ADVANCED ULTRASONIC FLAW SIZING OF STRESS CORROSION

CRACKING IN THICK STAINLESS STEEL

Y. Mizutani1, Y. Kurokawa1 and M. Mayuzumi1

1. Department of Mechanical Sciences and Engineering, Tokyo Institute of Technology, Tokyo, Japan

Abstract: The TOFD (Time-of-Flight Diffraction) method is the one of the most accurate flaw sizing methods. This
study aimed to increase the sophistication of the TOFD system by utilizing time-frequency analysis (wavelet
transform) for flaw sizing on thick stainless steel welds. A wavelet algorithm was implemented into our TOFD
system, and a preliminary test was conducted for a 46.2 mm-thick stainless steel specimen with an artificial
penetration slit (0.8mm width, 8.4-9.9mm height). As a result, we found that the frequency component of the three
main echoes observed used for sizing (lateral wave, diffraction echo and back surface echo) were different. This
feature can be effectively used for signal type classification during flaw size testing. The crack depth was measured by
using the extracted specific frequency component of TOFD signals by wavelet transform. We confirmed that the
sizing accuracy of the proposed method is similar to the traditional method using raw data. Finally, a weld joint
austenitic stainless steel specimen with stress corrosion cracking was prepared for a feasibility study. Crack tip
diffraction echoes can be separated from other waves by the proposed method, and the S/N ratio of the detected signal
was enhanced by our method. Overall, the method proposed in this study makes relevant contributions to the flaw
sizing of thick stainless steel (especially at welds).

1. INTRODUCTION TOFD signals detected on 9% of nickel steel weld joints.

Accurate ultrasonic flaw sizing is an important issue for They reported that flaw indications in D-scan images were
predicting the remaining life of industrial plants such as atomic remarkably clear using their approach. E. P. Moura et. al., [6]
power plants [1]. The TOFD (Time-of-Flight Diffraction) used continuous wavelet transform for preprocessing of TOFD
method is said to be one of the most accurate flaw sizing signals for pattern recognition of weld defects. In this study,
methods [2]. The TOFD method is not based on the amplitudes in order to enhance the S/N ratio of TOFD signals and to
of echoes, but uses the travel time of a diffracted wave at the provide additional significant information for discrimination of
crack tip to determine crack depth. However, the method is the main echoes, the usefulness of continuous wavelet
not applicable to thick steels due to the amplitude of the crack transform is examined.
tip diffraction echo being weak, and that this weak echo is
attenuated in thick materials. Moreover, when the TOFD 2. SPECIMEN AND PRINCIPLE OF TOFD TECHNIQUE
method is utilized for austenitic stainless steel welds, the Figure 1 schematically shows the principle of the TOFD
ultrasonic beam is subjected to many perturbations: deviation, technique (upper) and an A-scan image of a TOFD signal
partition, distortions and sometimes ghost echoes due to the (bottom). The technique utilizes two transducers, one as a
anisotropic and heterogeneous metallurgical structures of welds transmitter and the other as a receiver. The first echo to reach
[3]. When ultrasonic flaw sizing is carried out, it needs to the receiving transducer corresponds to the lateral wave
distinguish the main echoes (crack tip diffraction echo, surface (indicated by “L” in the figure). If there is a crack in the
back echo and lateral wave) from other many perturbations. specimen, a weak amplitude of the crack tip diffraction echo
Following this, signal type classification of the three types of (indicated by “D”) will be observed in front of a large
main echoes is also required. A high level of skill and amplitude of the back wall echo (indicated by “B” in the figure).
knowledge is required for this discrimination [4]. Hatanaka et. The crack height can be estimated by using the arrival time
al., [5] used discrete wavelet transform as post-processing for differences between these echoes. Figure 2 shows a picture of
transducers mounted on a specimen (45.3mmW X 200mmL X 1.0
T D B

Amplitude, V
46.2mm ) made of austenitic stainless steel (SUS304) with a D3
L2 L D1
penetration slit (0.8 mm width, 8.4-9.9mm height). A pair of 0
commercial longitudinal ultrasonic transducers for the TOFD
L1 D2
method (5MHz in the center frequency and 60 degrees of
incident angle, Japan probe: B5K5 X 5LA60) was used for -1.0
13 20 26
these measurements. The distance between the two Time, µsec
transducers was set at 96 mm. Detected signals were digitized
Fig. 3 Observed TOFD signal for the specimen shown in Fig. 2.
by an A/D converter (sampling frequency: 100 MHz) and fed to
a computer to be analyzed by software especially developed for Continuous wavelet transform [7] of 1 MHz to 7MHz was
this study. conducted and the wavelet contour map is shown in Fig. 4.
The main echoes of “L”, “D” and “B” clearly appear in the
contour map, and the frequency components of “L”, “D” and
“B” are clearly different. The wave “L” contains a low
frequency component; hence the “D” contains a high frequency
component. The wave “B” contains a wide band frequency.
This feature can be effectively used for signal type
classification during flaw sizing tests. Next, we attempted to
estimate crack depth by using the wavelet coefficient of the
TOFD signal. Since “L” and “B” waves have a 3MHz
component in common, we extracted the 3MHz wavelet
Fig. 1 Principle of the TOFD technique (upper) and typical component by wavelet transform as shown in Fig. 5. Unlike
A-scan image (bottom). raw signals (Fig. 3), “L” and “D” waves have a single peak.
This suggests that flaw sizing independent of inspector skill
and experience can be conducted using this treatment.
Transmitter Receiver Furthermore, there is a possibility to automate flaw sizing
96mm utilizing a computer, since it is easy to find peak arrival time
for a single peak with a computer. Crack depth was estimated
using arrival time differences between “L” and “D” coefficients.
46.2mm

Z Crack depth estimated by these two peaks was 37.7 mm.

Y 37.8 ～ 36.3 mm
7
Slit D B
Frequency, MHz

X L
0.8 mm in width
5
Fig. 2 Specimen with a penetration slit and the experimental
setup 3

1
3. ADVANCEMENT OF TOFD TECHNIQUE USING 13 20 26
Time, µsec
WAVELET TRANSFORM
Figure 3 shows TOFD signals observed on the specimen where Fig. 4 Wavelet contour map of a TOFD signal (shown in Fig. 3)
the distance of crack tip from surface is 37.5mm. The symbol
“L” indicates a lateral wave, “D” a diffraction echo and “B” the 4e-4
Wavelet coefficient

back wall echo. In spite of the small amplitude of noise B

signals observed between “L” and “D” signals, sharp peaks of
2e-4
“L” and “D” signals were observed. Crack depth (distance L D
between the crack tip and the front surface) was estimated
using arrival time differences between the “L” and “D” signals. 0
13 20 26
When the first peaks (L1 and D1 in Fig. 3) were used for the Time, µsec
estimation, the crack depth was estimated as 37.6 mm. On the
Fig. 5 Wavelet coefficient of a TOFD signal (shown in Fig. 3)
other hand, when the second peaks (L2 and D2) were used, a
at 3 MHz.
crack depth of 37.3 mm was estimated.
Figure 6 shows a D-scan image, which is constituted by

Position of sensors (Y), mm

7
stacking multiple raw signals that are acquired scanning the L D B
transducer-receiver pair in the Y direction (see Fig. 2) on the
specimen surface. The D-scan image represents a
cross-sectional view of the specimen and shows the presence of
defects (see “D” of Fig. 6). When the time scale of the
waveform is converted to a depth scale through the thickness of
the specimen, flaws can readily be sized directly from the 34
13 20 26
D-scan. In Fig. 6, the arrival time of D comes early as Y Time, µsec
(sensors position) becomes large. This represents that the
Fig. 8 D-scan TOFD image after peak emphasis processing.
crack depth becomes shallow at the point where Y is large, and
this agrees with the distribution of the actual crack depth.
Although it was a small amplitude, a noise signal is observed 4. FEASIBILITY STUDY USING A WELD JOINT
between L and D. Moreover, several peaks were observed in SPECIMEN WITH STRESS CORROSION CRACKING
each echo, which may cause trouble for accurate flaw sizing. To investigate the feasibility of our developed system,
In order to overcome these two points, wavelet transform is ultrasonic testing was conducted for a weld joint austenitic
used for constructing a D-scan image. Figure 7 shows a stainless steel specimen with SCC (Fig. 9). An enlarged side
D-scan image constituted by stacking multiple 3MHz wavelet view of the welded part is shown at upper left. The crack is
coefficients of TOFD signals. Although the peak of each echo traced and emphasized by a thick line. The distance of the
became single and the S/N ratio was improved, the image pair of transducers is set to 34 mm and ultrasonic testing is
became indistinct compared to the original image. This is due conducted from the upper surface. Since the crack height in
to the peak of the wavelet coefficient at this frequency range the specimen is unknown, the neighbourhood area on the side
becoming broad. (that is, the time resolution of wavelet (5mm away from the side) was inspected and the estimation
transform at this frequency range is low.) Since each echo has result is compared to the crack height observed from the side
a single peak, a threshold value is set and peaks are emphasized. (21.9 mm). The actual crack height at the inspected area was
The result is shown in Fig. 8. The image is reasonably estimated to be around 22 mm from this visual observation.
improved compared to the original one (Fig. 6). Figure 10 shows the TOFD signals detected at 5 mm away
from the side. Several peaks are observed and it is difficult to
distinguish each echo. Arrival times for the lateral wave (L)
Position of sensors (Y), mm

7
L D B and the diffraction echo (D) estimated from the assumed crack
height (around 22 mm) and the sensor distance are surrounded
by the dotted line. Figure 11 shows a wavelet contour map of
the TOFD signal (shown in Fig. 10). This result shows that
the observed signal consists of two frequency bands. If the
result of the fundamental study (Chap. 3, Fig. 4) is taken into
consideration, it can be estimated that the wave packet that
34
13 20 26 arrives 4.5～7.0μsec with a low frequency is a lateral wave,
Time, µsec while the packet that arrives 6.0 ～ 7.5 μ sec with a high
Fig. 6 D-scan TOFD image for the specimen shown in Fig. 2. frequency is a diffraction echo. Figure 12 shows the extracted
7 wavelet coefficient at 3 MHz. Two distinct peaks are
Position of sensors (Y), mm

L D B

34
13 20 26
Time, µsec
Fig. 7 D-scan image constituted by a wavelet coefficient of Fig. 9 Weld joint austenitic stainless steel specimen with SCC.
the 3MHz component.
observed and the crack height estimated from these two peaks 3) The crack depth can be measured using extracted specific
was 23.4 mm. Although the true crack height is unknown, the wavelet coefficients of TOFD signals. Single peaks for
measured crack height (23.4mm) is similar to the crack height each echo are observed and show the possibility of
(21.9mm) visually observed from the side. Further examination automating flaw sizing utilizing a computer.
will be needed for applying the developed system to a real 4) The D-scan image constituted by stacking wavelet
crack at an actual work site. coefficients with emphasis treatment is improved
0.5
compared to the original D-scan image.
L D 5) The sizing of stress corrosion cracking in a weld joint
Amplitude, V

austenitic stainless steel specimen was conducted. The

0 crack tip diffraction echo can be separated from other
echoes with the proposed method. The estimated crack
height was similar to the crack height visually observed
-0.5
4 7 10
from the side of the specimen.
Time, µsec
Fig. 10 Observed TOFD signal for the specimen shown in Fig. 6. ACKNOWLEDGEMENTS
9. Ishikawajima Inspection & Instrumentation Co., Ltd lent the
weld joint austenitic stainless steel specimen with SCC for this
7
research. Prof. H. Inoue of Tokyo Tech. gave in valuable
Frequency, MHz

D advices about wavelet transform calculations. This research

5
was partially supported by a Ministry of Education, Science,
L Sports and Culture, grant-in-aid for young scientists (B).
3

7. REFERENCES
1
1. K. Kashima, T. Nomura and K. Koyama, Current Status of
4 7 10
Time, µsec Japanese Code on Fitness-for-Service for Nuclear Power
Fig. 11 Wavelet contour map of the TOFD signal (shown in Plants, 2004 ASME/JSME PVP Conference (San Diego)
Proceedings PVP, Vol. 480, pp.227-234, 2004,
Fig. 10)
2. T. Mihara, M. Nomura and K. Yamanaka, Relations between
6e-5 Crack Opening Behavior and Crack Tip Diffraction of
Wavelet coefficient

L D Longitudinal Wave, Proceeding of SPIE, Vol. 4703, 137-145,

2002
3e-5
3. H. Yoneyama, Ultrasonic Testing of Austenitic Stainless
Steel Welds, J. of the Japanese Society for Non-destructive
0 Inspection, Vol. 53, No. 11, 687-692, 2004
4 7 10 4. H. Yoneyama, T. Sugibayashi and A. Yamaguchi, Results of
Time, µsec
UT Training for Defect Detection and Sizing Technique
Fig. 12 Wavelet coefficient of the TOFD signal (shown in Fig. using Specimens with SCC, J. of the Japanese Society for
10) at 3 MHz. Non-destructive Inspection, Vol. 54, No. 4, 187-191, 2004
5. H. Hatanaka, N. Ido, M. Furikoma and T. Arakawa, Signal

5. CONCLUSION Processing Method Based on Wavelet Transform for Testing

In this study, an advanced TOFD technique implementing a of 9% Ni Steel Welds Using Ultrasonic

wavelet algorithm was proposed for thick austenitic stainless Time-of-Flight-Diffraction Technique, J. of the Japanese

steel crack sizing. The results in summary are; Society for Non-destructive Inspection, Vol. 53, No. 2, 88-92,

1) A wavelet transform algorithm was implemented for the 2004

TOFD system and a preliminary test was conducted on a 6. E. P. Moura, R. R. Silva, M. H. S. Siqueira and J. M. A.

thick stainless steel specimen with an artificial penetration Rebello, Pattern Recognition of Weld Defects in

slit. Preprocessed TOFD Signals Using Linear Classifiers, J. of

2) It was found that the frequency components of the main Nondestructive Evaluation, Vol. 23, No. 4, 163-172, 2004

echoes used for sizing were different. This feature can 7. Chui.C.K., An Introduction to Wavelets, Academic Press,

be effectively used for signal type classification. 60-61, 1992