Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 188 (2017) 463 – 470

6th Asia Pacific Workshop on Structural Health Monitoring, 6th APWSHM

Automated Crack Detection and Crack Growth Rate Measurement

Using Thermoelasticity
Nik Rajica* and Chris Brooksa
a
Defence Science and Technology Group, 506 Lorimer Street, Fishermans Bend, Victoria 3207, Australia

Abstract

A new capability for automated crack detection and crack growth rate monitoring is described and experimentally
validated. The capability is based on a low-cost industrial grade microbolometer mounted to an x-y linear slider
assembly driven by a guidance algorithm that uses the thermoelastic quadrature signal to locate the crack tip. The
approach furnishes a high density record of the crack path, as well as thermoelastic response imagery in the vicinity
of the crack tip which can be used to determine stress intensity factors. The performance of the system is compared
to that of travelling microscopy and shown to be similar at crack growth rates above 10-7 m/cycle, but inferior at
lower rates due to increased scatter in the location estimates. This scatter is attributed in part to the limited spatial
resolution of the system in its present configuration.

Crown Copyright © 2016 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
© 2016 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review
Peer-review underunder responsibility
responsibility of the of the organizing
organizing committee
committee ofAPWSHM
of the 6th the 6th APWSHM.

Keywords: thermoelastic stress analysis; fatigue; crack detection; crack growth monitoring; crack growth rate measurement; infrared

* Corresponding author. Tel.: +61 3 9626 7193; fax: +61 3 9626 7089.
E-mail address: nik.rajic@dsto.defence.gov.au

1877-7058 Crown Copyright © 2016 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/).
Peer-review under responsibility of the organizing committee of the 6th APWSHM
doi:10.1016/j.proeng.2017.04.509
464 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470

1. Introduction

Current fatigue certification practice for safety critical load-bearing components relies on detailed knowledge of
material crack growth behaviour. Such knowledge is normally obtained empirically from exhaustive laboratory
testing carried out on standardised coupons under controlled conditions. Relatively large sample sizes are typically
required because of the variability in growth rate between samples, which can be up to a factor of 5 in certain
situations [1]. This sample-size requirement contributes to the generally high cost of material fatigue testing.
Automation of crack length measurement can reduce the cost burden of fatigue testing. Methods based on elastic
compliance [2] and electrical potential drop (EPD) [3] are the mainstays for such automation however they have
several disadvantages compared to visual inspection which is generally considered to be the benchmark method for
crack growth monitoring. Firstly, they infer rather than directly measure crack length; from a displacement in the
compliance method and a potential drop in the EPD method. For both methods, crack length estimates are obtained
from a functional relationship derived empirically or from theory. Even when such relationships are carefully
derived, errors can still arise from inconsistencies between the actual test and calibration conditions and in-lot
variation between specimens and probe/sensor placement. Another disadvantage is the inability of these methods to
identify irregular crack growth, such as angled cracking which if severe enough can invalidate a test result. The
relevant standard [4] stipulates that where irregular growth is possible a visual inspection must be used.
In the present paper an automated visual crack growth tracking capability based on thermoelasticity is described
and experimentally validated. Crack length measurements are derived from the thermoelastic response of the crack
tip stress singularity imaged using a low-cost thermal detector robotically controlled using a high precision x-y
translational stage under feedback control. The advantages of such an approach over established automation
methods derive chiefly from its direct determination of the crack tip coordinates. As is the case for visual inspection,
this enables testing of a wider range of specimen geometries and can cater for non-symmetric crack growth.
The proposed approach also offers advantages over visual inspection. Polishing of a specimen is normally
required to achieve optimal results with visual inspection, adding to specimen preparation time and overall cost.
Thermoelastic inspections observe emissions of infrared radiation from the surface rather than reflected light so
polishing is unnecessary and indeed counterproductive. Instead of high reflectivity the surface needs high infrared
emissivity which can be easily and quickly achieved with a thin coating of an appropriate matt paint. This difference
in surface preparation could confer advantages beyond just speed and convenience. For instance, where surface
modification has been applied to enhance fatigue resistance (e.g. bead peening), any further alteration of the surface
may influence fatigue behaviour. Crack growth monitoring of corroded samples is another example. The greatest
advantage of the proposed approach over visual inspection however is the ease with which it can be automated.

2. Crack Detection Using Thermoelasticity

The use of thermoelastic stress analysis in fatigue and fracture mechanics is well established. A relationship
between stress intensity factor (SIF) and bulk stress, the mechanical driving force for the thermoelastic effect, was
first reported in the 1980's [5]

ʹ‫ܭ‬ூ ߠ ʹ‫ܭ‬ூூ ߠ
ߪ௕ ൌ ߪଵ ൅ ߪଶ ൌ …‘• ൬ ൰ െ •‹ ൬ ൰ (1)
ξʹߨ‫ݎ‬ ʹ ξʹߨ‫ݎ‬ ʹ

Here, ‫ܭ‬ூ and ‫ܭ‬ூூ are the mode I and II stress intensity factors (SIF) respectively, and r and θ are polar coordinates
centred at the crack tip. For a solid deformed under adiabatic conditions a variation in bulk stress ߜߪ௕ leads to an
approximately linear variation in temperature given by,

ߙ
ߜܶ ൌ െ ܶߜߪ௕ (2)
ߩ‫ܥ‬௣
 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470 465

where ܶ is the absolute temperature, ߙ is the coefficient of thermal expansion, ߩ is the mass density, and ‫ܥ‬௣ is the
specific heat at constant pressure. The connection between the SIF and the thermoelastic response is obvious, and
there are numerous studies showing how SIF estimates can be obtained from this fundamental relationship [6–8]. In
all cases an important input parameter is the crack tip position. This can be deduced from the thermoelastic response
also, although not as simply as Equations 1 and 2 might suggest. Specifically, although Equation 1 implies a stress
singularity at the crack tip which in turn should produce a corresponding maximum in the thermoelastic response
according to Equation 2, this is predicated on linear elastic and adiabatic behaviour. Both conditions are violated in
the crack tip zone.
An alternative and more useful parameter is the thermoelastic phase signal. An experimental study by Diaz et al
[8] reported a strong correlation between the physical location of the crack tip as determined by visual inspection
and the location of a minimum in the phase signal. The existence of a phase variation in the thermoelastic response
is not obvious from Equation 2 as this equation ignores plasticity and heat conduction which are the primary drivers
for such variation. A non-adiabatic consideration of thermoelasticity shows that the thermoelastic response is in fact
complex valued and comprises real and imaginary components which relate respectively to fluctuations in phase and
in quadrature to the applied stress. Without heat conduction only the in-phase response exists, which is the situation
described by Equation 2. The present paper focuses on the quadrature component, which previous work has shown
yields a more definitive indication of the crack-tip position than the in-phase response. A reason for this was given
in [9].

3. Automated Crack Tip Monitoring

The notion of automating crack-growth tracking using thermoelastic response measurement is not new [7]. Its
prospects however have grown considerably in recent years as a result of a finding that high fidelity thermoelastic
stress analysis can be achieved with low-cost industrial grade thermal detectors [10]. For crack monitoring
applications the practical implications of this are significant, as even a cursory examination of the economics of
applying a cooled photon detector to such amply demonstrates. Cooled photon detectors are vastly more expensive
than thermal detectors and have a relatively short mean time between failure (MTBF) [11]. Since a single coupon
fatigue test can run for many days and given the statistical imperative for conducting multiple tests a cooled detector
would reach its MTBF rather quickly. Also, cooled devices are relatively large and heavy and therefore cumbersome
to apply particularly in situations where access is limited as can often be the case when testing smaller coupons.
A prototype autonomous crack growth tracking capability was developed by the Australian Defence Science and
Technology Group in 2006. It was based on a low-cost microbolometer-based thermoelastic stress analysis system
developed the year before. A proof of concept was successfully established but the relatively poor spatial resolution
of the detector employed, combined with the high cost of some of the other hardware components made the system
impractical. As a result further development was deferred. Efforts were resumed in 2015. The delay proved
serendipitous as advancements in microbolometer and motorised linear-slide technology during the intervening
period had ameliorated some of the original concerns. This resumed effort produced a significantly improved
capability.
466 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470

vertical
stage

microbolometer

edge-notched
g
coupon

horizontal
stage

Figure 1 Assembled crack growth tracking system and quadrature signal

corresponding to the starter notch (top right) and a 3 mm crack (bottom right).

The assembled system is shown in Figure 1 while Figure 2 provides a summary of how it works. There are four
basic steps to the process. Each test is begun by first manually positioning the linear slide assembly to centre the
stress concentration in the camera field of view. The objective of the automation algorithm is to maintain the stress
concentration at the centre of the image. Step 1 is a quasi-real time digital cross-correlation of the infrared video
stream and the reference load signal. This is implemented using a modified version of the MiTE software which
adaptively sets the cross-correlation length based on an estimate of the signal noise and the crack growth rate. In
step 2, the position of the quadrature minimum is determined using a robust image processing strategy. In step 3, the
x-y displacements required to re-centre the quadrature minimum are calculated and the motorised sliders
commanded to move accordingly. As these displacements reflect (by proxy) actual physical movements of the crack
tip, the crack trace can be recovered by integration. The final step allows for separate visual tracking of the crack. In
the present study, a USB microscope was fitted to an identical x-y motorised linear slide assembly attached to the
load frame on the other side of the specimen. This assembly is slaved to and moves in lock-step with the primary
assembly.
Camera
Load Signal Microscope

Correlate Detect Move Image

Data Crack Tip Stage Crack Tip

Figure 2 Flow diagram of automated crack tip tracking procedure.

The accuracy of the tracking system is likely to be determined by at least the following factors: (i) the spatial
resolution of the imagery which is a function of the instantaneous field of view of the camera/lens configuration and
the stand-off distance, (ii) the precision of the linear stage, (iii) the duration of the cross correlation (iv) the
relationship between the quadrature minimum and the actual location of the crack tip and (v) the frequency of
loading. In the present study, the first two parameters were fixed and the third varied adaptively in software as
described previously. The infrared imager used in the study was a Xenics Gobi 640 microbolometer equipped with a
standard 18 mm f1.0 lens. The unit has a 640 x 480 amorphous silicon detector and outputs GigE video at a 50 Hz
frame rate. A subject to camera distance of 157 mm was maintained throughout the study. At this distance the image
spatial resolution was approximately 152 μm/pixel. The motorised linear slides have a nominal step resolution of
less than 0.1 μm and an accuracy of better than 2 μm for step increments of less than 1 mm.
 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470 467

4. Experimental Evaluation

Fatigue testing was conducted on rectangular Al2024 aluminium alloy coupons (see Figure 1), 400 mm in length,
100 mm wide and 2 mm thick. Each had a single mechanically-milled edge-notch 10 mm in length and 1 mm in
width and a tip radius of 0.5 mm. One side of the coupon was sprayed with a high emissivity (~0.94) matt khaki
paint. The other side, which was inspected with the USB microscope, was left in the supplied condition. Testing was
done under fully-tensile constant-amplitude uniaxial cyclic loading in a digitally controlled servo-hydraulic
machine. The load range was 2–9 kN (r-ratio of 0.22) and the loading frequency was 5 Hz.
Six specimens were tested to failure. In each case two sets of data were generated: a crack trace distilled from the
thermoelastic response and a corresponding set of visual images from the microscope. Each visual image was taken
with the sample held at peak load, which served the purpose of eliminating motion blur in the captured image as
well as opening the crack, both of which assisted identification of the tip. This process, including control of the test
machine, was automated via algorithms implemented within the MiTE application. Locating the crack tip in the
visual imagery was done manually after completion of the test. Because the microscope moved in lock-step with the
infrared camera the crack tip was always located near the centre of the field of view, which helped speed up the
identification process. Another advantage in ceasing the dynamic load during capture of the visual image was that it
created arrest marks in the fracture surface [12]. This physical record of the crack tip position was used to check the
visual estimates of crack tip position, ensuring a reliable baseline against which the quadrature minimum could be
compared.

5. Results and Discussion

Figure 3 shows two representative results in which crack tip trace estimates are overlaid on photographs of the
fractured portion of the sample. Viewed at the larger scale (top two images) the visual and thermoelastic traces
appear in close agreement, however the close-ups reveal important differences. Before canvassing these, one should
note that the visual estimates coincide with the location of the arrest marks, which confirms the accuracy of those
measurements.
In Figure 3 red lines are used to connect corresponding visual and quadrature estimates of the crack tip position.
These lines reveal that the quadrature estimates are consistently ahead of and vertically displaced from the true tip
position. Some vertical discrepancy is expected given the obvious slant of the fracture surface which creates about 1
mm of difference in vertical position between the front and rear faces of the sample. That the quadrature estimates
correspond to positions inside the sample is interesting. There are two factors that might contribute to this; (i)
vertical motion of the crack due to the compliance of the specimen and (ii) through-thickness heat flow established
by the slant contributing to the quadrature signal. The first effect can be readily recovered by incorporating motion
compensation which was not done in the present study but will be implemented in due course. The second effect is
intrinsic to the crack geometry and likely to be frequency dependent. It is surmised that crack tip tunnelling should
produce similar behaviour except the offset would obviously be horizontal. This may present an opportunity to
retrieve the through-thickness profile of a tunnelling crack from quadrature measurements taken across a set of
frequencies or under a suitably frequency-rich variable load-amplitude spectrum. The enlarged view at the bottom of
Figure 3 confirms that when there is no slant in the crack face the visual and quadrature markers are at essentially
the same vertical position.
Heat conduction is also likely to play a role in the horizontal discrepancy between the quadrature and visual
markers, however there are additional factors to consider in this case. The first is crack tip plasticity. The presence
of a cyclic plastic zone violates the linear-elastic assumption underpinning the association between the quadrature
minimum and the crack-tip stress singularity. Also, cyclic plastic deformation is an irreversible process that leads to
the production of heat which itself will lead to temperature gradients and thus contribute to the quadrature signal.
For the fully tensile loading applied in the present study, such contributions should be at the frequency of loading.
Understanding the role of cyclic plasticity in the formation of the quadrature signal is an important topic for further
research. Another factor is the motion of the crack tip the effect of which is to average the crack tip position over the
cross-correlation period. One would expect this to produce a conservative estimate, however the crack growth data
468 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470

is seen to be biased in the opposite direction, suggesting that one or a combination of the aforementioned factors are
more significant in this particular case. Zanganeh et al [7] also reported y-component and phase estimates that were
consistently ahead of the actual crack tip position.

10m
10mm

5mm 5mm

10mm

Figure 3 Fractured sections from two different coupons with enlarged view revealing visible arrest marks. Red
and blue markers correspond respectively to visual and quadrature estimates of crack tip position.

5.1. Crack Growth Rate

The main purpose of fatigue crack growth testing is to produce crack growth rate curves that describe the rate of
crack growth (da/dN) as a function of the stress intensity factor range, ΔK. In the present study, crack growth rates
were calculated from the acquired crack progression data using the incremental polynomial method with a half
window length of 4 points. Representative crack growth rate curves are shown in Figure 4. Overall, the agreement
between the visual and thermoelastic estimates is excellent and the traces are virtually coincident at crack growth
rates above 10-6 m/cycle. Scatter is clearly an issue for the quadrature estimates and is the main limiting factor on
the crack growth rate resolution achieved in these results. However, the scatter is partly determined by the image
spatial resolution which is a function of the optical system and easily changed. In this preliminary study the spatial
resolution of the microbolometer was fixed at 152 μm, a relatively poor value and vastly inferior to the 3.6 μm
resolution of the microscope. The former value can be substantially improved (by an order of magnitude) with a
simple modification to the existing lens assembly, and specialised lenses are available that can do even better. An
investigation of the effect of spatial resolution, as well as correlation length and other factors on the threshold crack-
length sensitivity is currently underway.
At crack growth rates of 10-7 m/cycle and above the two methods yield similar results, which is encouraging
given the observed bias discussed earlier. This suggests that the bias is relatively stable. Of course, if as postulated
previously the bias is a function of the cyclic plastic zone size and heat conduction some variation in crack growth
rate is to be expected. In fact, a closer examination of the data reveals some deviation, albeit very slight and
probably insignificant for most practical purposes.
 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470 469

-3
10
-4

da/dN (m/cycle)
10
-5
10
-6
10
-7
10
Visual TSA
-8
10
1 2
10 10
ΔK (MPa m)
-3
10
-4
10
da/dN (m/cycle)

-5
10
-6
10
-7
10
-8
10
Visual TSA
-9
10
1 2
10 10
ΔK (MPa m)
-3
10
-4
10
da/dN (m/cycle)

-5
10
-6
10
-7
10
-8
10
Visual TSA
-9
10
1 2
10 10
ΔK (MPa m)
Figure 4 Crack growth rate curves as a function of stress intensity factor range for three specimens using visual
and thermoelastic crack tip location measurements.

6. Conclusion

A new capability for automated crack detection and crack growth rate monitoring has been described and
experimentally validated on single-edge notched coupons under constant amplitude cyclic loading. The capability is
based on a low-cost industrial grade microbolometer mounted to an x-y linear slider assembly driven by a guidance
algorithm that exploits the thermoelastic quadrature signal in the vicinity of a crack tip. The performance of the
system was shown to be comparable to that of visual tracking at crack growth rates above 10-7 m/cycle, but inferior
at lower rates due to scatter. This scatter was partly attributed to the relatively poor spatial resolution of the
470 Nik Rajic and Chris Brooks / Procedia Engineering 188 (2017) 463 – 470

microbolometer/lens assembly. Improved resolution can be achieved by decreasing the stand-off distance, or by
changing lens, strategies that are currently under investigation as part of a more comprehensive and systematic
follow-on study.

References

[1] W. G. Clark Jr., S. J. Hudak Jr., Variability in Fatigue Crack Growth Rate Testing, Journal of Testing and Evaluation, Vol 3, No. 6, pp. 454–
476, 1975.
[2] Marsh, Kenneth James, Roderick Arthur Smith, and Robert O. Ritchie. Fatigue crack measurement: techniques and applications. Engineering
Materials Advisory Services Ltd., 1991.
[3] M. A. Hicks and A. C. Pickard, A Comparison of Theoretical and Experimental Methods of Calibrating the Electrical Potential Drop
Technique for Crack Length Determination, Int. Journal of Fracture, 20, 1982, pp 91–101.
[4] ASTM International. Standard test method for measurement of fatigue crack growth rates. ASTM International, 2011.
[5] Stanley P. and Chan W. K., The determination of stress intensity factors and crack tip velocities from thermoelastic infrared emissions. In:
Proceedings of international conference on fatigue of engineering materials and structures, 1986, pp. 105–114. London: Institution of
Mechanical Engineers by Mechanical Engineering Publications.
[6] Tomlinson R. A., Nurse A. D. and Patterson E. A., On determining stress intensity factors for mixed mode cracks from thermoelastic data.
Fatigue & Fracture of Engineering Materials & Structures. 1997; 20(2): 217–226.
[7] Zanganeh M., Tomlinson R. A. and Yates J. R., T-stress determination using thermoelastic stress analysis. Journal of Strain Analysis for
Engineering Design. 2008; 43: 529–537.
[8] Diaz F. A., Patterson E. A. and Yates J. R., Assessment of effective stress intensity factors using thermoelastic stress analysis. Journal of
Strain Analysis for Engineering Design. 2009; 44: 621–631.
[9] Rajic N., and Galea S, Thermoelastic stress analysis and structural health monitoring: An emerging nexus, Structural Health Monitoring
(2014). DOI: 1475921714548936.
[10] Rajic N. and Rowlands D., Thermoelastic stress analysis with a compact low-cost microbolometer system. Quant. InfraRed Thermography
Journal. 2013; 10: 135–158.
[11] Rogalski, A., Progress in Focal Plane Array Technologies, Progress in Quantum Electronics 36 (2012), pp. 342–473.
[12] Ewalds, H. L., and Wanhill, R. J. H., Fracture Mechanics, Edward Arnold (Publishers) Ltd., 1984.