You are on page 1of 7

642

IEEE TRANSACTIONS ON MAGNETICS, VOL. 34, NO. 3, MAY 1998

A Novel Signal Processing Technique for


Eddy-Current Testing of Steam Generator Tubes
Gangzhu Chen, Atsunori Yamaguchi, and Kenzo Miya
AbstractIn eddy-current testing of steam generator tubes of
nuclear power plants, the signals of defects may be corrupted
by noise and other nondefect signals arising from the probe
lift-off and the structures attached to the tubes, resulting in
unreliable detection and inaccurate characterization of defects.
In this paper, a novel signal processing technique is presented to
reduce the noise and nondefect signals by the use of a wavelet
transform. The noise and nondefect signals are reduced by first
decomposing testing signals into wavelet components and then
modifying the wavelet coefficients. The defect signals embedded in
noise and nondefect signals are reconstructed through the inverse
wavelet transform of the modified wavelet coefficients. The results
of processing the one-dimensional and two-dimensional signals
from eddy-current testing of tube test pieces show that this signal
processing technique is effective for extracting defect signals
embedded in noise and nondefect signals.
Index Terms Defects, eddy-current testing, noise, nondefect
signals, signal processing, steam generator tubes, wavelet transform.

I. INTRODUCTION

DDY-current testing (ECT) [1][11] is one of the most


effective nondestructive testing techniques for detecting
defects in conducting materials. A typical application of the
ECT is the in-service inspection of the heat exchanger tubes
of a steam generator (SG) in a nuclear power plant [1][5]. By
the ECT inspection, it is necessary to detect and characterize
defects in the SG tubes reliably so that proper remedial
measures can be taken in time and the safe operation of
nuclear power plants can be guaranteed. When the ECT signals
from the SG tubes have a high signal-to-noise ratio (SNR),
defects can be detected and characterized by recognition
techniques such as those using impedance plane diagrams [5],
inverse analysis [7], artificial neural networks [8], and Fourier
descriptors [9]. The ECT signals of defects, however, may be
corrupted by noise and/or nondefect signals due to the probe
lift-off and the structures used to hold the SG tubes [2][6],
[10]. Because this corruption will adversely affect the detection
and characterization of defects, it is necessary to enhance the
SNR of the ECT signals by using signal processing methods
[3][6] before using the recognition techniques.
Fourier filtering and multi-frequency mixing (MFM) techniques [3], [6] are usually used to process noise and nondefect
signals in the ECT signals from the SG tubes. Fourier filtering
techniques are effective for removing noise and the nondefect
Manuscript received July 9, 1997; revised December 9, 1997.
G. Chen and A. Yamaguchi are with Tsurumi Research and Development
Center, Japan Power Engineering and Inspection Corporation, Yokohama, 230
Japan.
K. Miya is with Nuclear Engineering Research Laboratory, University of
Tokyo, Tokyo, 113 Japan.
Publisher Item Identifier S 0018-9464(98)01873-1.

signals due to probe lift-off, but in Fourier filtering, there


is a tradeoff between the SNR and useful information of
the signal being processed: some of the information of the
defect signals may be lost when the noise is reduced [12].
The nondefect signals due to the structures used to hold the
tubes, on the other hand, are conventionally processed by the
MFM techniques. Although these techniques can detect the
defect signals embedded in the nondefect signals, quantitative
evaluation of the defects is complicated because the result of
the MFM processing is the combination of defect signals from
several exciting frequencies [6].
This paper describes a novel signal processing technique
that reduces the noise and the nondefect signals by using
the discrete wavelet transform (DWT). The wavelet transform
[12][14] is a relatively new signal processing technique very
useful for analyzing the local characteristics of signals. The
signal processing algorithm presented here consists of three
steps: decomposition, filtering, and reconstruction. The ECT
signals are first decomposed into wavelet components in such a
way that noise and nondefect signals can be separated from the
defect signals by the wavelet coefficients. Then, the wavelet
coefficients are modified by using a filtering algorithm to
remove the undesired components. Finally, the defect signals
are reconstructed by inverse discrete wavelet transform of
the modified coefficients. Since the wavelet decomposition
provides information on not only frequency contents, but also
spatial positions, noise can be removed without losing the information of the defect signals. And the nondefect signals from
the structures attached to the tubes can be reduced without
evaluating the ECT signals from multiple exciting frequencies.
This wavelet processing method enhances the SNR of the ECT
signals and accurately reconstructs the defect signals so that
defects can be detected reliably and characterized accurately.
In addition, a fast implementation of the DWT makes possible
the real-time processing of the ECT signals.
The paper is organized as follows. Section II deals with collection of the ECT signals used in this study. Section III briefly
reviews the DWT, and Section IV describes the signal processing algorithm with the DWT. Section V gives some examples
of processing one-dimensional (1-D) and two-dimensional (2D) ECT signals using the algorithm. Section VI summarizes
the results of this study.
II. COLLECTION

OF

EDDY-CURRENT SIGNALS

A. Eddy-Current Testing of SG Tubes


Steam generators constitute an important link between the
primary and secondary loops in a nuclear power plant. They
contain thousands of heat exchanger tubes running parallel to

00189464/98$10.00 1998 IEEE

CHEN et al.: EDDY-CURRENT TESTING OF STEAM GENERATOR TUBES

643

Fig. 1. Experimental setup.

each other, and ECT is conducted by pulling an eddy-current


probe through each of the tubes (Fig. 1). An alternating current
in the exciting coil of the probe produces a varying magnetic
field that induces eddy currents in the surrounding wall of the
tube. The eddy currents, in turn, set up a secondary magnetic
field that opposes the primary magnetic field. The presence of
a defect will disturb the eddy currents and thus change the
magnetic field distribution, and changes in the magnetic field
can be monitored as changes in the impedance of the coil.
The probe signals, however, are also contaminated by noise
and factors such as the probe lift-off (the distance between the
probe and the inner surface of the tube) and the support plate
used to hold the tubes. This study discusses the use of wavelet
transform to reduce the noise and nondefect signals, and the
discussion is based on data collected in experiments designed
to produce ECT signals typical of those from SG tubes.
B. Data Collection
The test pieces used in the ECT experiments were Inconel tube test pieces with electro-discharge machining (EDM)
flaws. The inner diameter and the wall thickness of the test
pieces were 19.7 and 1.27 mm, respectively, the same as those
of the SG tubes in nuclear power plants. Flaws of various
depths were made from the outer surface of the tubes. The
flaws were 5 mm in length and 0.2 mm in width. They were
oriented in axial and circumferential directions since defects
in the SG tubes are usually axial and circumferential cracks.
The experimental setup for collecting ECT signals is shown
in Fig. 1. A pancake coil with an outer diameter of 3 mm was
used to generate eddy currents and to pick up the signals due
to changes in the eddy currents. The probe lift-off was about
0.5 mm and the frequency of the exciting current was set to
either 100 kHz or 600 kHz. The 1-D and 2-D ECT signals were
acquired by moving the probe axially and/or circumferentially
with the actuator controller. The probe moved in 0.5 mm
increments axially and 0.3 mm increments circumferentially.
The ECT signals picked up by the probe were digitized by
an A/D converter and collected in a personal computer for
analysis.
Figs. 2 and 3 show some examples of the 1-D signals picked
up when the frequency of the exciting current was 100 kHz.
The signal in Fig. 2 was obtained by rotating the probe across
a circumferential flaw with a depth of 30% of the tube wall

Fig. 2. ECT signal obtained by rotating the probe across a circumferential


flaw.

Fig. 3. Composite ECT signal from a support plate and a circumferential


flaw.

thickness. The abscissa represents the probe position and the


vertical axis shows the magnitude of the ECT signal. The lowfrequency change in the signal is the so-called lift-off noise
caused by the variation in the probe lift-off. It is seen that the
flaw signal is embedded by the lift-off noise.
The signal in Fig. 3 was obtained by moving the probe
axially across a flaw near which a simulated support plate
was arranged. The depth of the flaw is 60% of the tube wall
thickness. This probe signal is the combination of the flaw
signal and the support plate signal, and the flaw signal is
partially embedded by the support plate signal.
In the signals shown in Figs. 2 and 3, the lift-off noise
and the signal of the support plate are much stronger than
the flaw signals and may result in unreliable detection and
characterization of the defects. White noise may also interfere
with the recognition of the defects although this noise is not
obvious in Figs. 2 and 3 because of the large lift-off noise
and support plate signal. White noise, lift-off noise, and the
signal of the support plate are typical interference signals in
ECT of SG tubes, and the flaw signals corrupted by these
interference signals are taken as the objects being analyzed in
this study.

644

IEEE TRANSACTIONS ON MAGNETICS, VOL. 34, NO. 3, MAY 1998

III. DISCRETE WAVELET TRANSFORM


The DWT of a signal
is defined as the projection of
the signal on a set of wavelet functions
(1)
where

are generated from a mother wavelet

by
(2)

The wavelet functions are localized in space and frequency,


and their spatial locations and frequency contents are respectively determined by the shift parameter and the scale
parameter . The smaller the scale parameter , the higher
the frequency it corresponds to. The wavelet coefficients
reflect the local characteristics of
in space
and frequency.
is appropriately chosen, the signal can be reconIf
structed by the following inverse discrete wavelet transform
(IDWT)
(3)
satisfying (3), there
Associated with a mother wavelet
is usually a scaling function
by which
can be
approximated on multiresolution levels. The approximation of
on the level is defined by

parameter and it can be assumed without loss of generality


to be one. This assumption allows (9) to be rewritten as
(10)
.
where
This wavelet decomposition enables the noise in the signal
to be separated from the useful components. Further, the useful
components in the signal can be extracted by processing the
decomposing coefficients.
The wavelet decomposition specified by (1), (5), and (10),
however, are not feasible because in practice we deal with
discrete ECT signal sequences that need to be processed
rapidly. Fortunately, the discrete sequences can be efficiently
decomposed and reconstructed in the wavelet domain by an
algorithm referred to as the fast wavelet transform (FWT), a
detailed description of which can be found in [13] and [14].
IV. SIGNAL PROCESSING ALGORITHM

WITH THE

DWT

denote the probe signal picked up from a flaw,


Let
and some noise
which is the mixture of the flaw signal
or nondefect signal
. We reconstruct the
from the
as accurately as possible by using the DWT.
A. Wavelet Decomposition of Eddy-Current Signals
The probe signal
is first decomposed into
frequency
levels by using the fast algorithm of the DWT. From (10), we
have

(4)
where
(5)
and
(6)
The approximation can be decomposed into the next coarser
resolution level by the wavelet functions
(7)
where
(8)
The relationship in (7) enables the following equation to be
derived from (3)
(9)
where
represents the changes or details of the signal on
the level and
is a coarser approximation of the signal
on the level . Usually, there is a lower bound on the scale

(11)
are choWhen the mother wavelet and the frequency level
sen properly, the flaw signal
and the noise
can be
separated from each other by the wavelet decomposition. The
while the
low-frequency noise can be decomposed into
flaw signal and noise of higher-frequency can be decomposed
into
. In the
at middle frequency levels, the flaw
signal usually has larger components because it is usually
on the high frequency levels
a bandpass signal. The
contain high-frequency information of the flaw signal, but they
are dominated by high-frequency noise.
Fig. 4 shows an example of the wavelet decomposition of
an ECT signal. The waveform shown at the top of Fig. 4 is
the signal picked up by rotating the probe across an axial flaw
whose depth was 20% of the tube thickness. The frequency of
the exciting current was 600 kHz. The flaw signal is a bandpass
signal appearing around the position
. The probe signal
also contains white noise and an undesired low-frequency
change due to the probe lift-off. The other waveforms in Fig. 4
are a portion of the result of decomposing the signal into five
frequency levels (
) by using the Daubechies wavelet of
order ten [14]. It is evident that the lift-off noise is extracted
by the
and the white noise is mainly decomposed into high

CHEN et al.: EDDY-CURRENT TESTING OF STEAM GENERATOR TUBES

645

set to zero
(12)
come from both the flaw
The wavelet coefficients in
signal and higher-frequency noise. To remove the noise, it is
necessary first to identify the wavelet coefficients of the flaw
signal. At the middle frequency levels, the wavelet coefficients
of the flaw signal can be easily selected by setting thresholds
since their magnitudes are usually larger than those of the
wavelet coefficients from the noise. For a fixed frequency
level a threshold
for modifying the wavelet coefficients
is set. If the magnitude of
is larger than
, it is preserved as the flaw signal component; otherwise,
it is set to zero. Let the modified wavelet coefficient be
. Then we have
if
if

Fig. 4. Wavelet decomposition of an ECT signal obtained by rotating the


probe across an axial flaw.

frequency levels as shown in


. On the other hand, the main
. It is also evident
pattern of the flaw signal appears in
that in
and
the larger components are localized near
the flaw.
As shown by the above example, the lift-off noise can
easily be separated from the flaw signal by removing the
from the wavelet decomposition. On the other hand,
in
, although the frequency content of the white noise
overlaps with that of the flaw signal partially, the spacefrequency characteristics of the white noise are different from
those of the flaw signal. That is, across the frequency levels,
the wavelet coefficients
localized in the spatial
positions far from the flaw signal will have features different
localized near the flaw. Using
from those of the
this feature difference, we can reduce white noise effectively
by processing the wavelet coefficients as described in the
following subsection.

(13)

where
At the higher-frequency levels, on the other hand, the larger
wavelet coefficients correspond to both the white noise and
the flaw signal, and the coefficients of the flaw signal cannot
be identified by setting thresholds as in the above. As shown
in Fig. 4, however, the white noise decays rapidly as the
frequency level increases, but the flaw signal does not. This
characteristic allows us to identify the wavelet coefficients of
the flaw signal by defining the following spatial correlation:

(14)
where the th spatial region is localized in that denoted by
. A large value of the spatial correlation
means
that the wavelet coefficient
may correspond to the
flaw signal. Thus, the wavelet coefficients can be modified by
evaluating the spatial correlation as follows:
if
if

(15)

are thresholds defined in advance.


where
With the modified wavelet coefficients, the flaw signal can
reconstructed by the fast algorithm of the IDWT as
(16)

B. Processing of Wavelet Coefficients and


Reconstruction of Flaw Signals
After the wavelet decomposition, the decomposing coefficients are processed on the basis of the principle that those
featuring the flaw signal remain unchanged and the others are
set to zero to filter out noise.
The coefficients in the
are contributed by the lowfrequency noise, such as the lift-off noise, so all of them are

V. RESULTS

AND

DISCUSSION

The 1-D and 2-D ECT signals from the tube test pieces were
processed by using the noise reduction algorithm, and some
typical examples are given in this section. The mother wavelet
functions used in the following examples are those proposed
by Daubechies [14]. Hereafter, the Daubechies wavelet of
order
is denoted DN.

646

IEEE TRANSACTIONS ON MAGNETICS, VOL. 34, NO. 3, MAY 1998

(a)
(a)

(b)

(b)

(c)

(d)
Fig. 5. Noise reduction of the ECT signal illustrated in Fig. 4: (a) ECT
signal, (b)(c) wavelet processing of the ECT signal, (b) signal obtained by
removing lift-off noise, (c) reconstructed flaw signal, and (d) Fourier filtering
of the ECT signal.

A. Results of 1-D Signal Processing


The first example deals with the ECT signal shown in Fig. 4,
and the original signal waveform is redrawn in Fig. 5(a). The
lift-off noise and white noise were processed separately by
using different mother wavelets. First, the original signal was
decomposed into five frequency levels by the DWT with the
D10 wavelet. The lift-off noise was removed by setting to
. The signal in Fig. 5(b) was
zero the coefficients in the
obtained by the IDWT of the modified coefficients. That signal
was again decomposed into five frequency levels, this time by
the DWT with the D3 wavelet. The wavelet coefficients were
modified to reduce the white noise by using the algorithm
described in the previous subsection. The reconstructed flaw
signal shown in Fig. 5(c) was obtained by the IDWT of the
modified wavelet coefficients.
The D10 wavelet is used to remove the lift-off noise because
of its high regularity [14]. This high regularity ensures that
the lift-off noise can be approximated accurately by the
and thus can be removed effectively by filtering out the
.
The D3 wavelet, on the other hand, is used for reducing the
white noise because of its good space-frequency localization.
This good localization makes it possible to characterize the
flaw signal by using a small portion of wavelet coefficients
localized near the flaw signal so that the flaw signal can be
effectively separated from the noise.

Fig. 6. Result of wavelet processing of the ECT signal illustrated in Fig. 2:


(a) original signal waveform and (b) reconstructed flaw signal.

The signal shown in Fig. 5(d) is the result of reducing


the noise by using the conventional bandpass filter based on
Fourier analysis. Comparing the result shown in Fig. 5(c) with
that shown in Fig. 5(d), we can easily see that the wavelet
algorithm reconstructs the flaw signal more accurately than
Fourier filtering does. This is because Fourier analysis provides
only the frequency contents of the signals and there is a tradeoff between the SNR and the flaw signal information in Fourier
filtering. The wavelet analysis, on the other hand, allows us to
utilize the information of the signal at both frequency levels
and spatial positions. The basis functions in wavelet analysis
can be chosen with regard to the characteristics of the signal to
be processed, whereas the basis functions in Fourier analysis
are fixed as trigonometric functions. As a result, the flaw signal
can be extracted more accurately by wavelet filtering.
Figs. 6 and 7 show the results of processing the ECT signals
shown in Figs. 2 and 3. The waveform in Fig. 6(a) shows
the ECT signal picked up by rotating the probe across the
circumferential flaw with the depth of 30% of the tube wall
thickness, and the waveform in Fig. 6 (b) shows the flaw signal
extracted from the noise by wavelet processing.
The waveform in Fig. 7(a) shows the composite signal from
the support plate and the circumferential flaw with the depth
of 60% of the tube wall thickness. As illustrated in the Fig. 3,
the flaw is located near the end of the support plate. The
wavelet processing of the composite signal gave rise to the
result shown by the solid line in Fig. 7(b). Comparison of the
result with the dashed line, which shows the ECT signal of the
flaw only, reveals the flaw signal localized in the composite
signal is well extracted. When the flaw depth was less than
60% of the tube wall thickness, however, it was difficult to
extract the flaw signals from the composite signals.
B. Results of 2-D Eddy-Current Signal Processing
The signal processing algorithm for the 2-D ECT signals
is the same as that for 1-D signals except that the 2-D DWT
[13], [14] is used. Fig. 8 shows an example of 2-D signal
processing. The ECT signal shown in Fig. 8(a) is one that

CHEN et al.: EDDY-CURRENT TESTING OF STEAM GENERATOR TUBES

647

(a)

(a)
(b)
Fig. 7. Result of wavelet processing of the composite signal illustrated in
Fig. 3: (a) original signal waveform and (b) comparison of the reconstructed
flaw signal (solid-line) with the signal picked up in the presence of the flaw
only (dashed-line).

was picked up by helically rotating the probe in the tube with


an axial flaw whose depth was 20% of the tube wall thickness.
The frequency of the exciting current was 100 kHz. What can
be observed directly from the signal is white noise and liftoff noise (the circumferential low-frequency change). The flaw
signal is completely embedded in the noise. After removing
lift-off noise by using the 2-D DWT with the D10 wavelet,
we have the signal shown in Fig. 8(b). That signal was further
processed by using the 2-D DWT with the D3 wavelet, and
the flaw signal shown in Fig. 8(c) was obtained. Obviously the
SNR of the ECT signal is enhanced significantly after wavelet
processing.
VI. CONCLUSIONS
The signal processing technique using the DWT was proposed to reduce noise and nondefect signals in the ECT
signals from SG tubes. The noise and nondefect signals are
reduced by first decomposing the ECT signals into wavelet
components, and then modifying the decomposing coefficients.
The defect signals embedded in the noise and nondefect signals
are reconstructed through the inverse wavelet transform of the
modified coefficients.
The 1-D and 2-D ECT signals from tube test pieces with
EDM flaws were used to confirm the effectiveness of the
wavelet filtering technique, and the results show that noise and
nondefect signals are reduced effectively and flaw signals are
reconstructed successfully. In the processing of the composite
signals from the support plate and flaws, however, the flaw
signals can be extracted only when the depth of the flaw is
more than 60% of the tube wall thickness.
Compared to conventional Fourier filters, the wavelet filtering technique has the advantage that the SNR of the ECT
signals can be enhanced without losing the information of the
flaw signals. This ensures that defects can be detected reliably
and characterized accurately. And because the algorithm can
be implemented efficiently by using the fast wavelet transform,

(b)

(c)
Fig. 8. Wavelet processing of the 2-D ECT signal obtained by helically
rotating the probe across an axial flaw: (a) 2-D ECT signal, (b) signal obtained
by removing the lift-off noise, and (c) reconstructed flaw signal.

the wavelet technique provides a promising method for realtime processing of the signals from eddy-current testing of
SG tubes.
ACKNOWLEDGMENT
The authors would like to acknowledge the encouragement
and support of K. Kotama, President of JAPEIC, N. Uesugi,
Senior Director of Tsurumi R & D Center of JAPEIC, and H.
Yoneyama, Director of NDE Research Lab. of Tsurumi R &
D Center of JAPEIC.
REFERENCES
[1] T. Takagi, M. Hashimoto, H. Fukutomi, M. Kurokawa, K. Miya,
H. Tsuboi, M. Tanaka, J. Tani, T. Serizawa, Y. Harada, E. Okano,
and R. Murakami, Benchmark models of eddy-current testing for
steam generator tube: Experiment and numerical analysis, Int. J. Appl.
Electromagn. Mater., vol. 5, no. 3, pp. 149162, 1994.
[2] H. A. Sabbagh, R. K. Murphy, J. C. Treece, and L. W. Woo, Application of volume-integral models to steam generator tubing, Rev. Prog.
Quantitative Nondestructive Eval., vol. 14, pp. 283289, 1995.

648

[3] S. Kumano, N. Kawase, K. Kawata, and A. Kurokawa, Signal processing of rotating pancake eddy-current signal for steam generator tubes,
in Proc. 13th Int. Conf. NDE in the Nucl. Pressure Vessel Industries,
Kyoto, Japan, May 1995, pp. 413421.
[4] G. Chen and A. Yamaguchi, Enhancement of signal noise ratio of
eddy-current signals by wavelet transform, in Proc. 2nd Int. Workshop
Electromagn. NDE, Tokyo, Japan, Oct. 1996, pp. 140148.
[5] B. S. Wong and K. S. Tan, Software developed to analyze the signals
from the eddy-current inspection of heat exchanger tube, Insight, vol.
37, no. 2, pp. 8792, 1995.
[6] K. A. Bartels and J. L. Fisher, Optimal multidimensional multifrequency eddy-current mixing techniques, Rev. Prog. Quantitative
Nondestructive Eval., vol. 15, pp. 393400, 1996.
[7] G. Chen, Y. Yoshida, K. Miya, and M. Kurokawa, Reconstruction
of defects from the distribution of current vector potential T using
wavelets, Int. J. Appl. Electromagn. Mater., vol. 5, no. 3, pp. 189199,
1994.
[8] D. S. Forsyth, A. Fahr, and C. E. Chapman, An evaluation of artificial
neural networks for the classification of eddy-current signals, Rev.
Prog. Quantitative Nondestructive Evaluation, vol. 13, pp. 879886,
1994.
[9] S. S. Udpa and W. Lord, A Fourier descriptor classification scheme for
differential probe signals, Mater. Eval., vol. 47, pp. 11381141, 1984.
[10] S. Yamada, M. Katou, M. Iwahara, and F. P. Dawson, Planar-coil eddycurrent test probe, IEEE Trans. Magn., vol. 31, pp. 31853187, Nov.
1995.
[11] N. Ida and W. Lord, Finite element model for three-dimensional eddycurrent NDT phenomena, IEEE Trans. Magn., vol. 21, pp. 26352643.
Nov. 1985.
[12] Y. Xu, J. D. Weaver, D. M. Healy, Jr., and J. Lu, Wavelet transform
domain filters: a spatially selective noise filtration technique, IEEE
Trans. Image Processing, vol. 3, no. 6, pp. 747758, 1994.
[13] S. G. Mallat, A theory for multiresolution signal decomposition: The
wavelet representation, IEEE Trans. Pattern Anal. Machine Intell., vol.
11, no. 7, pp. 674693, 1989.
[14] I. Daubechies, Ten Lectures on Wavelets, CBMS-NSF Series in Applied
Mathematics, 1992.

IEEE TRANSACTIONS ON MAGNETICS, VOL. 34, NO. 3, MAY 1998

Gangzhu Chen received the Bachelors and Masters degrees in electrical


engineering from Xian Jiaotong University, China, in 1984 and 1989, respectively. She received the Ph.D. degree in nuclear engineering from the
University of Tokyo, Japan, in 1995.
Currently, she is a Researcher at Tsurumi Research and Development
Center, Japan Power Engineering and Inspection Corporation. Her research
interest includes computational applied electromagnetics, forward, and inverse
problems of eddy-current testing, and application of wavelet transform to the
signal processing methods for nondestructive testing.

Atsunori Yamaguchi received the Bachelors and Masters degrees in nuclear engineering from the University of Tokyo, Japan, in 1969 and 1971,
respectively.
He is the Deputy Director of Tsurumi Research and Development Center,
Japan Power Engineering and Inspection Corporation. He has been engaged
in the design of nuclear power plants, and his current research interest is
in the development and application of nondestructive testing techniques for
diagnosis and evaluation of the integrity of structural materials.

Kenzo Miya received the Masters and Ph.D. degrees in nuclear engineering
from the University of Tokyo, Japan, in 1966 and 1969, respectively.
He is a Professor of the Faculty of Engineering at the University of
Tokyo. He has contributed significantly to the development and application
of electromagnetosolid mechanics, computational applied electromagnetics,
fusion nuclear technology, high Tc superconductor, and inverse techniques
for nondestructive testing.
Dr. Miya is the President of The Japan Society of Applied Electromagnetics.

You might also like