Nuclear Engineering and Design 235 (2005) 14691480

Application of eddy current inversion technique to the sizing of

defects in Inconel welds with rough surfaces
Noritaka Yusa a, , Eiji Machida b , Ladislav Janousek a , Mihai Rebican a ,
Zhenmao Chen a , Kenzo Miya a
a International Institute of Universality, 2-7-17-7F Ikenohata, Taito-ku, Tokyo 110-0008, Japan
b Japan Atomic Power Company, Japan

Received 6 December 2003; received in revised form 24 November 2004; accepted 7 January 2005

Abstract

This paper evaluates the applicability of eddy current inversion techniques to the sizing of defects in Inconel welds with rough
surfaces. For this purpose, a plate Inconel weld specimen, which models the welding of a stub tube in a boiling water nuclear
reactor is fabricated, and artificial notches machined into the specimen. Eddy current inspections using six different eddy current
probes are conducted and efficiencies were evaluated for the six probes for weld inspection. It is revealed that if suitable probes
are applied, an Inconel weld does not cause large noise levels during eddy current inspections even though the surface of the
weld is rough. Finally, reconstruction of the notches is performed using eddy current signals measured using the uniform eddy
current probe that showed the best results among the six probes in this study. A simplified configuration is proposed in order
to consider the complicated configuration of the welded specimen in numerical simulations. While reconstructed profiles of the
notches are slightly larger than the true profiles, quite good agreements are obtained in spite of the simple approximation of
the configuration, which reveals that eddy current testing would be an efficient non-destructive testing method for the sizing of
defects in Inconel welds.
2005 Elsevier B.V. All rights reserved.

1. Introduction their design life. With increasing operating years, many

More than 30 years have passed since the first nu-
clear reactor began operation in Japan; thus, many
Japanese nuclear reactors are approaching the end of
Corresponding author. Tel.: +81 3 5814 5350;
fax: +81 3 3827 0682.
E-mail address: yusa@iiu.co.jp (N. Yusa).
nuclear reactor structural components have shown an
increased susceptibility to cracking (Horn et al., 1997).
Therefore, the role of non-destructive testing to verify
the structural integrity of nuclear reactor components
has significantly increased.
Recent experiences have revealed that Inconel welds
in the core internals are susceptible to cracking, and
thus require very reliable inspections. In general, visual

0029-5493/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.nucengdes.2005.01.005
1470 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480
inspection is used to locate cracks and subsequent
ultrasonic testing is used to estimate defect pro-
files for the inspections of these welds. However, it
has been pointed out that ultrasonic testing is not
effective for inspections of Inconel welds (Cheng
et al., 2003) because of the strong inhomogeneity and
anisotropy of Inconel welds. Since ultrasonic testing is
sensitive to grain structures of the target (Bouda et al.,
2003), Inconel welds significantly scatter ultrasonic
waves so that clear echoes due to any defects cannot
always be observed. This hampers not only the accu-
rate evaluation of defect profiles but also the detection
of defects when ultrasonic non-destructive testing is
employed.
For this reason, eddy current testing has recently
drawn considerable attention. Usually eddy current
testing is very sensitive to surface cracks, but not sen-
sitive to the anisotropy or microscopic structures of
a target. Thus, it is expected that small initial cracks
that cannot be detected with ultrasonic testing could
be detected with eddy current testing. In addition, it
is also expected that the profiles of such cracks may
be evaluated from the measured eddy current testing
signals. While eddy current testing signals themselves
do not provide direct information on defect profiles,
recent significant advances in physics-based computer
modeling have advanced eddy current testing as a non-
destructive testing method providing the capability to
size defects (Auld and Moulder, 1999). Recent publica-
tions have even reported the successful characterization
of natural stress corrosion cracks found in a steam gen-
erator tube of a nuclear power plant (Yusa et al., 2000,
2003a).
So far, eddy current testing has been mainly applied
to inspections of homogeneous materials with smooth
surfaces, such as steam generator tubes; because the
rough surfaces of welds can generate high levels of
noise in eddy current inspections. However, reports of
recent advances in eddy current studies have shown that
eddy current testing can be applicable to weld inspec-
tions. Quite a few publications document the develop-
ment of eddy current testing probes for the inspection of
welded surfaces, and some of these are now even being
marketed. Since Inconel welds usually do not exhibit
localized magnetism variability that causes high levels
of noise in eddy current inspections, it is expected that
eddy current testing will be effective in the detection
and sizing of cracks in Inconel welds.
On the basis of the discussion above, this study eval-
uates the applicability of eddy current testing to Inconel
weld inspection. Several publications have already pro-
posed methodologies for the non-destructive inspec-
tion of core internals (Light et al., 1996; Krzywosz,
1995). These methodologies consist of eddy current
tests that detect the presence of defects, and subse-
quent ultrasonic testing for evaluating the profile of
the detected defects. In contrast, this study evaluates
the ability of eddy current testing to accurately size
cracks.
2. Experimental verication of the applicability
of eddy current testing to weld inspection
2.1. Fabrication of specimen
In boiling water reactors, the core internals are
welded to the reactor pressure vessel using Inconel
welds, which include areas such as shroud supports,
control rod drive (CRD) stub tubes, and in-core
housings. Among these structural components, recent
experience has revealed that the enhancement of non-
destructive testing method for inspecting the CRD stub
tube welds is an important requirement. This study
involves the design and fabrication of a specimen that
models the CRD stub tube of a boiling water reactor.
The 1/2 scale model of a CRD stub tube and
the specimen fabricated in this study are shown in
Figs. 1 and 2, respectively. Numerical simulations had
been performed in advance of the fabrication of the
specimen. These simulations revealed that the curva-
ture of the surface does not significantly affect eddy
current signals. Therefore, this study modeled the weld-
ing of CRD stub tube as a welded plate specimen. The
materials of the specimen, both the base and welding
metal, are the same as those utilized in the actual stub
tubes: a low alloy steel, SFVQ1A, is welded with In-
conel 600 using Inconel 82 welding metal, after it is
buttered with Inconel 182. The surface of the Inconel
82 welding is quite rough as it was not polished after
TIG welding. Fig. 3 presents the cross-sectional profile
of the weld; Fig. 4 shows the surface condition of the
weld.
As cracks tend to occur perpendicular to the weld
line (METI, 2001), five electro-discharge machined
notches perpendicular to the weld line were machined
 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480 1471

between the Inconel 82 welding and the Inconel 600 is defined as the distance from the surface of the base
base material, as shown in Fig. 2. The notches are material.
rectangular; the dimensions of the notches are sum- Since the weld deposit is somewhat higher than
marized in Table 1. Note that the depth of the notches base material, the actual depth of the notches in

Fig. 1. BWR stub tube (1/2 scale model).

1472 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480

2.2. Eddy current probes and experimental setup

It is often pointed out that the conventional pan-

cake type probe is too sensitive to lift-off vibration, and
therefore it is not suitable to the inspection of welds
with a rough surface. To overcome this problem and
to make weld inspection using eddy current testing
possible, many eddy current testing probes have re-
cently been developed. However, a comparative study
of the efficiencies of these probes has not yet been fully
performed. Therefore, it is necessary to carry out ex-
perimental verification to choose a probe suitable for
the reconstruction of the notches considered in this
study.
This study performs eddy current inspections of the
specimen using the six eddy current testing probes il-
lustrated in Fig. 5. To fairly compare the probes under
basically the same conditions, all of the probes were de-
signed to have their best performance with low frequen-
cies in the range of 120 kHz. The following briefly
Fig. 2. Specimen (units are given in mm).
describes each of the probes.

Differential pancake probe: This probe consists of

the weld metal is deeper than the values shown in two pancake coils that work as both exciter and
Table 1. detector. The coils are connected differentially and
It should be emphasized that the height of the weld therefore the probe output is given as the difference
deposit is not constant along the weld line because of between the signals of two coils. In the experiments,
the lack of precision of the welding process. The weld this probe scans so that the plane that contains the
near Notch 1, which is the deepest notch among the axes of the two coils is parallel to the weld line
five notches fabricated, is much thicker than that near thereby reducing the noise signal due to the weld.
the other notches, and in addition, the welding height TR probe (mutual-induction differential type pan-
changes rapidly near Notch 1 as one observes in Fig. 2. cake probe): This probe consists of one exciter and
Since the weld configuration near Notch 1 is quite dif- two detectors connected differentially. The orienta-
ferent from that near the other notches, it was difficult tion of the probe during scanning is the same as that
to compare eddy current signals due to Notch 1 quan- of the differential pancake probe above.
titatively with those signals due to the other notches. Plus point type probe: This probe consists of two
For this reason, this study does not consider Notch 1 in rectangular coils perpendicular to each other. While
the inversion analysis described below. plus point type probe designs can be either self in-
duction differential type or mutual-induction type,
Table 1 this paper only presents the results of the self induc-
Dimension of machined notches tion differential type probe because the difference
Width Length Depth between the two designs was not significant.
Notch 1 0.9 10 10.0  probe: This probe, which is named after its top
Notch 2 0.9 10 1.0 view, consists of a pancake coil exciter and a rect-
Notch 3 0.9 10 0.5 angle detector oriented perpendicular to the exciter
Notch 4 0.9 10 3.0 (Hiroshima et al., 2003). While this probe can de-
Notch 5 0.9 10 6.0
tect notches of any orientation, this probe scanned
Units are given in mm. the notches so that the winding of detector coil was
 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480 1473

Fig. 3. Betweeen (a) notch 1 and 2, (b) notch 2 and 3, (c) notch 4 and 5. Cross-sectional profile of welded surface (across the weld line).
1474 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480
Fig. 4. (a) Top view, (b) side view. Surface condition of welding.
parallel to the notches because this arrangement pro-
vided the largest signals due to the notches.
Uniform eddy current probe: This probe consists
of a large exciter and a pancake pickup coil at-
tached to the bottom of the exciter (Koyama and
Hoshikawa, 1998). The exciter is situated so that in-
duced eddy currents flow uniformly below the probe,
which helps reduce lift-off effects.
Uniform eddy current probe with differential pick up
coils: This probe has the same exciter as the uniform
eddy current probe above, and has two detector coils
connected differentially. Because of the arrangement
of the detectors, this probe is expected to provide
smaller weld noise signals.
The experimental setup is illustrated in Fig. 6. A
probe is attached to an XYZ stage controlled with a
PC, and excited using an eddy current testing instru-
ment, aect2000s made by Aswan Ect Co., Ltd. The eddy
current testing instrument also measures the X and Y
components of the eddy current signals, and these sig-
nals are digitized and stored using the PC via an A/D
board.
2.3. Experimental results
Eddy current inspection was performed using the
probes and setup described in Section 2.2. All the
probes used the same low frequency of 10 kHz so that
induced eddy currents did not concentrate on the rough
surface of the specimen while reducing the weld noise.
The lift-off of the probes, which is defined as the dis-
tance from weld surface, was set to 1.0 mm.
Fig. 7 shows three-dimensional displays of the am-
plitude of the measured eddy current testing signals
due to Notch 5. The output signals of the eddy current
instrument are in volts and these are the units for all the
signals shown in the figure. None of the probes produce
output signals, as long as the probe did not detect any
discontinuity or anomalies. The signals displayed in the
figure are translated so that the minimum signal agrees
with the origin of the two-dimensional plane display-
ing the trajectory of the eddy current signals in order to
show the signals clearly in the three-dimensional dis-
play. The results shown in Fig. 7 reveal that the noise
signals due to the weld are small, and the notch signals
can be clearly observed in all cases. It is also shown
that there are two kinds of noise: one is noise due to
the height of weld bead, and the other noise is a result
of the weld surface roughness.
Among the six probes, the plus point type and uni-
form eddy current probes show large signals caused by
the weld bead. However, one can observe that these
weld bead signals are uniformly distributed along the
weld bead, and that the effect of the roughness of the
weld surface is very small. In contrast, the two pan-
cake type probes the differential pancake probe and
the TR probe how relatively large noise signals due
to the roughness of the weld surface, whereas they pro-
vide only a small noise due to the weld bead. The uni-
form eddy current probe with differential pickup coils
provides relatively small signals due to both weld bead
and the roughness of the weld surface. While the signals
measured with the  probe present a similar tendency
to those measured with the uniform eddy current probe
 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480 1475

Fig. 5. (a) Differential pancake probe, (b) TR probe, (c) plus point probe, (d)  probe, (e) uniform eddy current probe, (f) uniform eddy current
probe with differential pickup coils. The six eddy current probes utilized in this study. Upper image: side view, lower image: bottom view. Units
are given in mm.
1476 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480
Fig. 6. Experimental setup.
with differential pickup coils, the noise signal due to
the roughness of the weld surface is somewhat larger
than that of the uniform eddy current testing probe with
differential pickup coils.
3. Inversion of eddy current testing signals
Since the experimental results presented in the pre-
vious section have shown that eddy current testing does
not suffer from high levels of noise due to the Inconel
weld with a rough surface, this section attempts the re-
construction of the notches from the measured signals.
Note that weld bead signals, which are almost uniform
along the weld line, would be easily eliminated with the
use of a simple signal processing method and therefore
they do not essentially aggravate the detectability of
eddy current testing probes; whereas the signals asso-
ciated with noise due to the roughness of the weld sur-
face directly lead to a lower signal-to-noise ratio. Thus,
a probe with reduced noise due to the roughness of the
weld surface is preferable for the inversion performed
here.
To compare the probes quantitatively, signal-to-
noise ratios were calculated and the calculated values
are listed in Table 2. The signal-to-noise ratio, S/N, is
defined as
 characteristics. Note that the surface of the weld is
S |Vcrack |
max =  (1)
N max |Vweld |
Table 2
Signal-to-noise ratio
Probe S/N
Differential pancake probe 3.6
TR probe 2.0
Plus point probe 6.6
 Probe 3.8
Uniform eddy current probe 8.3
Uniform eddy current probe 14.3
With differential pickup coils
V  (x, y) = V (x, y) V (20, y) (2)

Vweld = V (20, y) (3)
where V(x, y) is the amplitude of measured eddy cur-
rent signals at (x, y) in Fig. 7. In order to consider only
the noise due to the roughness of the weld surface ac-
cording to the discussion above, the definition uses the
processed signals, V . The table reveals that the uniform
eddy current probe with differential pickup coils has the
highest signal-to-noise ratio. However, signals due to
the uniform eddy current probe with differential pickup
coils are not preferable from the viewpoint of signal in-
version. This is because the probe provides no signals
when it scans directly above the notches. Therefore
reconstructing notches from signals of a uniform eddy
current probe with differential pickup coils requires the
consideration of the two-dimensional spatial distribu-
tion of signals, which would make for a complex re-
construction. For this reason, the inversion performed
here uses eddy current testing signals measured with
the uniform eddy current testing probe that showed the
second highest signal-to-nose ratio. It should be noted
here that the definition of the signal-to-noise ratio tends
to overestimate the plus point probe because the plus
point probe provides only one signal peak whereas the
others provide two or four positive and negative peaks
with the same amplitude.
Since the inversion of eddy current signals is per-
formed on a computer, it is necessary to establish a
computational model in advance to perform the inverse
analyses. The exact configuration modeled is illustrated
in Fig. 8. The surface of the weld is somewhat higher
than that of the base materials, and the configuration
includes three materials with different electromagnetic
rough, as observed in Fig. 4. It is, however, unrealistic
to consider the exact configuration in numerical simula-
 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480 1477

Fig. 7. (a) Differential pancake probe, (b) TR probe, (c) plus point probe, (d)  probe, (e) uniform eddy current probe, (f) uniform eddy current
probe with differential pickup coils. Three-dimensional display of eddy current signals due to notch 5.

tions. Note that not only the exact surface condition but
also the distribution of electromagnetic characteristics
inside the weld is unknown at this moment. Therefore,
this study attempts to approximate the configuration as
a simple one, as shown in Fig. 9. The welded specimen
is modeled as a simple plate having a flat surface and
the uniform conductivity of Inconel 600, i.e., r = 1 and
= 1.0 MS/m. The height of the weld is approximated
as the change in the probe lift-off, as shown in Fig. 9.
In order to reconstruct the profiles of the notches,
this study adopts an inversion algorithm developed in
the authors previous study (Yusa et al., 2003b). The
algorithm is based upon the tabu search (Glover and
Laguna, 1997), and it modifies the profile of a defect
iteratively according to a given rule. While the algo-
rithm itself was developed to deal with stress corrosion
cracking and therefore it models a crack as a conductive
region, the reconstructions performed here are limited
to having the conductivity inside defects to be zero be-
cause this study only considers notches.
The defect model adopted is illustrated in Fig. 10.
The region where defects are assumed to be present is
divided into finite elements, and cracks are modeled as
an assembly of the elements. As this study considers
a single notch, the region consists of one layer of ele-
ments in its width direction and the conductivity within
1478 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480

Fig. 10. Defect model adopted in this study.

in this study are solved by finding the combination of

Fig. 8. Exact configuration.
the defect region is limited to zero. A defect is modeled
as an assembly of columnar regions, and parameterized
as a set of integers that indicate the depth of columns
using a finite element as an unit. For instance, the de-
fect shown in Fig. 10 is parameterized as {0 0 0 2 4 4 5
5 5 0 0 0}. Thus, the eddy current inversion problems
integers that minimizes the difference between mea-
sured eddy current signals and signals calculated from
a reconstructed defect profile.
Reconstructed profiles of the notches are depicted in
Fig. 11. Since the exciting current density in the mea-
surements, and the origin of the phase of the measured
signals were unknown, calibration of the measured sig-
nals was performed by using signals due to Notch 5.
As this study does not consider Notch 1, Notches 24
were reconstructed from measured eddy current sig-
nals. In the figure, hatched regions represent the re-
constructed profiles of the notches, and a small region

Fig. 9. Approximated configuration for numerical simulation.

 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480 1479

Fig. 11. (a) Reconstructed profile of notch 2 (true depth:

1.0 + 1 = 2 mm), (b) reconstructed profile of notch 3 (true depth:
0.5 + 1 = 1.5 mm), (c) reconstructed profile of notch 4 (true depth:
3.0 + 1 = 4 mm). Reconstructed profiles of the notches. Hatched re-
gion in the figure represents reconstructed profile.

formed by the grid is the unit element of the reconstruc-

tions.
It should be emphasized again the depth of the
notches, shown in Table 1, is given as the depth from
the surface of the base material, and the height of the
weld is approximately 1 mm. Therefore, the depths to
be reconstructed in Fig. 11(a)(c) are approximately
2, 1.5, and 4 mm, respectively. One can observe that
Notch 4 is accurtely reconstructed; whereas the other
two are reconstructed somewhat deeper and longer than
the true profiles; though the differences between the re-
constructed and true profiles are reasonable.
Fig. 12 shows a comparison between the measured
signals after the calibration and the signals from the re-
constructed profiles. The figure shows that the signals Fig. 12. (a) Notch 2, (b) notch 3, (c) notch 4. Comparison between
from the reconstructed profiles agree well with the mea- measured and reconstructed signals.
sured ones, in spite of the very simple approximation
of the configuration.
1480 N. Yusa et al. / Nuclear Engineering and Design 235 (2005) 14691480
4. Conclusion
This study has demonstrated that if suitable eddy
current testing probes and an efficient computational
strategy are adopted, eddy current testing would be ef-
fective in profile evaluation of notches in Inconel welds
with rough surfaces.
The profiles of the notches machined in the Inconel
weld specimen with a rough surface were reconstructed
from measured eddy current testing signals with the
aid of computer simulation. A simplified geometry was
used to model a specimen with a complicated configu-
ration for numerical simulations, and good agreements
between reconstructed and true profiles were obtained
in spite of the approximation of the configuration. It
would be reasonable to assume that it is not always
necessary for eddy current inversion problems to ex-
actly model the details of the configuration such as the
roughness of the weld surface, which is a great ad-
vantage for the practical application of eddy current
inversion.
This work utilized artificial notches to evaluate the
applicability of eddy current testing to weld inspection
because of the difficulty in the fabrication of artificial
stress corrosion cracking in thick Inconel welds. In re-
ality, however, Inconel welds are susceptible to stress
corrosion cracking and many publications have pointed
out that the characteristics of stress corrosion cracking
is very different from those of an artificial notch from
the viewpoint of eddy current inspections (Yusa, 2003).
Thus, this work needs to be expanded to address eddy
current inspections of stress corrosion cracking as fu-
ture work.
Acknowledgment
This study was supported in part by six Japanese
BWR utilities: Japan Atomic Power Company, Tohoku
Electric Power, Hokuriku Electric Power Company,
Tokyo Electric Power Company, Chubu Electric Power
Company, and Chugoku Electric Power Company. The
authors thank Hiroshi Iwami for technical advice on
the fabrication of the specimen. The stub tube model
whose picture is shown in Fig. 1 belongs to the Chubu
Electric Power Company.
References
Auld, B.A., Moulder, J.C., 1999. Review of advances in quantitative
eddy current nondestructive evaluation. J. Nondestruct. Eval. 18
(1), 336.
Bouda, A., Lebaili, S., Benchaala, A., 2003. Grain size influence on
ultrasonic velocities and attenuation. NDT&E Int. 36, 15.
Cheng, W., Shiwa, H., Yoneyama, H., Huang, H., Takagi, T., Uchi-
moto, T., 2003. Ultrasonic and eddy current testing of defects
in Inconel welding metals. In: Proceedings of the 12th MAGDA
Conference, Oita, Japan, pp. 187190.
Glover, F., Laguna, M., 1997. Tabu Search. Kluwer Academic Pub-
lishers.
Hiroshima, T., Fujimoto, T., Matsunaga, S., Kobayashi, Y., 2003. 
Probe and its application for maintenance inspection of welds.
In: Proceedings of the Symposium on NDT Techniques for Main-
tenance Inspection using Surface Methods and Quantitative Eval-
uation Techniques, Tokyo, Japan, pp. 2730.
Horn, R.M., Gordon, G.M., Ford, F.P., Cowan, R.L., 1997. Expe-
rience and assessment of stress corrosion cracking in L-grade
stainless steel BWR internals. Nucl. Eng. Design 174, 313325.
Koyama, K., Hoshikawa, H., 1998. Eddy current testing by uni-
form eddy current probe. In: Proceedings of the Second Asian
Joint Seminar on Applied Electromagnetics, May 1213, Tokyo,
Japan, pp. 16.
Krzywosz, K., 1995. Proceedings of the 13th International Confer-
ence on NDE in the Nuclear and Pressure Vessel Industries, Ky-
oto, Japan. Latest eddy current application in the nuclear indus-
try., 6165.
Light, G.M., Fisher, J.L., Tennis, R.F., Stolte, J.S., Hendrix, G.J.,
1996. Detection and sizing of defects in control rod drive mecha-
nism penetrations using eddy current and ultrasonics. J. Pressure
Vessel Technol. 118, 301307.
Japanese Ministry of Economy, Trade and Industry (Press release),
2001. Water leakage from penetration portion of control rod
derive housing into reactor pressure vessel of Unit-1. Novem-
ber 12, 2001 and available at http://www2.jnes.go.jp/atom-
db/en/trouble/individ/power/j/j01b121/.
Yusa, N., Chen, Z., Miya, K., 2000. Quantitative profile evaluation of
natural crack in a steam generator tube from eddy current signals.
Int. J. Appl. Electromagnet. Mech. 12, 139150.
Yusa, N., Chen, Z., Miya, K., Uchimoto, T., Takagi, T., 2003a. Large-
scale parallel computation for the reconstruction of natural stress
corrosion cracks from eddy current testing signals. NDT&E Int.
36, 449459.
Yusa, N., Chen, Z., Miya, K., 2003b. Rapid reconstruction of natural
stress corrosion cracks from eddy current NDE data. Key Eng.
Mater. 270-273, 593598.