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

net/publication/366599572

Study on corrugated wear on high-speed railways based on an improved finite

element model of wheel-rail rolling contact

Article in Tribology International · December 2022

DOI: 10.1016/j.triboint.2022.108199

CITATION READS

1 57

8 authors, including:

Jiabao Pan Dongdong Ye

Anhui Polytechnic University Anhui Polytechnic University
25 PUBLICATIONS 64 CITATIONS 31 PUBLICATIONS 248 CITATIONS

SEE PROFILE SEE PROFILE

Zhang Rongyun
Anhui University
30 PUBLICATIONS 61 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Advanced coating manufacture technology View project

PMSM control View project

All content following this page was uploaded by Jiabao Pan on 02 January 2023.

The user has requested enhancement of the downloaded file.

 Tribology International 179 (2023) 108199

Contents lists available at ScienceDirect

Tribology International
journal homepage: www.elsevier.com/locate/triboint

Study on corrugated wear on high-speed railways based on an improved

finite element model of wheel-rail rolling contact
Bowen Wu *, Wenjing Wang, Jiabao Pan *, Yan Hu, Rui Xu, Dongdong Ye, Wei Yan,
Rongyun Zhang
School of Mechanical Engineering, Anhui Polytechnic University, Wuhu 241000, China

A R T I C L E I N F O A B S T R A C T

Keywords: An improved method for simulation of the transient wheel-rail rolling contact (WRRC) was proposed. A 3D
Mixed Lagrangian/Eulerian method implicit/explicit finite element (FE) model was established to calculate the transient response of the wheel-rail
3D wheel-rail rolling contact system (WRS). The dynamic wheel-rail forces excited by rail weld joints were investigated to reveal the causes of
High-speed lines
corrugated wear on high-speed lines. The results show that the wheel-rail forces obtained by the proposed
Corrugated wear
method can quickly converge to the steady-state values without obvious initial disturbance, and the convergence
rate is almost independent of the wheel speed. The vibration in vertical direction of the track at 655.6 Hz excited
by the weld joint is identified as the main cause for the corrugated wear in 125–160 mm on high-speed lines.

1. Introduction
Corrugated wear (Rail corrugation or corrugation) is a common
defect observed on the rail surface and characterized by periodic and
non-uniform wear distribution in longitudinal direction. Strong vibra­
tion can be excited when a railway wheelset rolls over a corrugated rail,
which will lead to fatigue failure and shortened service lifespan of the
structural components of the railway system and safety threat to train
operation [1–3]. The treatments of rail corrugation increase the main­
tenance and operation costs of the railway system. Rail corrugation has
been mainly recorded in the subway system. In recent years, rail
corrugation has also been detected on some high-speed lines in China. At
high operation speed, the vibration caused by rail corrugation is more
violent and is one of safety risks of the high-speed trains [4]. Therefore,
the causes of corrugated wear on high-speed railways need to be
revealed to develop prevention and control measures.
The research on the causes of corrugated wear is mainly based on the
wavelength-fixing mechanism proposed by Grassie [5]. Namely, under
random irregularity or parametric excitation, the sensitive mode of the
vehicle-track system (VTS) is excited. Resonance occurs and leads to
fluctuated wear at the frequency of the sensitive mode. Initial corruga­
tion will generate on the rail surface and reversely promotes to reso­
nance development. Finally, visible corrugated wear will form. Based on
this theory, various resonances were identified as the wavelength-fixing
mechanisms of corrugated wear, such as the pinned-pinned resonance
[6–8]; the P2 resonance [9], the vertical resonance [10] and longitu­
dinal resonance [11] of the track; the rail bending and torsional vibra­
tion [12,13] and the wheelset torsional resonance [14,15] and bending
resonance [16]. Besides, the instability of the WRS is found to be another
important inducement for rail corrugation, such as the self-excited vi­
bration [17–19] and the stick-slip vibration [20], which are considered
to be the dominant inducement for corrugated wear on metro tracks
with tight radius. Excited by saturated tangential force, the self-excited
vibration will occur, without the need for the excitation of external ir­
regularity, and result in the fluctuating wear until corrugation is formed.
The self-excited vibration provides a good explanation for the phe­
nomenon that most corrugation occur on low rail [21].
Rail corrugation on Chinese high-speed lines has only been observed
in recent years and the research on its generation mechanism is still
insufficient. Ma et al. [22] set up a coupled dynamic model of the VTS
taking into consideration of multiple wheel-rail interactions and calcu­
lated the wheel-rail friction power taking into consideration of the wave
reflections between wheels. The 3rd-order bending of the rail between
bogie wheels and the pinned-pinned resonance were identified as the
wavelength-fixing mechanisms of the corrugated wear. Yu et al. [8]
investigated the transient response of the WRS of the high-speed railway
excited by initial corrugation by using a 3D explicit FE model. The
pinned-pinned resonance in vertical direction was considered as the
formation cause of the corrugated wear. Cui et al. [23] investigated the
corrugated wear on high-speed lines based on the coupling of multiple

* Corresponding authors.
E-mail addresses: wubowen@ahpu.edu.cn (B. Wu), panjiabao@ahpu.edu.cn (J. Pan).

https://doi.org/10.1016/j.triboint.2022.108199
Received 22 November 2022; Received in revised form 15 December 2022; Accepted 25 December 2022
Available online 26 December 2022
0301-679X/© 2022 Elsevier Ltd. All rights reserved.
B. Wu et al. Tribology International 179 (2023) 108199

vibrations. The wheel-rail tangential force is assumed to be saturated

under a large traction or braking torque when the vehicle passes a sharp
gradient section. The self-excited vibration is excited and results in the
corrugated wear. Bao et. al [24] established a FE model of the WRS with
the braking subsystem to study the corrugated wear on a braking sec­
tion. Stability analysis revealed that the self-excited vibration excited by
the friction of the braking system and the saturated tangential force is
the cause of the corrugated wear on the long-downhill braking section.
Rail corrugation originates from differential wear on the contact
patch and the wear depends on the relative sliding velocity and the
contact forces in contact patch. Therefore, the key to reveal the initiation
and evolution of the corrugated wear is to solve the WRRC [25,26]. At
present there are two major methods in solving the WRRC: one is based
on the multi-body dynamic model of the VTS and the other one is the 3D
FE model of WRRC. In the dynamic model, the wheel/rail normal con­
tact is treated as the Hertzian [27] or non-Hertzian [28] contact problem
and the tangential force is generally calculated by Kalker’s theories [29].
The dynamic model has the advantage of faster calculation and becomes
more comprehensive after decades of development from the
frequency-domain [30,31] to time-domain [32,33], from two-dimension
[28] to three-dimension [34], from steady-state [31] to non-steady state
Fig. 1. Roughness level spectra of rail corrugation on a high-speed line.
[26] and from only considering the vertical dynamics [33,35] to
considering vertical and lateral coupled dynamics [36,37]. However,
the rail is often simplified as Timoshenko beam and the wheel is introduced to the dynamic model, which will take a so-called “dynamic
simplified as a point mass without the consideration of the real contact relaxation zone” to decay until stable solutions are obtained. However,
geometry in the dynamic model. In addition, the dynamic model at higher speed, the initial disturbance will take a long time or even be
encounter difficulties in dealing with multi-point contact. impossible to decay.
The 3D FE model can solve the WRRC under the real contact ge­ In this paper, an improved explicit FE model for simulation of the
ometry without the assumptions, such as half space, constant friction WRRC is proposed. The proposed method has one more key step
and steady-state rolling, of the classical contact theories. The lateral, compared with the traditional method, namely, simulation of the wheel
longitudinal creepage and the spin effect are automatically included in acceleration in a steady-state analysis. The improved method shows
the FE model without increasing in the complexity of this model. Wen significant advantages over the traditional method in terms of the effi­
et al. [38] developed an explicit FE model of WRRC to investigate the ciency of solving the transient WRRC, especially at high speed. Based on
strain and stress caused by the wheel-force impact force at the weld the proposed method, the transient response of the WRS excited by the
joint. Li et al. [39] improved this model so that it can simulate the rail weld joint was investigated. The formation causes for rail corruga­
traction and braking of the wheelset by applying a torque on the wheel tion on Chinese high-speed lines was revealed.
axle. Zhao and Li [40] studied the wheel-rail contact based on the
improved explicit FE model and verified the results with Hertzian the­ 2. Rail corrugation on a high-speed line
ories and Kalker’s theories. Later, Zhao and Li extended the explicit FE
model of the WRRC to the cases of velocity dependent friction [41] and The irregularity on a corrugated rail surface of a high-speed line was
rolling contact in elasto-plasticity [42]. Yang et al. [43] tested the recorded [55]. The corrugated tracks investigated in this paper are
capability of the FE model of the WRRC in solving multi-point rolling straight tracks with gradient varying from 1‰ to 6.5‰, on which the
contact and capturing the transient dynamic effects of the transition trains operate at a speed of 300 km/h with a traction coefficient of 0.3.
from single-point to multi-point contact. In recent years, the FE model of The roughness levels of the rail surface at four sites of the measured
the WRRC is widely adopted to investigate the wheel/rail dynamic track are presented in Fig. 1. Two obvious peaks in the range of 125–160
interaction [44], the wheel-rail contact mechanics [45–47] and the mm and 63–80 mm respectively can be observed in the roughness level
material damage mechanisms of the rail or the wheel [48–52]. spectra. In site 1, site 2 and site 3, the wavelength around 63–80 mm is
Despite the advantages mentioned above, the computational effi­ dominant. However, in site 4, the wavelength around 125–160 mm
ciency of the explicit FE model for WRRC is low. Research shows that the dominates. A wavelength around 31 mm also can be observed in site 4,
element size on the contact zone must be refined to at least 1.3 mm × although not so obvious. As is shown in Fig. 1, the typical wavelengths of
1.3 mm to obtain the contact solutions with sufficient engineering ac­ the corrugated wear on the measured high-speed line are around 63–80
curacy [40]. Thus, up to millions of elements may have to be needed. mm and 125–160 mm. The passing frequencies of these two types of
Moreover, a very small time increment (microsecond level) in the typical corrugation at 300 km/h fall within 1041–1322 Hz and 521–667
explicit dynamic analysis is required to meet the Courant criterion [53], Hz, respectively.
which further reduces the efficiency of the FE model. To control the
calculation cost, most models only include half of a single wheelset and 3. Description of model
half of the track system with the sprung mass simplified as a lumped
mass. In that way, the application of this FE model is limited. Therefore, 3.1. 3D FE model for rolling contact
it is needed to develop a FE model for WRRC with high computing ef­
ficiency. The simulation of the WRRC based on the traditional explicit FE The 3D FE model illustrated in Fig. 2 for transient WRRC is devel­
model requires two steps. First, the static contact solution is obtained as oped with the commercial FEM software ABAQUS/Explicit. A righted-
the initial conditions for subsequent analysis with an implicit static hand Cartesian coordinate system is established and its z-axle is paral­
analysis. Then, roll the wheelset by applying a torque in an explicit lel to the track longitudinal direction with the y-axle parallel to the
dynamic analysis. Due to the differences between the results obtained lateral direction of the track and the x-axle perpendicular to the track.
from static and dynamic analysis and instantaneous impact of the wheel The positive direction of the x-axle points to the wheel. The model in­
centrifugal force in dynamic analysis [54], an initial disturbance is cludes a wheelset and a slab ballastless track. The investigated track is

2
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 2. 3D FE model for transient WRRC: (a) Structural diagram; (b) FE model in isometric view and (c) FE model in front view.

Table 1 Table 2
Main parameters in the FE model. Material parameters.
Parameters and symbols Values Parameters and symbols Values

Coefficient of friction ( f) 0.5 Wheel and rail Young’s modulus (E) 210 GPa
Sprung mass (M) 6350 kg Poisson’s ratio (ν) 0.3
Wheel radius (R) 430 mm Density (ρ) 7800 kg/m3
Vertical stiffness of the primary suspension (K) 880 kN/m Track slab Young’s modulus (E) 34500 MPa
Damping of the primary suspension (C) 4 kN⋅s/m Poisson’s ratio (ν) 0.3
Fastener stiffness (Kf) 22 MN/m Density (ρ) 2400 kg/m3
Fastener damping (Cf) 200 kN⋅s/m

taken from a straight section. The transverse movement, the roll angle of
the wheelset are almost zero in the straight track. In that case, the
transient WRRC on both sides of the track can be considered to be
symmetrical. Therefore, only half of the WRS is included in this model
for the frugality of simulation time. Symmetrical boundary is set on
longitudinal halved surfaces of the wheel axle and the track slab. The
established track in the FE model is 15.2 m long, including 23 group of
fasteners. The symmetrical boundary is applied to the rail ends. The
fastener system is simulated by several groups of vertical viscous spring.
Each fastener includes 21 viscous springs. Because the car body and the
bogie frame have a negligible influence on high-frequency wheel-rail
dynamic interactions, the sprung mass is replaced by a concentrated
mass (M in Fig. 1a), which is connected to the axle through a vertical
viscous spring simulating the primary suspension. To ensure that the Fig. 3. Traditional method for simulation of the transient WRRC.
longitudinal and lateral movement of the sprung mass is synchronized
with the wheel, the point mass is coupled with the axle in the longitu­ Fig. 2(a)) at the rail is 2.6 m long, to which refined mesh are applied. The
dinal and lateral directions. The mortar layer is not considered in this contact surface in AB and the wheel tread are discretized into small el­
model. The bottom of the slab is fixed in three directions. ements as fine as 1.1 × 1.1 mm2. Coarse mesh is applied outside the
The wheel, rail and the track slab are discretized by the 8-node, contact surface. The FE model has a total of 1391762 nodes and
reduced integration, 3D solid elements. The rolling zone (AB in the 1182856 elements. The wheel has a LMa-type tread profile, its nominal

3
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 4. Simulation scheme of the transient WRRC based on an improved method.

diameter is 860 mm. The wheel-rail contact based on the penalty

method is established to solve the normal contact solution. The Coulomb
law is selected to solve the tangential wheel-rail force. The friction co­
efficient at the contact interface is set to 0.5 [54]. Other parameters are
given in Table 1. The material parameters are given in Table 2.

3.2. Implicit-explicit Co-simulation of the transient WRRC

Implicit-explicit Co-simulation is implemented to solve the transient

WRRC at high speed, for which the traditional method [39–54] (see
Fig. 3) requires two steps. In step 1, a gravity load of 9800 mm/s2 (G in
Fig. 3) is applied to the FE model in an implicit static analysis. The stress
field and displacement field under the static equilibrium state are ob­
tained. In step 2, the obtained stress and displacement are imported into
the explicit dynamic model as the initial conditions. Meanwhile, the
predefined angular velocity (ω in Fig. 3) around the axle and translation
velocity (v in Fig. 3) in the longitudinal direction are applied to the
wheel at time t = 0. In addition, the same translation velocity is assigned
to the sprung mass. To simulate the driving of the wheel, a
time-dependent traction torque (T in Fig. 3) is applied to the axle. Then,
the explicit time integration is adopted to solve the transient WRRC. Due
to the differences between the displacements and stresses obtained from
the implicit static analysis and the explicit dynamic analysis, an initial
disturbance will be introduced into the dynamic model after the static
results are transferred, under which an initial instability will be
observed in the transient rolling contact solution. For this reason, a
so-called “dynamic relaxation zone” must be designed to decay the vi­
bration caused by the initial disturbance until the steady-state contact
solution is obtained. More importantly, the centrifugal force caused by
the sudden application of the initial rotation velocity will lead to radial
displacement and stress of the wheel, which will produce a stronger
initial excitation [56]. Given this factor, a longer dynamic relaxation
zone may be required to obtain the steady-state contact solution.
Especially at high speed, the required length of dynamic relaxation zone
may make the FE model too large and affect the efficiency of the
traditional model.
To address the shortcoming of the traditional method, an improved
method is proposed. The simulation of the transient WRRC based on the
proposed method is composed of four steps shown in Fig. 4. In step 1, the
gravitational acceleration is loaded into the whole model to solve the
static equilibrium state just like the traditional method does. Next,
applying a traction torque to the axle in an implicit static analysis step 2.
To avoid the disturbance caused by the centrifugal force, a wheel ac­
celeration step is arranged before rolling the wheel at a high speed. The
acceleration can be implemented in an explicit dynamic step. However,
it will be at the cost of simulation efficiency, because it requires a long
acceleration zone. In the proposed method, the wheel acceleration is
conducted with a so-called “steady-state analysis” based on mixed
Lagrangian/Eulerian method, in which Eulerian method is adopted to
described the rigid body rotation of the wheel, whereas Lagrangian
Fig. 5. Kinematic decomposition of the mixed Lagrangian/Eulerian
description.
method is selected to described the material deformation. In that way,
the mesh of the wheel does not need to be rotated, instead, the wheel
material flows in the mesh. One of the advantages of this way is that the
simulation cost is independent of the wheel speed and the steady-state
results can be obtained directly in this steady-state analysis. Thus, sig­
nificant reduction in simulation time can be achieved. In addition,
because the inertial effect of the wheel can be considered in the wheel
acceleration analysis, the centrifugal force from the wheel rotation is
automatically included in the model. Last, the steady-state results
including the rotation and translation speeds of the wheel obtained from
step 3 are imported into the transient dynamic model as the initial
conditions of the analysis step 4. Releasing the wheel after applying
initial forward speed to the sprung mass and torque to the wheel axle,
the transient dynamic analysis on the WRRC just can be carried out.
3.3. Steady-state rolling analysis with mixed Lagrangian/Eulerian method
In the mixed Lagrangian/Eulerian method, Eulerian method is
adopted to described the rigid body rotation of the wheel. The material
deformation of the wheel and the track are described in Lagrangian
method [57]. The kinematic decomposition of the mixed Lagrangia­
n/Eulerian description of the steady-state rolling of the wheel is shown
in Fig. 5. The wheel rotates around the axle at a rotation velocity ω0, and
moves at a translation velocity v0. Ω0 represents the initial configura­
tion. Ω is the current configuration. Ωis
̂ the reference configuration.
Assume the following one-to-one mappings exist

4
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 6. Comparison of vertical forces at different speeds: (a) Spatial domain (100 km/h), (b) Details of (a) in the range of 1500–2500 mm, (c) Frequency domain
(100 km/h); (d) Spatial domain (200 km/h), (e) Frequency domain (200 km/h).

χ = χ (X, t) ⎡ ⎤
̂ χ, t) (1) cos ωt − sin ωt 0
x = ϕ(
Rs = ⎣ sin ωt cos ωt 0⎦ (3)
When a material particle moves from initial position X to position Y 0 0 1
at t = Δt, the motion of a material particle is composed of the rigid
Y = x + v0 t (4)
rotation χ = χ (X, t) and the material deformation x = ϕ(
̂ χ, t). The rigid
rotation χcan be expressed as The velocity of the material particle can be written as
χ = Rs ⋅(X − X0 ) (2) ∂x ̂ ∂χ ∂ ϕ
∂ϕ ̂
v = Ẏ = + v0 = v0 + + (5)
∂t ∂χ ∂t ∂t
here Rs is the rotation matrix
the circumferential direction is defined as

5
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 7. Comparison of lateral forces at different speeds: (a) Spatial domain (100 km/h), (b) Frequency domain (100 km/h), (c) Spatial domain (200 km/h), (d)
Frequency domain (200 km/h).

T × (χ − X0 ) ⎧
S= (6) ⎪
⎨ fP fullslip
R √̅̅̅̅̅̅̅̅̅̅̅̅̅̅
τeq = τ1 + τ2 = fPΔv (11)
⎪ stick − slip
here T is direction vector of the rotation axis. R = |χ − X0 |is the rotation ⎩
2Ff ωR
radius of the material particle. The reference body velocity can be
written as ∂χ(X, t)/∂t = ω0 RS, so that here τeq is the resultant frictional stress, τ1 and τ2 are the lateral and
longitudinal frictional stress, respectively. P is the contact pressure and
̂
∂ϕ ̂
∂ϕ ̂ ∂ϕ
∂ϕ ̂
v = v0 + ω0 R ⋅S + = v0 + ω0 R + (7) Δv is the relative velocity at the contact interface. Ff is a user-defined slip
∂χ ∂t ∂S ∂t tolerance.
For steady state, ∂ ϕ/
̂ ∂t = 0, then The frictional stress increases linearly with the slip ratio defined as
Δv/ωR. When the relative slip velocity Δv reaches the maximum
̂
∂ϕ allowable value, full slip will occur. In that case, the frictional stress
v = v0 + ω0 R (8)
∂S equals to the sliding friction fP.
The acceleration of the material particle can be expressed as
3.4. Transient response analysis with explicit integration method
̂
∂2 ϕ
a = v̇ = ω0 R 2 2
(9)
∂S2 The explicit integration is selected to solve the dynamic equilibrium
equation of the WRS with the frictional rolling contact. The discretized
The virtual work contribution from the d Alembert forces is
′
dynamic equation of the WRS at time t is expressed as
∏ ∫ ̂ ∂δV
∂d ϕ (12)
dδ = ω0 2 ρ R2 ⋅ dV (10) Mt ẍ = Pt − It + Ft
V ∂S ∂S
where M, P, I and F are the mass matrix, the internal force vector, the
where ρ is the mass density. The wheel inertia can be considered auto­ external force vector and the contact force vector, respectively. ẍ is the
matically through the Eq. (10). acceleration vector. The displacement x at increment i + 1 is written as
The Coulomb friction law is used to calculate the frictional stress.
The frictional stress is calculated by

6
B. Wu et al.
Fig. 8. Wheel modes: (a) 1-order axle bending mode at 91 Hz, (b) Umbrella mode at 279 Hz and (c) Radial mode at 3266 Hz.
( ) ( )
i+12 i+12
Δt(i+1) + Δt(i) (i)
ẋ = ẋ + ẍ
2 (13)
( )
i+12
x(i+1) = x(i) + Δt(i+1) ẋ
where Δt is the time increment. The explicit time integration scheme is
conditionally stable. Convergence can be guaranteed only when Δt is
less than the critical value. The critical time increment is determined
based on the Courant criterion [53].
Lc
Δt ≤
Cd
where Lc can be taken as the minimum element dimension. Cd is the
speed of the dilatational wave, which is calculated by
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
E(1 − ν)
Cd = (15)
(1 + ν)(1 − 2ν)ρ
7
The critical time increment in this model is 1.513 × 10− s.
4. Simulation results
4.1. Comparison of results obtained from two methods
The simulation results predicted by the improved method are
compared with that obtained by the traditional method. Fig. 6 depicts
the comparison of the vertical wheel-rail forces (hereafter vertical force)
and Fig. 7 depicts the comparison of the lateral wheel-rail forces. Sig­
nificant initial oscillation in the vertical forces calculated by the tradi­
tional method can be found in Fig. 6(a) and Fig. 7(a), which are mainly
attributed to the instantaneous shock from the wheel centrifugal force.
The amplitudes of oscillation increase with the speed and attenuate with
the rolling distance due to damping. From Fig. 6(b) it is found that when
the wheel rolls to the longitudinal position of 1.5 m at 100 km/h, the
fluctuation amplitude of the vertical force is about 1.6 kN and is about
2.4% of the static value. The amplitude is beyond the acceptable range
and it can be considered that steady-state solution has been obtained.
Therefore, a dynamic relaxation zone with a length of 1.5 m is adequate
for 100 km/h. However, when the wheel rolls to the longitudinal posi­
tion of 2.5 m at 200 km/h, the oscillation amplitude of the vertical force
plotted in Fig. 6(d) is still up to 10 kN, approximately. It is too large for
steady-state solution. In other words, at 200 km/h, the vibration in
vertical force predicted by the traditional method cannot be effectively
attenuated in the dynamic relaxation zone. The dynamic relaxation zone
must be extended to obtained a stable solution. But as depicted in Fig. 6.
(d), the peak vertical force decays from the initial 83–71.5 kN, only
about 14% reduction is achieved. Slow decay rate of the vertical force
Tribology International 179 (2023) 108199
means that the extended dynamic relaxation zone will be very long,
which will significantly affect the efficiency of the traditional method. In
comparison, the vertical forces at both 100 km/h and 200 km/h pre­
dicted by the improved method are very stable from the beginning,
because the steady-state solutions have been obtained in the accelerated
analysis step.
Fig. 6(c) and (e) depict the vertical forces in frequency-domain ob­
tained from fast Fourier transform. Several obvious peaks can be
observed in Fig. 6(c) and (e). These two peaks at 44.4 Hz and 88.8 Hz
are caused by discrete fastener support or sleeper passing, which can be
predicted by both methods. Besides, two additional peaks at about
111 Hz and 3300 Hz are predicted by the traditional method at 100 km/
(14)
h, which are not able to find in the results from the improved method or
their amplitudes are relatively small. Another additional peak at 266 Hz
is found in the results from the traditional method at 200 km/h. More
significantly, the amplitude of the peak at about 3300 Hz is quite high,
even exceeds the one at the sleeper passing frequency, which will take a
very long dynamic relaxation zone to decay.
Similarly, the lateral forces (see Fig. 7(a) and (c)) predicted by the
improved method are very stable and the steady-state lateral forces are
about 5 kN. The lateral forces on the straight track are caused by the
wheel-rail contact angle. Again, the lateral forces predicted by the
traditional method experienced a strong initial vibration. From Fig. 7(b)
and (d), it is observed that the lateral forces in the frequency domain
present several peaks very similar to those of the vertical forces.
Fig. 8 shows three modes of the wheel including the 1-order bending
mode of the axle at 91 Hz, the umbrella mode at 279 Hz and the radial
mode at 3266 Hz, which have close frequencies to the peaks shown in
Figs. 6 and 7. The peaks at about 3300 Hz results from the radial mode.
The radial mode is excited by the instantaneous shock from the sudden
loading of the wheel centrifugal force at the beginning, which can be
avoided in the improved model. The peaks at 111 Hz depicted in Figs. 6
and 7 may be caused by the 1-order axle bending mode. The peaks at
266 Hz plotted in Fig. 7(d) may be attributed to the umbrella mode.
To further determine the causes for the two peaks at 111 Hz and
266 Hz, the frequency response of the wheel to harmonic excitation is
investigated by steady-state dynamic analysis. As depicted in Fig. 9(a),
the excitation location is selected on the wheel tread. The displacement
outputs of two nodes respectively located on the tread (marked as node
1) and the axle (marked as node 2) are selected to analyze the frequency
response characteristics of the wheel. Two kinds of harmonic excitation
are fed into the model, one considers the vertical force (excitation 1 in
Fig. 9(a)) and the other considers the lateral force (excitation 2). Fig. 9
(b) and (c) respectively plot the frequency response curves of the vertical
and lateral displacements of the nodes to excitation 1 and excitation 2.
In Fig. 9(b) three obvious resonance peaks at 91 Hz, 306 Hz and 872 Hz
can be found, of which the peaks at 91 Hz corresponds to the 1-order
axle bending mode. No peak is found near the frequency of the
7
B. Wu et al.
Fig. 9. Frequency response characteristics of the wheel to harmonic excitations: (a) Locations of excitation point and response points, (b) Frequency response to
excitation 1 and (c) Frequency response to excitation 2.
umbrella mode. In Fig. 9(c) three obvious resonance peaks can also be
found at 91 Hz, 279 Hz and 435 Hz. The peaks at 279 Hz corresponds to
the umbrella mode. The frequency response analysis shows that the 1-
order axle bending mode is excited by the vertical and lateral force.
The umbrella mode at 279 Hz is excited by the lateral force.
From the above analysis it is found that the improved method shows
good performance in simulating the transient WRRC. In the traditional
method, applying rotational speed to the wheel in the explicit analysis
step will excite the radial mode of the wheel, which will trigger a strong
initial vibration in the wheel-rail force, and then induce the 1-order axle
bending mode and the umbrella mode of the wheel. The resulted initial
vibration needs a long dynamic relaxation zone to decay. However, the
steady-state solution can be obtained at the beginning of simulation with
the improved method, which is a significant advantage for simulation of
the transient WRRC at higher speed.
8
Tribology International 179 (2023) 108199
4.2. Transient wheel-rail rolling contact on smooth rail
The above analysis shows that when the speed is increased to
200 km/h, the traditional method cannot quickly obtain stable wheel-
rail force, Therefore, the results at higher speeds will not be compared
in this paper. The performance of the improved model in solving the
transient WRRC at higher speed will be tested separately in this section.
The tested speed is 300 km/h, which is the operation speed of most
Chinese high-speed lines at present. The rail is smooth without any
external irregularity. The transient rolling contact for pure rolling and
traction are solved, respectively. The coefficient of traction μ is defined
as:
FL
μ= (16)
FN
where FL and FN are respectively the longitudinal and vertical force. In
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 10. Energy curves with the time during explicit dynamic analysis: (a) Ratio of the artificial strain energy to the internal energy and (b) Total energy.

Fig. 11. Wheel-rail force on smooth rail at 300 km/h: (a) Pure rolling (b) Accelerating with a traction coefficient of 0.3 and (c) Frequency spectrum of the verti­
cal force.

traction condition, the wheel is accelerated by a torque T applied to the
axle, which is balanced with the torque due to the longitudinal force,
that is, the applied torque T equals to the product of the longitudinal
force FL, traction coefficient μ and the rolling radius R.
Energy output is used to check the reliability of the results obtained
from explicit dynamic analysis. Artificial strain energy is an important
indicator to the development of hourglass deformation peculiar to the
reduced integration element. A large artificial strain energy indicates
severe mesh distortion and unreliable simulation results. In common
practice, a proportion less than 10% of the artificial strain energy in the
internal energy is acceptable. In addition, the total energy of the system
should be constant during the simulation due to conservation of energy.
Fig. 10 depicts the energy curves with the time during the explicit
dynamic analysis. From Fig. 10 (a) it is found that the maximum pro­
portion of the artificial strain energy in the internal energy is 0.13%. The
total system energy depicted in Fig. 10 (b) is almost constant. Therefore,
the aforementioned requirements can be met.
Fig. 11 depicts the wheel-rail forces on a smooth rail at 300 km/h.
Fig. 11 (a) represents the results under pure rolling condition, Fig. 11 (b)
represents the ones under traction condition with a coefficient of trac­
tion of 0.3 and Fig. 11 (c) represents the frequency spectrum of the
vertical forces. It can be seen that steady-state vertical forces at 300 km/
h under both pure rolling and traction conditions are obtained at the
beginning without obvious fluctuation. The steady-state values are
approximately 67.2 kN, which is very close to vertical forces obtained
from the static analysis. It is also found that the vertical forces fluctuate

9
B. Wu et al.
Fig. 12. Surface stresses on the contact patch: (a) Contact pressure, (b) Longitudinal shear stress (μ = 0.3), (c) Longitudinal shear stress (pure rolling) and (d) Stress
distribution along the longitudinal direction (μ = 0.3).
with a wavelength equals to the fastener spacing, and the peak vertical
forces appear in front of the fasteners. The vertical force under pure
rolling is almost identical with that under traction condition.
Under the pure rolling condition, the longitudinal wheel-rail force
quickly reaches its steady-state value after a sudden drop. In addition, a
slight fluctuation is found in the process of longitudinal force
rebounding. When the wheel rolls to about 0.8 m, the longitudinal force
become very stable. This sudden drop in the longitudinal force may
originate from the difference of the wheel rotation velocity corre­
sponding to pure rolling in the dynamic analysis and in the steady-state
analysis. As is shown in Fig. 11 (b), after an initial disturbance, the
longitudinal force under traction condition also reaches the steady-state
value of 10 kN at about 0.8 m. In Fig. 11 (c), except for a peak near the
sleeper-passing frequency, the curve of the frequency spectrum of the
vertical force demonstrates an overall downward trend, which is
reasonable because the modes with higher frequencies require more
energy to be excited. The amplitude of the vibrating vertical force
excited by the discrete supported track is about 0.55 kN.
Fig. 12 shows the surface stresses on the contact patch at 0.0288 s
when the steady-state solution is obtained. Fig. 12 (a) shows the contact
pressure. The contact patch is elliptical and these two axis are 13.9 mm
and 9.6 mm long respectively. The maximum contact stress is about
1080 MPa. Laterally the gradient of the contact stress at the side of the
rail gauge corner is bigger than that at the other side, which is related to
the gradual increase of the rolling circle radius along the direction
Fig. 13. Rail weld geometry in the FE model: (a) Side view and (b) Front view.
10
Tribology International 179 (2023) 108199
towards the wheel flange. Longitudinally the contact stress is basically
symmetrical. Fig. 12 (b) and (c) are the longitudinal shear stresses when
the traction coefficient μ = 0 and 0.3, respectively. Under the traction
condition, asymmetric shear stress distribution can be observed. The
maximum shear stress is about 328.3 MPa and located behind the rolling
direction, near the side of the wheel flange. However, the distribution
zone of the shear stress under pure rolling is divided into two patches,
which are symmetrical about the longitudinal direction. The direction of
the shear stress on the patch near the wheel flange is opposite to that of
the shear stress on the other patch. Thus, the resultant longitudinal shear
stress is zero. The stress distributions along the longitudinal direction
under traction condition are depicted in Fig. 12 (d). For comparison, the
static contact stress is plotted together. It is found that the maximum
contact stress from the static analysis is bigger than that from the dy­
namic analysis, which is one of the reasons for the initial disturbance
when the static results are transferred into the dynamic model.
Comparing the sliding friction and the shear stress, it is found that the
shear stress is divided into slip zone and stick zone, which is consistent
with the analytical results.
The above results show that the improved method performs well in
obtaining the steady-state solutions of transient rolling contact rapidly
at higher speed.
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 14. Wheel-rail forces excited by a weld joint with wavelength λ = 125 mm: (a) Wheel-rail forces in spatial domain, (b) Vertical force in frequency domain and
(c) Longitudinal force in frequency domain.

Fig. 15. Wheel-rail forces excited by rail weld joints with different wavelengths: (a) Vertical forces and (b) Longitudinal forces.

11
B. Wu et al.
Fig. 16. Peak values and valley values of the vertical and longitudinal forces: (a) Vertical forces and (b) Longitudinal forces.
4.3. Transient wheel-rail rolling contact excited by rail weld joint
The improved method is adopted to solve the transient WRRC when
the wheel rolls over a raised weld joint. The weld joint geometry is
introduced into the model by modifying the coordinates of element
nodes on rail surface. In the longitudinal direction the geometry of the
weld joint is simplified as a sine wave containing one wavelength. In the
lateral direction the amplitude distribution is parabolic. The vertical
coordinates of the weld joint are expressed as [8].
{ [ ( )]}
A 2π
x= 1 − cos z − zmin
2 λ
(17)
( ymin + ymax )2 Amax
A=−
2Amax
y− + joint with shorter wavelength. A linear upward trend in the peak vertical
2
(ymax − ymin ) 2 2
where A is the amplitude, Amax is the maximum amplitude. z is the
longitudinal coordinates of nodes and zmin is the longitudinal coordinate
of the starting position of the weld joint geometry. λ is the wavelength of
the weld point. y is the lateral coordinates and ymax and ymin are the
maximum and minimum lateral coordinates, respectively. Fig. 13 shows
the weld joint geometry introduced into the FE model. Note that the
weld joint in Fig. 13 is magnified to facilitate observation.
Fig. 14 plots the wheel-rail forces excited by a weld joint with
wavelength λ = 125 mm and maximum amplitude Amax= 0.2 mm at
300 km/h. The traction coefficient is 0.3. It is found that a strong impact
on wheel-rail forces occurs when the wheel rolls over the weld joint. The
peak vertical force is about 134.5 kN, which is about 2 times of the
steady-state value. The valley value of the vertical force is about 4.4 kN,
the wheel/rail separation almost occurs. The phases of the vertical and
Fig. 17. Frequency spectrum of the wheel-rail forces excited by weld joints with different wavelengths: (a) Vertical forces and (b) Longitudinal forces.
12
Tribology International 179 (2023) 108199
longitudinal forces are ahead of the weld joint geometry, and the vertical
force and longitudinal force are almost in phase. After passing the weld
joint, the fluctuation in wheel-rail forces continues. In Fig. 14 (b) and
(c), two peaks in the frequency spectrums of the vertical and longitu­
dinal forces are observed. The frequencies of the peaks in the vertical
force spectrum are 436 Hz and 610 Hz, respectively. For the longitudi­
nal force spectrum, the frequencies are 349 Hz and 610 Hz.
Fig. 15 shows the wheel-rail forces excited by weld joints with
wavelength varying from 160 mm to 65 mm. These weld joints have the
same amplitude, which is 0.2 mm. Fig. 16 depicts the peak values and
valley values of the vertical and longitudinal forces. From Fig. 15, it is
observed that the vertical force vibrates more violently over a rail weld
force with the decreasing of the wavelength can be observed. The peak
vertical force increase from 120.8 kN for wavelength λ = 160 mm to
160.6 kN for wavelength λ = 65 mm. When the wavelength is reduced
to 80 mm and 65 mm, the valley values of the vertical force are zero,
which indicates wheel-rail disengagement. However, the changes in the
peak longitudinal force with the wavelength is not obvious. The peak
longitudinal force varies nonlinearly in the range 13–15 kN with the
wavelength.
The frequency distributions of the vertical and longitudinal forces
excited by these weld joints are plotted in Fig. 17. From Fig. 17 (a), it is
found that two distinct peaks in 436–470 Hz and 610–657 Hz are
highlighted in the vertical force frequency spectrums, of which the peak
in 610–657 Hz dominants. It suggests that some modes of the wheel-rail
system within these frequency ranges are excited by the weld joint. In
addition, the vibration in the range of 436–470 Hz gradually weakens
B. Wu et al. Tribology International 179 (2023) 108199

Fig. 18. Wheel-rail forces excited by a weld joint with wavelength of 125 mm at different speeds: (a) Vertical forces and (b) Longitudinal forces.

Fig. 19. Peak values and valley values of the wheel-rail forces: (a) Vertical forces and (b) Longitudinal forces.

Fig. 20. Frequency spectrum of the wheel-rail forces at different speeds: (a) Vertical forces and (b) Longitudinal forces.

with the wavelength decreases. The vibration in the range of
610–657 Hz becomes strongest when the wavelength is 125 mm. It may
be because that the frequencies of the excited modes approach the
passing frequency (666.7 Hz) of the weld joint with wavelength of
125 mm. Additionally, when the wavelength is reduced to 80 mm and
65 mm, a weak peak appears in the range of 1440–1455 Hz, it indicates
that high-frequency vibration in the vertical direction will be excited
under the excitation of the weld joint with shorter wavelength. Similar
frequency distribution can be observed in the longitudinal force
spectrums shown in Fig. 17 (b). Similarly, two obvious peaks can be
found in Fig. 17 (b), but in 349–375 Hz and 563–648 Hz. When the
wavelength is reduced to 80 mm and 65 mm, two peaks in
1364–1440 Hz and 2092–2160 Hz appear.
Figs. 18–20 show the vertical forces and longitudinal forces excited
by a weld joint with wavelength of 125 mm at different speeds. From
Figs. 18 and 19 it is found that the vibration amplitudes of both the
vertical forces and longitudinal forces increase with the speed increases.
Both the peak vertical force and peak longitudinal force demonstrate a

13
B. Wu et al.
Fig. 21. Vertical vibration of the track system at 655.6 Hz.
linear upward trend with the speed increases and their valley values
demonstrate an overall downward trend. The peak vertical force in­
creases from 113.9 kN at 200 km/h to 143.3 kN at 350 km/h and the
peak longitudinal force increases from 13 kN to 15 kN. Two obvious
peaks in vertical force spectrum are observed in Fig. 20 (a), whose fre­
quencies fall within the range of 414–444 Hz and the range of
592–650 Hz, respectively. At 250–350 km/h, the vibration of vertical
forces in 592–650 Hz dominant. However, at 200 km/h, the vibration in
414–444 Hz dominants because the passing frequency (444 Hz) of the
weld joint at 200 km/h just falls within this frequency range. The lon­
gitudinal force spectrums also present two peaks in 348–414 Hz and
592–650 Hz, respectively.
4.4. Rail corrugation analysis
Two types of rail corrugation with wavelength within 65–80 mm and
125–160 mm (see Fig. 1) are recorded. Pinned-pinned resonance is
generally considered as the main inducement for the corrugated wear in
65–80 mm [8,22], but the formation causes of the rail corrugation in
125–160 mm are still controversial. In the reference [22], the 3rd-order
rail bending mode at 658 Hz excited by wave transmissions and re­
flections between bogie wheels was identified as the cause of the rail
corrugation in 125–160 mm. In the reference [23], however, the
self-excited vibration of the WRS at 607 Hz was considered as the cause
of the corrugated wear in 125–160 mm. The self-excited vibration is
excited by saturated creep force in braking section and can be an
explanation for rail corrugation in braking section, but rail corrugation
in non-braking section cannot be explained by friction-induced vibra­
tion. Field measurements show that initial rail corrugation on
high-speed lines first occurs near the weld joint [58], therefore, the ef­
fect of the rail weld joint on corrugated wear cannot be ignored.
From the results presented in Section 4.2 it is found that the vertical
vibration of the WRS is excited by the weld joint. The vertical vibration
results in the oscillation in vertical force, and thus causes the oscillation
of the tangential force or the longitudinal creep force. The dominant
frequencies of the vertical forces excited by weld joints with wavelength
varying from 65 mm to 160 mm all fall within the range of 610–660 Hz.
At 300 km/h, the dominant frequencies are close to the passing fre­
quency of the corrugated wear with wavelength within 125–160 mm.
Therefore, it is reasonable to consider that the main cause for the
corrugated wear in 125–160 mm on high-speed lines is the vertical vi­
bration of the WRS excited by the weld joint.
From the frequency responses of the wheel shown in Fig. 9, no
resonance peak can be found within 610–660 Hz. Therefore, the vertical
vibration should not be caused by the wheel resonance modes. Model
analysis was carried out to investigate the vibration characteristics of
the track system. A vertical vibration mode of the track system (see
Fig. 21) is found and its frequency falls exactly within 610–660 Hz. The
mode at 655.6 Hz is the vertical vibration of the rail relative to the slab,
which is expected to be main cause for the vibration of the vertical force
14
Tribology International 179 (2023) 108199
in 610–660 Hz. From Fig. 17 it is found that the amplitude of the vi­
bration in 610–660 Hz is maximum when the wavelength of the weld
joint is 125 mm. At 300 km/h, the passing frequency of the weld joint
with wavelength λ = 125 mm is about 666.7 Hz and is close to the fre­
quency of the mode shown in Fig. 21, which further proves that the
mode at 655.6 Hz is excited by the weld joints.
From the above analysis it is concluded that the main cause for the
corrugated wear in 125–160 mm on high-speed railway tracks is the
vertical vibration of the rail relative to the slab at 655.6 Hz. When the
wheel rolls over the weld joint with wavelength of 125 mm at different
speeds, the dominant frequencies of the vertical forces change slightly. It
suggests that the corrugated wear caused by the vertical vibration near
the weld joint is frequency-fixing.
5. Conclusions
A 3D FE model of the WRS was developed using the explicit FE
method to solve the transient WRRC. The traditional method for simu­
lation of the transient WRRC requires two steps. In step 1, the static
equilibrium solution is solved with an implicit finite element method
and imported into the explicit dynamic finite element model as the
initial conditions of the step 2. In step 2, an initial rotational velocity is
applied to the wheel to simulate the rolling of the wheel. Strong impact
will be introduced into the explicit dynamic model duo to instantaneous
loading of the centrifugal force originated from the wheel rotational
velocity. A so-called “dynamic relaxation zone” must be designed to
decay the initial disturbance caused by that impact until the steady-state
solution is obtained. At high speed, the initial disturbance caused by the
centrifugal force will take a very long dynamic relaxation zone to decay,
which will greatly affect the efficiency of the traditional method. In this
study, an improved method was proposed. In the proposed method, a
wheel acceleration step implemented in a steady-state rolling analysis
based on the mixed Lagrangian/Eulerian method is arranged before
rolling the wheel in the explicit dynamic analysis. Based on the proposed
method, the transient response of the WRS when the wheel rolls over a
weld joint was simulated. The effect of the rail weld joint on corrugated
wear was analyzed. The following conclusions were drawn.
(1) Strong vibration in the vertical wheel-rail forces calculated by the
traditional method can be caused by the radial mode at 3266 Hz
excited by instantaneous loading of the centrifugal force. The
vibration amplitude increases significantly with the vehicle
speed. At 100 km/h, the vibration amplitude can be attenuated to
an acceptable level within the designed dynamic relaxation zone.
When the speed is increased to 200 km/h, however, the vibration
amplitude is still up to 10 kN at the end of the dynamic relaxation
zone and only about 14% reduction in the amplitude is achieved.
It indicates that the vibration will take a long time or even is
impossible to attenuate at higher speed.
(2) The proposed method performs well in solving the transient
WRRC. Thanks to an additional wheel acceleration analysis step
arranged before the explicit dynamic analysis, the impact caused
by instantaneous loading of the centrifugal force can be avoided
in the proposed method. Therefore, the wheel-rail forces obtained
by the proposed method can quickly converge to their steady-
state values without obvious initial disturbance and the conver­
gence rate is almost independent of the vehicle speed.
(3) Strong vibration in the wheel-rail forces is excited by the rail weld
joint. The vibration amplitude increases obviously with the
wavelength of the weld joint decreases. When the wavelength is
reduced to 80 mm and 65 mm, the peak vertical wheel-rail forces
are almost two times of the steady-state values. The valley values
of the vertical wheel-rail forces are almost zero, which indicates
wheel-rail disengagement. The vibration amplitude increases
linearly with the vehicle speed.
B. Wu et al.
Table A1
Full names corresponding to abbreviations in the paper.
Abbreviations Full names
WRRC Wheel-Rail Rolling Contact
FE Finite Element
WRS Wheel-Rail System
VTS Vehicle-Track System
(4) The dominant vibration frequency of the wheel-rail force is
within 610–660 Hz. The dominant frequency varies slightly with
the wavelength of the weld joint and the vehicle speed and it is
very close to the passing frequency of the corrugated wear in
125–160 mm. The mode at 655.6 Hz in which the rail vibrates
vertically relative to the track system was found. Therefore, the
formation cause of rail corrugation was identified as the vertical
vibration of the rail at 655.6 Hz excited by the weld joint.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
Data availability
Data will be made available on request.
Acknowledgement
The authors thank the financial support from the Natural Science
Foundation of the Anhui Higher Education (KJ2020A0360), the Na­
tional Natural Science Foundation of China (51805057), the Industrial
Collaborative Innovation Special Fund Project of Anhui Polytechnic
University & Jiujiang District (2022cyxta1), Anhui Provincial Key
Research and Development Project (202104a05020002,
2022a05020004), the Natural Science Foundation of Anhui Provincial
(2108085ME169).
Appendix A
Abbreviations in this paper are summarized in Table A1.
References
[1] Liu XG, Wang P. Investigation of the generation mechanism of rail corrugation
based on friction induced torsional vibration. Wear 2021;468:203593.
[2] Wei ZL, Sun XF, Yang F, Ke ZT, Lu T, Zhang P, Chen Shen. Carriage interior noise-
based inspection for rail corrugation on high-speed railway track. Appl Acoust
2022;196:108881.
[3] Wei L, Sun Y, Zeng J, Qu S. Experimental and numerical investigation of fatigue
failure for metro bogie cowcatchers due to modal vibration and stress induced by
rail corrugation. Eng Fail Anal 2022;142:106810.
[4] Wang KY, Liu PF, Zhai WM, Huang C, Chen ZG, Gao JM. Wheel/rail dynamic
interaction due to excitation of rail corrugation in high-speed railway. Sci China
Technol Sci 2015;58:226–35.
[5] Grassie SL. Rail corrugation: characteristics, causes, and treatments. Proc Inst Mech
Eng Part F: J Rail Rapid Transit 2009;223:581–96.
[6] Hiensch M, Nielsen JCO, Verheijen E. Rail corrugation in The
Netherlands—measurements and simulations. Wear 2002;253:140–9.
[7] Correa N, Oyarzabal O, Vadillo EG, Santamaria J, Gomez J. Rail corrugation
development in high speed lines. Wear 2011;271:2438–47.
[8] Yu M, Wang WD, Liu JZ, Sun SC. The transient response of high-speed wheel/rail
rolling contact on “roaring rails” corrugation. Proc Inst Mech Eng Part F: J Rail
Rapid Transit 2019;233:1068–80.
[9] Grassie SL, Edwards JW. Development of corrugation as a result of varying normal
load. Wear 2008;265:1150–5.
[10] Xiao H, Yang S, Wang HY, Wu SX. Initiation and development of rail corrugation
based on track vibration in metro systems. Proc Inst Mech Eng Part F: J Rail Rapid
Transit 2018;232:2228–43.
[11] Li SG, Li ZL, Núñez A, Dollevoet R. New insights into the short pitch corrugation
enigma based on 3D-FE coupled dynamic vehicle-track modeling of frictional
rolling contact. Appl Sci 2017;7:807.
15
Tribology International 179 (2023) 108199
[12] Li W., Du X., Wang H.Y., Wu L., Li X., Wen Z.F., Jin X.S. Investigation into the
mechanism of type of rail corrugation of metro. Journal of Mechanical Engineering
201; 49: 26–32. (in Chinese).
[13] Li X, Li W, Shen YY, Wen ZF, Jin XS. Study on the rail corrugation of the ladder-
type sleepers track based on the track vibration theory. J Mech Eng 2016;52:
121–8.
[14] Grassie SL, Elkins JA. Rail corrugation on North American transit systems. Veh Syst
Dyn 1998;29:5–17.
[15] Matsumoto A, Sato Y, Tanimoto M, Qi K. Study on the formation mechanism of rail
corrugation on curved track. Veh Syst Dyn 1996;25:450–65.
[16] Carlberger A, Torstensson PT, Nielsen JCO, Frid A. An iterative methodology for
the prediction of dynamic vehicle–track interaction and long-term periodic rail
wear. Proc Inst Mech Eng Part F: J Rail Rapid Transit 2018;232:1718–30.
[17] Chen GX, Zhou ZR, Ouyang H, Jin XS, Zhu MH, Liu QY. A finite element study on
rail corrugation based on saturated creep force-induced self-excited vibration of a
wheelset–track system. J Sound Vib 2010;329:4643–55.
[18] Wu BW, Chen GX, Kang X, Zhu Q. Study on the origin of rail corrugation at a long
downhill braking section based on friction-excited oscillation. Tribology Trans
2020;63:439–52.
[19] Wu BW, Liu T, Pan JB, Zhang RY. Study on the effect of the fastener support
structure on rail corrugation in metros based on the friction-induce vibration. J Vib
Control 2022;28:3705–18.
[20] Sun YQ, Simson S. Wagon–track modelling and parametric study on rail
corrugation initiation due to wheel stick-slip process on curved track. Wear 2008;
265:1193–201.
[21] Chen GX, Zhang S, Wu BW, Zhao XN, Wen ZF, Ouyang H, Zhu MH. Field
measurement and model prediction of rail corrugation. Proc Inst Mech Eng Part F:
J Rail Rapid Transit 2020;234:381–92.
[22] Ma CZ, Gao L, Xin T, Cai XP, Nadakatti MM, Wang P. The dynamic resonance under
multiple flexible wheelset-rail interactions and its influence on rail corrugation for
high-speed railway. J Sound Vib 2021;498:115968.
[23] Cui XL, He ZQ, Huang B, Chen YC, Du ZX, Qi W. Study on the effects of wheel-rail
friction self-excited vibration and feedback vibration of corrugated irregularity on
rail corrugation. Wear 2021;477:203854.
[24] Bao PY, Cui XL, Ding HH, Yin Y, Du ZX, Yang ZC. Influences of friction self-excited
vibration characteristics of various types of high-speed trains on rail corrugations
in braking sections. Eng Fail Anal 2022;134:106087.
[25] Grassie SL. Rail corrugation: advances in measurement, understanding and
treatment. Wear 2005;258:1224–34.
[26] Knothe K, Groß-Thebing A. Short wavelength rail corrugation and non-steady-state
contact mechanics. Veh Syst Dyn 2008;46:49–66.
[27] Xie G, Iwnicki SD. Calculation of wear on a corrugated rail using a three-
dimensional contact model. Wear 2008;265:1238–48.
[28] Neilsen JB. Evolution of rail corrugation predicted with a non-linear wear model.
J Sound Vib 1999;227:915–33.
[29] Kalker JJ. Three-Dimensional Elastic Bodies In Rolling Contact. Springer Science &
Business Media,; 2013.
[30] Gómez J, Vadillo EG, Santamaría J. A comprehensive track model for the
improvement of corrugation models. J Sound Vib 2006;293:522–34.
[31] Gómez I, Vadillo EG. A linear model to explain short pitch corrugation on rails.
Wear 2003;255:1127–42.
[32] Jin XS, Wen ZF, Wang KY, Zhang WH. Effect of a scratch on curved rail on
initiation and evolution of rail corrugation. Tribol Int 2004;37:385–94.
[33] Jin XS, Wen ZF, Wang KY, Xiao XB. Effect of passenger car curving on rail
corrugation at a curved track. Wear 2006;260:619–33.
[34] Jin XS, Wen ZF, Wang KY, Zhou ZR, Liu QY, Li CH. Three-dimensional train–track
model for study of rail corrugation. J Sound Vib 2006;293:830–55.
[35] Igeland A, Ilias H. Rail head corrugation growth predictions based on non-linear
high frequency vehicle/track interaction. Wear 1997;213:90–7.
[36] Sun YQ, Dhanasekar M, Roach D. A three-dimensional model for the lateral and
vertical dynamics of wagon-track systems. Proc Inst Mech Eng Part F: J Rail Rapid
Transit 2003;217:31–45.
[37] Andersson C, Johansson A. Prediction of rail corrugation generated by three-
dimensional wheel–rail interaction. Wear 2004;257:423–34.
[38] Wen ZF, Jin XS, Zhang WH. Contact-impact stress analysis of rail joint region using
the dynamic finite element method. Wear 2005;258:1301–9.
[39] Li ZL, Zhao X, Dollevoet R, Molodova M. Differential wear and plastic deformation
as causes of squat at track local stiffness change combined with other track short
defects. Veh Syst Dyn 2008;46:237–46.
[40] Zhao X, Li ZL. The solution of frictional wheel–rail rolling contact with a 3D
transient finite element model: Validation and error analysis. Wear 2011;271:
444–52.
[41] Zhao X, Li ZL. A solution of transient rolling contact with velocity dependent
friction by the explicit finite element method. Eng Comput 2016.
[42] Zhao X, Li ZL. A three-dimensional finite element solution of frictional wheel–rail
rolling contact in elasto-plasticity. Proc Inst Mech Eng Part J: J Eng Tribology
2015;229:86–100.
[43] Yang Z, Li ZL, Dollevoet R. Modelling of non-steady-state transition from single-
point to two-point rolling contact. Tribol Int 2016;101:152–63.
[44] Zhao X, Li ZL, Dollevoet R. The vertical and the longitudinal dynamic responses of
the vehicle–track system to squat-type short wavelength irregularity. Veh Syst Dyn
2013;51:1918–37.
[45] Gao Y, Xu JM, Liu YB, Dong ZG, Wang P, Jiang ZQ. An investigation into transient
frictional rolling contact behaviour in a switch panel: validation and numerical
simulation. Veh Syst Dyn 2022;60:114–31.
B. Wu et al.
[46] Toumi M, Chollet H, Yin H. Finite element analysis of the frictional wheel-rail
rolling contact using explicit and implicit methods. Wear 2016;366:157–66.
[47] Yang Z, Deng X, Li Z. Numerical modeling of dynamic frictional rolling contact
with an explicit finite element method. Tribol Int 2019;129:214–31.
[48] Zhao X, Zhao XG, Liu C, Wen ZF, Jin XS. A study on dynamic stress intensity factors
of rail cracks at high speeds by a 3D explicit finite element model of rolling contact.
Wear 2016;366:60–70.
[49] Li ZL, Zhao X, Dollevoet R. An approach to determine a critical size for rolling
contact fatigue initiating from rail surface defects. Int J Rail Transp 2017;5:16–37.
[50] Zhao X, An BY, Zhao XG, Wen ZF, Jin XS. Local rolling contact fatigue and
indentations on high-speed railway wheels: Observations and numerical
simulations. Int J Fatigue 2017;103:5–16.
[51] Yang Z, Boogaard A, Wei ZL, Liu JZ, Dollevoet R, Li ZL. Numerical study of wheel-
rail impact contact solutions at an insulated rail joint. Int J Mech Sci 2018;138:
310–22.
16
View publication stats
Tribology International 179 (2023) 108199
[52] Zhao X, Zhang P, Wen ZF. On the coupling of the vertical, lateral and longitudinal
wheel-rail interactions at high frequencies and the resulting irregular wear. Wear
2019;430:317–26.
[53] Courant R, Friedrichs K, Lewy H. On the partial difference equations of
mathematical physics. IBM J Res Dev 1967;11:215–34.
[54] Zhao X, Wen ZF, Wang HY, Jin XS, Zhu MH. Modeling of high-speed wheel-rail
rolling contact on a corrugated rail and corrugation development. J Zhejiang Univ
Sci A 2014;15:946–63.
[55] Gu YL. Study on the Mechanism Of Rail Corrugation On High-speed Railway
Unballasted Track (Doctoral Thesis). Beijing Jiaotong University; 2017.
[56] He YP, Han J, Zhou Q, Sheng XZ. An improved finite element model for three-
dimensional wheel–rail rolling contact. Veh Syst Dyn 2022;60:4144–65.
[57] Wei YT, Feng XJ, Zhou FQ, Xiang DB. Simulation of rolling noise based on the
mixed Lagrangian–Eulerian method. Tire Sci Technol 2016;44:36–50.
[58] Yu M. Transient Simulation For High-speed Track/vehicle System And Study On
Rail Corrugation (Doctoral Thesis). China Academy of Railway Sciences; 2019.