You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/239406897

Rolling contact fatigue of rails - Finite element modelling of residual stresses,


strains and crack initiation

Article in Proceedings of the Institution of Mechanical Engineers Part F Journal of Rail and Rapid Transit · January 2000
DOI: 10.1243/0954409001531207

CITATIONS READS

57 761

4 authors, including:

Jonas W. Ringsberg B.L. Josefson


Chalmers University of Technology Chalmers University of Technology
287 PUBLICATIONS 3,629 CITATIONS 70 PUBLICATIONS 2,242 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Jonas W. Ringsberg on 08 November 2019.

The user has requested enhancement of the downloaded file.


7

Rolling contact fatigue of railsÐfinite element


modelling of residual stresses, strains and crack
initiation

J W Ringsberg1 , H Bjarnehed2 , A Johansson2 and B L Josefson1 


1
Department of Solid Mechanics, Chalmers University of Technology, GoÈteborg, Sweden
2
Frontec Institute of Science, FIoS, GoÈteborg, Sweden

Abstract: Numerical models that can be used to evaluate crack initiation in rails owing to rolling contact
loads are of great value. Very few models on rail fatigue assessments exist in the literature that consider the
non-linear stress and strain response in the rail caused by both global dynamic track response and local
wheel±rail contact loads. In the present investigation, such a tool has been developed using finite element
(FE) models. The tool can be used to calculate governing fatigue conditions such as residual stresses,
plastic strains and orientation of crack planes for crack initiation. The ability to perform parameter studies
of, for example, track properties, different materials and load cases is inherent in the tool. The tool was
used in a case study of commuter train traffic at a Swedish test site. In the FE simulations, a material
model with non-linear isotropic and kinematic hardening, which was able to consider ratchetting behaviour,
was used. Ratchetting material response near the railhead surface was found. A criterion for fatigue crack
initiation caused by ratchetting and a criterion for fatigue crack initiation caused by low-cycle fatigue
(LCF) were compared. The LCF criterion showed the lowest number of cycles to crack initiation, which,
given the models used, indicates that cracks will initiate owing to LCF rather than ratchetting. Good
agreement for the direction of surface cracks between the rail track at the test site and the results from the
simulations was achieved.

Keywords: finite element modelling, rolling contact fatigue, residual stresses, crack initiation, critical
planes, ratchetting

NOTATION Nf number of cycles to crack initiation (cycles)


p0 maximum normal contact pressure (N=m2 )
a9 longitudinal semi-axes of the Hertzian contact qi contact shear stress distribution in the i
ellipse (m) direction ˆ ì i p(x, y), where i ˆ x, y (N=m2 )
b fatigue strength exponent p(x, y) normal contact pressure distribution (N=m2 )
^b material parameter in the material model Pz normal contact force (N)
c fatigue ductility exponent Qi contact shear force in the i direction ˆ ì i Pz ,
c9 transversal semi-axes of the Hertzian contact where i ˆ x, y (N)
ellipse (m) Q1 material parameter in the material model (N=m2 )
^c material parameter in the material model (N=m2 ) R curve radius at the centre of gravity of the rail
cdi viscous damping coefficient in the i direction, cross-section (m)
where i ˆ x, y, z (N s=m) Rr rail radius at the railhead top at the axis of
d depth below the railhead surface (m) symmetry (m)
E elastic modulus (N=m2 ) Rw wheel radius at the running circle (m)
G shear elastic modulus (N=m2 ) v rolling direction of the running vehicle
ksi spring stiffness in the i direction, where i ˆ x, y, (x, y, z) local coordinate system in the rail
z (N=m) (X, Y, Z) global reference coordinate system with origin in
the centre of gravity of the rail cross-section

The MS was received on 9 September 1999 and was accepted after á backstress tensor, locus of the yield surface in
revision for publication on 13 October 1999. e
 Corresponding author: Department of Solid Mechanics, Chalmers stress space (N=m2 )
University of Technology, HoÈrsalsvaÈgen 7, SE-412 96 GoÈteborg, Sweden. árail cant angle (deg)
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
8 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

ârail rail inclination angle (deg) expected to decrease in number owing to improved rail
ã material parameter in the material model making technology, is not the main objective of this
Äi prescribed translation in the i direction, where work.
i ˆ X , Y, Z (m) Rolling contact fatigue has been studied extensively in
Ä~ã average shear ratchetting strain per cycle the literature [1], and a recent overview is given in the
Äår equivalent ratchetting plastic strain per cycle textbook by Suresh [2]. The different initiation mechan-
Ä~å average axial ratchetting strain per cycle isms occurring in railway tracks are described in Beynon
Äå1 =2 maximum principal strain amplitude et al. [3]. With engineering models for rolling contact
Äó1 =2 maximum stress amplitude on the maximum fatigue, it is possible to establish shakedown maps where
principal strain plane (N=m2 ) regions with different initiation mechanisms due to elastic
åc critical strain for failure by ratchetting obtained shakedown, plastic shakedown or ratchetting can be
from experiments identified for different combinations of friction coefficient
å9f axial fatigue ductility coefficient and maximum contact pressure. The material model
åpl
acc accumulated effective plastic strain according to normally employed is elastic±ideally-plastic, which
von Mises neglects hardening and allows a large number of cycles to
å pl |0 initial plastic strain be evaluated [4]. Jiang and Sehitoglu [5] use a semi-
å_ pl plastic strain rate tensor analytical model that only considers normal loads, together
e
è angle defining the crack plane (deg) with an advanced elastic±plastic material model that
ë plastic multiplier, equivalent plastic strain enables the analysis to include simulation of long-time
ìi coefficient of friction in the i direction, where ratchetting behaviour. However, the semi-analytical models
i ˆ x, y only give information about the stress and strain fields that
r mass density (kg=m3 ) are caused by the local contact loads. For the case of a
ó stress tensor (N=m2 ) railway track subjected to contact loads from railway
e
óy yield stress of the material (N=m2 ) wheels, there is also a global load component acting on
ó0 yield stress (size of the yield surface in stress the rail material.
space) in the material model (N=m2 ) By using non-linear finite element (FE) analyses, the
ó|0 initial yield stress in the material model (N=m2 ) residual stress and plastic strain fields can be calculated
ómax
1 maximum stress on the maximum principal in detail [6]. However, the computation time is normally
strain plane (N=m2 ) so long that only a few cycles may be covered in a
ó9f axial fatigue strength coefficient simulation. Furthermore, there are very few models in the
í Poisson's ratio literature on fatigue crack initiation in railway tracks. Fry
j angle defining the crack plane (deg) et al. [7] have presented a model consisting of a two-
Öi prescribed rotation in the i direction, where layered elastic half-plane. The upper layer constitutes the
i ˆ X , Y, Z (rad) rail and the lower layer represents the support system
beneath the rail. Local stresses are calculated using the
Hertz theories, and the critical planes for crack initiation
are determined in an elastic analysis assuming most
cracks are initiated subsurface where the stress field is
elastic, or at least where elastic shakedown conditions
1 INTRODUCTION prevail.
In the present investigation, a tool is developed for
Rolling contact fatigue of rails is a severe and growing the FE simulation of residual stresses and plastic strains
problem for many railways in the world. The annual cost in the rail owing to global bending and local contact
for reprofiling and repairing rail damage is very high. forces. The tool will provide detailed information of the
There are typically two sorts of surface initiated cracks stress and strain fields in the railhead, where rolling
that appear on the rail owing to rolling contact fatigue, contact fatigue cracks may develop. The proposed tool
and these are commonly known as head checks and employs a non-linear isotropic and kinematic hardening
squats; the former are associated with the high rail in material model that is able to consider ratchetting
curves and the latter with any rail in relatively straight behaviour. A specific load case simulating commuter
track. This type of damage does not originate from train traffic is studied, based on the real situations at a
metallurgical defects, but instead seems to be the result Swedish test site. The load case consists of four force
of an increasing traffic density and increased axle loads systems, corresponding to the passage of two bogies
on the railway lines. Wheel±rail contact may also lead to with two wheelsets each and an axle load of 12 t. The
the initiation of subsurface fatigue cracks which are often average velocity at the test site is 75 km=h, and the
associated with metallurgical faults, for example gauge friction coefficient was taken as 0.20 in the rolling
corner shelling and the detail fracture. Modelling the direction and as 0.41 in the transverse direction of the
initiation and growth of these types of cracks, which are rail.
Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 9

2 DESCRIPTION OF THE TOOL DEVELOPEDÐ [10]. The tool presented in the present investigation
FE MODELS consists of two FE models: a `track model' and a `rail
model'. The track model describes the global dynamic
The tool should consider both the dynamic bending effects behaviour of the track, and the rail model has an element
of the rail owing to `global' response from the track, and resolution that considers local distributions of stresses and
the `local' effects from the wheel±rail contact at the strains near the railhead surface. The models are connected
railhead surface. Numerical investigations show that by time dependent boundary conditions determined from
approximately 20 m of track should be modelled to the track model, which are applied to the rail model.
calculate the global dynamic behaviour of the track Therefore, the breakout model modelling technique is used.
accurately. The dynamic response at the middle of such a Section 2.2 describes the track model and Section 2.3
track model has been found to be almost independent of the describes the rail model, in detail.
applied boundary conditions at the track ends. Hence, the
model has to represent a relatively large number of sleeper 2.2 Track model
bays, and also be able to resolve very fine local distribu-
tions of stresses and strains. The dimension of the contact The track model has 32 sleeper bays (one sleeper
area between the wheel and the rail is of the magnitude of distance ˆ 0:60 m), and the track is assumed to be
100±500 mm2 , and the thickness of the top layers exposed horizontal and straight. A longitudinal plane of symmetry
to high gradients of stresses and strains is just a few is used in such a way that only one side of the track is
millimetres. Therefore, one requirement of the tool is that modelled. Hence, symmetry in loading between the two
it must be able to resolve the gradients of the stresses and rails is assumed. In Fig. 1, one sleeper of the track model is
the strains, from the wheel±rail contact, near the railhead presented, with the elements for the rail, the pads and the
surface. The material models used in the tool must also be ballast material. The rail and the sleepers are modelled by
able to consider elastic±plastic material response resulting undamped two-node linear Timoshenko beam elements
from the wheel±rail rolling contact. In addition, both FE using the I-DEAS beam section module, which can provide
models are created in the commercial FE package I-DEAS the elements for any desired shape of the rail and the
Master Series [8]. The model files are exported as input sleeper cross-sections. Thus, in principle, only the Saint-
files to the commercial FE code ABAQUS version 5.7 [9] Venant torsion of the rail is modelled. In addition, all the
by the interface provided by I-DEAS. Owing to the material in the track model is modelled as linear±elastic.
capability of ABAQUS to model (user-defined) elastic± The pads and the ballast material are modelled using
plastic material models, ABAQUS is used as the solver. linear spring±damper systems with equal properties in
both tension and compression. Thus, no load dependence,
or zero load transmission, when no load is present is
2.1 Submodelling technique
included in the model. The damper elements are
The modelling problem described is preferably treated by described by viscous damping; i.e. the force transmitted
`substructure' or by `breakout model' modelling techniques through a damper element is proportional to the relative

Fig. 1 Track model: one sleeper with the rail cross-section and the elements for the pad and the ballast material
(yellow elements: springs and dampers; green elements: rigid beam elements; blue colour: sleeper and rail
cross-sections)
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
10 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

velocity between its end-points. The system representing 2.3 Rail model
the pad attaches the rail to the sleepers by a discrete
The rail model represents 12 cm of a track arbitrarily
distribution in two dimensions of springs and dampers,
chosen and cut along the rail in the middle between two
using node-to-node elements over the connecting region.
sleepers. The model is divided into two parts: the `inner
A similar distribution is applied between the sleepers and
part' and the `outer part' (see Fig. 2). The inner part
the ballast material, but here node-to-ground elements are
used. The springs and the dampers are parallel coupled in incorporates 8 cm of the top of the railhead in the rolling
direction, and it is defined to a depth of 1.5 cm. An
three orthogonal directions at every discrete material
elastic±plastic material model (see Section 2.4) defines the
location. The spring±damper systems modelling the pads
material in the inner part, since this is the volume of the
and the ballast material are coupled by a two-dimensional
rail model of interest for the local material response from
distribution of a system of rigid elements, which is
the wheel±rail rolling contact. The outer part represents the
distributed through the centre of gravity of the sleeper
elastic surroundings of the rail, where the results show
beam.
elastic material response and are of less interest (in the
A system of rigid elements, similar to the system
modelling the connection between the pads and the ballast present investigation). The material within the volume of
the outer part is therefore modelled using a linear±elastic
material, enables the wheel contact forces to be applied at
material model. Hence, only a thin material layer in the
the top of the railhead. Rigid beam elements distribute the
vicinity of the wheel±rail contact region is assumed to
contact force from the top of the railhead, via the centre of
experience plastic deformation. In addition, both the parts
gravity of the rail beam, to the spring±damper system
are meshed by solid brick elements using a mapped option,
modelling the pads. The loads from the vehicle are
i.e. a so-called structured mesh. The inner part has a larger
represented by a distribution of point forces. Each point
mesh density in comparison with the outer part, whose
force is defined by three mutually orthogonal and, over
time, constant components, corresponding to the rail mesh density gradually decreases downwards from the
railhead surface. The mismatch between the element
normal, the longitudinal and the transverse force. Hence,
meshes in adjoining surfaces between the inner part and the
only one-point contact for each wheel is assumed in the
outer part is easily handled in ABAQUS by constraint
track model. The forces travel as a collective with a defined
equation routines. The constraints cause adjoining surfaces
velocity along the rail in the rolling direction. Hence, no
to deform into equal shapes.
inertia or damping forces from the vehicle are considered.
The distributions of contact pressure and shear stresses
However, a magnification factor applied to the wheel axle
from the wheel±rail contact are applied on the top surface
load can be used for approximate inclusion of dynamic
effects from the vehicle. Furthermore, symmetric boundary of the inner part. This investigation only treats one-point
contact (although one-point contact is an unlikely condition
conditions at the plane of track symmetry are applied to the
in a curved track), which is located at the symmetry line in
track model, and the track ends of the model are considered
the rolling direction. The `geometry-based' mesh option in
as free.
I-DEAS provides the capability to define arbitrary shapes
In principle, there are two methods that can be used to
of contact pressure distributions. Either a function using
solve the stated dynamic problem: the mode superposition
coordinates as variables or a table definition is offered as
method and the direct integration method. Numerical
investigations showed that the mode superposition method
was not appropriate for the present model, as, to achieve
accurate results, unreasonable numbers of modes had to
be included in the solution. Therefore, the direct integra-
tion method was chosen. In addition, the direct integra-
tion method is also available for non-linear dynamic
problems.
The modelling technique behind the track model can
easily be used to model a more general FE model of a
track that is not based on assumptions of symmetry of
the track, e.g. a model of a part of a track in a curve.
Also, non-linear material models describing the proper-
ties of the pads and the ballast material, such as load
dependent stiffness, can easily be included. In addition,
the software packages DIFF [11] and GENSYS [12]
treat the dynamic interaction between the vehicle and
track more generally. These software packages treat the
multibody train±track interaction, and an application
example and description of DIFF are presented in Fig. 2 Rail model with a cut to present the inner part (high mesh
reference [13]. density) and the outer part (low mesh density)
Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 11

an input alternative. In the present investigation, a function material model is subsequently referred to as the `Cha-
using coordinates as variables was used to describe the boche model'. The Chaboche model is able to simulate a
contact pressure and the shear stresses as Hertz distribu- ratchetting material response and is described by the von
tions. Furthermore, all degrees of freedom are restrained at Mises yield criterion, an associative flow rule, and non-
one cross-section of the rail model, while at the other linear isotropic and kinematic hardening laws. The non-
cross-section a set of six components (three translations linear isotropic and kinematic hardening laws proposed by
and three rotations) of prescribed displacements are Armstrong and Frederick [15, 16] are as follows:
applied. The prescribed displacements are applied at the
centre of gravity of the rail cross-section, and they are ^
ó 0 ˆ ó j0 ‡ Q1 (1 ÿ e ÿ bë ) (1)
calculated using the results from the track model as the
difference in displacement response between two cross-
sections at a distance of 12 cm. For every position of the ó ij ÿ á ij
contact load on the rail in the track model, corresponding á_ ij ˆ ^cë_ _ ij
ÿ ãëá (2)
ó0
sets of prescribed displacements are applied to the rail
cross-section in the rail model. The rail cross-section is A more thorough description of the Chaboche model with
assumed to be restrained as in the beam theory. This was details of the numerical treatment for uniaxial stress states
achieved by coating the faces of the solid elements at the is presented in reference [17]. Moreover, to determine
free rail cross-section with very stiff shell elements. proper values for the material parameters in the Chaboche
model, an optimization process using Matlab [18] was
carried out. In the Chaboche model there are five
2.4 Material model
parameters that can be optimized (^c, ã, ^b, Q1 and ó|0 ).
The material response of an elastic±plastic material to These were calibrated to uniaxial experimental data from a
cyclic loading can be divided into four categories (see Fig. cyclic tension±compression experiment on a cylindrical
3). At sufficiently small loads, so that no element of the specimen. The experiment was carried out by Bower and is
structure reaches the yield point, the response is perfectly described in reference [19]. The optimization routine used
elastic and reversible (a). If plastic flow takes place on the was the downhill simplex method [20], which minimizes
first loading, residual stresses, strain hardening or de- the objective function, i.e. the square root of the sum of the
formed contact geometry may `shake down' the structure to squared differences between experimental strain values and
a perfectly elastic response in the cyclic steady state. The the corresponding numerically computed strains. Thus, the
maximum load for which this is possible is known as the optimization was carried out in a least-squares sense. In
`elastic shakedown limit' (b). Above the shakedown limit, this method of calibration, all the parameters necessary to
plastic deformation takes place for every loading cycle. If describe the Chaboche model are determined simulta-
the material response is a closed cycle of plastic strain, the neously [21].
state is referred to as `plastic shakedown' (c). Increasing It is important to calibrate the material model for the
the load above the plastic shakedown limit may result in similar type of loading that is to be used in, for example,
ratchetting material response, where repeated increments of FE simulations. Optimization using uniaxial experimental
unidirectional plastic strain are accumulated at every cycle data is not the best option for FE simulations that show
of loading (d). multiaxial stress state response. However, uniaxial data are
Material models are described by constitutive equations often used when calibrating, as uniaxial experiments are
governing the stress or the strain response for a material for easy to perform. In the present investigation, the initial
a given stress or strain history. The material model used in yield stress ó|0 was determined from experiments as
the present investigation has been developed by Chaboche ó j0 ˆ 406 3 106 N=m2 . Therefore, only four of the materi-
and Lemaitre [14] and is available in ABAQUS [9]. The al parameters describing the Chaboche model in equations

Fig. 3 Illustration of different material responses that may occur for a material subjected to stress-controlled cyclic
loading with a constant non-zero mean load
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
12 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

(1) and (2) were determined by optimization: ^c ˆ 1:32 3 r ˆ 7820 kg=m3 and the radius of the railhead at the axis
1010 N=m2 , 㠈 3:12, ^b ˆ 3:97 and Q1 ˆ 1:52 3 of symmetry is Rr ˆ 0:3 m.
108 N=m2 . Based on uniaxial experimental data from 
The curve radius at the test site in AlvsjoÈ is R ˆ 303 m,
reference [19], the optimization was performed from cycles the track gauge of the rail is 1.44 m and the cant is 0.15 m.
1, 2, 15, 50, 100, 200, 400 and 600. The development of In Fig. 4, the definitions of significant track angles are
the tool used for the calibration is described in reference presented. The magnitudes of the angles árail and ârail for
[17] where also optimizations to the Bower experiment 
the high rail at Alvsjo
È are árail ˆ 5:958 and ârail ˆ 1:918.
[19] are carried out. The rail is fastened to the sleepers with Pandrol
fastenings with rail pads between the rail and the sleepers.
The rail pads are 10 mm thick Pandrol studded natural
rubber rail pads, which have a width of 0.15 m and a length
3 APPLICATION: TRACK SUBJECTED TO
of 0.16 m. The sleepers are 60 kg concrete monoblock
COMMUTER TRAIN TRAFFIC
sleepers S7 (StraÈngbetong S7), and the sleeper spacing at
the test site is 0.6 m. The material data for the sleeper
The FE models are applied to a case study of commuter
 material are E ˆ 41:3 3 109 N=m2 , G ˆ 17:2 3 109 N=m2
train traffic at a track test site in Alvsjo
È in Sweden. At this
and r ˆ 2400 kg=m3 . The ballast material consists of
test site, head checks have been found near the rail gauge
broken stones, and the properties of the ballast material and
corner (especially) on the high rail in one unlubricated
the rail pads are presented in Table 3 [25]. The properties
curve. It was therefore of interest to simulate the train
 are presented in the local coordinate system directions
traffic at AlvsjoÈ with the corresponding loads and material
(x, y, z) for a loaded track.
properties for the track.

3.1 Train and track properties at the Swedish test


site
The statistics for one year of train traffic at the test site are
presented in Table 1. Trains X1 and X10 are trains for
commuter train traffic [22]. The average speed of the
commuter trains at the test site is 75 km=h. According to
the statistics in Table 1, the X1 train traffic dominates the
traffic volume. Hence, the X1 train is chosen as a
representative train in the FE simulations, where the axle
load (120 3 103 N) for the motor coach was chosen as the
representative load. The wheel radius at the running circle
for the X1 motor coach is Rw ˆ 0:46 m.
The rail profile at the test site is UIC60 with a grade 900
rail steel material. In this investigation, the rail material in
the FE simulations is substituted by another pearlitic rail
steel, BS11 normal grade, since the mechanical properties
of this material are well documented in the literature Fig. 4 Definition of significant track angles árail and ârail for the
[23, 24] (see Table 2). The mass density of the rail steel is È lvsjoÈ
high rail at the test site in A

È lvsjoÈ, Sweden
Table 1 Statistics for the train traffic at the test site in A
Number of wheel pair passages per year with different
axle loads (1 3 103 N)
1 3 1010 N
Train traffic per year 200 150 120 105 75
X1 5.11 0 0 262 108 0 262 108
X10 2.56 0 100 218 0 100 218 0
Freight traffic 0.1 5000 0 0 0 0
 Based on an `average train' with three motor coaches and three trailing coaches.

Table 2 Mechanical properties of the pearlitic rail steel BS11 normal grade
ó y (N=m2 ) E (N=m2 ) í b c å9f ó 9f (N=m2 ) åc
406 3 106 209 3 109 0.29 ÿ0.089 ÿ0.559 0.103 936 3 106 11.5
 See Tyfour et al. [23].

Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 13

Table 3 Properties for the ballast material and the rail pads for a loaded track [25]
ksx (N=m) ksy (N=m) ksz (N=m) cdx (N s=m) cdy (N s=m) cdz (N s=m)
6 6 6 3 3
Rail pad 40 3 10 40 3 10 200 3 10 2.5 3 10 2.5 3 10 12.7 3 103
Ballast material 250 3 106 250 3 106 500 3 106 398 3 103 398 3 103 796 3 103

3.2 Load distribution in the track and the rail tions q i are calculated as q i ˆ ì i p(x, y), where ì x ˆ 0:20
models and ì y ˆ 0:41.
The load distribution in the track model is modelled as a p
p(x, y) ˆ p0 1 ÿ (x=a9)2 ÿ ( y=c9)2
moving force model, which simulates a running train
vehicle [26]. Hence, no inertia or damping forces from the
p0 ˆ 3Pz =(2ða9c9)
train are considered. A trainset consisting of two coupled
X1 motor coaches is studied, where the loads from the (3)
wheels of the bogie of the leading motor coach in the
traffic direction and the loads from the wheels of the
bogie of the trailing motor coach are considered. The
loading of the track is assumed to be symmetric with 3.3 Results from FE simulations
respect to an imaginary centre-plane between the two
rails. Consequently, only one side of the wheel pairs is The results from the track model were used to calculate
considered. the displacements between two rail cross-sections sepa-
The moving force model consists of four moving force rated by 12 cm. The two cross-sections were chosen in the
systems, which are moved with a velocity of 75 km=h. middle between two sleepers, located in the middle of the
Each of the force systems consists of three mutually track model, to prevent the sleepers and the boundaries of
orthogonal forces (Qx , Q y , Pz ), simulating the loads from the track model from affecting the results. In Fig. 5, the
one wheel in the longitudinal, the transverse and the relative displacements between the rail cross-sections
vertical (normal) direction. The magnitudes of the three resulting from one wheel passage are presented. These
forces are here presented as the forces on the high rail at results were used in the rail model as prescribed displace-
the test site for a running X1 motor coach. The forces ments applied at the centre of gravity of the free cross-
were calculated using the commercial codes MEDYNA section (see Section 2.3). Each circle represents the
and CONTACT. The magnitude of the vertical force is magnitude of a prescribed displacement for a specific load
Pz ˆ ÿ74:26 3 103 N, which corresponds to a dynamic step in the rail model; the circles symbolize the changes
factor of 1.26 of the moving vehicle, calculated on the in the displacements as the contact load distributions pass
basis of its static load. The dynamic factor considered is over the rail model.
the result of a combination of curving at non-equilibrium The FE simulation with the rail model was performed for
speed, vertical misalignment of the track top and wheel- 16 wheel passages. The material response near the railhead
flats. The longitudinal and transverse forces that arise surface from the wheel passages was ratchetting material
owing to, for example, friction forces and flange contact response, i.e. continuous accumulation of plastic strain
are Qx ˆ ÿ14:96 3 103 N and Q y ˆ 30:74 3 103 N. The after every wheel passage, with a low ratchetting rate. In
friction coefficients between the wheel and the rail were Fig. 6, the residual accumulated effective plastic strain
calculated on the basis of the forces Qx , Q y and Pz as according to von Mises [see equation (4)] is presented in
ì x ˆ 0:20 and ì y ˆ 0:41. the railhead of the rail cross-section after 16 wheel
The rail model only considers one-point contact between passages. The maximum accumulated effective plastic
the wheel and the rail. The contact position is fixed at the strain occurred at a depth d ˆ 2:8 3 10ÿ3 m below the
centre of the railhead on the axis of symmetry, and the load railhead surface.
distribution is modelled according to the Hertz theory for r
t
rolling contact between two elastic non-conforming solids 2 pl pl
åpl
acc
pl
ˆ å j0 ‡ å_ : å_ dt (4)
[1]. The distribution of the normal pressure p(x, y) is given 0
3e e
in equation (3), where the semi-axes of the contact ellipse
were calculated to a9 ˆ 6:25 3 10ÿ3 m and c9 ˆ 5:07 3 The residual von Mises effective stress in the railhead of
10ÿ3 m, using equations for elliptic integrals [27], the rail the rail cross-section is presented in Fig. 7 after 16 wheel
and wheel geometries and the axle load of the X1 motor passages. The maximum of the effective stress is located at
coach. The maximum normal pressure p0 is calculated as the same depth as the maximum of the accumulated
p0 ˆ 1119 3 106 N=m2 . The contact shear stress distribu- effective plastic strain in Fig. 6, i.e. at d ˆ 2:8 3 10ÿ3 m.
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
14 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

Fig. 5 Results from a track model simulation, presented as relative displacements between two rail cross-sections at
a distance of 12 cm

Fig. 6 Residual accumulated effective plastic strain according to von Mises in the railhead of the rail cross-section
after 16 wheel passages

4 CRACK INITIATION CRITERIA calculate the number of cycles to crack initiation at


ratchetting material response, given by
The material response determines which crack initiation
q
p
model should be used to estimate the number of cycles to
crack initiation. The FE simulation on the rail model N f ˆ åc =Äår , Äår ˆ (Äe å )2 ‡ (Äe ã= 3 ) 2 (5)
resulted in ratchetting material response for the tested load
case. In reference [28], a criterion was proposed to Ratchetting and low-cycle fatigue (LCF) can be treated as
Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 15

Fig. 7 Residual von Mises effective stress (N=m2 ) in the railhead of the rail cross-section after 16 wheel passages

independent and competitive modes for fatigue crack the maximum value of the product between ó max 1 and Äå1
initiation [28]. The criterion in equation (5) was therefore during one wheel passage for all possible planes at every
compared with an LCF crack initiation criterion with material point. The most damaged or critical element was
respect to the number of cycles to crack initiation. The found at the point of maximum accumulated plastic strain
initiation criterion that is satisfied in the smallest number (see Fig. 6). The number of cycles to crack initiation was
of cycles determines the failure of the material. The calculated using the ratchetting criterion as Nf ˆ 23 150
Smith±Watson±Topper (SWT) relation [30] [see equation cycles, and using the LCF criterion as Nf ˆ 2600 cycles.
(6)] can be interpreted as an energy-based approach and it Crack initiation will therefore occur because of LCF rather
is the LCF criterion that Ringsberg et al. [29] recommend than ratchetting. In addition, the values of the material
in rolling contact fatigue problems. The SWT relation parameters in equation (6) for the mechanical properties
allows mean stresses to be taken into account by ómax 1 in presented in Table 2 are taken from experiments that do not
equation (6), which is the maximum stress on the maxi- consider compressive stresses. The very limited experimen-
mum principal strain plane. However, this criterion is only tal data found in the literature indicate that compressive
valid for ó max
1 . 0. In rolling contact fatigue problems, ómax
1 stresses may influence material data. However, it has not
is very often negative during each cycle owing to the been possible to quantify such an influence for the present
compressive loads. In reference [29], a modification of rail steel. Moreover, the value of the critical strain to
ómax
1 according to reference [31] was suggested in order to failure, åc , in Table 2 was determined by twin-disc tests
use the criterion on problems with compressive mean conducted by Tyfour et al. [23]. Hence, the influence from
stresses [see equation (7)]. In the modification of ómax 1 , compressive stresses is considered in this mechanical
only the stress amplitude Äó 1 =2 is considered and the property and in the determination of the number of cycles
contribution from the (compressive) mean stress óm 1 is to crack initiation using equation (5). The maximum
neglected. This modification was used in the evaluation of contact pressure p0 was 1500 3 106 N=m2 in the tests
the results from the FE simulation: performed by Tyfour, and 1119 3 106 N=m2 in the present
FE simulation. The change in åc owing to the difference in
the maximum contact pressure in the test and the FE
ó max 2 2b
1 (Äå1 =2) ˆ [(ó f9 ) =E](2N f ) ‡ ó f9 å9f (2N f )
b‡c
(6) simulation has not been considered. The performed
calculation using åc from the Tyfour tests gives conserva-
tive results with respect to fatigue crack initiation life in the
ó max ˆ (ó m present case.
1 1 ‡ Äó 1 =2),
The crack plane (critical plane) for crack initiation was
Modification ) ó max
1 ˆ Äó 1 =2
calculated using the results from the FE simulation. The
(7) crack plane is here defined as the material plane that
accumulates the largest amount of damage according to
The criteria were applied to all elements in one rail cross- equation (6) during cycling. The angles that define the
section for every load step in the rail model. The damage direction of the crack plane are presented in Fig. 8, where v
parameter ó max
1 Äå1 in equation (6) was calculated, as denotes the rolling direction of the running vehicle. Angle
recommended by Socie [32] and Jiang and Sehitoglu [5], as è is the angle between the x axis and the crack plane, and j
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
16 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

Fig. 8 Definition of angles è and j for the crack plane (critical plane) for crack initiation (v denotes the rolling
direction of the running vehicle)

is the angle between the normal, ~n, of the crack plane and maximum stressed and strained point corresponds well with
the x axis seen from above. The angles were calculated to the results from shakedown maps (see, for example,
be è ˆ 348 and j ˆ 308. These are the (approximate) references [1] and [33]), where different initiation mechan-
angles that define the orientation of head checks, which isms are shown for different combinations of maximum
have occurred at the rail gauge corner at the test site in normal contact pressure and friction coefficient ì. The
AÈ lvsjoÈ. In practice, angle è is found by sectional cuts in the shakedown map shows that, for ì . 0:3±0:4 and the present
longitudinal rail direction, and angle j can be determined magnitude of the normal contact pressure, initiation will
by visual inspection at the rail gauge corner at the test site. take place at the surface, while, for lower values of ì,
initiation will take place subsurface. However, fatigue
cracks almost always initiate at the surface in railway rails,
while the present approach predicts crack initiation at some
5 DISCUSSION distance below the surface. The main cause for this
discrepancy is the low value of the friction coefficient (in
The presented FE models are a useful tool making it the longitudinal direction) used in the FE simulation.
possible to study the behaviour of a track, or the local Recent measurements at the A È lvsjoÈ test site (U. Olofsson,
material response in a rail, for arbitrary track and rail Department of Machine Elements, Royal Institute of
conditions. Case and parameter studies can easily be Technology, Stockholm, Sweden, 1999, personal commu-
performed by varying the material data for the components nication) clearly indicate that the friction coefficient is
in the track, the rail profile and its material data, and the higher, about ì ˆ 0:5, even in wet conditions. For this
loading conditions from a running train vehicle. For higher value of ì, crack initiation will be predicted at, or
instance, the track model can be used to calculate sectional very close to, the railhead surface.
forces and displacements at any rail cross-section. The rail The material response in the rail model after 16 wheel
model can be used to calculate residual stresses, plastic passages showed ratchetting material response. However,
strain fields, critical planes for crack initiation and shake- because of the low ratchetting rate, crack initiation will be
down limits. In addition, strain rates in the material can be governed by LCF, according to the comparison between the
estimated. results from the LCF and the ratchetting crack initiation
The distance from the railhead surface to the point of criteria. If a larger number of wheel passages is simulated,
maximum shear stress for elastic Hertzian contact without plastic shakedown may be expected. However, the combi-
traction forces can, as a rule of thumb, be estimated to be nation of the magnitudes of the axle load and the friction
half the semi-axis width in the rolling (longitudinal) coefficients in the rolling (ì x ) and the transverse ( ì y )
direction. Assuming no traction forces and elastic condi- directions directly affects the material response in the rail
tions, this distance should be 0:5a9 ˆ 3:1 3 10ÿ3 m for the model. The values of the friction coefficients in the present
present axle load. However, traction forces in the wheel± investigation were low, especially in the rolling direction.
rail contact move this point of maximum closer to the Larger values would probably have led to failure by
railhead surface. The results from the FE simulation using ratchetting in the comparison between the two criteria.
the rail model show that the distance from the railhead Also, the calculated directions for the crack plane are
surface to the point of maximum shear stress is equal to the similar to the directions of head checks at the rail gauge
distance from the railhead surface to the point of maximum corner observed at the test site in A È lvsjoÈ, Sweden. This
von Mises effective stress, i.e. at d ˆ 2:8 3 10ÿ3 m in Fig. indicates that the direction of the crack plane follows the
7. Thus, the point of maximum shear stress has been moved deformation of the material structure from the traction
closer to the railhead surface owing to the traction forces forces [29] that can be seen in twin-disc tests [23].
modelled in the rail model. In addition, the result for the The significance of taking the global bending from the
Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 17

track model into consideration in the rail model, on the temperature of the rail was estimated to be 214 8C, based
results from the rail model, has been studied. This was on a train velocity of 75 km=h, a maximum Hertzian
done by studying the sensitivity in results from changes in contact pressure of 1119 3 106 N=m2 , a friction coefficient
the boundary conditions in the rail model. Two cases using of 0.41 and creep of 2 per cent. The result from this
the rail model were compared: estimation is in agreement with predictions according to
Knothe et al. [35], which showed a temperature rise of
1. All degrees of freedom were restrained at one cross-
196 8C for a similar load case. Therefore, under these
section, and prescribed displacements were calculated
assumptions, the theory of thermally induced phase
from the track model at the opposite cross-section.
transformation, where a temperature of at least 600 8C is
2. All degrees of freedom were restrained at both cross-
needed, cannot be supported. However, in spite of the low
sections, i.e. global bending from the track response was
rise in temperature estimated beneath the wheel±rail
neglected.
contact, thermally induced stresses may still affect the
The results of the two FE simulations differed by less than residual stress field [35].
10 per cent with respect to the magnitude of the residual The Chaboche model is able to describe the correct
von Mises effective stress after 16 wheel passages. This ratchetting rate, but not the shape of the stress±strain
shows that the global bending from the track in the FE cycles accurately. This has led to the development of better
simulations with the rail model has little effect on the local material models. In order to describe ratchetting with
material response. In addition, both FE simulations showed decreasing ratchetting rate, which occurs for rail steel,
ratchetting material response, case 2 having a somewhat several attempts have been made by, for example, Bower
higher ratchetting rate owing to its stiffer constraints. This [19], Bower and Johnson [24, 36], McDowell [37] and
increase in ratchetting rate resulted in crack initiation by Jiang and Sehitoglu [38±40]. The model proposed by Jiang
ratchetting instead of LCF in the comparison between the and Sehitoglu, subsequently referred to as the `Jiang
crack initiation criteria. Consequently, fulfilment of the model', has been further investigated by Ekh et al. [41].
criteria for crack initiation by LCF and ratchetting was The Jiang model is more complex than the Chaboche
concluded to be very close for the present load case. model and includes a larger number of parameters in the
The influence from strain rate effects resulting in an optimization procedure. In Fig. 9, the result from an
increase in temperature within the material, i.e. considera- optimization of the material parameters in the Jiang model,
tion of the coupling between the thermal and the mechani- to experiments performed by Bower [19], is presented. The
cal fields, was not considered in the present FE simulation. line with circles denotes experimental data, and the solid
To estimate the relevance of this restriction in the FE line represents the numerical computation.
simulation, the strain rate in the material was estimated The Jiang model has been implemented numerically into
from the FE simulation using the rail model for the passage ABAQUS version 5.7 [9]. A more thorough description of
of one wheel. The results show that the maximum strain the Jiang model and details concerning the numerical FE
rate, here calculated on the basis of a velocity of 75 km=h implementation are given in reference [41]. To increase the
for the commuter trains at the test site, was approximately rate of convergence and decrease the time requirements for
12 sÿ1 . Strain rate effects may influence the results and will the analyses, algorithmic tangent stiffness can be used, i.e.
be investigated more thoroughly in future work. In addition, a consistent tangent modulus. One limitation of the Jiang
the rail was considered stress free at the first wheel passage model is the large number of parameters included in the
in the present work. The results from the performed FE optimization, and therefore it is difficult to find the optimal
simulation show that the first wheel passage over the rail set of parameters. Also, the structure of the model (the
has a considerable effect on the stress state after unloading. backstress is divided into several parts, and an evolution
Therefore, redistribution and relaxation of residual stresses law for the exponent is used in the evolution law for the
in the rail from, for example, the manufacturing process backstress) is not easy to motivate physically. Several of
will also be considered in future work. them are determined by the shape of the elastic±plastic
An estimation of the contact temperature rise due to stress±strain curve obtained from experiments. In future
frictional heating from the wheel±rail contact was per- work, the possibility of applying the Jiang model in FE
formed using the first-order method of contact temperature simulations using the rail model will be investigated.
estimation as described in reference [34]. The surface
temperature of the wheel and the rail are assumed to
depend on their speed relative to the source of heat, which 6 CONCLUSIONS
in the case of frictional sliding means relative to the contact
patch. It is assumed that both surfaces in the wheel±rail The observations and the results in the present investigation
contact have the same maximum surface temperature, since are summarized below:
the bodies are in close contact. The physical properties
needed for the present rail steel in the contact temperature 1. A tool consisting of two FE models has been developed
estimation were taken for a similar rail steel in reference for simulation of residual stress and plastic strain fields
[34]. The rise in surface temperature relative to the bulk in railheads.
F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F
18 J W RINGSBERG, H BJARNEHED, A JOHANSSON AND B L JOSEFSON

Fig. 9 Results from optimization of the Jiang model. The measure points used in the optimization are from cycles 1,
2, 15, 50, 100, 200, 400 and 600 [19]

2. A realistic load situation was simulated using a load REFERENCES


case in a case study of commuter train traffic at a test
site in A È lvsjoÈ, Sweden. The FE simulation showed
1 Johnson, K. L. Contact Mechanics, 1985 (Cambridge
ratchetting material response in the railhead after 16 University Press, Cambridge).
wheel passages. 2 Suresh, S. Fatigue of Materials, 2nd edition, 1998 (Cam-
3. A criterion for crack initiation by ratchetting and a bridge University Press, Cambridge).
criterion for crack initiation by LCF were compared. 3 Beynon, J. H., Brown, M. W. and Kapoor, A. Initiation,
The LCF criterion showed the lowest number of cycles growth and branching of cracks in railway track. In Proceed-
to crack initiation, which in this case indicates that LCF ings of Engineering Against Fatigue (Eds J. H. Beynon, M. W.
rather than ratchetting governs crack initiation. Brown, R. A. Smith, T. C. Lindley and B. Tomkins), 1999, pp.
461±472 (Balkema, Rotterdam, The Netherlands).
4. The FE models developed have the potential to
4 Jiang, Y. and Sehitoglu, H. An analytical approach to
determine angles for critical planes for crack initiation
elastic±plastic stress analysis of rolling contact. Trans. ASME,
in the railhead. The calculated angles for the direction J. Tribology, 1994, 116, 577±587.
of the crack plane in the simulated load case agreed well 5 Jiang, Y. and Sehitoglu, H. A model for rolling contact
È lvsjoÈ, Sweden.
with observations at the test site in A failure. Wear, 1999, 224, 38±49.
5. The influence of global bending from the track model 6 Hahn, G. T., Bhargava, V., Rubin, C. A., Chen, Q. and
on the results from the rail model was noticeable, about Kim, K. Analysis of the rolling contact residual stresses and
10 per cent of the calculated residual von Mises cyclic plastic deformation of SAE 52100 steel ball bearings.
effective stress, but not of major concern. The influence Trans. ASME, J. Tribology, 1987, 109, 618±626.
of global bending may not be of concern with respect to 7 Fry, G. T., Lawrence, F. V. and Robinson, A. R. A model for
crack initiation but may be significant with respect to fatigue defect nucleation in thermite rail welds. Fatigue and
Fract. Engng Mater. and Struct., 1996, 19, 655±668.
long crack propagation and rail fracture.
8 I-DEAS Master Series Version 5.0, 1997 (SDRC Inc., Milford,
6. Because of the low ratchetting rate in the FE simulation
Ohio).
using the rail model, it is expected that, for a larger 9 ABAQUS/Theory Manual Version 5.7, 1997 (Hibbitt, Karlsson
number of simulated wheel passages, plastic shakedown and Sorensen, Providence, Rhode Island).
material response will occur. This is also supported by 10 Bathe, K.-J. Finite Element Procedures, 1996 (Prentice-Hall,
the results from the comparison between the ratchetting Englewoods Cliffs, New Jersey).
criterion and the LCF criterion. 11 Nielsen, J. C. O. DIFF version 94.1Ða computer program for
numerical analysis of vertical dynamic train/track interaction.
Theory manual and user's manual. Publication 34, Department
of Solid Mechanics, Chalmers University of Technology,
Gothenburg, Sweden, 1994.
12 DEsolver AB, GENSYSÐOn-line Documentation, 1998
ACKNOWLEDGEMENTS (OÈ stersund, Sweden).
13 Nielsen, J. C. O. and Abrahamsson, T. J. S. Coupling of
This work has been funded by the European Union Brite- physical and modal components for analysis of moving non-
EuRam III project ICON, ref. BE96-3091 (Integrated study linear dynamic systems on general beam structures. Int. J.
of rolling CONtact fatigue), members of the ICON Numer. Meth. Engng, 1992, 33, 1843±1859.
Common Interest Group and also by UIC (The Interna- 14 Chaboche, J.-L. and Lemaitre, J. Mechanics of Solid
tional Union of Railways). Materials, 1990 (Cambridge University Press, Cambridge).

Proc Instn Mech Engrs Vol 214 Part F F02299 # IMechE 2000
ROLLING CONTACT FATIGUE OF RAILSÐFE MODELLING 19

15 Armstrong, P. J. and Frederick, C. O. A mathematical 29 Ringsberg, J. W., Loo-Morrey, M., Josefson, B. L., Kapoor,
representation of the multiaxial Bauschinger effect. CEGB A. and Beynon, J. H. Prediction of fatigue crack initiation for
Report RD/B/N731, 1966. rolling contact fatigue. Int. J. Fatigue, 2000, 22, 205±215.
16 Jiang, Y. and Kurath, P. Characteristics of the Armstrong± 30 Bannantine, J. A., Comer, J. J. and Handrock, J. L.
Frederick type plasticity models. Int. J. Plasticity, 1996, 12, Fundamentals of Metal Fatigue Analysis, 1990 (Prentice-Hall,
387±415. Englewood Cliffs, New Jersey).
17 Johansson, A. and Thorberntsson, H. Elastoplastic material 31 Beste, A. Elastisch plastisches Spannung-, Dehnungs- und
model with nonlinear kinematic hardening for rolling and Anriûverhalten in statisch und zyklisch belasteten Kerbschei-
sliding contact fatigue. Technical Report EX 1997:15, Depart- benÐEin Vergleich zwischen experimentellen Ergebnissen
ment of Solid Mechanics, Chalmers University of Technology, und NaÈherungsrechnungen (in German). No. 34, VeroÈffentli-
Gothenburg, Sweden, 1997. chung des Instituts fuÈr Stahlbau und Werkstoffmechanik der
18 Matlab Version 5.0, 1996 (The Math Works Inc., Natick, Technischen UniversitaÈt Darmstadt, Darmstadt, Deutschland,
Massachusetts). 1981.
19 Bower, A. F. Cyclic hardening properties of hard-drawn 32 Socie, D. Discriminating experiments for multiaxial fatigue
copper and rail steel. J. Mech. Phys. Solids, 1989, 37, damage models. In Fatigue Design, ESIS16 (Eds J. Solin, G.
455±470. Marquis, A. Siljander and S. SipilaÈ), 1993, pp. 201±211
20 Nelder, J. A. and Mead, R. A simplex method for function (Mechanical Engineering Publications, London).
minimization. Comput. J., 1965, 7, 308±313. 33 Ponter, A. R. S., Hearle, A. D. and Johnson, K. L.
21 Mahnken, R. and Stein, E. Parameter identification for Application of the kinematical shakedown theorem to rolling
viscoplastic models based on analytical derivatives of a least- and sliding point contacts. J. Mech. Phys. Solids, 1985, 33,
square functional and stability investigation. Int. J. Plasticity, 339±362.
1996, 12, 451±479. 34 Williams, J. A. Engineering Tribology, 1994 (Oxford Uni-
22 Diehl, U. and Nilsson, L. Svenska lok och motorvagnar (in versity Press, New York).
Swedish). Svenska JaÈrnvaÈgsklubben, Stockholm, Sweden, 35 Knothe, K. and Liebelt, S. Determination of temperatures for
1989. sliding contact with applications for wheel±rail systems. Wear,
23 Tyfour, W. R., Beynon, J. H. and Kapoor, A. The steady 1995, 189, 91±99.
state wear behaviour of pearlitic rail steel under dry rolling± 36 Bower, A. F. and Johnson, K. L. The influence of strain
sliding contact conditions. Wear, 1995, 180, 79±89. hardening on cumulative plastic deformation in rolling and
24 Bower, A. F. and Johnson, K. L. Plastic flow and shakedown sliding line contact. J. Mech. Phys. Solids, 1989, 37, 471±493.
of the rail surface in repeated wheel±rail contact. Wear, 1991, 37 McDowell, D. L. A two surface model for transient non-
144, 1±18. proportional cyclic plasticity, Part I. Trans. ASME, J. Appl.
25 Cox, S. J. Reference track proposalÐversion 2: 72105/3/ Mechanics, 1985, 52, 298±302.
PAND/T/Reference.Track.Proposal.2. Silent Track Consor- 38 Jiang, Y. and Sehitoglu, H. Rolling contact stress analysis
tium, Surrey, 1997. with the application of a new plasticity model. Wear, 1996,
26 Nielsen, J. C. O. Train/track interaction: Coupling of moving 191, 35±44.
and stationary dynamic systemsÐtheoretical and experimental 39 Jiang, Y. and Sehitoglu, H. Modeling of cyclic ratchetting
analysis of railway track structures considering wheel and plasticity, Part I: development of constitutive relations.
track imperfections. PhD dissertation, Department of Solid J. Appl. Mechanics, 1996, 63, 720±725.
Mechanics, Chalmers University of Technology, Gothenburg, 40 Jiang, Y. and Sehitoglu, H. Modeling of cyclic ratchetting
Sweden, 1993. plasticity, Part II: comparison of model simulations with
27 Timoshenko, S. P. and Goodier, J. N. Theory of Elasticity, experiments. J. Appl. Mechanics, 1996, 63, 726±733.
3rd edition, 1970 (McGraw-Hill, New York). 41 Ekh, M., Johansson, A., Thorberntsson, H. and Josefson, B.
28 Kapoor, A. A re-evaluation of the life to rupture of ductile L. Models for cyclic ratchetting plasticityÐintegration and
metals by cyclic plastic strain. Fatigue and Fract. Engng calibration. Trans. ASME, J. Engng Mater. Technol., 2000,
Mater. Struct., 1994, 17, 201±219. 122, 49±55.

F02299 # IMechE 2000 Proc Instn Mech Engrs Vol 214 Part F

View publication stats

You might also like