Vrije Universiteit Brussel

Fatigue weld crack detection using distributed fiber optic strain sensing
Mikhailov, Sergei; van Wittenberghe, Jeroen; Luyckx, Geert; Thibaux, Philippe; Geernaert,
Thomas; Berghmans, Francis
Published in:
International Journal of Fatigue

DOI:
10.1016/j.ijfatigue.2023.107851

Publication date:
2023

License:
CC BY-NC-ND

Document Version:
Final published version

Link to publication

Citation for published version (APA):

Mikhailov, S., van Wittenberghe, J., Luyckx, G., Thibaux, P., Geernaert, T., & Berghmans, F. (2023). Fatigue
weld crack detection using distributed fiber optic strain sensing. International Journal of Fatigue, 176(107851), 1-
9. [107851]. https://doi.org/10.1016/j.ijfatigue.2023.107851

Copyright
No part of this publication may be reproduced or transmitted in any form, without the prior written permission of the author(s) or other rights
holders to whom publication rights have been transferred, unless permitted by a license attached to the publication (a Creative Commons
license or other), or unless exceptions to copyright law apply.

Take down policy

If you believe that this document infringes your copyright or other rights, please contact [email protected], with details of the nature of the
infringement. We will investigate the claim and if justified, we will take the appropriate steps.

Download date: 24. apr. 2025

 International Journal of Fatigue 176 (2023) 107851

Contents lists available at ScienceDirect

International Journal of Fatigue

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

Fatigue weld crack detection using distributed fiber optic strain sensing
Sergei Mikhailov a, *, Jeroen van Wittenberghe b, Geert Luyckx c, Philippe Thibaux b,
Thomas Geernaert a, Francis Berghmans a
a
Department of Applied Physics and Photonics, Brussels Photonics (B-PHOT), Vrije Universiteit Brussel and Flanders Make, Brussels B-1050, Belgium
b
OCAS NV, Zelzate, Belgium
c
COM&SENS BVBA, Nazareth, Belgium

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

Keywords: We demonstrate the application of distributed fiber optic strain sensing based on optical frequency-domain
Fatigue testing reflectometry for early detection and location of fatigue cracks in welds in steel tubular test specimens. To do
Crack detection so, we have subjected welded tubular specimens instrumented with surface-mounted optical fiber sensors to
Distributed optical fiber sensor
resonant bending load, and we have measured the strain distributions in the test samples continuously and
Optical frequency-domain reflectometry
without any interruption of the test throughout its whole duration, with a 2.6 mm spatial resolution and a 1 με
Distributed strain sensing
strain resolution. We show that the fatigue cracks, which initiate from the inner surface of the wall of the
specimens, can be successfully detected and located in real time using our measurement method. We conclude
that the detection of the crack initiation may provide relevant information serving the estimation of the
remaining lifetime of the component.

1. Introduction local stage to warn about the degradation of a structure.

It is possible to monitor the crack growth at a weld location on a
Welding is a well-established fabrication process that is often used component by installing strain gauges perpendicular to the weld and to
for assembling metallic structures. The quality of the welds is of crucial follow the evolution of the strain signals [12,13]. The detected strain
importance to the overall fatigue characteristics of such structures. increases when the crack tip approaches the strain gauge, reaches a
Monitoring the fatigue response of a metallic structure is tradition­ maximum value when the crack tip is approximately in front of the
ally carried out using a global response system [1–4], for example, by strain gauge, and then decreases when the crack has passed the strain
means of an accelerometer fixed to the deck of a foundation or with gauge. One strain gauge can thus provide information about the inter­
strain gauges positioned close to the bottom of the monopile of a wind action with a crack at a certain location, with a detection range of the
turbine. Such measurements, however, provide information only on the order of the thickness of the material. The installation of the sensor re­
global behavior of the structure or on the total loads. It is therefore very quires connecting each strain gauge to the data acquisition system with
difficult to assess the integrity of the structure in detail based on these typically three wires per gauge. If one cannot anticipate the potential
data: one is often limited to verifying that the loads remain within the failure location, then the number of sensors needs to be increased in
design specifications throughout the usage time of the structure [5,6]. view of detecting the initial failure and enabling follow-up on the crack
Fatigue damage in metal structures is commonly caused by cyclic growth and the degradation of the structure.
loads at levels far below the failure load [7–9], and it is well-known that In a laboratory setting, locating crack initiation on-the-fly during a
local failures can occur before the global response is affected [10]. (cyclic) loading test is particularly interesting. This is possible with
Laboratory tests have shown, for example [11], that breakthrough discrete sensors such as strain gauges, but again their installation can be
cracks during a fatigue test on tubular components generally occur complex, whilst the spatial resolution remains limited. An alternative is
before any significant modification of the behavior of the component to combine acoustic emission to detect a cracking event with an in­
can be identified. For example, a breakthrough crack does not neces­ spection of the specimen during an interruption of the test for locating
sarily greatly affect the stiffness of a tube. This prompts the need for the crack. Such an inspection typically relies on dye penetrant inspection
detection and monitoring systems that operate at an earlier and more (DPI) [14] or magnetic particle inspection (MPI) [15] for surface cracks.

* Corresponding author.
E-mail address: [email protected] (S. Mikhailov).

https://doi.org/10.1016/j.ijfatigue.2023.107851
Received 27 April 2023; Received in revised form 21 June 2023; Accepted 18 July 2023
Available online 19 July 2023
0142-1123/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

Fig. 1. (a) Schematic illustration and (b) photograph of the HIPNOS test setup with installed test specimen, (c) photograph of the specimen instrumented with an
optical fiber sensor and strain gauges, (d) instrumentation scheme of the test specimen (the yellow line represents the optical fiber portion that acts as a sensor and
the orange rectangles represent the electrical strain gauges).

Whilst this provides information about the width of a crack at a given 2. Method
moment in time, it does not reveal the evolution of a crack, unless
repeated inspections are carried out. Furthermore, if the crack initiates 2.1. Experimental setup, test specimen and instrumentation
from the inside of a component and propagates towards the outer sur­
face, such intermediate inspections would not be possible. The fatigue tests were carried out on the HIPNOS resonant bending
A system that allows detecting a hidden crack before it breaks at the fatigue test setup located at the testing laboratory of OCAS NV in
surface and recording the size of the crack over time would clearly be Zwijnaarde, Belgium. This setup is illustrated in Fig. 1 (a) and Fig. 1 (b).
beneficial. In this paper, we demonstrate an on-the-fly fatigue crack It allows subjecting a tubular test sample to resonant bending load by a
detection method during resonant bending load fatigue tests, which cyclic excitation force at a frequency close to the first eigenfrequency of
exploits distributed optical fiber sensing technology. The distributed the test specimen. In addition, the specimen can be internally pressur­
strain sensing technique is based on optical frequency-domain reflec­ ized to create a wide range of combined loading conditions. Owing to
tometry (OFDR). We apply this technique to two tubular steel speci­ the use of the resonance principle, the maximal range of the applied
mens, and we demonstrate its applicability both for detecting and bending stress at the center of the test sample is virtually limitless.
locating crack initiation and its potential for continuous monitoring of Typical bending stress range values are between 50 MPa and 300 MPa.
crack propagation. At the midpoint of the test sample, a pipe joint is typically present. This
The paper is structured as follows. In Section 2, we describe the joint can be a girth weld, a threaded connection, a bolted flange, or any
experimental setup, the test specimens, and the test load scenario, other type of joint.
together with the principle of OFDR-based distributed strain sensing. We During the fatigue test, an excitation force is generated by the
also address the data processing algorithm. In Section 3, we describe the excenterweight attached to the end of the test sample. The excenter­
results of the fatigue tests and post-mortem analysis of the specimens. weight is powered via a cardan shaft by an electrical motor. The
Section 4 closes our manuscript with a summary and conclusions. excenterweight is bolted to the end of the test sample. The same
connection is used to attach a compensating endweight to the opposite
end of the test sample. The first bending mode of the test sample with

2
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

attached endweights is represented by the dashed line in Fig. 1 (a). Two

supports are positioned at the locations of the nodes of the excited
eigenmode. This minimizes the dynamic forces that are transmitted to
the supports. The tests are controlled by and the data are logged on an
integrated control and monitoring unit. Test samples can be instru­
mented with displacement sensors, accelerometers, and strain gauges.
The test was conducted on two X65 grade steel tubular specimens (as
shown in Fig. 1 (c) and Fig. 1 (d)). The specimens (273 mm outer
diameter × 20 mm wall thickness × 5 m length) featured a central girth
weld and were designed such that the fatigue cracks would initiate on
the inner surface of the weld. The specimen length was selected such Fig. 2. Data processing of the OFDR data. The blue line represents the
that the eigenfrequency of the sample (~26 Hz) was in the frequency instantaneous strain, the solid red lines indicate the maximal strain εmax and the
range of the test setup. Inside failure was promoted by improving the minimal strain εmin, and the dashed red lines give the average strain εDC over
each 1-sec time interval. The instantaneous strain corresponds to a cyclic load
surface finish on the outer surface, i.e., by grinding the weld cap flush on
applied at a 46.9 Hz frequency.
the outer wall. Such welds are optimized for steel catenary risers, which
connect a pipeline or a piping system laying on the seabed to a platform
floating on the surface. The environmental changes can be obtained from the cross-correlation
The test specimens were filled with water and pressurized. The of two recorded fiber fingerprints: one fingerprint is then considered a
pressure was monitored during the test. When the crack broke through reference and the second a measurement [19]. OFDR allows for dynamic
the surface, water leaked out and the internal pressure dropped. This distributed sensing with a sub-µε strain resolution and a temperature
pressure drop was detected by the monitoring system, which in its turn resolution of 0.01–0.1 K, with sub-cm spatial resolution over ~ 100 m
then halted the test. sensing distances. OFDR has already been used for crack monitoring in
Prior to installing the optical fiber-based sensor, the surface of each concrete structures [20,21], in adhesive joints between carbon fiber
specimen was prepared using a Bristle Blaster® surface preparation tool reinforced polymer adherends [22], as well as in cast iron pipes [23] and
and then cleaned with isopropyl alcohol. Each sensor, which simply stainless steel coupons [24]. However, the aforementioned measure­
consists of a length of optical fiber, was then mounted circumferentially ments were carried out under static or low-frequency loads, and real-
onto the surface of the specimen, such that the fiber was placed on top of time crack monitoring based on distributed strain sensing has not yet
the weld, as illustrated in Fig. 1 (c), which shows a test specimen been demonstrated in dynamic fatigue tests.
instrumented with an optical fiber sensor and strain gauges. The fatigue In our measurement campaign, we used a Luna Innovations ODiSI
crack propagates through the wall thickness with the crack’s length 6102 interrogating device that allows for a strain resolution of 1 με, a
parallel to the fiber. The sensor measured the hoop strain. With this spatial resolution of 2.6 mm and an acquisition rate of 250 Hz [25],
installation method, the spatial resolution of the measurement was enabling real-time strain measurements. This acquisition rate was suf­
defined by that enabled by the fiber sensor interrogation device (see ficient to ensure an accurate identification of the instantaneous values of
Section 2.2). Note that to install the sensor, we first pretensioned the the dynamic strains induced by the cyclic loads applied to the test
fiber using two UV-curable glue dots at both ends of the sensor part of specimens, with frequencies of 26.6 Hz (Sample 1) and 25.8 Hz (Sample
the optical fiber. This ensured that every point of the sensor was equally 2). The length of the sensor fibers was limited to 2.5 m. Given the 2.6
pretensioned in one step. The sensor was then attached to the specimen mm spatial resolution, this results in an equivalent of ~ 950 sensing
surface using the same UV-curable glue. As mentioned above, we also points. These sensing fibers were regular Corning® SMF-28® type op­
instrumented the specimens with reference strain gauges. These were tical fibers equipped with a standard polymer coating. One end of the
located in each quadrant of the circumference, on both sides of the weld fiber was connected to the interrogating unit using an LC/APC connector
at a distance d = 210 mm from said weld in Section A and Section B (see also Fig. 1 (d)), whilst the other end was spliced to a short (~15 cm)
(Fig. 1 (d)). This distance d was calculated as [11]: section of a coreless termination fiber to reduce the Fresnel reflection
√̅̅̅̅̅ from the fiber end facet. Prior to the tests, we carried out the baseline
tD measurement of the surface-mounted sensors by recording their refer­
d=4 (1)
2 ence fingerprints in a static situation.

where t is the wall thickness and D is the pipe outer diameter. The dis­ 2.3. OFDR sensor data processing
tance d corresponds to the so-called attenuation length of a bending
boundary layer in shell theory [16]. The bending boundary layer is a Since the data acquisition during the test was performed at 250 Hz
zone of bending stress around the weld induced by an abrupt change in with ~ 950 sensing points, measurement data were generated at ~ 6
pipe stiffness caused by weld cracks. By placing the strain gauges at the GB/h, or ~ 140 GB/day. To enable on-the-fly processing, we have
distance d from the weld, we avoided the influence of the propagating therefore implemented OFDR-data reduction and visualization using a
cracks on the gauge readings. The calculated axial strain values at the MATLAB® script.
distance d from the weld also differed by less than 1% from the calcu­ This processing of raw OFDR measurement data (see Fig. 2 for an
lated strain at the location of the weld. Therefore, the axial strains illustration) was done as follows.
measured by the strain gauges were used as reference values to control
the test. We give a detailed overview of the loading conditions in Section 1. The streamed strain data were decoded and stored in groups of 250
2.4. measurements buffers (equivalent to 1-second time intervals).
2. Maximal strain εmax and minimal strain εmin over this 1-second in­
2.2. Distributed strain sensing using OFDR terval were extracted for each sensing point.
3. Then the dynamic strain range εAC and the average strain εDC were
Our crack detection method relies on optical frequency-domain derived as:
reflectometry (OFDR) [17,18]. OFDR is a distributed fiber optic εAC = εmax − εmin ,
sensing method based on Rayleigh backscattering. The backscattered (2)
ε + εmin
signal along the fiber forms an actual fingerprint of the fiber that is εDC = max .
2
sensitive to environmental parameters, such as temperature or strain.

3
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

Fig. 3. Example of time-distance strain maps of (a) the dynamic strain range εAC and (b) the average strain εDC.

During the fatigue tests, large strain gradients appear along the
unbonded portion of the lead-in fiber. These gradients stem from the
movement of the unbonded fiber portion and weakened the cross-
correlation between the reference and measurement fingerprints [26].
This resulted, in its turn, in outlying data points that appeared in a
random manner in the measured strain profile. To remove these outliers
and to reduce the measurement noise, we applied median filtering to
each raw strain measurement.
In a cyclic load test, εAC represents the strain range induced by the
dynamic load, whilst εDC is indicative of the permanent strain changes in
the sensor or ambient temperature variations.
The strain range and the average strain were plotted as time-distance
maps, or so-called ‘waterfall’ maps. In these maps, each row represents a
position along the length of the sensing fiber (i.e., along the circum­
ference of the specimen), whilst one column corresponds to a certain
time. The cells contain the measured strain. These maps therefore allow
monitoring of the evolution of the strain at a certain position along the
Fig. 4. DNV-GL RP-C203 Category C average fatigue life curve.
circumference of the specimen. Fig. 3 shows an example of such
‘waterfall’ diagrams, i.e., the distribution of εAC and εDC along the fiber
to remove defects and weld overfill can be categorized as a Category C1
(y-axis) as a function of time (x-axis), with the strain components rep­
fatigue class according to the recommended practice of fatigue design of
resented on a color scale from blue (minimum) to yellow (maximum).
offshore steel structures DNV-GL RP-C203 [27]. The Category C1
The εAC and εDC values along the fiber were used as target values to
average life curve (with a 50% chance of survival) is shown in Fig. 4.
locate strain concentration regions, or ‘hotspots’, along the circumfer­
This curve is bilinear (when plotted in a log–log scale) and features a
ence to reveal the location of the fatigue cracks.
knee at 10 million cycles. We conducted the fatigue tests with two stress
ranges, covering a wide range of said curve.
2.4. Load scenario The low axial stress range of 97 MPa corresponds to a low internal
pressure of 221 bar and an expected fatigue life of 10.7⋅106 cycles,
As mentioned in section 2.1, the samples were prepared such that the which is just behind the knee in the C-curve. The high axial stress range
fatigue cracks were expected to initiate from the inner surface of the pipe of 250 MPa corresponds to an internal pressure of 570 bar and an ex­
at the root of the weld and to propagate through the wall thickness of the pected life of 607⋅103 cycles.
pipe towards the outer surface. The static axial stress σa and tangential stress σt at the outer surface
The tests were carried out as follows. of the specimens were calculated using eq. (3) below, which are valid for
closed thick-walled cylinders [28]:
1. Application of static internal pressure up to the desired pressure
pi ri2
level. σa = ,
2. Dynamic loading (bending) of the pipe in resonance with constant ro2− ri2
(3)
amplitude. 2pi ri2
3. The test was halted when a breakthrough crack occurred. σt = ,
ro2 − ri2

The static internal pressure acted on the capped ends and caused an where pi is the internal pressure, ro is the outer radius of the specimen,
axial preload. The internal pressure value was chosen such that the load and ri is the inner radius of the specimen. Using the values of σa and σt,
ratio (minimum load/maximum load) R of the cyclic axial stress during we calculated the static axial strain εsa as follows:
the fatigue test was R = 0.1. This means that the pipe was always under σa − νσ t
tension. Without internal pressure, the load ratio would be R = − 1, εsa = (4)
E
which means that the axial stress would cycle from tension to
compression. The load ratio R = 0.1 is representative for steel catenary where ν = 0.3 is the Poisson’s coefficient and E = 210 GPa is the Young’s
risers where the static weight of the riser causes axial tensile stress in the modulus of steel. We calculated the axial strain amplitude εaa using Eq.
pipe string. (4) based on minimal and maximal axial stress values.
A circumferential butt weld made from one side that is ground flush

4
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

Table 1
Overview of the loading conditions of the test program.
Sample Axial stress range Minimal axial stress Maximal axial stress εsa εaa Estimated fatigue life (C1 average Measured fatigue life
(MPa) (MPa) (MPa) (με) (με) curve) (Cycles) (Cycles)

Sample 250 28 278 291 595 607,636 918,858

1
Sample 97 11 108 113 231 10,680,063 12,314,598
2

Fig. 5. Strain distribution along the specimen circumference detected by the fiber sensors during static loading (a) in Sample 1 and (b) in Sample 2. The scale along
the radius on the radar plot indicates the detected strain values in microstrain.

the two specimens. Sample 1 was subjected to a high stress level (short
Table 2
fatigue life), whilst Sample 2 experienced a low stress level (long fatigue
Strain measured by the strain gauges during static loading.
life), as summarized in Table 1. In both cases, we continuously moni­
Sample 1 Sample 2 tored the distributions of εAC and εDC with our fiber sensors throughout
Section 0◦
90◦
180◦
270 ◦
0◦ 90◦ 180◦ 270◦ the whole duration of the tests.
A1 1140 1114 1078 – 433 485 475 –
B1 1033 1082 1105 1135 463 468 494 458
3.1. Static loading

An overview of the fatigue lives of the test specimens is provided in Prior to carrying out the actual resonant load tests, we first assessed
Table 1. Note that our experimental data points agree well with the whether the attachment of the fibers using the UV-curable glue had any
predictions of the DNV-GL RP C203 “C1” mean fatigue curve, as indi­ influence on the fiber sensor readings. To do so, we increased the static
cated in Fig. 4. internal pressure inside the test specimens up to that used in dynamic
tests (490 bar in Sample 1 and 190 bar in Sample 2) and we compared
3. Results and discussion the hoop strain detected by the strain gauges with that measured by the
fiber sensors. Fig. 5 shows radar plots that depict the hoop strain dis­
In this section, we detail the results obtained from the fatigue tests on tributions in polar coordinates detected by the fiber sensors. The strain

Fig. 6. (a) Waterfall map of the strain range εAC in Sample 1, (b) εAC evolution at the center of the detected hotspot.

5
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

Fig. 7. (a) Fracture surface in Sample 1, (b) crack location and geometry.

values recorded by the strain gauges are tabulated in Table 2. Note that
due to the failure of the strain gauges installed in Section A1 (270◦ ), we
did not include the strains measured by those gauges. The average
strains detected by the strain gauges were 1098 με and 468 με in Sample
1 and Sample 2, respectively, whilst the average strains detected by the
optical fiber were 1145 με and 445 με. These correspond to deviations of
4% and 5%, respectively. From these low values, we can safely conclude
that the attachment has little impact on the strain values detected by the
fiber sensors. The standard deviation of the strains detected by the fiber
sensors was 55.6 με in Sample 1 and 25.3 με in Sample 2. The standard
deviation of the strains detected by the strain gauges was 37.2 με in
Sample 1 and 19.8 με in Sample 2. These relatively large strain varia­
tions can be attributed to the load caused by the weight of the water
inside the pipes, the ovality of the pipes and the nonuniform thickness of
the pipe walls. Note that the standard deviations of the measurements
with the fiber sensors and with the strain gauges are of the same order of
magnitude.

3.2. Sample 1

3.2.1. Fatigue test

The test on Sample 1 was carried out with a high internal pressure of
490 bar, and the excitation force was applied at a 26.6 Hz frequency. The
corresponding applied stress range was around 250 MPa.
Fig. 6 (a) shows the waterfall map of the strain range εAC in Sample 1.
After approximately 870 × 103 cycles, we observed formation of a strain
concentration region with a length of 7 cm and center located at 41 cm Fig. 8. Image of the breakthrough crack in Sample 1.
on the waterfall map. The strain increase can be explained by a local loss
of stiffness due to the formation and propagation of a fatigue crack 3.2.2. Post-mortem analysis
through the pipe wall. We defined that a hotspot is detected when the After completion of the test, the girth weld was cut out of the pipe,
dynamic strain range at the location of the strain concentration exceeds and a DPI was carried out to inspect for the presence of cracks. We
the baseline value εACB = 360 με (i.e., the εAC at the beginning of the test) sectioned and opened the cracked area by forcing a brittle fracture after
by 5%. This detection threshold was selected based on the strain vari­ cooling. The resulting fracture surface is shown in Fig. 7 (a). We found
ations recorded by the fiber sensor before the hotspot initiation. The test two adjacent cracks: a breakthrough crack (indicated as Crack 1 in Fig. 7
was stopped following the breakthrough of the crack, more specifically (b)) and a shallow non-penetrating crack (indicated as Crack 2 in Fig. 7
after ~ 918 × 103 cycles after the start of the test. Hotspot initiation was (b)). The location of the breakthrough crack coincides with that of the
detected at ~ 45 × 103 cycles, which corresponds to ~ 5% of the strain concentration region detected by the fiber sensor. This is indica­
specimen lifetime. The axial strain variations recorded by the electric tive of the applicability of our crack detection method. The second crack
strain gauges throughout the whole test duration were less than ± 2% of did not result in a relevant strain increase at that location.
the average value, meaning that no significant strain changes were Cracks 1 and 2 clearly initiated from the internal wall of the pipe on
detected. Fig. 6 (b) shows the evolution of the dynamic strain at the the root pass of the weld. The length of Crack 1 (breakthrough crack) on
center of the detected hotspot. As the crack grows and propagates the inside was 58.0 mm and 17.8 mm on the outside. The length of Crack
through the specimen wall, this strain range monotonically increases 2 on the inside was 8.1 mm, and the crack propagated over a depth of
from the baseline value εACB = 360 με up to a maximum value 0.8 mm. A photograph of the outer surface of the pipe wall with the
εACmax ≈ 1250 με at the end of the test. breakthrough crack is shown in Fig. 8. Note that this crack was located
only ~ 5 mm away from the sensing fiber and in spite of this the fiber
remained intact upon failure of the specimen, which allowed for

6
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

Fig. 9. (a) Waterfall map of the strain range εAC in Sample 2, (b) εAC evolution in the center of the detected hotspot.

Fig. 10. (a) Fracture surface in Sample 2, (b) crack location and geometry.

continued sensor operation. presence of cracks. Apart from the breakthrough crack, no other cracks
were found. The breakthrough crack was cut out in the same way as for
3.3. Sample 2 Sample 1. The resulting fracture surface is shown in Fig. 10 (a). This
time, the crack was located in the first quadrant of the circumference
3.3.1. Fatigue test (Fig. 10 (b)), coinciding with the location detected by the fiber sensor.
The test on Sample 2 was carried out with a low internal pressure of The crack initiated from the inside of the specimen on the root pass of
185 bar and the excitation force applied at a 25.8 Hz frequency. The the weld and penetrated the wall. We measured a crack length of 36.1
corresponding stress range was around 97 MPa. mm on the inside and 6.8 mm on the outside of the specimen. Compared
A waterfall map of the strain range in Sample 2 is shown in Fig. 9 (a). to Sample 1, the stress range in Sample 2 was smaller, which resulted in
The length of the detected strain concentration region was ~ 5.2 cm, and a smaller size of the breakthrough crack. Fig. 11 shows the outer surface
its initiation was detected approximately 9 × 106 cycles after the start of of the specimen’s wall with the breakthrough crack. It was located ~ 2
the test. The hotspot formed at a position of 12 cm on the waterfall map. mm away from the fiber sensor, and as in the first test, the fiber
A breakthrough crack in this sample appeared after 12.3 × 106 cycles. remained intact, and the fiber sensor remained fully operational.
The fiber sensor was able to locate the hotspot at 3.3 × 106 cycles, i.e.,
~27% of the specimen’s lifetime. As in the test on Sample 1, the 3.4. Relation between detected strains and crack size
threshold for the hotspot detection in Sample 2 was also set at a 5%
strain range increase above its εACB. The axial strain variations recorded Table 3 provides an overview of the size of the breakthrough cracks
by the classical strain gauges during the test were less than ± 2% of the at the time of penetration. The experimental data suggests that both
average value. At the center of the ‘hotspot’ (Fig. 9 (b)), the strain range crack initiation and propagation are highly dependent on the applied
increased from the baseline value εACB ≈ 130 με up to a maximal value stress level. The crack length at breakthrough increases with increasing
εACmax ≈ 280 με. stress range. Hence for a low stress range, the crack will grow relatively
slower in the circumferential direction resulting in a crack shape with a
3.3.2. Post-mortem analysis higher aspect ratio LOD/LID, where LOD is the crack length on the outer
Similarly to Sample 1 and upon completion of the test, the girth weld surface of the wall and LID is the crack length on the inner surface of the
was also cut out of the pipe, and a DPI was carried out to inspect for the wall. If the stress level is known for a certain application, the detection of

7
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

welded area can be inspected by looking for hotspots in the strain profile
or significant deviations from the expected distributions.
Note that we carried out the fatigue experiments at a constant load
amplitude, and every heading of the weld experienced the same load
over the duration of the experiment. This simplified the crack detection
to some extent, since the strain hotspots are only measurable when the
weld region with the crack is loaded. This will typically not be the case in
field applications, and therefore supplemental knowledge about the load
amplitude and load direction over the lifetime of the component could
be required. For wind turbine structures, for example, this could include
data about wind speed and wind direction, which is sometimes avail­
able. In that case, assessments of the remaining lifetime in the context of
predictive maintenance strategies could become possible.

4. Summary

In this paper, we have studied the detection and location of fatigue

cracks in welds in steel tubular specimens using distributed strain
sensing based on optical frequency-domain reflectometry (OFDR).
OFDR technology offers greatly enhanced capabilities in terms of spatial
resolution, length of the measured region and the number of sensing
points compared to what can be achieved with traditional electrical
strain gauges.
We subjected two steel tubular test specimens instrumented with
surface-mounted optical fiber sensors to cyclic loads with applied stress
ranges of 250 MPa and 97 MPa, respectively. During the tests, we
Fig. 11. Image of the breakthrough crack in Sample 2. continuously measured the strain distributions in the weld along the
circumference of the specimens with a 2.6 mm spatial resolution and a 1
με strain resolution. For both specimens, we were able to detect the
Table 3 crack initiation and to locate the crack when its length was propagating
Overview of the features of the breakthrough cracks. parallel to the sensing fiber and was located at a distance of about 5 mm
Sample Axial stress LID LOD Detection (% of εACmax/ from the fiber. Furthermore, our surface-mounted distributed fiber
range (MPa) (mm) (mm) the lifetime) εACB sensors survived the loads applied to the specimens and continued
Sample 250 58.0 17.8 ~5 3.47
operating after the failure of the specimens.
1 The above shows that an OFDR-based sensing approach is applicable
Sample 97 36.1 6.8 ~27 2.15 to real-time weld crack detection during high-load large-scale fatigue
2 tests without prior knowledge about the location of the future crack, and
allows for:
crack initiation may thus provide relevant information that can serve the
estimation of the remaining lifetime of the component. 1. dynamic strain monitoring with a high spatial and strain resolution;
In its current state, the proposed technique allows for crack detection 2. detection of crack initiation with potential for continuous crack
and additional work is required to monitor the crack dimensions in the growth monitoring.
course of the test. The information on the crack features at certain
moments during the test (for example, at the initiation of the strain CRediT authorship contribution statement
hotspot) can be obtained from finite element method (FEM) simulations
or with a beach marking method, which can be used to visualize the Sergei Mikhailov: Methodology, Investigation, Visualization,
crack front propagation. The crack size can then be related with the Writing – original draft. Jeroen van Wittenberghe: Conceptualization,
measured strain hotspot. Note that we attempted to visualize the crack Methodology, Resources, Writing – review & editing. Geert Luyckx:
front propagation using the beach marking method. Unfortunately, no Investigation, Resources, Writing – review & editing. Philippe Thibaux:
macroscopic beach marks could be detected on the fracture surfaces, Methodology, Resources, Writing – review & editing. Thomas Geer­
possibly due to oxidation of the fatigue crack surfaces during contact naert: Investigation, Resources, Writing – review & editing. Francis
with water inside the specimens. We plan on carrying out FEM analyses Berghmans: Supervision, Writing – review & editing.
of the crack propagation in future work.
Declaration of Competing Interest

3.5. Interpretation of the results for the potential of crack weld monitoring The authors declare that they have no known competing financial
in the field interests or personal relationships that could have appeared to influence
the work reported in this paper.
Since our sensing approach allows for monitoring both the dynamic
strain range (εAC) and the static strain (εDC), information on the loading Data availability
conditions could also allow for comparing loading against non-loading
scenarios, by evaluating permanent deformations in the test compo­ Data will be made available on request.
nents via the εDC values. Furthermore, remaining lifetime estimations
could be refined by comparing certain conditions with computer simu­ Acknowledgements
lations (in view of creating a so-called ‘digital twin’). When the
component has a main loading direction, the strain distribution over the This work was performed in the framework of the BOPTIC project

8
S. Mikhailov et al. International Journal of Fatigue 176 (2023) 107851

funded by VLAIO and De Blauwe Cluster. Vrije Universiteit Brussel also [13] Otegui JL, Mohaupt UH, Burns DJ. A strain gauge technique for monitoring small
fatigue cracks in welds. Eng Fract Mech 1991;40:549–69.
acknowledges support Interreg (NWE758, Fotonica pilootlijnen); In­
[14] Shipway NJ, Barden TJ, Huthwaite P, Lowe MJS. Automated defect detection for
dustrial Research Fund (IOF); Methusalem; OZR of Vrije Universiteit Fluorescent Penetrant Inspection using Random Forest. NDT and E Int 2019;101:
Brussel. 113–23.
[15] Zhou ZF, Cheng YH. Magneto-optic microscope for visually detecting subsurface
defects. Appl Opt 2008;47(19):3463-3466.
References [16] Vinson JR. Structural mechanics: the behavior of plates and shells. Wiley-
Interscience; 1974.
[1] Papadimitriou C, Fritzen CP, Kraemer P, Ntotsios E. Fatigue predictions in entire [17] Bao X, Chen L. Recent progress in distributed fiber optic sensors. Sensors 2012;12:
body of metallic structures from a limited number of vibration sensors using 8601–39.
Kalman filtering. Struct Control Health Monit 2011;18:554–73. [18] Eickhoff W, Ulrich R. Optical frequency domain reflectometry in single-mode fiber.
[2] Weijtjens W, Iliopoulos A, Helsen J, Devriendt C. Monitoring the consumed fatigue Appl Phys Lett 1981;39:693.
life of wind turbines on monopile foundations. Proceedings of the EWEA Offshore [19] Froggatt M, Moore J. High-spatial-resolution distributed strain measurement in
Conference. 2015. optical fiber with Rayleigh scatter. Appl Opt 1998;37:1735–40.
[3] Hejazi R, Grime A, Milne I, Watson P, White E, Mariani A, et al. A Data-Driven [20] Design L, Berrocal CG, Fernandez I, Rempling R. Crack monitoring in reinforced
Framework for Fatigue Monitoring of Operating Flexible Risers. Proceedings of the concrete beams by distributed optical fiber sensors. Struct Infrastruct Eng 2021;17:
International Conference on Offshore Mechanics and Arctic Engineering -. 2022. 124–39.
[4] Pezeshki H, Adeli H, Pavlou D, Siriwardane SC. State of the art in structural health [21] Wu J, Liu H, Yang P, Tang B, Wei G. Quantitative strain measurement and crack
monitoring of offshore and marine structures. Proceedings of the Institution of Civil opening estimate in concrete structures based on OFDR technology. Opt Fiber
Engineers: Maritime Engineering. 2023. Technol 2020;60:102354.
[5] Li D, Ho S-C-M, Song G, Ren L, Li H. A review of damage detection methods for [22] Bernasconi A, Martulli LM, Carboni M. Fatigue crack growth analysis in composite
wind turbine blades. Smart Mater Struct 2015;24. bonded joints by back face distributed strain sensing and comparison with X-ray
[6] Lee YJ, Cho S. SHM-based probabilistic fatigue life prediction for bridges based on microtomography. Int J Fatigue 2022;154.
FE model updating. Sensors 2016;16:317. [23] Wong L, Rathnayaka S, Chiu WK, Kodikara J. Fatigue damage monitoring of a cast
[7] Cui W. A state-of-the-art review on fatigue life prediction methods for metal iron pipeline using distributed optical fibre sensors. Procedia Eng 2017;188:
structures. J Mar Sci Technol 2002;7:43–56. 293–300.
[8] Schijve J. Fatigue of structures and materials in the 20th century and the state of [24] De Pauw B, Hinderdael M, Moonens M, De Baere D, Geernaert T, Berghmans F,
the art. Int J Fatigue 2003;25:679–702. et al. Fatigue failure monitoring of 316L stainless steel coupons using optical fibre
[9] Hectors K, De Waele W. Cumulative damage and life prediction models for high- based distributed strain sensing. Smart Mater Struct 2019;28.
cycle fatigue of metals: a review. Metals 2021;11 [Page 204 11, 204 (2021)]. [25] ODiSI 6000 Series Optical Distributed Sensor Interrogators. https://lunainc.com/
[10] Wang P, Takagi T, Takeno T, Miki H. Early fatigue damage detecting sensors—a sites/default/files/assets/files/data-sheet/Luna ODiSI 6000 Data Sheet.pdf.
review and prospects. Sens Actuat A Phys 2013;198:46–60. [26] User’ s Guide, ODiSI 6100. https://lunainc.com/sites/default/files/assets/files/
[11] Van Wittenberghe J. Experimental Analysis and Modelling of the Fatigue resource-library/ODiSI 6100 User Guide.pdf.
Behaviour of Threaded Pipe Connections. Ghent University; 2011. [27] DNV-RP-C203 Fatigue design of offshore steel structures – DNV. https://www.dnv.
[12] Tikka J, Hedman R, Siljander A. Strain gauge capabilities in crack detection. 4th com/oilgas/download/dnv-rp-c203-fatigue-design-of-offshore-steel-structures.
International Workshop on Structural Health Monitoring (DEStech Publications html.
Inc. 2003. [28] Timoshenko SP. Strength of Materials: Part II Advanced Theory and Problems;
1956.