Engineering Structures 188 (2019) 553–563

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Experimental and finite element fatigue assessment of the spring clip of the T
SKL-1 railway fastening system
⁎
Diego Ferreño , José Antonio Casado, Isidro Alfonso Carrascal, Soraya Diego, Estela Ruiz,
María Saiz, José Adolfo Sainz-Aja, Ana Isabel Cimentada
LADICIM (Laboratory of Science and Engineering of Materials), University of Cantabria. E.T.S. de Ingenieros de Caminos, Canales y Puertos, Av. Los Castros 44, 39005
Santander, Spain

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

Keywords: The fastening system that links the rail and the sleeper in railway lines is an element of high responsibility, since
Railway fastening system its failure may lead to the derailment of the train. This paper combines an experimental approach and a Finite
Finite element Element model to assess the structural integrity of the spring clip of the SKL-1 fastening system, used assiduously
Fatigue test in high-speed lines. Our results prove that fatigue is not the failure mechanism that explains the fractures of clips
Failure analysis
as pointed out by other authors, provided the following requirements: the material has received the appropriate
Steel spring clip
thermomechanical treatments; the assembly of the clip has been carried out with precision; and the system has
not suffered anomalous in-service overloads.

1. Introduction model, the simplified description of the properties of the material or the
fact that they did not provide any experimental validation of the model.
Rail transport has experienced a remarkable boost in recent dec- In addition, their research did not investigate the mechanisms that led
ades. In particular, high-speed rail has accounted for a substantial to the breaking of the clips. Fatigue represents the most common me-
fraction of public transport investments worldwide. At present, high- chanism of failure in structural components. There is a large number of
speed trains can circulate at 350 km/h. However, public managers as- contributions dealing with fatigue assessment in railway components
pire to increase the in-service speed even more, in order to improve the such as [2–5]. For instance, in a recent contribution, Hong et al. [6,7]
quality of the transport. This would represent a technological challenge identified fatigue as the failure mechanism leading to the frequent
in many aspects, in particular with regard to the forces supported by the fractures appearing in e-type spring clips in Beijing subway lines, jeo-
fastening system that connects the rails and the sleepers and transmits pardizing the operational safety. They used a FE model to assess the
the static and dynamic forces exerted by the passing stocks to the fatigue life of an e-type fastening clip. The mechanical behavior of the
railway infrastructure. The fastening system maintains the gauge of the 60Si2Mn spring steel was described by means of an elastic-plastic bi-
track and the inclination of the rails within the admissible tolerances linear tensile curve (including hardening) [8,9]. Based on this FE
avoiding the overturning of the rail. The SKL-1 fastening system, see model, Hong et al. developed a complete parametric study of the clips
Fig. 1, is used assiduously on high-speed as well as on some conven- in order to optimize their fatigue performance. Hasap et al. [10] studied
tional tracks. the influence of the toe load on the deformation and fatigue resistance
The literature reports examples of failure of SKL-1 fastening spring of elastic rail e-clips using FE and fatigue loading of the clips. They
clips detected during the inspections carried out on the tracks. These obtained numerically the deformation, stress and strain states of the
failures could compromise the very functionality of the railway and the railway track system under different scenarios of toe and wheel loads.
safety of vehicles and passengers. For this reason, it is necessary to This information was employed to conduct a series of fatigue tests on
identify the underlying causes. Lakusic et al. [1], reported several clips under three installation conditions (low, normal, and high toe
fractures of SKL-1 fastening clips. These authors developed a Finite loads) and two cases of wheel loads (in the first case the wheel was on
Element (FE) model of the clip to determine the distribution and con- the base plate and in the other it was at the middle between two base
centration of stresses subjected to in-service conditions. Their work plates). Based on this experimental, they found that under normal
presented several limitations such as the number of elements of the FE wheel loads, the toe load does not influence the fatigue resistance of the

⁎
Corresponding author.
E-mail address: [email protected] (D. Ferreño).

https://doi.org/10.1016/j.engstruct.2019.03.053
Received 10 August 2018; Received in revised form 15 March 2019; Accepted 18 March 2019
Available online 22 March 2019
0141-0296/ © 2019 Elsevier Ltd. All rights reserved.
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 1. (a) Schematic description of the SKL-1 fastening system and its components. (b–d) Detailed drawings showing the geometry of the SKL-1 clip.

clip; in contrast, impact wheel loading reduces substantially the fatigue through quasi-static and fatigue tests, respectively. The experimental
lifespan of the clips. In [11], Mohammadzadeh et al. splitted the fatigue results were used to calibrate and validate the FE model (both under
lifespan of spring clips into initiation and propagation stages and ob- static and fatigue loading). Finally, once the numerical model was va-
served that the majority of the fatigue life was consumed in the crack lidated, it was used to simulate the response of the component under in-
generation stage and that the working conditions of spring clips cor- service conditions, in order to assess its structural integrity when sub-
responded to high cycle fatigue. In a later contribution, [12], they used jected to fatigue loading. To facilitate the reading of the paper, Table 1
a fracture mechanics-based approach to conduct a fatigue reliability includes a summary of the tests carried out, as well as their purpose.
analysis of SKL-14 spring clips under traffic loads. They considered
initial crack sizes between 0.1 and 0.7 mm and observed that in all
cases the reliability index is very low at the propagation stage, re- 2. Material and methods
commending the component to be replaced after detecting any crack.
Moreover, they obtained the evolution of the reliability index as a 2.1. Material and scope
function of time, concluding that this parameter falls to zero after a
short period of time that depends on the initial crack size. SKL-1 clips are fabricated in 38Si7 steel. Its specified chemical
In this work, we have developed a FE numerical model (using the composition is detailed in Table 2 [13].
ANSYS Workbench 18.0 software) reproducing the operating conditions The manufacture of the W-type clips consists of the mechanical
of the W-type clips of the SKL-1 fastening system. The study has been forming of the wire rod followed by a quenching and annealing heat
carried out in three different phases. In a first stage, the mechanical treatment (quenching in water from 910 °C and annealing at 455 °C for
behavior of the 38Si7 steel of the clip was characterized. To this end, one hour) followed by a shot-peening (to reduce the susceptibility of the
standardized specimens were heat treated and shot-peened (like the material to fatigue) [14]. To faithfully reproduce the response of the
actual spring clips), and subjected to tensile tests (to obtain the me- steel, two standard tensile specimens and 27 standard fatigue speci-
chanical properties) and fatigue tests (to determine the P-S-N curves, mens were machined from wire rod to determine, respectively, the
i.e., conventional fatigue S-N curves but including the probability of mechanical properties and the fatigue behavior (P-S-N curves) of the
failure, P). In a second phase, the behavior of the clip was characterized steel. After machining the specimens, one of the tensile specimens was
subjected to the aforementioned heat and shot-peening treatments at

554
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Table 1
Summary, description and purpose of the tests conducted in the study.
Type of test Type of sample Purpose

Microhardness Coupons taken from tensile specimens To compare the hardness profiles of clips and tensile specimens to Ensure the correct application of
Microhardness Coupons taken from SKL-1 clips heat treatment
Tensile Standard specimens (Φ6) after heat To obtain the mechanical properties, which are necessary for the FE model
treatment and shot-peening
Fatigue Standard specimens (Φ6) after heat To characterize the fatigue behavior of the material. Two experimental conditions (R = −1 and
treatment and shot-peening R = 0.1) were applied, to select the fatigue model (influence of mean stress)
Quasi-static compression test SKL-1 clip To validate the FE model (mechanical validation)
Fatigue test SKL-1 clip To validate the FE model (fatigue validation)

Table 2 These tests have been carried out by means of a Servohydraulic

Chemical composition of the clip (% wt.). Universal Instron 8501 test machine with a loading capacity of 100 kN
C Si Mn S P and following the requirements of the ASTM E466-96 standard [17]. To
describe the uncertainty inherent to fatigue, the confidence bands (as-
0.37–0.44 1.5–1.9 0.6 – 0.8 < 0.04 < 0.04 sociated with various probabilities of failure) of the experimental data
were determined using the statistical module of the MATLAB™ soft-
ware. For the fatigue characterization, two types of loading conditions
the same facilities employed to manufacture the clips. The non-treated have been imposed: 13 of the specimens were tested maintaining a
tensile specimen was used to observe the influence of the treatments on stress ratio R = σmin/σmax = −1 (which is the reference condition to
the material mechanical response. obtain the SN curves) while R = 0.1 was imposed for the remaining 14
For the validation of the FE models, 13 actual SKL-1 clips were samples. As will become evident later, having experimental results
manufactured for this research and tested under different conditions. obtained from two values of R (R = −1 and R = 0.1, respectively) will
One of them was subjected to a quasi-static compression test obtaining make it possible to identify the model to correct the effect induced by
the load-displacement curve. The remaining 12 clips were fatigue the mean stress during fatigue.
loaded, recording the number of cycles to failure. In both cases, the
predictions derived from the FE models were compared to the experi-
2.3. Experimental scope on skl-1 clips
mental response. Finally, a series of tests have been carried out on a
complete fastening system. The loads were applied on the rail, simu-
2.3.1. Quasi-static test
lating the passage of a train, recording various relevant displacements
As sketched in Fig. 2(a), this test consists in fixing the central region
experienced by the whole assembly that were subsequently employed
of the clip (which is in contact with the bolt) and compressing its ends
as boundary conditions for a FE model of the system. In light of the
(in contact with the rail), applying the force F, to obtain the load-dis-
experimental and numerical results obtained, an assessment of the
placement curve of the component. The test was carried out under load
possible fatigue failure of a fastening spring clip under in-service con-
control by means of an Instron 8500 Universal servohydraulic machine,
ditions has been carried out.
with a loading capacity of 250 kN and following the procedure [13].
First, the clip is compressed to a maximum force of 25 kN and then
2.2. Experimental scope on standard specimens completely unloaded. Fig. 2(b) shows a photograph of the test carried
out where the LVDT used to record the displacements can be seen. The
2.2.1. Microhardness tests experimental results obtained from this test were employed as a first
To verify the representativeness of the experimental results derived criterion to validate the FE model (see Section 3.2.1).
from tensile and fatigue tests performed on standard specimens, it is
necessary to guarantee that their final mechanical properties, after
2.3.2. Fatigue tests
thermal and shot-peening treatments, agree with those of the actual
The boundary conditions of the fatigue tests consists in fixing the
SKL-1 clips. For this reason, small coupons were extracted from the ends
central region of the clip and compressing the ends, see Fig. 3. Fatigue
of the tensile and fatigue samples and Vickers microhardness profiles
tests were carried out by means of a servohidraulic Universal Instron
were obtained. This same procedure has been carried out on coupons
8501 testing machine, with a loading capacity of 100 kN, following the
extracted from the ends of the W-type clips. Subsequently, the results
technical specification ET 03.360.566.8-Clips-ADIF [13]. The average
have been compared to verify the representativeness of the material
force was Fm = 2.5 kN and the force amplitude Fa = (Fmax − Fmin)/2
constituting the standardized specimens. The microhardness tests have
ranged between 2 and 1.2 kN, see Table 3. As explained above, the
been carried out with a Neurtek microdurometer model Qness Q10-30-
validation of the FE model requires reproducing these results from the
60, applying a load of 25 gf for 20 s following the procedure established
mechanical and fatigue properties of the material (see Section 3.2.2).
in the standard ASTM E 92-17 [15].

2.3.3. Reproduction of in-service conditions

2.2.2. Tensile tests Fig. 4(a, b) shows the experimental assembly used to reproduce the
The tensile tests have been carried out in an Instron 8501 Universal in-service conditions of the SKL-1 fastening system. It is made up of a
servo hydraulic machine with a loading capacity of 100 kN on speci- prestressed concrete sleeper, a segment of a rail and the SKL-1 fastening
mens with a diameter of 6 mm, following the procedure of the standard system that links both. The tests carried out are based on the specifi-
ASTM E 8M-16 [16]. cations [18,19] and consist of two parts. Firstly, the bolts are tightened,
recording their vertical displacement. Then, the test consists in re-
2.2.3. Fatigue tests peating a compressive cyclical sinusoidal load with constant amplitude
The fatigue tests on specimens with a diameter of 6 mm have al- for 3·106 cycles at a frequency of 5 Hz, a minimum load of 5 kN and a
lowed the P-S-N curves of the material to be obtained; that is, the curves maximum load of 70 kN. The angle between the load line and the
that relate the number of cycles to fracture (N) as a function of the vertical line was 26° (simulating the conditions of passage of a railway
stress amplitude (S ≡ σa) applied for a given failure probability (P). vehicle in a curve). The tests were performed with a bench equipped

555
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 2. (a) Schematic description of the boundary conditions imposed in the quasi-static test of the clip. (b) Picture showing the performance of the test on an SKL-1
spring.

cycles. In this experimental campaign, 12 fatigue tests of this type have

been carried out.

2.4. Statistics

The MATLAB Curve Fitting Toolbox™ software allows the con-

fidence bounds for the fitted coefficients or the prediction bounds for
new observations to be calculated. A confidence band is used in sta-
tistical analysis to represent the uncertainty in an estimate of a fitting
curve based on limited or noisy data [20]. Confidence bounds represent
the probability that the fitted function is within those limits. In contrast,
a prediction bound is used to represent the uncertainty about the value
of a new data-point of the population [20]. From the S-N curves ob-
tained experimentally (Section 3.1.3), the corresponding prediction
bounds for different probabilities of failure (1%, 5%, 95% and 99%;
respectively) have been calculated to obtain in this way the so-called P-
S-N curves [21]. These curves were implemented in ANSYS in order to
Fig. 3. Schematic description of the boundary conditions imposed in the fatigue develop a statistical assessment of the fatigue performance of the SKL-1
tests of the clips. clips.

Table 3 2.5. Finite elements numerical simulation

Parameters of the fatigue tests carried out of SKL-1 spring clips.
Fa (kN) Fmax (kN) Fmin (kN) # tests FE numerical simulations were developed in this study by means of
the software ANSYS Workbench 18.0. Fig. 5 shows the components
2 4.5 0.5 2 included in the model. The mechanical and fatigue properties of the clip
1.8 4.3 0.7 2
correspond to the experimental results obtained through tensile and
1.7 4.2 0.8 2
1.6 4.1 0.9 2 fatigue tests, including in this case the prediction bounds.
1.3 3.8 1.2 2 The tensile test provides the stress-strain curve expressed in en-
1.2 3.7 1.3 2 gineering variables, σ and ε. The true variables, σt and εt, were obtained
applying the well-known transformation formulas (1, 2) to the points of
the curve prior necking. A multilinear kinematic hardening material
with a dynamic servo-hydraulic actuator, brand INSTRON series 8505, was used to reproduce this curve in ANSYS.
with a loading capacity of ± 100 kN. During the test, the vertical dis-
placements undergone by the rail foot (because of the turning of the σt = σ (1 + ε ) (1)
rail) on the side corresponding to the outside of the curve (which is
εt = ln (1 + ε ) (2)
where the fatigue loading of the clip is more damaging) were con-
tinuously recorded by means of two LVDTs, see Fig. 4(a). In addition, The screw and the rail were modelled as Structural Steel (material
the horizontal displacements of the head of the rail were also measured included in the database provided by ANSYS Workbench). Finally, the
with another couple of LVDTs, see Fig. 4(b). The vertical displacements mechanical properties of the guide plate, which is fabricated in poly-
have been introduced into the FE model of the system to define the amide 66 reinforced with short glass fiber (35 wt%), were taken from a
boundary conditions (mean displacement, δm, and amplitude dis- previous research by the same authors [22].
placement, δa), as sketched in Fig. 4(c). According to the specification The final mesh consists of 190,060 nodes and 49,097 elements. A
[18], the test is considered acceptable if the system exceeds 3·106 sensitivity analysis was carried out to select elements size so that the

556
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 4. (a, b) Pictures showing the experimental setup to reproduce the in-service conditions of the fastening system. (c) Schematic description of the boundary
conditions imposed in the FE model.

models will yield accurate results with acceptable computing time. 3. Results
According to FE theory, stresses are calculated from displacements and,
for this reason, the sensitivity analysis was based on the Von Mises 3.1. Experimental results on standard specimens
stress. The size of the elements was reduced until the variation of the
Von Mises stress was below 5%. 3.1.1. Microhardness
Each of the components (clip, bolt, rail and guide plate) was meshed Fig. 6 shows a comparison between the Vickers microhardness va-
independently, but taking into account the mutual interactions (contact lues obtained along the 6 mm diameter of a cross-section of a tensile
regions). The general maximum size of the elements was 3 mm and a specimen and the 13 mm diameter of a SKL-1 clip. As can be seen, the
hex dominant method was applied to the whole model. Additionally, a hardness values are virtually indistinguishable. This result underlines
Multizone method was used on the bolt, the plate and the rail, aiming at the fact that the material of the tensile and fatigue specimens and that
obtaining a regular mesh. A face meshing was added to the plate to of the clip present the same microstructural nature and guarantees the
ensure the regularity of the elements in the area where the contact with representativeness of the experimental results. In addition, the surface
the clip occurs. The mesh was densified in the contact regions between condition of specimens and clips is the same, as they were subjected to
bodies, imposing a maximum element size of 1.5 mm. Three groups of the same shot-peening treatment (in the same facilities, indeed).
contacts have been defined in the model between the following bodies:
clip-bolt, clip-plate and clip-rail. In all cases, the contact was defined as
3.1.2. Tensile tests
a frictional contact with a coefficient of friction of 0.2. To avoid the
The stress-strain curves are reproduced in Fig. 7. As indicated pre-
interpenetration of the bodies, guaranteeing the contact compatibility,
viously, one of the specimens was tested without receiving any treat-
Augmented Lagrange formulation was used.
ment (which is referred to as “non-treated”) while the second had re-
ceived the quenching and tempering heat treatment as well as the shot-
peening (“treated” specimen in the figure). The relevant mechanical
parameters are summarized in Table 4. The differences between the two
curves are noteworthy: thanks to the applied treatments, a substantial
increase in the resistant ability of the material occurs (expressed in
terms of yield stress or tensile strength, which increase by 148% and

557
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 5. Description of the meshing of the different parts that compose the fastening system. (a) General perspective. (b) Contact regions where the density of elements
was increased. (c–f) show the meshing of the head of the bolt, the guide plate, the rail and the SKL-1 clip, respectively.

71%, respectively). The high value of the yield stress is an indispensable R = −1) and σm ≠ 0 (R = 0.1 in this case). In each case, a downward
requirement for a spring clip since an efficient performance is based on sloping and a horizontal region are observed. These regions follow a
its capacity to resist loads in the linear elastic regime. This improve- straight line in logarithmic scale. The equation of the line in the ne-
ment in strength takes place at the expense of a reduction in ductility gative slope region is known as the Basquin model. The horizontal line
expressed in terms of the strain under maximum load (−63%), al- represents the fatigue (or endurance) limit of the material, σe; below σe
though the ultimate strain is virtually not affected (in both cases, the life of the component is assumed to be infinite. The Basquin
∼25%). equations as well as the corresponding coefficients of determination,
R2, are included in the figures. The comparison between Fig. 8(a) and
3.1.3. Fatigue tests (c) allows the influence derived from the mean stress to be understood;
Fig. 8(a) and (b) shows the experimental data obtained after testing thus, the number of cycles to failure decreases as the stress ratio, R,
the fatigue specimens with, respectively, mean stress σm = 0 (or increases (or, equivalently, as the mean stress, σm, increases). This

558
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

σm ≠ 0 (σa(σm≠0)) that produces the same number of cycles until

failure. The expressions of the three models, (Eqs. (3)–(5)), are re-
produced next

σ
Soderberg: σa (σm≠ 0) = σa (σm= 0) ⎛1 − m ⎞
⎜ ⎟

⎝ σY ⎠ (3)

σ
Goodman: σa (σm≠ 0) = σa (σm= 0) ⎛1 − m ⎞
⎜ ⎟

⎝ σ TS ⎠ (4)

2
Gerber: σa (σm≠ 0) = σa (σm= 0) ⎡ ⎛ σm ⎞ ⎤
⎢1 − σTS ⎥
⎜ ⎟

⎣ ⎝ ⎠⎦ (5)

where σY is the material’s yield stress and σTS the tensile strength.
There is no universal criterion to decide what correction must be
Fig. 6. Comparison between the Vickers microhardness profiles obtained in one used for a certain metallic material. It is assumed that the Soderberg
tensile specimen and in one SKL-1 clip (after subjecting both of them to the criterion is overly conservative, that Goodman works well for brittle
same heat and shot-peening treatments). metals while Gerber is suitable for ductile metals. In this paper, a novel
procedure is proposed to select the most suitable corrective method for
the steel of the SKL-1 clip based on the experimental results obtained
from the fatigue tests.
First, the stress amplitudes for σm ≠ 0 have been corrected using the
Soderberg, Goodman and Gerber expressions, respectively. Then, using
the Basquin fitting for σm = 0, the predictions of the number of cycles
until failure were obtained. Both the experimental results (with R = −1
and R = 0.1) and the corrected values are shown in Fig. 9(a). As can be
seen, the experimental data with R = −1 fall between the Goodman's
and Gerber's predictions. The novel solution proposed in this study
consists in carrying out a linear combination of the Goodman and
Gerber methods, with the form “α·Goodman + β·Gerber”, where
β = 1 − α. Then, the value of α has been obtained through a least
squares procedure, obtaining α = 0.506 (and, consequently,
β = 0.494). Fig. 9(b) shows the result obtained after applying this
correction on the data with R = 0.1: as can be observed, this procedure
Fig. 7. Stress-strain curves obtained from a non-treated and a heat and shot- provides results that are virtually indistinguishable from the experi-
peened treated specimen. mental values obtained with R = −1.

Table 4 3.2. Experimental results on clips

Mechanical properties of the non-treated and treated 38Si7 steel.
Non-treated Treated Variation (%)
3.2.1. Quasi-static test
The experimental curve relating force and displacement is displayed
Yield Stress (MPa) 434 1077 148 in Fig. 10(a). Initially, the clip responds elastically until a force of
Tensile Strength (MPa) 756 1291 71 ∼8 kN; then, the plastic behavior begins. For a displacement of
Strain under maximum stress 0.139 0.052 −63
∼15 mm, the contact between the rail and the central region of the clip
Ultimate strain 0.265 0.257 −3
occurs (see Fig. 11); for this reason, the slope of the curve notably in-
creases. It is worth noting that this feature is not accidental but a
phenomenon is seen in the downward sloping region as well as in the consequence of the design of the clip, which is aimed at preventing the
fatigue limit, σe, of the material. For R = −1, σe = 301 MPa and for overturning of the rail. After unloading, some plastic deformations re-
R = 0.1 it reduces to σe = 291 MPa. The intersection between the main, given that the elastic limit had been exceeded during loading.
Basquin and the endurance regions occurs for Nf ∼ 490,000 cycles; this The comparison between the experimental results and the predictions
means that, disregarding the intrinsic uncertainties of fatigue, once a obtained from the FE model is displayed in Fig. 10(b). As can be seen,
component subjected to R = −1 uniaxial stresses exceeds this number the curve obtained numerically reproduces the experimental curve. In
of cycles, its life is infinite. Fig. 8(c) and (d) shows the prediction this way, it is concluded that the mechanical validation of the numer-
bounds (for the failure probabilities 1%, 5%, 95% and 99%, respec- ical model has been achieved satisfactorily.
tively) of the experimental data. The curves displayed in Fig. 8(c) re- Fig. 11 shows the von Mises stress field for a compression load of
present the P-S-N curves of the material, which were implemented in 25 kN (note the contact between the central part of the clip and the
ANSYS Workbench to provide a statistical analysis of the experimental rail). This figure allows the regions where the concentration of stresses
information. occurs to be identified, which correspond to the loops of the SKL-1
The specialized scientific literature collects various empirical spring clip. The maximum stress in the figure is 1511 MPa, but notice
models for metallic materials to correct the effect of the mean stress on that it occurs locally, in the contact between the ends of the clip and the
fatigue. The three most used models (which are also implemented in the rail; the maximum stress in the loops of the clips, which are depicted in
fatigue module of ANSYS Workbench) are those of Soderberg, Goodman the figure (≈1300 MPa), exceeds the yield stress of the material
and Gerber [23,24]. These models provide a relationship between the (1077 MPa, see Table 4). This result is consistent with the existence of
stress amplitude with σm = 0 (σa(σm=0)) and the stress amplitude with residual deformations detected in the compression test (see Fig. 10)
after unloading.

559
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 8. Experimental results obtained from the fatigue characterization, (a) R = −1, (b) R = 0.1, and the corresponding prediction bounds, (c) and (d), respectively.

3.2.2. Fatigue tests SKL-1 clips under in-service conditions to be assessed. Fig. 12(b) shows
The experimental results derived from the fatigue tests on spring a comparison between the experimental data and the predictions ob-
clips are collected in Fig. 12(a). Note that the points (Nf, Fa) follow the tained from the FE model. The fatigue module of ANSYS Workbench
same pattern as the S-N curves obtained on standard specimens which receives as input the S-N curves of the material. In order to carry out a
consists of a downward sloping and a horizontal region. The loads for statistical evaluation, the fatigue curves associated with different
these fatigue tests (Fm, Fa) were selected so that the material behaved in probabilities of failure have been introduced in the model. Besides, the
the linear elastic regime, which corresponds to the high cycle fatigue procedure for correcting the effect of the mean stress described in
regime, as stated by Mohammadzadeh et al. [11]. The coordinate of the Section 3.1.3 has been applied. Finally, to take into account the stress
horizontal line corresponds to the fatigue limit which is Fe = 1200 N (as triaxial state in the clip, the von Mises stress has been used. The level of
explained above, this value depends on the mean load during fatigue, agreement obtained is highly satisfactory: the experimental data are
Fm = 2500 N, in this case). One of the goals of this study is to develop scattered around the line corresponding with a failure probability of
and validate a numerical tool that allows the response to fatigue of the 50% and within the prediction bands. In addition, the model predicts

Fig. 9. (a) Experimental data obtained from the fatigue tests (with R = −1 and R = 0.1) as well as the corrected data after applying the Soderberg, Goodman, and
Gerber corrections. (b) Comparison between R = −1 data and R = 0.1 corrected using a linear combination of Goodman’s and Gerber’s models.

560
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 10. (a) Experimental curve obtained from the compression test of an SKL-1 spring clip. (b) Comparison between experimental and FE results.

well the fatigue limit of the clip. In conclusion, the FE model has been clips of the SKL-1 railway fastening system [1]. This is an element of
validated as a tool to evaluate the fatigue behavior of the SKL-1 clip. high responsibility since its failure might imply the derailment of a
train, with potentially fatal consequences for passengers and, in any
case, a substantial economic loss. This study, which arises from these
3.2.3. Reproduction of in-service conditions
premises, is aimed at evaluating the conditions of structural integrity of
As noted above, the result of this fatigue test is considered sa-
the SKL-1 spring clips under in-service conditions.
tisfactory if the component resists 3·106 cycles. In this study, 12 fatigue
One of the novel contributions of this study lies in the method ap-
tests have been carried out and, in all cases, the system has verified this
plied to model the fatigue behavior. It is a well-known fact that the
condition. After completing the tests, each of the clips has been care-
fatigue life of a structural element depends not only on the stress am-
fully inspected seeking any evidence that might indicate the existence
plitude but also on the mean stress. The results of this study prove that
of an incipient process of fatigue. No evidence of this type been ob-
the fatigue behavior of the constitutive steel of the SKL-1 spring clips is
served. The FE model developed applying the boundary conditions
accurately described using a linear combination of the Goodman's and
described in Section 2.3.2 (Fig. 4(c)) allowed the maximum equivalent
Gerber's corrective methods.
(thus to say, R = −1) stress amplitude, σa(σm=0) in the clip to be de-
Both the experimental results and the FE model predictions show
termined, obtaining σa(σm=0) ≈ 134 MPa. This result is well below the
that fatigue failure of a spring clip is a highly unlikely event. This result
fatigue limit for R = −1, σe = 301 MPa and in fair agreement with the
clashes head on with the evidence presented by Lakusic et al. [1]. Ex-
experimental results: as indicated previously, none of the assemblies
perimentally, it has been verified that none of the tests carried out re-
subjected to fatigue tests reproducing the in-service conditions, un-
producing the in-service conditions has led to the failure of the com-
derwent any failure nor showed any indication of fatigue damage.
ponent, despite having subjected each clip to 3·106 cycles. In fact, the
beginning of the endurance region of the fatigue curve can be ap-
4. Discussion proximately established in Nf ≈ 490,000 cycles; therefore, the experi-
mental conditions imposed by the specifications [18,19] guarantee that
The technical literature reports cases of in-service fractures of spring

Fig. 11. Perspective of the distribution of von Mises stresses for a compression force of 25 kN.

561
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

Fig. 12. (a) Results of the fatigue tests carried out on SKL-1 spring clips. (b) Comparison between the fatigue results obtained on the spring clip and the predictions
provided by the FE model.

the theoretical lifespan of the component should be infinite. The nu- The studies conducted by Hong et al. [6,7] focus on the large
merical results agree with this outcome. Thus, the FE model shows that number of fractures observed in e-clips in China. Their results show that
the maximum corrected stress amplitude in the clip amounts to installation parameters play a key role to promote the fatigue failure.
σa ≈ 134 MPa while the endurance limit of the material is The FE model shows a region of stress concentration in the contact
σe = 301 MPa. between this type of spring clips and the steel bottom plates. During
Given this situation, it is only possible to speculate on the causes operation, the friction at this zone leads to the initiation and propa-
behind the failure of spring clips reported by Lakusic et al. [1]. In gation of cracks that results in the fatigue fracture of the fastening clip.
principle, several possibilities can be pointed out. The most important This mechanism agrees with the on-site inspections conducted on actual
variable regarding the fatigue resistance is, without a doubt, the ma- broken e-clips. Another interesting outcome suggested after the on-site
terial itself. As shown in this paper, in order to achieve adequate me- examinations is that rail corrugation was found at the section of severe
chanical and fatigue properties, it is necessary to apply rigorously a fracture of fastening clips. The dynamic study carried out by Hong et al.
thermal and shot-peening treatments. Both processes must be optimized [7] proves that rail corrugation lead to the resonance of the fastening
to avoid compromising the structural integrity of the component. For clip increasing the loads and triggering the fatigue and final fracture of
example, if the yield stress does not reach the required value, the the clip.
component will work in the plastic regime, which will inevitably reduce Mohammadzadeh et al. [12] obtained the reliability index during
its fatigue lifespan as well as its fitness for service. Moreover, a suitable the propagation stage (e.g., after the initiation of the crack) of fatigue in
shot-peening treatment is crucial to optimize the fatigue performance. SKL-14 clips. They first considered an initial crack size of 0.1 mm and
In this sense, two extreme scenarios can be contemplated. In the case of observed that the reliability index falls from 1.20 at the beginning of
underpeening, the residual compressive stresses are insufficient to the process to 0 after 13. 5 years. It is worth noting that a 95% survival
prevent the initiation of the cracking process. In contrast, under over- probability is associated with a reliability index of 1.65. They extended
peening conditions, the material may have surface cracks that would the analysis to initial crack sizes of 0.3, 0.5 and 0.7 mm, finding that the
accelerate the initiation stage of cracking reducing its lifespan. index in the first year is 0.75, 0.60 and 0.55, respectively. These indexes
Our study assumes that the assembly of the fastening system has fall to zero after approximately 7.8, 5.5 and 4.5 years. Based on these
been carried out in accordance with the specifications of the standards findings, they recommend the replacement of any clip as soon as a crack
[18,19]. These documents detail the characteristics to be satisfied is detected. Therefore, the time elapsed between the initiation of the
during the prestressing of the clip as well as the testing loads, which are crack and the final fracture is extremely small, which makes it difficult
considered representative of the operating conditions. However, to detect the propagation during an inspection.
nothing avoids the existence of in-service overloads that would reduce We do not wish to end this paper without highlighting the fact that
the fatigue life of the spring clips. Regardless of the above, the only the FE numerical model developed and verified in this research would
procedure to elucidate the causes of failures reported by Lakusic et al. be extensible to any other spring clip geometry, allowing the design
[1] would be to carry out a rigorous forensic analysis, based on realistic stage to be optimized by decreasing deadlines and the costs associated
information about the material properties and in-service loads, in- with laboratory tests.
cluding a fractographic evaluation using SEM microscopy.
The role of in-service overloads was studied by Hasap et al. [10] 5. Conclusions
who analyzed the fatigue failure of e-clips and the influence of wheel
and toe loads. They took into consideration that under actual operation An exhaustive research has been carried out including the following
conditions of a railway track system the impact caused by irregularities sections:
of rails and wheels (due to weldments, wear, corrugated rails, etc.) can
add up to 60% to the wheel load. They found that, for normal wheel • Characterization of the mechanical and fatigue properties of the
loads, the fatigue lifespan of the clips is infinite (the stress amplitude is steel of the SKL-1 spring clips. To ensure the representativeness of
below the endurance limit). In contrast, impact loading reduces the the results, the specimens were subjected to the same thermal and
number of cycles to failure to 16,839 and 5468 under normal and high shot-peening treatments as the actual spring clips (in fact, these
toe loads, respectively. Disregarding the differences between studies, treatments were applied in the same industrial facility). The fatigue
the conclusions reached by Hasap et al. [10] agrees with ours since in performance of the material was analyzed following a probabilistic
both cases exceptional loading conditions are required to produce the description of the S-N curves (referred to as P-S-N approach).
fatigue failure. • Characterization of the mechanical (compression test) and fatigue

562
D. Ferreño, et al. Engineering Structures 188 (2019) 553–563

response until failure of a series of spring clips. Eng 2018;10:1–12. https://doi.org/10.1177/1687814018767249.

• Development and optimization of a FE model, using ANSYS [7] Hong X, Jia-bin W, Yan-Rong Z. The fractures of e-type fastening clips used in the
subway: theory and experiment. Eng Fail Anal 2017;81:57–68. https://doi.org/10.
Workbench, to reproduce the structural behavior of the SKL-1 fas- 1016/j.engfailanal.2017.07.006.
tening system (including the clip, the foot of the rail, the guide plate [8] Chinese National Standard - GB/T1222-2007:2007. Spring steels., n.d.
and the bolt). This numerical model has been exhaustively vali- [9] Yuan F, Liang B, Le J. Elastic-plastic mechanics. Devon: The yellow river water
conservancy press.; n.d.
dated, both from the mechanical and fatigue viewpoints, using the [10] Hasap A, Paitekul P, Noraphaiphipaksa N, Kanchanomai C. Influence of toe load on
experimental results described in the previous paragraph. the fatigue resistance of elastic rail clips. Proc Inst Mech Eng Part F J Rail Rapid
Transit 2017;232:1078–87. https://doi.org/10.1177/0954409717707834.
[11] Mohammadzadeh S, Ahadi S, Nouri M. Stress-based fatigue reliability analysis ofthe
The experimental results and the FE model prove that the fatigue rail fastening spring clip under traffic loads. Lat Am J Solids Struct
failure of a spring clip under normal conditions is a very unlikely event. 2014;11:993–1011. https://doi.org/10.1590/S1679-78252014000600006.
Therefore, the experimental evidence regarding in-service failures of [12] Mohammadzadeh S, Ahadi S, Keshavarzian H. Assessment of fracture reliability
analysis of crack growth in spring clip type Vossloh SKL14. Proc Inst Mech Eng Part
SKL-1 clips provided by some authors [1] must be explained on dif-
O J Risk Reliab 2014;228:460–8. https://doi.org/10.1177/1748006X14527926.
ferent grounds. [13] Grupo de Trabajo GT-202. Sujeciones. ET 03.360.564.3 - Clips. Madrid, Spain:
ADIF; 2017. p. 27.
Acknowledgments [14] REDALSA. Especificaciones clip elástico SKL-1 n.d. http://redalsa.com/productos/
clip-elastico-skl-1/.
[15] ASTM E92 - 17. Standard test methods for Vickers hardness and knoop hardness of
This paper describes the work performed by the Laboratory of metallic materials. West Conshohocken, PA: ASTM International; 2017http://doi.
Science and Engineering of Materials (LADICIM) of the University of org/10.1520/E0092-17.
[16] ASTM E8/E8M - 16. Standard test methods for tension testing of metallic materials.
Cantabria (Spain). The authors would like to express their gratitude to ASTM B. Stand., West Conshohocken, PA, 2017. ASTM International; 2016. p.
the companies Global Steel Wire (GSW), who contributed the rod wire 1–27http://doi.org/10.1520/E0008_E0008M-16A.
to machine the specimens, and REDALSA, which was responsible for [17] ASTM E466 - 15. Standard practice for conducting force controlled constant am-
plitude axial fatigue tests of metallic materials. B. Stand. Vol. 03.01. West
carrying out the heat and shot-peening treatments on the specimens as Conshohocken, PA: ASTM International; 2015http://doi.org/10.1520/E0466-15.
well as supplying the SKL-1 clips for their characterization. [18] British Standard Institution. BS EN 13481-2:2012+A1:2017. Railway applications.
Track. Performance requirements for fastening systems - Part 2: Fastening systems
for concrete sleepers. London: BSI; 2017. p. 22.
References [19] British Standard Institution. BS EN 13146-4:2012+A1:2014. Railway applications.
Track. Test methods for fastening systems. Effect of repeated loading. London: BSI;
[1] Lakusic S, Bartos D, Bajic AG. Experimental and numerical analysis of the rail fas- 2014. p. 20.
tening spring clips types SKL-1 and SKL-12. 22nd Danubia-Adria Symp. Exp Meth [20] Confidence and prediction bands 2018. https://en.wikipedia.org/wiki/Confidence_
Solid Mech 2005. and_prediction_bands (accessed January 21, 2019).
[2] Kang C, Schneider S, Wenner M, Marx S. Development of design and construction of [21] Huang BT, Li QH, Xu SL, Zhou BM. Tensile fatigue behavior of fiber-reinforced
high-speed railway bridges in Germany. Eng Struct 2018;163:184–96. https://doi. cementitious material with high ductility: experimental study and novel P-S-N
org/10.1016/j.engstruct.2018.02.059. model. Constr Build Mater 2018;178:349–59. https://doi.org/10.1016/j.
[3] Paixão A, Varandas JN, Fortunato E, Calçada R. Numerical simulations to improve conbuildmat.2018.05.166.
the use of under sleeper pads at transition zones to railway bridges. Eng Struct [22] Ferreño D, Carrascal I, Ruiz E, Casado JA. Characterisation by means of a finite
2018;164:169–82. https://doi.org/10.1016/j.engstruct.2018.03.005. element model of the influence of moisture content on the mechanical and fracture
[4] Imam BM, Righiniotis TD, Chryssanthopoulos MK. Numerical modelling of riveted properties of the polyamide 6 reinforced with short glass fibre. Polym Test
railway bridge connections for fatigue evaluation. Eng Struct 2007;29:3071–81. 2011;30:420–8. https://doi.org/10.1016/j.polymertesting.2011.03.001.
https://doi.org/10.1016/j.engstruct.2007.02.011. [23] Susmel L, Tovo R, Lazzarin P. The mean stress effect on the high-cycle fatigue
[5] Cheng B, Cao X, Ye X, Cao Y. Fatigue tests of welded connections between long- strength from a multiaxial fatigue point of view. Int J Fatigue 2005;27:928–43.
itudinal stringer and deck plate in railway bridge orthotropic steel decks. Eng Struct https://doi.org/10.1016/J.IJFATIGUE.2004.11.012.
2017;153:32–42. https://doi.org/10.1016/j.engstruct.2017.10.016. [24] Handbook ASM. Volume 19: Fatigue and Fracture. Novelty, Ohio: ASM
[6] Hong X, Xiao G, Haoyu W, Xing L, Sixing W. Fatigue damage analysis and life International; 1996.
prediction of e-clip in railway fasteners based on ABAQUS and FE-SAFE. Adv Mech

563