applied

sciences
Article
Observation and Simulation of Axle Box Acceleration
in the Presence of Rail Weld in High-Speed Railway
Boyang An 1,2 , Ping Wang 1,2 , Jingmang Xu 1,2, *, Rong Chen 1,2, * and Dabin Cui 1,3
1 Key Laboratory of High-Speed Railway Engineering, Ministry of Education, Southwest Jiaotong University,
Chengdu 610031, China; bk20081412@my.swjtu.edu.cn (B.A.); wping@home.swjtu.edu.cn (P.W.);
cdb1645@home.swjtu.edu.cn (D.C.)
2 School of Civil Engineering, Southwest Jiaotong University, Chengdu 610031, China
3 School of Mechanical Engineering, Southwest Jiaotong University, Chengdu 610031, China
* Correspondence: jingmang87@home.swjtu.edu.cn (J.X.); chenrong@home.swjtu.edu.cn (R.C.);
Tel.: +86-028-87600617 (J.X.); +86-028-87600623 (R.C.)

Received: 17 October 2017; Accepted: 30 November 2017; Published: 4 December 2017

Abstract: Rail welds are widely used in high-speed railways and short-wave irregularities usually
appear due to limitations in welding technology. These irregularities can excite a high wheel/rail force
and are regarded as the main cause of deterioration in track structures. To measure this fierce force (or
deterioration of the rail weld), axle box acceleration is treated as an effective and economic measure,
though an exact quantitative relation between these two quantities remains elusive. This paper aims
to develop such a relation in order to provide a new theoretical basis and an analysis method for
monitoring and controlling weld geometry irregularity. To better understand the characteristics of
axle box acceleration, the paper consists of two parts: an observation and a numerical simulation of
axle box acceleration by rail welds. Based on measured data from field tests, axle box acceleration at
rail welds was found to have high-frequency vibrations in two frequency bands (i.e., 350–500 Hz and
1000–1200 Hz). Upon analyzing the vibration characteristics in time–frequency domains, the exact
location of the rail weld irregularity could be identified. Subsequently, a 3D high-speed wheel/rail
rolling contact finite element model was employed to investigate the effect of rail weld geometry
on axle box acceleration, and led to the discovery that the weld length and depth determine the
vibration frequency and amplitude of the axle box acceleration, respectively. A quantitative relation
between axle box acceleration and wheel/rail force has also been determined. Finally, we propose an
approach for real-time health detection of rail welds and discuss the influence of other defects and
rail welds on the acceleration signal of the axle box.

Keywords: high-speed railway; rail welds; axle box acceleration; wheel/rail force; explicit finite
element method

1. Introduction
On a traditional railway track, a fishplate and bolt, as shown in Figure 1, are used to join rails [1].
The large dynamic force generated by geometric discontinuities between a wheel and the rail at joint
gaps is one of the weak points in track structure. However, as shown in Figure 1, rails of standard length
can be welded into rails of extended length, which is one important measure towards eliminating
rail joints.
Flash welding and thermite welding are mainly used in rail welding. However, absolute
smoothness of the geometric surface of the rail weld cannot be attained due to limitations in welding
technology and improper operation, etc. If the temperature of the rail surface and adjacent area is
lower than the melting temperature during flash welding, achieving complete contact between the
melt and parent metal is hard and crystalline lenses, micro-pores, and non-metallic impurities will

Appl. Sci. 2017, 7, 1259; doi:10.3390/app7121259 www.mdpi.com/journal/applsci

Appl. Sci. 2017, 7, 1259 2 of 16
Appl. Sci. 2017, 7, 1259 2 of 16
melt and parent metal is hard and crystalline lenses, micro-pores, and non-metallic impurities will
appear in the welds during forging [2]. In addition, the strength, hardness, and micro-structure of
appear in thematerials
the welding welds during forging from
are different [2]. In addition,
that the strength,
of the parent hardness,
metal [3], and micro-structure
and unaligned ends of welded of
the welding materials are different from that of the parent metal [3], and unaligned
rails, inequality of grinding, and welding parts by hand will cause irregularities on the surface of the ends of welded
rails, inequality
rail [4]. Gao andofZhai grinding, and welding
[5] carried parts by
out a detailed handon
survey will cause irregularities
geometric irregularitiesonattherailsurface
welds ofon
the
high-speed railways in China. They found that a short-wave irregularity with a wavelengthwelds
rail [4]. Gao and Zhai [5] carried out a detailed survey on geometric irregularities at rail range
on
fromhigh-speed
0.05 to 0.39 railways
m was in theChina. They found
main form. Initiation thatanda short-wave
development irregularity
of such anwith a wavelength
irregularity can be
range from 0.05 to 0.39 m was the main form. Initiation and development
from high-frequency excitation, which finally loses its energy by dissipation in the form of plastic of such an irregularity
can be from high-frequency
deformation and other processes excitation,
relatedwhich
to thefinally
growth loses its energy
of track damage,by asdissipation
reported inbytheWen form
[6],
of plastic[7],
Correa deformation
and Mandal andandother processes related
Dhanasekar to the growth
[8]. Therefore, of track
controlling thedamage, as reported
size of the short-waveby
Wen [6], Correa [7], and Mandal and Dhanasekar [8]. Therefore, controlling
irregularity of welds is of great importance to ensure secure, stable, and economic operation of the size of the short-wave
irregularity of welds Xiao
high-speed railways. is of [9]
great importance
investigated thetoinfluence
ensure secure,
of weldstable,
defectsand witheconomic
different operation
wavelengths of
high-speed railways.
on the wheel/rail Xiaoand
force [9] proposed
investigated the influence
safety limits underof weld defectsspeeds
different with different
based on wavelengths
the dynamics on
the
index or dynamic factors. Steenbergen and Esveld [10,11] evaluated the quality of rail welds usingora
wheel/rail force and proposed safety limits under different speeds based on the dynamics index
dynamic
geometric factors. Steenbergen
gradient, which can andbeEsveld
used [10,11] evaluated any
for considering the quality
geometricof rail welds
form using
of the a geometric
weld. Related
gradient, which can be used for considering any geometric form of the
standards have also been determined on the basis of numerical calculations and engineering weld. Related standards have
also been
experience. determined on the basis of numerical calculations and engineering experience.

(a) (b)

Figure 1. Rail joint (a) and rail weld (b).

Figure 1. Rail joint (a) and rail weld (b).

The precondition of carrying out the evaluation of wheel–rail force at rail welds stated in [9–11]
is toThe precondition
obtain its geometricof carrying
size, whichout the evaluation
is definitely of wheel–rail
time consumingforce at railintensive.
and labor welds stated Forin [9–11]
effective
ismeasurement
to obtain its geometric size, which
of irregularities in the is definitely time consuming
tracks, detecting and box
vertical axle laboracceleration
intensive. For effective
(hereinafter
measurement of irregularities in the tracks, detecting vertical axle
referred to as ABA) is a simple and economical method. An accelerometer sensor can bebox acceleration (hereinafter referred
easily
to
installed on the axle box (see Figure 2) without limitations of the vehicle’s running speed [12,13].on
as ABA) is a simple and economical method. An accelerometer sensor can be easily installed To
the axleon
reflect boxthe(see Figure
health 2) without
condition limitations
of track systems,of the vehicle’s
many running
scholars have, speed [12,13].
therefore, focusedTo reflect
their on
studies
the
on health condition
ABA. Liang [14] of track systems,
developed a set of many
indoorscholars
rollinghave, therefore,
test devices focused
that theirfor
was used studies on ABA.
simulating the
Liang [14] developed a set of indoor rolling test devices that was used
response of ABA under low-speed conditions, which ranged from 3.5 to 15 km/h, when there were for simulating the response of
ABA under low-speed conditions, which ranged from 3.5 to 15 km/h, when
defects on the wheel/rail surface. Obvious fluctuations in ABA were generated at the defects that there were defects on the
wheel/rail
were confirmedsurface. viaObvious fluctuations
time–frequency in ABAaccording
technology were generated at thefindings
to research defects that
to bewere
damageconfirmed
on the
via time–frequency technology according to research findings to be damage
wheels or rails. Molodova and Li [15] carried out a series of ABA tests targeted at monitoring on the wheels or rails.
the
Molodova and Li [15] carried out a series of ABA tests targeted at monitoring
health of some preliminary damage (e.g., rail squats, rail pad degradation, and fastening cracks) the health of some
preliminary
occurring atdamage (e.g.,on
weld joints railsite;
squats, rail padshowed
the results degradation, and fastening
an obvious difference cracks)
between occurring at weld
rail welds that
joints
were in good and poor condition. On this basis, Li [16] put forward three approaches to improveand
on site; the results showed an obvious difference between rail welds that were in good the
poor condition.
success On this basis,
rate of monitoring. Li [16] put
However, the forward
methodsthree approaches
are currently to improve
applied on linesthe success rate
of ordinary of
speed
monitoring.
(that is, 140 However,
km/h). Zhai’sthe methods
group [17] areconducted
currently applied
abundant ontests
lineson
of high-speed
ordinary speed (thatlines
railway is, 140
inkm/h).
China.
Zhai’s group [17] conducted abundant tests on high-speed railway
With obtaining the dynamic response of ABA up to 350 km/h, the results have shown that lines in China. With obtaining the
dynamic response of ABA up
high-frequency vibration exists in ABA.to 350 km/h, the results have shown that high-frequency vibration exists
in ABA.
Appl. Sci. 2017, 7, 1259 3 of 16
Appl. Sci. 2017, 7, 1259 3 of 16

Figure 2.
Figure 2. Installation
Installation of the acceleration
of the acceleration sensor
sensor on
on the
the axle
axle box.
box.

The above studies focused mainly on testing ABA. It is difficult to carry out tests under extreme
The above studies focused mainly on testing ABA. It is difficult to carry out tests under extreme
conditions due to many unknown factors in the measured work and due to safety considerations.
conditions due to many unknown factors in the measured work and due to safety considerations.
Consequently, it is vital to develop a numerical model for an in-depth investigation. In the past,
Consequently, it is vital to develop a numerical model for an in-depth investigation. In the past,
problems of dynamics were widely processed by adopting a rigid multi-body dynamics model [18].
problems of dynamics were widely processed by adopting a rigid multi-body dynamics model [18].
Results from the modal analysis introduced to the rigid model are regarded as a way to further
Results from the modal analysis introduced to the rigid model are regarded as a way to further consider
consider high-frequency dynamic effects of wheel/rail high-speed rolling contact. However, upon
high-frequency dynamic effects of wheel/rail high-speed rolling contact. However, upon considering
considering the vibration mode, the wheel/rail contact geometry relationship becomes complex and
the vibration mode, the wheel/rail contact geometry relationship becomes complex and difficult to
difficult to solve because of the bending and torsion of the wheel/rail systems. In recent years, a new
solve because of the bending and torsion of the wheel/rail systems. In recent years, a new method—the
method—the explicit finite element (FE) method—has been used to solve problems of
explicit finite element (FE) method—has been used to solve problems of high-frequency wheel/rail
high-frequency wheel/rail rolling contact [6,15,19–23]. This method abandons three basic
rolling contact [6,15,19–23]. This method abandons three basic assumptions including infinite half
assumptions including infinite half space, steady rolling, and linear elastic materials of the classic
space, steady rolling, and linear elastic materials of the classic contact algorithms such as Kalker’s
contact algorithms such as Kalker’s CONTACT [22]. Molodova and Li [19,20] developed a 3D
CONTACT [22]. Molodova and Li [19,20] developed a 3D explicit FE model based on this method to
explicit FE model based on this method to simulate the change in ABA at short-wave irregularities of
simulate the change in ABA at short-wave irregularities of rail at a speed of 140 km/h; their results
rail at a speed of 140 km/h; their results agreed with that of the test site. It should be noted that
agreed with that of the test site. It should be noted that Molodova and Li’s main purpose for developing
Molodova and Li’s main purpose for developing such a model was to monitor squat defects on the
such a model was to monitor squat defects on the rail surface and the degrees of damage. Therefore,
rail surface and the degrees of damage. Therefore, the relationship between ABA and wheel/rail
the relationship between ABA and wheel/rail force was not given and we cannot estimate degradation
force was not given and we cannot estimate degradation of the defect through ABA as the
of the defect through ABA as the evaluation index is always wheel/rail force or geometry size.
evaluation index is always wheel/rail force or geometry size.
The operating speeds of high-speed railway in China have reached 300 km/h, bringing higher
The operating speeds of high-speed railway in China have reached 300 km/h, bringing higher
requirements for safety. This paper aims to provide a new theoretical basis and analysis method
requirements for safety. This paper aims to provide a new theoretical basis and analysis method to
to monitor and control welds at at such high speeds. First, the S transform was used to analyze
monitor and control welds at at such high speeds. First, the S transform was used to analyze
vibration characteristics of ABA in the time–frequency domain at welds based on the measured data
vibration characteristics of ABA in the time–frequency domain at welds based on the measured data
of 300 km/h. Then, a validated 3D wheel/rail rolling contact FE model was employed to simulate
of 300 km/h. Then, a validated 3D wheel/rail rolling contact FE model was employed to simulate
dynamic responses of ABA induced by rail welds under 300 km/h through investigating the influence
dynamic responses of ABA induced by rail welds under 300 km/h through investigating the
of weld geometry on ABA. A quantitative relationship between ABA and wheel/rail force has also
influence of weld geometry on ABA. A quantitative relationship between ABA and wheel/rail force
been established. Based on these results, an approach to estimate rail weld is proposed. The differences
has also been established. Based on these results, an approach to estimate rail weld is proposed. The
in ABA induced by rail weld, rail corrugation, and wheel flat are then discussed.
differences in ABA induced by rail weld, rail corrugation, and wheel flat are then discussed.
2. Observation of Time–Frequency Vibration of the Measured Axle Box Acceleration
2. Observation of Time–Frequency Vibration of the Measured Axle Box Acceleration
High-frequency contact vibration generated on the wheel/rail surface under excitation of all kinds
High-frequency contact vibration generated on the wheel/rail surface under excitation of all
of irregularities can be easily transferred to the axle box due to large contact stiffness between wheel
kinds of irregularities can be easily transferred to the axle box due to large contact stiffness between
and rail. A trial measurement at a speed of 300 km/h on a certain Chinese high-speed railway was
wheel and rail. A trial measurement at a speed of 300 km/h on a certain Chinese high-speed railway
conducted by the State Key Laboratory of Traction Power of Southwest Jiaotong University. The track
was conducted by the State Key Laboratory of Traction Power of Southwest Jiaotong University. The
is non-ballast track containing rail, fastening, slab, and mortar and the car was CRH380. The ABA
track is non-ballast track containing rail, fastening, slab, and mortar and the car was CRH380. The
measurement (four accelerometer sensors) was installed on the axle box of a bogie (see Figure 3).
ABA measurement (four accelerometer sensors) was installed on the axle box of a bogie (see Figure
The data acquisition unit was connected with the sensor mounted on the tested components of the
3). The data acquisition unit was connected with the sensor mounted on the tested components of
vehicle. The sampling frequency of the axle box acceleration sensor was set to 16.4 kHz.
the vehicle. The sampling frequency of the axle box acceleration sensor was set to 16.4 kHz.
Appl. Sci. 2017, 7, 1259 4 of 16
Appl. Sci. 2017, 7, 1259 4 of 16

Appl. Sci. 2017, 7, 1259 4 of 16

Figure 3. A typical bogie and axle box acceleration (ABA) measurement on an axle box.
Figure 3. A typical bogie and axle box acceleration (ABA) measurement on an axle box.
FigureFigure
4 shows
3. Athe measured
typical ABA
bogie and axlechanging along with
box acceleration (ABA)the running time.
measurement on anNote that a 1500 Hz
axle box.
Figure
LPF (low 4 shows the has
pass filter) measured ABA changing
been imposed along
on the signals withwith the running
consideration time. Note
of typical that range
frequency a 1500 Hz
LPF (low Figure
of wheel/rail 4 shows
pass filter) hasthe
interaction measured
been and ABA
noises
imposed changing
oncontaining along
the signals with the running
high-frequency
with consideration time.
signals.
of Note
The
typicalABAthat amplitude
a 1500 range
frequency Hz of
LPF
wheel/rail (low
fluctuated pass filter)
within
interaction the has
and been
range of imposed
noises ±50 on the
m/s2 under
containing signals
300 km/h with
high-frequency consideration
on account
signals. of ABA
typical
of geometry
The frequencyon
irregularities
amplitude range
the
fluctuated
of
within wheel/rail
surface
the range ofinteraction
and elastic m/s2and
±50deformationundernoises
of 300 containing
the wheel/rail high-frequency
contact
km/h on account interface.
of geometry signals. The ABA amplitude
Rails irregularities
of non-ballast tracks
on theare
surface
fluctuated
welded every within
100the m.range of ±50 m/s
Short-wave 2 under 300occurred
irregularities km/h on easily
account in of geometry
the welds as irregularities
mentioned on in the
and elastic deformation of the wheel/rail contact interface. Rails of non-ballast tracks are welded
surface and elastic
introduction, and a deformation
pulse impulsionof the wheel/rail
occurs contact
every 1.2 s (it isinterface.
about 100Railsm asofmultiplied
non-ballastby tracks are
the speed
every 100 m. Short-wave irregularities occurred easily in the welds as mentioned in the introduction,
welded everyin100
of 300 km/h) m. 4,
Figure Short-wave irregularities
whose amplitude reachesoccurred
150 to 200 easily
m/s2,inasthe welds
shown as mentioned
in RW1. By contrast, in the
and aintroduction,
pulse impulsion
amplitude at RW2 and aisoccurs
pulse every
ratherimpulsion
1.2occurs
s (it isbecause
gentle, possibly
about1.2
every 100 mwere
s (it
there isasabout
multiplied
100 m as
no serious
by the speedby
multiplied
irregularities.
ofthe
300speed
km/h) in
Figure 4, whose amplitude reaches 150 to 200 m/s 2 , as shown in RW1. By contrast, the amplitude at
of 300 km/h) in Figure 4, whose amplitude reaches 150 to 200 m/s , as shown in RW1. By contrast, the
2

RW2 amplitude
is rather gentle, 300possibly because there were no serious irregularities.
at RW2 is rather gentle, possibly because there were no serious irregularities.
RW1
250
300
200
RW1
250
100m
150 RW2
) (m/s )
2

200
100
100m
ABA

150 RW2
2

50
ABA (m/s

100
0
50
-50
0
-100
-50 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

-100 Time (s)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Figure 4. Time-course response of measured ABA under 300 km/h. RW1 and RW2 are two examples
Time (s)
of rail welds.
Figure 4. Time-course response of measured ABA under 300 km/h. RW1 and RW2 are two examples
Figure 4. Time-course response of measured ABA under 300 km/h. RW1 and RW2 are two examples
Torail
of obtain
welds.a more detailed analysis and display the vibration of the ABA signal in the
of rail welds.
time–frequency domain (i.e., the changes of frequency components of acceleration at different times
To the
during obtain a more
process detailed
of a vehicle analysis
passing the and display
weld), the vibration
the S transform of the
method wasABA
used.signal in the
This method
time–frequency
wasobtain
To selected domain
a because
more (i.e., the
it inherits
detailed changes
and
analysis of frequency
develops
and localizedthe
display components
ideas of acceleration
of continuous
vibration ABAatsignal
of thewavelet different
in times
transform andtime–
the
during
frequency the Fourier
short-time
domain process of
thea vehicle
(i.e.,transform,
changes andpassing
ofadopts the weld),
a time
frequency the S transform
window
components of Gaussian method was
function
of acceleration used.
that
at Thistimes
presents
different method
a goodduring
was selected
resolution in because
the time it inherits
and and
frequency develops
domains localized
[24]. The ideas
S of continuous
transform can
the process of a vehicle passing the weld), the S transform method was used. This method was selected be wavelet
describedtransform
by and
Equations
short-time
(1) and (2): Fourier transform, and adopts a time window of Gaussian function that presents a good
because it inherits and develops localized ideas of continuous wavelet transform and short-time
resolution in the time and frequency domains ∞ [24]. The S transform can be described by Equations
Fourier transform, and adopts a time
(1) and (2): ) =  x(t )of
S (τ , fwindow ω (τGaussian
− t , f ) exp(function
− j2π ft ) d that
t presents a good resolution
(1) in
−∞
the time and frequency domains [24]. The S transform can be described by Equations (1) and (2):
∞
S (τ , f Z) =  x(t )ω (fτ − t , f ) exp(
f 2 (−τ j2π 2ft ) d t
− t) (1)
ω (τ −∞t , f ) =
−∞
exp(− ) (2)
S(τ, f ) = x (t)ω2(πτ − t, f ) exp 2 (−j2π f t)dt (1)
−∞
f f (τ − t)
2 2
ω (τ − t , f ) = exp(− ) (2)
| f2|π f 22(τ − t)2
ω (τ − t, f ) = √ exp(− ) (2)
2π 2
Appl. Sci. 2017, 7, 1259 5 of 16
Appl. Sci. 2017, 7, 1259 5 of 16

where f is frequency; ω (τ − t , f ) is the Gaussian window function; τ is a parameter controlling the

Gaussian
where f iswindow
frequency; (τ − t, f of
in aωposition ) istime
the t.Gaussian window function; τ is a parameter controlling the
Time–frequency
Gaussian window in avibration
position characteristics
of time t. of the ABA at rail welds are presented in Figure 5 by
taking RW1 in Figurevibration
Time–frequency 4 as an characteristics
example. Figure 5a shows
of the ABA atthe railtime-domain variationinof
welds are presented the ABA
Figure 5 by
between
taking RW1 11.1inand
Figure 11.2 s. an
4 as Asexample.
we can see, Figurean 5a
obvious
shows impact was generated
the time-domain with
variation of athe
high-frequency
ABA between
vibration when
11.1 and 11.2 s. Aspassing
we canRW1.see, anThe time–frequency
obvious dynamic response
impact was generated through the vibration
with a high-frequency S transformwhen is
presented
passing RW1. in Figure 5b; the contour
The time–frequency has been
dynamic response through Sthe
normalized (τ , Sf transform
) and the ismean vibration
presented energy
in Figure 5b;
the contour has been normalized S ( τ, f )
changed from small to large in blue and red. By this figure, we know that the vibration of ABA atin
and the mean vibration energy changed from small to large a
blue and
speed red. km/h
of 300 By this figure,a we
covers wide know that thedomain.
frequency vibration of ABA
Energy at a speed
excited of 300
by weld km/h covers
geometry a wide
irregularity
frequency
around domain.
11.14 s gathers Energy
mainly excited
in twobyfrequency
weld geometrybands: irregularity
from 350 to 500 around
Hz and11.14 s gathers
from 1000 tomainly
1200 Hz.in
two frequency bands: from 350 to 500 Hz and from 1000 to 1200 Hz. High-frequency
High-frequency vibration lasting about 2 to 5 ms is dissipated by the wheel/rail contact interface, vibration lasting
about 2isto
which 5 ms iscause
a major dissipated by the
of fatigue wheel/rail
generation for contact
all kinds interface,
of rolling which is a major
contacts. cause the
In addition, of fatigue
welds
generation
excite for allatkinds
vibration of rolling
a frequency contacts.
band from 50 In to
addition,
130 Hz,the welds
which excitethe
reflects vibration at aoffrequency
vibration the bogieband
and
from 50 to 130 Hz, which reflects the vibration of the bogie and
sleeper (the main components of the vehicle–track system). With the long duration, such sleeper (the main components of the
vehicle–track system).
low-frequency vibration With
notthe
onlylong duration,
transfers to thesuch low-frequency
vehicle but also tovibration not onlyof
the substructure transfers
the tracktoand
the
vehicle but also to the substructure of the track and is an important cause
is an important cause of performance deterioration of infrastructures such as ballast, slabs, and of performance deterioration
of infrastructures
bridges, etc. such as ballast, slabs, and bridges, etc.

250

200 RW1

150
ABA (m/s )
2

100

50

-50

-100
11.1 11.11 11.12 11.13 11.14 11.15 11.16 11.17 11.18 11.19 11.2
Time (s)
(a)
2000

RW1
1500
Frequency (Hz)

1000

500

0
11.1 11.11 11.12 11.13 11.14 11.15 11.16 11.17 11.18 11.19 11.2
Time (s)
(b)
Figure 5. The response of ABA in (a) time and (b) time–frequency domain when a vehicle passes the
Figure 5. The response of ABA in (a) time and (b) time–frequency domain when a vehicle passes the
rail
rail welds at
welds at aa speed
speed of
of 300
300 km/h
km/hon
onaacertain
certainhigh-speed
high-speedline.
line.

As shown in Figure 4, the signal contains 17 groups of acceleration responses of the axle box at
As shown in Figure 4, the signal contains 17 groups of acceleration responses of the axle box at
rail welds. CDF (cumulative distribution function) has been used to create statistics on the amplitude
rail welds. CDF (cumulative distribution function) has been used to create statistics on the amplitude
of axle box acceleration at these rail welds, as shown in Figure 6. It is obvious that the 17 groups of
of axle box acceleration at these rail welds, as shown in Figure 6. It is obvious that the 17 groups of axle
axle box accelerations change between 66 and 2002 m/s2 with nearly 70.6% of the ABAs less than 1482
box accelerations change between 66 and 200 m/s with nearly 70.6% of the ABAs less than 148 m/s .
m/s2.
Appl. Sci. 2017, 7, 1259 6 of 16
Appl. Sci. 2017, 7, 1259 6 of 16

1.0

Appl. Sci. 2017, 7, 1259 6 of 16

0.8
70.6%
1.0

0.6

CDF
0.8
70.6%
0.4
0.6

CDF 0.2
148m/s
2

0.4

0.0
0.2 60 80 100 120 140 160 2180 200 220
148m/s 2
ABA induced by rail weld (m/s )
Figure 6. Cumulative 0.0
distribution function (CDF) of 17 measured ABAs induced by rail welds.
Figure 6. Cumulative distribution
60 80function
100 (CDF)
120 of 17160
140 measured
180 ABAs220
200 induced by rail welds.
2
Rail welds have been maintained mainly ABA induced by rail the
by limiting weldgeometry
(m/s ) to reduce the wheel/rail force
Rail
[25]. welds
However,have been
general maintained
testing of weld mainly
geometryby limiting
is labor the geometry
intensive, demanding
Figure 6. Cumulative distribution function (CDF) of 17 measured ABAs induced to by
reduce
on the wheel/rail
railresources,
welds. and
time consuming. To the authors’ best knowledge, there are currently no
force [25]. However, general testing of weld geometry is labor intensive, demanding on resources,efficient techniques to
measure
and timeRail wheel/rail
welds haveTo
consuming. force,
been especially
themaintained for
authors’ mainly high-speed situations.
by limiting the
best knowledge, Therefore,
geometry
there clearly
to reduceno
are currently mapping
theefficient the
wheel/railtechniques
force
relationship
[25].
to measure However,between
wheel/railgeneralABA
force, data
testing ofand geometric
weld
especially geometry
for sizeis of
high-speed thesituations.
labor weld, as well
intensive, as the wheel/rail
demanding
Therefore, on force,
resources,
clearly can the
and
mapping
help railway
time consuming.departments
To the assess
authors’ weld geometry
best knowledge, from numerous
there are as data of
currently ABA, which will improve
relationship between ABA data and geometric size of the weld, well as no
theefficient
wheel/rail techniques to help
force, can
the efficiency
measure of maintaining
wheel/rail the railway
force, especially forsystem.
high-speed situations. Therefore, clearly mapping the
railway departments assess weld geometry from numerous data of ABA, which will improve the
relationship between ABA data and geometric size of the weld, as well as the wheel/rail force, can
efficiency of maintaining
3. 3Drailway
High-Speed the railway
Wheel/Rail system.
Rolling Contact Finite
help departments assess weld geometry from Element
numerous Model
data of ABA, which will improve
the efficiency of maintaining the railway system.
3. 3D3.1.
High-Speed
Overview of Wheel/Rail Rolling
the Finite Element ModelContact Finite Element Model
3. 3DFigure
High-Speed Wheel/Rail
7 shows Rolling Contact
a 3D high-speed Finite Element Model
3.1. Overview of the Finite Element Model wheel/rail rolling contact finite element (FE) model and
simulates a vehicle–track system on the high-speed railway line. Only half of a wheelset and track
3.1. Overview
Figure of the a
7 shows Finite Element Model wheel/rail rolling contact finite element (FE) model and
are simulated since 3Dthe high-speed
relative longitudinal and vertical planes of a vehicle–track system are
simulates Figure 7 shows a 3D high-speed wheel/rail railway
a vehicle–track
symmetric. All componentssystem on
above the
the high-speed
primary suspension
rolling of line. Only
a vehicle
contact half
are
finite of a wheelset
simplified
element as mass
(FE) and
model track are
points
and
and
simulated connected
since
simulates with wheels through
the relative longitudinal
a vehicle–track the
system on theand primary suspension.
verticalrailway
high-speed planes line. The high-frequency
of a vehicle–track feature
system are
Only half of a wheelset between
andsymmetric.
track
the wheelset
are simulated
All components and
above track
sincethethewas fully suspension
relative
primary considered
longitudinalthrough
of and 3D modeling.
vertical
a vehicle areplanes It of
canabe
simplified approximately
vehicle–track
as mass pointssystemassumed
and are
connected
that the acceleration
with symmetric.
wheels of primary
All components
through the the outer end of the wheel
abovesuspension.
the primary Theaxle
suspension is of
consistent
a vehiclewith
high-frequency are that ofbetween
theasaxle
simplified
feature massbox.
the The
points
wheelset
trackconnected
and is a non-ballast
with track, which
wheels through wasthemodeled
primary with consideration
suspension. of the rails, fasteners, slab, and
and track was fully considered through 3D modeling. It canThe behigh-frequency
approximately feature
assumedbetween that the
cement
the asphalt
wheelset and(CA)
trackmortar.
was fullyTheconsidered
fastening system
throughcontains support
3D modeling. stiffness
It can and dampingassumed
be approximately that are
acceleration of the outer end of the wheel axle is consistent with that of the axle box. The track is
relevant
that to the dynamic
the acceleration of thebehavior.
outer endThe track
of the length
wheel axleofis15.6 m waswith
consistent considered in axle
that of the the model
box. The to
a non-ballast
eliminate
track, whichofwas
the influence
modeled
boundary
with consideration of the rails, fasteners, slab, and cement
conditions.
track is a non-ballast track, which was modeled with consideration of the rails, fasteners, slab, and
asphalt (CA)asphalt
cement mortar. Themortar.
(CA) fastening system contains
The fastening supportsupport
system contains stiffness and damping
stiffness and dampingthat that
are relevant
are
to therelevant
dynamic to the dynamic behavior. The track length of 15.6 m was considered in the model to the
behavior. The track length of 15.6 m was considered in the model to eliminate
influence of boundary
eliminate the influenceconditions.
of boundary conditions.

(a) the schematic diagram (b) the mesh

Figure 7. 3D wheel/rail rolling contact fine element model.

(a) the schematic diagram (b) the mesh

Figure 7. 3D wheel/rail rolling contact fine element model.
Figure 7. 3D wheel/rail rolling contact fine element model.
Appl. Sci. 2017, 7, 1259 7 of 16

The 3D wheelset and rail in real geometry were meshed by an 8-node hexahedral element. The rail
was the CHN60 model with the rail cant set as 1:40. “Face-to-face” contact algorithm simulation based
on the penalty function method was used for solving contact behaviors among rails. A non-uniform
grid was used when meshing the wheel and rail in order to reduce the calculation time without
losing precision in the solution. The finest contact zone (element size was 1 mm) was used for solving
wheel/rail contact; the farther the contact zone, the larger the grid. The track slab and CA mortar were
also stimulated by the 8-node hexahedral elements. The primary suspension and rail fastener of the
vehicle were represented by a distributed spring and damping unit. The total number of nodes and
elements were 1.46 and 1.29 M, respectively.
The parameters of the model are listed in Table 1. Taking into account the elastic shakedown
response of the wheel and rail’s material in the field, a linear elastic material model was used for the
wheel and the rail, though more accurate material models can be easily introduced. For an effective
computation, the slab and mortar materials were also assumed to be linear elastic.

Table 1. Values of the parameters involved in this work.

Parameters Values
Lumped sprung mass 7500 kg
Stiffness of primary suspension 880 kN/m
Damping of primary suspension 4 kNs/m
Stiffness of fastenings 22 MN/m
Damping of fastenings 200 kNs/m
Young’s modulus 205.9 GPa
Wheel and rail material Poisson’s ratio 0.3
Density 7790 kg/m3
Young’s modulus 34.5 GPa
Material of pre-fabricated slabs Poisson’s ratio 0.25
Density 2400 kg/m3
Young’s modulus 8 GPa
Mortar material Poisson’s ratio 0.2
Density 1600 kg/m3

3.2. Rail Weld in 3D

With a smooth wheel/rail surface upon mesh generation, the form of 3D geometry irregularity
at rail welds in the calculated area was applied by modifying the coordinates of related nodes on
the rail surface by a self-compiled program. For easy understanding, a 3D right-handed Cartesian
coordinate system is defined, of which the origin O is located at the initial position of the contact patch
(i.e., position A as shown in Figure 7a). The x-axis is defined along the rolling direction and the y-
and z-axes are in the lateral and vertical directions. The rail weld model used in this paper for the
calculations is shown in Figure 8. Depth (or thickness) D at any element of the simulated weld is cosine
in the direction of x and expressed in Equation (3):

D = [1 − cos2π · ( x − x0 )/λ] · δ/2 (3)

For any x within the welding zone, D is parabolic in the y direction,

D = [1 − (y/w)2 ] · δ (4)

where λ and δ represent wavelength and maximum depth of a rail weld, respectively; x0 is the starting
position of a rail weld applied; w is one-half width of the weld and assumed as 15 mm in this paper
according to field observations.
Appl. Sci. 2017, 7, 1259 8 of 16
Appl. Sci. 2017, 7, 1259 8 of 16

Appl.(a)
Sci. 2017, 7, 1259 λ (b) 8 of 16
v
δ
z
(a) x0 λ (b)
v
xδ
z

z
x0
2w

z
z

2w

y x
z

Figure 8. Simulated rail weld geometry in (a) 2D and (b) 3D.

y and (b) 3D. x
Figure 8.y Simulated rail weld geometry in (a) 2D
3.3. Simulation Process
3.3. Simulation Process Figure 8. Simulated rail weld geometry in (a) 2D and (b) 3D.
The explicit integral has been widely applied in the solution of the high-frequency phenomenon
due
The to its small
explicit
3.3. Simulation time has
integral
Process step been
(the integration
widely applied time step
in the forsolution
the model in this
of the high-frequency × 10 −8 s).
paper was 8.9phenomenon
due to Implicit–explicit
itsThe
small timeanalysis
explicit step (the
integral
wasintegration
adopted in the
has been time
widely applied
simulation
step inthe
order
forsolution
in the
to smooth
model
of thein thisthepaper
dynamic
high-frequency was 8.9 × 10−8 s).
interaction
phenomenon
of the wheel
Implicit–explicit and rail
analysis by reducing the excitation of the initial energy in the rolling process. Figure 9 of
due to its small time was
step adopted in the simulation
(the integration time step for in the
order to smooth
model in this the
paperdynamic
was 8.9interaction
× 10 −8 s).
shows a schematic diagram of the simulation process; positions A and B are the same as that of
the wheel and rail byanalysis
Implicit–explicit reducing wastheadopted
excitation of simulation
in the the initial inenergy
orderinto the rolling
smooth the process.
dynamic Figure 9 shows
interaction
Figure 7a. It firstly adopts an implicit algorithm to solve the static contact in the initial position of the
a schematic diagram
of the wheel of the
and rail simulation
by reducing the process;
excitationpositions
of the initialA and
energyB are therolling
in the same process.
as that of Figure
Figure 9 7a.
wheel, and then it initializes the explicit rolling contact calculation with the vehicle rolling forwards
shows
It firstly a schematic
adopts an diagram
implicit of the simulation
algorithm to solve process;
the static positions
contact A and
in the B are the
initial same as
position of that
the of
wheel,
at a constant angular w and forward speed v and driven by a constant traction coefficient μ . The
Figure
and then it 7a. It firstly adopts
initializes an implicit
the explicit rollingalgorithm
contacttocalculation
solve the static
withcontact in the initial
the vehicle position
rolling of the at a
forwards
traction
wheel, and coefficient defined by
then it initializes theEquation (5) is 0.15
explicit rolling for calculation
contact a typical traction
with thescenario and theforwards
vehicle rolling friction
constant angular wisand forward speed v and driven by situation.
a constant traction
that nocoefficient µ. The in traction
at a constantf angular
coefficient 0.5 for a typical dry wheel–rail contact Note force is applied
w and forward speed v and driven by a constant traction coefficient μ . The
the
coefficient
lateral defined
direction. by Equation (5) is 0.15 for a typical traction scenario and the friction coefficient f is
traction coefficient defined by Equation (5) is 0.15 for a typical traction scenario and the friction
0.5 for a typical dry wheel–rail contact situation. Note that no force is applied in the lateral direction.
μ = Fcontact
coefficient f is 0.5 for a typical dry wheel–rail L / FN ≤ f
situation. Note that no force is applied in(5)
the
lateral direction. ≤ f transmitted in the longitudinal direction.
where FN is the normal load and FL is µ the=traction
FL /FN force (5)
μ =
Zone AB between the initial position A andF / FN ≤ f
L position (5)
B is designed to ensure that the wheel
FN is the normal
whereachieves load and FLbeforeis the traction the force transmitted in theB).
longitudinal direction.
where FNsteady
is therolling
normalcontact
load and FL isenteringthe traction welding zone (after
force transmitted Boundary conditions
in the longitudinal direction.
Zone
imposed by the model include constraint of lateral movement on the front of the wheel axle, achieves
AB between the initial position A and position B is designed to ensure that the wheel the
Zone AB between the initial position A and position B is designed to ensure that the wheel
steady rolling contact
longitudinal front before entering
end of rails the by
imposed welding zone (after
the longitudinal B). Boundary
restraint, the fixedconditions
bottom of imposed
CA mortar,by the
achieves steady rolling contact before entering the welding zone (after B). Boundary conditions
modeland include
primaryconstraint of lateral
suspension movement
and fastening on theonly
reserving front
theofvertical
the wheel
degree axle,
of the longitudinal
freedom. front end
More details
imposed by the model include constraint of lateral movement on the front of the wheel axle, the
about
of rails the model
imposed by arelongitudinal
the referred to in restraint,
a previousthe study [22].
fixed bottom of CAthe mortar, and primary suspension
longitudinal front end of rails imposed by the longitudinal restraint, fixed bottom of CA mortar,
and fastening
and primaryreserving only and
suspension the vertical
fasteningdegree
reservingof freedom. More details
only the vertical degreeabout the model
of freedom. Moreare referred
details
to in aabout
previous study
the model are[22].
referred to in a previous
Implicit to explicit study [22].

Implicit to explicit Wheel Welding zone

ω
Wheel Welding zone
A
Rail
ωB
Implicit solution at Explicit solution during Explicit solution after
Rail A B
position A: zone AB: position B (welding zone):

Implicit
Obtain solution
initial at
stress field Explicit
Relax solution
initial during
turbulence to Explicitwheel-rail
Acquire solution after
force

position
under static A:
load. get a zone AB:
steady rolling position
and axle Bbox
(welding zone):
acceleration.

Obtain initial stress field Relax initial turbulence to Acquire wheel-rail force
Figure 9. Schematic diagram of the simulation process.
under static load. get a steady rolling and axle box acceleration.

Figure
Figure 9. 9. Schematicdiagram
Schematic diagram of
of the
thesimulation
simulationprocess.
process.
Appl. Sci. 2017, 7, 1259 9 of 16
Appl. Sci. 2017, 7, 1259 9 of 16

4. Results and Discussion

4.1. The Influence of Rail Weld Geometry on ABA

Time-course changes of ABA excited by rail welds with different wavelengths on a high-speed
railway line (300 km/h)
km/h) are are presented
presented in in Figure
Figure 10a.
10a. The
The data
data has
has been
been processed
processed by a 1500 Hz LPF
and conforms to the measured data shown in Figure 4. Considering the typical size of the measured
weld geometry
geometry [5][5]and
andthe thefeatures
featuresofofshort wavelengths
short wavelengths concerned
concerned in the paper,
in the the wave
paper, the wavedepth was
depth
assumed
was assumedas 0.2as
mm 0.2and
mmthreeandwavelengths including
three wavelengths 0.05 m, 0.1
including 0.05m,m,and
0.10.2
m,mandwere0.2selected.
m wereTaking as
selected.
an example
Taking the
as an workingthe
example condition
workingatcondition
a wavelength at a of 0.05 m, theof
wavelength acceleration
0.05 m, theofacceleration
the axle boxof fluctuated
the axle
within 2 as a result of elastic
±20 m/swithin deformation between the wheel and rail.
box fluctuated ±20 m/s2 as a result of elastic deformation between theThis value
wheel andis rail.
less than
This
the ± 2 , see Figure 4) due
value is less than the measured data (±50 m/s , see Figure 4) due to the geometry irregularities of not
measured data ( 50 m/s 2 to the geometry irregularities of the actual line the
being
actual considered.
line not beingWhen passingWhen
considered. the rail weld,the
passing it excites strong
rail weld, vibrations
it excites strongof ABA. The
vibrations of first
ABA. wave
The
crest ABAFcrest
first wave was ABA
reachedF wasat 204.6
reachedm/sat2 204.6
and wasm/scaused
2 and was by the impact
caused of the
by the rail weld.
impact of theTherailcontinuous
weld. The
fluctuation depends on coupled vibration of the vehicle–track system with
continuous fluctuation depends on coupled vibration of the vehicle–track system with a vibrationa vibration wavelength of
72 mm. It canofbe72observed
wavelength mm. It can thatbeABA V is significantly
observed that ABAVhigher than the first
is significantly wave
higher crest
than excited
the by thecrest
first wave rail
weld.
excitedTheby phenomena
the rail weld.can Thebephenomena
found in thecan measured
be found data of RW1
in the in Figure
measured data5a.
of RW1 in Figure 5a.

400 50
(a) ABAV 0.05m (b)
72mm 0.1m f 2= 1150Hz
300 ABAF 0.05m
0.2m
0.1m
40
200 0.2m
Amplitude (m/s )
2

100
ABA(m/s )

30
2

-100 20
f 1= 350Hz
-200
10
238mm
-300

-400 0
28 29 30 31 32 33 34 35 0 250 500 750 1000 1250 1500
Time (ms) Frequency (Hz)

Figure 10.
Figure 10. The
The response
responseofofABA
ABAunder
underexcitation
excitation
of of
railrail welds
welds at different
at different wavelengths
wavelengths intime
in (a) (a) time
and
and (b) frequency domains.
(b) frequency domains.

The ABAF decreased from 204.6 m/s2 to 102.4 m/s2 when the wavelength increased from 0.05 to
The ABAF decreased from 204.6 m/s2 to 102.4 m/s2 when the wavelength increased from 0.05
0.2 m. When the wavelength was 0.2 m, the resonance wavelength (238 mm) caused by the
to 0.2 m. When the wavelength was 0.2 m, the resonance wavelength (238 mm) caused by the
vehicle–track system was significantly greater than that of 0.05 m and 0.1 m. This is because rail
vehicle–track system was significantly greater than that of 0.05 m and 0.1 m. This is because rail welds
welds with long wavelengths do not easily excite high-frequency vibrations in vehicle–track
with long wavelengths do not easily excite high-frequency vibrations in vehicle–track systems, as
systems, as shown in Figure 10b. This figure shows the distribution of three signals in the frequency
shown in Figure 10b. This figure shows the distribution of three signals in the frequency domain.
domain. There were two obvious dominant frequencies: 350 Hz and 1150 Hz marked as f1 and f2. The
There were two obvious dominant frequencies: 350 Hz and 1150 Hz marked as f 1 and f 2 . The two
two frequencies corresponding to the wavelengths of 238 mm and 72 mm at a running speed of 300
frequencies corresponding to the wavelengths of 238 mm and 72 mm at a running speed of 300 km/h
km/h agree with the vibration wavelength shown in Figure 10a.
agree with the vibration wavelength shown in Figure 10a.
The variations of ABA under more wavelengths are presented in Figure 11. Figure 11a indicates
The variations of ABA under more wavelengths are presented in Figure 11. Figure 11a indicates
that ABAF decreased accordingly from 204.6 m/s2 to2 24.7 m/s2 with increasing wavelengths from 0.05
that ABAF decreased accordingly from 204.6 m/s to 24.7 m/s2 with increasing wavelengths from
m to 0.5 m. In Figure 11b, a corresponding amplitude of ABA at frequency f1 first increased and then
0.05 m to 0.5 m. In Figure 11b, a corresponding amplitude of ABA at frequency f 1 first increased and
decreased along with an increment of the wavelength, reaching a peak at 0.2 m. By contrast, the
then decreased along with an increment of the wavelength, reaching a peak at 0.2 m. By contrast,
amplitude at f2 decreased continuously. When the wavelength reached 0.3 m, no dynamic response
the amplitude at f 2 decreased continuously. When the wavelength reached 0.3 m, no dynamic response
at f2 was excited. Note that f2 corresponding to vibration wavelengths of 72 mm and 65~80 mm are
at f 2 was excited. Note that f 2 corresponding to vibration wavelengths of 72 mm and 65~80 mm are
typical wavelengths of high-speed rail corrugations. Therefore, keeping the wavelength of a rail
typical wavelengths of high-speed rail corrugations. Therefore, keeping the wavelength of a rail weld
weld irregularity to no less than 0.3 m can help prevent the occurrence of short-wave rail
irregularity to no less than 0.3 m can help prevent the occurrence of short-wave rail corrugations,
corrugations, which might also explain why rail corrugations only appeared in certain sections of
which might also explain why rail corrugations only appeared in certain sections of lines.
lines.
Appl. Sci. 2017, 7, 1259 10 of 16
Appl. Sci. 2017, 7, 1259 10 of 16
Appl. Sci. 2017, 7, 1259 10 of 16
220 50
(a) 220 (b) 50
(a) 200 (b)
200
180 40
180 40
160

))
160
f1

2
(m/s
ff1
) )

140 30
2

2
(m/s

(m/s
140 30 2
2

120 f2
(m/s

Amplitude
120
F F

20
ABA

100

Amplitude
20
ABA

100
80
80
60 10
60 10
40
40
20 0
20 0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
0.2 0.0
0.3 0.1
0.0 0.1 0.2 0.3(m)
Wavelength 0.4 0.5 Wavelength (m) 0.4 0.5
Wavelength (m) Wavelength (m)
Figure 11. The influence of wavelength on (a) ABA and (b) variation of amplitude at f1 and f2.
Figure 11. The
Figure11. The influence
influence of
ofwavelength
wavelengthon
on(a)
(a)ABA
ABAand
and(b)
(b)variation
variationof
ofamplitude
amplitudeat atff11 and
and ff22..
Wave depth is another important parameter of rail weld irregularities, and was also calculated
Wave depth
depthisisanother
anotherimportant
important parameter of of
railrail weld irregularities,
andand
waswas
alsoasalso calculated
here Wave
to know its influence on ABA; parameter
a typical weld irregularities,
wavelength was considered calculated
0.1 m. hereThe
here
to knowto its
know its influence
influence on ABA; aon ABA;wavelength
typical a typical was wavelength
considered was
as considered
0.1 m. The as 0.1 m. The
frequency-domain
frequency-domain responses of ABA under wave depths of 0.1 and 0.2 mm are presented in Figure
frequency-domain
responses responses ofdepths
ABA under wave depths of 0.1 and 0.2 in
mm are presented
It canin Figure
12a. It canofbeABAseenunder
that wave
the wave of 0.1affects
depth and 0.2onlymmtheare presented
amplitude of Figure
ABA but12a.with be seen
identical
12a.
that It can
the wave bedepth
seen affects
that the
onlywave
the depth affects
amplitude of only but
ABA thewith
amplitude
identicaloffrequency
ABA but under
with identical
different
frequency under different wave depths. Thereby, ABAF in Figure 12b grows almost linearly when
frequency
wave under
depths. different
Thereby, ABA wave depths.
F in Figure 12b Thereby,
grows ABAF in Figure 12b grows almost linearly when
the wave depth increased from 54.3 m/s 2 to 272.1 almost
m/s2. linearly when the wave depth increased from
the
54.3wave 2
m/s depth 2
increased
to 272.1 m/s . from 54.3 m/s to 272.1 m/s2.
2

22 280
(a) 22 (b) 280
(a) 20 0.1mm (b)
20 240
18 0.1mm
0.2mm
240
18 0.2mm
16 200
) )

16
2

14 200
(m/s

) )
2
2

(m/s

14 160
(m/s

12
2
(m/s

160
Amplitude

12
10
F F
Amplitude

120
ABA

10
8 120
ABA

8 80
6
6 80
4
40
4
2 40
2 0
0
0 0 0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
300 600 900 1200 1500
0 300 600 900 1200 1500 0.00 0.05 0.10 0.15depth
Wave 0.20
(m) 0.25 0.30 0.35
Frequency (Hz)
Frequency (Hz) Wave depth (m)
Figure 12. The influence of wave depth on ABA in (a) frequency and (b) time domains.
Figure
Figure 12.
12. The
The influence
influence of
of wave
wave depth
depth on
on ABA
ABA in
in (a)
(a) frequency and (b)
frequency and (b) time
time domains.
domains.
4.2. The Relationship between ABA and Wheel/Rail Force
4.2. The Relationship between ABA and Wheel/Rail Force
4.2. The Relationship between ABA and Wheel/Rail Force
Time-course changes of variation of dynamic wheel/rail force Fdyn (the difference between the
Time-course changes of variation of dynamic wheel/rail force Fdyn (the difference between the
wheel/rail force and
Time-course the static
changes wheel load)
of variation of the wheel/rail
of dynamic three rail welds
force Fin dynFigure 10a are presented
(the difference between thein
wheel/rail force and the static wheel load) of the three rail welds in Figure 10a are presented in
Figure 13. When
wheel/rail the wheel
force and passed
the static wheelover the of
load) railthe
welds,
threeFrail
dyn initially
welds inincreased
Figure 10a to aare
peak and then
presented in
Figure 13. When the wheel passed over the rail welds, Fdyn initially increased to a peak and then
decayed
Figure 13. gradually
When the to complete
wheel passedthe impact;
over the therail
amplitude
welds, Fof dynF fluctuation
initially
dyn was
increased small
to a once
peak leaving
and then
decayed gradually to complete the impact; the amplitude of Fdyn fluctuation was small once leaving
the rail weld.
decayed By comparing
gradually to completeto Figure 10a, itthe
the impact; is easy to discover
amplitude of Fdynthat the rail weld
fluctuation wascan excite
small onceonly one
leaving
the rail weld. By comparing to Figure 10a, it is easy to discover that the rail weld can excite only one
impact
the rail of the wheel/rail
weld. By comparing force,tobut it can10a,
Figure induce continuous
it is easy fluctuations
to discover that the of ABA.
rail weldThecanphenomena
excite only can
one
impact of the wheel/rail force, but it can induce continuous fluctuations of ABA. The phenomena can
be explained
impact of theby the wheel/rail
wheel/rail force,force
but itthat
canisinduce
determined mainlyfluctuations
continuous by local geometry
of ABA. irregularity.
The phenomenaThus,
be explained by the wheel/rail force that is determined mainly by local geometry irregularity. Thus,
no
canobvious impact
be explained by action will appear
the wheel/rail forceupon
that ispassing through
determined mainly the by
weld.
localMoreover, continuous
geometry irregularity.
no obvious impact action will appear upon passing through the weld. Moreover, continuous
fluctuations
Thus, no obvious arise impact
since ABAactionbelongs to theupon
will appear structural
passingvibration
through the withweld.
energy that is continuous
Moreover, difficult to
fluctuations arise since ABA belongs to the structural vibration with energy that is difficult to
dissipate
fluctuationsquickly by primary
arise since suspension.
ABA belongs to theThe maximum
structural of Fdyn
vibration withwillenergy
decrease
thatin
is turn along
difficult with the
to dissipate
dissipate quickly by primary suspension. The maximum of Fdyn will decrease in turn along with the
wavelength increase,
quickly by primary consistent The
suspension. withmaximum
the change of trend of ABA
Fdyn will F.
decrease in turn along with the wavelength
wavelength increase, consistent with the change trend of ABAF.
increase, consistent with the change trend of ABAF .
Appl. Sci. 2017, 7, 1259 11 of 16
Appl. Sci. 2017,
Appl. Sci. 2017, 7,
7, 1259
1259 11 of 16
11 of 16

100
100
0.05
0.05
80
80 0.1
0.1
60 0.2
0.2
60
40
40

(kN)
FFdyndyn(kN)
20
20
0
0
-20
-20
-40
-40
-60
-60
-80
-80
28 29 30 31 32 33 34
28 29 30 31 32 33 34
Time (ms)
Time (ms)

Figure
Figure 13.
Figure 13. The
13. The response
Theresponse of
response of wheel–track
of wheel–track force
wheel–track force under
force under excitation
under excitation of
excitation of welds
of welds with
welds with different wavelengths.
with different wavelengths.

As mentioned
As mentioned above, the the first wave
wave crest of of ABA at at rail welds
welds represents the the wheel/rail
wheel/rail impact.
impact.
As mentioned above,
above, the first
first wave crest
crest of ABAABA at rail rail welds represents
represents the wheel/rail impact.
Eighteen
Eighteen rail welds
railwelds (including
welds(including
(including wave
wave depths
depths of 0.1, 0.2, and 0.3 mm that correspond to six
six
Eighteen rail wave depths of 0.1,of0.2,0.1,
and0.2,
0.3 and 0.3 correspond
mm that mm that correspond to
to six wavelengths
wavelengths from
wavelengths from 0.05
0.05 toto 0.5m
0.5m in
in Figure
Figure 11)
11) were
were selected
selected forfor calculation.
calculation. The
The relationship between
between
from 0.05 to 0.5m in Figure 11) were selected for calculation. The relationshiprelationship
between ABAF and
ABA
ABA F and maximum of Fdyn (Fdyn,max) has been established, as shown in Figure 14. Discrete points in
F and maximum of Fdyn (Fdyn,max) has been established, as shown in Figure 14. Discrete points in
maximum of Fdyn (Fdyn,max ) has been established, as shown in Figure 14. Discrete points in the figure
the figure
the figure are
are the
the calculation
calculation results
results and
and the
the corresponding
corresponding fitted fitted curve
curve is
is also shown
shown in in the
the figure;
figure;
are the calculation results and the corresponding fitted curve is also shownalsoin the figure; please see
please see
please see Equation
Equation (6),
(6),
Equation (6),
Fdyn,max =
F
Fdyn,max = 0.41
= 0.41⋅⋅ ABA
0.41 ABA
· ABA F (6)
(6)
dyn,max F F (6)

140
140
Original data
Original data
120
120 Fitting curve
Fitting curve
yy =
= 0.41x
0.41x
100
100 Safety limitation
limitation
Safety
(kN)
(kN)

80
80
dyn,max
FFdyn,max

60
60

40
40

20
20

00
00 50
50 100
100 150
150 200
200 250
250 300
300
ABAF (m/s
(m/s )) 22
ABAF

Figure 14.
Figure 14. The
The relationship
relationship between
between ABA
ABAF and
and FFdyn,max
dyn,max.
Figure 14. The relationship between ABAFF and Fdyn,max. .

Obviously, aa linear
Obviously, linear relationship
relationship is is presented
presented between
between ABA ABAFF and
and FFdyn,max
dyn,max. The measured data in
. The measured data in
FigureObviously,
4 shows athat
linear
the relationship
ABA excited is by
presented
rail weld between ABAF and
irregularities Fdyn,maxwithin
fluctuated . The measured
range data
the range in
of ±50
±50
Figure 4 shows that the ABA excited by rail weld irregularities fluctuated within the of 2
Figure
m/s 2 at4ashows
vehiclethat the ABA
speed of 300excited
300 km/h. byThus,
rail weld
the irregularities
excitedfluctuated
ABA excited by rail within
rail welds
welds ofthe range
long of ±50 m/s
wavelengths or
m/s 2 at a vehicle speed of km/h. Thus, the ABA by of long wavelengths or
at a vehicle
shallow wavespeed
depthsof 300 km/h.
is submerged,
submerged,Thus,as theRW2
ABAinexcited
Figureby 4.rail welds ofa long
However, low ABAwavelengths
ABA cancan mean
meanor shallow
shallow wave depths is as RW2 in Figure 4. However, a low aa small
small
wave
dynamic depths is
effect. submerged,
For instance, as
theRW2ABA in Figure
of 50 4.
m/s However, a low ABA can mean
2 corresponds merely to the Fdyn of 20 kN—fara small dynamicless
dynamic effect. For instance, the ABA of 50 m/s corresponds merely to the Fdyn of 20 kN—far less
2
2 corresponds
effect.
than the For instance,
the safety the
safety threshold—as ABA
threshold—as shown of 50
shown inm/sin Figure
Figure 14. merely
14. According
According to to the F
to maintenance of 20 kN—far
maintenance regulations
dyn less
regulations on than
on Chinese
Chinesethe
than
safety threshold—as
high-speed railways, shown
railways, the in Figure
the FFdyn,max 14. According to maintenance regulations
dyn,max shall be no more than 90 kN; we can find the corresponding
on Chinese high-speed
high-speed shall be no more than 90 kN; we can find the corresponding
railways, the
threshold Fdyn,max
value of ABA shall be no more than 90 kN; we can find the corresponding threshold value of
F to be 220 m/s22. High ABA is easily identified from the measured data, as
threshold value of ABA F to be 220 m/s . High ABA is easily identified from the measured data, as
2 High ABA is easily identified from the measured data, as shown in Figure 4,
ABA
shown F to
inbe 220 m/s
Figure so .that
4, so that ABA
ABA is is aa good
good measure
measure for for monitoring
monitoring thethe health
health conditions
conditions of of rail
rail welds.
welds.
shown in Figure 4,
Hence, 17 groups of welds in Figure 6 met the safety threshold. However,
Hence, 17 groups of welds in Figure 6 met the safety threshold. However, a portion of these were a portion of these were
Appl. Sci. 2017, 7, 1259 12 of 16

so that ABA is a good measure for monitoring the health conditions of rail welds. Hence, 17 groups of
welds in Figure 6 met the safety threshold. However, a portion of these were within an inch of the
danger threshold of 220 m/s2 , indicating the need for maintenance as soon as possible.

4.3. Discussion of This Paper

4.3.1. Qualitative Analysis on the Effectiveness of the Model

The FE modeling approach employed in this study originated from Li and Zhao [26].
This approach has been validated by Li and Molodova for high-frequency vehicle–track interaction at
squats [19,20], and by Zhao and An for local rolling contact fatigue occurring on high-speed wheel
tread [22]. This section tries to further validate it through measured ABA in time and frequency
domains. It should be noted that no corresponding weld geometry for measuring ABA in Section 2
was available so only a qualitative analysis will be performed to discuss the effectiveness of the model.
Gao and Zhai [5] indicated that on Chinese high-speed railway lines the average weld geometry
irregularity in 0.11 m (0.05~0.39 m) accounted for 78%. Meanwhile, weld straightness should be no
more than 0.2 mm for tracks allowing speeds of 300 km/h, according to requirements of maintenance
regulations. Therefore, we assumed that geometry wavelength of rail welds corresponding to the
measured data in Figure 4 is from 0.05 to 0.4 m with a depth less than 0.2 mm. Figure 5b shows
that the wavelength of the rail weld excites two high-frequency vibrations that are well reflected in
Figure 10b; similar to the frequency domain response of the 0.2 m wavelength welds. Figure 10a
indicates that the wavelength raises the ABA amplitude to about 100 m/s2 in the time domain, less than
the measured data (148 m/s2 ) in Figure 5a. The difference might be caused by the low frequency
response generated without the consideration of an actual irregularity in the model built in this paper.
For example, a high-energy low-frequency response is seen in Figure 5b, while there is no such response
in Figure 10b. It should be emphasized that the results of two measured ABAs in the literature [18]
did not overlap completely due to vehicle hunting. In Figure 11a, variations in ABA ranged from
35.6 to 204.6 m/s2 under wavelengths from 0.05 to 0.4 m. It agrees well with the measured data in
Figure 6 when we consider that the measured ABA induced by other irregularities is about 50 m/s2 .
Therefore, the 3D transient finite element model established in Section 3 can effectively simulate the
ABA response that is excited by the rail welds.
The effect of differences between actual parameters and model parameters of a vehicle–track
system on ABA should be further investigated in the future. These include the change in stiffness of
rail pads and fatigue damage of the fastening system [1], the location of the rail weld between the
adjacent sleepers [27], existence of track lateral irregularity [28], noise from braking pads, and damage
appearance due to an increase in total train mass, etc.

4.3.2. Estimation of Rail Weld Based on ABA Signal Using Time–Frequency Techniques
We note that the maximum acceleration of the axle box (202 m/s2 ) in Figure 5a is not the peak
excited by weld RW1, but is the wave crest (148 m/s2 ) appearing after 11.15 s. In other words,
considering merely the maximum amplitude of the ABA in the time domain, the degradation might
be overestimated. As the rail weld can excite high frequency vibrations in ABA, which can be found
through S transform, we can identify the true location of the rail weld by time–frequency techniques
and get corresponding amplitude, as shown in Figure 5. Further, we can estimate the rail weld through
a safety limitation (220 m/s2 ) proposed in Section 4.2. However, a question we should pay more
attention to is whether short wavelength irregularity in a rail weld can inspire a higher contact force
and increase more rapidly than that of longer wavelengths [9]. Therefore, there is a need to identify
the geometries of rail welds.
By parametric analysis in Section 4.1, it can be found that the weld length and depth determine
the vibration frequency and amplitude of ABA, respectively. Based on these characteristics, we present
an approach for obtaining rail weld geometry from ABA, expressed in Figure 15. Step 1: develop a
Appl. Sci. 2017, 7, 1259 13 of 16

Appl. Sci. 2017, 7, 1259

Appl. Sci. 2017, 7, 1259 13 of 16
13 of 16
database of ABA and rail weld geometries through the FE model established in Section 3.1; Step 2:
analyze the
technique, and measured
and identify ABA
identify the in time
the wavelength and
wavelength and frequency
and wave domains
wave depth
depth of using
of the a
the railtime–frequency
rail weld
weld from
from the technique,
the database
database
technique,
and identify
established in the wavelength
in step
step 1;
1; finally, and
finally, make wave depth
make aa decision of the
decision whether rail
whether the weld
the railfrom
rail weld the database
weld should
should be established
be maintained. in
maintained. Taking step
Taking
established
1;RW1
finally,
asanmake
anexample, a decision
example,ititmay whether
maybe be0.1–0.2 the
0.1–0.2m rail
min weld
inlength should
lengthas be
asititcan maintained.
canexcite
excitefrequenciesTaking
frequenciesof RW1
of350
350Hzas
Hzand an example,
and1150
1150Hz Hz
RW1 as
itshown
may be in0.1–0.2
Figures m5bin and
length
10b.asFurther,
it can excite
the frequencies
true amplitude of 350
of Hz and
ABA by 1150
RW1 Hz
is shown
148 m/s 2in
, Figures
and we 5b
can
shown in Figures 5b and 10b. Further, the true amplitude of ABA 2 by RW1 is 148 m/s2, and we can
and
judge 10b. Further,
that thewave
wave thedepth
true amplitude
shouldbe of ABA
beabout
about 0.2by
mmRW1 is 148 m/s
according , and 11a.
toFigure
Figure we can judge that the wave
judge that the depth should 0.2 mm according to 11a.
depth should be about 0.2 mm according to Figure 11a.

MeasuredABA
Measured ABA FEmodel
FE model

Time–frequency
Time–frequency
techniques
techniques

Frequency
Frequency Amplitude
Amplitude

Developaadatabase
Develop database
Identifywavelength
Identify wavelength
ofABA
of ABAand
andrail
rail
anddepth
and depth
welds’geometries
welds’ geometries

Maintenance?
Maintenance?

Figure 15.
Figure15. Flowchart
15.Flowchart of
Flowchartof an
ofan approach
anapproach to
approachtotoestimate
estimaterail
railweld.
weld.
Figure estimate rail weld.

4.3.3. Comparison
4.3.3.Comparison
4.3.3. of
Comparisonof Other
ofOther Defects
Defectsand
OtherDefects andRail
and RailWeld
Rail Weldon
Weld onABA
on ABA
ABA
The
The above
The above results
above results were
results were obtained
were obtained by
obtained by assuming
by assuming that
assuming that the
that the wheel
the wheel tread
wheel tread is
tread isis smooth
smooth and
smooth and that
and that welds
that welds
welds
only
only exist on the rail surface; the influence of other flaws in the wheel/rail contact interface was not
only exist
existon
on the
the rail
rail surface;
surface;the
theinfluence
influence of
ofother
other flaws
flaws ininthe
thewheel/rail
wheel/rail contact
contact interface
interface was
was not
not
considered.
considered. In
considered. In fact,
In fact, a variety
fact, aa varietyof
variety of defects
of defects such
defects such as
such asrail corrugation,
as rail
rail corrugation, wheel
corrugation, wheel polygonal
wheel polygonal wear,
polygonal wear, wheel flats,
wear, wheel etc.,
wheel flats,
flats,
exist
etc., in operation,
exist in as
operation, shown
as in
shown Figure
in 16.
FigureIt is
16.necessary
It is to
necessary identify
to these
identify interfering
these signals.
interfering
etc., exist in operation, as shown in Figure 16. It is necessary to identify these interfering signals. signals.

(a)
(a) (b)
(b)

Figure16.
Figure 16.Rail
Railcorrugation
corrugation(a)
(a)and
andwheel
wheelflat
flat(b).
(b).
Figure 16. Rail corrugation (a) and wheel flat (b).

AA schematic
schematic diagram
diagram of of ABA
ABA induced
induced byby rail
rail corrugation
corrugation and and aa wheel
wheel flat
flat are
are presented
presented in in
Figure
A 17 with
schematic horizontal
diagram and
of vertical
ABA axes
induced representing
by rail the
corrugation running
and a time
wheel
Figure 17 with horizontal and vertical axes representing the running time and acceleration of the and
flatacceleration
are of
presented the
in
axle box,
Figure
axle box, respectively.
17 with horizontal
respectively. As
As thevertical
and
the geometric form
axesform
geometric of rail
rail corrugation
representing
of corrugation
the running is is continuously
timecontinuously fluctuating,
and acceleration the
of the axle
fluctuating, the
excited
box, ABA inin corrugation
respectively.
excited ABA corrugation geometric
As the geometric
geometricformsize
of λ
size λ (a
rail(a typical size
corrugation
typical sizeisisiscontinuously
65~80 mm)
65~80 mm) isisfluctuating,
continued wavelength
continued wavelength
the excited
vibration, as shown in Figure 17a. The vibration mode is different from that
vibration, as shown in Figure 17a. The vibration mode is different from that in Figure 10a. ABA in Figure 10a. ABA at at
welds will gradually decay upon three to four wave crests due to the lack of
welds will gradually decay upon three to four wave crests due to the lack of a continued excitationa continued excitation
source,while
source, whilethe
therail
railcorrugation
corrugationcancanexcite
excitelonger
longerandandwithwithcontinued
continuedfluctuation.
fluctuation.Similarly,
Similarly,wheel
wheel
Appl. Sci. 2017, 7, 1259 14 of 16

ABA in corrugation geometric size λ (a typical size is 65~80 mm) is continued wavelength vibration,
as shown in Figure 17a. The vibration mode is different from that in Figure 10a. ABA at welds
will gradually decay upon three to four wave crests due to the lack of a continued excitation source,
Appl. Sci. 2017, 7, 1259 14 of 16
while the rail corrugation can excite longer and with continued fluctuation. Similarly, wheel polygonal
wear is continuous
polygonal with circular
wear is continuous withgeometry fluctuation
circular geometry appearing
fluctuation in the wheel
appearing in thetread.
wheelThe excited
tread. The
acceleration of the axle box should resemble that of rail corrugation. Thus, the same method
excited acceleration of the axle box should resemble that of rail corrugation. Thus, the same method can be
adopted to identify its influence.
can be adopted to identify its influence.

(a) (b)

2π R

ABA (m/s )
ABA (m/s )

2
2

λ
Rail corrugation

Time (s) Time (s)

Figure
Figure 17.
17.Schematic diagrams
Schematic of axle
diagrams boxbox
of axle acceleration induced
acceleration by (a)by
induced rail(a)corrugation and (b)and
rail corrugation wheel
(b)
flat.
wheel flat.

Differing from rail corrugation, wheel flat is a local defect existing in the wheel tread. An impact
Differing from rail corrugation, wheel flat is a local defect existing in the wheel tread. An impact
will be generated on the track system with the wheel rotating every circle. The change in ABA with
will be generated on the track system with the wheel rotating every circle. The change in ABA with
time is presented in Figure 17b. It can be observed that the distance between two wave crests is the
time is presented in Figure 17b. It can be observed that the distance between two wave crests is the
circumference of the wheel. Wheel circumference is generally about 2.7~2.9 m, which is far less than
circumference of the wheel. Wheel circumference is generally about 2.7~2.9 m, which is far less than
the distance (100 m) of two adjacent rail welds. On the other hand, according to site measurements,
the distance (100 m) of two adjacent rail welds. On the other hand, according to site measurements,
the typical wavelength range of the wheel flat in high-speed railways is from 15 to 40 mm, less than
the typical wavelength range of the wheel flat in high-speed railways is from 15 to 40 mm, less than the
the size of a rail weld, which means that a higher ABA can be raised by wheel flats compared to rail
size of a rail weld, which means that a higher ABA can be raised by wheel flats compared to rail welds.
welds. So, interference of a wheel flat in the signal of ABA can be identified by the above two
So, interference of a wheel flat in the signal of ABA can be identified by the above two differences.
differences. To be sure, a more reliable method is to install the acceleration sensors in different
To be sure, a more reliable method is to install the acceleration sensors in different wheels to avoid the
wheels to avoid the influence of a flat in a certain wheel.
influence of a flat in a certain wheel.
5. Conclusions
5. Conclusions
In
In this
this paper,
paper, the
the vibration
vibration characteristics
characteristics of
of an
an ABA
ABA at
at rail
rail welds
welds in
in the
the time–frequency
time–frequency domain
domain
have been analyzed at the speed of 300 km/h according to experimental data. A
have been analyzed at the speed of 300 km/h according to experimental data. A correspondingcorresponding FE
model was developed to simulate ABA and wheel/rail force under different rail
FE model was developed to simulate ABA and wheel/rail force under different rail welds, which welds, which
established a quantitative relation between ABA and the wheel/rail force. Based on the results of this
established a quantitative relation between ABA and the wheel/rail force. Based on the results of this
study, the following conclusions can be drawn:
study, the following conclusions can be drawn:
(1) According to field tests, the acceleration of an axle box changed within the range of ±50 m/s22
(1) According to field tests, the acceleration of an axle box changed within the range of ±50 m/s
under normal operation, which significantly increased at the welds to amplitudes ranging from
under normal operation, which significantly increased at the welds to amplitudes ranging from
66 to 200 m/s2. 2High-frequency vibrations at welds existed for a very short time, around 2 to 5
66 to 200 m/s . High-frequency vibrations at welds existed for a very short time, around 2 to
ms. Energy of the high-frequency vibration centralized mainly in two frequency bands: 350 to
5 ms. Energy of the high-frequency vibration centralized mainly in two frequency bands: 350 to
500 Hz and 1000 to 1200 Hz.
500 Hz and 1000 to 1200 Hz.
(2) ABA at rail welds exhibited drastic fluctuations both in measured data and in simulations. The
(2) ABA at rail welds exhibited drastic fluctuations both in measured data and in simulations.
first wave was caused by the impact of the wheel/rail force, while the continued vibration
The first wave was caused by the impact of the wheel/rail force, while the continued vibration
depended on the resonance of the vehicle–track system, for which the amplitude was greater
depended on the resonance of the vehicle–track system, for which the amplitude was greater
than that generated by the wheel/rail force in certain conditions. Therefore, the maximum value
than that generated by the wheel/rail force in certain conditions. Therefore, the maximum
of ABA in the time domain always overestimated the wheel/rail dynamic interaction. To deal
with this problem, the position of the rail welds should be determined first by means of
time–frequency analysis.
(3) A linear relationship was found between the ABAF and Fdyn,max, and corresponding expressions
were established. A value of 220 m/s2 after 1500 Hz LPF can be used as a safety value for
real-time onsite monitoring in judging the necessity of maintaining the rail welds.
Appl. Sci. 2017, 7, 1259 15 of 16

value of ABA in the time domain always overestimated the wheel/rail dynamic interaction.
To deal with this problem, the position of the rail welds should be determined first by means of
time–frequency analysis.
(3) A linear relationship was found between the ABAF and Fdyn,max , and corresponding expressions
were established. A value of 220 m/s2 after 1500 Hz LPF can be used as a safety value for
real-time onsite monitoring in judging the necessity of maintaining the rail welds.

For a further study, a more detailed database of rail weld geometries and axle box accelerations
based on measurements should be established. In addition, simulations should be performed, taking
into account the contribution of track stiffness and its variation on ABA, which can be used to detect
deterioration of track structure.

Acknowledgments: The present work has been supported by the National Key R & D Plan of China
(2016YFC0802203-4), National Natural Science Foundation of China (51425804, 51608459, 51605394 and 51778542)
and Doctoral Innovation Fund Program of Southwest Jiaotong University.
Author Contributions: Boyang An and Jinmang Xu conceived and designed the idea and numerical study;
Boyang An, Ping Wang, Rong Chen and Dabin Cui analyzed the data and contributed in writing the paper;
Ping Wang, Jingmang Xu, Rong Chen and Dabin Cui contributed in obtaining the funding for performing
the research.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

ABA Axle Box Acceleration

LPF Low Frequency Filter
CDF Cumulative Distribution Function
3D Three Dimensional

References
1. Oregui, M.; Li, S.; Núñez, A.; Li, Z.; Carroll, R.; Dollevoet, R. Monitoring bolt tightness of rail joints using
axle box acceleration measurements. Struct. Control Health Monit. 2017, 24, 2. [CrossRef]
2. Esveld, C. Modern Railway Track; MRT-Productions: Zaltbommel, The Netherlands, 2001.
3. Mutton, P.; Alvarez, E. Failure modes in aluminothermic rail welds under high axle load conditions.
Eng. Fail. Anal. 2004, 11, 151–166. [CrossRef]
4. Steenbergen, M. Rolling contact fatigue in relation to rail grinding. Wear 2016, 356, 110–121. [CrossRef]
5. Gao, J.; Zhai, W.; Guo, Y. Wheel–rail dynamic interaction due to rail weld irregularity in high-speed railways.
Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit 2016. [CrossRef]
6. Wen, Z.; Xiao, G.; Xiao, X.; Jin, X.; Zhu, M. Dynamic vehicle–track interaction and plastic deformation of rail
at rail welds. Eng. Fail. Anal. 2009, 16, 1221–1237. [CrossRef]
7. Correa, N.; Vadillo, E.G.; Santamaria, J.; Herreros, J. A versatile method in the space domain to study
short-wave rail undulatory wear caused by rail surface defects. Wear 2016, 352, 196–208. [CrossRef]
8. Mandal, N.; Dhanasekar, M. Sub-modelling for the ratchetting failure of insulated rail joints. Int. J. Mech. Sci.
2013, 75, 110–122. [CrossRef]
9. Xiao, G.; Xiao, X.; Guo, J.; Wen, Z.; Jin, X. Track dynamic behavior at rail welds at high speed. Acta Mech. Sin.
2010, 26, 449–465. [CrossRef]
10. Steenbergen, M.; Esveld, C. Rail weld geometry and assessment concepts. Proc. Inst. Mech. Eng. Part F J. Rail
Rapid Transit 2006, 220, 257–271. [CrossRef]
11. Steenbergen, M.; Esveld, C. Relation between the geometry of rail welds and the dynamic wheel-rail response:
Numerical simulations for measured welds. Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit 2006, 220,
409–423. [CrossRef]
12. Haigermoser, A.; Luber, B.; Rauh, J.; Gräfe, G. Road and track irregularities: Measurement, assessment and
simulation. Veh. Syst. Dyn. 2015, 53, 878–957. [CrossRef]
Appl. Sci. 2017, 7, 1259 16 of 16

13. Grassie, S. Measurement of railhead longitudinal profiles: A comparison of different techniques. Wear 1996,
191, 245–251. [CrossRef]
14. Liang, B.; Iwnicki, S.; Zhao, Y.; Crosbee, D. Railway wheel-flat and rail surface defect modelling and analysis
by time–frequency techniques. Veh. Syst. Dyn. 2013, 51, 1403–1421. [CrossRef]
15. Molodova, M.; Oregui, M.; Núñez, A.; Li, Z.; Dollevoet, R. Health condition monitoring of insulated joints
based on axle box acceleration measurements. Eng. Struct. 2016, 123, 225–235. [CrossRef]
16. Li, Z.; Molodova, M.; Núñez, A.; Dollevoet, R. Improvements in axle box acceleration measurements for the
detection of light squats in railway infrastructure. IEEE Trans. Ind. Electron. 2015, 62, 4385–4397. [CrossRef]
17. Zhai, W.; Liu, P.; Lin, J.; Wang, K. Experimental investigation on vibration behaviour of a CRH train at speed
of 350 km/h. Int. J. Rail Transp. 2015, 3, 1–16. [CrossRef]
18. Zhai, W. Vehicle–Track Coupling Dynamics, 3rd ed.; Science Publishing House: Beijing, China, 2007.
19. Molodova, M.; Li, Z.; Dollevoet, R. Axle box acceleration: Measurement and simulation for detection of
short track defects. Wear 2011, 271, 349–356. [CrossRef]
20. Molodova, M.; Li, Z.; Núñez, A.; Dollevoet, R. Validation of a finite element model for axle box acceleration
at squats in the high frequency range. Comput. Struct. 2014, 141, 84–93. [CrossRef]
21. Zong, N.; Dhanasekar, M. Sleeper embedded insulated rail joints for minimising the number of modes of
failure. Eng. Fail. Anal. 2017, 76, 27–43. [CrossRef]
22. Zhao, X.; An, B.; Zhao, X.; Wen, Z.; Jin, X. Local rolling contact fatigue and indentations on high-speed
railway wheels: Observations and numerical simulations. Int. J. Fatigue 2017, 103, 5–16. [CrossRef]
23. Li, S.; Li, Z.; 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. [CrossRef]
24. Stockwell, R.; Mansinha, L.; Lowe, R. Localization of the complex spectrum: The S transform. IEEE Trans.
Signal Process. 1996, 44, 998–1001. [CrossRef]
25. Xu, J.; Wang, P.; Gao, Y.; Chen, J.; Chen, R. Geometry evolution of rail weld irregularity and the effect on
wheel-rail dynamic interaction in heavy haul railways. Eng. Fail. Anal. 2017, 81, 31–44. [CrossRef]
26. Li, Z.; Zhao, X.; Dollevoet, R.; Molodova, M. An investigation into the causes of squats—Correlation analysis
and numerical modeling. Wear 2008, 265, 1349–1355. [CrossRef]
27. Askarinejad, H.; Dhanasekar, M.; Cole, C. Assessing the effects of track input on the response of insulated
rail joints using field experiments. Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit 2013, 227, 176–187.
[CrossRef]
28. Zhang, Z.; Dhanasekar, M. Dynamics of railway wagons subjected to braking torques on defective tracks.
Veh. Syst. Dyn. 2012, 50, 109–131. [CrossRef]

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).