Journal Pre-proofs

Damage detection in rotating objects using position-triggered thermography

Pedro J. Sousa, Francisco Barros, Nuno V. Ramos, Paulo J. Tavares, Pedro

M. G. P. Moreira

PII: S1350-6307(20)30467-2
DOI: https://doi.org/10.1016/j.engfailanal.2020.104642
Reference: EFA 104642

To appear in: Engineering Failure Analysis

Received Date: 25 March 2020

Revised Date: 25 May 2020
Accepted Date: 27 May 2020

Please cite this article as: Sousa, P.J., Barros, F., Ramos, N.V., Tavares, P.J., M. G. P. Moreira, P., Damage
detection in rotating objects using position-triggered thermography, Engineering Failure Analysis (2020), doi:
https://doi.org/10.1016/j.engfailanal.2020.104642

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier Ltd.

 Damage detection in rotating objects using position-triggered
thermography

Pedro J. Sousaa,b, Francisco Barrosa, Nuno V. Ramosa, Paulo J. Tavaresa, Pedro

M. G. P. Moreiraa
aINEGI, Universidade do Porto, Rua Dr. Roberto Frias, 400, Porto 4200-465, Portugal
bFaculdade de Engenharia da Universidade do Porto, Rua Dr. Roberto Frias, s/n, Porto 4200-465, Portugal

Corresponding author
Pedro J. Sousa
INEGI
Rua Dr. Roberto Frias, 400
4200-465 Porto
Portugal
Tel: +351 225082151
E-mail address: [email protected], [email protected]

Declarations of interest: none

Abstract

Damage detection using non-destructive testing techniques is important to prevent damage in structures
with long operating times. In the case of rotating structures with minimal downtime, such as wind turbines,
it is advantageous to be able to detect these damages during operation. Thermography is a method that is
capable of performing these measurements, but can still have some difficulties in field application. One
such difficulty resides in the necessity of properly aligning each image with the previous ones to obtain
difference thermograms with good quality. Thus, the present work proposes a method to properly align
images by synchronizing the termographic camera with the rotation of the object, using only external non-
contact devices, all positioned on the same side of the specimen. In this case, a photodetector is used in a
high-amplification mode, working like a single-pixel camera. It outputs a grey level signal that will vary
when a blade passes in front of it. By detecting this change, it is possible to trigger the camera and/or a heat
source in the correct angular position, and use the acquired images to obtain good quality thermograms.
This approach was applied to defect detection in rotating carbon fibre blades from a 450-size RC helicopter.
It was possible to obtain results in these dynamic experiments that were in agreement with the ones obtained
in a static approach.

Keywords: Structural Integrity; Structural Monitoring; Damage detection; Non-destructive testing;

Infrared thermography;

1. Introduction

Detecting damage in its early stages is important to prevent failure of structures with long operating
times. This should be done using non-destructive testing techniques, as the intention is for them to keep
working as long as possible. There are several methods that can be successfully used for these procedures
[1,2]. Among them, Thermography is a widely used non-contact technique with usage in multiple fields,
from aeronautics and energy generation to electronics [3–6], and it is well suited for non-destructive
qualitative detection with few, although costly [2], requirements. The basic approach to damage detection
using thermography involves obtaining two thermograms in differing stages of cooling/heating and
analysing their subtraction. Thus, it only requires the use of a thermographic camera and the presence of a
thermal load that can be either artificial or natural.
This method is commonly used in composites to detect the presence of voids and delamination [7]. These
are characterized by an enclosed air pocket, which has a thermal conductivity that is significantly different
from the surrounding regions. This will be noticeable in the thermogram as a region that stays hot for longer,
while the surrounding region cools down [4]. It should be noted, however, that similar behaviour could also
be caused by incoming heat being concentrated in a region. Thus, multiple evaluations should be made
before concluding about the presence of a defect, if possible in different heating situations.
In the case of structures with minimal downtime, such as wind turbines, it is advantageous to be able to
detect these damages during operation. However, if the inspected components are moving, this will cause
difficulties in obtaining the difference thermograms, as the two thermal images that generate them should
be well aligned.
This problem can be approached using digital image processing, by registering two images in different
points in time and different angular positions and afterwards subtracting them. For example, M.
Doroshtnasir et al. [8] use this approach successfully to analyse wind turbine blades on-site.
Another possible approach is to synchronize the image acquisition with the rotation of the blades. This
way, it is possible to obtain sequential images that are already aligned and, as such, do not require any
image processing except subtraction. This is a natural extension towards damage detection of the approach
previously undertaken for digital image correlation in [9–13]. It uses similar triggering procedures,
including an upgraded version of the trigger controller and a custom photodetector, focused on improving
the versatility of the system by sparing the need for a laser beam. By not requiring this laser, it is possible
to place the whole acquisition system on the same side of the target object and thus improve its usability
for large-scale operations, as everything can be placed on the ground and close to each other.
For the present work, this method was implemented in a laboratorial test, using only external non-contact
devices to trigger the thermographic camera, and with the complete setup positioned on the same side of
the specimen.
2. Methodology

The main goal of this work is to develop a method for defect detection in rotating parts using position-
triggered thermography. In order to assess its feasibility, a blade was analysed in both static and dynamic
situations, with the aim of finding the same potential defects in both. The same setup was used in both
situations, using position-triggering in the dynamic one and time-based triggering in the static.
Both analyses were performed using simple thermogram subtraction, where a colder situation is
subtracted from a hotter one. This requires only the use of two images for each analysis. Nonetheless, a
larger amount was acquired in order to choose the best pair to clearly show the defects.

2.1. Experimental setup

The used experimental setup can be seen in Fig. 1, and a schematic representation of it is presented in
Fig. 2. The evaluated specimen is a carbon fibre blade, typically used in 450-size RC helicopters.
The thermograms were acquired using a FLIR SC7000 series thermographic camera, which is connected
to the control computer by Gigabit Ethernet and is capable of accepting LVTTL (0 to 3.3V) trigger pulses
[14]. This is an important feature for the present work, as without triggering, it would be extremely difficult
to obtain aligned images without performing additional digital processing.
In this case, a custom photodetector is used in a high-amplification mode, working like a single-pixel
camera as it outputs a grey level signal that will vary when a blade passes in front of it. The trigger controller
will compare this level with a configured reference value, and convert these variations into input pulses,
which are then counted. When a configurable value is reached, e.g. once every rotation, the controller will
generate two independent output pulses, one of which is used to trigger the camera in the correct angular
position.
The trigger controller, responsible for triggering the camera, was previously developed for use with
digital image correlation applications and high-speed or industrial cameras [9–13], which typically accept
5V TTL logic. Therefore, it was necessary to convert these voltage levels using an external converter
module, which was included between them. This converter is a simple solution based on a voltage divider,
which has the advantages of a fast response and easy implementation with only two input and two output
conductors.
The different cold and hot situations were created using a heat source in the form of an incandescent
light source. This creates a thermal load that flows differently through air pockets and other inclusions and
thus enables the detection of detects. In this work, the heat source was manually controlled, but the trigger
controller still provides a free output channel that could be used to control it.
 Fig. 1. Photo of the experimental setup

Fig. 2. Schematic of the experimental setup

2.2. Triggering

The most important components of the triggering system are the previously mentioned photodetector
and trigger controller and they are responsible for synchronizing the camera’s image acquisition with the
rotation of the object.
Although using a laser beam and a standard photodetector would have a significantly faster response,
when evaluating large structures for damage, it is difficult to place a laser beam on the other side of the
rotating blades and it is also disadvantageous to stop the device’s operation just to place a reflector. Thus,
a photodetector that does not need such measures was developed.
2.2.1. Photodetector

In a “single pixel” approach, the photodetector receives an amount of incoming light that is several order
of magnitude smaller when compared to a laser-based one, with expected currents in the nanoampere range
which require the use of much larger amplification, becoming susceptible to picking up significant amounts
of noise that can easily become larger than the signal.
This system is built around a SFH 2201 from OSRAM Opto Semiconductors [15] in photovoltaic mode.
This mode is achieved by using a zero bias voltage on the diode and prioritizes high sensitivity [16–18],
which is what is needed in this situation.
The amplifier's requirements are also important, namely small input bias current and noise in general.
The AD8615 Precision Operational Amplifier from Analog Devices was selected, as it features a low input
bias current of just 1 pA, an input offset current of 0.5 pA and a common mode rejection ratio of 100 dB
[19].
In Fig. 3, the amplification circuit is shown, where it is possible to see that this system operates between
0 and 5 V, while standalone photodetector solutions usually work up to 10 V. Due to the small currents
involved, a Faraday cage was integrated as part of the enclosure design to protect the system from any
potential noise sources.

Fig. 3. Schematic of the employed amplification circuit

As shown in Fig. 4, this cage was implemented using aluminium foil, and a screw and two nuts were
used to connect it to ground. This allows for a sturdy connection without the need for soldering. The
aluminium foil is wrapped around an inner enclosure where the PCB is placed, avoiding any possible short
circuits between the ground-connected foil and any other trace. After assembly, the Faraday cage is
protected against tear using an external enclosure.
 Fig. 4. Exploded view of the developed photodetector, using folded aluminum foil as a Faraday cage to reduce noise

As sensitivity is of utmost importance in this system, an interchangeable lens adapter was included,
implemented using a 3D printed adapter. This allows for the use of any commercial F-mount lens to either
restrict or augment the detection area, gathering a larger amount of light from the selected area.
The photodetector module is placed in the final enclosure, Fig. 5, which holds not just the lens and the
diode, but also the supporting electronics, such as a voltage regulator and the input and output connectors.
It also features plenty mounting options, in the form of M4, M6 and ¼”–20 screw connections.

Fig. 5. The developed photodetector, using a “single pixel” approach

2.2.2. Trigger controller

 As previously mentioned, the trigger controller, Fig. 6, was developed focused on the use in high speed
digital image correlation applications [9–13], but it is versatile enough for other types of application, such
as damage detection and any other situation that requires generating precise pulses by counting input pulses.
However, an upgrade was necessary for this usage, in order to replace the previously fixed threshold levels
with customizable ones. This is due to varying lighting conditions and different contrast levels between the
target object and its background.

Fig. 6. Trigger controller

Therefore, the input logic of the controller had to be reworked, and an LT1713 high-speed comparator
from Linear Technology [20] was introduced to compare the grey level output of the photodetector with a
customizable reference value. Although this is changed manually using a trimmer, its value can be seen in
the developed graphical user interface, as well as the current state of the comparator’s output.
In order to maintain feature compatibility with the previous experiments, the previous 10V-compatible
transistor-based input processing chain was kept and it is possible to switch between them using a physical
selector. Additionally, the inputs of the comparator were protected using a Zener diode based voltage
limiting circuit, which protects the controller against possible usages of 10V photodetector systems with
the new processing circuit.

2.3. Approach

The experiment was divided in two main stages: static and dynamic. In the static phase, the target
specimen was kept immobile and was subjected to a thermal load. In the dynamic phase, the target
specimen, while rotating at 530 rpm, was subject to a similar thermal load. In both cases, thermal images
are acquired both before heating and during the cooling period.
The static phase does not pose significant challenges, as it is a standard application of thermography for
damage detection, used here mainly as a comparison point.
On the other hand, the dynamic one poses significant challenges, mainly in acquiring accurately
positioned thermograms of a rotating object with sufficient quality. Considering that the camera’s
integration time affects not only motion blur, but also the range of temperatures that can be detected, it is
then necessary to reconcile the two. The integration time cannot be so small as to be unable to detect the
blade’s temperature of approximately 22 ºC and require unrealizable temperatures nor so large that motion
blur makes any defect undetectable.
After finding a workable middle ground, it is possible to start image acquisition. The RC helicopter is
started and the blades’ rotation speed is set. While it is rotating, the heat source is turned on for a few
seconds, in order to cause an increase in temperature on the blades. After it is turned off, the image
acquisition continues until the blades are cold again.
The obtained thermal images can then be subtracted to highlight temperature differences, and thus detect
damage. In order to confirm the presence of defects, these were detected not just on a single experiment but
in multiple ones, obtaining similar results.
3. Results and Discussion

During the experiment, the first stage was the static one, as it was important to find if there were any
defects on the target blade and, if so, where. Thus, similar to the dynamic stage, the blade was heated with
incandescent lighting and both the heating and cooling periods were recorded. From this set, two images
were selected to calculate the difference thermogram, one hotter (Fig. 7a) that the other colder (Fig. 7b).
The static difference thermogram of Fig. 8a, clearly shows the presence of multiple defects, encircled in
red, the most noticeable of which is the circular one very close to the tip. Nonetheless, it should be noted
that the hotter region towards the centre of the blade, i.e. the larger black ellipse, is hotter mainly due to
concentrated heat from the loading phase and the presence of defects in that area is thus uncertain. However,
in other experiments it was possible to distinguish that there were two independent potential defects in that
region.
Afterwards, the dynamic stage was performed. The first step was to select the integration time,
considering the limitations of minimum temperature and motion blur. It was noticed that, for a value of 349
µs it was possible to obtain thermal images with temperatures around 22ºC, with some motion blur, but not
as to make it unusable. This value was not decreased further because the acquisition software recommended
minimum temperatures of 30ºC for that integration time value.
Nonetheless, it was possible to acquire over 150 thermograms in the dynamic situation, with the blades
rotating at approximately 530 rpm and acquiring one thermal image per rotation. Two of these, a few
seconds apart, were chosen to calculate the difference thermogram, with one being hotter (Fig. 7c) than the
other (Fig. 7d).
These thermal images have temperatures that are much closer to the environment temperature than the
ones in the static stage, as they are rotating and therefore will not heat up as much and will cool down faster,
which contributes to reduced contrast between the blades and the background.
The obtained dynamic difference thermogram, Fig. 8b, also shows the presence of multiple possible
defects, also encircled in red. The three defects that are noticeable in this thermogram are also clearly seen
in the static stage. Once again, the most noticeable among these defects is the circular one on the tip. Besides
these defects, there is also a region, encircled in black, where there is some accumulated heat and where
the presence or not of defects is uncertain.
Comparing the two, it is noticeable that the smaller defects around the top edge that are detected in the
static situation could not be detected here, which can possibly be both due to the very fast cooling of the
edges and to motion blur. Nonetheless, the larger potential defects are clearly detectable, and are in
agreement with the static results.

(a) (b)

(c) (d)

Fig. 7. Thermograms obtained using the mentioned methods, where (a) and (b) are the hot and cold situation in static measurement
and (c) and (d) are the hot and cold situations in dynamic measurement at 530 rpm
 (a) (b)

Fig. 8. Difference thermograms obtained in: (a) Static measurement; (b) Dynamic measurement, rotating at 530 rpm. Potential
defects are marked in red and uncertainty regions are marked in black.

4. Conclusions

The presented results demonstrate that it is possible to detect potential defects in rotating objects using
position-based triggering.
The developed methodology employs a custom photodetector, with a working principle similar to a
single pixel camera, where its high amplification provides a grey level of a particular region. This level was
processed with the developed controller to trigger the acquisition of thermal images in the same position in
every rotation. And consequently enable a direct comparison of the acquired images through subtraction
without the need for image registration. With this, the procedure also becomes fast enough that it can be
performed live, if necessary.
Despite the difficulty in reconciling motion blur and minimum temperatures when selecting the
integration time, the obtained dynamic and static results are in agreement and the largest potential defect
can be clearly seen in the same position in both. These issues could possibly be mitigated using a faster and
more sensitive thermographic camera.
The presented approach is also compatible with other image-based damage detection methods, as long
as that method can work with position-triggered instead of time-triggered images and the stability
requirements are not too tight. For example, it should be compatible with single-shot spatial phase shift
shearography methods, but it would probably struggle with phase-stepping shearography, as in the latter
there is a tight stability requirement during the phase-stepping phase.
It should be possible to apply this procedure to in-situ evaluations of full-scale rotating wind turbines.
The whole setup can be assembled in the ground in front of the wind turbine, and the trigger system would
be able to distinguishing the wind turbine’s blade from the background and trigger the thermographic
camera in the correct position, like in this work. The most significant difference between these applications
relates to the thermal load. In laboratory conditions, it is possible to use an artificial heat source. However,
due to the sheer size of the wind turbine, this would be difficult to perform in fieldwork. Nonetheless, it is
possible to obtain two different thermal situations by considering the naturally occurring thermal variations,
such as the cooling that the wind turbine undergoes around sunset, which is a similar approach to [8]. Future
work should include the test of this system with full-scale wind turbines.

Acknowledgements

Pedro J. Sousa gratefully acknowledges the FCT (Fundação para Ciência e Tecnologia) for the funding
of the PhD scholarship SFRH/BD/129398/2017.

References

[1] R. Raišutis, E. Jasiuniene, R. Šliteris, A. Vladišauskas, The review of non-destructive testing

techniques suitable for inspect manufacturing of the wind turbine blades., ULTRAGARSAS
(ULTRASOUND). 63 (2008) 63(1):26–30.
[2] W. Yang, P.J. Tavner, C.J. Crabtree, Y. Feng, Y. Qiu, 1, Wind turbine condition monitoring:
 technical and commercial challenges, Wind Energy. 17 (2014) 673–693.
https://doi.org/10.1002/we.1508.
[3] X. Li, J. Sun, N. Tao, L. Feng, J. Shen, Y. He, C. Zhang, Y. Zhao, An Effective Method to
Inspect Adhesive Quality of Wind Turbine Blades Using Transmission Thermography, J.
Nondestruct. Eval. 37 (2018) 1–11. https://doi.org/10.1007/s10921-018-0473-5.
[4] V. Dattoma, R. Marcuccio, C. Pappalettere, G.M. Smith, Thermographic investigation of
sandwich structure made of composite material, NDT E Int. 34 (2001) 515–520.
https://doi.org/10.1016/S0963-8695(00)00082-7.
[5] J. Wu, J. Zhu, H. Xia, C. Liu, X. Huang, G.Y. Tian, DC-Biased Magnetization Based Eddy
Current Thermography for Subsurface Defect Detection, IEEE Trans. Ind. Informatics. 15 (2019)
6252–6259. https://doi.org/10.1109/TII.2019.2891107.
[6] J.L. Bartlett, F.M. Heim, Y. V. Murty, X. Li, In situ defect detection in selective laser melting via
full-field infrared thermography, Addit. Manuf. 24 (2018) 595–605.
https://doi.org/10.1016/j.addma.2018.10.045.
[7] A.P. Chrysafi, N. Athanasopoulos, N.J. Siakavellas, Damage detection on composite materials
with active thermography and digital image processing, Int. J. Therm. Sci. 116 (2017) 242–253.
https://doi.org/10.1016/j.ijthermalsci.2017.02.017.
[8] M. Doroshtnasir, T. Worzewski, R. Krankenhagen, M. Röllig, On-site inspection of potential
defects in wind turbinerotor blades with thermography, Wind Energy. (2016) 1407–1422.
https://doi.org/10.1002/we.1927.
[9] P.J. Sousa, F. Barros, P.J. Tavares, P.M.G.P. Moreira, Displacement analysis of rotating RC
helicopter blade using coupled CFD-FEA simulation and digital image correlation, Procedia
Struct. Integr. 17 (2019) 812–821. https://doi.org/10.1016/J.PROSTR.2019.08.108.
[10] P.J. Sousa, F. Carneiro, N.V. Ramos, P.J. Tavares, P.M.G.P. Moreira, Development of LED-
based illumination system for high-speed digital image correlation, Procedia Struct. Integr. 17
(2019) 828–834. https://doi.org/10.1016/J.PROSTR.2019.08.110.
[11] P.J. Sousa, F. Barros, P. J. Tavares, P.M.G.P. Moreira, Displacement measurement and shape
acquisition of an RC helicopter blade using Digital Image Correlation, Procedia Struct. Integr. 5
(2017) 1253–1259. https://doi.org/10.1016/j.prostr.2017.07.097.
[12] P.J. Sousa, F. Barros, P.J. Tavares, P.M.G.P. Moreira, Digital image correlation displacement
measurement of a rotating RC helicopter blade, Eng. Fail. Anal. 90 (2018) 371–379.
https://doi.org/10.1016/j.engfailanal.2018.04.005.
[13] P.J. Sousa, J.M.R.S. Tavares, P.J.S. Tavares, P.M.G.P. Moreira, Correction of rigid body motion
in deformation measurement of rotating objects, Measurement. 129 (2018) 436–444.
https://doi.org/10.1016/J.MEASUREMENT.2018.07.049.
[14] FLIR Systems Inc., SC7000 User Manual, (n.d.).
[15] OSRAM Opto Semiconductors, SFH 2201 - TOPLED - Silicon PIN Photodiode with Enhanced
Blue Sensitivity, (2019).
[16] Thorlabs Inc., Photodiode Tutorial, (n.d.).
https://www.thorlabs.com/tutorials.cfm?tabID=787382FF-26EB-4A7E-B021-BF65C5BF164B
(accessed January 30, 2020).
[17] AP Technologies Ltd, Photodiode Theory of Operation, (n.d.).
https://www.aptechnologies.co.uk/support/SiPDs/operation (accessed January 30, 2020).
[18] OSI Optoelectronics, Photoconductive Series - Planar Diffused Silicon Photodiodes, (n.d.).
[19] Analog Devices Inc., AD8615/AD8616/AD8618 - Precision 20 MHz CMOS Rail-to-Rail Input /
Output Operational Amplifiers, (2004).
[20] Linear Technology Corporation, LT1713/LT1714 - Single/Dual, 7ns, Low Power, 3V/5V/±5V
Rail-to-Rail Comparators, (2000).