact
problem is solved with the use of the approximate method. This is connected with the problem
of temperature in contact [4, 5] and wear. Some works in this direction have already been
published [34, 52].
The other property, which would be useful to include in the framework of the simpli-
fied methods, is the roughness of the contacting surfaces [21]. This is important for vehicle
dynamics and wear.
As the fast, approximate methods are important for wheel/rail wear simulations, the possi-
bility of approximately accounting for inelastic displacements in on-line calculations would be
an important step towards improving wear simulations. Such a task would need comprehensive
introductory studies of the problem with the use of the finite element method.
In the last 20 years, from the time of the equivalent conicity approach to vehicle dynamics
up to now, the contact models have been generally complicated. At the same time the speed and
capacity of computers was developing. But something has not really changed, the challenge
of simulating vehicle system dynamics in real time in order to offer the railway industry an
efficient tool to help design and evaluate the performance of the railway systems of tomorrow.

References
[1] Ayasse, J.B. and Chollet, H., 2005, Determination of the wheel rail contact patch for semi-Hertzian conditions.
Vehicle System Dynamics, 43(3), 159–170.
[2] Duffek, W., 1982, Contact geometry in wheel/rail mechanics. Proceedings of Contact Mechanics and Wear of
Rail/Wheel Systems, Vancouver, Canada.
[3] Elkins, J.A., 1992, Prediction of wheel/rail interaction: the state of the art. In: G. Sauvage (Ed), The Dynamics
of Vehicles on Roads and on Tracks, Proceedings of the 12th IAVSD Symposium, Lyon, France, August 1991,
Vehicle System Dynamics, 20 (suppl.), 1–27 (Amsterdam/Lisse: Swets & Zeitlinger).
[4] Ertz, M. and Knothe, K., 2000, Die Temperaturentwicklung im Rad-Schiene-Kontakt und ihre Auswirkungen
auf Kraftschluss und Materialverhalten. Tagungsband zur VDEI Fachtagung BahnBau 2000, Berlin, 12–15
September, pp. 232–237.
[5] Ertz, M. and Bucher, E., 2001, Improved creep force model for wheel/rail contact considering roughness and
temperature. Paper presented at the 17th Symposium Dynamics of Vehicles on Roads and on Tracks, Copenhagen
(Lyngby), 20–24 August.
[6] Esveld, C. and Gronskov, L., 1996, MINIPROF wheel and rail measurement. In: I. Zobory (Ed). Proceedings
of 2nd Mini Conference on Contact Mechanics and Wear of Rail/Wheel Systems, TU Budapest, pp. 69–75.
482 J. Piotrowski and H. Chollet

[7] Gostling, R.J., 1971, The measurement of real wheel and track profiles and their use in finding contact conditions.
British Rail Research Technical Note DA 22.
[8] Gostling, R.J., Piotrowski, J. and Rose, K.A., 1981, Comparison of two theoretical methods of calculating
contact area between measured wheelset and track. British Rail Technical Memorandum TMDA 31.
[9] Hashemi, J. and Paul, B., 1979, Contact stresses on bodies with arbitrary geometry. Application to Wheels and
Rails. Report FRA-ORD 79/23 (Washington DC).
[10] Hertz, H., 1882, Uber die Beruhrung fester elasticher Korper. J. f. Reine u. Angew. Math., 156–171.
[11] Johnson, K.L., 1985, Contact Mechanics (Cambridge: Cambridge University Press).
[12] Kalker, J.J., 1967, On the rolling contact of two elastic bodies in the presence of dry friction. Dissertation, TH
Delft, Delft.
[13] Kalker, J.J., 1986, Users Manual of the Fortran program CONTACT. Delft University of Technology, Department
of Mathematics and Computer Science.
[14] Kalker, J.J., 1990, Three dimensional elastic bodies in rolling contact. (Dortrecht EA: Kluwer Academic
Publishers).
[15] Kalker, J.J., 1996, Book of tables for the Hertzian creep force. In: I. Zobory (Ed) Proceedings of Second Mini
Conference on Contact Mechanics and Wear of Rail/Wheel Systems, TU Budapest, pp. 11–20.
Downloaded by [University of Arizona] at 00:42 15 November 2012

[16] Kalker, J.J., 2001, Rolling contact phenomena – linear elasticity. In: B.O. Jacobson and J.J. Kalker (Eds), Rolling
Contact Phenomena, CISM Courses and Lectures, No. 411, Chap. l, (Wien, New York: Springer), pp. 1–84.
[17] Kalker, J.J. and Piotrowski, J., 1988, The elastic cross-influence between two quasi-Hertzian contact zones.
Vehicle System Dynamics, 17, 337–355.
[18] Kik, W. and Piotrowski, J., 1997, Einfluss des kontaktmechanischen Modells auf die Zugkraftentwicklung im
Bogen. Yortrag zum Workshop Bahntechnische Software und Knowhow aus Berlin, Berlin, October 24.
[19] Kik, W. and Piotrowski, J., 1996, A fast, approximate method to calculate normal load at contact between wheel
and rail and creep forces during rolling. In: I. Zobory (Ed), Proceedings of the Second Mini Conference on
Contact Mechanics and Wear of Rail/Wheel Systems. Budapest, Hungary.
[20] Kik, W. and Moelle D., 2005, Verschleiß in einer S-Bahn Wendeschleife, Messung und Simulation. Eisenbahn
Kalender, in press.
[21] Knothe, K. and Theiler, A., 1997, Normal and tangential contact problem with rough surfaces. In: I. Zobory (Ed)
Proceedings of the Second Mini Conference on Contact Mechanics and Wear of Rail/Wheel Systems, Budapest,
August 1996, pp. 34–43, Technical University of Budapest.
[22] Knothe, K. and Le, T.H., 1984, A contribution to calculation of contact stress distribution between elastic bodies
of revolution with non-elliptical contact area. Computers and Structures, 10(6).
[23] Knothe, K., Wille, R. and Zastrau, B.W., 2001, Advanced contact mechanics – road and rail. Vehicle System
Dynamics, 35(4–5), 361–407.
[24] Le, T.H., 1987, Normal- und Tangentialspannungsberechnung beim rollenden Kontakt fur Rotationskorper mit
nichtelliptischen Kontaktflachen. VDI Verlag, Dusseldorf.
[25] Linder, Ch., 1997, Verschleiss von Eisenbahnradern mit Unrundheiten. Dissertation ETH Nr. 12342.
[26] Linder, Ch. and Brauchli, H., 1996, Prediction of wheel wear. In: I. Zobory (Ed) Proceedings of the Second
Mini Conference on Contact Mechanics and Wear of Rail/Wheel Systems, TU Budapest, pp. 215–223.
[27] Pascal, J.P. and Sauvage, G., 1993, The available methods to calculate the wheel/rail forces in non Hertzian
contact patches and rail damaging. Vehicle System Dynamics, 22, 263–275.
[28] Piotrowski, J., 1982, A Theory of Wheelset forces for two point contact between wheel and rail. Vehicle System
Dynamics, 11(2), 63–87.
[29] Birger, I.A., 1968, Procnost’, ustojcivost’, kolebanija. Handbook in 3 volumes (Publishing House “Masinos-
troijenje”) (in Russian).
[30] Piotrowski, J., 1993, On an approach to prediction of contact loading of rail for two-point contact between wheel
and rail. In: J.J. Kalker et al. (Ed) Rail Quality and Maintenance for Modern Railway Operation, (Dordrecht,
Boston, London: Kluwer Academic Publishers), pp. 215–224.
[31] Piotrowski, J., 2001, On application of theory of rolling contact in modelling dynamic interaction between rail
vehicle and track. In: W. Szczesniak (Ed) Polska mechanika u progu XXI wieku, Ofic. Wyd. PW, 363–374.
[32] Piotrowski, J., 1998, Contact loading of a high rail in curves. Physical simulation method to investigate shelling.
Vehicle System Dynamics, 17, 57–79.
[33] Piotrowski, J., 1998, Calculation of creep forces for weakly curved area of contact. Proceedings of Xth Polish–
German Seminar, (Warsaw, Poland: Warsaw University of Technology).
[34] Piotrowski, J. and Kik, W., 2000, Wheel-rail creep forces with friction coefficient depending on sliding velocity.
Calculation with the simplified theory. Proceedings of Transcomp Conference, Pol. Radomska, pp. 217–224.
[35] Piotrowski, J., 1997, A description of contact between wheel and rail using pre-calculated data. Zesz. Naukowe
Inst. Pojazdow, 1(23), 7–32.
[36] Piotrowski, J., 1981, Non-Linear quasi-static curving theory for railway vehicles with coupled bogies and
two-point contact between guiding wheels and rails. British Rail Research, Technical Memorandum TM DA 36.
[37] Reusner, H., 1977, Druchflachenbelastung und Oberflachenverschibung in Walzkontakt von Rotationskorper.
Thesis Karlsrue, SFK Schweinfurt.
[38] Shen, Z.Y., Hedrick, J.A. and Elkins, J.A., 1983, A comparison of alternative creep force model for rail vehicle
dynamic analysis. Vehicle System Dynamics, 12, 79–82.
[39] Vohla, G., 1996, Werkzeuge zur Realitätsnahen Simulation der Laufdynamik von Schinenfahrzeugen. Disser-
tation ETH 11300, VDI Verlag Düsseldorf.
Wheel–rail contact models 483

[40] Aknin, P., Ayasse, J.B., Chollet, H., Maupu, J.L., Gautier, B. and Paradinas, M., 1993, Quasi static derailment
of a railway vehicle, comparison between experimental and simulation results. Proceedings of the 13th IAVSD
Symposium, Chengdu, Vehicle System Dynamics, 23, (suppl.).
[41] Ayasse J.B., Chollet H. and Pascal J.P., 1987, Modelisation simplifiee d’un bogie avec frottement sec. INRETS
report Nr. 33, ISBN 2-85782-179-4.
[42] Ayasse J.B., Chollet H., Maupu J.L., 2000, Paramètres caractéristiques du contact roue-rail. INRETS report Nr.
225, ISBN 0768-9756.
[43] Ayasse J.B., Chollet H., Fleuret J.S. and Lévêque, E., 2002, CAF – a generalised conicity criteria for the wheel rail
contact. Example of a switch blade safety study. In: I. Zobory (Ed) Proceedings of the Fifth VSDIA Symposium,
B.U.T.E, Budapest, November, in press.
[44] Iwnicki, S., 1999, The Manchester benchmark for rail vehicle simulation, Vehicle System Dynamics, 31 (Suppl.),
1–48.
[45] Pascal J.P. and Sauvage G., 1991, New method for reducing the multicontact wheel/rail problem to one equivalent
rigid contact patch. Proceedings of 12th IAVSD Symposium, Lyon, 26–30 August, (Lisse: Swets & Zeitlinger
B.V.).
[46] Pascal J.P., 1993, About multi Hertzian contact hypothesis and equivalent conicity in the case of S1002 and
Downloaded by [University of Arizona] at 00:42 15 November 2012

UIC60 analytical wheel/rail profiles’, Vehicle System Dynamics, 22(2), 57–78.


[47] Pascal J.P. 1993, The railway Dynamic Codes “VOCO”, 137–139, and “Benchmark to test Wheel/Rail Contact
Forces”, Nov. 1990, Reports from the IAVSD workshop, Herbertov. Vehicle System Dynamics 22, (suppl.),
pp. 169–173.
[48] Quost, X., 2005, Sensibilité des motrices TGV au vent traversier, PhD thesis, Ecole Centrale de Lyon.
[49] Sebes, M., Ayasse, J.B.P., Pouligny, P. and Pirat, B., 2005, Application of a semi-Hertzian method to the
simulation of vehicles in high speed switches, XIX IAVSD Symposium, Milano.
[50] Shen G., Ayasse J.B., Chollet H. and Pratt I., 2003, A unique design method for wheel profiles by considering
the contact angle function, Proceedings of the Institution of Mechanical Engineers Part F: Journal of Rail and
Rapid Transit, 217, No. F1.
[51] Piotrowski, J. and Kik, W., 1999, The influence of spin on creep forces for non-flat contact area. Vehicle System
Dynamics, 31, (Suppl.), 158–177.
[52] Nielsen, J.B. and Theiler, A., 1996, Tangential contact problem with friction coefficients depending on sliding
velocity. Proceedings of Second Mini Conference on Contact Mechanics and Wear of Rail/Wheel Systems.
I. Zobory (Ed) TU Budapest, pp. 44–51.

You might also like