J Fail. Anal. and Preven.

https://doi.org/10.1007/s11668-019-00696-w

TECHNICAL ARTICLE—PEER-REVIEWED

Experimental Analysis of Rail Contact Fatigue Damage on Frog

Rail of Fixed Common Crossing 1:12
Mykola Sysyn . Franziska Kluge . Dmitri Gruen . Vitalii Kovalchuk . Olga Nabochenko

Submitted: 15 July 2019

Ó ASM International 2019

Abstract Assembled fixed common crossings of the steel of the calculated impact angles along the frog rail
grade R350HT with high switch angles are subjected to demonstrated the presence of a small cluster of high-impact
quick rail contact fatigue (RCF) deterioration. The cross- angles outside of the zone subjected to the highest number
ings are a significant cost driver in railway maintenance of impacts. The results of the geometrical measurements
and have an important influence on reliability and avail- correspond to those obtained by the acceleration
ability in railway transportation. The present paper is measurements.
devoted to the experimental study of the deterioration
causes that is based on inertial and geometrical measure- Keywords Railway turnout  Common crossing 
ments on one crossing during its full lifecycle. The causes Rolling contact fatigue  Deterioration causes 
are classified into primary technical ones and secondary Inertial measurements  Statistical analysis 
ones, which depend on the operational work and mainte- Dynamic loadings
nance of track and rolling stock. The primary causes are
studied using analyses of the impact position and inertial
measurements on the frog nose during the crossing life- Introduction
cycle, the end of which is marked by RCF defects. An
application of the machine learning technique t-distributed Railway turnouts with fixed common crossing are the most
stochastic neighbor embedding (t-SNE) allowed to deter- commonly used turnout construction in European railway
mine a small cluster of high-impact loadings that networks. Despite substantial progress in the development
correspond to the future damage zone location. The of turnouts with swing-nose crossing or moveable point
appearance of the high-impact loadings group on a certain frog, there is still no cost-effective alternative to stiff
location is explained by the secondary causes: rail and crossings for common use. The turnouts with curve radii up
wheel wear and wheel lateral position. A multivariate to 500 m and switch angles lower than 1/14 amount to 90%
kinematic modeling of wheel passages from wing to frog of the 72,000 turnouts that are used in the German Rail-
rail provided the wheel trajectories, impact position, and ways (DB AG) network [1]. Common crossings are the
trajectory kink angle in the impact point. The distribution most loaded part of turnouts and are subject to relatively
quick deterioration compared to ordinary track. This results
in high maintenance costs and a significant limitation of
M. Sysyn (&)  F. Kluge  D. Gruen railway transportation reliability and availability. The
Institute of Railway Systems and Public Transport, Technical
University of Dresden, Hettnerstraße 3/353, 01069 Dresden, financial costs of turnout maintenance amount to more than
Germany 30% of overall maintenance costs for railway track
e-mail: [email protected] superstructure [2]. The lifecycle of stiff common crossings
that are assembled of steel R350HT is usually 50 Mt,
V. Kovalchuk  O. Nabochenko
Department of the Rolling Stock and Track, Lviv Branch of which is more than ten times lower than the lifecycle of
Dnipro National University of Railway Transport, I. rails with the same steel in the ordinary track [3]. The high
Blazhkevych Str. 12a, Lviv 79052, Ukraine

123
 J Fail. Anal. and Preven.

maintenance costs due to short lifecycles are amplified by and traffic operation. The first-level primary causes of RCF
cost expensive manual inspections that are necessary dur- damage are contact stresses and accumulated loading. The
ing the lifecycle to prevent the formation of failures. second-level primary causes are considered the contact
Nevertheless, there are unexpected failures on crossings area and the impact force that directly influence the contact
that result in extensive consequences of progressive train stress. The contact surfaces of wing and frog rail substan-
delays over the whole network. The consequences are tially vary along and across the common crossing, leading
estimated in the study [4] as 55% of the train delays in the to varying contact areas depending on the longitudinal and
overall railway network of the Netherlands due to 6% of the lateral wheel position. The third-level primary causes
unplanned turnout faults. are those that influence the dynamic interaction, namely
Rail contact fatigue (RCF) failure mode is the first kind geometric irregularity, track stiffness, contact stiffness and
of lifecycle limiting failures on assembled common the dynamic masses of wheel and crossing elements.
crossings from rail steel. Other failure modes of the com- The secondary causes of RCF on common crossing rails
mon crossings, such as rail wear, sleeper fastening can be directly controlled within the crossing maintenance
damages, and ballast settlements, are much easier to predict norms or by influencing the traffic operation. These factors
and usually appear much later than RCF failures. The RCF are the most interesting for practical aims, such as pre-
development is to a great extent difficult to predict. This is diction of the service life or optimization of the structure of
confirmed by the large spread in the lifecycle length of maintenance norms for a prolongation of the lifecycle. Rail
common crossings, ranging from 25 to up to 120–200 Mt, and wheel wear, as well as the wheel’s lateral position,
even for equal geometrical design and steel grade, as well influence both the wheel trajectory and the contact area,
as for similar operational loading [5]. The difficulty in the depending on the position of the wheel impact. The geo-
assessment of the causes of RCF can on the one hand be metrical irregularity together with train masses and
explained by the complex nonlinear, dynamical interaction velocity influences the dynamic loading of the rails in the
in the contact of wheel and rails. On the other hand, the common crossing. The bold marked secondary factors in
mechanism of RCF is very complicated with many inter- Fig. 1 and the chain of the primary factors that lead to the
connected factors. A simplified mechanism of the factor RCF are considered in the study.
influence on RCF damage is shown in Fig. 1. The mech-
anism does not take into account a possible influence of Critical Review of the Current Studies
wear on RCF development, as described in [6], and rolling
contact damage mechanism. The research topic of the rolling contact fatigue damages
The factors in Fig. 1 are divided into primary technical on crossing rails is not young. Many works have been done
factors and secondary or root ones, such as rail and wheel by many researchers of Chalmers University of Technol-
wear and wheel lateral position, depending on maintenance ogy, TU Delft, University of Birmingham, Beijing Jiaotong
University, and others.

Fig. 1 Influence diagram of the appearance of RCF damage on stiff common crossings

123
J Fail. Anal. and Preven.

The detailed research of the influence mechanism of the crossing rail that are based on FEM and multibody meth-
factors on the RCF and wear in common crossing rails is ods. The EEM model was validated with experimental
presented in the studies [7–11]. An initiation of subsurface measurements of accelerations on common crossing.
damage on differently deformed crossing nose geometries Multibody models of dynamic interaction between train
is studied by [7] using FEM crossing model. Thereby, three and a standard turnout design were developed by [18, 19].
different materials of crossing nose are considered under The variation in rail profile and contact between the back
the impacts of new and worn wheels with different of the wheel flange and the check rail are considered. The
velocities, axle loads, and other parameters. The ability of model validation was based on wheel–rail contact forces
manganese steel crossings to withstand unfavorable load that were measured using test train with an instrumented
situations is explained within the study. Authors in [8, 12] wheelset. The wheelset contained pairs of strain gauges
developed a three-dimensional explicit finite element glued to the wheel disk with step 45° that enabled mea-
model of a wheel rolling over a crossing rail with two- suring of vertical and lateral contact forces. The similar
dimensional geometrical contact analyses. The model was multibody modeling with application of commercial soft-
validated via transition regions form and crossing accel- ware packages but without experimental validation is
erations. The RCF life is analyzed with damage/fatigue presented in [20–23]. The common thing for most papers is
parameter that is based on different fatigue models. the absence or relative poor model validation that is based
Authors in [9] presented numerical predictions of the long- on the average measured values for many usually unknown
term accumulation of plastic deformation and wear using model parameters. The validation of accelerations shows
simulations of dynamic vehicle–track interaction, wheel– some similarity between the mean measured acceleration
rail contact, and rail damage. Variations in worn wheel process and one case of modeling. However, the deviation
profile, vehicle speed, and wheel–rail friction coefficient of the measured accelerations is as high as their mean
are taken into account for two different steel grades. An value. In this respect, even the most complicated models
elastic–plastic finite element simulation of wheel and show not better validation results as the simplest analytical
common crossing contact interaction is presented by [10]. models as shown in [24]. Thereby, the simple models with
The variation of stress and deformation in frog and wing little unknown parameters have the advantage of simple
rail depending on axle loading and train speed are found. A interpretation.
numerical prediction of rail degradation taking into account Parameter study of maintenance and operation factor
3D profile and hardness measurements is proposed in [11]. influence is presented in many research papers. The influ-
The prediction is based on finite element simulation of ence of 14 factors on the train–track interaction at the
wheel/rail interaction. A decrease in contact stress during switch panel is shown in the study [25] that is based on
the lifecycle of crossing is noted together with the increase multibody simulation. The factors of axle load, wheel/rail
in dynamic loading. An application of different elastic non- profile, running direction, and friction coefficient were
Hertzian contact models for simulating rail damages in found to be the most significant. The similar model-based
railway turnout switch panel is presented in [13]. The research but for wheel transition behavior at crossing panel
results are compared with Kalker’s 3D theory and an is presented in [26]. The studied parameters were local
elastic–plastic FEM model. The comparison shows quite crossing geometry, the track alignment, and the wheel
different calculation results for compared models for both profiles. The wheel profile and the crossing longitudinal
wear and RCF damage estimation. The RCF initiation height profile are identified to be the most significant fac-
estimation is based on different hypotheses that used in tors. The fatigue life influence of this factors and additional
various theoretical models that are presented by many factors such as velocity, railpad stiffness, and friction
studies [6, 14–16]. In general, the most research of wheel coefficients is estimated in the subsequent study [4]. Both
and rail contact problem and RCF damages on common the experimental measurements and the fitted to the
crossing is highly theoretical and has low experimental experiment model demonstrate the highest fatigue damage
substantiation. The substantiation is usually indirect and is in the most frequent loaded points of the frog rail. The
based on some cases of damages without taking into influence of wheel profile on the dynamic interaction
account factor influence during overall lifecycle. The between a railway freight vehicle and a switch and crossing
studies point out the complexity of the contact crack ini- and their damage is analyzed in [27]. One hundred and
tiation and development process that is based on nonlinear twenty wheel profiles from freight wagons in regular traffic
plastic contact interaction and a complex contact geometry. have been measured and categorized with respect to wear.
The detailed studies of the influence mechanism of the It is indicated that equivalent conicity is the wheel profile
maintenance factors on dynamic interaction of wheel and parameter correlating best to damage in the S&C panels.
common crossings are presented in [17, 18]. The studies of More detailed study of wheel profile influence together
[8, 17] present the models of a wheel rolling over a with wheel lateral position and crossing rails longitudinal/

123
 J Fail. Anal. and Preven.

cross profiles is presented in the further study [28]. The parameters are considered. One of the most complete
frog/wing rail optimization is performed within the study, experimental investigations on turnout monitoring is pre-
but the rails wear during the lifecycle is not taken into sented in [42], where the multi-point strain measurements
account. The estimation of the modeled damage distribu- in switch panel rails were taken. The measurements show
tion and the impact point position distribution along the the significant increase in strains in the transfer zone from
frog rail shows that their maxima coincide. A study of the switch rail to stock rail during the turnout lifecycle.
crossing rails to degradation relating to the dynamic fric- The performed review of the current studies on common
tional contact is presented in [29]. In the validated FEM crossings shows that most of them are theoretical based on
model, the adhesion–slip regions, shear traction, and numerical simulation of dynamic and contact interaction.
microslip are taken into account to calculate the plastic Many of the studies have either no experimental validation
deformation and frictional work. The high influence of or the vague one that is based on the similarity to mean
frictional work on RCF degradation is indicated as well as measured values of wide scattered dynamic interaction
the out-of-sync of the maximal normal and tangential values. Even less validation is presented in the studies
contact force. contact interaction that is interpreted within many theo-
A wide range of recent experimental and theoretical retical approaches. There are disproportionately low
studies are devoted to a relatively new research field— number of experimental measurements and almost no ones
prognostic health monitoring (PHM) and predictive main- that are devoted to the continuous monitoring during the
tenance [30–32]. The authors in [33, 34] studied the common crossing lifecycle. That leads to the result that the
possibilities of onboard and track-side monitoring systems calculation models quality is assessed not by their ability to
for the prediction of the service life of common crossings. explain the certain cases of measurements but by their
An analysis of the growth of vertical accelerations and the complexity and the number of considered parameters.
change in the impact position during the lifecycle and up to The present paper deals with an explanation of a RCF
the appearance of the first RCF damages are presented. The deterioration case using two independent measurements of
studies [35, 36] deal with the same analysis but with the same common crossing during its lifecycle. The
application of machine learning methods. The study [36] explanation is based on the experimental evidence with
demonstrates a significant improvement in the correlation application of modern statistical analysis and simple
of the considered indicators to the crossing’s lifetime, with mathematical modeling of wheels passing through the
the indicators taking into account all the components of the crossing. The experimental evidence is provided in two
measured accelerations, rather than only their maxima. A steps. In the first one, the location of RCF damage and its
time-spectral analysis of axle box accelerations (ABA) on casual relation to the measured accelerations and impact
track measurement trains for an evaluation of railway position is presented. In the second one, the secondary
crossing degradation is presented in [37]. The ABA mea- causes of the measured acceleration and their position
surements were verified via 3D profile measurements for distribution are explained based on the independent mea-
three conditions: nominal, degraded, and repaired. The surement of crossing geometry during its lifecycle. Such
results indicate that the axle box acceleration measurement two-step experimental explanation indicates different cau-
system is effective for the monitoring of wing and frog rail ses of the RCF damage cases of frog rails as considered in
degradation. However, the method allows to detect geo- other studies.
metrical degradation due appeared RCF/wear damages
without their prediction. A data-driven fault diagnosis
method for railway turnouts, based on microcomputer
measurements, is considered in [38]. A methodology for Statistical Study of the Primary Causes of RCF
applying the railway structural deterioration monitoring of on the Frog Rail Using Track-Side Inertial Monitoring
turnout common crossings is presented in [39]. It is based
on the geometrical and inertial measurements of track The common crossing selected for the investigation is
inspection vehicle over the lifecycle of common crossing. assembled of rails UIC60, steel grade R350HT on concrete
The study shows a significant growth of condition indica- sleepers. The switch angle is 1/12, and the common
tors during the lifecycle. However, it should be noted that crossing design type is BWG. The crossing service life
the dataset is limited with a few measurement points from ended after 315 days of operation with multiple big RCF
the inspection vehicle with low velocity that cannot ade- damages in a zone in 350–430 mm distance from the frog
quately depict real dynamic loading from operation trains. point (Fig. 2, bottom). The first visible cracks on rolling
Experimental studies of geometrical degradation of railway surface had already appeared before the 275th day of
turnout longitudinal profile are presented in [40, 41]. The operation. The operation conditions of the common
long-term trends of different time-domain and spectral crossing are 29 Mt/year, mixed traffic with maximal

123
J Fail. Anal. and Preven.

velocities of 160 km/h for passenger trains and 90 km/h for The scatter diagram shows that at all times within the
freight trains. lifecycle the points of acceleration components are dis-
The study of primary causes of the RCF damages con- tributed rather homogenously. An exception is formed by a
sists in finding the causal relationship between the group of red points with high impacts (260–300 g),
accelerations and impact positions measured throughout observed at the beginning of the lifecycle and concentrated
the lifecycle to the position of the zone subjected to RCF at on the 210–250 mm x-coordinate of the frog rail. The
the end of the lifecycle. The influence factor contact area impacts have all three high acceleration components. The
(Fig. 1) is discarded in advance and not considered further, histogram of the impact position shows the most frequently
since the investigated RCF damage was located in a zone loaded zone between 240 and 320 mm that comprises more
of relatively low curvature. that 65% of impacts. The most frequent vertical impact
The accelerations of the frog rail are measured using the loadings are those with accelerations of 130–240 g (Fig. 4,
track-side measurement system ESAH-M (Electronic top). The most of lateral impact loadings have accelera-
Analysis System of Crossing—Portable, Fig. 2, right), tions of 40–75 g (Fig. 4, middle). The longitudinal
providing 3D acceleration measurements together with accelerations are the lowest and are almost all less than
wheel positioning sensors. The latter allow to determine 50 g (Fig. 4, bottom).
train velocities and impact positions [33, 43]. The mea- Thus, the vertical impacts are much higher than the
surements were taken 11 times during the lifecycle with 4– lateral and longitudinal ones. However, the RCF damage
6 trains passing for each measurement. zone 340–430 mm is subjected to a relatively low intensity
A typical signal of the vertical rail acceleration during of loadings that is less than five times lower that the most
the passage of an ICE train (German high-speed train) with frequently loaded zone. These diagrams cannot evidentially
a velocity of 160 km/h is shown in Fig. 3. The maximal explain the causes of RCF damages. Statistical analysis
negative dynamic acceleration reaches 300 g and varies in methods of acceleration components are used to find out
a range of 100 g for different axles. The loading has a the differences in accelerations that are particular only to
distinct impact form with the maximal acceleration in the the damage zone.
first oscillation. The maximal values of the 3D acceleration A statistical analysis using the t-SNE algorithm is used
components signals are used for the following analysis. to reveal the latent relations in the statistical data. The
The analysis of the changes in maximal acceleration machine learning algorithm t-SNE is often used for the
over the lifetime of a common crossing that was carried out analysis of statistical structures. It is useful for visualiza-
in other studies [36, 44] showed no statistically significant tion of high-dimensional data [45] in low-dimensional
relation with the lifetime of the crossing. The same con- spaces with the help of dimensionality reduction. Accord-
clusion can be done from Fig. 4, in which the scatter plots ing to the similarities between neighboring points, the high-
of impact acceleration components vs. the longitudinal dimensional data are embedded in only two or three
position of the impact are depicted together with their dimensions. A plot of the data points, grouped in clusters of
frequency distributions and the indication of the RCF zone. the same similarity, is the result of the embedding. How-
The lifetime that the points correspond to is marked with ever, this process delivers a merely qualitative expression
different colors. of the similarities in the dataset. The algorithm is divided

Fig. 2 Passing of wheels from

wing rail to frog rail (left) and
the track-side inertial
measurement system (right)
[33]

123
 J Fail. Anal. and Preven.

into two calculation phases. Within the first phase, the Fig. 4 The distributions of the longitudinal coordinates of the impact c
similarity matrix is calculated with the determination of a and of the corresponding maximal vertical (top), lateral (middle) and
longitudinal (bottom) accelerations for the frog rail of a common
distance probability distribution over the pairs of dataset. crossing
There is a range of distance metrics that can be used for
this aim: Chebyshev, Mahalanobis metrics, etc. The
Euclidean metrics is used for the present study. To control
the value of similarity, a parameter of perplexity is intro- points, except for some more densely located red points
duced that is the effective number of local neighbors of (Fig. 5, left) corresponding to the high loading on the front
each point. Within the second phase, similar probability zone of the frog rail (Fig. 4). The post-processing analysis
distributions over the points are iteratively projected into of the points of the small cluster showed that almost all the
the low-dimensional plot. An iterative optimization pro- points have their longitudinal position of the impact at an
cedure is used to minimize the Kullback–Leibler x-coordinate higher than 340 mm and show vertical
divergence between a t-distribution in the low-dimensional accelerations more than 250 g. Thus, the results of t-SNE
space and a Gaussian distribution in the high-dimensional analysis highlight the existence of a separate group of high
space [46]. impacts that are located in the future RCF damage zone,
Figure 5 presents the results of the t-SNE analysis for but are not visible in the overall impact distributions
the same dataset that is used for Fig. 4. The impact (Fig. 4).
accelerations in the lateral and longitudinal directions are To explicitly outline the highly loaded zone in the
used additionally in this analysis. The same data points as impact position distribution histogram, it is divided into
in Fig. 4 are presented in the two-dimensional space of the different acceleration classes. Figure 6 shows the overall
first two principal components Pc1 and Pc2 with three impact distribution histogram and four other histograms,
different colors according to the lifetime. Remarkable in depicting the distributions of the classified impacts, using a
the t-SNE plot are two isolated clusters. The smaller cluster class width of 100 g.
comprises about 10% of the points that mostly belong to The diagram shows that the distribution of the impact
the middle or the late lifetime of the crossing. The bigger position for 0–100 g and 100–200 g show the most fre-
cluster shows relatively uniformly distributed lifetime quent loading in the same zone of the frog rail as the

Fig. 3 The signal of the vertical rail acceleration on the frog rail of a common crossing for an ICE train at 160 km/h (top: whole train, bottom:
two axles)

123
J Fail. Anal. and Preven.

123
 J Fail. Anal. and Preven.

Fig. 5 The principal

components of t-SNE analysis
for acceleration components and
impact distribution statistics

overall statistic shows. At the same time, both classes crossing’s lifetime. This proves that the high impacts were
together contain more than a half of the overall amount of not caused by the RCF damages themselves. The presented
data points. The distribution of the impact position for the statistical analysis reveals the primary causes of the RFC
acceleration level 200–300 g already differs from the first damages—a small group of high loadings, but it cannot
two distributions with a much wider zone of impacts. explain for which reason those loadings appear.
Finally, the distribution of impact positions for the highest
acceleration level of more than 300 g is quite different and
indicates the group of highest loadings in the frog rail zone Studying the Root Causes of RCF on Frog Rails Using
with an x-coordinate of more than 325 mm. However, the Track-Side Geometrical Monitoring
number of impacts in the group is very low. A more
detailed classification of the acceleration, using a more The study of the causes of the loadings or the root cause of
narrow class width, would provide a better isolation of the RCF damage on frog rails is performed using simple
highly loaded impact zone that corresponds to the impact kinematical modeling of the wheel and frog/wing rail
cluster in the t-SNE analysis. contact during the wheel’s passage over the common
The results of the analysis clearly show that the RCF crossing. There are three main possible secondary causes of
damages appear not in the most frequently loaded zone of dynamic loadings: wing/frog rail wear, wheel profile wear,
the frog rail, but in the zone of about 10% of impacts that and the wheel’s lateral position (Fig. 1). The geometric
show a high vertical loading. Thereby, the horizontal and irregularity of the vertical wheel trajectory, depending on
lateral acceleration components in the damage zone have those three factors, is determined by kinematical modeling.
no special features. Moreover, the mean lateral and longi- The dynamic loading is assumed to be proportional to the
tudinal accelerations in the damage zone are even less than angle in the trajectory kink point. The usage of a simple
those outside. The mechanical interpretation of the fact approach like this, instead of dynamic modeling, is justified
corresponds to the common concept of fatigue mechanics, by the possibilities of a following interpretation and prac-
where the level of loading is inversely and nonlinearly tical application.
related to the number of loading cycles. The study [44] The wing/frog rail wear over the overall lifecycle of
estimates the influence of one impact with an acceleration the crossing is known owing to measurements using the
of more than 250 g to be equivalent to more than one surface scanning device SCORPION [47] (Fig. 7). The
thousand impacts of 50 g. The statistical analysis allows to device is based on laser scanning technique and is
estimate also the time aspects of the primary causes and designed for periodic measurements of rail and turnout
their causality with RFC damages. Most of the high-impact profiles, providing a measurement accuracy of
loadings of the damage zone occurred in the middle of the ± 0.1 mm.

123
J Fail. Anal. and Preven.

Fig. 6 Distribution of the maximal vertical accelerations along the frog nose for different acceleration levels

corresponds to the admissible wear range. The wheel pro-

file is variated in 15 steps from a new S1002 profile (R00),
over a wheel profile subjected to average wear (R08), to a
maximum worn-out profile (R14) (Fig. 9).
The lateral position of the wheel depends on many
factors (Fig. 10, top): wheel flange wear, flange gap
between frog and wing rail, track gauge, and distance
between inner wheel flanges. The check rail position and
its wear limit the lateral position of the wheel relatively to
the frog rail. The lateral position of the wheel is variated in
five steps from the most close to the frog rail (L0) to the
farthest possible from it (L4). Due to the wheel conicity, as
well as the radii of the wing and frog rails, the wheel’s
lateral position influences its vertical position as shown in
Fig. 7 Surface scanning device SCORPION [47] for common
crossings Fig. 10, bottom.
The vertical displacement of the wheel from its original
position can be calculated, delivering the quasistatical
The results of the measurement of the wing and frog rail vertical trajectories of the wheel contact point on the wing
rolling surfaces with wear cross sections during the life- zc;wr ðx; ylat;i Þ and the frog rail zc;fr ðx; ylat;i Þ, respectively. The
cycle are shown in Fig. 8. trajectories depend on the case of the lateral wheel position
Wheel profile wear and the wheel’s lateral positions are ylat;i . The overall wheel trajectory is found as the maximum
the unknown factors in this study. Therefore, multivariate line of the both trajectories taking into account the vertical
calculations of the wheel trajectory are done with a uniform difference between the contact points on wheel profile
distribution of both unknown factors. The distribution Dzc;pr ðx; ylat;i Þ:

123
 J Fail. Anal. and Preven.

Fig. 8 The surface of wing and frog rail at the end of the lifecycle (top) and their cross sections for x = 200 mm during the lifecycle (bottom)

Fig. 9 The cross section of wheel profiles for new, average, and fully worn-out wheels


  z ðx; ylat;iÞ position, kink point, and kink angle u. The left column of
zc x; ylat;i ¼ max c;wr ðEq 1Þ
zc;fr x; ylat;i þ Dzc;pr ðx; ylat;i Þ Fig. 11 corresponds to the new crossing, the right one—to
the crossing at the end of its lifecycle. The rows are the
Figure 11 shows the trajectories for the main different variations of the maximal and minimal wheel
combinations of rail wear, wheel wear and the lateral wear and lateral position.

123
J Fail. Anal. and Preven.

Fig. 10 The lateral position of a wheelset in a common crossing (top) and influence of the wheel position on the vertical displacement of the
wheel (bottom)

The analysis of the influence factors on the kink angle influence of each factor, the visualization is divided into
shows the following: The increase in the frog/wing rail three plots for each factor. The plots clearly show the same
wear over the lifecycle of the crossing equivocally influ- statement as in the statistical analysis: There are two
ences the angle. Looking at the trajectories of the new clusters of impact points. The smaller one is located far
wheel profile (R0), the angle u somewhat decreases with from the frog point and has a few points with a much
the rail wear during the crossing’s lifetime. However, for higher kink angle than the majority of points. The plots in
the worn-out wheel profile, an increase in the angle of up to Fig. 12 can very well explain the reasons for the appear-
0.3° can be observed for the worn frog/wing rails at the end ance of those big angles. The bottom plot in Fig. 12 shows
of the crossing’s lifetime. that the smaller cluster of points is formed with the wheel
The increase in the lateral wheel position from the profiles R09–R14. The absence of the profiles can exclude
closest to the farthest from the frog rail causes a significant the cluster. The middle plot in Fig. 12 demonstrates that
growth of the kink angle of up to 2.5 times, independently the point clouds in the smaller cluster show different
of the condition of the common crossing. An even higher inclinations relative to the x-axis depending on the wheel
increase in the kink angle of up to three times is caused by position L0–L4. Therefore, the wheel position factor can-
the variation of the wheel profile. not exclude the small cluster but can significantly reduce
The worn wheel and far wheel position summarize their the impact angles. The top plot in Fig. 12 shows that the
influence on the kink angle that can reach up to 4.5 times high angles appear for both the beginning and the end of
more than in the best case of the factors. The worst case is the lifecycle. The rail wear has no influence neither on the
the worn wheel, rail, and the farthest wheel position, kink position nor on its angle.
showing a kink angle of u ¼ 2:3 . The distributions of the kink angle and position clearly
The influence of the considered factors on the longitu- show a data structure similar to that obtained by the sta-
dinal distribution of the kink point position and the kink tistical analysis of the inertial measurements. However,
angle is visualized in Fig. 12. The number of factor com- there are also significant differences in the modes of the
binations is presented with 450 points. To make clear the distributions. The position of the major point cluster can be

123
 J Fail. Anal. and Preven.

Fig. 11 Wheel trajectory on frog and wing rail for the main combinations of rail wear, wheel wear, and lateral position of the wheel

explained by the fact that the wheel profiles and their lat- mean values of those distributions. The zone of maximal
eral positions were assumed to be distributed uniformly. kink angles lies in a range of the x-coordinate of 320–
However, the minor point cluster has no relation to the 350 mm, whereas the maximal points showing the

123
J Fail. Anal. and Preven.

Fig. 12 Influence of rail wear (top), lateral position of the wheel (middle) and wheel wear (bottom) on the kink position and the kink angle of the
vertical wheel trajectory

maximal accelerations lie between 350 and 400 mm DlC ¼ R  tgðu=2Þ; ðEq 2Þ
(Fig. 6). The reasons for this discrepancy can be explained rffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffi me
in Fig. 13. Due to the wheel radius, the wheel and rail Pim ¼ u  V t kH m e 1 þ ; ðEq 3Þ
contact point IC is shifted to a position behind the kink M
point by the distance DlC ; which depends on the kink angle where u—wheel trajectory angle, Vt —train velocity, kH —
u and the wheel radius. Additionally, the impact point of contact stiffness, me —equivalent track mass and M—train
the maximal loading IIm is shifted due to the wave char- unsprung mass.
acter of the loading, as depicted in the path acceleration Figure 14 presents the diagram of the impact loading vs
diagram in Fig. 13. Both shifting distances can be summed the impact position. The maximal impact loadings more
up to 20–50 mm. than 160 kN are located in a zone of the frog surface at a
The vertical first impact loadings Pim and the distance longitudinal coordinate x of 340–410 mm. This corre-
DlC are determined by the formulas [48]: sponds to the location of the observed RCF damage

123
 J Fail. Anal. and Preven.

(Fig. 4) and the group of maximal vertical accelerations in Studying the case of rail contact fatigue damage shows
Fig. 6. The group of maximal impact loadings contains that the primary cause of the RCF is a concentration of a
about 9% of the overall amount of impacts. small cluster of high vertical loading impacts in the zone of
the frog rail with x-coordinates than 340–430 mm. The
small cluster of the vertical accelerations is the only feature
Discussion of the damage zone that differs it from the other contact
zone of the frog rail. The vertical and longitudinal accel-
The present experimental study is based on the monitoring erations in the future damage zone have no special features
of impact accelerations and rail wear on a common and are on average even less than outside the damage zone.
crossing during its lifecycle. Different to the other mostly Most of the high impacts appear in the middle and late
theoretical studies, the research results present direct lifetime of the crossing. This indicates the causality of RCF
experimental evidence of factor influence and in their damages as a result of high-impact loadings and not vice
causality. Both two independent measurements of impact versa.
accelerations and rail wear lead to the same conclusions of The most frequently loaded zone between 240 and
the RCF causes but in the different level of factor 320 mm indicates no RCF damages. Thereby, the frog rail
influence. maximal curvature in the most frequently loaded zone is
about 1.6 times higher than that in the damage zone. Thus,
the dominating factor is not condition nor the number of
accumulated impacts but the height of the impacts. This
fact does not comply with the results of the other model-
based studies [4, 28], where the maximal RCF damage
location is predicted to be in the same zone as the maximal
impact occurrence. Moreover, the high influence of the
tangential dynamic contact forces on the RCF as stated in
[4, 29] cannot be confirmed within the present experi-
mental measurements. The reasons why the theoretical
models of the other studies do not with the present
experimental measurements can be different. The models
are validated with fragmental experimental data not cov-
ering the overall lifecycle. The validation data are for
different crossing with different operational conditions
than in the present study. Though the similar as in the
present paper accumulation of high-impact angles was
found in [28], its influence was leveled with other more
weighting factors depending on the number of accumulated
impacts and contact shear forces.
The secondary maintenance causes of the high loadings
are wheel profile wear, lateral position of the wheel, and
Fig. 13 Determining the impact position and impact velocity wear of the crossing rails. The most significant cause is the

Fig. 14 Dynamic loading

depending on the longitudinal
position of the impact

123
J Fail. Anal. and Preven.

wheel wear. A limitation of the heavily worn wheel profiles 4. Frog/wing rail vertical wear plays a secondary role in
higher than R9 could completely eliminate the high-impact appearing the early RCF failures for the considered
loading and would avoid the appearance of RCF damages crossing case.
in the zone of the frog rail behind 340 mm. The lateral
The presented study shows that the main relations
wheel position plays an auxiliary role in the formation of
between crossing RCF and the influencing factors can be
the high-impact loadings. The factor cannot prevent the
determined using simplest kinematic modeling. However,
appearing of high loadings, but it can significantly reduce
it is not able to reflect all possible dynamic interactions and
their values. The wear of frog and wing rail ambiguously
the impact position. In some cases of the wheel trajectory,
influences the loadings. The high-impact loadings appear
the maximal loadings could be caused not only by the kink
both at the beginning of the crossing’s lifecycle and at its
angle, but also by the influence of the overall geometry of
end. In some cases, rail wear can even reduce the impact
the wing and frog rails. This can be observed analyzing the
loadings. However, on average, the rail wear facilitates
acceleration measurements. Thus, a dynamic and contact
increasing the impact loadings. The low influence of the
modeling could be performed in subsequent studies.
rail wear could be explained with its relative low value
within the short lifecycle of the crossing compared to the Acknowledgments The authors acknowledge the German Railway
initial structural geometry of crossing rails. Company (DB Systemtechnik GmbH) and WITT Elektronik GmbH
The secondary causes can be more or less influenced in for their financial supports.
the course of maintenance. The potential consists in the
limitation of the admissible wear in the railway standards.
The wheel wear is the most significant factor, but it can References
only be limited to a certain degree due to cost reasons. LCC
1. A. Zoll, Werkstoffauswahl für Weichenhetzstücke durch Prüfs-
studies of the topic could demonstrate the expediency of a tandversuche (Material selection for point frog points with test
limitation of the admissible wheel wear. The lateral posi- bench tests.), Ph.D. thesis, TU Berlin, Shaker Verlag Aachen,
tion of the wheel, even if exerting only a minor influence, 2016 (in German)
2. C. Letot, P. Dersin, M. Pugnaioni, P. Dehombreux, G. Fleurquin,
could be a good way for preventing RCF damages. It can
C. Douziech, P. La-Cascia, A data driven degradation-based
be influenced by the control of the check/stock rail position model for the maintenance of turnouts: a case study. IFAC
and wear, so that the wheel’s lateral position could be set PapersOnLine 28(21), 958–963 (2018)
close to the frog rail. However, a change in the mainte- 3. B. Lichtberger, Track Compendium: Formation, Permanent Way,
Maintenance, Economics (Eurailpress, Hamburg, 2005)
nance rules in this direction would also cause higher
4. L. Xin, Long-term behaviour of railway crossings: wheel–rail
maintenance costs. interaction and rail fatigue life prediction, Ph.D. thesis, TU Delft,
Delft, 2017
5. U. Gerber, A. Zoll, W. Fengler, Verschleiß und Fahrflächen-
ermüdung an Weichen mit starrer Herzstückspitze (Wear and
Conclusion and Subsequent Studies
rolling contact fatigue on common crossings of railway turnouts).
ETR Eisenbahntech. Rundsch. 1, 36–41 (2015). (in German)
The paper has presented an experimental study of the pri- 6. M.C. Burstow, Whole Life Rail Model Application and Devel-
mary and secondary causes of RCF damages on a common opment for RSSB—Continued Development of an RCF Damage
Parameter (Rail Standards and Safety Board, London, 2004)
crossing. An analysis of the influence mechanisms has been
7. J. Wiedorn, W. Daves, U. Ossberger, H. Ossberger, M. Pletz,
performed. The study is based on the acceleration and Finite element model for predicting the initiation of subsurface
geometry monitoring results of a common crossing over damage in railway crossings—a parametric study. Proc. Inst.
the overall lifecycle. The primary causes of RCF were Mech. Eng. Part F J. Rail Rapid Transit 233(6), 614–628 (2019)
8. V. Markine, X. Liu, A. Mashal, Y. Ma, Analysis and improve-
determined using a statistical analysis. The influence of the
ment of railway crossing performance using numerical and
secondary causes on the primary ones is studied using a experimental approach: application to 1:9 double crossovers.
simple kinematic model. The model enables a clear inter- Dyn. Veh. Road Track 2, 717–722 (2018)
pretation of the wheel trajectory and the resulting impact 9. R. Skrypnyk, M. Ekh, J.C.O. Nielsen, B.A. Pålsson, Prediction of
plastic deformation and wear in railway crossings—comparing
loading. The main results of the experimental research can
the performance of two rail steel grades. Wear 428, 302–314
be highlighted in the following four theses: (2019)
10. J. Xiao, F. Zhang, L. Qian, Numerical simulation of stress and
1. A minor cluster of high impacts is located outside of deformation in a railway crossing. Eng. Fail. Anal. 18(8), 2296–
the overall impact distribution on frog rail. 2304 (2011)
2. The minor cluster location corresponds to location of 11. Z. Wei, A. Núñez, A. Boogaard, R. Dollevoet, Z. Li, Method for
later appeared RCF failures. evaluating the performance of railway crossing rails after long-
term service. Tribol. Int. 123, 337–348 (2018)
3. The main causes of the high-impact cluster are wheel 12. L. Xin, V. Markine, I. Shevtsov, Numerical analysis of rolling
wear and its lateral position. contact fatigue crack initiation and fatigue life prediction of the

123
 J Fail. Anal. and Preven.

railway crossing, in CM 2015: 10th International Conference on 31. Q. Mahboob, E. Zio, Handbook of RAMS in Railway Systems:
Contact Mechanics of Wheel/Rail Systems, 2015 Theory and Practice (CRC Press, Boca Raton, 2018)
13. X. Ma, P. Wang, J. Xu, R. Chen, J. Wang, Assessment of non- 32. S. Panchenko, I. Siroklyn, A. Lapko, A. Kameniev, D. Buss,
Hertzian wheel–rail contact models for numerical simulation of Critical failures of turnouts: expert approach. Proc. Comput. Sci.
rail damages in switch panel of railway turnout. Wear 102912, 149, 422–429 (2019)
432–433 (2019) 33. U. Gerber, W. Fengler, A. Zoll, Das Messsystem ESAH-M (The
14. A. Ekberg, E. Kabo, H. Andersson, An engineering model for measurement system ESAH-M). EIK Eisenb.ing. Kal., 2016, 49–
prediction of rolling contact fatigue of railway wheels. Fatigue 62 (2016) (in German)
Fract. Eng. Mater. Structu. 25(10), 899–909 (2002) 34. U. Gerber, W. Fengler, A. Zoll, Vehicle-based assessment of frog
15. Y. Jiang, H. Sehitoglu, A model for rolling contact failure. Wear wear: modern measuring system allow new inspection concept.
224, 38–49 (1999) Przeglad komunikacyjny 7, 26–27 (2015)
16. L. Huang, H.S. Lai, K.L. Liu, A novel method to predict the low- 35. M. Sysyn, D. Gruen, U. Gerber, O. Nabochenko, V. Kovalchuk,
cycle fatigue life. J. Fail. Anal. Prevent. 18(6), 1484–1489 (2018) Turnout monitoring with vehicle based inertial measurements on
17. C. Wan, V. Markine, I. Shevtsov, Analysis of train/turnout ver- operational trains: a machine learning approach. Commun. Sci.
tical interaction using a fast numerical model and validation of Lett. Univ. Zilina 21(1), 42–48 (2019)
that model. Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit 36. M. Sysyn, U. Gerber, O. Nabochenko, Y. Li, V. Kovalchuk,
228(7), 730–743 (2014) Indicators for common crossing structural health monitoring with
18. E. Kassa, C. Andersson, J.C.O. Nielsen, Simulation of dynamic track-side inertial measurements. Acta Polytech. 59(2), 170–181
interaction between train and railway turnout. Veh. Syst. Dyn. (2019)
44(3), 247–258 (2006) 37. Z. Wei, A. Núñez, Z. Li, R. Dollevoet, Evaluating degradation at
19. E. Kassa, J.C.O. Nielsen, Dynamic interaction between train and railway crossings using axle box acceleration measurements.
railway turnout: full-scale test and validation of simulation Sensors (Switzerland) 17(10), 2236 (2017)
models. Veh. Syst. Dyn. 46, 521–534 (2008) 38. D. Ou, R. Xue, K. Cui, A data-driven fault diagnosis method for
20. Y.Q. Sun, C. Cole, M. McClanachan, The calculation of wheel railway turnouts. Transp. Res. Rec. 2673, 448–457 (2019)
impact force due to the interaction between vehicle and a turnout. 39. S. Kaewunruen, Monitoring structural deterioration of railway
Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit 224, 391–403 turnout systems via dynamic wheel/rail interaction. Case Stud.
(2010) Nondestruct. Test. Eval. 1, 19–24 (2014)
21. X. Shu, N. Wilson, C. Sasaoka, J. Elkins, Development of a real- 40. I.A. Khouy, P.-O. Larsson-Kråik, A. Nissen, J. Lundberg, U.
time wheel/rail contact model in NUCARS and application to Kumar, Geometrical degradation of railway turnouts: a case study
diamond crossing and turnout design simulations. Veh. Syst. from a Swedish heavy haul railroad. Proc. Inst. Mech. Eng. Part F
Dyn. 44(1), 251–260 (2006) J. Rail Rapid Transit 228(6), 611–619 (2014)
22. A. Blanco-Saura, J. Velarte-González, F. Ribes-Llario, J. Real- 41. N. Minbashi, M. Bagheri, A. Golroo, I. Arastehkhouy, Geomet-
Herráiz, Study of the dynamic vehicle-track interaction in a rical degradation analysis of railway turnouts using power
railway turnout. Multibody Syst. Dyn. 43(1), 21–36 (2018) spectral density. J. Transp. Eng. Part A Syst. 143(8), 04017035
23. V. Salajka, M. Smolka, J. Kala, O. Plášek, Dynamical response of (2017)
railway switches and crossings. MATEC Web Conf. 107, 00018 42. A. Cornish, R.A. Smith, J. Dear, Monitoring of strain of in-ser-
(2017) vice railway switch rails through field experimentation. Proc.
24. D. Jiang, D. Grin, Dynamical loading on a turnout with a rigid Inst. Mech. Eng. Part F J. Rail Rapid Transit 230(5), 1429–1439
frog, in Civil-Comp Proceedings, 110, 2016 (2016)
25. E. Kassa, J.C.O. Nielsen, Stochastic analysis of dynamic inter- 43. http://witt-solutions.de/
action between train and railway turnout. Veh. Syst. Dyn. 46(5), 44. M. Sysyn, U. Gerber, O. Nabochenko, V. Kovalchuk, Common
429–449 (2008) crossing fault prediction with track based inertial measurements:
26. C. Wan, V. Markine, Parametric study of wheel transition at statistical vs mechanical approach. Pollack Period. (2019).
railway crossings. Veh. Syst. Dyn. 53(12), 1876–1901 (2015) https://doi.org/10.1556/606.2019.14.x.x
27. A. Pålsson, J. Nielsen, Wheel–rail interaction and damage in 45. M. Wattenberg, F. Viégas, I. Johnson, How to use t-SNE effec-
switches and crossings. Veh. Syst. Dyn. 50(1), 43–58 (2012) tively. Distill 1(10) (2016). https://distill.pub/2016/misread-tsne/.
28. A. Pålsson, Optimisation of railway crossing geometry consid- Accessed 15 July 2019
ering a representative set of wheel profiles. Veh. Syst. Dyn. 53(2), 46. V. Maaten, G. Hinton, Visualizing data using t-SNE. J. Mach.
274–301 (2015) Learn. Res. 9, 2579–2605 (2008)
29. Z. Wei, C. Shen, Z. Li, R. Dollevoet, Wheel–rail impact at 47. http://www.graw.com/downloadprod/product-leaflet-scorpion.pdf
crossings: relating dynamic frictional contact to degradation. J. 48. M. Verigo, A. Kogan, Interaction of Track and Rolling Stock
Comput. Nonlinear Dyn. 12(4), 041016 (2017) (Transport Publishing House, Moscow, 1986)
30. N.O. Attoh-Okine, Big Data and Differential Privacy: Analysis
Strategies for Railway Track Engineering (Wiley, New York, Publisher’s Note Springer Nature remains neutral with regard to
2017) jurisdictional claims in published maps and institutional affiliations.

123