You are on page 1of 13

Wear 428–429 (2019) 302–314

Contents lists available at ScienceDirect

Wear
journal homepage: www.elsevier.com/locate/wear

Prediction of plastic deformation and wear in railway crossings – Comparing T


the performance of two rail steel grades
Rostyslav Skrypnyka,∗, Magnus Ekhb, Jens C.O. Nielsena, Björn A. Pålssona
a
Department of Mechanics and Maritime Sciences/CHARMEC, Chalmers University of Technology, SE-412 96, Gothenburg, Sweden
b
Department of Industrial and Materials Science/CHARMEC, Chalmers University of Technology, SE-412 96, Gosthenburg, Sweden

A R T I C LE I N FO A B S T R A C T

Keywords: Railway crossings are subjected to a severe load environment leading to damage and degradation of rail profiles.
Dynamic vehicle–track interaction The damage is the result of the high magnitudes of contact pressure and slip generated in the wheel–rail contacts
Switches & crossings during each wheel transition between wing rail and crossing nose. In this paper, numerical predictions of the
S&C long-term accumulation of plastic deformation and wear are presented and compared for two rail grades used in
FEM
crossings. The comparison is performed using a framework that includes simulations of dynamic vehicle–track
Plastic deformation
Wear
interaction, wheel–rail contact and rail damage. A load sequence generated by means of Latin hypercube
sampling, taking into account variations in worn wheel profile, vehicle speed and wheel–rail friction coefficient,
is considered. A Hertzian-based metamodel for wheel–rail normal contact accounting for inelastic material re-
sponse is used for the simulation of wheel–rail contact. The methodology is demonstrated by calculating the
plastic deformation for one cross-section of the crossing nose using an Ohno-Wang cyclic plasticity model, while
wear is predicted by means of the Archard model for sliding wear. For the studied conditions and after the
simulation of 41400 load cycles, representing an accumulated traffic load of 0.8 MGT, it is concluded that the
fine-pearlitic rail grade R350HT experiences half of the ratchetting strain compared to the austenitic hot-rolled
manganese steel Mn13. Based on a wear model calibrated for Mn13, the maximum wear of the studied cross-
section is about 2% of the maximum plastic deformation.

1. Introduction was observed that the plastic flow stabilised at a constant rate. An at-
tempt to quantify plastic deformation in a numerical simulation fol-
In Sweden, the estimated cost for the maintenance of railway lowed almost a decade later, see Bower and Johnson [3]. The predicted
turnouts (switches & crossings, S&C) in 2014 was 400–450 million SEK, cyclic strain accumulation over 100 load cycles showed reasonable
corresponding to about 12% of the overall maintenance costs for the agreement with experimental data from tension-torsion tests. A non-
railway network. One important reason for the high maintenance costs linear purely kinematic hardening law for the material subjected to a
is the need to repair and replace crossings and switch rails due to plastic Hertzian stress field was utilised.
deformation and wear. An efficient measure to reduce the maintenance Nowadays, non-linear finite element (FE) analysis is the standard
costs for a turnout design under given traffic conditions is to select the tool for the simulation of plasticity in structures in general, which en-
most appropriate rail steel grade. To this end, a comparison of materials ables applications of sophisticated material models capable of accu-
in terms of rail damage is needed. The aim of this paper is to perform rately capturing cyclic plasticity (see e.g. Ref. [4]). An example where
such a comparison in terms of plastic deformation and wear, which are the FE method was applied to predict cyclic plasticity in turnouts can be
the main damage mechanisms in turnouts together with rolling contact found in Ref. [5]. The influence of material selection on the deforma-
fatigue. tion of the crossing nose due to repeated wheel passages was in-
A literature survey on the modelling of surface plasticity and wear vestigated, although only one wheel profile was used to roll over the
in railway applications was performed by Enblom [1]. Laboratory same region. The study suggested that the contact force can be reduced
testing of plastic flow in rails can be traced to the early 1980s when a for the manganese crossing compared to a maraging steel crossing,
two-roller test rig was used to simulate wheel–rail contact on tangent which was explained by the higher adaptability of the manganese steel.
and curved tracks [2]. For the stock rail made of standard material, it The second damage mechanism considered in this paper is wear. As


Corresponding author.
E-mail address: rostyslav.skrypnyk@chalmers.se (R. Skrypnyk).

https://doi.org/10.1016/j.wear.2019.03.019
Received 6 July 2018; Received in revised form 21 December 2018; Accepted 19 March 2019
Available online 30 March 2019
0043-1648/ © 2019 Elsevier B.V. All rights reserved.
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

concluded in Ref. [6], wear may lead to amplification of impact loads in (facing move or trailing move). The recent study [18] showed how the
a turnout. A comparison of responses of some materials used in turnouts choice of the rail material affects crossing degradation. The growing
when subjected to high impact forces can be found in Ref. [7]. Railway needs for faster and heavier transportation intensifies the contact con-
turnouts have also been identified to be the main driver of wheel wear ditions between wheel and rail. Furthermore, if the wheel profiles are
[8]. The predicted wheel wear, generated in a turnout without track sternly worn and/or the rail profiles are damaged, the degradation rate
irregularities, was found to be one order of magnitude higher than the of the rail material increases. This is why, in this study, the variability
wear on tangent track and curves with irregularities. Two main types of in traffic conditions is represented by three random variables: vehicle
wear models were identified in Ref. [1]: (a) models that relate wear to speed, wheel–rail friction coefficient and wheel profile wear.
the dissipated friction energy in the contact area, and (b) Archard-based It is concluded in Ref. [1] that most studies of wheel/rail damage
[9] models that attribute wear to sliding distance, normal force and focus on one of several damage modes and that models where wear and
hardness of the softer material. The latter type has attracted a broad plastic deformation are integrated are needed. Such a multidisciplinary
field of applications (e.g. used for the prediction of wear in roller methodology was proposed in Ref. [20] allowing to simulate long-term
bearings). The wear rate in these types of models depends on the wear damage in S&C and accounting for variability in traffic conditions. In
regime that can be specified using a wear map that relates the wear each iteration step, the methodology consists of four parts: (I) Simula-
coefficient to contact pressure and sliding velocity, see e.g. Lim and tion of dynamic vehicle–track interaction; (II) Analysis of wheel–rail
Ashby [10]. However, the generation of a wear map requires extensive normal contact; (III) Prediction of accumulated damage; (IV) Updating
testing and is usually only valid for a given material pair. Due to dif- of rail profiles, which are then used as input in the next iteration step.
ferent wear regimes, wear maps are often discontinuous and a small The methodology starts by feeding rail profiles to vehicle dynamics
variation in the contact conditions may lead to one order of magnitude simulations that are carried out for a number of measured wheel pro-
variation in the wear coefficient. Recently, Cremona et al. showed how files and assuming statistical distributions for the friction coefficient
a metamodelling technique can be used to construct a continuous wear and vehicle speed. Then, based on the calculated contact forces and
map [11]. contact positions at selected cross-sections, the three-dimensional (3D)
A comparison of the two types of wear models can be found in Ref. wheel–rail normal contact problem is solved to estimate the contact
[12]. The Archard-based models used by Jendel [13], and Enblom and patch size and the maximum von Mises stress, assuming the wheel to be
Berg [14] were compared with the models of Zobory [15], and Pearce elastic and the rail elasto-plastic. Thereafter, the simulations of plastic
and Sherratt [16] that are based on the energy dissipation approach. deformation are carried out in 2D, where the transition from 3D to 2D is
The models were selected because of their capability of predicting wear accomplished by adjusting the normal load to preserve the maximum
in the mild and severe wear regimes. Even though all the models are von Mises stress. Both the normal and tangential loadings are applied to
dependent on experimental coefficients that may not have been esti- the width of the contact patch, assuming the same spatial distribution
mated in similar test conditions, a reasonable agreement was found for the normal and tangential pressures and where the magnitude of the
between the models of Zobory and Jendel/Enblom. The application of tangential pressure is scaled by the utilised friction coefficient. In the
the models to a case of an urban transport vehicle travelling on a net- simulation, the accumulated load history is taken as a multiple of the
work with a distribution of curve radii showed that the model of Pearce load sequence stemming from the vehicle dynamics simulations. In
and Sherratt predicted wear rates one order of magnitude higher in the each iteration step the output from the simulations of dynamics is as-
severe wear regime than the other models. It was concluded that sumed not to be influenced by the continuous change of rail profiles to
models based on different approaches can yield good agreement in reduce computational effort. The wear calculation is also carried out
some cases, but that significant discrepancies may appear because based on the results of the vehicle dynamics simulations. Finally, the
transitions between wear regimes are not introduced consistently for all rail profile is updated for the next iteration of the methodology.
models. The methodology has been validated by comparing predicted rail
A turnout is generally subjected to the influence of various types of profiles with those measured in the field at Härad in Sweden [20] and
vehicles, passing at different speed and in different weather conditions. Haste in Germany [21]. To reduce the computational effort, the
The parametric study presented by Xin et al. [17] suggests that vehicle methodology has recently been updated with a Hertzian-based meta-
speed and wheel–rail friction coefficient have a significant effect on the model of wheel–rail normal contact [22]. Another procedure to predict
impact position on the rail and on the magnitude of the impact forces the life of a turnout based on an explicit FE model has been presented in
and stresses. Vehicles induce impact loads either on the crossing nose or Ref. [23]. It also predicts the elasto-plastic material response assuming
on the wing rail (see Fig. 1) depending on the direction of the vehicle non-linear kinematic hardening, however, for one wheelset only

Fig. 1. Components of a turnout (from Ref. [19]).

303
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

without considering the variability in traffic conditions. Furthermore,


unlike in Ref. [20], neither wear nor cyclic loading were accounted for,
which may result in an overestimation of rail life.
The following sections describe how the methodology described
above is set up to simulate the performance of two different crossing
materials: Section 2 presents the geometry and settings for the vehicle
and track models. Section 3 describes how the spread in traffic condi-
tions is accounted for. Section 4 shows the difference in the mechanical
response of the two compared materials when subjected to a uniaxial
cyclic load and how the response can be modelled. The simulation of
wheel–rail normal contact is explained in Section 5. Sections 6 and 7
present results from the calculations of plasticity and wear. Finally,
Section 8 completes the study by summarising the findings.

2. Simulation of dynamic vehicle–track interaction

The traffic situation considered in this study is a passenger vehicle


running in the through route of the crossing panel of a 60E1-1:15-R760 Fig. 3. Illustration of crossing geometry comprised of 90 rail cross-sections.
turnout in the facing move (from switch panel to crossing panel), see
Fig. 1. The vehicle–track interaction model was developed in the Swedish Transport Administration (Trafikverket), while the rail profiles
commercial multibody simulation code SIMPACK 2017 [24]. were optimised to minimise the wheel–rail normal contact stress ac-
The considered vehicle is the Manchester benchmarks passenger cording to Hertz [28]. In total, 90 rail cross-sections are used to re-
vehicle model [25]. The standard vehicle configuration is utilised, but present the crossing geometry (see Fig. 3). The cross-section spacing
the axle load was increased from 15.1 to 19.1 tonnes by increasing the varies depending on the position in the crossing panel, from 100 mm at
car body mass to better represent the load environment generated by the ends to 10 mm in the transition zone, where the wheels make the
modern passenger vehicles. For each simulation, a unique combination transition from wing rail to crossing nose or vice versa. In SIMPACK, the
of wheel profile, vehicle speed and wheel–rail friction coefficient are full crossing geometry is created via an interpolation of adjacent cross-
used as input based on a sampled load collective that will be described sections. It is therefore a requirement that adjacent cross-sections have
in Section 3. the same number of data points (unless there is a defined discontinuity
The track model is of the co-running type with one planar system of in the interpolation) to allow for a well-defined interpolation and
rigid masses, springs and dampers following each wheelset to represent geometry description. SIMPACK's interpolation procedure was con-
the track flexibility (see Fig. 2). sidered in the generation of the cross-sections. Care was taken to make
This study uses the “TM2” track model configuration described in the interpolation as physical as possible and to avoid surface warping
Ref. [26]. This model was tuned to measured track receptance data up that could result if the interpolation lines are not aligned with the rail
to 200 Hz. Each track model consists of five bodies: two rails, one surface geometry between cross-sections. SIMPACK's Hertzian model is
sleeper and two ballast masses. The track model has nine degrees of used to model the wheel–rail normal contact and FASTSIM [29] to
freedom in total which are distributed as follows: each rail has one model the tangential contact. The simulations are performed without
vertical and one lateral degree of freedom; the sleeper body has one track geometry irregularities. The simulation outputs are sampled at a
vertical, one lateral and one in-plane rotational degree of freedom; each frequency of 8 kHz.
ballast mass has one vertical degree of freedom.
The crossing rail geometry representation is taken from Ref. [27].
The overall layout of the crossing is from a standard design used by the 3. Load sequence

Vehicle speed, wheel–rail friction coefficient and the geometry of


the wheel profile are important input parameters in the simulations of
damage and degradation of rail profiles as they affect the location and
magnitude of the wheel–rail contact forces. A load sequence that ac-
counts for a variation in these parameters is therefore generated to
resemble the varying traffic conditions that a crossing would experience
in field. Here, this is achieved using a sample of 276 measured wheel
profiles from Bombardier Regina passenger trains [30] to represent the
variation in wheel profile geometry, while the vehicle speed and
wheel–rail friction coefficients are assumed to be normally distributed
with means 160 km/h and 0.35, and standard deviations 5 km/h and
0.07, respectively, see Fig. 4. 276 unique random combinations of these
parameters for simulations of vehicle–track interaction were con-
structed using the method implemented by Pålsson [31]. It utilises the
Latin Hypercube Sampling (LHS) technique [32], which has been
proven (see e.g. Ref. [33]) to be more efficient than a standard Monte
Carlo sampling.
Fig. 5 depicts a top view of the transition zone in the crossing, where
the wheels travel from the wing rail over to the crossing nose (or vice
versa). The longitudinal distance is measured from the theoretical
crossing point (TCP). A schematic lateral view of the wheel trajectory is
shown in Fig. 6. A wheel travelling over a crossing generally induces an
Fig. 2. Track model (arrows denote degrees of freedom). impact load on the crossing nose (for traffic in the facing move) or on

304
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

Fig. 4. Input data for simulations in SIMPACK: (a) three examples of measured wheel profiles, and assumed distributions of (b) friction coefficient μ and (c) vehicle
speed v in terms of number of occurrences (276 combinations in total).

Fig. 6. Lateral view of the wheel trajectory over crossing.

the wing rail in the running direction. On the crossing nose, the lateral
contact position is relatively constant but the crossing nose has a ver-
tical inclination.
One output from each of the 276 simulations in SIMPACK is the time
history of the wheel–rail normal contact force, corresponding to the
wheel passage along the length of the crossing nose. Examples of force
time histories are shown in Fig. 7, indicating high-magnitude impact
loads occurring on the crossing nose. The load maxima are marked with
dots. Results from the 276 load cases are summarised in Fig. 8, where
Fig. 5. Top view of crossing geometry, adopted from Ref. [27].
the frequency (number of occurrences) and the locations of the tran-
sition point are plotted in Fig. 8(a), whereas the magnitudes of the
the wing rail (for traffic in the trailing move), see also Fig. 1. The im- normal and tangential load maxima using discretisation of the rail
pact load is caused by the downwards-upwards motion experienced by surface are plotted in Fig. 8(b) and (c). For a nominal rail profile, it is
the wheel as it rolls along the wing rail and over to the crossing nose or observed that most of the wheel transitions occur within a 5 mm wide
vice versa. On the wing rail, the vertical motion is caused by the band on the crossing nose. After the transition, the load from each
conicity of the wheel tread and the significant change in lateral wheel passage builds up to a maximum in the interval 170–280 kN.
wheel–rail contact position that occurs due to the lateral deviation of However, the locations of the load maxima are more distributed over

305
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

the surface of the crossing nose with maximum levels occurring at


about 800 mm from the TCP. For each situation with maximum normal
load according to Fig. 8(b) and (d) presents the ratio between the uti-
lised friction coefficient μeff (that is used in the plasticity calculations,
see Section 6) and the friction coefficient assigned for each load case by
the LHS. It is observed that full slip (friction ratio equals to one) is never
reached. On average, the friction ratio is 0.47.
From each simulation of dynamic vehicle–track interaction, the
time histories of the wheel–rail normal contact force and contact po-
sition are used as input to part II of the methodology [20].

4. Material modelling

In this section, the adopted cyclic plasticity model is presented and


calibrated for two rail steel grades used in railway crossings. The two
materials are the fine-pearlitic rail grade R350HT and the austenitic
hot-rolled manganese steel Mn13. The calibration of the plasticity
Fig. 7. Three samples of calculated time history of normal contact force N along model was done by using test data from the uniaxial cyclic stress-con-
the length x of the crossing rail. Dots indicate maximum values. trolled ratchetting experiments reported in Ref. [34]. The modelling
was performed by using the cyclic plasticity model proposed by Ohno

Fig. 8. (a) Location and number of occurrences for the first contact point on the crossing nose, (b) maximum normal contact force N in each cell, cf. Fig. 5 (c)
maximum tangential force Ft , and (d) friction ratio μeff / μ .

306
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

Fig. 9. Experiments and simulations with cyclic plasticity model for 525 load cycles for rail grades (a) R350HT and (b) Mn13.

Table 1
Identified material parameter values (Poisson's ratio ν is not calibrated) for the Ohno-Wang model for rail grades R350HT and hot-rolled Mn13 for 525 load cycles.
Grade E ν σytens C1 γ1 m1 C2 γ2 m2 C3 γ3 m3 σycompr

[−] [GPa] [−] [MPa] [MPa] [−] [−] [MPa] [−] [−] [MPa] [−] [−] [MPa]

R350HT 183 0.3 433 432000 2400 2.12 45700 229 3.80 11300 18.4 3 379
Mn13 200 0.3 351 31760 708 2.40 15990 718 2.56 2310 0 2 425

and Wang [35] (summarised in Appendix A). By using three back 5. Elasto-plastic wheel–rail normal contact
stresses, it was found that a reasonably good agreement with the ex-
perimental data is obtained. For both materials, 525 load cycles were In part II of the simulation methodology, the normal contact pro-
considered in the calibration and the results are shown in Fig. 9. The blem is solved with local 3D FE models of the wheel–rail contact using
calibration procedure was performed by minimising the least-square the FE software Abaqus, cf [20]. The goal here is to perform compu-
error using the Nelder-Mead simplex method [36]. The obtained ma- tationally efficient and realistic predictions of contact patch sizes ac-
terial parameter values are stated in Table 1. counting for plasticity in the rail material. Since the influence of the
By examining the results in Fig. 9, clear differences between the tangential load on the size of the contact patch is small (see e.g. Ref.
studied materials can be observed. For example, at 1% strain in cycle 1, [37]), the tangential load is disregarded for simplicity. However, in part
the stress value is almost twice as high for R350HT as compared to III of the methodology the tangential load is accounted for (see Section
rolled Mn13. The rolled Mn13 shows a very soft response (most pro- 6).
nounced during the first load cycles) above the yield limit, which gives The local 3D models are based on the assumption that the curvature
significant plastic strains. This implies that the material can reshape radii of the rail and wheel are constant near the contact point.
significantly during the first loadings, which is also observed for Mn13 Furthermore, the wheel material is assumed to be linear isotropic
rails in the field. After this significant plastic strain, the model shows a elastic with Young's modulus E = 183 GPa and Poisson's ratio ν = 0.3,
plastic shakedown behaviour, i.e. the ratchetting strain rate decreases whereas the rail material behaviour is assumed to be elasto-plastic with
and the ratchetting stops. The experiment shows a slow transition of the the Ohno-Wang model [35] as described in Section 4.
shape of the stress-strain cycles towards an elastic shakedown. This is To further speed up the calculations, a metamodel replacing the 3D
not captured well by the model. Nevertheless, the ratchetting strain that local FE models was proposed in Ref. [22]. It was shown that the error
the experiment saturates towards is very well captured by the model of the metamodel compared to the FE model in estimation of contact
which is believed to be of major importance for simulations of changes semi-axes is up to 6%. The metamodel is based on the Hertz's solution
of cross-section shapes. The R350HT experiment shows that the for contact problems [38]. The metamodel adopts Hertz's results that
ratchetting strain is still increasing after 525 cycles, but the ratchetting the contact patch is elliptical and that the ratio α between the semi-axes
rate (per cycle) has decreased from the first loading cycles. The cyclic a and b is determined by the radii of the contacting bodies, see details in
plasticity model can capture this decrease very well and also the shape Ref. [22]. The Hertz's contact model has been shown to be capable of
of the stress-strain cycles. This shape remains similar during the load predicting accurate long-term damage when the variation in traffic
cycles for R350HT in contrast to the shape change for the rolled Mn13. conditions is accounted for, see e.g. Ref. [39]. The assumption of el-
liptical contact patch is strengthened by the recent publication [40],
where the contact patches from FE modelling using realistic wheel and
rail profiles resemble elliptic shapes also for elasto-plastic materials. In

307
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

the metamodel, the expressions for the semi-axes a, b and the maximum {β0, β1, β2, β3} were determined from).
contact pressure p0 are inspired by the Hertz's solution:
β1 6. Plasticity in cross-sections
⎡ α 2 E2 (α ) N ⎤
a = β0 ⎢ 1 1 1 1

⎢ wheel + wheel + rail + rail
⎥ In part III of the simulation methodology, the plastic deformation
⎣ Rx Ry Rx Ry
⎦ (1) and wear of the two crossings with different materials are predicted for
a one contact point of the leading wheel. The R350HT and rolled Mn13
b=
α (2) crossings are both subjected to the same traffic situation represented by
the load sequence described in Section 3. The corresponding normal
N β3 contact conditions in terms of contact patch size a, b and maximum
p0 = β2 ⎡ ⎤
⎣ πa b ⎦ (3) von Mises stress σvM,n
max are predicted by the Hertzian-based meta-

where E2 is the complete elliptic integral of the second kind, N is the model. A realistic long-term traffic situation for a crossing can only be
normal force and {R xwheel , R ywheel , R xrail , R yrail} are the local radii of the obtained if the load sequence is repeated sufficiently many times. To
wheel and rail models. The difference as compared to the Hertz's so- obtain a reasonable computational time for the plasticity predictions,
lution is that the coefficients {β0, β1, β2, β3} in the metamodel are these are performed using 2D plane strain models of sampled cross-
identified from 3D local FE simulations accounting for the elasto-plas- sections along the crossing using an in-house FE software. The cross-
ticity in the rail. This identification is based on a least-square fit via an sections are loaded on a segment with length 2b (where b is the contact
optimization algorithm against sampled results from a number of 3D semi-axis in the direction orthogonal to the rolling of the wheel) with a
local FE simulations, each considering one load cycle with the material pressure distribution according to the contact theory of Hertz for a line
response calibrated for 525 load cycles. In this sampling, the geometry contact:
parameters {R xwheel , R ywheel , R xrail , R yrail} and the normal force N are varied
according to the procedure described in Ref. [22]. 2Fn s−c 2
p (s, Fn ) = 1−⎛ ⎞ c−b≤s≤c+b
In the proposed simulation methodology (see further discussion in πb ⎝ b ⎠ (8)
Section 6), not only the semi-axis b is needed as an output from the
where Fn is the normal force per unit thickness, s is the position along
metamodel, but also the maximum von Mises stress of the normal
the boundary of the cross-section and c is the centre of the contact
contact problem. Therefore, the metamodel has been extended here
patch. The centre of the contact patch is given by the SIMPACK simu-
based on the expressions for the stresses σx , σy and σz below the centre of
lations. In order to ensure that the 2D model results in a similar stress
the contact from the Hertz's solution, see Appendix B, as follows (shear
condition as a 3D model would (for pure normal loading), the load level
stresses are zero due to symmetry):
Fn is adjusted such that the same value of the maximum von Mises stress
3N n n
σx = − [C1 Nx (α, z ) + C2 Mx (α, z ) + C3 Mz (α, z )] σvM, max is obtained as in the metamodel. This means σvM, max is only used
2π ab (4) for the 3D to 2D conversion. This was performed for all 276 load cases
3N for both of the considered materials.
σy = − [C4 Ny (α, z ) + C5 My (α, z ) + C6 Mz (α, z )] Plastic deformation is simulated by the 2D FE models for a load
2π ab (5)
collective of 150 load sequences. The total number of loadings (wheel
3N passages) then sums up to 41400. In these simulations, the tangential
σz = − C7 Q (α, z )
2π ab (6) traction pt is assumed to have the same spatial distribution as the
with all the functions {Nx , Mx , Mz , Ny , My , Mz , Q} defined in Ap- normal pressure p, and where the magnitude is scaled by the utilised
pendix B. For a given set of coefficients {C1, …, C7} and α, it is possible to friction coefficient μeff defined by the ratio of the tangential and normal
compute the equivalent von Mises stress: loads provided by the vehicle dynamics simulations (recall Fig. 8(d)).
The simulations were performed for the rail cross-section at 750 mm
n 1
σvM = (σx − σy )2 + (σy − σz )2 + (σz − σx )2 after the TCP, which is the closest cross-section to where the impact
2 (7)
load is maximum, cf. Fig. 7. Fig. 10(a) shows the range of contact point
n positions c for the 276 loadings, as well as their number of occurrences.
as a function of depth z. The largest von Mises stress is found
σvM, max
n
from the optimality condition dσvM /dz = 0 . The coefficients {C1, …, C7} It is evident that the wheel–rail contact most often happens within a
are determined from the least-square fit with sampled results of σvM,
n
max 5 mm band, between 22 mm and 27 mm along the lateral coordinate y.
from a number of 3D local FE simulations (the same sampling as This is why it is possible to use a non-uniform FE mesh for the cross-

Fig. 10. (a) Distribution of contact points (centre of contact patch) and (b) complete potential contact zone comprised of the range of contact points (in red) and the
semi-axis b (in yellow) for Mn13 concatenated to it. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of
this article.)

308
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

Fig. 11. Plastic deformation (without scaling) of rail profile for steel grades (a) R350HT and (b) Mn13 after 150 load sequences (41400 wheel passages). Dashed
black lines show undeformed mesh.

section, see Fig. 10(b)), with a fine mesh region covering the complete R350HT. Another important conclusion is that it is necessary to simu-
potential contact area (marked in yellow) that encompasses the centres late many load sequences before any stable nodal displacement rate is
of the wheel–rail contact (marked in red), and a coarser mesh outside of obtained for both of the studied materials.
t
it. Here, for the sake of computational efficiency within each iteration The distributions of accumulated plastic strain λ = ∫0 2/3 |ε˙ p | dt
of the simulation methodology, a simplification is made by disregarding (see the material model in Appendix A) in the cross-section are shown
the change in the position of the contact point after each load appli- in Fig. 13. The accumulated plastic strain is larger in the R350HT
cation. Otherwise, the parts I and II of the simulation methodology material than in the rolled Mn13, in particular a few millimetres below
would need to be repeated after each load cycle. the surface. This indicates that the stress-strain loops are wider for the
The resulting deformation of the cross-section at 750 mm after the R350HT material as compared to the loops for the rolled Mn13. This is
TCP is shown for both materials in Fig. 11. It can be noted that the in accordance with the ratchetting experiments in Fig. 9. Also, due to
cross-section made of rolled Mn13 steel experiences a larger shape different contact conditions caused by plasticity, the stress levels in the
change as compared to its counterpart. This can be quantified by materials are different. The larger shape change of the rolled Mn13
studying, for example, the maximum (permanent) nodal displacement cross-section, in particular close to the contact surface, is caused by the
and how it evolves during the first load sequence and during all load higher ratchetting behaviour in the stress-strain loops. This is also il-
sequences, see Fig. 12. During the first sequence the maximum nodal lustrated in Fig. 14 where the ratchetting strain field εrat = 2/3 ε: ε
displacement increases significantly for Mn13 for some of the load cy- (see Ref. [41]) after 150 load sequences is shown.
cles, while for R350HT the increase is more continuous. The reason is On a computer with Intel 2650v3 (“haswell”) CPU and 6.4 GB RAM,
that for cases with high stresses the rolled Mn13 has a lower stiffness the simulation of 150 load sequences required 20 h for the R350HT
(see Fig. 9(b)), which results in large strains and thereby a considerable crossing and 56 h for the rolled Mn13 crossing. To reduce the compu-
shape change. The maximum nodal displacement for Mn13 is in- tational time, the algorithm proposed in Ref. [42] was adopted to in-
creasing significantly during the first 75–80 load sequences but then the clude an extrapolation technique when solving the boundary value
shape change of the crossing nose starts to stabilise. The corresponding problem. The current state S is defined to include all displacements u ,
results for R350HT show that the displacement in the first load se- stresses σ and internal variables q (in the plasticity model) of the FE
quences is significantly smaller, but after 30–40 sequences the dis- model. An extrapolation is adopted when 10 load sequences have been
placement growth stabilises to a higher rate as compared to the rolled simulated. The extrapolation of the state from load sequence N to load
Mn13. Nevertheless, it can be concluded that after 150 load sequences sequence N + ΔN is performed as
(41400 load cycles) the crossing nose made of Mn13 shows around 3.5
times larger displacement as compared to the crossing nose made of SN + ΔN = SN + (SN − SN − 1) ΔN , (9)

Fig. 12. Comparison of maximum permanent nodal displacement for rail steel grades R350HT and Mn13 over (a) the cycles of the first load sequence and (b) over
150 load sequences. Each load sequence is 276 load cycles.

309
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

Fig. 13. Distribution of accumulated plastic strain for (a) R350HT and (b) Mn13 after 150 load sequences.

where also the state SN − 1 from the load sequence N − 1 is used to softer material H:
approximate the current change of state per load sequence. The size of
k
the load sequence increment ΔN is chosen as (here a simplification of V= Ns,
H (11)
the procedure in Ref. [42])
where k is a non-dimensional wear coefficient. Based on the distribu-
ε0 tions of contact pressure and sliding distance obtained from the FAS-
ΔN = floor ⎛⎜ ⎞⎟
⎝ 2/3 maxΩ |εN − εN − 1| ⎠ (10) TSIM discretisation of the contact area, the wear depth at each grid
point within the contact patch is computed as
where floor(•) is an operator that returns the largest integer smaller or
equal to •, ε 0 is a chosen tolerance (strain increment per load sequence), k
Δz (x , y ) = p (x , y )Δs (x , y ),
and maxΩ |εN − εN − 1| is the maximum absolute strain increment among H z (12)
all finite elements. If the equilibrium (solution of the FE problem) with where Δs is the distance a particle on the rail slides with velocity s˙ (x , y )
the extrapolated state SN + ΔN is found, then load sequence N + ΔN + 1 relative to the wheel through a given grid element with length Δx
is also simulated before the extrapolation procedure is repeated to ap- during the time Δt = Δx / v :
proximate the tangent line for the next extrapolation, cf. Eq. (9).
Otherwise, the extrapolated state SN + ΔN is discarded and load sequence Δs (x , y ) = s˙ (x , y ) Δt (13)
N + 1 is simulated followed by a new attempt of the extrapolation. The vector of sliding velocity ṡ is defined as
Another feature of the algorithm is that the solution at certain se-
quences can be requested (for post-processing purposes), whereby the s˙ x = v (γx − ϕy ) (14)
extrapolation algorithm cannot “jump over” these sequences. Fig. 15
s˙ y = v (γy + ϕx ), (15)
shows a comparison of the predicted maximum nodal displacement
with and without the extrapolation algorithm. It also illustrates that where v is the vehicle speed, γx and γy are the longitudinal and lateral
large extrapolation steps may lead to the absence of convergence at a creepages, respectively, and ϕ is the spin creepage. Based on a con-
later stage, as observed for Mn13. However, for a reasonable value of vergence study for the present application, it was concluded that a
the extrapolation parameter ε 0 , the extrapolation showed good preci- FASTSIM discretisation of each contact area into 50 by 50 elements is
sion at the time reduction of 400%. Thus, this extrapolation procedure sufficient.
helps to further improve the methodology in terms of computational In an extensive field test campaign at Niklasdorf in Austria, 20
efficiency. turnouts with five different rail materials have been monitored in terms
of plastic deformation and wear [45]. The turnout geometry is 60E1-
7. Wear 1:12-R500, corresponding to a smaller curve radius and larger crossing
angle than the geometry considered in this paper. Two of these cross-
The Archard model [9] is used to simulate wear due to sliding. It has ings are manufactured from Mn13 and explosively depth hardened (see
been used in several previous applications within the field of railway Ref. [7] for laboratory tests and a comparison with other commonly
mechanics, see e.g. Refs. [13,43,44]. The model suggests that the vo- used rail steels), and subjected to traffic in the facing move at maximum
lume V of material being worn is proportional to the normal force N and speed 140 km/h. Based on measurements of rail profiles in the interval
the sliding distance s, and inversely proportional to the hardness of the 197–677 mm from the TCP, where most transitions from wing rail to

Fig. 14. Distribution of ratchetting strain for (a) R350HT and (b) Mn13 after 150 load sequences.

310
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

Fig. 15. Investigation of strain tolerance ε0 on performance of extrapolation algorithm for (a) R350HT and (b) Mn13.

crossing nose have taken place, and evaluating the corresponding re-
duction in material volume, the average wear rate for these two
crossings is 140 mm3 /MGT. Assuming k is independent of contact
pressure p and sliding velocity ṡ (since no wear map data is available),
this wear rate has been used to calibrate the ratio k / H in Eq. (12). In the
simulation of wear, one load sequence (276 wheel passages) described
in Section 3 was considered and the wear was scaled linearly to cor-
respond to 1 MGT of traffic. For explosively depth hardened Mn13, the
tuned value for k / H = 8.7⋅10−14 m2/N .
Unfortunately, a corresponding tuning of the ratio k / H for R350HT
has not yet been possible as no such turnouts were included in the field
test campaign in Austria. A similar extensive field test campaign on
crossings has been performed at Haste in Germany [46].
The distribution of accumulated wear on the crossing nose after 150
load sequences is determined in a post-processing step by mapping the
wear from each load cycle to a mesh that describes the rail surface (‘rail
mesh’). The mesh of the rail surface is rectangular and covers all po-
tential contact positions in the interval 0.25–1.25 m from the TCP. The
element size is 1 mm and 0.2 mm in the longitudinal and lateral di-
rections, respectively. In each time step Δt and for each wheel–rail
contact, the calculated wear depth for the FASTSIM discretisation Fig. 16. Distribution of accumulated wear for Mn13 after 150 load sequences.
(‘contact mesh’) is mapped to the elements of the rail mesh by linear
interpolation.
Based on the calibrated wear model for Mn13, and considering only
one contact point on the leading wheel, the calculated wear depth
distribution after 150 load sequences is illustrated in Fig. 16. The
maximum wear depth is 134 μm and positioned at 0.53 m from the TCP
and at lateral coordinate 27 mm, which is in the region where most
transitions from the wing rail to the crossing nose take place. Due to the
large spread in worn wheel profiles, the accumulated wear is dis-
tributed over a band that is increasing in width along the crossing nose.
However, the maximum wear is generated in a narrow band located
around lateral coordinate 27 mm, see Fig. 17. At 0.75 m from the TCP,
the maximum accumulated wear depth is 30 μm, cf. Fig. 11. This cor-
responds to about 2% of the maximum plastic deformation, cf.
Fig. 12(b).
Fig. 18 shows the nominal and worn profiles of the manganese
crossing. It is apparent that the worn profile has a non-physical shape.
This is due to the aforementioned simplification of the contact point
location, which within each iteration is assumed to be constant for each
load case. Part IV of the methodology was designed to address this Fig. 17. Accumulated wear for Mn13 after 150 load sequences at three selected
issue. The influence of this simplification is yet to be quantified. cross-sections.

311
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

change of profile and the amount of ratchetting strain were smaller for
the crossing made of R350HT, while less accumulated plastic strain was
predicted for the rolled Mn13. This situation might alter if more load
cycles are considered since the stabilised ratchetting rate is higher for
R350HT as compared to the stabilised ratchetting rate for the rolled
Mn13. The results of ratchetting strain show that many load cycles
(around 20000) need to be computed for both materials before stabi-
lised ratchetting rates are obtained. When it comes to wear, the simu-
lations suggest that the maximum wear occurs soon after the transition
from wing rail to crossing nose. Also, it was observed that the predicted
wear after the initial 41400 load cycles was 2% of the predicted max-
imum plastic deformation for rolled Mn13. The wear model for R350HT
needs to be calibrated once wear map data are available.
In future work, several iterations with the simulation methodology
will be performed to simulate the life of a turnout and to be able to
compare long-term damage of each material. The quality of the pre-
diction will be assessed by comparing results of the simulations with
field measurements for one of the materials. It will also be investigated
Fig. 18. Nominal and worn rail profiles for Mn13 at 0.75 m from TCP after 150 how often the results of the dynamic simulations need to be updated to
load sequences. The worn profile was generated by scaling the wear by a factor capture the changes in the rail profile, and how often the metamodel
50 corresponding to a traffic load of 40 MGT. needs to be recalibrated. Further, the profile change due to plasticity
and wear may result in reduced contact stresses, which has not been
8. Conclusions and outlook accounted for in this paper.

A comparison of two rail steel grades, R350HT and hot-rolled Mn13,


in terms of plastic deformation and wear for one cross-section in a Acknowledgements
crossing has been presented. The damage (plasticity and wear) was
predicted by means of a multidisciplinary simulation methodology as The performed work is part of the activities of the Centre of
described in Ref. [20]. However, in this paper the methodology has Excellence CHARMEC (CHAlmers Railway MEChanics). It is partly fi-
been modified by replacing the FE solution of the wheel–rail contact nanced within the European Horizon 2020450 Joint Technology
problem for all the applied loadings with a metamodelling approach Initiative Shift2Rail through contract no. 730841. The authors would
[22]. The applied loadings are collected in a load sequence, which is like to thank voestalpine VAE GmbH and the Swedish Transport
defined by using Latin hypercube sampling with random combinations Administration (Trafikverket) for their support. The simulations were
of wheel profile, vehicle speed and wheel–rail friction coefficient to performed on resources at Chalmers Centre for Computational Science
mimic a realistic traffic variability. and Engineering (C3SE) provided by the Swedish National
The comparison revealed that after 41400 load cycles the overall Infrastructure for Computing (SNIC).

Appendix A. Ohno–Wang plasticity model

The adopted plasticity model is briefly summarised. For further details, see Ref. [35]. The model is formulated in the small strain framework and
the strain ε is assumed to be additively decomposed into an elastic εe and a plastic εp part:
ε = ε e + εp (A.1)

Linear isotropic elasticity is assumed whereby the following relations for the volumetric stress σvol = I : σ and the deviatoric stress
σdev = σ − σvol/3 I hold in terms of the elastic strain:
σvol = 3 Kb εevol , σdev = 2 G εedev , (A.2)
where G is the shear modulus and Kb is the bulk modulus. The von Mises yield function Φ is adopted which is defined as:
3
Φ= (σdev − X ): (σdev − X ) − σy (σvol )
2

3
= |σdev − X | − σy (σvol ),
2 (A.3)
where X is the kinematic hardening (back stress) and σy is the (initial) yield stress. To improve the model's capability to fit experimental data, the
yield stress σy is chosen to be dependent on σvol such that σy = σytens if σvol ≥ 0 , whereas σy = σycompr if σvol < 0 . The evolution of plastic strain is
assumed to be of associative type (disregarding the discontinuity at σvol = 0 ):
∂f 3 σdev − X
ε̇p = λ˙ = λ˙ ,
∂σ 2 |σdev − X | (A.4)

where the plastic multiplier λ̇ is determined from the loading/unloading conditions:


Φ≤ 0, λ˙ ≥ 0, λ˙ Φ= 0 (A.5)
nX
The kinematic hardening X is assumed to be additively decomposed into several hardening stresses Xi , i.e. X = ∑i = 1 Xi . Following the proposal in
Ref. [35], the evolution equation for the kinematic hardening is given by:

312
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

mi
2 γ mi + 1 3 2 Xi
X˙ i = Ci ε˙ p − i mi ⎛ ⎜ |Xi |⎞ 〈ε˙ p : Xi 〉

3 Ci ⎝ 2 ⎠ 3 |Xi | (A.6)
To increase the identifiability, the values of the parameters: Ci , mi and pi = γimi + 1/ Cimi are determined in the calibration against the experimental
data.

Appendix B. Stresses below the centre of the contact according to Hertz

According to Ref. [38], the stresses below the centre of the contact can be expressed as:

3N ⎡
σx = − (1 − 2ν ) Nx (α, z ) − 2 (1 − ν ) Mx (α, z )
2π ab⎢

+ 2 ν Mz (α, z )] (B.1)

3N ⎡
σy = − (1 − 2ν ) Ny (α, z ) − 2 (1 − ν ) My (α, z )
2π ab⎢

+ 2 ν Mz (α, z )] (B.2)
3N
σz = − Q (α, z )
2π ab (B.3)
and τxy = τyz = τzx = 0 . In these expressions the following functions are used:
α
Q=
(α 2 + ζ 2)(1 + ζ 2) (B.4)

α ⎡ 1 + ζ2 ⎤
Nx = 1−
α2 − 1⎢
⎣ α2 + ζ 2 ⎥
⎦ (B.5)

α ⎡ α2 + ζ 2
Ny = − 1⎤
α2 − 1 ⎢
⎣ 1+ζ
2 ⎥
⎦ (B.6)

ζ (E1 − E2)
Mx =
α2 − 1 (B.7)

ζ (α 2E2 − E1) ζ2 α
My = −
α2 − 1 (α 2 + ζ 2)(1 + ζ 2) (B.8)

1 + ζ2
Mz = α − ζ E2
α2 + ζ 2 (B.9)
where ζ = z / b , E1 and E2 are the complete elliptic integrals of first and second kinds, respectively, with the modulus k2 =1− 1/ α 2 and the argument
arctanα / ζ .

References 981–988, https://doi.org/10.1063/1.1721448.


[10] S.C. Lim, M.F. Ashby, Overview no. 55 Wear-Mechanism maps, Acta Metall. 35 (1)
(1987) 1–24, https://doi.org/10.1016/0001-6160(87)90209-4.
[1] R. Enblom, Deterioration mechanisms in the wheel–rail interface with focus on [11] M.A. Cremona, B. Liu, Y. Hu, S. Bruni, R. Lewis, Predicting railway wheel wear
wear prediction: a literature review, Veh. Syst. Dyn. 47 (6) (2009) 661–700, under uncertainty of wear coefficient, using universal kriging, Reliab. Eng. Syst. Saf.
https://doi.org/10.1080/00423110802331559. 154 (2016) 49–59, https://doi.org/10.1016/j.ress.2016.05.012.
[2] S. Kumar, V. Aronov, B.R. Rajkumar, R. Margasahayam, Experimental investigation [12] J. De Arizon, O. Verlinden, P. Dehombreux, Prediction of wheel wear in urban
of plastic flow in rails for a laboratory wheel rail simulation, Can. Metall. Q. 21 (1) railway transport: comparison of existing models, Veh. Syst. Dyn. 45 (9) (2007)
(1982) 59–66, https://doi.org/10.1179/cmq.1982.21.1.59. 849–866, https://doi.org/10.1080/00423110601149335.
[3] A.F. Bower, K.L. Johnson, Plastic flow and shakedown of the rail surface in repeated [13] T. Jendel, Prediction of wheel profile wear – comparisons with field measurements,
wheel–rail contact, Wear 144 (1) (1991) 1–18, https://doi.org/10.1016/0043- Wear 253 (1) (2002) 89–99, https://doi.org/10.1016/S0043-1648(02)00087-X.
1648(91)90003-D. [14] R. Enblom, M. Berg, Simulation of railway wheel profile development due to wear –
[4] K.A. Meyer, M. Ekh, J. Ahlström, Modeling of kinematic hardening at large biaxial influence of disc braking and contact environment, Wear 258 (7) (2005)
deformations in pearlitic rail steel, Int. J. Solids Struct. 130–131 (2018) 122–132, 1055–1063, https://doi.org/10.1016/j.wear.2004.03.055.
https://doi.org/10.1016/j.ijsolstr.2017.10.007. [15] I. Zobory, Prediction of wheel/rail profile wear, Veh. Syst. Dyn. 28 (2–3) (1997)
[5] M. Wiest, W. Daves, F. Fischer, H. Ossberger, Deformation and damage of a crossing 221–259, https://doi.org/10.1080/00423119708969355.
nose due to wheel passages, Wear 265 (9) (2008) 1431–1438, https://doi.org/10. [16] T.G. Pearce, N.D. Sherratt, Prediction of wheel profile wear, Wear 144 (1) (1991)
1016/j.wear.2008.01.033. 343–351, https://doi.org/10.1016/0043-1648(91)90025-P.
[6] S. Alfi, S. Bruni, Mathematical modelling of train–turnout interaction, Veh. Syst. [17] L. Xin, V. Markine, I. Shevtsov, Numerical analysis of the dynamic interaction be-
Dyn. 47 (5) (2009) 551–574, https://doi.org/10.1080/00423110802245015. tween wheel set and turnout crossing using the explicit finite element method, Veh.
[7] S. Eck, H. Oßberger, U. Oßberger, S. Marsoner, R. Ebner, Comparison of the fatigue Syst. Dyn. 54 (3) (2016) 301–327, https://doi.org/10.1080/00423114.2015.
and impact fracture behaviour of five different steel grades used in the frog of a 1136424.
turnout, Proc. Inst. Mech. Eng. - Part F J. Rail Rapid Transit 228 (6) (2014) [18] U. Ossberger, S. Eck, E. Stocker, Performance of different materials in a frog of a
603–610, https://doi.org/10.1177/0954409713511078. turnout, Proceedings of the 11th International Heavy Haul Conference, 2015, pp.
[8] C. Casanueva, E. Doulgerakis, P.-A. Jönsson, S. Stichel, Influence of switches and 329–336, , https://doi.org/10.13140/RG.2.1.2549.7442 Perth, Australia.
crossings on wheel profile evolution in freight vehicles, Veh. Syst. Dyn. 52 (sup1) [19] E. Kassa, C. Andersson, J.C.O. Nielsen, Simulation of dynamic interaction between
(2014) 317–337, https://doi.org/10.1080/00423114.2014.898779. train and railway turnout, Veh. Syst. Dyn. 44 (3) (2006) 247–258, https://doi.org/
[9] J.F. Archard, Contact and rubbing of flat surfaces, J. Appl. Phys. 24 (8) (1953) 10.1080/00423110500233487.

313
R. Skrypnyk, et al. Wear 428–429 (2019) 302–314

[20] A. Johansson, B.A. Pålsson, M. Ekh, J.C.O. Nielsen, M.K.A. Ander, J. Brouzoulis, [33] A. Olsson, G. Sandberg, O. Dahlblom, On Latin hypercube sampling for structural
E. Kassa, Simulation of wheel-rail contact and damage in switches & crossings, Wear reliability analysis, Struct. Saf. 25 (1) (2003) 47–68, https://doi.org/10.1016/
271 (1–2) (2011) 472–481, https://doi.org/10.1016/j.wear.2010.10.014. S0167-4730(02)00039-5.
[21] D. Nicklisch, J.C.O. Nielsen, M. Ekh, A. Johansson, B.A. Pålsson, A. Zoll, [34] M. Schilke, J. Ahlström, Laboratory Tests of Material Properties. INNOTRACK
J. Reinecke, Simulation of wheel-rail contact and subsequent material degradation Deliverable Report D3.1.6, Tech. Rep. Department of Materials and Manufacturing
in switches & crossings, Proceedings 21st International Symposium on Dynamics of Technology, Chalmers University of Technology, Gothenburg, Sweden, 2010http://
Vehicles on Roads and Tracks, Stockholm, Sweden, 2009, pp. 1–14 (available www.charmec.chalmers.se/innotrack/deliverables/sp3/d316-f3-recommendation_
on CD). of_and_scientific_basis_for_optimisation_of_switches_and_crossings_part2.pdf.
[22] R. Skrypnyk, J.C.O. Nielsen, M. Ekh, B.A. Pålsson, Metamodelling of wheel–rail [35] N. Ohno, J.-D. Wang, Kinematic hardening rules with critical state of dynamic re-
normal contact in railway crossings with elasto-plastic material behaviour, Eng. covery, part I: formulation and basic features for ratchetting behavior, Int. J. Plast.
Comput. 35 (1) (2019) 139–155, https://doi.org/10.1007/s00366-018-0589-3. 9 (3) (1993) 375–390, https://doi.org/10.1016/0749-6419(93)90042-O.
[23] L. Xin, V. Markine, I. Shevtsov, Numerical procedure for fatigue life prediction for [36] J.A. Nelder, R. Mead, A simplex method for function minimization, Comput. J. 7 (4)
railway turnout crossings using explicit finite element approach, Wear 366 (2016) (1965) 308–313, https://doi.org/10.1093/comjnl/7.4.308.
167–179, https://doi.org/10.1016/j.wear.2016.04.016. [37] K.L. Johnson, Contact Mechanics, Cambridge University Press, 1987.
[24] Dassault Systèmes Simulia Corp., SIMPACK, (2017) Available from: http://www. [38] G. Lundberg, H. Sjövall, Stress and Deformation in Elastic Contacts, the Institution
simpack.com , Accessed date: 16 March 2018. of Theory of Elasticity and Strength of Materials, Chalmers University of
[25] S. Iwnicki, Manchester benchmarks for rail vehicle simulation, Veh. Syst. Dyn. 30 Technology, Gothenburg, Sweden, 1958.
(3–4) (1998) 295–313, https://doi.org/10.1080/00423119808969454. [39] S.H. Nia, C. Casanueva, S. Stichel, Prediction of RCF and wear evolution of iron-ore
[26] B.A. Pålsson, J.C.O. Nielsen, Dynamic vehicle-track interaction in switches and locomotive wheels, Wear 338–339 (2015) 62–72 https://doi.org/10.1016/j.wear.
crossings and the influence of rail pad stiffness - field measurements and validation 2015.05.015.
of a simulation model, Veh. Syst. Dyn. 53 (6) (2015) 734–755, https://doi.org/10. [40] Z. Wei, C. Shen, Z. Li, R. Dollevoet, Wheel–rail impact at crossings: relating dynamic
1080/00423114.2015.1012213. frictional contact to degradation, J. Comput. Nonlinear Dyn. 12 (4) (2017) 11,
[27] B.A. Pålsson, Optimisation of railway crossing geometry considering a re- https://doi.org/10.1115/1.4035823.
presentative set of wheel profiles, Veh. Syst. Dyn. 53 (2) (2015) 274–301, https:// [41] A. Esmaeili, M.S. Walia, K. Handa, K. Ikeuchi, M. Ekh, T. Vernersson, J. Ahlström, A
doi.org/10.1080/00423114.2014.998242. methodology to predict thermomechanical cracking of railway wheel treads: From
[28] H. Hertz, Über die Berührung fester elastischer Körper (On the contact of elastic experiments to numerical predictions, Int. J. Fatigue 105 (2017) 71–85, https://doi.
solids), J. für die Reine Angewandte Math. (Crelle's J.) 92 (1881) 156–171 (For org/10.1016/j.ijfatigue.2017.08.003.
English translation see Miscellaneous Papers by H. Hertz, London: Macmillan [42] G. Johansson, M. Ekh, On the modeling of large ratcheting strains with large time
(1896) 146-162). increments, Eng. Comput. 24 (3) (2007) 221–236, https://doi.org/10.1108/
[29] J.J. Kalker, A fast algorithm for the simplified theory of rolling contact, Veh. Syst. 02644400710734945.
Dyn. 11 (1) (1982) 1–13, https://doi.org/10.1080/00423118208968684. [43] A. Asadi, M. Brown, Rail vehicle wheel wear prediction: a comparison between
[30] B.A. Pålsson, J.C.O. Nielsen, Track gauge optimisation of railway switches using a analytical and experimental approaches, Veh. Syst. Dyn. 46 (6) (2008) 541–549,
genetic algorithm, Veh. Syst. Dyn. 50 (sup1) (2012) 365–387, https://doi.org/10. https://doi.org/10.1080/00423110701589430.
1080/00423114.2012.665167. [44] J.C.O. Nielsen, B.A. Pålsson, P.T. Torstensson, Switch panel design based on si-
[31] B.A. Pålsson, Robust evaluation of rail damage and track forces using representative mulation of accumulated rail damage in a railway turnout, Wear 366–367 (2016)
load collectives, The Dynamics of Vehicles on Roads and Tracks, CRC Press, 2016, 241–248, https://doi.org/10.1016/j.wear.2016.06.021.
pp. 841–850, , https://doi.org/10.1201/b21185-90. [45] U. Ossberger, Voestalpine VAE GmbH, Austria, Private Communication, (February
[32] M.D. McKay, R.J. Beckman, W.J. Conover, Comparison of three methods for se- 2018).
lecting values of input variables in the analysis of output from a computer code, [46] A. Zoll, Werkstoffauswahl für Weichenherzstücke durch Prüfstandsversuche, Ph.D.
Technometrics 21 (2) (1979) 239–245, https://doi.org/10.1080/00401706.1979. thesis Technische Universität Berlin, Shaker Verlag, Germany, 2016, p. 186.
10489755.

314

You might also like