Engineering Failure Analysis 135 (2022) 106142

Contents lists available at ScienceDirect

Engineering Failure Analysis

journal homepage: www.elsevier.com/locate/engfailanal

Estimation of residual stress distribution in railway axles

Pavol Dlhý a, b, Jan Poduška a, Pavel Pokorný a, Michal Jambor a, Luboš Náhlík a, *,
Daniel Kajánek c, Rostislav Fajkoš d, Pavel Hutař a
a
Institute of Physics of Materials Czech Academy of Sciences, v.v.i., Žižkova 513/22, 616 00 Brno, Czech Republic
b
CEITEC - Brno University of Technology, Purkyňova 656/123, 612 00 Brno, Czech Republic
c
Research Centre of the University of Žilina, University of Žilina, Univerzitná 8215/1, 010 26 Žilina, Slovakia
d
BONATRANS GROUP, a. s., Revoluční 1234, 735 94 Bohumín, Czech Republic

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

Keywords: The manufacturing process of railway axles usually includes axle surface treatment to induce
Residual stress compressive residual stress on the axle surface to increase impact resistance and fatigue lifetime.
X-ray diffraction A proper determination of residual stress enables to obtain correct input data for other proced­
Railway axles
ures, e.g., optimization of specialized treatment or lifetime estimation, which sets safe, but not
Residual lifetime prediction
Neutron diffraction
unnecessary frequent, maintenance intervals of the axles. The most common methods for residual
Numerical modelling stress determination used by R&D centers of axle manufacturers are the hole drilling method and
X-ray diffraction. However, we can use these methods for the determination of surface or close-to-
surface residual stress only. In the case of large components like railway axles, it is essential to
have information about the residual stress in the whole cross-section, not only from the axle
surface, especially if the residual stress is developed by induction hardening, which can influence
the residual stress distribution in a considerable depth.
The work presented in this paper aims to develop a reliable methodology for determining
residual stress in the whole cross-section of a railway axle with reasonable equipment prices or
commonly used equipment. Two presented methods follow the defined methodology. They
combine well-known destructive methods (layer removal and sectioning methods) with X-ray
diffraction and numerical simulations to evaluate correct residual stress distribution. Both ways
are applied to the case of an induction-hardened railway axle from the EA4T steel. Results of both
methods are then compared to results obtained by neutron diffraction technique and other
experimental methods to validate the plausibility of the proposed scenarios.

1. Introduction

Examining and improving the fatigue resistance of railway axles has been an essential issue in the industry since the time of Wöhler
[1]. The safety and durability of axles during operation are the primary targets of axle manufacturers. These properties are tightly
linked to the resistance against the onset of defects and subsequent crack propagation. Also, the better the resistance against defects is,
the longer the axle can stay in operation without any problems, which lowers its maintenance costs. Besides safe and reliable axles, the
low maintenance costs and reasonable inspection intervals are significant for the success of the railway axle on the global market.
Improving the resistance against external impacts is possible by the traditional procedures of surface quenching [2,3] or deep

* Corresponding author.
E-mail address: nahlik@ipm.cz (L. Náhlík).

https://doi.org/10.1016/j.engfailanal.2022.106142
Received 21 October 2021; Received in revised form 26 January 2022; Accepted 8 February 2022
Available online 12 February 2022
1350-6307/© 2022 Elsevier Ltd. All rights reserved.
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

rolling [4]. They are accompanied by the formation of residual stress under the axle surface. Residual stress is a crucial component
load, especially in thermo-mechanically treated components [5–8]. Nowadays, technologists are looking for a way to use surface
treatment to increase the surface hardness and the overall fatigue resistance of the axle because a suitable compressive residual stress
distribution can prevent fatigue cracks from growing. Hence it prolongs the residual fatigue life of the axle [9]. One of the promising
technologies for treating high-quality axles is induction hardening, which we can control quite well. This technology can induce high
residual stresses (commonly in hundreds of MPa) in the hardened part. The choice of the parameters of the induction hardening is a
crucial issue because the resultant microstructure and residual stress profile should guarantee both sufficiently hard, impact-resistant
surface and excellent fatigue properties, i.e., high resistance to fatigue crack initiation and growth. This paper is devoted to the
possibilities of precise determination of residual stress in railway axles, which is always the first step in further optimizing the residual
stress distribution to increase the fatigue lifetime.
We can use various surface treatment techniques to induce compressive residual stress at the surface of a railway axle. The most
common are the shot peening [10], grinding [11], deep rolling [3], and quenching [12]. However, the depth of induced compressive
residual stress by these standard techniques is relatively low (e.g., shot peening less than 0.5 mm [13], deep rolling up to 4 mm [3,14]).
On the other hand, the induction hardening can induce compressive residual stress up to 20 mm under the surface [15–18] and thus
significantly influence the residual stress distribution in the whole cross-section. Another benefit of induction hardening is the high
hardness of the railway axle surface. We know that a ballast impact can cause a defect during train operation where a fatigue crack then
initiates [19–21]. Therefore, high hardness and high compressive residual stress near the surface of railway axles may prevent fatigue
failure from surface defects, especially in the case of high-speed trains [2,3].
The investigation of residual stress in railway axles is usually done only on the surface of the component [2,11,13,14,22–26]. It is
caused by the technical complexity and considerable residual stress costs if investigated in depth. For a railway axle, the penetration
depth of the measurement should ideally be around at least 60 mm for hollow axles up to about 90 mm for solid axles. Even though the
essential compressive residual stress may be located only in a shallow depth, from a physical point of view, the residual stress dis­
tribution is balanced through the whole body (the sum of forces and moments must be zero). Therefore, for example, compressive
residual stress close to the free surface induces tensile stress further from the surface and closer to the axle core [18,27,28]. Deter­
mining residual stresses in the depth of a few millimeters can provide only partial information about the overall residual stress dis­
tribution and, consequently, about the natural stress gradient influencing the residual fatigue lifetime.
Many different methods have been developed for residual stress determination. They differ in the basic principle and the possible
penetration depth (meaning the distance under the surface at which the stress can be effectively measured). Fig. 1 shows the list of the
most common available residual stress measurement methods and their penetration depth. The approximate penetration depth in­
terval singles out those potentially suitable methods for the application on railway axles.
One of the most common residual stress measurement techniques is the ultrasonic method. This non-destructive method can
measure a through-thickness average of residual stress in a sample [33]. Although the ultrasonic method allows measuring residual

Fig. 1. List of common residual stress measurement methods [29–32].

2
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

stresses in quite significant depths in some configurations, it is unsuitable for railway axles due to the averaging.
The neutron diffraction is the most powerful non-destructive method in terms of penetration depth. However, there are only some
places in the world [34] where this method can be performed in the required depth and even fewer where we can measure the whole
axle in one piece. The preparation for the measurement is time-consuming, and the measurement itself is very expensive, which makes
the method not economically viable for a repeated residual stress measurement during the development of railway axles. There are no
other non-destructive techniques capable of penetrating the integral portion of the railway axle cross-section. The destructive methods
capable of deeper penetration, the deep hole drilling method, the slitting (crack compliance) method, the contour method, the layer
removal, or the sectioning method seem appropriate. All of these methods are based on stress relaxation [30,35]. It means that some
part of the examined body is removed while deflection and/or relaxed strains are recorded. The residual stress is then calculated.
The deep hole drilling method requires drilling the reference hole in the studied sample and trepanning the ring core [36]. Residual
stress is then evaluated from deflection measured inside of the reference hole before and after trepanning to get stresses in two di­
rections, perpendicular to the drilling direction. This method requires specific hardware for manufacturing tiny hole as well as for high
accuracy deflection measurement. Such hardware is not commonly available in R&D centers. Using the slitting method, a part of the
body under study is cut off, which causes the body to deform. The strains and/or deflection are recorded for the residual stress
evaluation [37]. The residual stress is evaluated as the average value per cut increment in the cut-off part in the direction perpendicular
to the cut. The tricky part of this method lies in the choice of the correct formula for residual stress calculation. Such procedure de­
scribes measured specimen geometry, the position of strain gauges, plasticity effects, etc. The contour method of residual stress
evaluation is based on Bueckner’s superposition principle [38]. It means that the measured sample is cut into two parts along the
investigated plane, and a high accuracy displacement sensor records the deflection of newly created surface. Numerical simulations
then seek the residual stress distribution. Like the deep hole drilling method, the hardware necessary for accurate measurement is rare
in R&D centers. The original principle of the layer removal method [39] is subsequent removing of layers of material from one side of
the measured part, while recording changes in strain by a strain gauge glued to another location on the piece (for flat details, it is
usually the opposite side to the one from which material is removed). The sectioning method [40] is very similar in principle to the
layer removal method, but the part is cut into larger sections instead of eliminating layers. The removal or cutting of the material
continues until no further deflection is present. Residual stress is then calculated from the measured deformation using the theory of
elasticity. Both methods typically deal with strain gauge measurement and multiple machining operations. Therefore, methods are
time-consuming and do not fit well in the optimization process for new designs. From the above, the most promising method for
measurements on railway axles is the layer removal method, but only in its modification published by Moore and Evans [41]. They
modified the layer removal method for cylindrical bodies by combining the classic layer removal method with X-ray diffraction
measurements of residual stress. They also devised a correction calculation based on the elasticity theory that enables obtaining the
correct residual stress distribution in a cylindrical part. It is the only available method directly applicable to railway axles. It is
seriously time-consuming due to the large number of machining operations needed in the experimental procedure.
This work addresses the lack of precise yet time- and cost-effective residual stress determination methods for railway axles. The use
of two methods is presented here that combine experimental measurement and numerical simulations. Both of them were designed to
be economically viable in the conditions of an axle manufacturing company and possibly to be carried out using equipment commonly
available in the industry. The first method is not entirely new – it is the layer removal method published by Moore and Evans [41],
which is modified by the use of numerical simulations for the residual stress evaluation. For the purpose of this article, it is called the
layer-removal-based method so as not to confuse it with the original. The other method was developed for this work as a less time-
consuming alternative. It is called the sectioning-based method because it is loosely based on the sectioning method. The methods
both employ X-ray diffraction to measure residual stress in specimens prepared from a real axle. Calculations then correct the
measured residual stresses to account for the stress redistribution during specimen preparations. Numerical simulations by finite
element method (FEM) are used for the calculations. These two methods are demonstrated here by investigations of residual stress
done on the same railway axle.
The results obtained by the proposed approaches were also compared to referential experiments by the neutron diffraction, the
slitting method, and the X-ray diffraction carried out at other facilities to verify that the proposed scenarios are plausible.

2. Materials and methods

The railway axle examined in this work was made by the BONATRANS GROUP from the EA4T (25CrMo4) steel grade, standardized
steel to produce railway axles used in Europe [42]. The total length of the investigated railway axle was 2180 mm. The outer diameter
in the studied area was 168 mm, and the hole inside the axle had a diameter of 70 mm (see Fig. 2). The axle was forged, machined, and
then treated by induction hardening and final turning to precise dimensions.

Fig. 2. Scheme of considered railway axle with highlighted segments for residual stress measurements.

3
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

We applied two methods for residual stress determination on one railway axle: the layer-removal-based method (LRBM) and the
sectioning-based method (SBM). Both follow the same general methodology that is schematically illustrated by a diagram in Fig. 3. The
first step is the design of the method, followed by destructive specimen preparation and X-ray diffraction measurement of residual
stress on the surface of the specimen. An integral part of both methods is a numerical simulation of the experiments by FEM that
corrects the measured values and thus provides the accurate, original residual stress distribution in part before any machining.
The principles and details of the methods are described in the following sections.

2.1. The Xray diffraction

The Xray diffraction (XRD) itself was not listed among the methods suitable for residual stress determination in large depths.
However, despite its low penetration depth, it can be used for a sizable part when combined with appropriate specimen preparation
and, more importantly, the essential numerical procedure for analyzing and correcting the measured data.
XRD was used to measure residual stress at the surface of the railway axle when performing both the LRBM and the SBM – step 2 in
Fig. 3. The setting of the method follows. We used the Stresstech Xstress 3000 robot with Chromium (Cr) radiation, current of 9 mA,
and voltage of 30 kV for X-ray diffraction. The X-Ray beam was collimated to a diameter of 2 mm. The measurements were carried out
using 30◦ detector with four inclinations between ±40◦ . Diffraction peak from {2 1 1} planes at 2θ = 156.4◦ was used for the mea­
surements. Young’s modulus E = 205 GPa and Poisson ratio ν = 0.3 were used for stress calculation [17]. Surfaces were cleaned and
electrolytically etched by KRISTALL 650 machine before every measurement to remove RS introduced by machining. The depth of
etching was set to 0.3 mm (the choice of this parameter will be discussed later).

Fig. 3. Scheme of the developed optimization process of residual stresses determination in sizeable components.

4
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Before the actual measurement of stress by XRD, the surface has to be prepared by etching. It is desirable to etch such an amount of
material to get rid of subsurface residual stress caused by machining. We tested different etching depths on the studied axle. In Fig. 4,
residual stress values obtained by XRD for different etched depths are plotted. After etching 0.1 mm of the material, the residual stress
results are still influenced by machining. With 0.2 mm of the etched depth, there is still some observable influence in the axial residual
stress values. For depths of 0.3 and 0.4 mm, most of the results fall within the uncertainty range of the method. Therefore, we set the
depth of etching to 0.3 mm for all the measurements.
As shown in Fig. 4, the residual stress components determined at the three predefined positions around the segment circumference
are almost the same. Such regular distribution of residual stress around the railway axle segment we found for all the etched depths.
Therefore, axisymmetric residual stress distribution can be assumed, so it is possible to use the average of the three measured values as
the residual stress magnitude for a given depth.

2.2. The layer-removal-based method

The original layer removal method for cylindrical parts that served as the basis for the LRBM is described by Moore and Evans [41].
The experimental procedure of LRBM is the same as in Moore and Evans’s work.
The specimen for the LRBM is a full segment of the axle (see Fig. 2 and a detail in Fig. 5). The LRBM is a series of residual stress
measurements by the XRD followed by removal of a layer of material by machining (lathe turning). The layer removal is schematically
shown in Fig. 5. After each layer removal, another round of stress measurements follows. These steps are repeated until the specimen
no longer deforms. The residual stress measurements are always carried out in the same 3 points on the surface spaced by 120◦ – see
Fig. 5. The particular parameters of the experiment (like the length of the section, number of the layers, etc.) are the subject of the first
step in the procedure (see Fig. 3) – the design of the experiment – and they should be chosen to fit the particular part under study. The
choice of the parameters is addressed in section 3 of this paper.
Note that, measured stress generally does not correspond to the original residual stress in axle due to removed layers of material
with residual stress, see, e.g. [17]. Therefore, there is a redistribution of residual stress in the machined segment. Hence, after the
measurements, the residual stress values must be corrected by a calculation. For this, Moore and Evans provided an analytical solution
based on the theory of elasticity. In this work, numerical simulations were used instead.

2.3. The sectioning-based method

The original sectioning method for the determination of longitudinal residual stress in bars was described by Kalakoutsky [43]. Is
required to slit the specimen into many small strips and measure their deformation. Ueda et al. [44] updated the method for a three-
dimensional residual stress determination using inherent strains. However, this methodology requires strain recording while cutting
out specimens from the original component. The sectioning-based method (SBM) is a new method designed to need just two sections
from the railway axle. On those, residual stress is measured by the XRD, without the need for further machining.
The SBM needs two specimens manufactured from the studied railway axle (see Fig. 2). The first specimen, segment A, serves for
the axial residual stress determination, and it is obtained by cutting a part of the axle into two halves along the axial direction. The
second specimen, segment R, has a flat ring shape and serves for the tangential (circumferential) and radial residual stress determi­
nation. The XRD measures the stresses in points spaced along straight paths on the surface of the specimens – shown in Fig. 6. The
parameters of the SBM for this work are described further in section 3.
The main advantages of the SBM are that it requires fewer measurements than the LRBM and much less time, as no additional
manufacturing of the measured segments is needed between the measurements. The disadvantage is that the information of residual
stress distribution is determined just in one or two paths (180◦ rotated to each other) across the cross-section. Therefore, its accuracy
can be lower than in the case of LRBM, where every stress value is determined as the average of three values on the circumference.
Like in the case of the LRBM, the measured values must be corrected to obtain the original residual stress distribution. In the case of

Fig. 4. Measured residual stress versus the etched depth for different circumferential positions determined on the original outer surface.

5
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 5. The measuring points position for the LRBM.

Fig. 6. Measuring paths positions for segment A and segment R for SBM.

SBM, a numerical simulation is the only possible choice for the calculation because no analytical solution is possible for the specimen
shapes.

2.4. Numerical simulations

Every time a portion of the material is removed from a part with residual stress, it causes a redistribution of stress in the rest.
Therefore, gradually removing layers causes a discrepancy – the measured residual stress does not equal the original residual stress
state in a component before the machining. The original residual stress distribution is still to be determined based on the measured
values after a successful residual stress measurement. A procedure of such determination, which uses numerical simulations by the
FEM, was developed, and its application is described here. This part of the procedure corresponds to steps 3 to 10 in Fig. 3.
Finite element models of the specimens used in the experiments are created. In the case of LRBM, the models must correspond to all
the stages of the investigation – i.e., one model for every number of layers removed. For the SBM, just two models (segment A and
segment R) are necessary.
Residual stress in the models is defined as the initial state of the body. The residual stress is defined relative to the original outer
surface. Therefore, regardless of actual specimen diameter (e.g., the number of removed layers), the residual stress distribution that is
put in the model keeps a defined shape. When a model with a defined initial state is solved, the initial stress redistributes and assumes
an equilibrium state. It corresponds to the actual stress redistribution when specimens are cut out of the axle and machined. The
redistributed stress is then the stress that is possible to be measured by XRD. The goal is to find such residual stress distribution that has
caused the measured values.
As the initial approximation, the residual stress distribution obtained from the corresponding measurements is defined in the
models. The models are then solved. The resultant redistributed residual stress is then compared to the measured values. The input data
are changed and again used as the initial state in the next round of calculations based on the obtained difference. This procedure was
repeated several times until such initial stress distribution was found to be similar to measured values. Such distribution can then be
considered the accurate, original residual stress distribution in the examined part.
The stress distribution found by the numerical simulations can then be used in the simulation of the whole axle.

3. Residual stress determined by the LRBM

The layer-removal-based method consists of two main parts as outlined by the general methodology. The first part is the actual
experiment (corresponding to the first two steps in Fig. 3) that involves the design of the experimental setting (dimensions of spec­
imens, number of layers, etc.) and measurements by XRD. The second part (corresponding to steps 3 to 10 in Fig. 3) is a set of numerical
simulations that finally lead to the corrected residual stress distribution in the whole railway axle.

6
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

3.1. Experimental setting and measurements

We chose the specimen dimensions to bear in mind that the manipulating the whole axle is very difficult. Therefore, we studied the
influence of the free surface on the stress state to avoid the necessity to use the entire axle for experiments. It was then decided that the
symmetrical 433 mm long part (shown in Fig. 2) is compact enough. At the same time the investigated cross-section in the middle of
such a segment stays free from the effects of the free surfaces, according to Saint-Venant’s principle [45]. The residual stress state is
preserved in its original form in the whole axle.
The number of layers to subsequent machining away was designed to 17 (the layers did not have uniform thickness). The first
residual stress measurement was carried out on the original surface of the axle before cutting the 433 mm long segment out. Then,
another measure was performed on the original surface after the specimen cut out of the axle. These measurements would reveal the
potential influence of free surfaces of segment cut from the axle. After that, three measurements for three removed layers with the
thickness of 5 mm were performed (to remove remaining pairs of wheel seats), followed by 11 measurements and 11 removed layers
with the step of 2 mm and the last four measurements and four removed layers were carried out with the degree of 5 mm.
The measurements were always carried out at the same spots spaced by 120◦ on the circumference in the middle of the segment, see
Fig. 5. The total number of measurements was 20 × 3 = 60. XRD was employed for all the measurements. Axial and tangential
components of stress were recorded in each of the three spots on the surface.
The measured residual stress values cannot be considered as the original residual stress state – a correction using an analytical or
numerical calculation must be applied to them.

3.2. Numerical evaluation of the original residual stress state

We used the ANSYS software and its capability of parametrical programing (APDL) for the simulations. The discretization was made
very fine to be able to describe the residual stress gradient. The biggest model contained about 21 000 quadratic, quadrilateral 2D
elements. The material model of the segments was linear elastic isotropic, defined by Young’s modulus of 205 GPa and Poisson’s ratio
of 0.3. The model’s initial state was definedusing the INISTATE function available in the ANSYS [46].
We created a 2D axisymmetric representation (see examples of the geometry in Fig. 7 (a)) for each measured geometry (one
measured depth point corresponds to one FEM model). In the next step, the same input residual stress was defined in the model – this
state is illustrated in Fig. 7 (b). Note that the distribution is defined always starting at the original outer surface of the axle and
removing a layer, thus removing the corresponding parts of the stress distribution. After solving the models, the residual stress

Fig. 7. Illustration of LRBM (a) segment models; (b) defined initial stress state; (c) redistributed equilibrium stress state.

7
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

(originally is not in equilibrium) redistributes, resulting in the distribution in Fig. 7 (c). These values are then compared to the
measured values on corresponding surfaces. The input stress is then changed accordingly, and the steps are repeated.
Fig. 8 shows the evolution of input residual stress (green lines) concerning the measured data (black crosses). The goal is to find
such input stress (green line) that, after redistributing and achieving equilibrium, the resultant stress (red line) corresponds to the
measured data. Residual stresses in the last two measured depths are close to zero. There is a small region between the depths of 42 mm
to 49 mm marked as “estimated” in Fig. 9, where the extrapolation of a constant value estimates the residual stress values.
We found compliance between numerically and experimentally determined values of residual stress after several iterations. Fig. 9
shows the measured data points (black crosses) compared to the equilibrium state of axial residual stress in each FEM model of the
corresponding segment. It can be seen that the measured residual stress obtained on the surface of each segment corresponds to the
simulation result at the surface, while we used the same initial stress state data for each of the models – that means we can take this
initial stress distribution for the original stress state in the specimen.
Fig. 10 shows results axial, tangential, and radial residual stress components. These data correspond to actual residual stress
distribution in the axle. In the following, they were used as input values for a numerical model of the whole axle.
We can estimate the residual stress state in the whole axle after loading the model of the entire axle by the initial residual stress
distribution obtained in the previous FEM analysis. It is important to note here that the residual stress in the whole axle corresponds to
the measured values in the parts cut out and measured. However, in other parts of the axle, the residual stress distribution differs (e.g.,
the ends or the rounded transitions) because it assumes the most suitable equilibrium. Fig. 11 shows the resultant distribution of
residual stress in the whole axle.

4. Residual stress determined by the SBM

However functional, the LRBM is still considerably time-consuming due to the many repeated machining and measuring cycles. The
SBM was developed as a less time-consuming alternative with the same capacity to describe the residual stress distribution as the
LRBM. The basic scheme of the methodology shown in Fig. 3 was preserved.

4.1. Experimental setting

The SBM requires two different specimens – one for the axial stress measurement (segment A) and the other for the radial and
tangential stress measurements (segment R). Fig. 12 (a) shows segment A in detail. It is an exact half of an axle segment cut along the
longitudinal axis. The length of the segment is 433 mm, the same as for LRBM. Fig. 12 (d) shows segment R in detail. It is a 10 mm thick
ring – we chose a lower thickness to ensure axial residual stress is relaxed and keep the tangential and radial residual stress in the ring.
As shown in Fig. 6, the location for the measurement was considered as a path connecting the outer surface with the inner surface
for both segments. The path in segment A was located precisely in the middle of the segment. We set the number of measured points to
21. The spacing of the points is the following: from the outer diameter to the depth of 6 mm by the step of 1 mm; from 6 mm to the
depth of 26 mm by the step of 2 mm and from 26 to 46 mm (inner surface) by step of 5 mm. Preparation of the surface before
measurement and residual stress measurement as described in section 2 followed.
One measurement of residual stress on the surface of the original axle was also performed before cutting the segments A and R out
of the axle for later verification of the surface values. The residual stress near the original outer and inner diameter (close to the edges
of the specimens) cannot be reliably measured by the given method due to the dimensions of the probe. Therefore, the first point is
measured at a depth of 2 mm under the outer surface and the last 2 mm before reaching the inner surface. Missing input data from the
original outer and inner surfaces were estimated by constant extrapolation of the closest available value.

Fig. 8. Schematic evolution of first two iterations of initial state input data concerning measured axial residual stress.

8
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 9. Optimized solution for simulated segments.

Fig. 10. Results of residual stresses determined by LRBM.

4.2. Numerical evaluation of the original residual stress state

Again, the numerical procedure for the determination of the original residual stress field was performed. The process was similar to
the case of LRBM – repeated cycles of initial state definition, followed by equilibrium state determination and comparison to measured
stress values. Instead of having to simulate 16 different specimens, just the two segments A and R were modelled. In this case, 3D
models of the specimens were used because segment A is not axisymmetrical and, therefore not suitable for 2D modelling. Segments A
and R discretization was done by 57,000 and 2300 quadratic, hexahedral 3D elements, respectively. The material model of the seg­
ments was linear elastic isotropic, defined by the Young’s modulus of 205 GPa and Poisson’s ratio of 0.3 as well as was for LRBM.
The geometry of the models is shown in Fig. 12 (a) and (d). The initial state of the models before solving for the equilibrium is
shown in Fig. 12 (b) and (e). Fig. 12 (c) and (f) then demonstrate the same model after achieving equilibrium. This state corresponded
to the state of the actual specimens when the XRD measured residual stress. As soon as the measured state and the determined
equilibrium agree with a sufficiently slight deviation, the initial state of stress that caused this equilibrium is used for full-scale axle
where the original residual stress is obtained.
The correlation of the measured data (points) with the calculated equilibrium for segments A and R (solid lines) is plotted in Fig. 13
and Fig. 14, respectively.
LRBM results show small values of radial stress in comparison with axial and tangential stress. Therefore, radial stress was
neglected for further optimization, and presented radial stress results are obtained as a side effect of the equilibrium stress state.
Fig. 15 shows the already corrected, original axial, tangential, and radial residual stress distributions determined by the SBM.

9
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 11. Final residual stress distribution in the whole axle.

Fig. 12. Illustration of SBM (a), (b), (c) segment A - axial cut; (d), (e), (f) segment R - radial cut; (a), (d) models for measurement; (b), (e) dis­
tribution of input stress; (c), (f) redistributed stress after numerical calculation.

Fig. 13. Optimized stress components for segment A (axial cut).

10
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 14. Optimized stress components for segment R (radial cut).

Fig. 15. Results of residual stress measurement by SBM.

5. Discussion and validation

5.1. Discussion of the results

Two different methods for determining residual stress through a cross-section of railway axles were performed that combine well-
established experimental procedures with numerical simulations. Experimental specimens for both of the methods were manufactured
from the same axle.
The axial residual stress distributions obtained from both methods are plotted in Fig. 16. There is substantial compliance between
the two methods. The most important residual stress component is the axial component because the most dangerous cracks grow in the
radial direction with tensile (positive) axial stress acting as the opening force (the compressive residual stress acts against the opening,
thus preventing cracks from growing). For investigated commercially manufactured axle with induction hardening treatment, the
compressive (negative) axial residual stress of about − 600 MPa was found up to the depth of 8 mm below the axle surface. Then, the
axial residual stress rises in the interval from 10 mm to 24–27 mm, where it reaches its maximum of 490 MPa. It is very important to
note here that the effort to induce a high compressive residual stress close to the surface of the axle also causes high positive stress
deeper in the cross-section.
The distributions of the other two components of stress, tangential and radial, are shown in Fig. 10 and Fig. 15 for LRBM and SBM,
respectively.
There is a cross in Fig. 16 marking the measurement of residual stress distribution on the surface of segment A before it was cut in
half and processed by the SBM. This could be used as the stress value in the region on edge, where it was not possible to be measured,
instead of using the extrapolated value. However, even with the extrapolations, the residual stress distribution determined by the SBM
is sufficiently accurate.
The axial and tangential stress in the railway axle was measured quite precisely, with a relatively low scatter in the values. Thanks
to the shape of the specimens for the SBM, it was possible to measure also the radial components (this is impossible in the LRBM). There
is a noticeable scatter in the values of radial stress (see Fig. 13 and Fig. 14). However, radial stress is not very important regarding crack
propagation. In a model of the whole axle, the radial component is usually calculated automatically to match the state of the other two

11
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 16. Comparison of axial residual stress measured by different methods.

components, which are also much more significant in magnitude.

Both of the presented methods are destructive, so it is impossible to repeat the measurements in production very frequently. That is
their biggest drawback. However, it is expected that the measured axle represents the whole series of manufactured axles (with the
same technological processes and parameters), which will have similar residual stress distribution. Therefore, sacrificing one axle to be
destroyed by the experimental procedure is acceptable.
The advantage of the LRBM, in particular, lies in need for one specimen only. Also, if using FEM is not possible for some reason,
there is always the possibility to use the original analytical method of evaluation by Moore and Evans [41]. The analytical method was
also applied to the data measured here, and the residual stress distribution was compared to the numerically obtained distribution. The
comparison is plotted in Fig. 17 (a) – the obtained distributions are practically identical. However, suppose only some measured values
are used for the evaluation, simulating a coarser sampling. In that case, the numerical procedure can produce results closer to reality,
see Fig. 17 (b). The LRBM proposed here has a little advantage over the original layer removal method with the analytical evaluation.
Should the residual stress be non-uniformly distributed around the axis (not axisymmetric)? The LRBM can show, because of the
measurements in 3 places (it could be even more) on the circumference. The main disadvantage is the repeated time-consuming
machining.
The main advantage of the SBM is that no additional machining of the specimens is needed once they are cut out of the axle, which
saves a considerable amount of time and costs. FEM software is necessary for the final residual stress evaluation, but such equipment is
quite standard in larger industrial manufacturers like railway axle producers. Another minor disadvantage is that measuring residual
stress by XRD near original outer and inner surfaces is practically impossible. However, an additional measurement performed before
the cutting can reduce this problem for the outer surface.
Successful application of the methods described above shows that the XRD applied to an appropriate specimen can provide in­
formation about the residual stress distribution not only close to the body surface (ca 10 µm, see Fig. 1 – X-ray diffraction method), but
also in the whole cross-section, provided that a numerical FEM simulation is applied to obtain the correct results. The LRBM and the
SBM require standard tools available in industrial companies, i.e., a set of XRD equipment and FEM software.

Fig. 17. Comparison of axial residual stress evaluated by different methodologies from the same measured data with a) fine depth sampling b)
coarse depth sampling.

12
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

Fig. 18. (a) Comparison of tangential residual stresses measured by different methods on (b) ring specimens.

5.2. Validation of the results

The described methods both rely on the use of the same procedure of residual stress measurement – the XRD. Although their
compliance is acceptable and an analytical calculation verified the evaluation procedure, this alone cannot be taken for sufficient
validation. Therefore, to validate obtained results, additional experiments by different techniques were carried out.
To perform the referential measurements, preferably on the same geometry used by one of the two methods above (concerning the
necessity of transporting the specimens), the shape of a ring specimen (the same as the section R for the SBM) was chosen, see Fig. 18
(b). The ring specimens were manufactured from a different axle of the same type as the one studied in this work. The following
methods measured tangential residual stress distribution in these specimens:

• neutron diffraction – performed by the Nuclear Physics Institute of the Czech Academy of Sciences, Řež, Czech Republic
• slit ring method – performed at the Institute of Physics of Materials of the Czech Academy of Sciences by the authors of this work; this
experiment was already published [47], where the detailed description can be found
• X-ray diffraction – additional XRD measurement performed by the Research Centre of the University of Žilina, Žilina, Slovakia.

The comparison of tangential residual stress obtained by these methods is shown in Fig. 18 (a). The solid lines in Fig. 18 (a) stand for
the referential results from specimens manufactured from a different railway axle. The red points represent the results obtained by the
SBM on segment R, already presented in this paper. All the methods show similar results to each other.

6. Conclusions

The paper focuses on the accurate determination of the magnitude of residual stresses in the railway axle. This knowledge is crucial
for determining the effect of heat treatment and can be used in subsequent analyses, e.g. for determining the residual fatigue lifetime of
railway axles. Two different methods for the determination of residual stresses across the cross-section of a railway axle using available
X-ray diffraction and FEM software were presented. Both methods can be used by producers of railway axles for quick investigations of
residual stress distribution using a piece of commonly available equipment.
Knowledge of the axial component of the residual stress is the most important for further axle design. The compressive axial re­
sidual stress was found to be as low as − 600 MPa up to 8 mm under the surface. These compressive residual stresses can play a
significant role in the case of fatigue crack propagation from the axle surface. The significant magnitude of the stresses near the axle
surface makes the propagation of a potential crack more difficult and contributes to the good fatigue resistance of the axle. However,
the induction hardening severely influenced the residual stress distribution in the whole axle. At a depth of about 18 mm, the nature of
the residual stresses changes from compressive to tensile. A peak of positive residual stress at 490 MPa was found at a depth between 20
and 25 mm under the axle surface. This value is even higher than the external mechanical load of the axle considered during train
operation.
A distribution of tangential and radial components of residual stresses was also determined. The tangential stress distribution is
similar to that of the axial component. The magnitude of the radial component is not significant from the point of view of the axle load.
It varies smoothly between zero values on the inner and outer surface of the axle, reaching a maximum at a similar depth to that at
which the axial and tangential residual stresses change their character from compressive to tensile, i.e. between 18 and 20 mm.
Although the tensile axial and tangential stresses reach significant values deep below the surface, this does not affect the positive
effect of surface hardening on the fatigue resistance of the railway axle. The complete residual stress distribution is then essential for
residual fatigue lifetime evaluation, where internal loading by residual stresses is as important as external loading from bending or
press-fit loading.
It was shown that the induction hardening creates a residual stress gradient in the railway axle cross-section that can work for the
benefit of the part in terms of fatigue resistance to typical surface cracks because the compressive axial stress at the surface prevents the
onset of defects and crack growth.
The applicability of the presented methodology is not limited to railway axles only. It can be applied to the determination of

13
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

residual stress distribution in sizeable cylindrical bodies (with the assumption of axisymmetrical residual stress distribution) in
general, which can be helpful in many engineering applications.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Acknowledgement

This work has been created using the research infrastructure of the Institute of Physics of Materials of Academy of Sciences of the
Czech Republic, v.v.i., CEITEC IPM and in close cooperation with BONATRANS GROUP a.s. This work was supported by project No.
FV40034 “Development of new design of railway axles with high operational reliability” of Ministry of Industry and Trade of the Czech
Republic. This work was also partially supported by Operational Program Integrated Infrastructure 2014 - 2020 of the project:
Innovative Solutions for Propulsion, Power and Safety Components of Transport Vehicles, code ITMS 313011V334, co-financed by the
European Regional Development Fund.
The authors would like to thank Dr. Charles Hervoches from Neutron Physics Laboratory, Nuclear Physics Institute, v. v. i., The
Czech Academy of Sciences, Řež, Czech Republic, for the experimental measurements performed by neutron diffraction and the
Research Centre of the University of Žilina, Slovakia for the referential X-ray diffraction measurements.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.engfailanal.2022.106142.

References

[1] A. Wöhler, Versuche zur Ermittlung der auf die Eisenbahnwagenachsen einwirkenden Kräfte und die Widerstandsfähigkeit des Wagen-Achsen, Zeitschrift Für
Bauwes. 10 (1860) 583–616.
[2] J. Zhang, H. Li, B. Yang, B. Wu, S. Zhu, Fatigue properties and fatigue strength evaluation of railway axle steel: Effect of micro-shot peening and artificial defect,
Int. J. Fatigue 132 (2020) 105379, https://doi.org/10.1016/j.ijfatigue.2019.105379.
[3] D. Regazzi, S. Beretta, M. Carboni, An investigation about the influence of deep rolling on fatigue crack growth in railway axles made of a medium strength steel,
Eng. Fract. Mech. 131 (2014) 587–601, https://doi.org/10.1016/j.engfracmech.2014.09.016.
[4] T. Oevermann, T. Wegener, A. Liehr, L. Hübner, T. Niendorf, Evolution of residual stress, microstructure and cyclic performance of the equiatomic high-entropy
alloy CoCrFeMnNi after deep rolling, Int. J. Fatigue 153 (2021) 106513, https://doi.org/10.1016/j.ijfatigue.2021.106513.
[5] P. Castellini, L. Stroppa, N. Paone, Laser sheet scattered light method for industrial measurement of thickness residual stress distribution in flat tempered glass,
Opt. Lasers Eng. 50 (5) (2012) 787–795, https://doi.org/10.1016/j.optlaseng.2011.12.008.
[6] R. Rego, C. Löpenhaus, J. Gomes, F. Klocke, Residual stress interaction on gear manufacturing, J. Mater. Process. Technol. 252 (2018) 249–258, https://doi.org/
10.1016/j.jmatprotec.2017.09.017.
[7] S.K. Tippabhotla, N.G. Diesta, X. Zhang, S. Sridhara, C.V. Stan, N. Tamura, A.A.O. Tay, A.S. Budiman, Thermomechanical residual stress evaluation in multi-
crystalline silicon solar cells of photovoltaic modules with different encapsulation polymers using synchrotron X-ray microdiffraction, Sol. Energy Mater. Sol.
Cells 193 (2019) 387–402, https://doi.org/10.1016/j.solmat.2019.01.016.
[8] J. Zhang, B.o. Song, Q. Wei, D. Bourell, Y. Shi, A review of selective laser melting of aluminum alloys: Processing, microstructure, property and developing
trends, J. Mater. Sci. Technol. 35 (2) (2019) 270–284, https://doi.org/10.1016/j.jmst.2018.09.004.
[9] P. Hutař, P. Pokorný, J. Poduška, R. Fajkoš, L. Náhlík, Effect of residual stresses on the fatigue lifetime of railway axle, Procedia Struct. Integr. 4 (2017) 42–47,
https://doi.org/10.1016/j.prostr.2017.07.005.
[10] J.W. Zhang, L.T. Lu, K. Shiozawa, X.L. Shen, H.F. Yi, W.H. Zhang, Analysis on fatigue property of microshot peened railway axle steel, Mater. Sci. Eng., A 528 (3)
(2011) 1615–1622, https://doi.org/10.1016/j.msea.2010.10.086.
[11] X. Lorang, Y. Cheynet, P. Feraud, Y. Nadot, A study on lifetime of a railway axle subjected to grinding, Procedia Eng. 213 (2018) 255–261, https://doi.org/
10.1016/j.proeng.2018.02.026.
[12] M. Luke, I. Varfolomeev, K. Lütkepohl, A. Esderts, Fatigue crack growth in railway axles: assessment concept and validation tests, Eng. Fract. Mech. 78 (5)
(2011) 714–730, https://doi.org/10.1016/j.engfracmech.2010.11.024.
[13] X. Li, J. Zhang, B. Yang, J. Zhang, M. Wu, L. Lu, Effect of micro-shot peening, conventional shot peening and their combination on fatigue property of EA4T axle
steel, J. Mater. Process. Technol. 275 (2020) 116320, https://doi.org/10.1016/j.jmatprotec.2019.116320.
[14] S.M. Hassani-Gangaraj, M. Carboni, M. Guagliano, Finite element approach toward an advanced understanding of deep rolling induced residual stresses, and an
application to railway axles, Mater. Des. 83 (2015) 689–703, https://doi.org/10.1016/j.matdes.2015.06.026.
[15] K. Ishii, Fatigue Strength and Maintenance of the Wheel-Axle Assembly for the Japanese Fast Train (Shinkansen), J. Eng. Mater. Technol. 100 (1978) 227,
https://doi.org/10.1115/1.3443484.
[16] S. Isomura, K. Yomoda, Manufacturing history of axles for Shinkansen, Proc. 11th Int Wheel. Congr. 2 (1995) 51–54.
[17] P. Pokorný, P. Dlhý, J. Poduška, R. Fajkoš, T. Vojtek, L. Náhlík, M. Grasso, P. Hutař, Influence of heat treatment-induced residual stress on residual fatigue life of
railway axles, Theor. Appl. Fract. Mech. 109 (2020) 102732, https://doi.org/10.1016/j.tafmec.2020.102732.
[18] J. Holmberg, A. Steuwer, A. Stormvinter, H. Kristoffersen, M. Haakanen, J. Berglund, Residual stress state in an induction hardened steel bar determined by
synchrotron- and neutron diffraction compared to results from lab-XRD, Mater. Sci. Eng., A 667 (2016) 199–207, https://doi.org/10.1016/j.msea.2016.04.075.
[19] U. Zerbst, S. Beretta, G. Köhler, A. Lawton, M. Vormwald, H.T. Beier, C. Klinger, I. Černý, J. Rudlin, T. Heckel, D. Klingbeil, Safe life and damage tolerance
aspects of railway axles – a review, Eng. Fract. Mech. 98 (2013) 214–271, https://doi.org/10.1016/j.engfracmech.2012.09.029.
[20] R.A. Smith, S. Hillmansen, A brief historical overview of the fatigue of railway axles, Proc. Inst. Mech. Eng. Part F J. Rail Rapid Transit. 218 (4) (2004) 267–277,
https://doi.org/10.1243/0954409043125932.
[21] K. Hirakawa, K. Toyama, M. Kubota, The analysis and prevention of failure in railway axles, Int. J. Fatigue 20 (2) (1998) 135–144, https://doi.org/10.1016/
S0142-1123(97)00096-0.

14
P. Dlhý et al. Engineering Failure Analysis 135 (2022) 106142

[22] M. Rieger, C. Moser, P. Brunnhofer, D. Simunek, F.-J. Weber, A. Deisl, H.-P. Gänser, R. Pippan, N. Enzinger, Fatigue crack growth in full-scale railway axles –
influence of secondary stresses and load sequence effects, Int. J. Fatigue 132 (2020) 105360, https://doi.org/10.1016/j.ijfatigue.2019.105360.
[23] D. Regazzi, S. Cantini, S. Cervello, S. Foletti, A. Pourheidar, S. Beretta, Improving fatigue resistance of railway axles by cold rolling: process optimisation and
new experimental evidences, Int. J. Fatigue 137 (2020) 105603, https://doi.org/10.1016/j.ijfatigue.2020.105603.
[24] D. Simunek, M. Leitner, M. Rieger, R. Pippan, H.P. Gänser, F.J. Weber, Fatigue crack growth in railway axle specimens – transferability and model validation,
Int. J. Fatigue 133 (2020) 105421, https://doi.org/10.1016/j.ijfatigue.2019.105421.
[25] J.-W. Gao, X.-N. Pan, J. Han, S.-P. Zhu, D. Liao, Y.-B. Li, G.-Z. Dai, Influence of artificial defects on fatigue strength of induction hardened S38C axles, Int. J.
Fatigue 139 (2020) 105746, https://doi.org/10.1016/j.ijfatigue.2020.105746.
[26] H. Gong, Y. Wu, X. Feng, Y. Li, Analysis of quenching stresses in 35CrMo axle, J. Wuhan Univ. Technol. Mater. Sci. Ed. 31 (3) (2016) 630–635, https://doi.org/
10.1007/s11595-016-1421-9.
[27] H.J. Schindler, Effect of residual stresses on safe life prediction of railway axles, Procedia Struct. Integr. 4 (2017) 48–55, https://doi.org/10.1016/j.
prostr.2017.07.008.
[28] H.P. Gänser, J. Maierhofer, R. Tichy, I. Zivkovic, R. Pippan, M. Luke, I. Varfolomeev, Damage tolerance of railway axles - the issue of transferability revisited,
Int. J. Fatigue 86 (2016) 52–57, https://doi.org/10.1016/j.ijfatigue.2015.07.019.
[29] F.A. Kandil, J.D. Lord, A.T. Fry, P. V. Grant, A Review of Residual Stress Measurement Methods - A Guide to Technique Selection, 2001.
[30] G.S. Schajer, Practical Residual Stress Measurement Methods, John Wiley & Sons, Ltd, Chichester, UK, 2013. Doi: 10.1002/9781118402832.
[31] P.J. Withers, H.K.D.H. Bhadeshia, Residual stress. Part 1 – measurement techniques, Mater. Sci. Technol. 17 (4) (2001) 355–365, https://doi.org/10.1179/
026708301101509980.
[32] N.S. Rossini, M. Dassisti, K.Y. Benyounis, A.G. Olabi, Methods of measuring residual stresses in components, Mater. Des. 35 (2012) 572–588, https://doi.org/
10.1016/j.matdes.2011.08.022.
[33] R.B. Thompson, W.-Y. Lu, A.V.J. Clark Jr., Ultrasonic Methods, in: Handb. Meas. Residual Stress., 1996, pp. 153–182.
[34] Neutron centres map, 2020. https://neutronsources.org/neutron-centres/.
[35] G. Totten, M. Howes, T. Inoue, Handbook of Residual Stress and Deformation of Steel, 2002.
[36] D.J. Smith, Deep Hole Drilling, in: Pract. Residual Stress Meas. Methods, 2013, pp. 65–87. Doi: 10.1002/9781118402832.ch3.
[37] M.R. Hill, The Slitting Method, in: Pract. Residual Stress Meas. Methods, 2013. Doi: 10.1002/9781118402832.ch4.
[38] M.B. Prime, A.T. DeWald, The Contour Method, in: Pract. Residual Stress Meas. Methods, 2013. Doi: 10.1002/9781118402832.ch5.
[39] J.F. Flavenot, Layer Removal Method, in: Handb. Meas. Residual Stress., 1996: pp. 35–48.
[40] Y. Ueda, Sectioning Methods, in: L. Jian (Ed.), Handb. Meas. Residual Stress., Farnibt Press, 1996, pp. 49–70.
[41] M.G. Moore, W.P. Evans, Mathematical correction for stress in removed layers in X-ray diffraction residual stress analysis, SAE Tech. Pap. 45 (1958) 340–345,
https://doi.org/10.4271/580035.
[42] EN 13261:2011, Railway Application – Wheelsets and Bogies – Axles – Product Requirements, 2010.
[43] N. Kalakoutsky, The Study of Internal Stresses in Cast Iron and Steel, London (1888), also St. Petersburgh (1889), 1888.
[44] Y. Ueda, K. Fukuda, Y.C. Kim, New measuring method of axisymmetric three-dimensional residual stresses using inherent strains as parameters, J. Eng. Mater.
Technol. Trans. ASME 108 (1986) 328–334, https://doi.org/10.1115/1.3225890.
[45] R.v. Mises, On Saint Venant’s principle, Bull. Am. Math. Soc. 51 (8) (1945) 555–562, https://doi.org/10.1090/S0002-9904-1945-08394-3.
[46] ANSYS Inc., ANSYS Help Release 2020 R2, (n.d.).
[47] P. Dlhý, J. Poduška, P. Pokorný, L. Náhlík, R. Fajkoš, P. Hutař, Methodology for estimation of residual stresses in hardened railway axle, Procedia Struct. Integr.
23 (2019) 185–190, https://doi.org/10.1016/j.prostr.2020.01.084.

15