IDENTIFICATION OF THERMAL AND OPTICAL EFFECTS FOR
THE DETECTION OF OPEN-CRACK IN PHOTOTHERMAL NON
DESTRUCTIVE TESTING
S. Hermosilla-Lara1'2, P.-Y. Joubert1, J.-M. Decitre1
[
SATIE (CNRS-UMR 8029), Ecole Normale Superieure de Cachan, 94 235 Cachan
France
2
AREVA/Framatome-ANP, Technical Center, ZIP sud 71380 Saint Marcel
France
ABSTRACT. In this paper, the images provided by a flying-spot camera dedicated to open-crack
detection are considered. In this type of active thermography, thermal effect and the optical effects
contribute to the elaboration of photothermal images. In this contribution a signal processing method,
based on multiple principal component analysis is proposed in order to separately identify both
thermal and optical effects from the raw images. The detection performance enhancement is
characterized thanks to Receiver Operating Characteristic curves.
INTRODUCTION
Among the existing non-destructive testing techniques, the active thermography
appears to be a good alternative to penetrant testing for open-crack detection in metallic
structures, because it is a contactless and automatizable technique. A photothermal camera
prototype (flying-spot camera) was realized and presented in [1] and is implemented here
for the testing of an industrial mockup showing open fatigue cracks with severe surface
conditions. After a short description of the basic principle of the implemented
photothermal camera, the mockup and the obtained photothermal images are presented.
Then an image processing method, based on multiple principal component analysis is
implemented in order to improve the open crack detection.
THE PHOTOTHERMAL CAMERA
Basic Principle
The photothermal camera used in this study was presented in [1]. It consists in a
simple and portable device, composed of a laser used as a thermal excitation source and an
array of infrared detectors as thermal sensors. The images are obtained in an active
thermography scheme [1] thanks to a simultaneous scanning of the inspected structure
surface by the excitation and detection beams at a constant speed and with different offsets,
as shown in figure 1.
CP657, Review of Quantitative Nondestructive Evaluation Vol. 22, ed. by D. O. Thompson and D. E. Chimenti
© 2003 American Institute of Physics 0-7354-0117-9/03/$20.00
674
Inspection i
Detection bmm
IR Camera
Phototterraai signature of an open-crack
FIGURE 1. Basic principle of the photothermal camera for open-crack detection.
Two main physical phenomena take place in the this type of active thermography:
the thermal effect which is a thermal barrier effect produced by the presence of a crack,
and the optical effects which mainly consist of a variation of the absorptivity or diffusivity
factor of the inspected surface. The signals provided by the camera are resulting from a
combination of all these effects, the proportions of the combination depending on the
nature of the defects and of the inspected surface. The combination is assumed to be linear
and this assumption was experimentally verified. In the case of open-cracks on a surface
with a good absorptivity, the thermal barrier is the dominating effect, and the defect signal
leads to an easily interpretable signature, called "thermal signature". In the case of surface
defects such as corrosion areas, rough surface etc, the dominating effects are the optical
ones, which produce "optical signatures". In the case of open-cracks on metallic structure
showing rough and rusty surface conditions, optical and thermal effects are competitive
and lead to hardly interpretable signatures which make the technique hardly efficient if
applied without any processing. In what follows, the images obtained by the inspection of a
mockup containing open fatigue cracks and showing severe surface conditions are
considered.
Inspected Mockup And Photothermal Images
The mockup used in this study is a structure with plasma-sprayed chromium cast iron on
ferritic steel, showing open fatigue-cracks and rough surface conditions (presence of rust
and high variations of the surface topology - up to 5mm -). An example of photothermal
image obtained by multiple ID scanning of the mockup with a 0 mm offset between the
excitation and detection beams, is shown in figure 2. Large open cracks (15jim width) are
easily seen. However, because of the surface roughness, the presence of highly perturbing
optical effects (presence of rust) makes an automatic crack detection difficult, and also
makes small cracks (lum width) hardly visible. Therefore, in order to enhance the
visibility of the cracks an image processing technique, based on the identification of the
physical effects is proposed in the following section.
675
Open-cracks
Rust stains
Rust on beads
FIGURE 2. Raw photothermal image obtained at speed 5mm/s and for a Omm offset
FIGURE 3. Normalized photothermal image
PHYSICAL EFFECTS IDENTIFICATION
A first identification method, called normalization was proposed in [1]. The
normalization is based on the morphological differences between the thermal and optical
signatures, according to the scanning direction of the structure under test for a given offset.
Indeed, by subtracting the raw image obtained by a forward scanning from the raw image
obtained by a backward scanning, a resulting image is obtained showing highlighted
thermal signatures (figureS), since the subtraction is mainly constructive in the case of
thermal signatures and destructive in the case of optical signatures. However, this
technique allows to enhance only the thermal effect at each offset [1].
In this paper, in order to improve the crack detection, the authors generalize this
technique in a multi-image scheme to separate and identify several physical effects. The
proposed method is based on a multiple principal component analysis (PCA),
simultaneously applied to the whole set of photothermal images of different offsets.
Principle Of Principal Component Analysis
Let us define M as a matrix of raw data measurement mi provided by the camera.
The signals of M result from the combination of the physical source siy (thermal effect,
optical effects, noise...) through the unknown transfer matrix T of the camera. If r sources
676
si constituting the source matrix Z are considered and p>r measures signals w/ are
available, the matrix M can be written as :
x~
Af =
m2
V
= T-
s2
pxr
mp_
(1)
= r-s
pxr
_Sr_
The PCA separates the different sources constituting the signals of the matrix M.
This separation is realized by the projection of M on the subspace formed by the
eigenvectors of the covariance matrix MA/ [2], as expressed by:
Kl
o
0
U
(Q }M =
ACP
0
0
0
0
s=
= RA
I'll
where QACP is the eigenvector matrix of MA/ classified by descending order, \tjt
th
(2)
is the
ACP
norm of they/ column of T. The lines of the result matrix R , called components of
RACP, are constituted of the separated sources obtained within the multiplicative factor
± I tji I . However, if the sources are separated they are not identified, that is to say one do
not a priori know on which component of RACP the different physical effects will appear.
But by judiciously choosing the covariance matrix MA/ it is possible to force a particular
physical effect to appear on the first component of RACP as describe in the following
section.
Identification Of The Optical Effects
Choosing a specific raw data measurement matrix Mop essentially relative to optical
effects, the first component of the result matrix RAfp will inevitably correspond to the
separated optical effects. Therefore, the first vector Vop of the eigenvector matrix QA^P is
the best projection operator for identifying the optical effects. The optical effects of the
whole photo thermal image set is then obtained by the projection of the whole image set
along Vop.
Here, the matrix Mop is build by the inspection of a reference surface containing a
black line drawn with a pencil, i.e. a high absorptivity effect without thermal effect.
The implementation of the optical identification on the studied structure is shown in
figure 4, where it can be seen that the optical effects (rust on cast bead, rust stains...) are
clearly identified, separately from the thermal effect.
677
FIGURE 4. Identification of the optical effect
FIGURE 5. Identification of the thermal effect
Identification Of The Thermal Effect
The same identification method is implemented for the thermal effect, thanks to the choice
of a data measurement matrix Mth essentially relative to the thermal effect (i.e. relative to
an a priori known open-crack in absence of perturbation effects). Then, the first vector Vth
of the obtained eigenvector matrix Qfhcp is used as the projection operator for identifying
thermal effect of the whole image set.
However, the efficiency of the thermal effect identification depends on the chosen
reference. Indeed, the chosen open crack essentially provides thermal effect but also
optical effects, due to black body effect [3]. Therefore, in order to fully reject the optical
effects from the thermal effect identification, the projection operator Vth is modified to be
orthogonal to the projection vector Vop determinated in the previous section, thanks to the
Gram-Schmidt orthogonalization scheme [4].
The implementation of the thermal identification technique on the studied structure
is shown in figure 5. It can be seen that the open cracks visibility is highlighted
(comparatively to the raw image of figure 2 and the optical effects are significantly
reduced comparatively to the normalized image shown in figure 3.
APPLICATION TO OPEN-CRACK DETECTION
The detection consists in obtaining a binary image representative of the presence of
open-cracks. The detection is implemented using a ID Continuous Wavelet Transform
(CWT) applied on the identified thermal effect image Ith shown in figure 5. For each line
Ilth(x) of Ith, the obtained wavelets coefficients c% are expressed by [5]:
678
= f/;U*)K*
(3)
where x denotes the spatial position, a denotes the dilatation scale, b the spatial shift. The
wavelet set y/atb(x)is build from the mother wavelet i//(x) by means of time shift and
dilatation, as expressed by:
x
~b
(4)
As the thermal effect exhibits a bipolar signature, the mother wavelet \j/(x) is
chosen to be a first derivative of a gaussian, because of the similarity of its shape.
The crack-detection image is a binary image resulting from the following decision
scheme applied on each line:
(5)
0 otherwise
where A, is a threshold value. The detection performances are characterized using Receiver
Operating Characteristic (ROC) curves [6]. The ROC curves exhibit the Good Detection
Rate (GDR) versus the False Alarm Rate (FAR) (Figure 7). Best detection performances
are obtained when the distance of the curve to the point (0,1) is minimum. Both values of/I
and a are chosen in order to minimize this distance, as shown in figure 6.
The implementation of the detection method leads to a 95% GDR for a 4.4% FAR,
when applied to the thermal image, as shown in figure 7, and to a 86% GDR and 5.5%
FAR when applied to the normalized image. The corresponding binary detection images
are presented in figure 8 and figure 9 respectively.
Global minimum
value of d
<D
o
Threshold value A
Scale a
FIGURE 6. Retrieving of the minimum of d in the (h,a) plane.
679
Thermal Image
•Normalized image
0.1S
0.2
0-JS
False Alarm Rate
FIGURE 7. ROC curves for open crack detection.
FIGURE 8. Open-crack detection binary image obtained from the "Thermal Image"
FIGURE 9. Open-crack detection binary image obtained from the "Normalized Image".
CONCLUSION
In this contribution, a physical identification effects method is proposed in order to
improve the open-cracks detection. This signal processing method permits to separately
identify both thermal and optical effects from the raw images. This method provides
improved detection performances. Furthermore, the presented identification scheme could
be extended to a multi-source identification process in order to provide a larger
characterization of the target under test.
680
ACKNOWLEDGEMENTS
The authors thank the Technical Center of FRAMATOME-ANP and the
Department of the mechanic of the solid and the damage of ONERA (Chatillon, France)
for technical Assistance.
REFERENCES
1.
2.
3.
4.
5.
6.
Legrandjacques L., Krapez J.C., Lepoutre F. and Balageas D. "Nothing but the
Cracks: A New Kind of Photothermal Camera", 7th European Conference on NonDestructive Testing, Copenhagen, Denmark, pp. 2651-2656, 26-29 May 1998.
Saporta G., Probabilites, analyse de donnees et statistique, Technip, 1990.
Kaufman L, Chang P.T., Hsu H.S., Huang W.Y., Shyong D.Y., « Photothermal
radiometric detection and imaging of surface cracks », J. Nondestructive Evaluation,
vol. 3,n°2, 1987, p. 87-100.
Rice J. R., « Experiments on Gram-Schmidt Orthogonalization », Mathematics of
Computation, Volume 20, Number 94, pp. 325-328, April 1966.
Cohen, J. Kovacevic, "Wavelets the Mathematical Background". Proceedings of the
IEEE, vol. 84 n°4, pp. 514-522. April 1996.
Ergan J.P., Signal detection theory an ROC analysis, Series in cognition and
perception, New-York: Academic Press, 1975.
681