S. Takahashi - Electromagnetic Nondestructive Evaluation (X) (Studies in Applied Electromagnetics and Mechanics) - IOS Press (2007)

ELECTROMAGNETIC NONDESTRUCTIVE
EVALUATION (X)
Studies in Applied Electromagnetics
and Mechanics
Series Editors: K. Miya, A.J. Moses, Y. Uchikawa, A. Bossavit, R. Collins, T. Honma,
G.A. Maugin, F.C. Moon, G. Rubinacci, H. Troger and S.-A. Zhou
Volume 28
Previously published in this series:
Vol. 27. A. Krawczyk, S. Wiak and X.M. Lopez-Fernandez (Eds.), Electromagnetic Fields in
Mechatronics, Electrical and Electronic Engineering
Vol. 26. G. Dobmann (Ed.), Electromagnetic Nondestructive Evaluation (VII)
Vol. 25. L. Udpa and N. Bowler (Eds.), Electromagnetic Nondestructive Evaluation (IX)
Vol. 24. T. Sollier, D. Prémel and D. Lesselier (Eds.), Electromagnetic Nondestructive
Evaluation (VIII)
Vol. 23. F. Kojima, T. Takagi, S.S. Udpa and J. Pávó (Eds.), Electromagnetic Nondestructive
Evaluation (VI)
Vol. 22. A. Krawczyk and S. Wiak (Eds.), Electromagnetic Fields in Electrical Engineering
Vol. 21. J. Pávó, G. Vértesy, T. Takagi and S.S. Udpa (Eds.), Electromagnetic Nondestructive
Evaluation (V)
Vol. 20. Z. Haznadar and Ž. Štih, Electromagnetic Fields, Waves and Numerical Methods
Vol. 19. J.S. Yang and G.A. Maugin (Eds.), Mechanics of Electromagnetic Materials and
Structures
Vol. 18. P. Di Barba and A. Savini (Eds.), Non-Linear Electromagnetic Systems
Vol. 17. S.S. Udpa, T. Takagi, J. Pávó and R. Albanese (Eds.), Electromagnetic
Nondestructive Evaluation (IV)
Vol. 16. H. Tsuboi and I. Vajda (Eds.), Applied Electromagnetics and Computational
Technology II
Vol. 15. D. Lesselier and A. Razek (Eds.), Electromagnetic Nondestructive Evaluation (III)
Vol. 14. R. Albanese, G. Rubinacci, T. Takagi and S.S. Udpa (Eds.), Electromagnetic
Nondestructive Evaluation (II)
Vol. 13. V. Kose and J. Sievert (Eds.), Non-Linear Electromagnetic Systems
Vol. 12. T. Takagi, J.R. Bowler and Y. Yoshida (Eds.), Electromagnetic Nondestructive
Evaluation
Vol. 11. H. Tsuboi and I. Sebestyen (Eds.), Applied Electromagnetics and Computational
Technology
Vol. 10. A.J. Moses and A. Basak (Eds.), Nonlinear Electromagnetic Systems
Vol. 9. T. Honma (Ed.), Advanced Computational Electromagnetics
Volumes 1–6 were published by Elsevier Science under the series title “Elsevier Studies in
Applied Electromagnetics in Materials”.
ISSN 1383-7281
Electromagnetic Nondestructive
Evaluation (X)
Edited by
Seiki Takahashi
Iwate University, Morioka, Iwate, Japan
and
Hiroaki Kikuchi
Iwate University, Morioka, Iwate, Japan
Amsterdam • Berlin • Oxford • Tokyo • Washington, DC

© 2007 The authors.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, without prior written permission from the publisher.
ISBN 978-1-58603-752-9
Library of Congress Control Number: 2007927275
Publisher
IOS Press
Nieuwe Hemweg 6B
1013 BG Amsterdam
Netherlands
fax: +31 20 687 0019
e-mail: order@iospress.nl
Distributor in the UK and Ireland Distributor in the USA and Canada

Gazelle Books Services Ltd. IOS Press, Inc.
White Cross Mills 4502 Rachael Manor Drive
Hightown Fairfax, VA 22032
Lancaster LA1 4XS USA
United Kingdom fax: +1 703 323 3668
fax: +44 1524 63232 e-mail: iosbooks@iospress.com
e-mail: sales@gazellebooks.co.uk
LEGAL NOTICE
The publisher is not responsible for the use which might be made of the following information.
PRINTED IN THE NETHERLANDS

Electromagnetic Nondestructive Evaluation (X) v
S. Takahashi and H. Kikuchi (Eds.)
IOS Press, 2007
© 2007 The authors. All rights reserved.
Preface
The 11th International Workshop on Electromagnetic Nondestructive Evaluation
(ENDE) was held at the Hotel APPI Grand in Hachimantai-shi, Iwate, Japan, on June
14th–16th, 2006. The workshop was organized by NDE & Science Research Center,
Faculty of Engineering, Iwate University, and financially supported and co-sponsored
by Japan Society of Maintenology, the Iron and Steel Institute of Japan and the Japa-
nese Society for Non-Destructive Inspection, and sponsored by the Japan Society of
Applied Electromagnetics and Mechanics, the Institute of Electrical Engineers of Japan
and the Magnetic Society of Japan.
Following welcoming remarks by Prof. Seiki Takahashi, workshop chair, there
were 46 presentations given at the workshop: 3 invited talks, 29 oral and 14 poster
presentations. Three invited talks were given by Dr. Masaaki Kurokawa, Mitsubishi
Heavy Industries Ltd., Prof. Anthony Moses, Cardiff University, and Dr. Gábor
Vértesy, Hungarian Academy of Sciences.
The workshop was organized into 9 oral sessions and 1 poster session: “ECT mod-
eling” chaired by D. Lesselier, A. Tamburrino, “ECT Modeling and Simulation”
chaired by G. Rubinacci, T. Takagi, “Eddy Current Testing” chaired by J. Pàvò,
T. Theodoulidis, “New Methods” chaired by K. Ara, S. Udpa, “Industrial Applications”,
Y. Tsuchida, T. Uchimoto, “NDE by Magnetism I” chaired by D.G. Park, G.Y. Tian,
“NDE by Magnetism II” chaired by P. Novotný, S. Takahashi, “Inverse Problem”
chaired by Z. Chen, L. Janousek, “Inverse Problem and Benchmark” chaired by
G. Berthiau, F. Kojima. The workshop was concluded with closing remarks by Prof.
Fumio Kojima.
The main theme of the ENDE workshop has been on “Eddy Current Testing” to
identify cracks in metals and alloys. Since the first ENDE workshop was held in Lon-
don 1995, the seeing of workshops have contributed the technical advance in ECT
through our competition and collaboration. ECT is put to practical use in industry now
as one of the approved methods of crack detection in steels and metallic structures.
We added the new topics in APPI meeting, i.e. magnetic NDE method according
to the concept of NDE & Science Research Center. Two of three invited talks are on
the magnetic NDE.
Forty-eight participants were registered for this workshop from different parts all
over the world; France, Italy, Hungary, China, Korea, UK, USA, Czech Republic,
Germany, Greece, Poland, Slovakia and Japan. Short versions of all contributed papers
have been published in the workshop abstracts, and 33 full papers were accepted after
review and are published in “Electromagnetic Nondestructive Evaluation (X)” pub-
lished by IOS Press in the series “Studies in Applied Electromagnetic and Mechanics.”
We could concentrate our interest on the academic exchange and the sincere dis-
cussion for three days, since APPI is located in an out-of-the way place. All the par-
ticipants enjoyed the beautiful green, fresh air, good taste water and the lovely concert.
The workshop organizers gratefully acknowledge the financial support of the
sponsors and would like to thank the session chairs and the participants. Thanks are
also due to the members of the standing committee. The editors are truly thankful to
vi
referees listed below, especially regional editors, Professors G. Rubinacci, T. Takagi

and S.S. Udpa.
The next ENDE meeting will be held from June 19th to 21st 2007 in Cardiff under
the auspices of Wolfson Center for Magnetics Technology, Cardiff School of Engi-
neering, Cardiff University.
Seiki Takahashi Hiroaki Kikuchi

Iwate University Iwate University
Morioka, Iwate, Japan Morioka, Iwate, Japan
Workshop Chair and Co-Editor Co-Editor
vii
List of Referees
K. Ara – Iwate University, Japan
K. Arunachalam – Michigan State University, USA
J. Bowler – Iowa State University, USA
N. Bowler – Iowa State University, USA
Z. Chen – Xi’an Jiaotong University, China
G. Dobmann – Fraunhofer-IZFP Institute, Germany
M. Hashimoto – Polytechnic University, Japan
H. Hashizume – Tohoku University, Japan
Y. Kamada – Iwate University, Japan
S. Kanemoto – Aizu University, Japan
H. Kikuchi – Iwate University, Japan
F. Kojima – Kobe University, Japan
K. Koyama – Nihon University, Japan
D. Lesselier – DRE-LSS CNRS-SUPÉLEC, Ecole Superieure d’ectricite, France
C. Lo – Iowa State University, USA
V. Melpaudi – Michigan State University, USA
O. Mihalache – Japan Atomic Energy Agency, Japan
A.J. Moses – Cardiff University, UK
N. Nair – Michigan State University, USA
J. Pávó – Budapest University of Technology and Economics, Hungary
G. Pichenot – CEA/DRT/LIST/DETECS/SYSSC/LSM, Centre CEA de Saclay, France
P. Ramuhalli – Michigan State University, USA
A. Razek – LGEP CNRS-SUPÉLEC, Ecole Superieure, France
G. Rubinacci – DIEL, Università degli Studi di Napoli Federico II, Italy
M.J. Sablik – Southwest Research Institute, USA
S.J. Song – Sung Kwan University, South Korea
T. Takagi – Tohoku University, Japan
S. Takahashi – Iwate University, Japan
A. Tamburrino – DAEIMI, Università degli Studi di Cassino, Italy
I. Tomáš – Institute of Physics, ASCR, Czech Republic
Y. Tsuchida – Oita University, Japan
T. Uchimoto – Tohoku University, Japan
L. Udpa – Michigan State University, USA
S.S. Udpa – Michigan State University, USA
G. Vértesy – Hungarian Academy of Sciences, Hungary
F. Villone – Università degli Studi di Cassino, Italy
N. Yusa – International Institute of Universality, Japan
Z. Zeng – Michigan State University, USA
viii
Editors
S. Takahashi – Iwate University
H. Kikuchi – Iwate University
Regional Editors
G. Rubinacci – Universita di Napoli Federico II, Italy
Standing Committee
G. Rubinacci – Universita di Napoli Federico II, Italy
R. Albanese – Universita Reggio Calabria, Italy
J. Bowler – Iowa State University, USA
N. Bowler – Iowa State University, USA
G. Dobmann – Fraunhofer Institute for NDT, Germany
R. Grimberg – National Institute of R&D for Technical Physics, Romania
H.K. Jung – Seoul National University, South Korea
F. Kojima – Kobe University, Japan
D. Lesselier – DRE-LSS CNRS-SUPELEC-UPS, France
V. Lunin – Moscow Power Engineering Institute, Russia
K. Miya – Keio University, Japan
G.Z. Ni – Zhejing University, China
J. Pávó – Budapest University of Technology and Economics, Hungary
A. Razek – LGEP CNRS-SUPELEC-UPS-UPMC, France
T. Sollier – CEA Paris, France
S. Takahashi – Iwate University, Japan
L. Udpa – Michigan State University, USA
V. Vengrinovich – Institute of Applied Physics, Belarus
Organizing Committee
S. Takahashi, Workshop Chairman – Iwate University
T. Takagi – Tohoku University
F. Kojima – Kobe University
Y. Ogura – Japan Probe Co., Ltd.
T. Imanaka – Iwate University
M. Kurokawa – Mitsubishi Heavy Industries, Ltd.
N. Sato – Asahi Kasei Engineering Co., Ltd.
K. Ara – Iwate University
Y. Kamada – Iwate University
H. Kikuchi, Secretariat – Iwate University
S. Kobayashi – Iwate University
ix
List of Participants
Mr. Kavoos Abbasi Dr. Ladislav Janousek
Department of QSE, Tohoku University, Department of Electromagnetic and
Japan Biomedical Engineering, Faculty of
Electrical Engineering, University of Zilina,
Dr. Katsuyuki Ara Slovakia
NDE & SRC, Faculty of Engineering,
Iwate University, Japan Dr. Yuichiro Kai
Faculty of Engineering, Oita University,
Dr. Hossein Bayani Japan
Department of Applied Science for
Electronics and Materials, Kyushu Dr. Yasuhiro Kamada
University, Japan NDE & SRC, Faculty of Engineering,
Iwate University, Japan
Dr. Gérard Berthiau
Research Institute in Electrotechnology Dr. Hiroaki Kikuchi
and Electronics of Nantes Atlantique, NDE & SRC, Faculty of Engineering,
France Iwate University, Japan
Dr. Pierre Calmon Dr. Satoru Kobayashi
CEA-LIST, France NDE & SRC, Faculty of Engineering,
Prof. Zhenmao Chen Iwate University, Japan
School of Aearospace, Xian Jiaotong
Prof. Fumio Kojima
University, China
Graduate School of Science and
Dr. Weiying Cheng Technology, Kobe University, Japan
NDE Center, Japan Power Engineering
and Inspection Corporation, Japan Dr. Masaaki Kurokawa
Mitsubishi Heavy Industries, Ltd., Japan
Prof. Masato Enokizono
Faculty of Engineering, Oita University, Dr. Ken Kurosaki
Japan Division of Sustainable Energy and
Environmental Engineering, Graduate
Dr. Szabolcs Gyimóthy School of Engineering, Osaka University,
Budapest University of Technology and Japan
Economics, Hungary
Prof. Jinyi Lee
Dr. Yusuke Imai Department. of Inf., Cntl and
National Institute of Advanced Inst./Chosun University, South Korea
Industrial Science and Technology,
Japan Dr. Dominique Lesselier
Laboratoire des Signaux et Systemes –
Dr. Satoshi Ito Departement de Recherche en
Department of QSE, Grad. School of Electromagnetisme, France
Eng., Tohoku University, Japan
x
Prof. Luming Li Prof. Seiki Takahashi

Aerospace School, Tsinghua University, NDE & SRC, Faculty of Engineering,
China Iwate University, Japan
Dr. Ovidiu Mihalache Prof. Tokuo Teramoto

Japan Atomic Energy Agency, Fast Graduate School of Systems &
Breeder Reactor Research and Information Engineering, University of
Development Center, Japan Tsukuba, Japan
Dr. Kenzo Miya Prof. Theodoros Theodoulidis

Japan Society of Maintenology, Japan Department of Electrical
Engineering/Technological Educational
Prof. Anthony Moses Institute of West Macedonia, Greece
Wolfson Centre for Magnetics
Technology, School of Engineering, Prof. Gui Yun Tian
Cardiff University, Wales, UK School of Electrical, Electronic and
Computer Engineering, Newcastle
Dr. Pavel Novotný University, UK
Institute of Chemical Technology
Prague, Czech Republic Dr. Haiyan Tian
Institute of Fluid Science, Tohoku
Dr. Toshihiro Ohtani
University, Japan
Ebara Research Co. LTD, Japan
tian@wert.ifs.tohoku.ac.jp
Dr. Duck-Gun Park
KAERI, South Korea Prof. Masaaki Tsushima
Iwate University, Japan
Dr. József Pávó
Budapest University of Technology and Prof. Hajime Tsuboi
Economics, Hungary Department of Information Engineering,
Fukuyama University, Japan
Dr. Stéphane Perrin
IIU, Japan Dr. Yuji Tsuchida
Faculty of Engineering, Oita University,
Dr. Gregoire Pichenot Japan
CEA, France
Prof. Antonello Tumburrino
Dr. Madalina Pirlog DAEIMI, Italy
Materials
Characterization/Fraunhofer-Institut Dr. Tetsuya Uchimoto
IZFP, Germany Institute of Fluid Science, Tohoku
University, Japan
Prof. Guglielmo Rubinacci
Dipartimento di Ingegneria Elettrica, Prof. Lalita Udpa
Università degli Studi di Napoli Department of Electrical and Computer
Federico II, Italy Engineering, Michigan State University,
USA
Prof. Toshiyuki Takagi
Institute of Fluid Science, Tohoku Prof. Satish Udpa
University, Japan Department of Electrical and Computer
xi
Engineering, Michigan State University, Environmental Engineering, Graduate

USA School of Engineering, Osaka University,
Japan
Dr. Gábor Vértesy
Hungarian Academy of Sciences, Mr. Tomoharu Yasutake
Research Institute for Technical Physics Department of Electrical and electronic
and Materials Science, Hungary Engineering, Graduate School of
Engineering, Oita University, Japan
Prof. Shinsuke Yamanaka
Division of Sustainable Energy and
This page intentionally left blank
xiii
Contents
Preface v
Seiki Takahashi and Hiroaki Kikuchi
List of Referees vii
Organization viii
List of Participants ix
Invited Talks
Inspection Experience of Steam Generator Tubes with Intelligent ECT Probe 3

Masaaki Kurokawa
Origin, Measurement and Application of the Barkhausen Effect in Magnetic Steel 4
Anthony J. Moses and David C. Jiles
A New Initiative: Universal Network for Magnetic Non-Destructive Evaluation 9
Gábor Vértesy
ECT Modeling and Simulation
Skin and Proximity Effects in ECNDT Sensors 13

Vincent Doirat, Gérard Berthiau, Javad Fouladgar and Anthony Lefevre
Eddy Current Modelling for Inspection of Riveted Structures in Aeronautics 25
S. Paillard, G. Pichenot, M. Lambert and H. Voillaume
Numerical Modeling of a Phase Sensitive Eddy Current Imaging System 33
Guglielmo Rubinacci, Antonello Tamburrino, Salvatore Ventre,
Pierre-Yves Joubert and Jean Pinassaud
Developments in Modelling Eddy Current Coil Interactions with a Right-Angled
Conductive Wedge 41
Theodoros Theodoulidis, Nikolaos Poulakis and John Bowler
Volumetric and Surface Flaw Models for the Computation of the EC T/R Probe
Signal Due to a Thin Opening Flaw 49
Léa Maurice, Denis Prémel, Jozsef Pàvò, Dominique Lesselier and
Alain Nicolas
Application of Eigenfunction Expansions to Eddy Current NDE: A Model of
Cup-Cored Probes 57
Hossein Bayani, Theodoros Theodoulidis and Ichiro Sasada
xiv
Eddy Current Testing and Technique
Experimental Extraction of Time-of-Flight from Eddy Current Test Data 67

Antonello Tamburrino, Naveen Nair, Satish Udpa and Lalita Udpa
A Probe Array for Fast Quantitative Eddy Current Imaging 75
Carmine Abbate, Maxim Morozov, Guglielmo Rubinacci,
Antonello Tamburrino and Salvatore Ventre
Superposition of Several Phase-Shifted Exciting Fields for Crack Evaluation 83
Ladislav Janousek, Noritaka Yusa and Kenzo Miya
Automatic Discrimination of Stress Corrosion and Fatigue Cracks Using Eddy
Current Testing 91
Stéphane Perrin, Noritaka Yusa and Kenzo Miya
Conductivity and Permeability Evaluation on Type IV Damage Investigation by
Electromagnetic Method 99
Haiyan Tian, Tetsuya Uchimoto, Toshiyuki Takagi and Yukio Takahashi
Industrial Applications and New Methods
Evaluation of Fatigue Loaded Conducting Structures Using Selected

Electromagnetic NDT Methods 109
Tomasz Chady, Ryszard Sikora, Grzegorz Psuj, Przemysław Łopato,
Masato Enokizono and Yuji Tsuchida
Evaluation of Circumferential Crack Location in Pipes by Electromagnetic
Waves 117
Kavoos Abbasi, Satoshi Ito, Hidetoshi Hashizume and Kazuhisa Yuki
Inspection of Cement Based Materials Using Microwaves 125
Kavitha Arunachalam, Vikram R. Melapudi, Lalita Udpa and
Satish S. Udpa
Defect Profiling Using Multi-Frequency Eddy Current Data from Steam
Generator Tubes 131
Kavitha Arunachalam, Oseghale Uduebho, Ameet Joshi, Shiva Arun Kumar,
Lalita Udpa, Pradeep Ramuhalli, Satish S. Udpa and James Benson
Electromagnetic Reading of Laser Scribed Logistic Markers on Metallic
Components 139
Szabolcs Gyimóthy, József Pávó, Imre Kiss, Antal Gasparics,
Zoltán Kalincsák, Imre Sebestyén, Gábor Vértesy, János Takács and
Hajime Tsuboi
Design of a Remote Field Eddy Current Probe Dedicated for Inspection of
a Magnetic Tube from Its Outer Surface 147
Tomas Marek, Daniela Gombarska, Ladislav Janousek, Klara Capova,
Noritaka Yusa and Kenzo Miya
xv
Application of Thermoelectric Power Measurement to Nondestructive Testing 153

Shinsuke Yamanaka, Yasuhiro Kawaguchi, Toshihiro Ohtani and
Ken Kurosaki
NDE by Magnetism and Magnetics
Accurate Detection of Material Degradation of Stainless Steel by ECT Sensor 163

Tokuo Teramoto
Micromagnetic Characterization of Thermal Degradation in Cu-Rich Alloys and
Results of Neutron-Irradiation 170
Madalina Pirlog, Iris Altpeter, Gerd Dobmann, Gerhard Hübschen,
Melanie Kopp and Klaus Szielasko
Challenges in Quantifying Barkhausen Noise in Electrical Steels 178
Anthony J. Moses, Harshad V. Patel and P.I. Williams
Magnetic Adaptive Testing: Influence of Experimental Conditions 186
Gábor Vértesy, Tetsuya Uchimoto, Toshiyuki Takagi and Ivan Tomáš
Magnetic and Acoustic Barkhausen Noise for the Characterisation of Tensile
Deformation and Stresses in Steel 193
Gui Yun Tian, John Wilson and Jiri Keprt
Electromagnetic Acoustic Resonance to Assess Creep Damage in a Martensitic
Stainless Steel 201
Toshihiro Ohtani
NDE Method Using Minor Hysteresis Loops in Ferromagnetic Materials 209
Seiki Takahashi, Satoru Kobayashi, Yasuhiro Kamada,
Hiroaki Kikuchi and Katsuyuki Ara
Investigation of Neutron Radiation Effects on Fe Model Alloys by Minor-Loop
Analysis 217
Satoru Kobayashi, Hiroaki Kikuchi, Seiki Takahashi,
Katsuyuki Ara and Yasuhiro Kamada
Numerical Analysis for Non-Destructive Evaluation of Hardening Steel Taking
into Account Measured Magnetic Properties Depending on Depth 225
Yuichiro Kai, Yuji Tsuchida and Masato Enokizono
Development of Metal Detection System for Reuse of Dismantled Wood
from Houses 233
Tomoharu Yasutake, Tomasz Chady, Yuji Tsuchida and Masato Enokizono
Inverse Problem and Benchmark
Nondestructive Evaluation for Material Degradation of Steel Sample Using

Minor Hysteresis Loop Observations 243
Fumio Kojima and Ryou Nishiyama
Identification of Defects from ECT Signals Using Linear Discriminant Function 251
Weiying Cheng, Shigeru Kanemoto and Ichiro Komura
xvi
Sizing of Volumetric Stress Corrosion Crack from Eddy Current Testing Signals
with Consideration of Crack Width 259
Zhenmao Chen, Noritaka Yusa and Kenzo Miya
Reconstruction of Fatigue Cracks Using Benchmark Eddy Currents Signals 267
Maxim Morozov, Guglielmo Rubinacci, Salvatore Ventre and
Fabio Villone
2D Axisymmetric ECT Simulation of the World Federation’s First Eddy Current
Benchmark Problem 275
Ovidiu Mihalache, Masashi Ueda and Takuya Yamashita
Author Index 283

Invited Talks
Electromagnetic Nondestructive Evaluation (X) 3
IOS Press, 2007
Inspection Experience of Steam Generator

Tubes with Intelligent ECT Probe
Masaaki KUROKAWA
Takasago R/D Center, Mitsubishi Heavy Industries, Ltd., Japan
An intelligent ECT probe was developed to perform a high speed and accurate
inspection of SG tubes. The probe consists of an ordinary bobbin probe and a newly
developed thin-film probe in which nine to twelve drive coils and pick-up coils are
arrayed along the circumferential outer surface of the probe to face the inner surface of
the SG tube. This multi-coil system can carry out high-performance flaw detection
without probe rotation.
Field tests were experienced in Japan, USA and Taiwan, and the intelligent ECT
technique was qualified by the Japanese regulatory authority in August ’03, and has
been adopted for actual inspection since December ’03 in 13 units in total. The total
number of inspected tube has reached about 120,000 in Japan. After field tests in USA
and Taiwan, the intelligent ECT technique obtained EPRI Appendix H qualification of
the residual damage mechanisms and was selected to perform all tube inspection with
full length in Taiwan.
(summarized by reviewer, K. Ara)

4 Electromagnetic Nondestructive Evaluation (X)
IOS Press, 2007
Origin, Measurement and Application of

the Barkhausen Effect in Magnetic Steel
Anthony J. MOSESa and David C. JILESa
a
Wolfson Centre for Magnetics, Cardiff University, Wales, UK
Introduction
The magnetic Barkhausen effect was discovered towards the beginning of the last
century [1] but its origin and interpretation is still open for debate. It is generally
accepted to be due mostly to microscopic discontinuities in domain wall motion due to
the presence of defects. Various methods of measurement have been developed
although quantification of results obtained using different sensor approaches is tenuous.
The phenomenon is very sensitive to internal stress and microstructure and the
direct correlations found between such parameters and Barkhausen emissions has made
it a useful tool in Non Destructive Evaluation (NDE). Today we have more powerful
data collection and analyzing systems which, combined with sensors capable of
accurate measurements, gives us the opportunity to exploit the effect in a broader field
of applications
This paper reviews the origin of the effect and challenges in its measurement and
interpretation. Applications of the Barkhausen effect, particularly in magnetic steels,
are then briefly discussed.
1. Origin of Barkhausen Noise
The Barkhausen noise (BN) is produced in a magnetic material while its magnetization
is being changed. There are several contributing factors to Barkhausen effect emissions
and these include: domain wall motion, domain rotation, domain nucleation and
annihilation. The effect is largely due to interaction of moving domain walls with
defects in the material. This action produces time varying irregularities in the shape and
instantaneous velocity of walls on a microscopic scale. The associated flux change
within the material in turn causes small electromotive forces (emfs) to be randomly
generated. This is usually schematically illustrated as discontinuities in the B-H curve
of the material [2]. In fact the Barkhausen emissions have both deterministic and
stochastic components [3], and it is the irreproducibility and randomness of the
stochastic component that causes the most difficulty in describing Barkhausen effect
using a theoretical model.
Early workers were able to estimate the magnetization changes due to single
Barkhausen jumps from the size of emf pulses produced during BN activity and
A.J. Moses and D.C. Jiles / Origin, Measurement and Application of the Barkhausen Effect 5
concluded reversals were occurring in volumes in the nanometer range [4] It is

interesting to note that Weiss [5] effectively predicted the existence of domains in 1907,
Barkhausen published his experimental results in 1918 but apart from indirect evidence
from emf.s induced in coils embracing magnetic wires and rods containing large
moving domain walls, the existence of domains was not confirmed until the powder
pattern work of workers such as Hámas and Thiessen [6] and by Bitter [7] in the early
1930’s.
The discontinuities were attributed to rotation of magnetization within a domain

but now it is accepted that the discontinuous domain boundary motion is the most
significant effect [3, 8]. Energy is dissipated at the time of a Barkhausen jump. It is
normally assumed that this event takes place on a time scale over which the applied
field does not vary significantly so it is independent of the field rate of change [9].
From this it might be assumed that the BN per cycle, although non-repeatable from
cycle to cycle, does not depend on the magnetization frequency, i.e. it is frequency in-
dependent.
In practice BN does vary with magnetizing frequency [10] simply because domain
activity, such as bowing, nucleation, annihilation, etc themselves vary with frequency.
Therefore the area of domain walls interacting with an individual dislocation varies
with frequency and hence the BN per cycle increases with frequency far more than
would be expected just from its stochastic nature. It is also possible that when clusters
of submicron domains are redistributed during the magnetization process they cause
BN like emissions which add to the static BN, effectively meaning the process
comprises a nano-scale frequency independent component and a micro scale frequency
dependent component
Some correlation has been found between BN measured in electrical steel

magnetized at 50 Hz and static hysteresis loss obtained by extrapolating curves of loss
per cycle against frequency to zero frequency. A linear increase in BN with hysteresis
loss of non-oriented steels endorses this finding and suggests the phenomena have
similar origins.
2. Measurement of Barkhausen Noise
An emf is induced in a search coil wound around a magnetized sample due to the
internal flux changes that constitute the BN. The effect is conveniently demonstrated
by connecting a loud-speaker via an amplifier to such a coil and listening to the
acoustic noise output as the magnetizing field is slowly changed. The emf is small so it
is a challenge not only to accurately measure it but also to ensure that what is measured
is due solely to BN. For example, because BN is stochastic in nature, it is important to
distinguish it from other forms of radiated or conducted noise in the measurement
system.
The BN has mostly been studied at very low frequency (at exciting field
frequencies of less than 5 Hz). However now there is more attention being placed on
higher frequency measurements because of its potential for indirect assessment of the
characteristics of electrical steels at power frequency. In either case the Barkhausen
signal is captured within a given frequency bandwidth typically around 3 kHz to 300
6 A.J. Moses and D.C. Jiles / Origin, Measurement and Application of the Barkhausen Effect
kHz to avoid high and low frequency noise errors. Some BN signal is undoubtedly lost
this way but it is not believed to be a significant factor in interpreting or practically
using the phenomenon since most BN activity occurs in this region.
When the enwrapping search coil method of measurement is used under A.C.
magnetizing conditions the BN emf must be separated from the normal Faraday emf
whose magnitude, at 50 Hz, is more than a thousand times higher than the average BN
signal. A convenient way of eliminating the Faraday emf is to connect two search coils
in series opposition so their output voltage ideally is just due to BN. The signals
induced in the coils are very similar when the coils are placed very near to each other
whereas a progressive decrease of their correlation occurs when the coils are moved
apart [11].
Recent improvements in digital technology, particularly with data acquisition cards,

make the AC measurements accurate and more convenient to process and analyze.
Questions over the necessary specification of magnetization waveform quality and data
sampling rate still need to be considered. Typically AC BN signals are captured at rates
of 100-200 kHz and there is no advantage in using higher values since under constant
magnetizing conditions the measured BN becomes constant at higher frequencies [12].
The IEC standard for single sheet testing of electrical steel stipulates that the form
factor of the secondary induced voltage should be maintained to within ±1 % of 1.11 to
ensure sufficiently accurate magnetic measurement. However, for Barkhausen noise
measurements this is not sufficient since even this small deviation from a perfect
controlled sinusoidal flux density is sufficient to cause poor repeatability of the
measured BN. Incorporating digital feedback into the measurement system to reduce
the form factor variance to within ±0.01 % of 1.11 results in a significant improvement
for Barkhausen analysis.
There are several ways in which the BN signal can be analyzed. Today data
acquisition systems can be used to analyze a signal in terms of an average value, its
RMS value, a power spectrum, summation of BN peaks, etc. It is impossible to relate
any of these quantities directly to the number of BN events occurring in any particular
region of a test sample but they do all normally follow similar trends as magnetizing
field or frequency is changed or when one sample is compared with another. Other
methods which have been found to be useful in interpreting Barkhausen noise data
include the mean, standard deviation, pulse height distribution, and FFT [13].
An alternative method for Barkhausen measurement is to use a wound ferrite core

whose detection coil axis is placed perpendicular to the surface of the sample [14]. The
ferrite core essentially acts as a magnetic amplifier of the Barkhausen signal induced in
the coil. A localized BN measurement can be made or the complete surface of a steel
sample can be scanned. Other high initial permeability materials such as amorphous
alloys have been successfully demonstrated as the probe core [15]. The magnitude of
the detected BN signal depends on the core dimensions and permeability as well as the
number of coil turns but it is interesting that the variation of BN obtained using probes
follows similar treads to those found using enwrapping search coils in the same region
of a specimen under the same magnetizing conditions.
A.J. Moses and D.C. Jiles / Origin, Measurement and Application of the Barkhausen Effect 7
Other methods reported for measuring Barkhausen noise include the use of a
commercial magnetic head sensor [16] and a commercial system supplied by Stresstech
Inc, the Rollscan 200-1 [17].
3. Applications
The BN effect has been used in many NDE applications already [18] and competes
strongly with ultrasound and eddy current testing in many cases. Its sensitivity to
microstructure including dislocations, and grain boundaries are the basis for its use in
internal stress and hardness measurement. It’s potential for use in assessment of in-
depth properties such as in case hardening has been demonstrated widely.
There are many opportunities for fully exploiting BN measurements and analysis
as a competitive NDT tool. However, more research is necessary to assess its full
potential in studies of magnetic steels under AC magnetization. It is a strong candidate
for use in measurement of magnetic properties such as power loss and physical
properties such as localized regions of irregular grain structures or mechanical hardness
of such materials where other techniques are difficult to apply or do not produce
sufficient information.
4. Conclusions
In conclusion, although there are many factors which contribute to BN it has been
found to correlate closely with mechanical stress and microstructure. The various
sensing techniques used to detect BN picks up different features of the phenomena. BN
measurement at power frequency in electrical steel may help in obtaining a better
understanding of loss processes and surface imperfections which affect the magnetic
properties.
The BN signal is rich in information and undoubtedly reflects many aspects of the
magnetizing process [9]. Better knowledge of its underlying features may lead to
greater understanding of effects of material parameters on magnetizing processes which
in turn may lead to improved magnetic materials as well as identifying further
applications of BN in NDE.
References
[1] H.Barkhausen, Two phenomena uncovered with help of the new amplifiers Phys Z., 20, (1919), 401-3
[2] D.Jiles, Introduction to magnetism and magnetic materials, Second edition, Chapman and Hall, London,
1998.
[3] D.C.Jiles, Dynamics of domain magnetization and the Barkhausen effect, Czechoslovak Journal of
Physics, 50, 893, 2000.
[4] E.P.T.Tyndall, The Barkhausen effect, Phys. Rev., 24, (1924) 439-51
[5] P.Weiss, Hypothesis of the molecular field and ferromagnetic properties, J. Phys., [4], 6, (1907), 661-90
[6] L.Hámos and P.A. Thiessen, Making visible the regions of different magnetic states in solid bodies, Z.
Physik, 71, (1931), 442-4
8 A.J. Moses and D.C. Jiles / Origin, Measurement and Application of the Barkhausen Effect
[7] F.Bitter, On homogeneities in the magnetization of ferromagnetic materials, Phys. Rev., 38, (1931),
1903-5
[8] K.Schroeder and J.C.McClure, The Barkhausen effect, CRC Critical Reviews of Solid State Science, 6,
45, (1976), 45
[9] G.Bertotti, Hysteresis in magnetism, Academic Press, San Diego, 1998
[10] A.J.Moses, F.J.G.Landgraf, K.Hartmann and T.Yonamine, Correlation between angular dependence of
A.C. Barkhausen noise and hysteresis loss in non-oriented electrical steel, Stahleisen, (2004), 215-9
[11] E.Puttin, M.Zani, and A Ventura, A double coil apparatus for Barkhausen noise measurement, Rev. Sc.
Instrum.,72, (4), (2001)
[12] B.Zhu, M.J.Johnson, C.H.Lo and D.C.Jiles, Multifunctional magnetic Barkhausen emission
measurement system, IEEE Trans. Magn., 37, (3), (2001), 1095-1099
[13] H.V.Patel, A.J.Moses and P.I.Williams, The dependence of AC Barkhausen noise measurement on data
acquisition parameters Proc. of 9th Int. Workshop on 1 and 2 Dimensional Magnetic Measurements and
Testing Czestochowa ,Poland, (2006), 68-69
[14] D.M.Stewart, K.J.Steven, and A.B.Kaisser, Magnetic Barkhausen noise analysis of stress in steel,
Current Applied Phys., 4, (2004), 308-311
[15] J.Pal’a, J.Bydzovsky, and P.Svec, Influence of magnetising frequency and construction of pick-up coil
on Barkhausen noise, J. of Electrical Eng., 55, No 10/S, (2004), 38-40
[16] A.J.Perez-Benitez, L.R. Padovese, J.Capo-Sanchez, and J.Anglada-Rivera, J Investigation of the
magnetic Barkhausen noise using elementary signals parameters in 1000 commercial steel, J. Mag.
Magn. Mater. 263 (2003) 72-77
[17] M.Lindgren and T.Lepisto, Effect of cyclic deformation on Barkhausen noise in a mild steel, NDT&E
International 36 (2003) 401-409.
[18] D.C. Jiles, Review of magnetic methods for nondestructive evaluation, NDT International, 21, 311,
1988.
IOS Press, 2007
A new initiative: Universal Network for

Magnetic Non-destructive Evaluation
Gábor VÉRTESY
Research Institute for Technical Physics and Materials Science,
Hungarian Academy of Sciences
H-1525 Budapest, P.O.B. 49, Hungary
Objective
Magnetic measurements are suitable for characterization of changes in structure of

ferromagnetic materials, because their magnetization processes are closely related to
the microstructure of the materials. This fact also makes magnetic measurements an
evident candidate for non-destructive testing, for detection and characterization of any
modification and/or defects in materials and manufactured products made of such
materials. Structural non-magnetic properties of ferromagnetic materials have been
non-destructively tested by several magnetic methods for a long time with fair success.
However, the application of magnetic methods in everyday nondestructive inspection
practice is not satisfactory. Because if this an informal network of many research
workers in many countries all over the World, interested in this area, has been
organized on Prof. Seiki Takahashi’s (Iwate University) initiative
(http://www.ndesrc.eng.iwate-u.ac.jp/UniversalNetwork/).
Main goal of the Universal Network is to concentrate research power in this
important area, to improve efficiency of the information exchange, to prove
applicability of magnetic ND methods, to develop new methods, to investigate
theoretically the observed phenomena, to find new application possibilities, to fasten
cooperation with industrial companies, to organize workshops for experts in this area
and to organize new projects for introduction of magnetic methods into industrial
application.
The main research directions are to clarify and quantify the relationship between
microstructure and magnetism in materials and to investigate and develop in-situ
magnetic inspection techniques for quantitative nondestructive evaluation of
components and structures including: i) Methods for evaluating performance-related
properties of materials from their structure-sensitive magnetic properties, ii) New
techniques and instrumentation for evaluation of material condition using magnetic
properties and iii) Models for description of magnetic properties and their dependence
on structure.
NDE of steel degradation before any crack initiation would be one of the targets in
Universal Network concerning nuclear plants, thermal electric plants, chemical plants,
mass transportation, bridges and gas pipelines.
10 G. Vértesy / A New Initiative: Universal Network for Magnetic Non-Destructive Evaluation
Methods
A large number and quite different methods exist in this area. They are the following:
Magnetic hysteresis loop measurements (including classical methods and the recently
developed minor-loop analyzing method (MAM) and Magnetic Adaptive Testing
(MAT)), Barkhausen noise measurement, magnetic acoustic emission (MAE),
micromagnetic, multiparameter, microstucture and stress analysis (3MA), magnetic
flux leakage measurements, combination of conventional eddy current technique with
magnetic field measurement, magnetooptical methods, magnetostrictive delay line
technique and classic low frequency ac magnetometry.
Planned Projects
Pressure Vessel
The age degradation in pressure vessels is one of the most important and urgent
problems in the world. No NDE methods exist for the ductile-brittle transition.
Magnetic method seems to be an effective and good solution for this problem.
However only few magnetic data, concerned with neutron irradiation, exist. Within this
project the facilities of Halden reactor are going to be used. Measurements before and
after neutron irradiation are planned to be performed. Among magnetic measurements
Barkhausen noise, hysteresis loops measurement (MAT, MAM), magnetic acoustic
emmission, microstucture and stress analysis will be performed, together with the
investigation of mechanical properties (Vickers hardness, tensile deformation, Charpy
impact test). Defects (precipitates, dislocation loops, vacancies, interstitial, void
(neutron irradiation damages) will also be studied.
Degradation of Gas Pipelines, Railways, Bridges and Other Steel Constructions

The aim and methods of this project is similar to the previous one, but its target is to
study age degradation in low carbon steels by magnetic NDE and to observe defects
(dislocations and micro-cracks induced by metal fatigue test and tensile deformation)
by magnetic methods. TEM, SEM, FMM and other observations will also be performed.
Standardization of Magnetic Properties

The aim of the project is: standardization of magnetic properties connected with
degradation in steels. Fundamental study will be performed on the relationships
between magnetic properties and defects. Urgency of this project is motivated by the
fact that only few models exist, which are proved theoretically as well as
experimentally, and frequently experimental data depend on the investigators. It means
that we need a standardized data base of the magnetic properties. We want to compose
the quantitative relationships between them depending on the standard data and give
the physical meaning to the relationships.
The project has already started, three series of round robin samples have been
prepared, which are now being circulated among the participating laboratories. The
same series of samples (prepared and distributed according an agreed time-table and
order) were, are or will be measured by each lab within the network. Each member
performs his own measurements by use of the same samples and all the measuring
results will be compared.
ECT Modeling and Simulation
IOS Press, 2007
Skin and Proximity Effects in ECNDT

Sensors
DOIRAT Vincent a , BERTHIAU Gérard a,1 , FOULADGAR Javad a ,
LEFEVRE Anthony a
a
37 Bd de l’université, BP 406, 44602 Saint Nazaire, France
Abstract. A semi-analytical forward model based on the electromagnetic coupled

circuits method is presented. This model allows to simulate the impedance varia-
tion of a ferrite-cored sensor above a controlled material. The computational speed
is increased by determining the current density in coil using the Kelvin analytical
distribution. This forward model provides a good accuracy with respect to experi-
mental measurements.
1. Main Objectives
ECNDT is generally used in such frequency domain that intrinsic skin and proximity
effects of the sensor are rather important (freq. between 100Hz and some MHz). Most
of well-used models do not consider these effects which can change considerably the
impedance of the sensor [1,2]. Using Finite Element Method (FEM) in nodal formula-
tion does not well evaluate the potential gap due to the ferrite core of the sensor. In edge
element formulation, there is an increased number of unknowns and it is heavier to pro-
ceed. Semi-analytical Dodd & Deeds’ formulation [3] is fast and accurate but does not
take into account neither the skin and proximity effects nor the presence of ferritic core.
In this paper, we propose therefore a semi-analytical forward model which takes into
account these phenomena. Our axisymmetrical model is based on the Coupled Circuits
Method (CCM) [4]-[7] which allows to determine the sensor impedance. As this model
is fast-running, we can use it in an inversion process in order to calculate the lift-off
distance between the probe and the controlled plate and/or the material conductivity. For
the probe discretization, we propose a formulation which largely reduces the number of
unknowns without accuracy degradation.
2. Model Description
The CCM is used to solve axisymmetric configurations. The sensor is constituted of a

ferrite core and a bobbin with N w wires. The bobbin is supplied by a sinusoidal current
with constant amplitude (I bob ) and pulsation (ω). That sensor is above a metallic plate
(μ = μ0 ,σ). The CCM consists in associating the integral form of the solution to a sub-
1 Corresponding Author: Gérard Berthiau, 37 Bd de l’université, BP 406, 44602 Saint Nazaire, France; E-
mail: gerard.berthiau@univ-nantes.fr
14 V. Doirat et al. / Skin and Proximity Effects in ECNDT Sensors
division in elementary coils. The unknowns are the current densities in the different parts
of the system (Q ⇒Bobbin, K ⇒Ferrite and P ⇒Plate). To determine these unknowns,
material (Bobbin+Ferrite+Load) are discretized in elementary loops with constant cur-
rent densities [6]. The ferrite core is discretized only on the surface Fig. 1)[7,8]. Each
Axis of symmetry
Ferrite (μ) Bobbin (σ)
Only sector discretization
Layer and sector discretization
Lift-off Load (σ, μ)
Figure 1. System discretization.
Axis of symmetry
b M
i1 R1
i2
Sb
U1
R2 L2 L1
a c
Discretization 2 Discretization 1
Sa
Figure 2. Equivalent circuit for two discretizations.
Ibob Bobbin
Wire 1 Wire 2 Wire Nw

RB1,1 LB1,1 RB2,1 LB2,1 RBNw ,1 LBNw ,1
Ubob Q1,1 Q2,1 QNw ,1
RB1,2 LB1,2 RB2,2 LB2,2 RBNw ,2 LBNw ,2

Q1,2 Q2,2 QNw ,2
MB12 ,B1Ndw
RB1,Ndw LB1,Ndw RB2,Ndw LB2,Ndw RBNw ,Ndw LBNw ,Ndw
Q1,Ndw Q2,Ndw QNw ,Ndw
MB2Ndw ,BNw Ndw
U1 U2 UNw
MB1Ndw ,L1
RL1 LL1 LF1
P1 K1 • Nw : Number of wires in a bobbin
• Ndw : Number of discretizations in a wire
RL2 LL2 LF2 •
P2 K2 Ndl : Number of discretizations in the load
• Ndf : Number of discretizations in the ferrite
• Qi,j : Constant current density in the (i, j)th Bobbin discretization
• Ki : Constant current density in the (i)th Ferrite discretization
RLNdc • Pi : Constant current density in the (i)th Load discretization
LLNdc LFNdf
PNdc KN df
• L : Self inductance of a discretization
• R : Resistance of a discretization
Load Ferrite • M : Mutual inductance between two discretizations
Figure 3. Equivalent electrical scheme.

V. Doirat et al. / Skin and Proximity Effects in ECNDT Sensors 15
elementary loop is in magnetic interaction with itself and with the other ones. The inter-
action between two loops can be explained with the electric transformer model (Fig. 2).
The resistances (R1 , R2 ), self inductance (L 1 , L2 ) and mutual inductance (M ) depend on
the geometrical parameters (a, b, S a , Sb , c) and physical characteristics (σ, μ) of the dis-
cretization. Our discretization system is an extension of this electric transformer model.
So, the coils inductive components are computed according to the equivalent electrical
scheme (Fig. 3)– all the mutual inductances are not represented to increase clarity. The
magnetic component of the bobbin is constituted by the summation of the self induc-
tances and all the mutual inductances for each discretization element. For the bobbin’s
resistance, we determine the active power in the complete system in order to get a better
accuracy than with the Ohm’s law.
2.1. Skin Effect in a Cylindrical Wire
We aim at determining the current density distribution Q(r) in the cylindrical wire, with
external radius R, depending on the distance from the axis r :
r
1
H= Q(x)x dx (1)
r 0
Differentiating (1) with respect to the radius r and using the Maxwell-Faraday’s law, we
obtain :
d2 Q 1 dQ
+ − jωσμQ = 0 (2)
dr2 r r.
where ω is the angular frequency, μ the magnetic permeability and σ the electrical con-
ductivity. With k 2 = ωσμ, (2) can be written as a differential equation (3), the general
solution of which is a linear combination of zero-order Bessel function (4):
d2 Q 1 dQ
+ − jk 2 Q = 0 (3)
dr2 r r.
Q = AJ0 (k r j 3/2 ) + BK0 (k r j 1/2 ) (4)
where J0 is the zero order Bessel function and K 0 the zero order modified Bessel func-
tion. Considering the boundary conditions and the phase origin at the conductor surface
(R : external radius), (4) becomes the Kelvin distribution [9,10]:
M0 (k r) j θ0 (k r)
Q(r) = Q0 e (5)
M0 (k R)
where M0 (X) is the modulus of J 0 (X j 3/2 ), θ0 (X) the angle and Q 0 the current density
M0 (k r)
on the surface of the wire. Q 0 M 0 (k R)
represents the current density magnitude in the
conductor.
The Kelvin distribution is used in this work to reduce the discretization number of
the bobbin (section 3).
Fig. 4 and Fig. 5 present the ratio of inner current density to surface current density
r
versus the normalized radius ( R ). The higher are the frequencies, the more important is
1 1
25 kHz 25 kHz 50 kHz
0.9 0.9
100kHz
Normalized current density |Q(r)/Q |

Normlized current density |Q(r)/Q0|
0
0.8 0.8
50 kHz
0.7 0.7
200 kHz
0.6 0.6
0.5 0.5 300 kHz

100 kHz
0.4 0.4 400 kHz
0.3 0.3 500 kHz

200 kHz 0.2
0.2
300 kHz 400 kHz
0.1 0.1
500 kHz
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Normalized Radius r/R Normalized Radius r\R
Figure 4. Outer Radius R = 0,5mm. Figure 5. Outer Radius R = 0,25mm.
the skin effect. Consequently, as ECNDT works on high frequencies (some 100kHz), the
sensor impedance evaluation must take into account the skin effect because it generates
an increasing variation of the conductor resistance.
2.2. Proximity Effect in a Multi Coil Sensor
The second main effect to be taken into account is the proximity effect of the coil wires.
First of all, we consider a circular sensor with N w elementary coils and without ferritic
core. Each coil is discretized in layers and sectors (N dw discretizations) in order to deter-
mine the current density distribution using the CCM. So, the total number of unknowns
is Nw × Ndw .
With respect to Fig. 6, the magnetic potential vector A ϕ (which only has the az-
imuthal component) generated in P by the current I bob which circulates around C is:

μ0 Ibob a k12
Aϕ = 1− L1 (k1 ) − L2 (k1 ) (6)
k1 π b 2

4ab
where k1 = (a+b) 2 +c2 , L1 , L2 are Legendre elliptical integrals.
• The mutual and self inductancies are computed with magnetic potential vector (6):

Φ section
BdS C
Adl
M= = = (7)
Ibob Ibob Ibob
However, the unknown in the CCM presented in this paper is the current density. So we
used the next inductance formulation where the "surface" of the discretization element
(Sd ) is used:
√
2μ0 ab k12
Ms = 1− L1 (k1 ) − L2 (k1 ) Sd (8)
k1 2
If both layer and sector discretization are used, the current in the surface can be supposed
constant. In this case, S d is the real surface of the element. This method however, leads to
a great number of unknowns. To reduce the calculation time, one can suppose that radial
variation of current is defined by Kelvin function (5). In this case, the discretization is
done only in sectors. The current is supposed to be constant on the circle’s arc and the
z dA
C P(b,0,c)
y
c
dl
a x
0 a b H
C
Ibob dl
Figure 6. Circular coil inducing the magnetic potential vector dA in P.
inductance M s is calculated by integrating (7) with respect to r. The "surface" S d in (8)

is then given by :
3
2πR J1 (k R j 2 )
Sd = 3 3 (9)
ns kj 2 J0 (k R j 2 )
where ns is the number of sectors. In order to avoid numerical singularities in the inte-
gration computing, the Gauss X-point integration formula is used, taking care that the
Gauss points do not coincide with the discretization barycenter points.
• The "resistance" for each discretization is computed with:
2πr
Rs = (10)
σ
where r is the distance between the barycenter of the discretization and the revolution
axis (a in Fig 2 for example).
Finally, the "impedance" full square matrix (Z) is built using (10) and (8). Then, in
each coil the current is:

N dw
Ibob = Sdi,k Qi,k (11)

k=1
where Ndw is the number of discretization by coil, Q i,k the unknown current density in
the ith coil and the k th element with the surface S di,k .
The surface matrix S Nw ×(Nw .Ndw ) is built from (11) :

Sd1,1 . . . Sd1,Ndw 0 . . . . . . 0
S= (12)
0 . . . . . . 0 SdNw ,1 . . . SdNw ,Ndw
Considering the voltage constant between all discrete elements of the same coil
(eg. coil i), the difference of potential between the two elements (eg. element 1 and
k ∈ (2, Ndw )) is null (13) and we finally get the following system (14):

Nw N dw
(Rsi,1 + j ωMsi,1 )Qi,1 − Rsi,k Qi,k − j ωMsp,q Qp,q = 0 (13)

p=1 q=1
T T T
[[S] [D]] [Q1,1 · · · QNw ,Ndw ] = [I · · · I 0 · · · 0] (14)
Then, this system is solved to obtain the current densities for each sector of each coil.
Fig. 3 shows the repartition of the current density in the coils of the sensor. The skin
effect is visible and, the proximity of the conductors also has an influence on the current
repartition as far as the neighbors are concerned.
3. Main Interest
The CCM needs the inversion of a quasi full square matrix the dimensions of which
are linked to the discretization number. So, it is important to reduce this number while
keeping the accuracy as best as possible. To do this, the coils are only discretized in
sectors where the unknown current density is on the conductor’s surface. The current
densities inside the conductors are then calculated with the Kelvin distribution (5).
By introducing the Kelvin distribution rather than discretizing the wire in layers, we
obtain a remarkable reduction of the discretization number and then the computational
time is divided by a factor 36 for similar accuracies (Fig.3, Tab.1).
The measurements are obtained with an "Agilent 4294" impedance analyser. Simu-
lations are obtained with a Matlab R
program on a computer (Pentium 4, CPU 2.8GHz,
RAM 1G).
Altitude [mm]
×105 Altitude [mm]
4 ×105
4 4
4
3.5 3.5
3.5 3.5
3
3 3
3
2.5
2.5
2.5
2 2.5
2
2 1.5 2 1.5
1 1
1.5 1.5
0.5 0.5
1 1
9 9.5 10 10.5 11 11.5 12 9 9.5 10 10.5 11 11.5 12
Radius [mm] Radius [mm]
(a) Layer and sector discretization. (b) Only sector discretization.

Figure 7. Distribution of current density in the bobbin: Skin and proximity effects.
Table 1. Reduce discretization results and comparison with the experiment.

Nb discretization Resistance (Ω) Reactance (Ω) Inductance (μH) time(s)
Layer and sector 1800 1,3 40,97 13,04 170
Only sector 300 1.22 40,99 13,05 4.7
Measures × 1.34 41 13.05 ×
3.1. Introduction of a Ferritic Core
Most of EC sensors have a ferritic core to focus the magnetic flux, so it is important to
be able to take it into account in a model. We considered the magnetic permeability of
the ferritic core as linear, homogeneous and isotropic and its electrical conductivity as
null (no EC loss). The core surface is discretized and N df fictitious surface currents K’
are considered [7,8]. In this part, only the system "bobbin+ferritic core" is taken into
account. The characteristic magnetic equation of the system can be written as:

1 μr + 1
(μ0 K’(M )) + n × dB’ = −n × B0s (15)
2 μr − 1 (l)
which corresponds to a second kind Fredholm equation in K’. The magnetic induction
B0s on the ferritic core surface is generated by the sensor coils (sensor discretization in
sectors) and B’ by the ferritic core itself (fictitious current K’) as shown on Fig. 8. This
gives the following discrete relation:

1 μr + 1 Ndf

N dw
(μ0 K’(M )) + n × dB’ =− n × dB0 (16)

2 μr − 1 (lk ) (lk )
k=1 k=1
For this determination, we used the expression of the magnetic potential vector (6) for
each coil given by:
μ0 k1 c
dB’r = √ [−J1 (k1 ) + C J2 (k1 )]K’(M ) dl (17)
4π b a b
μ0 k1
dB’z = √ [J1 (k1 ) + D J2 (k1 )]K’(M ) dl (18)
4π a b
2 2 2 2 2 2
a +b +c a −b −c
with C = (a−b) 2 +c2 and D = (a−b) 2 +c2
Finally, we have to solve the following matrix system representing the set "ferritic core
+ bobbin" :
T
[[[S] [0]] [D] [F ]]T Q1,1 · · · QNw ,Ndw K1 · · · KN

df
= [I · · · I 0 · · · 0]T (19)
Axis of symmetry
Ferritic core Inductor
n
K(M)
dB0(M) c
J0

dB (M)
a
b
Figure 8. Inductor + Ferritic core contributions to magnetic induction.

3.2. Addition of a Load
A conducting plate, which can be constituted by several layers (i) (σ(i), μ(i)), is set
under the sensor. The load length is long enough in order to have no boundary effect. The
discretization mesh is refined in the interesting zones under the sensor. Thus, the load is
considered as a set of rectangular section circular coils where the voltage is constant in
each discretization part.
As previously, the matrix system (19) is completed by addition of the impedance
matrix of the load (C). In the system (20), P represents the current densities in load
discretization parts.
T
Q1,1 · · · QNw ,Ndw K1 · · · KN

T T
[[[S] [0]] [D] [F ] [C]] df
P1 · · · PN dl
= [I · · · I 0 · · · 0]
(20)
4. Discretization Number
The accuracy of the CCM method depends on the discretization number. However, if
the discretization number is very important, this method can not be use in an inversion
process. We should do then a compromise between the precision and the computation
time.
• As far as the bobbin is concerned, the number of discretization has been reduced
by using the Kelvin distribution (section 3). The impact on the computational time is
remarkable since it has been divided by a factor 36.
• As far as the ferrite core is concerned, the number of elements and the type of dis-
cretization have an effect on accuracy. On figure 9(a), discretization is equally distributed
12
Altitude [mm]
10 12
Altitude [mm]
8 3 10
6 8 3
4 2 4 6
2 4 2 4
0 2
1 5 0 1 5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Radius [mm] Radius [mm]
(a) Discretization of the ferrite with w = 1. (b) Discretization of the ferrite with w = 2.
Figure 9. Influence of the discretization sort to the ferrite core (Ndf = 100).
on the whole surface. On figure 9(b), a weighting coefficient w is used to discretize finely
the segment 1 to 5 in which the field gradient is more important. The number and the
type of discretization’s impact on the resistance and the inductance simulation are pre-
sented on Fig. 10. The resistance and the inductance variations versus the number of dis-
cretizations are presented on 10(a) and 10(c), respectively. The resistance and inductance
relative deviation from the measures versus the number of discretizations are presented
on 10(b) and 10(d), respectively. We notice that the accuracy increases with the number
of discretizations. Table 2 shows the number of discretizations which are used to reach a
5% accuracy as far as resistance and inductance measures are concerned. It also notifies
62
Measurement
60 10
Simulations w=1
w=1.5 9
58 w=2
[m :]
[%]
Simulation w=1
56 w=1.5
7
w=2
54 6
5
52
4
240 314 474
50 3
0 200 400 600 800 1000 0 200 400 600 800 1000
Number of discretizations Number of discretizations
(a) Resistance versus discretization number. (b) Relative erreur of the resistance between
measures and simulations.
Measurement 14
56
12
54
52 10
50 Simulations w=1 Simulations w=1

w=1.5 8
[%]
w=1.5
[P H]
w=2 w=2
48
6
46
4
44
2 202 273
42 422
40 0
0 200 400 600 800 1000 0 200 400 600 800 1000
Number of discretizations Number of discretizations
(c) Inductance versus discretization number. (d) Relative erreur of the inductance between
measures and simulations.
Figure 10. Influence of the ferrite discretization number on the resistance and the inductance.
the simulation time. The bobbin used in the ferrite core is constituted of 5*5 coils, which
are discretized in 6 sectors only (using Kelvin distribution). The results prove that simu-
lation time and ferrite discretization number have been divided by 2 because segments 1
to 5 have been given priority.
• As far as the load is concerned, the number of discretization could be very great.
But the interesting zone is localized underneath the sensor. This zone is discretized finely
according to the radius and the elements are enlarged after the external sensor radius.
Moreover, the higher are the frequencies, the more important is the skin effect. Thus,
the current densities are located on the surface of the metallic plate. So, an exponential
discretization which depends on the skin depth is used according to the thickness (Fig.
1).
Table 2. Ferrite discretization to obtain an uncertainty of 5% between measurement and simulation.

w=1 w = 1.5 w=2
Discretization number to R 474 314 240
Discretization number to L 422 273 202
times (s) 7.5 4.6 3.6
Resistance (Ohm) Reactance (Ohm)

10 50
Lift Off = 1 mm Lift Off = 0.5 mm Lift Off = 0 mm

5
12
0 0
10 0 200 400 600 0 200 400 600
10 100
8
Altitude [mm]
6 5 50
4
0 0
2 0 200 400 600 0 200 400 600
10 100
0
0 5 10 15 20 25
Radius [mm] 5 50
×104 0 0
2 4 6 8 10 12 14 16 0 200 400 600 0 200 400 600
Frequency (kHz) Frequency (kHz)
Figure 11. Currents density in the load and the bob- Figure 12. Measure(–) and simulation(+).
bin.
5. Measurements and Simulations
Tests have been carried out on different conducting materials. The sensor is built with 25
coils (wire diameter 1mm), so skin effects are not negligible for frequencies greater than
50 kHz.
In the following example, a 2mm thick copper plate has been used (Fig. 11).
A comparison between measured and simulated resistance and reactance is pre-
sented on Fig. 12. The frequency moves from 1kHz to 500kHz and three different lift-off
(0mm, 0,5mm and 1mm) have been taken into account.
The relative error is less than 2% for reactance and 5% for resistance.
6. Inversion Process
The presented model allows to take into account the coil intrinsic skin and proximity ef-
fects. This forward model is fast-running and accurate; it can be used easily in inversion
process to determine geometrical and/or physical parameters (lift-off, conductivity,. . . ).
Optimization methods with a gradient were discarded, because a gradient cannot be ac-
curately computed in the case of non-analytical objective functions, as its first derivative
is not assured to be continuous. In this perspective, we implemented a Particle Swarm
Optimization which is a recent heuristic [11,12] able to escape from local minima. PSO
is a population based on the stochastic optimization technique developed by Eberhart
and Kennedy in 1995, inspired by the social behavior of bird flocking or fish schooling.
PSO shares many similarities with evolutionary computation techniques such as Genetic
Algorithms (GA). The system is initialized with a population of random solutions and
searches for optima by updating generations. However, unlike GA, PSO has no evolu-
tion operators such as crossover and mutation. In PSO, the potential solutions, called
particles, fly through the problem space by following the current optimum particles.
Most of evolutionary techniques have the following procedure:
• Random generation of an initial population
• Reckoning of a fitness value for each subject. It will directly depend on the distance
to the optimum.
• Reproduction of the population based on fitness values.
• If requirements are met, then stop. Otherwise go back to 2.

In our case, the optimization problem consists in determining sensor geometrical
characteristics (distance between the bobbin inner radius and the ferritic core(Rint), var-
nish thickness (ev), wire diameter (Dcu),...) with respect to measurements (Fig.13).
On Fig.14, we have represented the resistance and the inductance of a sensor. The
typical intrinsic data given by the manufacturer for the sensor have been taken as initial
parameters in the model (crosses), the solid line shows the measurements and the circles
illustrate the result obtained after optimization using PSO and our forward model.
Resistance (Ohm) Inductance (μH)

ev 1.4 15.5
Measurements
Initial
Dcu 1.2
15 Optimal
1
14.5
0.8
14
0.6
13.5
0.4
13
0.2 Measurements
Initial
Optimal
0 12.5
Rint 0 200 400 0 200 400
Frequency (kHz) Fréquency (kHz)
Figure 13. Parameters optimization. Figure 14. Impedance variation before and after
optimization.
Table 3 shows that the optimization process allows to reduce the gap between simu-
lations and measurements, providing the sensor parameters are closer to the real values.
Table 3. Optimization result.

Rint (mm) Dcu (mm) ev (μ m) erRmean (%) erXmean (%)
Initial 8,75 1 50 5.6037 1.2532
Optimal 8,789 0,972 77,3 1.9396 0.4051
7. Concluding Remarks
The presented model can be used for probe coil design (e.g. geometrical parameters),
for implementation of industrial NDT methods (e.g. frequency choice), for evaluation
of the influence of perturbation parameters (e.g. lift-off) and/or for physical properties
evaluation of a controlled material (σ, μ), thus limiting the number of experimental tests.
As it is fast running, it can be also driven by an optimization heuristic like Particle Swarm
Optimization for inversion problem in respect with measurements. In our inversion case,
the results for sensor geometrical parameters have been achieved due to the accuracy and
speed of the forward model and to the relevant choice of the objective functions allowing
to fit the measurements provided by the sensor with accuracy less than 2%.
References
[1] F. Buvat et al., Eddy-current modeling of ferrite-cored probes, Review of Progress in Quantitative Non-
destructive Evaluation, Vol. 24, pp. 463-470, 2004.
[2] G. Berthiau, B. de Barmon, MESSINE : Eddy current modeling in CIVA, 15th WCNDT, Roma, pp.
15-21, Oct 2001.
[3] C. V. Dodd, W. E. Deeds, Analytical Solution to Eddy-Current Probe-Coil Problems, J. Appl. Phys,
Vol.39, No.6, pp.2829-2838, may 1968
[4] A. Lefèvre, L. Miègeville, J. Fouladgar, G. Olivier, 3-D Computation of transformers overheating under
nonlinear loads, IEEE Trans. on Magn., Vol. 41, No. 5, May 2005.
[5] B. Maouche, M. Feliachi, N. Khenfer, A half-analytical formulation for the impedance variation in
axisymmetrical modelling of eddy current non destructive testing, Eur. Phys. J. Appl. Phys, Vol.33,
pp.59-67, 2006.
[6] D. Delage, R. Ernst, Prédiction de la répartition du courant dans un inducteur à symétrie de révolution
destiné au chauffage par induction MF et HF, RGE No.4/84, pp.225-230, Apr. 1984
[7] R. Ernst, A. Gagnoud, I. Leclercq, Etude du comportement d’un circuit magnétique dans un système de
chauffage par induction, RGE No.9, pp.10-16, Oct. 1987
[8] E.Durand, Magnétostatique, Ed. Masson et Cie, 1968
[9] A. Angot, Compléments de Mathématiques à l’usage des ingénieurs de l’électrotechnique et des télé-
communications, Ed. Masson et Cie, Edition 6, 1972
[10] G. Gaba, M. Abou-Dakka, A simplified and accurate calculation of frequency dependence con-
ductor impedance, IEEE/PES and NTUA 8th International Conference of haronics and quality of
power(ICHQP’98), pp. 939-945, Athens, Oct. 1998
[11] Kennedy, J. and Eberhart, R. C. Particle swarm optimization. Proc. IEEE int’l conf. on neural networks
Vol. 4, pp. 1942-1948. IEEE service center, Piscataway, NJ, 1995.
[12] R. C. Eberhart, J. Kennedy, A new optimizer using particle swarm theory. Proceedings of the sixth inter-
national symposium on micro machine and human science pp. 39-43. IEEE service center, Piscataway,
NJ, Nagoya, Japan, 1995.
IOS Press, 2007
Eddy Current Modelling for Inspection of

Riveted Structures in Aeronautics
S. PAILLARD a,1 , G. PICHENOT a , M. LAMBERT b , H. VOILLAUME c
a
CEA/LIST, CEA Saclay 91191 Gif-sur-Yvette France
b
L2S (CNRS-Supélec-UPS), 3 rue Joliot-Curie, 91192 Gif-sur-Yvette France
c
EADS CCR, DCR/SP/PN, 12 rue Pasteur, 92152 Suresnes France
Abstract In the framework of a collaborative project with EADS, a semi-analytical

model based on a volume integral method has been developed to simulate eddy
current (EC) inspection of riveted structures in aeronautics. The model handles a
layered structure by considering a dyadic Green’s approach where a fastener and a
flaw are introduced as a variation of conductivity in a stack of slabs. Experimental
data are used to validate the model.
Keywords. Eddy Current Testing, Aeronautic inspection
1. Introduction
EC technique is currently the operational tool used for fastener inspection which is an im-
portant issue for the maintenance of aircraft structures. The industry calls for faster, more
sensitive and reliable NDT techniques for the detection and characterization of potential
flaws nearby rivets. In order to reduce the development time, to optimize the design and
to evaluate the performances of an inspection procedure, CEA and EADS have started
a collaborative work aimed to extend the modelling features of the CIVA non destruc-
tive simulation platform to the simulation of multilayer assembly with fasteners. CIVA
(a) Bobbin coil placed (b) Bobbin coil placed on a (c) Three ferrite cores
inside a conducting tube configuration defined by CAD placed on a conducting slab
Figure 1. Representation of several configurations affected by a parallelepiped flaw in the CIVA user interface.
1 Corresponding Author: Séverine Paillard, CEA/LIST, CEA Saclay 91191 Gif-sur-Yvette France; E-mail:
Severine.PAILLARD@cea.fr
26 S. Paillard et al. / Eddy Current Modelling for Inspection of Riveted Structures in Aeronautics
is a powerful multi-technique platform for industrial NDT (see [1,2,3]). The developed
EC simulation models are mainly based on the volume integral method using the dyadic
Green’s formalism detailed in [4]. Several examples of CIVA for eddy current testing are
presented in Fig. 1. This paper describes the progress in developing a 3D computer code
for fastener modelling based on the volume integral equations which has the capability
to quickly predict the response of an eddy current probe to 3D flaws.
2. Description of the Model
2.1. Theoretical Formulation
A typical configuration of interest is depicted in Fig. 2. It consists of a layered planar

structure with a fastener and a semi-elliptical flaw nearby the lower part of the rivet. The
EC probe is moved along the surface, above the fastener assembly. This configuration
Figure 2. Typical aircraft configuration.
can be attacked in two steps: (i) modelling the response of a probe to a layered structure
with fastener without flaw; (ii) taking into account the flaw. Results of the first step are
given below. Those of the second one are not yet obtained and will be presented later.
The configuration is described as follows : the space is divided in two air half-spaces
numbered 0 and N + 1 with, in between, a N -layer slab, each layer being numbered i
and having a conductivity σ i (all materials are supposed to be non magnetic and of air
permeability μ0 ). The slab is affected by a defect of volume Ω and conductivity σ (r)
crossing one or more layers (as depicted Fig. 2). Let us denote with index m (resp. n)
n (m < n), the latter being sliced into
the first (resp. last) layer affected by the defect
as many layers as necessary such as Ω = k=m Ωk (note that, in the case of a rivet
crossing the N layers without his foot, m = 1 and n = N ). An exemple of the Ω domain
for a two-layered slab is shown in Fig. 3(a). A time-harmonic source (circular frequency
ω and implied time-dependence exp (jωt)) –a coil probe for example– is placed in the
upper half-space 0. The so-called vector domain integral formulation of the electric field
Ek (r) in the layer k in such a configuration is obtained by application of the Green’s
theorem onto the diffusive vector wave equation and is given by
n

(0) (ee)
Ek (r) = Ek (r) − jωμ0 Gkl (r, r’) [σl − σ(r’)] El (r’) dr’ ∀r’ ∈ Ωk
l=m Ωl
(1)
(0) (ee)
where Ek (r) is the primary field in the layer k and G kl (r, r’) the electric-electric
dyadic Green’s functions defined as the field response for a unit point source and solution
of
S. Paillard et al. / Eddy Current Modelling for Inspection of Riveted Structures in Aeronautics 27
(ee) (ee)
∇ × ∇ × Gkl (r, r’) − kk2 Gkl (r, r’) = δkl Iδ(r − r’). (2)
In the above equations k, l denote the index of the layer of the observation r and of the
source r’ point, respectively, I is the unit dyadic and δ kl stands for the Kronecker delta.
kl is the wave number in the l th layer defined as kl2 = jωμ0 σl . The Green’s dyad satisfies
the appropriate boundary conditions at the interfaces between the different layers in the
same way as the electric fields do. The response of the probe is given by its impedance
variation is obtained via the reciprocity theorem, where I 0 is the feeding current of the
probe, as
n
(0)
I02 ΔZ = [σl − σ(r)] El (r) · El (r)dr. (3)
l=m Ωl
2.2. Numerical Considerations
Once the model has been chosen and the equations established, the numerical formula-
tion can be implemented. Equation (1) is discretized using a Galerkin’s version of the
method of moments where the contrast zone Ω is sliced in N cell parallelepipeded voxels.
The voxels are chosen in order to have an homogeneous conductivity inside each voxel,
and in each voxel, the electric field is a constant-valued. This approach leads to a linear
system (4)
⎡ ⎤ ⎛ ⎡ ⎤⎞ ⎡ ⎤
(0)
Em Gm,m · · · Gm,n Em
⎢ . ⎥ ⎜
⎢ . ⎥ = ⎝I − ⎢ . . ⎥⎟ ⎢ . ⎥
⎣ .. . . . .. ⎦⎠ ⎣ .. ⎦ (4)
⎣ . ⎦
(0)
En Gn,m · · · Gn,n En
where Gi,i are the electromagnetic self-coupling terms of the i th region of the sliced rivet
onto itself and where G i,j are the mutual coupling terms of the j th over the ith .
An example is given for a two-layered slab (n = 1 and m = 2) in Fig. 3. The
rivet illustrated in Fig. 3(a) is here sliced into two parts, each one entirely contained
in a single layer of conductivity σ k . The self-coupling terms G i,i with i ∈ {1, 2} are
represented in Fig. 3(a) and the mutual-coupling terms G i,j with (i, j) ∈ {1, 2} and i = j
are represented in Fig. 3(b).
(a) Rivet and the contrast zone Ω (a) Self-coupling (b) Mutual coupling
sliced in two inhomogeneity zones
Figure 3. Example of a rivet in a two-layered slab.
For building this multi-layer model, two main improvements have been made:
• Self-coupling terms: the planar stratification of the work piece is taken into ac-
count by introducing generalized reflection and transmission coefficients at each
interface in the Green’s dyads.
• Mutual coupling terms: the mutual Green’s functions are written in explicit ana-
lytical expressions [4] and implemented to reconstruct the entire matrix of equa-
tion (4).
In the applications which we are interested in, the typical size of the domain Ω may be
more than ten skin-depths which leads to a large number of voxels and to a too large
linear system to invert (the memory size can be estimated as O (9 N cell 2
)). Taking into
account the convolution structure of the integral equation (1) with respect to the two lat-
eral directions via appropriate fast Fourier transforms, an iterative solution of the system
4/3
allows us to treat larger defects by reducing the memory size to O (9 N cell ).
3. Validations
On one hand, the model is developed to handle a defined configuration –a rivet within a
laminated slab– and from this point of view, we have to validate the two first aspects of
the fastener modelling illustrated in Fig. 4. On the other hand, this model is a multi-layer
model –an inhomogeneity zone embedded in a laminated work piece– and therefore,
we have to validate this multi-layer modelling also. In order to focus on these different
aspects, and to avoid errors in rivet shape simulation, the rivet with its typical flat head
shape is assumed in all validations to be a cylindrical through-wall hole. The flat head
shape of the rivet can be obtained by introducing volume ratios in the calculation zone.
Several validations have been done to improve the two approaches -handle a fastener in
(a) Rivet hole (b) Rivet crossing (c) Rivet in a multi-layer slab
in one slab a multi-layer slab and a flaw nearby
Figure 4. Different aspects of fastener modelling.
a laminated slab and a multi-layer configuration- of this multi-layer model:

• Fastener approach: through-wall inhomogeneity zone in one slab (first aspect,
Fig. 4(a)) and in a two-layer slab (second aspect, Fig. 4(b)). The third aspect
(Fig. 4(c)) is not treated yet.
• Multi-layer approach: inhomogeneity zone contained successively in the different
layers of a two-layer slab.
For all such studies, the same air-cored probe is used (an inner radius of 1 mm, an outer
radius of 1.6 mm, a lift-off of 0.32 mm and a height of 2 mm with 320 turns) and is
displaced along the diameter of the hole.
Figure 5. Cylindrical through-wall hole in one layer of aluminium (— experimental data, +++ CIVA results).
3.1. One-Layer Validation
An impedance meter HP4194 is used to measure the impedance of the air-cored probe
working in absolute mode at the frequency of 10 kHz on a through-wall hole in an alu-
minium slab (Fig. 5, left). The hole diameter is 4.9 mm and the slab thickness is 4 mm
with a conductivity of 30 MS/m. The agreement between the model and the experimental
data is better than 1% for the amplitude and 8 ◦ in phase (Fig. 5, right).
3.2. Multi-Layer Validations
The multi-layer modelling has been validated on a two-layer slab (mock-up inconel-
aluminium) described as follows: an inconel slab with a conductivity of 1 MS/m and
a thickness of 1.27 mm lies above an aluminium slab with a conductivity of 30 MS/m
and a thickness of 4 mm. A cylindrical hole of 4.9 mm is crossing one (inconel slab,
Fig. 7(a)) or the other (aluminium slab, Fig. 8(a)) or both (Fig. 9(a)). The air-cored probe
is working here at 75 kHz.
3.2.1. Calibration
In most industrial applications, the measured EC signal is calibrated over a reference
flaw. Preliminary to these validations, a calibration experiment has been made; the refer-
ence flaw is a surface breaking notch in an inconel slab with a conductivity of 1 MS/m.
The EDM notch is 0.1 mm in width, 20 mm in length and 0.93 mm in depth and the
thickness of the slab is 1.55 mm as shown in Fig. 6 (left). The impedance variation mea-
sured in the impedance plane calibrated at 500 mV and 135 ◦ is presented in Fig. 6 (right).
Figure 6. Response of the probe to a breaking notch in a slab (— experimental data, +++ CIVA results).
(a) Cylindrical hole in (b) Cylindrical hole in

the inconel layer with the inconel layer with
perfect matching imperfect matching
(c) Calibrated signals in the impedance plane
Figure 7. Cylindrical hole in the inconel layer of a two-layer slab perfectly matched (— experimental data,
*** CIVA results with perfect matching, +++ CIVA results with imperfect matching).
3.2.2. Imperfect Matching Slabs Influence

For the hole in the inconel slab (like for the others but we will come back to them later
on), the result is not completely satisfactory (Fig. 7); even if the agreement for the mea-
surement of the EC signal in the impedance plane between the model and the experi-
mental data is better than 4% for the amplitude and 2 ◦ in phase, the shapes of the signal
are different (Fig. 7(c)). One of the reasons can be that the simulated configuration does
not correspond exactly to the reality of the experimental configuration. As a matter of
fact, in the experiment, the two slabs could not be fastened in perfect fashion (like on
the Fig. 7(a)), causing the occurrence of a thin air layer in between. A study has been
carried out to evaluate the thickness of the layer of air to be taken into account and the
best results have been obtained with a thickness of 50 μm (Fig. 7(b)). All the results
presented in the next subsection take into account this air gap.
3.2.3. Results of Validations

For the hole in the inconel slab, the agreement between the model and the experimental
data is better than 6% for the amplitude and 2 ◦ in phase (Fig. 7(c)) whereas, for the
hole in the aluminium slab, the agreement is better than 4% for the amplitude and 3 ◦ in
phase (Fig. 8(b)). For the through-wall hole in the two-layer slab (Fig. 9), the agreement
between the model and the experimental data is better than 2% for the amplitude and 3 ◦
in phase (Fig. 9(b)).
(a) Hole in the alu- (b) Calibrated signals in the impedance plane with imperfect
minium layer with im- matching
perfect matching
Figure 8. Cylindrical hole in the aluminium layer of a two-layer slab (— experimental data, +++ CIVA re-
sults).
(a) Hole crossing a two- (b) Calibrated signals in the impedance plane with imperfect
layer slab with imperfect matching
matching
Figure 9. Cylindrical through-wall hole in a two-layer slab (— experimental data, +++ CIVA results).
4. Application in Aeronautics
Once the model has been validated, we can consider a realistic case: two identical multi-
layered slabs held together by a rivet. We have applied the model to the calculation of
the impedance variation of a ferrite-cored probe used to test the aeronautical work piece
illustrated in Fig. 10. One slab is decomposed in three thin layers of aluminium alloy,
Figure 10. Cylindrical through-wall hole in two multi-layer slabs.
bonded together with non-conductive material. The aluminium slabs are 0.3 mm in depth
and the non-conductive slabs are 0.25 mm, the fastener hole has a diameter of 4.9 mm.
The cylindrical ferrite-cored probe used for these studies has an inner (resp. outer) radius
of 3.74 mm (resp. 7.325 mm), and a height of 3.46 mm with 926 turns and works at
2.6 kHz. The results presented in Fig. 11 have to be validated with experimental data,
Figure 11. Simulated response of the probe to a cylindrical through-wall hole in two multi-layer slabs.
however they are coherent with what is expected in such a configuration: (i) when the
centre of the probe is exactly above the centre of the cylindrical hole, the signals are
almost null because the inner diameter of the probe is larger than the diameter of the hole,
and so the currents are almost undisturbed. (ii) when the centre of the probe is at 5 mm
from the centre of the hole, the signals (Fig. 11) are at their maximum (resp. minimum)
for the real part (resp. for the imaginary part) corresponding to the positions where most
of the winding is above the hole.
5. Conclusion and Perspectives
The extension of the CIVA platform to the simulation of riveted structures is currently
in progress. The multi-layer model is now validated, with a good agreement between the
model and the experimental data, for a cylindrical through-wall hole in a set of two slabs,
a cylindrical hole either in the top slab or in the bottom slab of the stack. A first milestone
has been reached with the development of a model taking into account the presence of
a rivet in a layered slab assembly. Validations with experimental data of the 3D model
developed here for fastener modelling have been carried out successfully. Work is in
progress to calculate the probe response due to the presence in a fastened structure of
both a rivet and an embedded flaw located nearby as shown in Fig. 4(c).
Acknowledgements
This research is supported by the Paris Ile-de-France Region.
References
[1] Buvat F., Pichenot G., Prémel D., Lesselier D., Lambert M. and Voillaume H., Eddy current modelling
of ferrite-cored probes, in Review of Progress in QNDE Vol. 24, 2005, pp. 463-470.
[2] Sollier T., Buvat F., Pichenot G. and Premel D., Eddy current modelling of Ferrite-Cored Probes, ap-
plication to the simulation of Eddy current signals from surface breaking flaws in austenitic steel, Proc.
16th World Conf. on NDT, Montreal, 2004.
[3] Pichenot G., Buvat, F., Maillot V. and Voillaume H., Eddy current modelling for non destructive testing,
Proc. 16th World Conf. on NDT, Montreal, 2004.
[4] Chew W.C., Waves and Fields in Inhomogeneous Media, Van Nostrand Reinhold, New York, 1990.
IOS Press, 2007
Numerical Modeling of a Phase Sensitive

Eddy Current Imaging System
Guglielmo RUBINACCI a,1, Antonello TAMBURRINO b, Salvatore VENTRE b,
Pierre-Yves JOUBERT c and Jean PINASSAUD c
a
Ass. EURATOM/ENEA/CREATE, DIEL,
Università degli Studi di Napoli Federico II Italy
b
Ass. EURATOM/ENEA/CREATE, DAEIMI, Università degli Studi di Cassino, Italy
c
SATIE, ENS Cachan, CNRS, Universud, F-94230 Cachan, France
Abstract. This work focuses on an innovative Eddy Current Imager dedicated to

the high-speed and high-resolution non-destructive testing of large metallic
structures, such as the riveted lap joints of aircrafts. The system produces
time-harmonic in – phase and in – quadrature eddy current images, thanks to a
specific linear magneto-optic set-up, which actually provides a true measurement
of the 2-D spatial distribution of the magnetic flux density at the surface of the
inspected structure. In view of quantitative imaging of defects in planar structures,
an appropriate numerical model, based on an efficient integral formulation, is
presented and validated against the measurements.
Keywords. Eddy current imaging, nondestructive evaluation, 3D numerical

modeling experimental and computed data, surface and buried defects.
1. Introduction
Magneto-Optic (MO)/eddy current (EC) imagers appear to be a good alternative to

conventional EC sensors such as pencil probes or array sensors, for the non-destructive
evaluation (NDE) of large metallic structures such as the riveted lap joints of aircrafts.
Indeed, these imagers provide real time and possibly high resolution images relative to
the integrity of the structure, without intensive mechanical scanning. MO imagers
dedicated to NDE were firstly introduced by [1]. However, this type of imager only
provide “two-level” images resulting from the comparison to an adjustable reference
threshold. These features actually limit both the efficiency and the defect
characterization possibilities. In this paper, we focus on an original eddy current (EC)
imager [2] able to provide true in-phase and in-quadrature EC images which are
linearly related to the spatial distribution of the magnetic flux density at the surface of
the inspected structure. These EC images are suitable to be processed by a quantitative
inversion algorithm in order to carry out an exhaustive defect characterization.
However, the success of this approach requires the precise knowledge of the
interactions between the imager and the inspected structure. To this purpose, it is
1
Corresponding Author: Guglielmo Rubinacci, Euratom/Enea/Create, Dipartimento di Ingegneria Elettrica,
Università degli Studi di Napoli Federico II, Via Claudio, 21 – 80125, Napoli, Italy; E-mail:
rubinacci@unina.it
34 G. Rubinacci et al. / Numerical Modeling of a Phase Sensitive Eddy Current Imaging System
essential to develop numerical formulations, tailored for nondestructive testing

applications, that satisfy the following three requirements: (i) the numerical method has
to be capable of handling different scales ranging from the defect scale (small scale) to
the imager scale (large scale), (ii) numerical errors must be small enough otherwise,
due to the ill-posedness of the inverse problem, they may alter significantly the
reconstruction and (iii) the numerical method has to be fast enough to be incorporated
into an iterative inversion procedure. Finally the numerical model should also take
efficiently into account the magnetic circuit of the EC inductor.
In this paper, we present the numerical simulation of a quantitative MO/EC
inspection of a riveted lap joint with a surface breaking or a buried crack in the
proximity of the rivet hole. The numerical model is briefly illustrated and validated
against the measurements.
2. Description of the Eddy Current Imager
The diagram of the imager [2] is presented in Figure 1. Its working principle is based
on the combination of an EC inductor used to excite the material under inspection, with
a specific MO set-up used to image the spatial distribution of the normal magnetic field
at the surface of the inspected area, in real and imaginary parts. The specific
configuration of the EC inductor allows a uniformly oriented eddy current flow to be
generated in a large inspection area. The presence of a defect induces the rise of a
non-zero normal component of the magnetic field in the vicinity of the defect. This
component is sensed by a dedicated MO garnet film relying on the Faraday effect. The
garnet features a linear and hysteresis free magnetization loop. It is integrated to an
optical set-up so that the variation of the magnetic field is translated into the variations
of the intensity of light beam measured by a CCD camera. The acquisition rate of the
used camera (25 images/s) being far smaller than the excitation frequency of the EC
inductor (100Hz up to 20kHz), the images are obtained using a stroboscopic approach,
followed by a digital lock-in used to obtain the real and imaginary parts of the magnetic
field. The EC imager prototype used in this study allows 45mm diameter images to be
obtained in 15 seconds, and was designed to provide a 100μmu100μm spatial
resolution.
3. The Numerical Model
The numerical model here implemented is based on the numerical formulation

presented in [3]-[7]. The problem under consideration is the calculation of the magnetic
field perturbation induced on a conducting specimen by a time harmonic magnetic field
in the presence of a perfectly insulating defect and linear magnetic materials. The direct
solution of this problem may require a very heavy computational effort, since the defect
size is usually smaller than the other relevant characteristics of the system (size of the
specimen and of the eddy current inductor). Consequently, a proper discretization of
the system calls for a very large number of unknowns. Moreover one has to take into
account that the signal due to the defect is usually weak so that numerical errors can
compromise the numerical solution.
G. Rubinacci et al. / Numerical Modeling of a Phase Sensitive Eddy Current Imaging System 35
Our approach exploits a volume integral formulation, to efficiently reduce the

discretization only to the sources region, and the superposition principle, to separately
compute the current density perturbation due to holes and defects.
Light proof box
Synchronisation
board
CCD camera
Light source
Polarizer PC
Analyzer
Current controlled
AC power.
& 100Hz-20kHz
z
Coils
& &
x y MO sensor EC inductor
2 layer lap joint mockup

defect
Figure 1. General diagram of the eddy current imager

In our volume integral formulation, the unknowns are the two-component vector
potential T defined in the conducting region Vc (where the current density J is given by
its curl) and the magnetization vector M defined in the magnetic region Vf, that in the
present case does not coincide with Vc. Assuming a sinusoidal excitation, we represent
the current density J in terms of edge shape functions Tk and the magnetization vector
as a piecewise uniform function in terms of elementary pulse functions Pk(x)
n m
J (x) ¦I
k 1
k u Tk (x) , M (x) ¦M
1
k Pk ( x ) (1)
The gauge based on the tree-cotree decomposition of the mesh [8] assures the
uniqueness of Tk.
Applying the Galerkin's approach to the electric and magnetic constitutive
equations, the following linear system of equations is obtained:
³uT k (KJ jZA)dV 0 Tk (2)

Vc
³P
Vf
k [M kB]dV 0 Pk (3)
where B is the magnetic flux density, A is the magnetic vector potential, K is the
resistivity and k P r 1 P 0 P r . A and B are calculated from J and M via
Biot-Savart law.
Having defined I and M as the column vectors made by the complex coefficients
of the expansions (1), equations (2) and (3) lead to a linear system of equations, that
* *
when solved for M, reduces to ZI jZ U , where Z is a n u n full matrix, and U
is a n u 1 column vector related to the external sources, whose explicit expressions can
be found in [3, 6].
Using superposition, the forward problem is reformulated [3], [5], [6] as the
determination of the modified eddy current pattern J = J0 + GJ. Here, J0 is the
unperturbed current density in the presence of the hole, whereas GJ = 6k=1,n GIk Jk is
the perturbation due to the crack.
The crack is assumed to be thin, so that it may be computationally convenient to
treat it as a surface 6d, discretized via a set of finite element facets, with the constraint
J. n̂ = 0, leading to GJ. n̂ = J0 . n̂ , where n̂ is the normal unit vector on the
face.
We then make a change of variables:
GI = K GX - S G0 (4)
where GX is an auxiliary variable which gives current densities with zero flux through
the crack and G0 is a particular set of values giving a net flux through each facets of the
crack according to the constraint GJ. n̂ = J0. n̂ . The definition of the matrices K and
S is given in [5].
Galerkin’s procedure in terms of the new variables yields:
KTZK GX = KTZS G0 (5)
This approach, introduced in [3, 5], was also applied to the treatment of volumetric
cracks [4, 6, 7]. The possibility of treating volumetric defect allows to efficiently
compute also the effect of the hole on the unperturbed current density J0. Specifically,
J0 can be represented as the superposition of J0,P and GJ0,H where J0,P is the current
density when only the plate is present and GJ0,H its perturbation due to the presence of
the hole. Moreover, the interaction between GJ0,H and the iron yoke is negligible (GJ0,H
flows in a region that is relatively “far” from the yoke) and, therefore, the presence of
the magnetic material can be limited to the computation of J0,P only.
One of the advantages of the proposed approach is that it is possible to define a
region where the crack could be located ad to pre-compute all the unknowns external to
that region in terms of the unknowns in the tentative crack region. In this way, the
magnetic field associated to any set of faces belonging to the possible crack region can
be computed in a very fast way by solving a very small linear system.
4. Results
The eddy current inductor is shown schematically in Figure 2a. The magnetic poles
(panels) are vertical (without any tilt angle as opposed to the drawing) and directly in
contact with the inspected mockup. Two induction windings are winded around 10mm
diameter ferritic rods. Each winding features 2 coils of 120 turns each. (the total
number of turns is 4 u 120 turns). The four windings are connected in parallel. The
magnetic circuit is made out of 3C90 with a relative magnetic permeability μr| 2300
and a resistivity of 5:m, that has been assumed infinite in the numerical simulation.
The active area of the MO sensor film is a circle with a diameter of 45 mm.
The tested mock-up is a laboratory made riveted lap joint with a thin notch
(surface or buried) in the proximity of a rivet hole, simulating a crack, as shown in
Figure 2b. The two 1.5 mm thick non-magnetic plates are made in Aluminium with a
conductivity V| 20 MS/m, according to the manufacturer. The coils are fed with
sinusoidal current, the excitation frequency being 1kHz.
Due to the symmetry of the eddy current inductor and of the mock-up, the
computation of the eddy currents in absence of the holes and notch can be made by
discretizing only one forth of the system. The finite element mesh, shown in Fig. 3a is
made of 3476 elements in the conducting region and 796 elements in the magnetic
region, leading to 5106 complex Degrees of Freedom (DOF) for the eddy currents and
2388 complex DOF for the magnetization. The finite element mesh to compute the
field due to the hole is again symmetric, in the hypothesis that the hole is at the center
of the eddy current inductor. This is not exactly true (the holes in Figure 2b are not
exactly at the center of the eddy current inductor), but it is a good approximation
because the inductor produces an almost uniform field in the hole region. This finite
element mesh is shown in Figure 3b and is made of 2196 elements in the conducting
region, leading to 3226 complex DOF for representing the eddy current density. The
magnet is discretized as in the previous case, although it has been checked that its
contribution in this part of the computation is absolutely negligible, as already
remarked in Section 3. A third finite element mesh is necessary for computing the
crack contribution.
In this case, we again assume that the crack is located at the center of the inductor,
so that only half of the mock-up can be discretized. Again, this is a good approximation
in the reasonable hypothesis of an almost uniform inducing field in the region of the
crack and the hole. The finite element mesh is shown in Figure 3c and is made of 4320
elements in the conducting region leading to 6328 complex DOF. In this case half of
the EC inductor should be discretized, leading to twice the magnetic unknowns than
previously. Of course, it can be verified that also in this last case the presence of the
a)
12 mm x 1.5 mm x 0.5mm
(length x depth x width ) 5 mm x 1.5 x 0.5
2 Aluminium Alloy plates (length x depth x width )
V | 20 MS/m ; μr=1
1.5 mm
Notches on the first layer
4 mm diameter
1.5 mm
20 mm
b)
Figure 2. Eddy current inductor (a) and riveted lap joint mock-up (b).
inductor does not produce any appreciable contribution to the field. The measurements
are available for the surface defect and the buried defect.
The first simulation refers to the hole without defect. In particular, in Figure 4 it is
shown the flux density component normal to the specimen corresponding to both
experimental and numerical data on a rectangular region above the hole and along a
line passing through the center of the hole in a direction orthogonal to the induced
current density. The data are normalized to the maximum value of the field modulus.
The computed results are also shown for another value of the conductivity, namely V|
35 MS/m, corresponding to the pure aluminum. The slight asymmetry in the
measurements can be ascribed to a possible tilting of the garnet film. The results for the
surface and buried defects are shown in Figures 5 and 6, showing a satisfactory
agreement with the measurements. In the pictures the dimensions of each square pixel
have been assumed to be actually 97.7μmu97.7μm, rather than 100μmu100μm.
a) b) c)
Figure 3. The finite element mesh (a) in the unperturbed case, (b) for the hole and (c) for the crack
perturbation.
V=20MS/m
a) b) c)
Figure 4. The real part of Bz/max(|Bz|) for both experimental (a) and numerical data (b) on a rectangular
region above the hole; The Lissajous plot (c) of Bz/max(|Bz|) along a line passing through the center of the
hole. The computed results for V=20MS/m (diamonds) and V=35MS/m (crosses) are compared with the
measurements.
a) b)
Figure 5. The Lissajous plot of Bz/max(|Bz|) along a line passing through the center of the hole for a surface
(a) and a buried (b) defect. The computed results for V=20MS/m (diamonds) and V=35MS/m (crosses) are
compared with the measurements.
V=20MS/m V=35MS/m V=20MS/m V=35MS/m
(a) (b) (c) (d) (e) (f)

Figure 6. The real and imaginary part of Bz/max(|Bz|) for both experimental (a), d)) and numerical data (b),
c),e),f)) on a rectangular region above the hole, in the presence of a buried defect; The computed results
refers to V=20MS/m (b),e)) and V=35MS/m (c),f)).
V=20MS/m V=35MS/m V=20MS/m V=35MS/m
(a) (b) (c) (d) (e) (f)

Figure 7. Same as Figure 6, but in case of a surface defect.
It should be explicitly mentioned that the presence of the magnetic core in the
perturbed models does not play any role, so that the computational model is very fast
and the shape of the crack can be very efficiently detected with a very limited
computational effort, as in the non magnetic case. For instance the field variation due
to the introduction in the crack shape of additional boundary facets is an almost real
time computation (few ms of a PC CPU time).
5. Conclusions
In this paper we have presented a numerical tool able to fully simulate a new
quantitative MO/EC imager. Usually, MO/EC imagers provide only “two-level”
images that are not well suited to give enough information for NDE, as in cases when
one needs to discriminate surface and buried defects and evaluate their size and shape.
This two-level feature actually limits both the efficiency and the defect characterization
possibilities. In this paper, we have shown that the magnetic flux density data provided
by a new eddy current imager can be reproduced by an effective numerical tool leading
to a very fast and accurate evaluation procedure for the detection of cracks in a riveted
lap joint. As a matter of fact, a linear model of the sensor has been coupled to a
numerical scheme able to simulate the effects of cracks in conducting bodies in terms
of magnetic field perturbation. The resulting tool is extremely useful in understanding
more deeply the behaviour of the sensor in an advanced ECT application. This is
fundamental if the inverse problem (given the measurements provided by the sensor,
find the characteristics of the crack) must be solved.
Acknowledgments
This work was supported in part by the Italian Ministry of University (MIUR) under a Program for the
Development of Research of National Interest (PRIN grant # 2004095237) and in part by the CREATE
consortium, Italy.”
References
[1] G. L. Fitzpatrick et al. Magneto optic/eddy current imaging of ageing aircrafts, Mat. Eval. (1993), pp.
1402-1407.
[2]. P.-Y. Joubert and J. Pinassaud, Linear magneto-optic imager for non-destructive evaluation, Sensors and
Actuators A: Physical, Volume 129, Issues 1-2, 24 May 2006, pp. 126-130.
[3] R. Albanese, G. Rubinacci, F. Villone, “Crack simulation in the presence of linear ferromagnetic
materials using an integral formulation”, Electromagnetic NDE (V), (J. Pavo et al. Eds.), pp. 16-21, IOS
press, 2001.
[4] G. Rubinacci, A. Tamburrino, S. Ventre, F. Villone, Numerical Modelling of Volumetric Cracks, Int. J.
Appl. Electromag. Mech, vol. 19, pp. 345-349, 2004.
[5] R. Albanese, G. Rubinacci, F. Villone, “An Integral Computational Model for Crack Simulation and
Detection via Eddy Currents”, J. Comp. Phys., Vol 152, 736-755, 1999.
[6] R. Albanese, G. Rubinacci, A. Tamburrino, F. Villone, "Phenomenological approaches based on an
integral formulation for forward and inverse problems in eddy current testing", Int. J. Appl. Electromag.
Mech., vol. 12, pp. 115-137, 2000.
[7] M. Morozov, G. Rubinacci, A. Tamburrino, S. Ventre, Numerical Models with Experimental Validation
of Volumetric Insulating Cracks in Eddy Current Testing, IEEE Trans. Mag, Vol. 42, no. 5, May 2006,
pp 1568-1576.
[8] R. Albanese and G. Rubinacci., Finite Element Methods for the Solution of 3D Eddy Current Problems,
Advances in Imaging and Electron Physics 102 (1998) 1-86.
IOS Press, 2007
Developments in Modelling Eddy Current

Coil Interactions with a Right-Angled
Conductive Wedge
Theodoros THEODOULIDIS a,b,1 , Nikolaos POULAKIS a and John BOWLER c
a
TEI of West Macedonia, Electrical Engineering Department, Greece
b
University of West Macedonia, Energy Department, Greece
c
Iowa State University, Center for Nondestructive Evaluation, USA
Abstract. Recently we presented an analytical solution for the 3D configuration of

a cylindrical coil at the edge of a conductive block and calculated the impedance
variation with position relative to the edge. Since then we have been seeking ways
to improve and extend the analytical and numerical treatment of this canonical
problem. In the present paper, several extensions to previous work are presented
including a modification to the expressions for the field potentials in double series
form and the generalization of the field source to coils of arbitrary shape and orien-
tation. Experimental results involving a cylindrical coil of arbitrary tilt are shown
to verify the calculations.
Keywords. Eddy current testing, analytical modelling, conductive wedge
1. Introduction
The problem of evaluating the impedance change of a coil at the edge of a conductive
block modelled as a conductive quarter space has been solved recently by using the Trun-
cated Region Eigenfunction Expansion (TREE) method [1]. The problem was formulated
using a Cartesian coordinate system with the z-direction perpendicular to one surface of
the block and the edge in the y-direction, Figure 1. The solution was made possible by
introducing artificial boundaries that limit the problem domain in the x-direction. The
truncated domain is divided into regions, the solution expressed as series expansions in
each region and the expansion coefficients found from the continuity conditions govern-
ing the field at the interfaces between each region. The approach is based on the principle
of mode matching and the creative use of truncation boundaries to find quasi-analytical
solutions to boundary value problems that would be intractable without modified bound-
aries.
Using symmetry considerations, the solution was extended to the case of a plate [2]
and a through thickness slot [3]. In the present study, we report on further developments
including (i) the reformulation of the problem by expressing the electromagnetic field
1 Corresponding Author: Theodoros Theodoulidis, University of West Macedonia, Energy Department,
Bakola & Sialvera, 50100 Kozani, Greece; E-mail: theodoul@uowm.gr
42 T. Theodoulidis et al. / Developments in Modelling Eddy Current Coil Interactions
and coil impedance in double series form which is advantageous in terms of computer
implementation and convergence control and (ii) extension to a cylindrical coil of arbi-
trary tilt. In particular, we provide general expressions for the magnetic field, eddy cur-
rent density and impedance change of the coil in terms of source coefficients that char-
acterize the coil. These coefficients depend only on the isolated coil magnetic field and
can be found by using the Biot-Savart law.
Figure 1. Problem geometry. A cylindrical coil is moved across the edge of a right-angled conductor.
2. Analysis
In the new approach, the solution domain for the boundary value problem is truncated in
both x and y directions. Formally the truncation was in the x-direction only [1]. Thus,
the solution domain extends from 0 to hx in the x-direction and from 0 to hy in the y-
direction. The presence of four boundary surfaces at x = 0, hx and y = 0, hy means that
we have a wide choice of boundary condition combinations. The choices include one
which defines a perfect magnetic insulator, Bn = 0 (n stands for normal component) and
one which defines a perfect electric insulator, Bt = 0 (t stands for tangential component).
For the case examined here where the edge at x = c as well as the coil are located far
from the boundaries these choices have a negligible effect on numerical values of the
coil response. However, they determine the form of expressions for the solution and the
combination of eigenfunctions-eigenvalues used to represent it. In this work we consider
a magnetic insulator at x = 0, x = hx , y = hy and an electric insulator at y = 0. The
choice of magnetic insulation is consistent with previous work [1]. The reason for the
discrepancy in the boundary condition at y = 0 is clearly logistic. In this way we end up
with sines instead of cosines in the Y -dependence of the expressions for the potentials
used in the solution. The use of sines means the absence of a dc term in the series which
introduces a small simplification. From the physical point of view it does not have any
effect on the solution since all the boundaries are located far from the source coil.
T. Theodoulidis et al. / Developments in Modelling Eddy Current Coil Interactions 43
(a) (b)
Figure 2. (a) Normal cylindrical coil above a conductive quarter-space (b) Tilted cylindrical coil above the
edge of a conductive quarter-space.
2.1. Field Expressions
Consider Figure 2 which shows a coil located above a right-angled conductive non-
magnetic quarter-space with a conductivity σ. The coil is excited by a time harmonic
current varying as the real part of I exp(jωt). The analysis of the electromagnetic field
problem is based on the use of potentials. In the air-region between the lowest point of
the coil and the upper conductor surface, the magnetic field can be expressed as the gra-
dient of a scalar potential B = ∇φ where φ satisfies the Laplace equation. The poten-
tial can be considered as the superposition of the isolated coil potential and the potential
originating from the eddy currents in the conductive quarter-space φ = φ(s) + φ(ec) . The
expressions for these two potentials are then written as:
∞
∞
(s)
φ(s) (x, y, z) = cos(ui x) sin(vj y)eκij z Cij (1)
i=0 j=1
∞
∞
cos(ui x) sin(vj y)e−κij z Dij
(ec)
φ(ec) (x, y, z) = (2)
i=0 j=1
The magnetic flux density in the region below z = 0 can be written using the second
order vector potential as B = ∇ × ∇ × W where W = Wa x0 + x0 × ∇Wb and Wa , Wb
satisfy either the Laplace or Helmholtz scalar equations according to the conductivity of
the sub-region. Here x0 is a unit vector. The eddy current density in the quarter-space is
written as J = −jωσ∇ × W. Expressions for the two potentials satisfy the continuity
conditions on the magnetic field at the z = 0 plane and the x = c half-plane (z < 0).
By ensuring continuity and satisfying the insulator boundary conditions at the truncation
boundaries, it is found that
⎧
⎪ ∞
(a)
⎪
⎪ sin(vj y) xevj y C0j +
⎪
⎪
⎪
⎨
j=1

∞
(a)
Wa (x, y, z) = + sin(pi x)e γij z
ai Cij ; 0≤x≤c (3)
⎪
⎪
⎪
⎪ ∞ ∞
i=1
⎪
⎪ (a)
sin[qi (hx − x)] sin(vj y)eγij z Cij ; c ≤ x ≤ hx
⎩
j=1 i=1
∞
(b)
Wb (x, y, z) = C0j cosh[k(hx − x)] cos(vj y)evj z
j=1
∞
∞
(b)
+ cos[ri (hx − x)] cos(vj y)esij z Cij ; c ≤ x ≤ hx (4)
j=1 i=1
where ui = iπ/hx , vj = (2j − 1)π/(2hy ), κ2ij = u2i + vj2 , k 2 = jωμ0 σ, γij 2

=
qi + vj + k , pi = γij − vj = qi + k , ri = (2i − 1)π/[2(hx − c)], sij = ri + vj + k 2 . A
2 2 2 2 2 2 2 2 2 2 2
solution of (4) in the region 0 ≤ x ≤ c is not necessary since B depends only on Wa in

nonconductive regions [1]. The values of qi and hence pi are sought from the continuity
of the magnetic field at x = c which requires
qi tan pi c + pi tan qi (hx − c) = 0 (5)
the roots of which give the eigenvalues for Eq. (3). These complex eigenvalues do not
depend on the variable vj , which means that their numerical computation needs to be
carried out only once. In addition in Eq. (3)
sin qi (hx − c) qi cos qi (hx − c)

ai = =− i = 1, 2, 3, ... (6)
sin pi c pi cos pi c
and
(a) (b) (b)
C0j = k 2 cosh[k(hx − c)]C0j = a0 C0j (7)
The source coefficients in Eq. (1) are considered to be known and hence all other
coefficients Eqs. (2)-(4) are calculated in terms of them. This is done by imposing the
interface conditions at the surface z = 0, see [1] for details of the procedure. The final
expressions are:
∞
(ec) k2 (−1)i+1 (b) (ec) T (b)
D0j = (sij − vj ) Cij = Rj Cj (8)
2vj hy i=1 ri
(s) k2

∞ i+1
(b)
hx vj C0j + 2 (sij + vj ) (−1)
ri Cij
(b) i=1 (b) (s) (b) T (b)
C0j = = λj C0j +Rj Cj (9)
k2 cosh[k(hx − c)]cvj + k sinh[k(hx − c)]vj
while the other terms (i = 0) are calculated from the solution of the following matrix
system, for each value of the index j
hx ! (s) (ec)
"
(a)
u Cj + Dj = M s Cj (10)
2
hx ! (s) (ec)
"
(a)
!
(b)T
"
(b)
vj Cj + Dj = vj Mc Cj + k 2 Mr sj + vj Lj Rj Cj (11)
2
(b) (s)
+ k 2 vj λj Lj C0j
hx ! (s) (ec)
"
(a)
!
(b)T
"
(b)
κj Cj − Dj = γ j Mc Cj + k 2 vj Mr + Lj Rj Cj (12)
2
(b) (s)
+ k 2 vj λj Lj C0j
The dimensions of the vectors and matrices are dictated by the number of terms Ns used
in the double summation expressions that represent the electromagnetic field. In Eqs.
(10)-(12) the unknown coefficients for each value of j are described by a Ns × 1 vector
and Ms , Mc , Mr are square matrices Ns × Ns , the elements of which are defined by
the following (here k is index)

sin[(pk − ui )c] sin[(pk + ui )c]
Ms [i, k] = p2k ak − (13)
2(pk − ui ) 2(pk + ui )

2 ui cos(ui c) sin[qk (c − hx )] − qk sin(ui c) cos[qk (c − hx )]
+ pk
qk2 − u2i

sin[(pk − ui )c] sin[(pk + ui )c]
Mc [i, k] = ak pk + (14)
2(pk − ui ) 2(pk + ui )

−qk cos(ui c) sin[qk (c − hx )] + ui sin(ui c) cos[qk (c − hx )]
− qk
qk2 − u2i
rk cos(ui c)(−1)k+1
Mr [i, k] = (15)
rk2 − u2i
Finally, L is again a vector Ns × 1 defined by:
sin(ui c)
L[i] = cosh[k(hx − c)] (16)
ui
−ui sin(ui c) cosh[k(hx − c)] + k cos(ui c) sinh[k(hx − c)]
+
k 2 + u2i
Note that all of the above matrices have a common characteristic: they are independent
of the variable vj and therefore they need to be formed just once.
2.2. Impedance Change
The magnetic field in all regions as well as the eddy current density in the conductor can
be calculated from the expressions that relate B and J to W. The general expression for
the impedance change caused by the presence of the conductive edge can be derived by
using a reciprocity relation and written in the following form [2]:
hx hy
−jω ∂φ(s) ∂φ(ec)
ΔZ = φ(ec) − φ(s) dxdy (17)
μ0 I 2 ∂z ∂z z=0
0 0
Substituting from Eqs. (1)-(2) and using Parseval’s theorem for Fourier series gives
∞ ∞
jωhx hy (s) (ec)
ΔZ = − (2 − δi )κij Cij Dij (18)
4μ0 I 2 i=0 j=1
(s) (ec)
where Cij represents the source coefficients characterizing the isolated coil and Dij
represents the reflection coefficients characterizing the contribution of the eddy current
density induced in the right-angled conductor.
3. The Source Coefficient for a Tilted Cylindrical Coil
In addition to the case where the coil axis is normal to a surface of the conductor, as
studied in [1], it is of interest to compute the response of a tilted coil to the presence
of an edge, Figure 2(b). An analytical model for the tilted coil above a conductive half-
space was recently presented in order to study the effect of the tilt angle on the coil’s
impedance and moreover its effect on surface crack inspection signals [4]. In the context
of our analysis, the source coefficient of the tilted coil is
(s) μ0 i0 e−κij d
Cij = 2π (2 − δi ) sin(vj yd ) · (19)
hx hy κij

1 ψ1 l 1 ψ2 l −jui xd
· sin M1 e jui xd
+ 3 sin M2 e
ψ13 2 ψ2 2
where i0 = N I/[(r2 −r1 )l] is the excitation current density with N denoting the number
of wire turns, ψ1 = ui sin ϕ − jκij cos ϕ and ψ2 = ui sin ϕ + jκij cos ϕ and Mi =
ψi r 2
ψi r1 xI1 (x)dx with I1 (x) denoting the modified Bessel function of order 1. When the
lift-off l0 of the coil is known, i.e. the distance of the lowest point of the coil to the
upper conductor surface, the height of the coil center is given by d = l0 + r2 sin (|ϕ|) +
(l/2) cos (ϕ). The tilt angle ϕ is positive for an anti-clockwise rotation. The case of
a cylindrical coil, whose axis is normal to the upper surface of the conductor, can be
derived from Eq. (19) by setting ϕ = 0.
4. Results
Code was written in Mathematica to compute the impedance change of the tilted coil as
it is moved across the edge. The issue to be decided is the extent of the truncated domain
defined by hx and hy and the number of terms in the x and y-summations. Reference to
the case of the half-space conductor was very helpful in this respect. Theoretical results
from the double series expressions were compared to results from the exact double in-
tegral expressions in [4] and it was observed that for hx = hy = n · r2 and Ns = 2n
we obtained an agreement of the order of 1% for all frequencies. Hence, for n = 15,
we only need 30 terms in the series expansions and the square matrices are 30 × 30.
This approach makes the quarter-space model very efficient in terms of numerical im-
plementation since, in addition, the matrices inversions do not depend on coil position.
Thus, the calculations of the whole coil-position scan above the edge (41 points), takes
about a second when using Mathematica in a typical Pentium class PC. The particular
value for n ensures that the coil is always far from the boundaries at x = 0, hx and thus
its impedance it not affected by them. Taking into account that the coil is moved at a
distance of ±20mm from the edge, the choice of n = 15 ensures that the coil is never
located closer than 5r2 to the boundaries.
0.04 0
0.035
0.05
0.03
0.025
−0.1
ΔR/X0
0
ΔX/X
0.02
0.15
0.015
0.01
−0.2
0.005
0 0.25
−20 −10 0 10 20 −20 −10 0 10 20
Coil center position [mm] Coil center position [mm]
−3
x 10
5 0
4.5 0.005
4 −0.01
3.5 0.015
3
ΔX/X0
−0.02
0
ΔR/X
2.5
0.025
2
−0.03
1.5
0.035
1
−0.04
0.5
0.045
0 −20 −10 0 10 20
−20 −10 0 10 20 Coil center position [mm]
Coil center position [mm]
−3
x 10
3.5 0
3
005
2.5
.01
2
0
ΔX/X
0
ΔR/X
1.5 015
1
.02
0.5
025
0 −20 −10 0 10 20
−20 −10 0 10 20 Coil center position [mm]
Coil center position [mm]
Figure 3. Comparison of theoretical (lines) and experimental results (circles) for the normalized resistive and
inductive part of the impedance change as the coil moves across the edge. Excitation frequency is 10kHz. From
page top to page bottom the tilt angle ϕ is 0, 38 and 90 degrees respectively.
Table 1. Test parameters for the numerical computations in Figure 3.
Coil Testpiece Tilt data

r1 7.04 mm σ 25.51 MS/m ϕ 0o 38o 90o
r2 12.20 mm c hx /2 d 4.6 mm 12.17 mm 14.0 mm
l 5.04 mm
N 544
L0 5.55 mH
The theoretical results were also compared to experimental measurements for three
coil orientations, Figure 3. Coil and testpiece data are given in Table 1. In all cases the
coil former is in contact with the conductor upper surface and thus the distance of the
coil center changes with tilt angle. Good agreement is observed for all cases except for
some discrepancy in the resistive part of the tilted coil. This is attributed to the fact that
it was a very small signal and thus it was subject to coil thermal drift since the position
scan was time consuming.
5. Conclusions
An existing model for a coil above a right-angled conductor has been extended in terms
of both numerical implementation and scope. We are now able to formally express the
edge effect for any coil and have given explicit expressions for a cylindrical coil of arbi-
trary orientation. The model is open for still further developments including (i) the use
of closed form expressions for source coefficients for non-cylindrical coils (ii) the cal-
culation of the magnetic field and eddy current density and (iii) the extension to driver
pick-up probes. Furthermore, the tilted coil expressions can be combined with the plate
edge analysis [2] and the through-slot model [3].
Acknowledgements
The authors would like to thank Rob Ditchburn of the Defence Science and Technology
Organization, Australia, for providing the experimental data. This work has been par-
tially funded by the Greek Ministry of National Education & Religious Affairs in the
framework program “Archimedes II: Promotion of Research Groups in TEI”.
References
[1] T.P. Theodoulidis and J.R. Bowler, Eddy current coil interaction with a right-angled conductive wedge,
Proc. R. Soc. Lond. A 461 (2005), 3123–3139.
[2] J.R. Bowler and T.P. Theodoulidis, Coil impedance variation due to induced current at the edge of a
conductive plate, J. Phys. D: Appl. Phys. 39 (2006), 2862–2868.
[3] F. Fu, J.R. Bowler and T.P. Theodoulidis, The effect of opening on eddy current probe response for
an idealized through crack, Review of Progress in Quantitative Nondestructive Evaluation 25 (2005),
330–336.
[4] T.P. Theodoulidis, Analytical model for tilted coils in eddy current nondestructive inspection, IEEE
Trans. Magn. 41 (2005), 2447–2454.
IOS Press, 2007
Volumetric and Surface Flaw Models for

the Computation of the EC T/R Probe
Signal due to a Thin Opening Flaw
Léa MAURICE a,1 , Denis PRÉMEL a and Jozsef PÀVÒ b and
Dominique LESSELIER c and Alain NICOLAS d
a
CEA Saclay, LIST/SYSSC, Bât. 611, 91191 Gif-sur-Yvette, France
b
Budapest University of Technology and Economics, H-1521 Budapest, Hungary
c
L2S-DRE (CNRS-Supélec-UPS), 91192 Gif-sur-Yvette Cedex, France
d
ECL CEGELY, 36 avenue Guy de Collongue, Bât. H9, 69134 Ecully
Abstract This paper is concerned with the dyadic Green formalism in order to de-
velop simulation tools dedicated to Eddy Current Non Destructive Testing (EC-
NDT). The Volume Integral Method (VIM) is useful when considering volumet-
ric flaws. The Surface Integral Method (SIM) is more appropriate for thin open-
ing flaws. This latter fast method provides accurate results, except in some crit-
ical Transmitting / Receiving (T/R) configurations. In such cases, we propose to
combine VIM and SIM to obtain satisfactory results.
Keywords. Eddy current, non destructive testing, ideal crack, CIVA software
1. Introduction
The interaction between eddy currents and a thin crack has been studied by many authors
in the last two decades. It is convenient to predict the EC signal due to the changes in
the impedance of an absolute probe investigating a flawed region by considering an ideal
crack [1].
The “ideal crack model” consists in considering that the crack width is very small
compared to its other dimensions and the skin depth [1]. The surface model leads to
represent the ideal crack by a current dipole surface density, which is a scalar quantity
depending on two spatial variables, in the crack plane; the third spatial variable, along
the thickness of the flaw, disappears. Even if this surface dipole density p is solution of an
integral equation with an hypersingular kernel on the crack surface [2,3], this kernel may
be evaluated in an alternative way in the spectral domain [4]. Some numerical difficulties
coming from specific boundary conditions satisfied by p [5] may be overcome by using
a global approximation [6]. A fast numerical model has been implemented [7], it is very
favorable for the development of a commercial software dedicated to ECT engineers in
probe design or in the goal to perform some parametric studies [8].
1 Corresponding Author: Léa Maurice, CEA Saclay, LIST/SYSSC, Bât. 611, 91191 Gif-sur-Yvette, France.
E-mail: maurice@satie.ens-cachan.fr
50 L. Maurice et al. / Volumetric and Surface Flaw Models
Most of simulation results are in good agreement with experimental results, except
in some critical T/R NDT configurations implying a transmitting and a receiving coil. In
such a situation, we propose to combine the volumetric and surface integral approaches
in order to obtain suitable results.
The paper is organized as follows. A review of the volumetric (VIM) and the surface
(SIM) semi-analytical models based on integral and dyadic formulations is given. Then,
the hybrid method (HybM) is introduced in order to take advantage of each method and
to retrieve more accurate results than those obtained by the surface model (SIM) but
with a computational time smaller than the one required by using the volumetric model
(VIM). In most of applications in NDT, the planar approximation gives quite good results
and the integral formalism using green’s dyads provides a good accuracy and a very short
computational time compared to more general FEM-BEM methods.
2. Description of the Three Models
Let us consider a conducting slab, constituted by an homogeneous non magnetic media

of conductivity σ 0 and permeability μ 0 . The slab is assumed to be infinite in the x and
y directions with a finite depth, and it is affected by a thin crack represented by a planar
defect. A driving time-harmonic current of angular frequency ω and of magnitude I T
is applied to the transmitting coil. The induced primary field is denoted by E P (r). The
current in the receiving coil has a magnitude of I R . The implicit time dependence is
exp(iωt).
2.1. Volumetric Model
The VIM model has shown its efficiency for the prediction of the probe response in
presence of a volumetric flaw [9]. If we call σ(r) its conductivity, this kind of flaw is
described [11] by a fictitious current density P(r) = (σ(r) − σ 0 )E(r), which is solution
of the integral equation:

E(r) = EP (r) + iωμ0 G(r|r ) P(r ) dr (1)
Vd
where G(r|r ) is the Green dyad calculated for a slab of finite thickness, and V d is the
volume of the flaw. According to the reciprocity theorem [10], the probe response is
obtained by:

I · I ΔZ = −
T R
ER (r) · P(r) dr (2)
Vd
where ER is the electric field which would be due to the receiving coil assumed to
operate in the source mode. Since the transmitting and the receiving coils are identical
in the case of an absolute probe, I T = I R and ER = EP , this leads to the usual
formula of impedance [1]. A numerical model has been developed using a Method of
Moments (MoM) decomposition, and a set of pulse testing functions to approximate
L. Maurice et al. / Volumetric and Surface Flaw Models 51
P(r). When using this model to simulate the case of a thin-opening flaw, the number of
cells has to be increased to reach satisfactory accuracy, and this leads to a quite significant
computational time. This is the reason why the SIM has been developed, which embodies
specific assumptions enabling to significantly reduce the computational load overall.
2.2. Surface Integral Model
To develop this dedicated model, an ideal crack is defined [1]: its opening is negligible,
and no current is allowed to flow across it. The defect is assumed to be a void of zero
conductivity : σ(r) = 0. The existence of a scalar potential quantity p(r), defined by
Eq. (3), can be demonstrated [1].
− 1
t (r) − Et (r) = −
E+ ∇t p(r) (3)
σ0
−
where E+ t (r) and Et (r) are the tangential components of E(r) on both sides of the flaw,
and ∇t is the tangential gradient. The scalar surface dipole density p(r) can be related
to the projected part of P(r) on n, the unit vector orientated normal to the surface of
the crack. It takes into account the fact that the idealization of the flaw means that it is
equivalent to a source layer of current dipole orientated along n. Then, it can be shown
that p(r) is solution of Eq. (4):

EP (r0 ) · n = − lim iωμ0 Gnn (r|r ) p(r )dr (4)
r→r0 ∈Sf Sf
where Gnn (r|r ) = n · G(r|r ) · n is the projected dyad and S f is the surface of the flaw.
The probe response is then again given by:

I T · I R ΔZ = − ER (r) · n p(r) dr (5)
Sf
with the same remarks for the case of an absolute probe as previously. A numerical model
has been developed using a MoM decomposition and a global approximation [6] of p(r).
It gives accurate results for the simulation of most NDT configurations in a very short
time. Eq. (5) shows that only the normal component of the primary field contributes to the
probe response, so we propose a hybrid model (HybM) which considers all components
of the primary field. This model is built up in order to be better suited to a thin crack than
VIM and less restrictive than SIM.
2.3. Hybrid Model
Let us assume that we are faced with a thin planar flaw which volume V d that is perfectly
non conducting. This flaw can be represented by a volumetric current density P(r) =
−σ0 · ET (r) where ET (r) is the electric field due to the interaction between the primary
field and the flaw. Let us approximate E T (r) with the electric field of a corresponding
ideal thin crack with surface S f (a mid-cross-section of V d ) as [1]:
Coils C1 C2 C3
Absolute mode × T/R ×
T/R1 mode R T ×
T/R2 mode × T R
Figure 1. The EC probe is constituted by three Figure 2. The functioning mode depends on the
coils. part of each coil (Transmitter or Receiver).

ET (r) = EP (r) + iωμ0 G(r, r ) · n p(r )dr (6)
Sf
Then, the probe response is given by :

I T · I R ΔZ = σ0 ER (r) · ET (r) dr. (7)
Vd
To perform this HybM, the SIM model must be launched in a first step in order to get the
surface dipole density p(r) on S f , then the total electric field E T (r) in Vd is computed
as in Eq. (6) considering a reduced VIM version. Only three dyads are required, instead
of nine for VIM, and the dimensions of each matrix operators amount to N × N 1 with
N = nx × ny × nz and N1 = ny × nz instead of N × N . N is the total number of
discretization cells of the volumetric flaw, n x , ny and nz are respectively the number of
discretization cells in the x, y and z directions. By this arrangement, the "HybM" method
can get more accurate results than SIM, but with a computational time nearly comparable
to the one obtained by SIM.
3. Experimental Validations
Simulated data provided by VIM, SIM and HybM are then compared to experimental
data. Three arrangements of three coils are tested, as illustrated in Figure 1. The table in
Figure 2 summarizes three functioning modes of the probe: the first one corresponds to
an absolute mode (the same coil is transmitting and receiving), the two other configura-
tions assume a receiving coil separated from the transmitting coil. The table in Figure 2
summarizes the modes for each coil. It arises that two different T/R orientations are con-
sidered according as the axis passing by the center of the two coils is parallel or normal
to the length of the flaw. These two configurations are respectively denoted by T/R1 and
T/R2.
Each figure, except specific mention, displays four curves, one for each tested model,
and one for the experimental data. For VIM and HybM, the number of cells assuming
pulse testing functions for the approximation of P is given into parentheses in the cap-
tion, with the format (n x x ny x nz ). In the same manner, the number of global approx-
imating functions [6] is given for SIM. These numbers have been optimized by a set of
numerical experiments.
Figure 3. Real part of the absolute Figure 4. Imaginary part of the abso-
probe response. lute probe response - see Fig. 3.
3.1. Experimental Validations with an Absolute Probe - First Arrangement
The tested specimen consists of a 1.55-mm-thick slab of conductivity σ 0 = 1.02 MS/m,

which is containing a flaw 0.61 mm deep, 4 mm long, and 0.11 mm wide.
The absolute probe is characterized by an inner radius of 1 mm, an outer radius of
1.6 mm, a height of 2 mm, and a number of turns of 328. The real and imaginary parts of
the actual experimental absolute probe response are compared to simulated data obtained
with the three models in Figure 3 and Figure 4, respectively.
We can observe a good agreement between simulated data and experimental data
but simulated data obtained by VIM are closer to experimental data. The value of the
thickness of the flaw and its small size are not favorable factors for the approximation
due to surface current density.
3.2. Experimental Validations with a T/R Probe - Second Arrangement
We now carry out two sets of experimentations, each one involving the same T/R probe
with two different orientations arising from the functional mode of the T/R probe.
The operating coils are characterized by an inner radius of 1.15 mm, an outer radius
of 1.39 mm, a number of turns of 90, a height of 1.2 mm, and a lift-off of 0.1 mm. They
are separated by a distance of d = 6 mm.
We perform tests on two EDM notches in a 1.55 mm-thick slab of inconel 600, with
an opening of 0.1 mm, a length of 7 mm, and respective depths 1.23 mm (80%), for
the so-called "N1" one, and 0.92 mm (60%) for "N2", at two frequencies, 1 MHz and
500 kHz. As experimental data obtained at 1 MHz present a higher magnitude, we prefer
to present these results.
For each method, the signal obtained on "N2" in the "T/R 1" orientation is normal-
ized with a complex value coefficient which is then used for calibration. Figure 5 and
Figure 6 show the real and imaginary parts of the simulated and actual probe response
in the presence of notch "N1" (80%), with the "T/R 1" configuration. In Table 1 are re-
ported in the first line, referred to as "Error", the difference in % on the maximum mag-
nitude between the data obtained with each model and the experimental data. The second
and third lines feature the corresponding computational time required for the 1D scan,
and for a 2D scan including 18 lines and 31 rows respectively, on the same standard PC
(Pentium R, 3.20 GHz, RAM : 512 Mo).
Table 1. Comparative elements on the “N1” / “T/R 1” configuration.

VIM HybM SIM
Error [%] 5 9 11
CPU Time [minutes] 15’ 2’ 1’40
CPU Time (2D) [minutes] 5 × 60’ 4’ 2’
Figure 5. Real part of the probe re- Figure 6. Imaginary part of the probe
sponse for “N1” (1.23 mm depth, 80%) response for “N1” and “T/R1” - see Fig.
and “T/R1” 5
We therefore conclude to a good agreement between the three models used and the
experimental data.
Let us consider now the configuration defined by the "N1" notch, with the probe in
the "T/R 2" orientation. The real and imaginary parts of the results are given in Figure 7
and Figure 8. A first feature of these results is that the magnitudes of the signals are more
than 10 times smaller than those obtained in the previous configuration. That is the reason
Figure 7. Real part of the probe response for “N1” Figure 8. Imaginary part of the probe response for
(1.23 mm depth, 80%) and “T/R2” “N1” and “T/R2” - see Fig. 7.
why actual testing is seldom performed in this way. The differences on the magnitude
for the line scanning between the data computed with each model and experimental data
are reported in Table 2.
Table 2. Comparative elements on the “N1” / “T/R 2” configuration.

VIM HybM SIM
Error [%] 16 4 17
CPU Time [minutes] 9’ 2’ 1’15
The VIM signal underestimates the other results with a cumulative magnitude of
0.1031 mV against 0.1173 and 0.1222 for HybM and experimental data, respectively.
This underestimation often occurs when the number of cells is not sufficiently large. The
increasing of the number of cells provides a higher computational time. However, an
overestimation of SIM (0.1424 mV of magnitude) can be noticed for all tested cases with
this “T/R 2” configuration. Moreover, we sometimes observe a real shape difference (see
e.g. Figure 9). In all such cases, HybM enables to retrieve more suitable results, as illus-
trated here. Although the “T/R 2” configuration is not a favorable NDT configuration, it
turns out to be a critical configuration for SIM, because of too restrictive assumptions,
and it exemplifies the needed corrections brought by HybM by taking the total electrical
field ET into account for the computation of the probe response.
The second line of Table 2 features the computational times required to compute the
1D data set with each method.
We observe the same behaviour, regarding the shape of the signals and the compu-
tational time, at a frequency of 500 kHz.
4. Conclusive Remarks and Open Questions
The already existing VIM and SIM models provide two means of predicting the electro-
magnetic interactions with a thin flaw. VIM has been developed for general volumetric
flaws, whereas SIM is based on assumptions coming from the ideal crack. These assump-
tions turn out to be too restrictive in some T/R configurations. However the main benefit
of SIM remains its fast computation, while VIM requires a high computational time and
a high memory space to simulate the thin flaw case.
The new developed hybrid model gives quite good results for volumetric flaws as
well as for planar defects. This model enables to join VIM and SIM. The thickness of
the flaw is better taken into account by HybM rather than by SIM.
Let us consider the particular arrangement when the probe is constituted by two
identical coils functioning in the T/R2 configuration. One interesting question is what
happens when the distance between the two coils is decreasing comparatively to the
diameter of the coils.
Figure 9 displays some simulated results obtained by the three methods VIM, SIM
and HybM. One can observe a real shape discrepancy between these results. We do not
observe such a shape discrepancy for the other T/R1 functioning mode. Now, if we con-
sider an increasing of the distance separating the coils, denoted by d, the shape discrep-
ancy does not come out. Figure 10 shows some results considering a distance which
reaches a value of 7 mm. The results obtained with the three models are rather satisfying.
Therefore, we can observe that above the SIM model can not “see” the effect of the two
coils as it should, if we refer to the results yielded by other models otherwise validated
by experimental experiments.
The HybM model that we propose takes advantage of the SIM and the VIM and
it provides a suitable correction for the T/R2 configuration. HybM seems to be more
reliable than the SIM model as it provides accurate results close to the experimental data,
for all configurations, even for those appear to be critical for SIM.
Moreover, the CPU time is almost similar than the one we obtain by SIM because a
fewer number of discretization cells is required compared to VIM.
Figure 9. Magnitude of the e.m.f. Figure 10. Magnitude of the e.m.f.

yielded by the SIM for a distance of yielded by the SIM for a distance of
4 mm between the coils. 7 mm between the coils.
References
[1] J. R. Bowler, “Eddy-current interaction with an ideal crack. I. The forward problem,” J. Appl. Phys.,
Vol. 75, no 12, pp. 8128-8137, 1994.
[2] P. Beltrame and N. Burais, “Computing methods of hypersingular integral applied to eddy-current test-
ing,” IEEE Trans. Magn., Vol. 38, no. 2, pp. 1269-1272, 2002.
[3] P. Beltrame and N. Burais, “Application of regularization method of quasi-singular integrals to compute
eddy-current distribution near cracks,” Int. J. Comp. Math. Engng, Vol. 21, no. 4, pp. 519–533, 2002.
[4] J. Pávó and K. Miya, “Reconstruction of crack shape by optimization using eddy current field measure-
ment,” IEEE Trans. Magn., Vol. 30, no. 5, pp. 3407-3410, 1994.
[5] J. R. Bowler, Y. Yoshida and N. Harfield, ”Vector potential boundary integral evaluation of eddy current
interaction with a crack,“ IEEE Trans. Magn., Vol. 33, no. 5, pp. 4287-4294, 1997.
[6] J. Pávó and D. Lesselier, “Calculation of eddy current testing probe signal with global approximation,”
IEEE Trans. Magn., Vol. 42, no. 4, pp. 1419-1422, 2006.
[7] J. Pávó, D. Prémel, and D. Lesselier, “Application of volumetric and surface defect models for the anal-
ysis of eddy current non-destructive testing problems,” URSI International Symposium on Electromag-
netic Theory, May 2004, Pisa, Italy, pp. 400-402.
[8] Y. Deng, X. Liu, Z. Zeng, L. Udpa, W. Shih, and G. Fitzpatrick, “Numerical studies of magneto-optic
imaging for probability of detection calculations,” Electromagnetic Nondestructive Evaluation, Vol. IX,
L. Udpa and N. Bowler (eds), IOS Press, pp. 33-40, 2005.
[9] P. Calmon and D. Prémel, “Integrated NDT models in CIVA,” IInd international Workshop of NDT
Experts, 2003.
[10] B. A. Auld and J. C. Moulder, “Review of advances in quantitative eddy current non destructive evalua-
tion,” J. Nondestruct. Eval., Vol. 18, no. 1, pp. 3-10, 1999.
[11] J. R. Bowler, S. A. Jenkins, L. D. Sabbagh, and H. A. Sabbagh, “Eddy-Current probe impedance due to
a volumetric flaw,” J. Appl. Phys., Vol. 70, no. 3, pp. 1107-1114, 1991.
IOS Press, 2007
Application of Eigenfunction Expansions

to Eddy Current NDE:
A Model of Cup-Cored Probes
Hossein BAYANI a,1, Theodoros THEODOULIDIS b and Ichiro SASADA a
a
Dept. of Applied Science for Electronics and Materials, Kyushu University, Japan
b
Dept. of Engineering and Management of Energy Resources, University of West
Macedonia, Greece
Abstract. An axisymmetric cup-cored coil placed above a layered conductive half-

space is analyzed by using the truncated region eigenfunction expansion method.
Closed-form expressions are presented for the coil impedance as well as the
induced eddy current density. The results are in very good agreement with results
from 2D-FEM and with experimental ones.
Keywords. Cup-cored Probe, Eddy current, Eigenfunction Expansion.
1. Introduction
In order to decrease an eddy current probe’s magnetic reluctance one must wind it
either on or inside ferrite cores. The latter reduces the leakage field from the test area
too. In this case, where the Dodd model for air-cored coils [1], cannot be utilized
numerical models have been used instead [2, 3]. Nevertheless, a closed-form
expression can be derived by using the Truncated Region Eigenfunction Expansion
(TREE) method [4] which involves a modification of the solution domain in order to
replace the integral expressions with more convenient series ones. In this paper, the
model is extended to the case of a cup-cored probe as shown in Figure 1. The probe
consists of a circular coil of rectangular cross section confined coaxially by a cup-cored
ferrite, and is located above a layered half-space of conductive material. The cup-cored
probe not only gives a much higher flux density but it also shields the flux into the test
area thus producing stronger signals when it comes to crack inspections. As in the
classical approach, the method uses separation of variables to express the
electromagnetic field in the various regions of the problem in an analytical form. It
differs, however, from the classical approach in the truncation of the solution domain in
order to limit the range of a coordinate that would otherwise have an infinite span. As a
result, the solution dependence on the coordinate is expressed as a series form, rather
than as an integral. In this case the numerical implementation is usually more efficient
and the error control is easier.
1
Corresponding Author: Hossein Bayani, Dept. of Applied Science for Electronics and Materials,
Interdisciplinary Graduate School of Engineering Sciences, Kyushu University, Kasuga-Koen 6-1,
Fukuoka, 816-8580, Japan; E-mail: hbayani@yahoo.com
58 H. Bayani et al. / Application of Eigenfunction Expansions to Eddy Current NDE
Figure 1. A cup-cored coil of finite cross section above a layered conductive half-space.
In this paper truncation of the problem region means that the cylindrical surface
r=t has now become the outer boundary. On this boundary we impose a homogeneous
Dirichlet condition for the magnetic vector potential although a homogeneous
Neumann can also be used. The magnetic cup-core is treated as a homogeneous and
isotropic region having relative magnetic permeability μf and together with the outer
boundary these are about the only additional assumptions that have to be made for this
model compared to the Dodd and Deeds models. The solution proceeds as follows: The
only component of the vector potential (azimuthal) is expressed in the form of a series
of orthogonal eigenfunctions involving discrete eigenvalues. These eigenvalues as well
as the series coefficients are computed by imposing the continuity conditions on the
various boundaries and interfaces of the solution. In the vertical boundaries the
continuity is imposed in a term by term manner while in the horizontal boundaries this
is done using mode matching. The whole analysis is very lengthy and will not be
repeated. The reader is referred to [4] for details of the approach. Here we will provide
the final expressions for the coil impedance and the induced eddy current densities
together with all other expressions necessary for their computation.
2. Solution
In the beginning we solve the electromagnetic field problem for a delta-function coil at
(r0, h) as in Figure 2, driven by a harmonic current I exp( jZt ) . The coil is located
above a nonmagnetic conductive half space comprising two layers with conductivities
ı6 and ı7. The plane z = 0, coincides with the bottom of the ferrite core. Following the
separation of variables, the expressions for Aĳ in the various regions of the problem
have the following form which in addition is given in matrix notation:
T 1 qz
A1 ( r , z ) J 1 (q r )q e C1 ; 0dr dt, (1)
H. Bayani et al. / Application of Eigenfunction Expansions to Eddy Current NDE 59
Figure 2. A cup-cored coil of delta-function circular current above a layered conductive half-space.
T
J 1 (m r ) 1 mz mz
0drdb
A2 ( r , z ) T
m (e C2 e B 2 ) ; , (2)
L1 (m r ) bdrdt
T
J 1 (p r ) 0drda
T 1 pz pz
A3 ( r , z ) R1 (p r ) p (e C3 e B 3 ) ; a d r d b, (3)
R1c(p r )
T
bdrdt
T
J 1 (p r ) 0drda
T 1 pz pz
A4 ( r , z ) R1 (p r ) p (e C 4 e B 4 ) ; a d r d b, (4)
R1c(p r )
T
bdrdt
T 1 qz qz
A5 ( r , z ) J 1 (q r )q ( e C 5 e B 5 ) ; 0dr dt, (5)
T 1 sz sz
A6 ( r , z ) J 1 (q r )s ( e C 6 e B 6 ) ; 0dr dt, (6)
T 1 uz
A7 ( r , z ) J 1 (q r )u e B 7 ; 0dr dt, (7)
where si qi2 jZP0V 6 and ui qi2 jZP0V 7 , Jn, Yn are Bessel functions of
order n, Ln, Rn, R’n are defined next, the superscript T denotes a row vector, p-1, q-1, m-1,
s-1, u-1 and exponentials are diagonal matrices and C, B are unknown vector
coefficients. These unknown coefficients and the discrete eigenvalues are to be
determined from the boundary and interface conditions. The eigenvalues for each
region of Figure 2 are defined as follows:
For regions 1, 5, 6, 7 the qi are the positive real roots of the equation:
J 1 ( qi t ) 0. (8)
For region 2 the mi are the positive real roots of the equation:
L1 ( mi t ) 0, (9)
where
Ln ( mi r ) B2 F J n ( mi r ) C2 F Yn ( mi r ) , (10)
S mi b ª J 0 ( mi b )Y1 ( mi b ) º
B2 F « J1 ( mi b)Y0 ( mi b) », (11)
2 ¬ Pf ¼
S mi b ª J 0 ( mi b ) J 1 ( mi b ) º
C2 F « J 1 ( mi b) J 0 ( mi b) ». (12)
2 ¬ Pf ¼
For regions 3 and 4 the pi are the positive real roots of the equation:
R1c( pi t ) 0, (13)
where
Rnc ( pi r ) B3ca J n ( pi r ) C3caYn ( pi r ) , (14)
S pi b ª R0 ( pi b)Y1 ( mi b) º
B3ca « R1 ( pi b)Y0 ( pi b) », (15)
2 ¬ Pf ¼
S pi b ª R0 ( pi b) J 1 ( mi b) º
C3ca « R1 ( pi b) J 0 ( pi b) », (16)
2 ¬ Pf ¼
Rn ( pi r ) B3 a J n ( pi r ) C3 aYn ( pi r ) , (17)
S pi a
B3 a > J ( p a )Y ( p a) P
1 i 0 i f @
J 0 ( pi a )Y1 ( pi a ) , (18)
2
S pi a
C3 a > J ( p a) J ( p a) P
1 i 0 i f @
J 0 ( pi a ) J 1 ( pi a ) . (19)
2
The final expressions for the eddy current densities in regions 6, 7 as well as the
impedance of the coil are:
J 6eddy ( r , z ) jZV J1 (qT r ) s 1 (e s z C 67 e s z B67 )

1 ,(20)
ª¬ (T U )e md C 27 (T U )e md B27 º¼ ª¬ (T U )e md L2 (T U )e md L1 º¼
1 1 1 1
J 7eddy ( r , z ) jZV J1 (qT r ) u 1e u z

1 , (21)
ª¬ (T U )e md C 27 (T U )e md B27 º¼ ª¬ (T U )e md L2 (T U )e md L1 º¼
1 1 1 1
2
jZP0S N
Z 2 2
F ( pT r1 , pT r2 ) p
4
( r2 r1 ) ( h2 h1 )
^ª¬(e p h1
e
p h2
)C 47 (e
p h2
e
p h1
) B47 ) º¼
1
ª¬(T U )e md C 27 (T U )e md B27 º¼
1 1 ,(22)
(T U )e md 1 e md F 1 ª ( H G )e pd (e ph e ph ) ( H G )e pd (e ph e ph ) º ½
1 2 2 2 1 2 1 2
°° 2
¬ ¼ °°
® ¾
°(T U )e md 1 e md F 1 ª( H G )e pd (e ph e ph ) ( H G )e pd (e ph e ph ) º °
1 2 2 2 1 2 1 2
°¯ 2
¬ ¼ °¿
[2( h2 h1 ) p e p h2 e p h1 e p h2 e p h1 ] D 1 p 3 F ( pr1 , pr2 ) `
where
L1 P0 i0 ª( H r G )e pd2 (e ph2 e ph1 ) º 3 1

e r md2 F -1 « pd 2 ph1 ph2 » p D F ( pr1 , pr2 ) , (23)
L2 4 ¬ ( H # G )e (e e )¼
pr2
F ( pr1 , pr2 ) ³ pr1
x J 1 ( x ) dx , (24)
B67 1 r sy 2 uy 2
e (1 r su 1 )e , (25)
C 67 2
B57 1 r qy1 sy1

[(1 # qs 1 )e C 67 (1 r qs 1 )e
sy1
e B67 ] , (26)
C 57 2
B47 1
D 1 > ( H c # G c)C 57 ( H c r G c) B57 @ , (27)
C 47 2
B27 1 # md2 pd 2
F 1 ª¬ ( H # G )e
pd 2
e C 47 ( H r G )e B47 º¼ , (28)
C 27 2
and all other matrices are defined in the Appendix. Many special cases can be obtained
by the general expression for the impedance change (22). For example for a half-space
conductor we can either set V6 =V7 or y2-y1 or y2ĺ. For an isolated cup-core probe we
can set V6 =V7 = 0. For an air-core probe we can set μf =1.
3. Results
We conducted a series of experiments besides applying 2D-FEM package for two cases
of cup-cored and air-cored probe in order to test the validity of our solutions. In the
experiments, we scanned a range of frequencies between 100 Hz to 100 kHz with an
Agilent HP4284 impedance analyzer. The experimental parameters used in the
calculations are given in Table 1. First, we calculated the impedance for a conductive
plate to find the normalized impedance plane and then we conducted the first
experiment to compare the measurements to the theoretical result with the result of the
calculations. We also applied the parameters used in the calculation to a 2D-FEM
package. The obtained results are shown in Figure 3. In the second experiment we
calculated the impedance based on the special case of the air-cored coil, to find the
normalized impedance plane as well as the Dodd model. The results are depicted in
Figure 4. For the ferrite-cored coil, the inductance L0 is calculated 21.48 mH, while the
measured value is 21.59 mH. For the air-cored coil, the relative values are 18.93 mH
calculated and 18.78 mH measured and 19.01mH based on the Dodd and Deeds model.
The calculations are carried out by selecting t to be 6 times the outer radius of the cup-
cored ferrite. In the case of cup-cored we used Ns=33 and for the case of air-cored we
used Ns=45. In all cases the relative error between theoretical results and measurements
is less than 1.22%, which shows a very good agreement. Even better agreement can be
achieved by increasing t and Ns. In Figure 5 we depict the real part of the eddy current
density computed by using the TREE method and a 2D-FEM package and the results
are also in very good agreement.
1.0
1.0
0.9
TREE
TREE
f EXP
X / X0
X / X0
EXP f
DODD
FEM
0.9 0.8
Cup-cored Coil Air-cored Coil

Pf == 2300
Pf == 11

F 2300 0.7 F
0.01 0.02 0.03 0.04 0.02 0.04 0.06 0.08
(R-R0) / X0 (R-R0) / X0
Figure 3. Impedance plane diagram showing Figure 4. Impedance plane diagram showing
variation of normalized Z with frequency for a variation of normalized Z with frequency for an
cup-cored coil above a conductive 5-mm-thick air-cored coil above a conductive 5-mm-thick
plate. X0 stands for coil’s isolated reactance. plate. X0 stands for coil’s isolated reactance.
Eddy -current Density ( MA/ m )

2
FEM LAB
TREE
-1
-2
-3
0 5 10 15 20 25
R adial D istance ( m m )
Figure 5. Real part of eddy-current density amplitude at 2 mm below the conductor surface.
Table 1. Coil, cup core, and plate parameters used in calculations and experiments.
Coil Cup core Aluminum plate
Inner radius r1 4.64 mm Inner radius a 7 mm Thickness (y2-y1) 5 mm

Outer radius r2 6.5 mm Outer radius b 10 mm Conductivity V6 35.36 MSm-1
Offset h1 0.74 mm Rel. permeability μf 2300 Conductivity V7 0 MSm-1
Length (h2 - h1) 2.9 mm Liftoff y1 0.82 mm
Number of turns 1240
4. Conclusion
Important magnetic induction quantities such as impedance and eddy current density
are derived by using the method of eigenfunction expansion for the case of an
axisymmetric cup-cored coil above a layered half-space. Comparison between the
results of the experiments and 2D-FEM package with the results of our solution
showed a very good agreement. The proposed method is extremely fast and it can also
be used to model a cup-cored coil above a half-space having an axisymmetric hole, or
can be used to solve the problem of an E-cored probe.
Appendix
The matrices E, T, U, F, G, H, D, G`, and H` are computed as
0 iz j
ij °
E ®t 2
2
, (A1)
°¯ J ( q t ) i j
0 i
2
ij bqi § · 1
T 2 2
J 1 ( qi b ) J 0 ( m j b ) ¨ 1 ¸, (A2)
qi m j © P ¹ f
ij bqi § · 1
U 2 2
J 0 ( qi b ) J 1 ( m j b ) ¨ 1 ¸, (A3)
qi m j © P ¹ f
0 iz j
ij °
b ª J ( m b) º§ 1 · , (A4)
2 2 2
F ®t 2 0 i 2
° 2 L (m t ) 2 « P 0
J (m b) » ¨ 1
i ¸ i j 1 i
¯ ¬ ¼© P ¹ f f
ij ami § 1 ·
G 2 2
J 1 ( mi a ) J 0 ( p j a ) ¨ 1¸ , (A5)
mi p j ©P f ¹
ij ami § 1 ·
H 2 2
J 0 ( mi a ) J 1 ( p j a ) ¨ 1¸ , (A6)
mi p j ©P f ¹
i t
2
a §2
1 ·
R0c ( pi t )
2 2 2
D ¨1 ¸ ª¬ J ( p a ) J ( p a ) P º¼
1 i 0 i f
2 2 © P f ¹
, (A7)
§b 1 · ª R ( p b)
2
º 0
2
i 2
¨1 ¸« R ( p b) » 1 i
2 © P ¹¬ P ¼ f f
§ 1 · pi
¨ 1 ¸ > aJ ( p a ) J ( q a ) bR ( p b ) J ( q b ) @ ,
ij
Gc 2 2 1 i 0 j 1 i 0 j
(A8)
p q © i
P ¹ j f
§ 1 ·
pi
¨ 1 ¸ > a P J ( p a ) J ( q a ) bR ( p b) J ( q b ) @ ,
ij
Hc 2 2 f 0 i 1 j 0 i 1 j
(A9)
p q © i
P ¹ j f
where the matrices T, U, G, H, G`, and H` are full and the matrices E, F, and D are
diagonal.
References
[1] C.V. Dodd and W.E. Deeds, Analytical solutions to eddy-current probe-coil problems, Journal of
Applied Physics 39 (1968), 2829-2838.
[2] H.A. Sabbagh, A model of eddy-current probes with ferrite cores, IEEE Trans. Magn. 23 (1987), 1888-
1904.
[3] F. Buvat, G. Pichenot, D. Lesselier, M. Lambert and H. Voillaume, A fast model of eddy-current ferrite-
cored probes for NDE, Electromagnetic Nondestructive Evaluation (VIII), IOS Press (2004), 44-51.
[4] T.P. Theodoulidis, Model of ferrite-cored probes for eddy current nondestructive evaluation, Journal of
Applied Physics 93 (2003), 3071-3078.
Eddy Current Testing and Technique
IOS Press, 2007
Experimental Extraction of Time-of-Flight

from Eddy Current Test Data
Antonello TAMBURRINOa,1, Naveen NAIRb, Satish UDPAb, Lalita UDPAb

a
Associazione EURATOM/ENEA/CREATE, DAEIMI, Universita’ di Cassino, Italy
b
Electrical and Computer Engineering, Michigan State University, USA
Abstract. This paper presents a contribution in the framework of a novel method

to extract defect location in conductive materials using data obtained from
diffusive nondestructive evaluation (NDE) techniques such as eddy current testing.
The method is based on the definition and extraction of the Time of Flight (TOF)
for diffusive phenomena by using the Q-transform, a mapping that associates a
wave propagation problem to a diffusion problem. In this work we present the
method for extracting the TOF from diffusive data together with experimental and
numerical tests for a simple configuration. The results clearly demonstrate the
application of Q-Transform based approach in a realistic NDE setting.
Keywords. Eddy current testing, time-of-flight, Q-Transform
1. Introduction
The Q transform, first defined in [1]-[4] as
Q : u ( x, q ) → v ( x, t ) = ( 4π t 3 ) q exp ( −q 2 / 4t ) u ( x, q ) dq
−1/ 2 +∞
∫
0
(1)
relates the solution of a wave equation to the solution of a corresponding diffusion

equation. It has been shown to be useful to properly associate the time of flight to
diffusion domain data [5]. For wave problems, the TOF is a clearly defined quantity
and so, in the light of the Q-Transform, it is reasonable to define the time of flight for a
diffusion problem as the (usual) time of flight for the corresponding fictitious wave
problem. In particular the extraction of the TOF from the diffusion data can be
approached either through an inverse Q-transform [5, 6] or by a careful choice of the
excitation waveforms [7-9]. The inverse Q-transform based method involves the
evaluation of a Fredholm integral of the first kind and therefore requires regularization
techniques. On the contrary, the freedom in the choice of the excitation waveform can
be exploited to relate the TOF to an easily measurable quantity in the measured time-
domain signal. Specifically, in past work we related the TOF to the peak position of the
1
Corresponding Author: Antonello Tamburrino, DAEIMI, Università degli Studi di Cassino; E-mail:
tamburrino@unicas.it. Antonello Tamburrino is also with Electrical and Computer Engineering,
Michigan State University.
68 A. Tamburrino et al. / Experimental Extraction of Time-of-Flight from Eddy Current Test Data
measured signal under proper conditions. In this work we will present a new method
that is more general than the one based on the peak position, together with numerical
results validating the proposed approach and moreover, an experimental setup designed
to test the method on a simple canonical problem.
The organization of the paper is as follows. The next section will discuss the Q-
transform and the idea behind the extraction of the time of flight. The details of the
numerical and experimental setup will be provided in section 3, whereas section 4 will
attempt to draw some conclusions and point out directions for possible future work.
2. The Q-Transform for TOF Extraction
2.1. A Relationship between Parabolic and Hyperbolic Differential Equations
Consider the following two scalar initial value problems defined in Ω⊆ℜN
∇ 2 v (x, t ) − k (x) ∂v(x, t ) / ∂t = F (x, t ) in Ω × ( 0, +∞ )

a (x)v(x, t ) + b(x) ∂v(x, t ) / ∂n = G (x, t ) on ∂Ω × ( 0, +∞ ) (2)
v(x, 0) = h(x) in Ω,t = 0
and
∇ 2 u (x, q) − k (x)∂ 2 u (x, q ) / ∂q 2 = f (x, q) in Ω × ( 0, +∞ )

a (x)u (x, q ) + b(x)∂u (x, q ) / ∂n = g (x, q ) on ∂Ω × ( 0, +∞ )
(3)
u (x, 0) = 0 in Ω
∂u (x, 0) / ∂q = h(x) in Ω
where ∂ ∂n represents the normal derivative w.r.t. the spatial co-ordinate x. Then the
Q-transform relates the solutions of the two problems above in the following manner
[1]-[4]
F = Qf and G = Qg ⇒ v = Qu (4)
For instance, for a scalar (magneto-quasi-static) diffusion problem in a non-

magnetic conductor (Eddy Current Testing) k (x ) = μ 0σ (x ) ( μ 0 is the magnetic
permeability and σ is the electrical conductivity) and v is a magnetic flux density
component. The “wave velocity” for the fictitious field u is, therefore, given by
c(x ) = 1 / k (x ) = 1 / μ 0σ (x ) .
Eq. (4) provides the Q transform relationship for the scalar case. For vector
electromagnetic equations, the relationship can be expressed as described in [9, 10].
Once the connection between a diffusion and a wave propagation problem has been
established (see (4)), the TOF for a diffusive measurement v x= x in the time-domain is
0
A. Tamburrino et al. / Experimental Extraction of Time-of-Flight from Eddy Current Test Data 69
defined as the TOF for the waveform u x = x = Q −1 v x =x

0
[ ] corresponding
0
to the
associated fictitious wave propagation problem. As discussed in the introduction, this
definition of TOF is not practical to be applied to experimental data because of the
noise affecting the measured and the ill-posedness of the problem of computing the
inverse Q-Transform.
However, the problem can be cast in an equivalent form. Let qTOF be the TOF
associated to the waveform u x= x and let qi be the (known) instant when the field
0
source is turned on. Therefore, u x= x is vanishing for q<q0 where q0= qi+qTOF; in
0
addition we assume that u x= x at q 0+ is different from zero. It is worth noting that,

0
regardless the particular setting, qTOF is related to the optical path between source and
receiver. For instance, in an homogeneous media qTOF=l/cσ where l is the length of the
optical path.
Omitting the explicit dependence of v and u on the spatial position x0, the problem
of extracting the TOF qTOF = q0−qi can be cast as:
find q0 such that v = Q[u ] where v is given and u is vanishing for q<q0 and non-
zero for q>q0.
The extraction of the TOF relies on the following properties (see [9]) of the Q-
Transform valid for a waveform u that (i) is vanishing for q<q0 and (ii) satisfies
( )
u (q ) exp − q 2 / 4t → 0 for q → +∞ :
1 ⎡ ⎤
2
q q2
− 0 ∞ −
v(t ) := Q [u (q) ] = ⎢u (q0 )e + ∫q+ u '(q)e 4t dq ⎥
+ 4t
(5)
π t ⎢⎣ 0
⎥⎦
that can easily be proven by integration by parts applied to (1). In (5) u (q0 + ) is the
limit from right of u evaluated at q0.
2.2. TOF from Small-time Waveform
The idea behind the TOF extraction is extremely simple and is based on the fact that the
integral at the r.h.s. of (5) is rapidly vanishing for t→0 and the first addend (at the r.h.s.
of (5)) depend in an explicit way upon q0. Here we prove that for small t the integral at
the r.h.s. of (5) can be neglected. Specifically, under the rather ‘mild’ assumption that
u’ is bounded for q>q0, it is also possible to find an upper bound to the integral at the
r.h.s. of (5) for small t. Indeed, if u ' ≤ M for q>q0, then
q2 q2
1 ∞ − M ∞ − ⎛ q ⎞
πt ∫ πt ∫
u '(q )e 4t
dq ≤ e 4t
dq = Merfc ⎜ 0 ⎟ (6)
⎝2 t ⎠
q0 q0
where erfc(⋅) is the well known complementary error function. For large arguments the
complementary error function can be approximated as
( )
−1
erfc ( x ) = x π e − x ⎡⎣1 + O ( x −2 ) ⎤⎦ for x → +∞ (see [11]).
2
Thanks to (6), it results that for small t, v(t) can be approximated by

u (q0 ) exp ( − q02 / 4t ) / π t . The error in this approximation is bounded by:
+
v(t ) −
( )e
u q 0+ −
q02
4t
⎛ q
≤ Merfc⎜⎜ 0
⎞ 2M t − q40t ⎡
⎟≅
2
⎛ t ⎞⎤
e ⎢1 + O⎜⎜ 2 ⎟⎟⎥ (7)
⎟
πt ⎝2 t ⎠ q0 π ⎢⎣ ⎝ q 0 ⎠⎦⎥
therefore, the error is negligible with respect to u (q0 + ) exp ( − q02 / 4t ) / π t for t small
( ( ) )
enough t << u q 0+ / (M / q 0 ) . Once the (small time) representation for the measured
quantity v is known in a parametric form depending explicitly on the TOF q0, its
extraction can be carried out by a best fit. Specifically, the estimate of q0 follows from
the following minimization problem:
t*
exp ( −q02 / 4t ) ⎤ dt
2
min ∫ ⎡v(t ) − a0 (π t )
−1/ 2
(8)
a0 , q0 ⎣ ⎦
0
where t* is small “enough” so that v can be appropriately approximated by a function of

the type a0 exp ( −q02 / 4t ) / π t . Let a 0 (q0 , v ) be the (least-square) solution of (8) for
fixed q0; then q0 can be obtained by the one-dimensional minimization of
t*
[
Ψ (q0 ) = ∫ v(t ) − a0 (q0 , v )exp − q02 / 4t / πt dt .
0
( ) ]2
As final problem, we have to find the condition such that the waveform u for the
fictitious wave propagation problem presents a TOF, i.e. is vanishing for q<q0. This is
the case if the source for the fictitious problem is, for instance, of the type H (q − qi )
where H is the Heaviside function and qi is the fictitious q-time when the source is
switched on. From (4) it follows that the waveform of the source for the diffusive
problem must be Q[H (q − qi )] = exp − qi2 / 4t / πt at least for t<t*. ( )
2.3. TOF from Peak Location
Another technique for extracting the Time of Flight from the peak location of the
received signal has been detailed in [7-9]. This essentially uses the argument that for
signals whose q-domain is vanishing for q<q0 and having a bounded first derivative for
q<q0, the peak location of the received waveform will have a direct correspondence to
the time of flight. In particular, the time of flight ( q0 ) is given by q0 = α tmax where α
is an appropriate constant depending on the driving waveform. This approach can be
proven to be less general with respect to the approach based on small time analysis.
2.4. TOF and Standard Pulsed Eddy Current Testing Experiment.
Typical PEC (pulsed eddy current testing) experiments on conductive material often
rely on the time to peak (that is the time when the measured response shows a peak) and
peak height. Time to peak is not comparable to TOF because it is not strictly related to
geometrical distance as the TOF we introduced. This is due to the fact that the group
velocity depend on the frequency of the signal considered. Moreover, time to peak
depend on the shape of the driving waveform, whereas the TOF introduced in section
2.2 is a unique feature of the measured waveform that holds under an extremely weak
condition that is u ' bounded for q>q0. The TOF introduced by means of the Q-
transform has the unique property to be related to geometrical distances (it is
proportional to the length of the optical path between source and receiver) because is
the TOF for an appropriate wave propagation problem.
3. Experimental Setup
In this section we present a simple configuration used to test experimentally the idea of
extracting the TOF, from eddy current testing measurements, by processing the small
time response as described in section 2.2. In the test case the TOF is related to the
thickness of a conductive slab as described in the following section. Experimental tests
to locate, by means of the proposed approach, a small anomaly in the slab are beyond
the scope of this paper and are currently in progress.
3.1. Test Geometry
The reference geometry is described in figure 1. A metallic plate is kept over two
parallel line sources and a probe is located on the top. The excitation is provided by two
parallel line conductors located at ± x0 and carrying currents ± I o zˆ that induce eddy
currents in the sample. The length of the current carrying wires and the width of the
sample is large enough, in comparison to its thickness to justify a 2-dimensional
assumption for the problem. The wires generate a magnetic field in the y-direction and
this is measured at the location x=0 using a GMR based magnetic flux density sensor.
Therefore, v(t ) = B y (x GMR , t ) where x GMR = (0, y GMR ) is the position of the sensor.
As mentioned before, the q-domain wave velocity is inversely related to the
conductivity of the medium ( c(x ) = 1 / μ 0σ (x ) , see section 2). In particular, in air, the
q-domain velocity is infinite and, therefore, it can be easily shown that an optical path
crosses the plate perpendicularly. Consequently, the TOF is calculated as qTOF=h/cσ
where h is the thickness of the plate and cσ is the q-domain velocity in the slab, given
by cσ=(μ0σ0)-1/2 where, in turn, μ0 and σ0 are the permeability and conductivity of the
material, respectively. It is worth noting that the part of the optical path in air gives no
contribution because c(x) in air is infinite. In the present case cσ = 0.176 m/s1/2
( μ 0 = 4π ⋅10−7 H/m and σ 0 = 2.57 ⋅107 S/m ), qTOF = 0.0144s1/2 for the single plate
configuration and qTOF = 0.0288s1/2 for the two plates configuration.
y y
GMR
2.5mm
sensor
x x
15mm
Source
Figure 1. Test Geometry. Left : single plate. Right : two plates.
1 1.2
1
0.8
0.8
0.6 0.6
a.u.
a.u.
0.4 0.4
0.2
0.2
0
0 -0.2
0 0.5 1 1.5 2 0 0.5 1 1.5 2
t (ms) t (ms)
Figure 2. The measured waveform of the driving current (solid) is almost superimposed to the related
analytical expression (dotted). The waveform of the measured quantity v(t ) = By (xGMR , t ) (dashed). Left:
single plate case. Right: double plate case.
The validation experiment is carried out with two plates of known thickness. First
one plate is placed between the source and the sensor and the TOF measured. Then, a
second plate is kept on the top of the first. The TOF should ideally be twice the value
measured in the first case. The excitation waveform used in the experiment is shown in
figure 2 below. The signal is designed to meet the criteria set forth in section 2. It is
important to be noted here that, again, since the conductivity of air is low enough, the
measured TOF can be assumed insensitive to the air gap between the two plates. The
experiment was carried out using currents of 6A peak in the two wires.
3.2. Numerical Simulation and Experimental Results
To test the validity of the experimental data, some initial numerical simulations were
carried out using FEMLAB®. The geometry used for simulation is identical to the
figures shown in figure 1 above. The source waveform was chosen as the Q-transform
of a Heaviside step function. A normalized plot of the input signal is shown in figure 2.
(
Specifically, the input waveform is of the type Q[H (q − qi )] = exp − qi2 / 4t / πt for )
small t ( t ≤ 3ms ) and then is extended to zero in the interval 3ms ≤ t ≤ 8ms . For
t ≥ 8ms the input signal is identically zero.
The results of the numerical simulation are shown in figure 3. In this figure, the
result of the numerical simulation is shown along with the measured waveform. We
found a slight disagreement that we believe due to the measurement setup (the GMR
sensor and signal conditioning electronics).
1
0.8
0.8
0.6
By (a.u.)
By (a.u.)
0.6
0.4 0.4
0.2
0.2
0
0 -0.2
0 0.5 1 1.5 2 0 0.5 1 1.5 2
t (s) x 10
-3 t (s) x 10
-3
Figure 3. The measured waveform v(t ) = By (xGMR , t ) (solid) together with its small-time
approximation (dotted) and the numerically computed waveform (dashed). Left: single plate case. Right:
double plate case.
1 1
0.8 0.8
0.6 0.6
Ψ (a.u.)
Ψ (a.u.)
0.4 0.4
0.2 0.2
0 0
0 0.05 0.1 0.15 0.2 0.25 0 0.1 0.2 0.3 0.4
q (s1/2) q (s1/2)
0 0
Figure 4. The error functional used to extract the time of flight q0. Left: single plate case. Right:
double plate case.
In the same figure it is also shown the plot (dash-dot line) of the small time
approximation that, as discussed, is given by a 0 exp − q 02 / 4t / πt for appropriate ( )
values of a0 and q0. The values of a0 and q0 have been obtained by means of the best fit
approach presented in section 2.2. The parameter q0 is connected to the time of flight
through the q0 = qi + qˆTOF where qi is a parameter controlling the shape of the input
waveform and qˆTOF is the estimated time of flight. Here qi = 0.0102s1/2 for the single
plate configuration and qi = 0.0108s1/2 for the two plates configuration. Thus, the
theoretical values of q0 are 0.0246 s1/2 (single plate configuration) and 0.0397 s1/2
(double plates configuration).
In figure 4 are shown the plot of the error functional Ψ for the two cases under
analysis. The position of the minimum provide the estimated value for q0. As can be
seen clearly from the figure 3, the values of q0 obtained in this way provide small time
waveforms that are in a perfect agreement with the measured waveform up to the peak.
The related time of flight are 0.0315 s1/2 and 0.0484 s1/2 , corresponding to thickness of
3.75mm and 6.72mm. These two values are greater than the theoretical values of about
47% and 30% respectively. Again, we believe that the reason is in the measurement
setup.
4. Conclusions
In this paper we have presented an innovative approach that allows to introduce the
time of flight in the context of eddy current testing. Specifically, we have shown that
from small time analysis of the time domain waveform of an eddy current testing
measurement, associated to a proper driving waveform, it is possible to extract a
parameter that is the time of flight for a (fictitious) wave propagation problem where
the dielectric permittivity is proportional to the electrical conductivity of the real eddy
current testing problem. This entails to process eddy current testing data in terms of
time of flight.
Preliminary experiments on a simple test configuration (a conductive slab) have
confirmed the validity of the small time approximation. However, the time of flight has
been retrieved with an error that is presumably due to the measurement setup. Work is
in progress to apply the approach to more complicated test problems.
Acknowledgments
This work was supported in part by the Italian Ministry of University (MIUR) under a
Program for the Development of Research of National Interest (PRIN grant #
2004095237) and in part by the CREATE consortium, Italy.
References
[1] Bragg L R, Dettman J W, An operator calculus for related partial differential equations, J. Math. Analysis
and Applic. 22 (1968), 261-271.
[2] Bragg L R, Dettman J W, A class of related Dirichlet and initial value problems, Proc. Amer. Math. Soc.
21 (1969), 50-56.
[3] Bragg L R, Dettman J W, Related partial differential equations and their applications, Siam J. on Appl.
Math. 16 (1968), 459-67.
[4] Reznitskaya K G, The connection between solutions of the Cauchy problem for equation of different
types and inverse problems, Mat. Problemy Geofiz. Vyp. 5 part 1 (1974), 55-62 (in Russian).
[5] Lee K H and Xie G, A new approach to imaging with low frequency electromagnetic fields, Geophysics
58 (1993), 780-96.
[6] Ross S, Lusk M, Lord W, Application of a diffusion-to-wave transformation for inverting eddy current
nondestructive evaluation data, IEEE Trans. on Magnetic 32 (1996), 535-546.
[7] Tamburrino A, Fresa R, Udpa S, Tian Y, Three-dimensional defect localization from time-of-flight/eddy
current testing data, IEEE Trans. on Magnetics 40 (2004), 1148-1151.
[8] Tian Y, Tamburrino A, Udpa S and Udpa L, Time-of-flight Measurements from Eddy Current Tests
Review of Progress in Quantitative Nondestructive Evaluation, D.O. Thompson and D.E. Chimenti
(Eds.), American Institute of Physics, 22 (2003), 593-600.
[9] Tamburrino A, Udpa S, “Solution of inverse problems for parabolic equations using the Q-Transform:
Time domain analysis” , Internal report, Department of Electrical and Computer Engineering, Michigan
State University, 2002. https://www.egr.msu.edu/ece/Technicalpapers/.
[10] Lee K H, Liu G and Morrison H F 1989 A new approach to modeling the electromagnetic response of
conductive media Geophysics 54 1180-1192.
[11] Abramowitz M and Stegun I A, Handbook of mathematical functions with formulas graphs and
mathematical tables, Dover, New York, 1972.
[12] “FEMLAB User’s Guide and Introduction,” Comsol AB, 2002.
IOS Press, 2007
A Probe Array for Fast Quantitative

Eddy Current Imaging
Carmine ABBATEa, Maxim MOROZOVb, Guglielmo RUBINACCIc, Antonello
TAMBURRINOa,1, Salvatore VENTREa
a
Ass. EURATOM/ENEA/CREATE, DAEIMI, Università degli Studi di Cassino, Italy
b
CREATE Consortium, v. Claudio, 21 - 80125 Napoli, Italy
c
Ass. EURATOM/ENEA/CREATE, DIEL, Università di Napoli “Federico II”, Italy
Abstract. This work focuses on a eddy current system based on a new inversion
principle requiring the processing of low frequency measurements of the matrix of
self and mutual impedances among coils of an array of coils. Numerical
simulations presented in another work have proven that this algorithm is suitable
for fast inversion of ECT data. This work is focused on a first experimental
prototype tailored for the particular inversion algorithm. Specifically, we show that
measurements of the impedance matrix can distinguish between volumetric defects
of different lengths and we discuss the main issues to be solved in view of the
inversion of the experimental data.
Keywords. Eddy current system, fast imaging method, experimental system
1. Introduction
The contribution of this work is in the framework of the development of fast

quantitative imaging method for the non-destructive testing of conductive materials.
Nowadays, the current state-of-the-art in eddy current testing (of conductors) for on
field applications is able to provide the presence of anomalies (defects) and, in some
cases, the sizes of the anomalies too. The point is that the eddy current data cannot be
easily interpreted and connected to the anomalies that may be present in the specimen
under test. Moreover, systems operating in a real world setting have the constraint to
provide diagnostic information in a fast way. This is the reason why many
sophisticated inversion algorithms, that are usually time-consuming, are still subject of
research in the scientific community and have not yet been massively introduced in real
world applications. On the other hand, in practical applications, eddy current testing
data are usually interpreted either by trained operators, or by means of calibration
charts or by methods based on artificial intelligence. In all these cases the system
(eventually the operator) must be trained on a proper set of responses associated to
selected anomalies. New configurations that are not enough similar to configurations
present in the training set, may be more difficult to be correctly recognized. Starting
from these motivations, in the past year we have developed an automated quantitative
1
Corresponding Author: Antonello Tamburrino, DAEIMI, Università degli Studi di Cassino; E-mail:
tamburrino@unicas.it. Antonello Tamburrino is also with Electrical and Computer Engineering,
Michigan State University.
76 C. Abbate et al. / A Probe Array for Fast Quantitative Eddy Current Imaging
i1 i2 iM
v1 v2 … vM
Defect
Figure 1. Left: the conducting slab with the array of coils and a possible defect. A PC controls the currents
circulating into each coil and measures the voltages on each coil. Right: main geometrical dimensions (in mm)
of the 4 coils sensor used in the experiments. The magnetic yoke (the square support plate) has a thickness of
5mm. Each magnetic pole (leg) has a height of 7.5mm. Each coil consists of 700 turns and it has an internal
radius of 2.5mm, an external radius of 5.3mm, and height of 6.5mm.
imaging algorithm (image of the anomalies in addition to their detection) [1]-[3] that,
moreover, requires a modest computational effort and has shown real-time capabilities
during numerical tests [1].
This work presents preliminary results for a first prototype of the experimental
system that will be employed together with the fast imaging method. The paper is
organized as follows: in section 2 we briefly describe the imaging method, in section 3
we describe the experimental system, in section 4 we present sensitivity tests, in section
5 we highlight the main issues to be tackled to use the imaging algorithm on this
system, and in section 6 we drawn some conclusions.
2. The Underlying Inversion Method
The inversion method is described in [1] and [2]. Here we briefly summarize the main
issues for the sake of completeness.
The reference problem consists of identify the shape of volumetric anomalies in a
conductor (zero-thickness anomalies can be treated as well). We assume that the
resistivity of the anomaly ηi is greater than the resistivity of the background material ηb.
Both the anomaly and the host conductor are homogeneous; the extension to non-
homogeneous configurations is treated in [2].
The sensor we consider is an array of coils (see figure 1) and the data are
measurements of the self and mutual impedance between pairs of coils at several
frequencies. These measurements form the impedance matrix Z( ω) seen from the coils.
This is an M×M symmetric matrix where the element kj is given by [Z(ω )]kj = v k i j
when in=0 for n≠j, v k and i j being the complex voltage and current on the coils k and j,
respectively, and M being the number of coils. The impedance matrix Z( ω) admits the
following expansion valid for small ω (see [1], [3]):
( )
Z(ω ) = R 0 + jωL 0 + ω 2 PD( 2 ) + jω 3 PD( 2 ) + ω 4 PD( 2 ) + O ω 5 (1)
where R 0 is the dc (diagonal) resistance matrix due to the resistivity of the wires and
L 0 is the magnetostatic self and mutual inductance matrix between the coils. Both R 0
and L 0 are evaluated in the absence of the conductor (free space). Matrices PD( 2 ) , PD(3)
etc., take into account the presence of the conductor (here D represents the domain
occupied by the anomaly).
C. Abbate et al. / A Probe Array for Fast Quantitative Eddy Current Imaging 77
Vc Vc
D DTest
Figure 2. Left: an anomaly (dark grey) occupies the region D contained in the host conducting material (light
grey) Vc. Right: the a test anomaly DTest (dark grey) that is considered in the numerical problem where the
(2 )
quantity P is computed.
DTest
The imaging algorithm processes the variations of PD( 2 ) due to the presence of
anomalies, whereas the measurement system measures the matrix Z( ω). The extraction
of PD( 2 ) from measurements of Z( ω) is discussed in section 5.
The inversion method is based on the following property of the second order
moment (see figure 2):
(2 ) (2 )
P D − P D is not positive semi - definite ⇒ DTest ⊄ D (2)
Test
where D is the region occupied by the anomaly and PD( 2 ) its related second order
moment, whereas DTest is a test region that can be placed in an arbitrary position in the
(2 )
conducting domain Vc and P D is its related second order moment. Therefore, by
Test
(2 ) (2 )
evaluating the eigenvalues of P D − P D it is possible to establish if DTest is part or not
Test
of the anomaly and, by repeating the test for different position of DTest, it is possible to
reconstruct the image of the anomaly D. From the practical perspective, PD( 2 ) is
(2 )
extracted from the measured data whereas P D is numerically computed for different
Test
position of DTest.
The efficiency of the inversion method is due to the fact that to decide if a test
region, a voxel for instance, is contained in the domain D of the unknown anomaly, we
can ignore completely all the other remaining voxels. In practice the noise make the
processing more difficult; here we report the complete algorithm (see [2] for details):
(2 )
1. extract the noisy second order moment P from the measurements of Z;
set sk = f ⎛⎜ P − P ⎞⎟ for each k;

(2 ) (2 )
2.
⎝ Ωk ⎠
(2 ) (2 )
3. set Dεext = U Ω k and D ext = Dεext
ext
where ε ext = arg min P − P ext ;
s k ≥1−ε 0≤ε ≤ 2 Dε
set tk = f ⎛⎜ P − P ext ⎞⎟ for k such that Ωk⊆ D ext ;

(2 ) (2 )
4.
⎝ D \Ωk ⎠
(2 ) (2 )
5. set Dεint = U Ω k and D int = Dεintint where ε int = arg min P − P int .
t k <1−ε 0 ≤ε ≤ 2 Dε
Ω k ⊂ D ext
Here {Ω 1 , K} is a partition of the region under test, ⋅ is a matrix norm such as the
Frobenius norm and f, evaluated on a symmetrical and non-vanishing matrix A, is
( )
given by f A = ∑i λi ∑i λ i where λi is the i-th eigenvalue of A. A faster version of
the method (see [1]) consists of steps 1 and 2 only.
Figure 3. The first prototype of the hardware of the measurements circuit. The hardware is controlled by a
personal computer through a parallel port. Measurements are completely automated.
3. The Hardware and Software System
(2 )
The measurement/estimation of the variation of P due to the presence of anomalies
is a challenging problem due to the low level (bad signal-to-noise ratio) of the related
signals. To increase the SNR, the coils are mounted on ferrite cores. Figure 3 shows the
picture of a first prototype of the measurement system completely controlled by a PC.
The coil array consists of several induction coils (see figure 1) coupled via a
magnetic yoke with corresponding number of poles (legs). The coils are excited one by
one with an AC current. When the n-th coil is being driven, an induced voltage is
measured on every coil and divided by the excitation current, therefore producing a
symmetric impedance matrix. The signal generation and acquisition are carried out
using PCI-expansion boards installed in a PC. The analog output (AO) board NI PCI-
6733 generates the excitation and compensation signals. It has 8 AO channels and
sampling frequency up to 1MS/s. The signal conditioning and control board is a bench-
top device engineered in our laboratory. It provides the following functionality:
filtering and amplifying the excitation signal, subtracting the compensation signals,
amplifying the measured signals, switching coils between the excitation and
measurement channels and setting gains and auxiliary controls. The analog input (AI)
board NI PCI-6254 performs acquisition of the measured signals. It has 16 differential
AI channels and sampling frequency up to 1.25 MS/s (shared between channels).
Scanning is performed by an XY-surface scanner which moves the measuring head
above a test piece.
4. Sensitivity of the System
The first experimental test carried out on the system was a sensitivity test. This test is
aimed to understand the possibility of the system to detect the presence of simple EDM
cracks of different length. In this test we have not considered the imaging of the shape
of the anomaly by the inversion algorithm presented in section 2 and, therefore, we
have considered a frequency such that the skin-depth is of the order of the thickness of
the sample. The sample was a 20cm×20cm (thickness 2mm, electrical resistivity
5.61 ⋅ 10 -8 Ω m ) Aluminum sample having at its center a 100 % surface breaking EDM
−Z
mn
Figure 4. The spatial distribution of Z obtained by means of a spatial scan (2D scan with
BG
constant lift-off) on a 2cm×2cm region centred on the defect (the horizontal white line). Left: 3mm long
defect, the scale is from 0Ω (black) to 2.18Ω (white). Right: 5mm long defect, the scale is from 0Ω (black) to
5.28Ω (white). Here Z = 194Ω . The origin of the reference system is at the uppermost left corner.
BG
0.25 0.5
l=5mm l=5mm
0.2 l=3mm 0 l=3mm
0.15 -0.5
0.1 -1
X (Ω)
X (Ω)
0.05 -1.5
0 -2
-0.05 -2.5
-0.1 -3
-0.1 -0.05 0 0.05 0.1 0.15 -0.5 0 0.5 1
R (Ω) R (Ω)
Figure 5. The Lissajous patterns obtained considering the measurement taken on line 6 from below.
Left: pattern related to the element 4,1 of the matrix Z − Z . Right: pattern related to the element 4,4 of
m,n
BG
the matrix Z
m,n
−Z
BG
( )
. Notice that Z
BG
4 ,1
= 2.83 + j 5.12Ω and Z ( )
BG
4 ,1
= 45.3 + j143.5Ω .
crack of thickness 0.1mm and length either 3mm or 5mm. The sensor (lift-off equal to
0.5mm) consists of four coils mounted on a ferrite support (μr=2000, see figure 1). The
impedance matrix was measured at a frequency of 2.5kHz (skin-depth 2.38mm) in
different positions. Specifically, the impedance matrix was measured onto a regular
grid of 2cm×2cm (21×21 measurements were collected) symmetrically centered on the
defect. Figure 4 present the norm (the standard L2 norm) of the difference Z − Z BG
mn
is the impedance matrix measured at position x = mΔ , y = nΔ (here

mn
where Z
Δ=1mm and m,n = −10, -9, ..., 10) and Z BG is the impedance matrix measured in a
defect free part of the specimen.
From figure 4 it is evident that the system has the capability of detecting the target
defects. Moreover, the system is also capable to discriminate between these two
different defects as follows from the Lissajous patterns (figure 5) obtained by
considering the 21 measurements taken on a horizontal line parallel to the defect (row 6
from below).
Table 1. Example of real part of the impedance matrix at 200Hz. δ is the ratio between the elements of ΔR
and RBG. This matrix has been measured at the location corresponding to row 17 and column 11 of the 21×21
scanned area (see figure 4 for the reference system).
5. Toward the Fast Imaging Algorithm
In section 4 we have shown that the experimental system has “standard” capabilities of
detecting and discriminating defects. In this section we discuss the problems to be
assessed in order to make possible to apply to the measured data the imaging algorithm
described in section 2. The application of the imaging algorithm to the measured data is
beyond the scope of this paper.
The main issue is that the inversion algorithm process the second order moment
(2 )
P extracted from the measured impedance matrix at several frequencies. The way
(2 )
P is estimated starting from the measured impedance matrix is by means of a
weighted least square approach. Specifically, in [1], [3] we proposed to truncate the
(2 )
series expansion (1) up to the fourth order term and to recover P from spectral data
by means of a weighted least square approach applied element-wise, i.e. by minimizing
( )
NF
Θ(α , β , γ ) = ∑ wijk α + βωk2 + γωk4 − dijk
2
(3)
k =1
and setting P ( )
(2 )
ij
= β . In (3) NF is the number of frequencies where measurements of
{ }
the impedance matrix are available, d ijk = Re [Z(ω k )]ij and wijk is a weight. Usually
(in maximum-likelihood approach) wijk is set equal to the standard deviation of the data.
Here, since measurements carried out at different frequencies have significant different
magnitude, we set wijk = ω k− nw where nw is, usually, a small integer. Numerical test
showed that a value of about 8 gives the best possible condition number for the linear
system arising from the minimization of (3). Numerical tests have shown that the
approach based on (3) is effective as long as the considered frequencies are such that
the electromagnetic field penetrate in the material and low frequencies are considered.
For example in the experimental case described in section 4 the frequencies to be used
(2 )
for extracting P are in the range 100Hz-800Hz. As well known, inductive
measurements at these low frequencies are critical. Table 1 reports at a frequency of
200Hz for the 3mm long defect the real part of the background impedance matrix (RBG)
and the real part of the impedance variation due to the defect (ΔR).
From Table 1 it is evident that the quantity of interest (ΔR) at this “low” frequency
of 200Hz presents elements between one tenth and few tenth of mΩ and, moreover,
these correspond to variation (w.r.t. RBG) between 0.1% and 15%. In addition, matrices
RBG and ΔR are not exactly symmetric as expected from a theoretical point of view.
800 0
700
-200
600
500
-400
ΔR (mΩ)
ΔX (mΩ)
400
300 -600
200
-800
100
0 -1000
0 200 400 600 800 1000 0 200 400 600 800 1000
f (Hz) f (Hz)
5 0
-2
0
-4
ΔR (mΩ)
ΔX (mΩ)
-5 -6
-8
-10
-10
-15 -12
0 200 400 600 800 1000 0 200 400 600 800 1000
f (Hz) f (Hz)
Figure 6. The errors introduce by the experimental system for a test load made by 4 equal resistors. Left: real
part. Right: imaginary part. Top: element 1,1 of the measured impedance matrix. Bottom: element 1,2 of the
measured impedance matrix. Values of the test resistors are: RL=0Ω (solid), RL=1Ω (+), RL=2Ω (*), RL=5Ω
(◊), RL=25Ω (o), RL=50Ω (×), RL=100Ω (□).
We have, therefore, carried out a preliminary study to evaluate the errors

introduced by the experimental system. Specifically, we have applied four resistors of
equal (and known) value as load. In this case, the expected resistance matrix is a 4×4
diagonal matrix having on the diagonal the values of the 4 test resistors and, therefore,
it is easy to quantify the errors introduced by the system (in this configuration). The
experimental results are shown in figure 6 with respect to element 1,1 (self-impedance)
and 1,2 (mutual-impedance) of the measured impedance matrix.
The measurements errors depend on the load and are larger on the self-impedance
measurements. Moreover, from figure 6 it is evident that large variations of the load
resistors (between 0Ω and 2Ω) produce only a slight variation of the errors affecting
the measurement.
The strategies that can be used to properly measure the real part of the impedance
are essentially two: (i) to create a low frequency EMC model to explain and correct (at
least partially) the undesired coupling between different parts of the power bus (used to
energize the coils) and the measurement bus (used to read the voltage across the
sensing coils) and (ii) to measure the impedance variation with respect to the
background impedance. These strategies are currently under study.
Regarding strategy (i), we notice that the (almost) constant error on the real part
can be explained as a common mode impedance coupling [4], whereas the error on the
imaginary part increasing (almost) linearly with the frequency can be explained by
considering spurious capacitive and/or inductive couplings [4].
Regarding strategy (ii), we notice that, in general, the systematic error depends on
the measured impedance matrix. In addition, if we consider that, in this critical
situation, the impedance variation with respect to the background is relatively “small”
(less than the previously discussed 2Ω value), we have that the systematic error is
almost unaffected by the presence of the defect, therefore:
Meas True
(
Z BG = Z BG + δ Z E Z BG + N BG , Z
True
) Meas
≅Z
True
(
+ δ Z E Z BG + N
True
) (4)
Meas Meas
where N and NBG represent random noise terms, Z BG and Z represent the
True True
measured impedance matrices (BG stays for background), Z BG and Z represent
the noise-free impedance matrices, δ Z E Z BG ( True

) represents the systematic error
introduced by the system. From (4) it follows that δ Z =δZ + N − N BG where
Meas True
δ Z True = Z True − Z True

BG
, i.e. the measured impedance variation results to be less affected
by the systematic error.
6. Conclusions
In this paper we have presented an experimental system, designed and developed in our
laboratories, to be used with a fast imaging method aimed to achieve real-time
quantitative imaging. Several experimental tests have been performed and, in particular,
we have clearly shown that this system is capable to detect small surface breaking
crack and is also capable to discriminate different defects.
In view of the use of the experimental system with the fast imaging method, we
have also highlighted the weakness of the system (low frequency measurements) and
described possible techniques to overcome this drawback.
Acknowledgments
References
[1] A. Tamburrino, G. Rubinacci “Fast methods for quantitative eddy current tomography of conductive
materials”, IEEE Trans. on Magnetics, vol. 42, no. 8, pp. 2017-2028, August 2006.
[2] A. Tamburrino, G. Rubinacci, “A new non-iterative inversion method for Electrical Resistance
Tomography”, Inverse Problems, 18, pp. 1809-29, 2002.
[3] G. Rubinacci, A. Tamburrino, “A non-iterative ECT data inversion algorithm”, in E’NDE,
Electromagnetic Non-destructive Evaluation (VII), G. Dobmann (Ed.), IOS Press, 2006.
[4] Clayton R. Paul, Introduction to Electromagnetic Compatibility, Second Edition, John Wiley & Sons,
2006.
IOS Press, 2007
Superposition of Several Phase-Shifted

Exciting Fields for Crack Evaluation
Ladislav JANOUSEK a,1 , Noritaka YUSA b and Kenzo MIYA b
a
University of Zilina, Slovak Republic
b
IIU, Japan
Abstract. The paper proposes a novel method for crack evaluation based on electro-
magnetic induction phenomenon. The principle of the method is to realize a unique
distribution of induced eddy currents by superposition of several phase-shifted ex-
citing electromagnetic fields. The ratio of the superposition is varied to inspect a
crack using different distributions of eddy currents without changing the frequency.
Numerical and experimental results reveal that the method provides clear indica-
tion about the depth of cracks. In addition, the method is applicable even for cracks
which are much deeper than the standard depth of penetration.
Keywords. Electromagnetic induction, eddy currents, phase-shifted fields, superpo-
sition, defect depth evaluation
1. Introduction
Scheduled in-service inspection is necessary for the maintenance of structural compo-

nents in many industrial fields. Usually, a component is allowed to stay in service even if
a crack is found in the component. However, the crack has to be properly characterized
to assure that it will not be larger than the critical flaw size until the next inspection.
Frequently, ultrasonic-based methods are used for this purpose. However, such methods
are not effective in the inspection of highly anisotropic or inhomogeneous materials [1].
Recently, several studies have proposed to employ electromagnetic-based methods for
non-destructive inspection of such materials and structures [2].
Electromagnetic-based methods are usually used only for detection of a crack be-
cause their signals do not carry explicit information about the crack dimensions. Several
papers employing numerical inversion techniques to estimate the dimensions have been
published [3]. However, as the ill-posedness of the problem has not been fully under-
stood yet, practical application of the numerical inversions is quite difficult [4]. Improve-
ments in the interpretation of the data gained using electromagnetic-based methods are
therefore still necessary.
The paper proposes a novel electromagnetic method for crack evaluation to over-
come the problem mentioned above. The key idea is to inspect a crack using various
1 Corresponding Author: Ladislav Janousek, Department of Electromagnetic and Biomedical Engineering,

Faculty of Electrical Engineering, University of Zilina, Univerzitna 1, 010 26 Zilina, Slovak Republic; E-mail:
janousek@fel.utc.sk
84 L. Janousek et al. / Superposition of Several Phase-Shifted Exciting Fields for Crack Evaluation
distributions of alternating currents flowing inside an inspected body to gather relevant

information about the depth of a defect.
The authors have mainly worked for the enhancement of eddy current techniques
and thus the novel method is applied to the eddy current non-destructive testing (ECT),
however, the principle can be employed in other electromagnetic-based non-destructive
techniques (i.e. AC potential drop).
2. Principle of the Novel Method
The principle of the novel electromagnetic method for crack evaluation is to realize a
unique distribution of eddy currents flowing inside an inspected conductive body. A new
ECT probe shown in Fig.1, originally proposed by the authors for the non-destructive
inspection of near-side deep cracks in thick structures [5], is used for the purpose. The
probe consists of four coaxial rectangular exciting coils positioned tangentially to the
surface of a specimen. The four coils are divided into two detached sets separated by
a space of 50 mm. The inner exciting coils and the outer ones are connected in series,
respectively, and they are driven independently by phase shifted currents of 180 o . The
signal is sensed by a circular coil positioned normally to the surface of a specimen in the
center between the two sets of the exciting coils.
Differences in position of the inner exciting coils and the outer ones from the pick-up
coil (the difference is 10 mm) make it possible to locally drive eddy currents of different
distributions under the pick-up coil. The situation is shown in Fig. 2 for an SUS316L
plate specimen with a thickness of 25 mm. The two dependences of the absolute value of
eddy current density along material depth differ depending on whether the outer exciting
coils or the inner ones are driven. Therefore, superposing eddy currents induced by the
inner and the outer exciting coils with different ratios enables one to vary the distribution
of eddy currents under the pick-up coil. The ratio of superposition relates to a ratio of
the inner and the outer exciting currents’ densities: J i /Jo , and thus the distribution of
eddy currents can be controlled by changing the densities of the driving currents. It has
front view side view

exciters
outer
inner
28
30
detector
1
10 10 50 10 10 28
30
top view
Figure 1. Arrangement and dimensions of the ECT probe

L. Janousek et al. / Superposition of Several Phase-Shifted Exciting Fields for Crack Evaluation 85
16
absolute value of eddy current density [mA/m2]

outer
inner
14
12
10
0
0 5 10 15 20 25
material depth [mm]
Figure 2. Dependence of the absolute value of eddy current density on the material depth for the outer exciting
coils and for the inner exciting coils, respectively
been found out that the distribution of eddy current density influences the amplitude, the
phase as well as the shape of a crack signal. Therefore, by changing the ratio during the
inspection one can obtain more information about the crack.
The ratio of the superposition can be controlled in two different ways. At first, the
exciting coils with tap winding can be utilized. However, many scans are needed to in-
spect the crack using different distributions of eddy currents and the ratio can be changed
only in certain steps given by configuration of the taps.
The second approach is more simple but applicable only for linear problems. The
crack is inspected twice: once with driving only the outer exciting coils (real part of the
signal Reo , imaginary part of the signal Im o ) and the second time with driving only
the inner exciting coils (real part of the signal Re i , imaginary part of the signal Im i ).
Both the obtained signals for one crack are then numerically mixed; the real (Re) and
the imaginary (Im) parts of the mixed signal are given:
Re = Ci · Rei − Co · Reo , Im = Ci · Imi − Co · Imo , (1)
where Ci and Co are arbitrary numbers representing the ratio of the exciting currents’
densities: Ji /Jo = Ci /Co . In this case, the ratio can be changed in a wide range with
small steps. The latter approach is used in this paper due to its simplicity and due to the
linearity of the investigated problem.
3. Influence of the Eddy Current Distribution on Crack Signal
Influence of the eddy current density distribution along material depth under the pick-
up coil on the crack signal is investigated in this section by numerical means. A three
dimensional finite element code is used for calculations.
A plate specimen, Fig. 3, made of a stainless steel SUS316L is inspected in this
study. Thickness of the specimen is 25 mm and electromagnetic characteristics of the
material are: conductivity of σ = 1.4 MS/m and relative permeability μ r = 1. A non-
SUS316L
0.5
40
dc
σ = 1.4 MS/m
μr= 1
25
Figure 3. Configuration of the specimen
conductive crack of rectangular shape with a length of l c = 40 mm, a width of wc = 0.5

mm and with a variable depth d c simulates an electro-discharge machined (EDM) notch.
An exciting frequency of 10 kHz is applied for the inspection.
The crack is inspected using the probe shown in Fig. 1. Two scans are made over the
crack along its length; only one group of the exciting coils, i.e. the inner or the outer, is
driven during a particular scan. Both the signals for one crack are then mixed according
to (1).
Five signals of the crack with a depth of d c = 10 mm for five different values of
the ratio of the exciting currents’ densities J i /Jo are shown in Fig. 4. As it can be seen,
the crack signal rotates clockwise increasing the ratio while the amplitude of the crack
signal changes. The crack signals obtained for the outer exciting coils and the inner ones
are mixed numerically, therefore it is possible to change the ratio with small steps in a
wide range. Figure 5 shows the dependences of the crack signal amplitude and its phase
on the ratio Ji /Jo for the crack with a depth of d c = 10 mm. It can be observed that
the crack signal amplitude decreases until a certain point when increasing the ratio and
1
Ji/Jo=0.0/1.0
Ji/Jo=0.5/1.0
Ji/Jo=1.0/1.0
Ji/Jo=1.0/0.5
Ji/Jo=1.0/0.0
0.5
imaginary [mV]
-0.5
-1
-1 -0.5 0 0.5 1
real [mV]
Figure 4. Signals of the crack with a depth of dc = 10 mm for different values of the ratio Ji /Jo
0.9
amplitude
phase
0.8
100
0.7
0.6
amplitude [mV]
phase [degree]
0.5 50
0.4
0.3
0
0.2
0.1
-50
0
0 0.5 1 1.5 2 2.5 3
ratio Ji/Jo [-]
Figure 5. Dependences of the crack signal amplitude and its phase on the ratio Ji /Jo for the crack with a
depth of dc = 10 mm
1
dc=10mm
dc=12mm
dc=15mm
dc=20mm
0.8
relative signal amplitude [-]
0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3
ratio Ji/Jo [-]
Figure 6. Dependences of the crack signal relative amplitude on the ratio Ji /Jo for the crack with depths of
dc = 10, 12, 15, 20 mm
then increases again while the phase of the crack signal changes almost 180 o . Similar
characteristics of the crack with different depths of d c = 10, 12, 15, 20 mm are shown
in Fig. 6, 7. It is evident that the ratio where the amplitude of the crack signal reaches
its minimum as well as rotation of the crack signal depend on the crack depth. Thus,
two criteria can be extracted from these characteristics: 1) value of the ratio where the
crack signal amplitude reaches its minimum; 2) value of the ratio where the crack signal
rotates at an angle defined as a half value of the total crack signal rotation. Dependences
of the ratio Ji /Jo on the crack depth for the two extracted criteria are shown in Fig. 8.
The dependences for both the criteria are nearly the same and they are almost linear.
The numerical results are experimentally verified in the next section.
160
140
120
phase change [degree]

100
80
60
40
dc=10mm
20 dc=12mm
dc=15mm
dc=20mm
0
0 0.5 1 1.5 2 2.5 3
ratio Ji/Jo [-]
Figure 7. Dependences of the crack signal phase change on the ratio Ji /Jo for the crack with depths of
dc = 10, 12, 15, 20 mm
1.6
1.4
1.2
1
ratio Ji/Jo [-]
0.8
0.6
0.4
0.2
amplitude
phase
0
0 5 10 15 20 25
crack depth [mm]
Figure 8. Dependences of the ratio Ji /Jo on the crack depth dc for the two extracted criteria
4. Experimental Verification
Four rectangular EDM notches are experimentally inspected to confirm the numerical
results. The notches measure l c = 40 mm in length, w c = 0.5 mm in width and d c =
10, 12, 15, 20 mm in depth. The notches are introduced into an SUS316L plate specimen
with a thickness of 25 mm. The parameters of the specimen and of the crack are the same
as ones used in the numerical investigations. The probe (configuration and dimensions
are given in Fig. 1) scans twice at the near side over each crack along their length. A
function synthesizer and an amplifier are utilized to drive the inner and the outer exciting
coils. A frequency of 10 kHz is used for the inspection. The crack signal is picked-up by
a lock-in amplifier and stored in a PC through an A/D board. Both the signals for each
1
dc=10mm
dc=12mm
dc=15mm
dc=20mm
0.8
relative signal amplitude [-]

0.6
0.4
0.2
0
0 0.5 1 1.5 2 2.5 3
ratio Ji/Jo [-]
Figure 9. Dependences of the crack signal relative amplitude on the ratio Ji /Jo for the crack with depths of
dc = 10, 12, 15, 20 mm
160
140
120
phase change [degree]
100
80
60
40
dc=10mm
20 dc=12mm
dc=15mm
dc=20mm
0
0 0.5 1 1.5 2 2.5 3
ratio Ji/Jo [-]
Figure 10. Dependences of the crack signal phase change on the ratio Ji /Jo for the crack with depths of
dc = 10, 12, 15, 20 mm
crack are then linearly mixed based on (1). It should be noted that all the instruments
used in the experiments are the ones that are also utilized in the conventional ECT.
The experimentally gained dependences of the crack signal relative amplitude and
the crack signal phase’s change on the ratio J i /Jo for the four EDM notches are shown
in Fig. 9 and 10, respectively. The behavior of the experimental crack signals is very
similar to the one obtained using the numerical simulations (Fig. 6, 7). Comparison of
the experimental data with the simulated dependence of the ratio J i /Jo on the crack
depth for the phase criterion is shown in Fig. 11. It can be seen that there is a difference
between the numerical and experimental results. However, the experimental results prove
that the proposed novel electromagnetic method for crack evaluation works efficiently. It
1.8
1.6
1.4
1.2
ratio Ji/Jo [-]

1
0.8
0.6
0.4
0.2
simulation
experiment
0
0 5 10 15 20 25
crack depth [mm]
Figure 11. Dependences of the ratio Ji /Jo on the crack depth dc for the phase criterion, comparison of
numerical and experimental results
has been revealed that the method provides clear indication about the depth of a detected
crack. Moreover, the dependence of the ratio on the crack depth is almost linear and thus
the method is applicable also for cracks which are much deeper than the standard depth
of penetration.
5. Conclusion
The paper proposes a novel electromagnetic method for crack evaluation. A new eddy
current testing probe was used to drive eddy currents with variable distribution by super-
position of two exciting electromagnetic fields without changing the exciting frequency.
It was shown that the amplitude and the phase of a crack signal strongly depend on the
eddy current distribution and thus on the ratio of the superposition. Considered cracks
were inspected with different values of the superposition and the obtained crack signals
were processed to extract two criteria. It has been proved that the method provides clear
indication about the depth of a crack and it is applicable also for cracks which are much
deeper than the standard depth of penetration.
References
[1] W. Cheng et al.: Ultrasonic and eddy current testing of defects in Inconel welding metals, Proceedings
of 12th MAGMA conference, Oita, Japan, 2003, 187–190.
[2] N. Yusa et al.: Application of eddy current inversion technique to the sizing of defects in Inconel welds,
Nuclear Engineering and Design 235 (2005), 1469–1480.
[3] B.A. Auld and J.C. Moulder: Review of advances in quantitative eddy current nondestructive evaluation,
Journal of Nondestructive Evaluation 18 (1999), 3–36.
[4] N. Yusa et al.: Caution when applying eddy current inversion to stress corrosion cracking, Nuclear
Engineering and Design 236 (2006), 211–221.
[5] L. Janousek et al.: Excitation with phase-shifted fields - enhancing deep defect evaluation using eddy
currents, NDT&E International 38 (2005), 508–515.
IOS Press, 2007
Automatic Discrimination of Stress

Corrosion and Fatigue Cracks Using Eddy
Current Testing
Stéphane PERRIN 1, Noritaka YUSA and Kenzo MIYA
IIU, 2-7-17-7F Ikenohata, Taito-ku, Tokyo 110-0008, Japan
Abstract. This paper presents a methodology using Eddy Current Testing (ECT)
for the discrimination of two types of cracks commonly found in Japanese nuclear
power plants: Stress Corrosion Cracks (SCCs) and Fatigue Cracks (FCs). The
discrimination is based on the extraction of relevant features that are classified
using two classifiers. These classifiers are trained with simulated features and
tested with experimental ones. More meaningful results with regard to the
discrimination problem are obtained through the introduction of a so-called
cautious classifier.
Keywords. Stress Corrosion Cracks, Fatigue Cracks, Features Extraction,

Automatic Classification, Cautious Classifier.
1. Introduction
According to the present regulations on Japanese nuclear power plants, the owner of a
plant where a crack appeared must investigate the reason why it appeared. Because it is
necessary to determine the type of the crack as the first step of the investigation,
various tests, including metallographic and chemicals ones, are requested. If a
nondestructive method can reveal the type of a crack, these costly and time-consuming
tests may be avoided.
Previous studies [1-2] have shown the ability of two EC sensors, the uniform and
the differential plus-point probes, to discriminate SCCs and FCs. In these studies,
discrimination was performed through simple threshold methods on extracted features.
Nonetheless, it was not possible to discriminate the totality of the cracks without errors.
In the context of nuclear plant inspection, such results are insufficient and
improvements were necessary.
The present paper aims to ameliorate the previously obtained results by improving
both the physical system itself and the classification method. The first point is
addressed by proposing a multi-probe configuration instead of two individual probes.
The second point, amelioration of the classification method, is solved by the use of
more complex classifiers [3-4], k-Nearest Neighbors and Artificial Neural Network [5-
6], and synthetic EC signals to train them [7]. Finally, two ways to further improve the
classification results are introduced. It is claimed that the results obtained are more
meaningful in the specific context of the discrimination of SCCs and FCs.
1
Corresponding Author: Stéphane Perrin, IIU, 2-7-17-7F Ikenohata, Taito-ku, Tokyo 110-0008, Japan;
E-mail: sperrin@iiu.co.jp
92 S. Perrin et al. / Automatic Discrimination of Stress Corrosion and Fatigue Cracks
2. Features Extraction
The discrimination of SCCs and FCs with uniform and plus-point probes has been
exhaustively studied in previous articles [1-2]. In these studies, the main differences
between SCCs and FCs were presented from an ECT point of view. An SCC is
considered as a conductive wide crack while an FC is considered as a narrow crack that
acts like an impermeable barrier to eddy currents. The differences in SCC or FC
responses to eddy currents, generated by a given ECT sensor and that flow parallel and
perpendicular to the crack, were studied and used to extract discriminative features [7]
depending on the probe used. Features are assumed to be more representative of the
underlying regularities of the data and more relevant according to the task to be
performed than raw data that is extracted prior to classification itself.
A single feature, called Į, has been extracted from the signals of the uniform probe
[1]. This feature was defined as the ratio of the maximum of signals due to the
disturbance of perpendicular currents over the maximum of signals due to the
disturbance of parallel currents. Four features were extracted from the signals of the
differential plus-point probe [2]. The extraction of these features relies on at least two
scan lines, one perpendicular to the crack length (x direction, see Figures 1 and 2) that
goes through the center of the crack and the other one parallel to the crack length that
goes too through the center of the crack (y direction, see Figure 3). The two EC sensors,
uniform and differential plus-point probes, were individually evaluated on the same 82
specimens with similar threshold methods. 87.8% of the specimens were classified
correctly in the case of the uniform probe and 84.1% in the case of the plus-point probe.
The first proposition to improve the performance in terms of discrimination of
SCCs and FCs is to make use of complementarity through a multi-probe system. The
multi-probe configuration means that all the extracted features from several probes are
aggregated in a single features vector that is later classified by a classification
algorithm. In this study the physical system itself remains unchanged; the data is
collected by sequentially using each of the probes individually. The second proposition
to improve the performance is to use a more complex classification method.
Figure 1. Extraction of the first feature of the plus-point probe.

S. Perrin et al. / Automatic Discrimination of Stress Corrosion and Fatigue Cracks 93
Figure 2. Extraction of the two spatial spreading features of the plus-point probe.
Figure 3. Extraction of the fourth feature of the plus-point probe.
Table 1. Parameters of the cracks models used for numerical simulations.

Width (mm) Length (mm) Depth (mm) Conductivity (%)
SCC 0.6, 1.0 10, 15, 20 3, 4, 5 5.0, 7.5
FC 0.1, 0.2 10, 15, 20 3, 4, 5 0.0, 1.0
Overall, five features are extracted from the two probes. These five features are
combined in a single vector Vfeat of features that is extracted for each of the inspected
specimens. To have features that are all of the same order, a first normalization is
performed. The first three features are inverted so that values of each of the five
features are comprised between zero and one.
3. Classification
3.1. Classification Algorithms
Once the features vectors are obtained for all the specimens, they are classified in order
to determine the type of crack, depending on the values of their elements. Different
kinds of classifiers are available [5] and the choice of a given classifier depends on the
problem to be solved. Because of the relatively low amount of available data and the
absence of knowledge about eventual probabilistic properties of the data, two structural,
supervised classifiers are chosen and compared: the k-Nearest Neighbors (k-NN)
algorithm and a feed forward back propagation Artificial Neural Network (ANN).
3.2. Simulated and Experimental Features
Whatever is the chosen classifier, k-NN or ANN, it must be trained with data similar to
the ones that will be classified later. It can be trained with part of the available
experimental data, the other part being kept for performance evaluation, or testing.
Another solution is to train the classifier with simulated data that represents the real
ones keeping all the experimental data for performance evaluation [7]. Using simulated
ECT signals allows for the possibility of a large amount of data in a lot of different
configurations. Moreover, experimental ECT signals can be kept for performance
evaluation.
The simulations provided signals that are numerical simulations of the ECT signals
obtained with the two probes. From these simulated signals, features are extracted in
the same manner that they are extracted from experimental signals. The experimental
signals are obtained from 82 specimens constituted of 37 SCCs and 45 FCs. The
specimens were plates made of alloy 600, SUS304, or SUS316. All the cracks were
artificial. The simulated signals are obtained from numerical models of the crack that
are representative of the real ones. The models chosen are shown in Table 1. The
values of the parameters of the crack models used during simulation are validated by
comparing the simulated signals to the ECT measurements through forward analysis.
Moreover, numerous destructive tests have provided valuable information about the
physical parameters of the two types of cracks. Generally speaking, FCs are defined as
narrow, relatively non-conductive cracks and SCCs as wide, relatively conductive
cracks. The depth of 1mm for FC crack models is not included as it represents a
physically implausible depth. These models are used for simulating the signals of the
two probes with three different lift-off values (0.1, 0.5 and 1 mm). Frequency is always
50kHz for both probes. The total number of simulated features vectors is 324 (108 for
each lift-off value, constituted of 48 FCs and 60 SCCs).
The main drawback for the use of simulated features is that they differ from
experimental ones for several reasons. As the difference can be significant for some of
the features, a calibration is necessary to make the simulated features as representative
as possible of the experimental features. The calibration is done by using part of the
experimental features vectors. As it may be noticed, this calibration eliminates one of
the advantages of using simulated features for training purposes: all experimental
features can be kept for performance evaluation. For this reason, the performance will
be evaluated by comparison with the classification using only experimental features
that do not require calibration. If the performance is higher with simulated features for
an equal part of experimental features used for calibration (when simulated features are
used for training) or for training (when only experimental features are used), then there
is an advantage in using them for training.
The calibration itself is performed by first dividing each of the training features by
its mean computed using all of the simulated features. Then each training feature is
multiplied by the mean of the corresponding experimental feature computed using all
of the features of the calibration set. With such a calibration, each simulated feature of
the training set has the same mean than the corresponding experimental feature taken
from the calibration set. Moreover, in order to have the training features, simulated or
experimental, and testing features in the same range, a second normalization is
performed. The minimum and maximum values of each element of the features vectors
of the training set are computed. These two values are used to normalize the features of
both the training and testing sets so that each feature is comprised between zero and
one.
Table 2. Classification results using k-NN and ANN, respectively, with experimental (Exp.) and simulated
(Sim.) features.
Pc (%) SME Pc (%) SME

k-NN ANN
Exp./Sim. Exp./Sim. Exp./Sim. Exp./Sim.
LOOCV 95.1 / 98.8 2.4 / 1.2 LOOCV 97.6 / 96.3 0.15 / 0.2
CV 95.6 / 98.3 0.15 / .09 CV 98.3 / 95.7 0.015 / 0.02
3.3. Performance Evaluation
If simulated features are used or not, the experimental features set must be partitioned
into two sets (calibration/testing or training/testing, respectively). This partition has a
strong influence on the performance evaluation because the testing set is dependant on
it. Several methods exist for evaluating the performance depending on the partition of
the data set (the experimental features set). The method used in this paper is based on
the K-Fold Cross-Validation [5] with K=2. The data set is divided in two subsets, the
elements of each subset being randomly chosen. For each random partition,
performance is evaluated and given by Pi (i=[1..m], with m the number of random
partitions), defined as the fraction of elements correctly classified over the total number
of elements. The overall performance of the classifier Pc is given by the mean of Pi.
The error on Pc is evaluated by the Standard Mean Error (SME). If ı is the standard
deviation of Pc, the SME is V m .
Hereafter, this method will be called Cross-Validation (CV). If n is the size of the
data set, the K-Fold CV can be applied with K=n-1. In this case, evaluation is
performed n times with only one element in the testing set. This method is called
Leave-One-Out Cross-Validation (LOOCV, [5]). The CV and the LOOCV methods are
chosen for evaluating and comparing the performance of the classifiers.
3.4. Classification Results
As said previously, the two classifiers are evaluated using respectively simulated
features and only experimental features vectors for training. The testing set is
constituted of the remaining experimental features vectors that were not selected for
calibration or training. In the case of simulated features used for training, the training
set size is the size of the simulated features vectors set. Performance results for both
classifiers are given in Table 2. The parameters of the k-NN are m=2000, k=10 and the
Euclidean distance is used. The chosen ANN is a feed forward back propagation
network with two layers in addition to the input layer. There are five neurons in the
input layer, three in the second layer and one in the output layer. In the case of the
ANN, m=2000 as well.
The training set of simulated features is constituted of the simulated features
obtained for a lift-off equal to 0.1mm that is in good agreement with the real lift-off
value of the experiments. Training with simulated features obtained with other lift-off
values or the combination of several lift-off values gives lower values for Pc. The same
remarks are valid for the ANN classifier. Results obtained with the CV method are
given for a calibration or training set size that is 90% of the experimental features
vectors set. The testing set is thus 10% of the experimental features vectors set.
As it can be seen in Table 2, the overall performance with simulated features or

only experimental features is similar, regardless of the classifier. In the case of only
experimental features used for training, the total amount of data is very small (82
features vectors), which explained the lower performance of the k-NN applied with
k=10. With k=5, a slight improvement can be observed.
4. Improvement of Classification
4.1. Cautious Classifier
Depending on the cost of an error, when one is unsure, it is sometimes better not to
make a decision at all than to make a wrong decision. In the case of the discrimination
of SCCs and FCs, the cost of an error is very high, including safety concerns. If no
decision is taken, the cost is merely financial, due to the need to conduct more
inspections for determining the type of the crack. It is thus preferable that a classifier
takes no action if in doubt rather than making a erroneous decision.
A cautious classifier [8] prefers abstention if confidence in a decision is below a
given threshold. Two notions characterize the performance of a cautious classifier, the
abstention and the accuracy. The abstention is the number of examples, here an
example is a features vector, for which the classifier says “I don't know” divided by the
total number of examples. The accuracy is the number of correctly classified examples
divided by the total number of classified examples, which is the difference that results
from the total number of examples and the number of abstentions. The higher the
accuracy and the lower the abstention, the better is the cautious classifier. If T is the
threshold on the confidence, the higher T is, the more cautious is this classifier. The
definition of the confidence as well as the different values for this threshold depend on
the classifier (see Table 3).
Depending on the classifier used, it is possible to easily achieve 100% of correct
classification either if simulated features or only experimental features are used. But
the same pattern can be observed: a high accuracy is obtained at the expense of a high
abstention. With the k-NN and the most cautious behavior possible, it is possible to
achieve 100% or almost 100% of accuracy with a significantly lower abstention using
simulated features than when using only experimental features. Abstention levels are
then comparable between k-NN and ANN. The introduction of a cautious classifier
allows obtaining 100% of correct discrimination at the cost of some abstentions. In the
context of discrimination of SCCs and FCs in nuclear plants, it is a crucial advantage
compared to the use of the usual classification method.
Table 3. Classification results in terms of accuracy (Acc.) and abstention (Abs.) with a cautious classifier
using k-NN and ANN, respectively, with experimental (Exp.) and simulated (Sim.) features
k-NN ANN
Acc. (%) Abs. (%) Acc. (%) Abs. (%)
T T
LOOCV .9 100 / 98.7 12.2 / 8.5 .9999 98.75 / 100 2.4 / 14.6
CV .9 100 / 98.9 14.6 / 8 .9999 98.7 / 99.95 2.5 / 15.3
LOOCV 1 100 / 100 20.7 / 12.2 .99999 98.7 / 100 3.6 / 19.5
CV 1 100 / 99.75 21.95 / 12 .99999 98.85 / 100 3.6 / 21.1
4.2. Pre-Classification
The idea is to exploit the a priori knowledge of the problem for selecting the
calibration or the training set, avoiding the problem of random partitions and hopefully
selecting examples that are more pertinent for training or calibration. The partition is
done through a method similar to a rule based classifier that has been used in a
previous work in ECT [9] for a different purpose. It has been shown in [1] that the
feature Į is a fairly good classifier. Going into detail, if Į is lower than a given
threshold ĮSCC, the inspected specimen is always an SCC. If it is higher than another
given threshold ĮFC (ĮSCC<ĮFC), the specimen is always an FC. Once these two
thresholds are heuristically defined, some of the experimental features vectors can be
directly classified. These directly classified features vectors are then used either to form
the calibration set or the training set. All the experimental features vectors which the
value of Į is included in the interval [ĮSCC; ĮFC] are kept for testing purposes and are
classified as it was previously done; however, the CV and LOOCV methods cannot be
applied anymore as the partition is now determined by the thresholds on Į. Results are
given in Table 4.
Results with simulated features are obviously better than using only experimental
features, regardless of the classifier. The reason is that during pre-classification using
only experimental features, more features vectors representative of FCs than SCCs are
pre-classified, introducing a bias. By increasing the value of the highest threshold, the
number of FCs in the training set is decreased, and the bias is reduced. This leads to
better results even if the abstention is usually higher (see Table 4). Increasing the lower
bound is impossible without misclassifying FCs as SCCs. When using simulated
features, the training set is constituted only of simulated features that are more evenly
distributed between FCs and SCCs. Moreover, the pre-classification acts on the
calibration, not directly on the training set.
When the k-NN algorithm is considered, slightly better results can be obtained
mainly through a lower abstention, by using fewer neighbors when using only
experimental features. With simulated features it is possible to reach 100% accuracy
and no abstention with a threshold as low as 0.6 (with ĮSCC=1.6 and ĮFC=4), which is
nonetheless very incautious behavior.
Using a pre-classification step allows to easily reach 100% of accuracy when using
simulated features. The high sensitivity of ANN and the relatively low amount of data
don’t allow for results as good as in the case of the k-NN from the point of view of the
abstention. Nonetheless, a decrease of the threshold on the confidence leads to
significantly lower abstention while maintaining 100% of accuracy. For example, with
T=.999, ĮSCC=1.6 and ĮFC=4, abstention is 20.6%.
Table 4. Classification results in terms of accuracy (Acc.) and abstention (Abs.) with a pre-classifier and a
cautious classifier using k-NN and ANN, respectively, with experimental (Exp.) and simulated (Sim.) features.
k-NN ANN
Acc. (%) Abs. (%) Acc. (%) Abs. (%)
Į Interval T T
[1.6;4] .9 44.1 / 100 10.5 / 13.2 .9999 74.1 / 100 65.7 / 29.2
[1.6;5] .9 59.5 / 100 10.6 / 19.1 .9999 99.4 / 98.6 53.3 / 29.2
[1.6;4] .7 44.4 / 100 5.3 / 5.3 .99999 89 / 100 76.3 / 38.4
[1.6;5] .7 58.1 / 97.6 8.5 / 12.8 .99999 100 / 100 69.3 / 36.5
5. Conclusion
The proposed methodology, from features extraction to decision, based on simple

models for cracks and a training of the classifiers with simulated data has been
validated. Results obtained with synthetic EC signals have been compared to the ones
obtained by using only experimental data. It has been shown that to extract features
from two ECT sensors rather than a single one and to make use of more elaborate
classification techniques than a threshold lead to significantly improved results in terms
of discrimination of SCCs and FCs. Nonetheless, a multi-sensor system has some
drawbacks such as an increased burden of the measurement task.
By taking into account the specificities of the problem to be solved, more adapted
classification procedures have been introduced: the cautious classifier and the pre-
classification. It has been shown that these two methods improve the results of the
classification itself and allow for more effective decision-making.
Future works include the reduction of the features vector dimension, the
improvement of the classification by using better models, combination of classifiers,
assessment of others classifiers such as Support Vector Machines [5], definition of a
true multi-probe EC sensor, and a more subtle rule-based pre-classification.
Acknowledgment
This study was performed under the sponsorship of JNES (Japan Nuclear Safety
Organization) open application project for enhancing the basis of nuclear safety.
References
[1] L. Janousek et al., “A Novel Non-destructive Method For Distinguishing Between Fatigue and Stress
Corrosion Cracks Using Electromagnetic Induction”, 13th Int. Conf. Nucl. Eng., Beijing, China, May
16-20, 2005
[2] S. Perrin, N. Yusa, K. Miya, “Electromagnetic Nondestructive Discrimination of Stress Corrosion and
Fatigue Cracks”, 5th Int’ l Conf. NDE, San Diego, U.S.A., May 10-12 2006.
[3] R. Polikar, L. Udpa, S. Udpa, V. Honavar, “An Incremental Learning Algorithm With Confidence
Estimation for Automated Identification of NDE Signals”, IEEE Trans. Ultrasonics, Ferroelectrics and
Frequency Control, vol. 51, no. 8, pp. 990-1001, August 2004.
[4] R. Smid, A. Docekal, M. Kreidl, “Automated Classification of Eddy Current Signatures During Manual
Inspection”, NDT&E International, vol. 38, pp. 462-470, 2005.
[5] S. Theodoridis, K. Koutroumbas, Pattern Recognition Third Edition, Academic Press, 2006
[6] B.P.C. Rao, B. Raj, T. Jayakumar, P. Kalyanasundaram, “An Artificial Neural Network for Eddy
Current Testing of Austenitic Stainless Steel Weld”, NDT&E International, vol. 35, pp. 393-398, 2002.
[7] Y.H. Kim et al., “Inversion of Experimental Eddy Current Testing Signals Obtained from Steam
Generator Tubes – Approach and Automated System”, Studies Appl. Electromag. Mech., vol. 25, pp.
85-92, 2005.
[8] C. Ferri, J. Hernández-Orallo, “Cautious Classifiers”, Proc. ROC Analysis in Artificial Intelligence, pp.
27-36, Valencia, Spain, 2004
[9] M. Das, H. Shekhar, X. Liu, R. Polikar, P. Ramuhalli, L. Udpa, S. Udpa, “A Generalized Likehood
Ratio technique for Automated Analysis of Bobbin Coil Eddy Current Data” , NDT&E International,
vol. 35, pp. 329-336, 2002.
IOS Press, 2007
Conductivity and Permeability Evaluation

on Type IV Damage Investigation
by Electromagnetic Method
Haiyan TIANa, Tetsuya UCHIMOTOa, 1, Toshiyuki TAKAGIa,

and Yukio TAKAHASHIb
a
Institute of Fluid Science, Tohoku University,
2-1-1 Katahira, Aoba, Sendai, Miyagi 980-8577, Japan
b
Central Research Institute of Electric Power Industry,
Iwado Kita 2-11-1, Komae, Tokyo 201-8511, Japan
Abstract. Type IV damage in power plants is a premature failure that occurs in the
softened heat-affected zone (HAZ) of weldment. Eddy-current method is applied
to evaluate electromagnetic properties of high-chromium ferritic steels in order to
investigate the material changes of HAZ and base metal in Type IV damage
evaluation. A pancake-coil impedance model based upon eddy current theory is
proposed to evaluate conductivity and permeability. A series of HAZ and base
metal specimens of Modified 9Cr-1Mo steel in different heat treatment conditions
are experimentally measured and then their conductivity and permeability are
estimated. The evaluated results illustrate electromagnetic properties between heat
treatment specimens in different test time can be clearly distinguished by presented
approach.
Keywords. Electromagnetic, conductivity, permeability, HAZ, Type IV damage
1. Introduction
In electric power industry, high-chromium ferritic steel (such as Modified 9Cr-1Mo

steel) is widely used due to low thermal expansion, high thermal conductivity and good
steam corrosion resistance. On the other hand, some studies for this steel have pointed
out that Type IV damage, which is happened in the softened heat-affected zone (HAZ)
of weldment and is caused by different creep strength between weld metals HAZ and
base metal, is a premature failure and often termed mid-life cracking in service [1-4]. In
order to investigate material changes of HAZ and base metal in high temperature
service environment and prevent weldment from Type IV damage, evaluation of the
electromagnetic properties such as conductivity and permeability by non-destructive
evaluation method is proposed as one of investigation approaches to monitor the
microstructure change.
1
Corresponding Author: Tetsuya Uchimoto, Institute of Fluid Sciene, Tohoku University, 2-1-1
Katahira, Aoba-ku, Sendai, Miyagi 980-8577 Japan; E-mail: uchimoto@ifs.tohoku.ac.jp
100 H. Tian et al. / Conductivity and Permeability Evaluation on Type IV Damage Investigation
r2
Pancake coil
r1
l2
l1
Metal plate d
Fig. 1 Pancake impedance coil model.
It is possible for eddy-current method to provide a non-destructive testing in

service to estimate conductivity and permeability, and carry out a direct measurement
of electromagnetic properties without calibration standards. In this paper, a pancake-
coil impedance model is proposed, which is derived from partial differential equations
(PDE). Comparing with well established theory by Dodd and Deeds [5], the present
model is simplified, but can solve the "overflow" problem that sometimes occurs in the
numerical calculation using conventional Dodd’s formula. In Dodd’s formula, an
exponential term that includes relative permeability exceeds the maximum floating-
point number of a given precision (single or double) in the case of ferritic materials
because of its higher relative permeability than linear material.
Furthermore, 16 pieces of Modified 9Cr-1Mo steel specimens of HAZ and base
metal with different heat treatment conditions are measured by impedance analyzer.
Then, their conductivity and permeability are evaluated by searching for an impedance
match with theoretical valuescalculated by proposed formula. The evaluated results
indicate material’s electromagnetic properties in different heat treatment conditions. It
is prospected for eddy-current method as a non-destructive evaluation to estimate the
conductivity and permeability of ferromagnetic materials.
2. Pancake-Coil Impedance Model
For a pancake coil parallel to a metal plate, as shown in Fig.1, a closed-form solution in
Eqs. (1) and (2) can be derived from Maxwell’s equations considering both
ferromagnetic and non-ferromagnetic conductors by solving the magnetic vector
potential A in an electromagnetic field.
j ZSP 0 n 2 f 1
Z ³ I 2 ( r2 , r1 ) [ 2 ( l 2 l1 ) D 1
( 2 e D ( l 2 l1 ) ) 2
( l 2 l1 ) 2 ( r2 r1 ) 2 0 D5 (1)
D E1
( e 2 D l2 e 2 D l2 2 e D ( l2 l )
1
) ]d D
D E1
jZSP 0 n 2 f 1 (2)
Z air 2 2 ³ I 2 ( r2 , r1 ) [ 2(l 2 l1 ) D 1 ( 2e D ( l 2 l1 ) 2)]dD
(l 2 l1 ) ( r2 r1 ) 0 D5
H. Tian et al. / Conductivity and Permeability Evaluation on Type IV Damage Investigation 101
Table 1. Numerical comparison between proposed model and Dodd’s formula
Assumed ZP 0 P rV Impedance () Impedance ()

parameters (proposed model) (Dodd’s formula)
5.25649163714297 + 5.25648012001690 +
106
96.49174352595570i 96.49173915532485i
10.1623168357060 + 10.16231683570060 +
d = 8 mm 107
r1 = 1 mm 115.5950043219815i 115.5950043219816i
6.56731369303464 + 6.56731369303463 +
r2 = 1 mm 108
l1 = 0.1 mm 129.3217344475126i 129.3217344475127i
2.79398351489306 + 2.79398351489307 +
l2 = 2.1 mm 109
f = 100 kHz 135.5543623534413i 135.5543623534413i
0.9880977197488859 +
T = 300 turns 1010 Overflow
137.7556024650322i
11 0.3245567948216825 +
10 Overflow
138.4676091849045i
j ZSP 0 n 2 f 1
Z Dodd 2 2 ³ I 2 ( r2 , r1 )[ 2 ( l 2 l 1 ) D 1
( 2 e D ( l 2 l1 ) )
( l 2 l 1 ) ( r2 r1 ) 0 D5 (3)
2 2 2 2 2D d
( E 1 D ) (D E 1 )e
2 ( e 2 D l 2 e 2 D l1 2 e D ( l 2 l1 ) ) ]d D
( E 1 D ) 2 (D E 1 ) 2 e 2 D 1 d
Here
E1 D 1 / P r , D1 (D 2 jZP 0 P r V ) 1 / 2 ,
and
r2
I ( r2 , r1 ) D 2
³ r0 J 1 (D r0 ) dr 0 .
r1
P0 and Pr are permeability of free space and relative permeability of metal plate
respectively. V is conductivity of metal plate, n is the number of turns in the pancake
coil, J1 ( x) is the first-order Bessel function of the first kind. Eq.(2) is the air-
impedance without measuring a metal plate. The impedance formula in Eq.(3) was
derived by Dodd and Deeds. Since the nonlinear and hysteresis effects are fairly small
for low currents case, this formula can be used for ferrite material [5][6]. However, on
the other occasions when relative permeability is big, the term of e 2D1d that includes
relative permeability is too large so as to exceed the maximum floating-point number
and then the “overflow” error in numerical calculation happens. In order to improve
this flaw, equivalent equations shown in Eq.(1) and (2) are proposed to overcome
“overflow” through solving partial differential equations. In this paper, mathematic
demonstration will not be presented since the process to solve PDE is very complex.
The analogical deduction method is available in Ref. [5] and [6]. On behalf of
mathematic demonstration, a comparison between Eq.(1) and Eq.(3) by numerical
calculation shown in Table 1 validates the equivalence of proposed equation.
3. Impedance Coil and its Experiment Correction
An air-cored probe is fabricated by winding a coil. Parameters of the coil are listed in
Table 2. Parameters of pancake-coil
r1 Inner radius 0.5 mm

r2 Outer radius 1.0 mm
l1 Lift-off 0.1 mm
l2 Height of coil 2.1 mm
n Number of turns 300

f Exciting frequency 62 kHz
Z L0 air-reactance 23.008 ȍ
R 0 DC resistance 15.02 ȍ
Table 2. Sixteen specimens of Modified 9Cr-1Mo steel in different heat treatment

conditions are listed in Table 3. Because of the non-ideal coil behavior, two corrections
have to be considered when processing the experiment data. First, in Eq. (1), the DC
resistance of a coil (when Z 0 ) is zero. However, any practical coil has DC
resistance R0 shown in Table 1. Therefore the experimental resistance R exp should be
corrected to match the theoretical value R cal calculated by Eq.(1). The corrected
resistance R exp in Eq.(4) is that measured resistance R mea minus DC resistance R0 .
Secondly, since the coil was wound in layers, it is not an absolute ideal pancake
inductance coil. The self-reactance between the measured and theoretically calculated
value by Eq.(2) are different. The error between them is 1.32%. This error can be
solved after normalizing the reactance by dividing the self-reactance in both calculated
and measured data. In Eq.(5), ZL0 mea is measured air-reactance, X norm exp is normalized
experimental reactance. In Eq.(6), Z air cal is calculated air-reactance by Eq.(2), ZLcal is
imaginary part calculated by Eq.(1). X norm cal is normalized calculated reactance. Finally,
depending on seeking an approximation between theoretical and experimental
impedance in Eqs.(7) and (8) with assumed conductivity and permeability, the
evaluated conductivity and permeability are determined.
R exp R mea R0 (4)

exp mea mea
X norm jZL / jZL0 (5)
cal cal cal
X norm jZ L / Z air (6)
R cal # R exp (7)
cal exp
X norm # X norm
(8)
4. Experiment Data
The measurement device is Hewlett Packard 4294A impedance analyzer. An exciting

current of 10 mA with frequency 62 kHz was applied to the pancake impedance coil.
Table 3. Specimen list of Modified 9Cr-1Mo steel
Heat treatment Base material HAZ

No. No.
condition Size (mm) Size (mm)
500 0C600 hr B01 33156t H01 25155t

500 0C1400 hr B05 20155t H05 15105t
550 0C 600 hr B02 33155t H02 25145t
550 0C1400 hr B06 20155t H07 15105t
600 0C 600 hr B03 34155t H03 25134t
600 0C1400 hr B07 20155t H07 15105t
650 0C 600 hr B04 33156t H04 15145t
650 0C1400 hr B08 20155t H08 15105t
㪉㪌㪅㪏㪌 Base 600h 500C

Base 600h 550C
㪉㪌㪅㪏㪇 Base 600h 600C
Base 600h 650C
㪩㪼㪸㪺㫋㪸㫅㪺㪼㩷㩿㱅㪀
㪉㪌㪅㪎㪌 HAZ 600h 500C

HAZ 600h 550C
HAZ 600h 600C
㪉㪌㪅㪎㪇 HAZ 600h 650C
Base 1400h 500C
㪉㪌㪅㪍㪌 Base 1400h 550C
Base 1400h 600C
㪉㪌㪅㪍㪇 Base 1400h 650C
HAZ 1400h 500C
HAZ 1400h 550C
㪉㪌㪅㪌㪌
HAZ 1400h 600C
㪈㪌㪅㪌㪇㪈㪌㪅㪌㪌㪈㪌㪅㪍㪇㪈㪌㪅㪍㪌 HAZ 1400h 650C
㪩㪼㫊㫀㫊㫋㪸㫅㪺㪼㩷㩿㱅㪀
Fig. 2 Measured impedance
The measured resistance is shown in x-axis of Fig.2. The data needs to be

processed by Eq.(4) to obtain corrected experimental resistance for comparing with
theoretical values. The measured reactance is shown in y-axis of Fig.2. The data also
needs to be normalized by air-reactance using Eq. (5). Both of corrected resistance and
normalized reactance are compared with theoretically calculated value using Eq. (1).
When the best match between corrected experiment value and theoretically calculated
value is found with an assumed conductivity and permeability, the evaluated
conductivity and permeability are accepted.
5. Evaluated Results and Discussion
The evaluated conductivity and permeability are shown in Figs.3 and 4, respectively.
The electromagnet properties of heat treatment specimens in 600 hours are obviously
different from ones in 1400 hours. When adding the time of heat treatment, the
Fig. 3 Evaluated conductivities of heat treatment specimens.
Fig. 4 Evaluated relative permeabilities of heat treatment specimens.
conductivity increases but the permeability decreases. The change of electromagnet

properties can be clearly distinguished in accordance to the time of heat treatment.
However, in the condition of having the same time of heat treatment, a qualitative
conclusion still can not be given yet even though the evaluated results differs each
other. The further studies to improve the accuracy in this condition are required and
still in progress.
In order to validate the evaluated conductivities, a four-point conductivity
measurement device is applied to measure these 16 specimens. The composition of this
device and the dimension of the test sample are illustrated in Fig. 5 (a) and (b). The
conductivities are calculated in accordance to Eq.(9) and are supplied in Fig.6.
Conductivi ty (V )
1

Length ( L ) I L

(9)
Resis tance ( R ) Aera ( S ) V S
The measured conductivities by the four-point measurement device are in the same
quantitative level with evaluated results. This information to some extent supports the
evaluated results by the proposed impedance model. But on the other hand, this device
can not provide enough accuracy digits to distinguish the difference from 600 and 1400
hours heat treatment conditions.
6. Summary
An impedance coil model and its experiment correction method are presented for a
prediction of the conductivity and permeability of ferritic steels in Type IV damage
investigation. A series of heat treatment HAZ and base metal of Modified 9Cr-1Mo
steel in different experiment conditions were measured and then their conductivities
Fig. 5 (a) Composition of four-point conductivity measurement device and (b) sample dimension.
Fig. 6 Evaluated conductivities by four-point measurement device.
and permeabilities are evaluated by the proposed approaches. The evaluated results
reveal that the conductivity increases but the permeability decreases when adding the
time of heat treatment. The change of electromagnet properties can be clearly
distinguished according to the heat treatment time. The evaluated results are validated
by a four-point conductivity device. Therefore the proposed eddy-current method in
this paper is efficient as a nondestructive evaluation approach to estimate the
electromagnetic properties of ferritic steel.
Acknowledgement
This work was performed under the sponsorship of Ministry of Education, Culture,
Sports, Science and Technology.
References
[1] Y. Takahashi, “Study on Type-IV damage prevention in high temperature welded structures of next-
generation reactor plants, part II fatigue and creep-fatigue behavior of welded joints of modified 9Cr-
1Mo steel”, Proceeding of ASAEM pressure vessels and piping division conference, 2006.
[2] Y. Takahashi, M. Tabuchi, “Study on Type-IV damage prevention in high temperature welded structures
of next-generation reactor plants, part I fatigue and creep-fatigue behavior of welded joints of modified
9Cr-1Mo steel”, Proceeding of ASAEM pressure vessels and piping division conference, 2006.
[3] S.K. Albert, M. Matsui, T. Watanabe, H. Hongo, K. Kubo, and M. Tabuchi “Variation in the Type Φ
cracking behaviour of a high Cr steel weld with post weld heat treatment”, International Journal of
Pressure Vessels and Piping 80, 2003.
[4] C.D. Lundin, P. Liv and Y. Cui, “A Literature review on Characteristics of High Temperature Ferritic Cr-
Mo Steels and weldments”, WRC Bulletin, No.454, 2000, pp.1-36.
[5] C.V. Dodd and W.E. Deed, “Integral Solutions to some Eddy Current Problems”, International Journal of
Nondestructive Testing, Vol.1, 1969, pp. 29-90.
[6] Jack Blitz, Electrical and magnetic methods of nondestructive testing, IOP Press, 1991, pp.89-116.
Industrial Applications and New Methods
IOS Press, 2007
Evaluation of Fatigue Loaded

Conducting Structures Using Selected
Electromagnetic NDT Methods
Tomasz CHADY a,1, Ryszard SIKORA a, Grzegorz PSUJ a, Przemysáaw àOPATO a
Masato ENOKIZONO b, Yuji TSUCHIDA b
a
Szczecin University of Technology, Department of Electrical Engineering , Poland
b
Oita University, Faculty of Engineering, Oita, Japan
Abstract. The purpose of this paper is to present the results of observation of

fatigue failures forming under cyclic dynamic loading using selected
electromagnetic NDT methods. To evaluate fatigue damage a multi-frequency
eddy current and induced current field measurements methods were used.
Keywords. Multi-frequency eddy current method, Injected current field

measurements method
1. Introduction
It is known that metal structures subjected to a repetitive or fluctuating stress will

eventually fail. Considering that almost all today’s constructions are subjected to
repeated loading and vibration, it becomes a necessity to estimate a fatigue failure
before it can lead to a catastrophe. In this paper selected electromagnetic
nondestructive testing methods are utilized to detect changes in conducting
nonmagnetic structures exposed to a cyclic dynamic loading. Inspection of such
structures is more difficult than ferromagnetic, because fatigue has influence only on
one property – conductivity. Therefore only a limited number of NDT methods is
applicable. Later, results of all techniques can be fused to enhance the overall
performance of an inspection and to achieve a more accurate assessment of a structural
integrity.
In order to select the best methods, extended experiments with a fatigue loaded
sample made of aluminum were carried out. During the evaluation process several
factors were taken into consideration: sensitivity, spatial resolution, versatility and
noise tolerance.
1
Corresponding Author: Tomasz Chady, Szczecin University of Technology, Department of Electrical
Engineering , ul. Sikorskiego 37, 70-313 Szczecin, Poland; E-mail: tchady@ps.pl
110 T. Chady et al. / Evaluation of Fatigue Loaded Conducting Structures
Digital Radiography Unit

DR
X-Ra y
Excitation
Unit Control & Fusion Center Data Acquisition Unit
Function PC A/D
Generator Computer Converter
Field Measuring Unit (ICFM , MF-ICFM)
Power
Amplifier Bartery Amplifier
DC & AC current
Eddy Current Measuring Unit (MF-EC)

LP HP
Filter Amplifier Filter Amplifier
Figure 1. Block scheme of the computerized NDT system
2. Measuring System
Measurements were carried out using a computerized multi-purpose laboratory NDT

system developed at Technical University of Szczecin [1]. It is based on the PC class
computer connected to generators, amplifiers and various data acquisition devices
(Fig. 1). The system can be used to perform nondestructive tests using various
electromagnetic techniques and is fully scaleable. It means that the user may add
additional devices and change methods or transducers to fulfill wide range of
requirements. Most of the experiments were carried out utilizing multi-frequency
excitation and spectrogram technique which was proposed in [2]. Idea of this technique
is to use a complex signal containing selected sinusoidal components as an excitation
signal and a spectrogram for precise defect characterization. Very promising results
were achieved using a flux leakage and eddy current method in the case of
ferromagnetic structures, leading us to utilize similar methods for conducting structures
evaluation.
2.1 Multi-frequency Eddy Current Method (MF-EC)
A multi-frequency excitation method and an eddy current differential transducer were

used for testing. The transducer consists of a cylindrical ferrite core with five
symmetrically placed columns. A pickup coil is wounded on the central column and
four excitation coils are placed on remaining columns in pairs, on two perpendicular
axes (Fig. 2). Both pairs generate in the pickup coil oppositely directed magnetic fluxes.
The resulting flux in the pickup coil is close to zero in equilibrium state. An output
signal depends on difference of fluxes Ix and Iy. The proposed probe has a very good
sensitivity. Differential configuration of the transducer enables us to detect minor flaws
with the SNR rate higher than in the case of the traditional eddy current probes (SNR =
20 dB was achieved in case of 10% outer flaw in 3 mm thick aluminum plate). The
optimal testing frequency range (from 700 Hz to 4 kHz) was selected after preliminary
tests of artificial EDM flaws having different depths. The probe was moved over the
specimen in x and y-direction in steps of 0.5 mm. Detailed description of the
advantages of the proposed transducer can be found in [3].
T. Chady et al. / Evaluation of Fatigue Loaded Conducting Structures 111
a) b)
I P
Ip Iy
Ix
EA EB EC ED
S
n
Ix Iy
Uout= - j Z n (Ix - Iy)
Figure 2. View and simplified electrical scheme of the transducer; F – ferrite core,
EA…ED – excitation coils, (S) – pickup coil, (Ix, Iy) - magnetic fluxes, P – potentiometer that controls
magnetic fluxes Ix and Iy, n – number of turns in the pickup coil S
2.2 Injected Current Field Measurement Method (ICFM)
All magnetic methods of flaw detection rely on the measurement of the magnetic flux
leakage field near the surface of the material, which is caused by the presence of the
flaw. The ACFM (Alternating Current Field Measurement) technique is a widely used
NDE method. It was developed during the 1980's from the ACPD (Alternating Current
Potential Drop) technique. The method can be used to detect and size defects in both
magnetic and non-magnetic materials. The basis of the ACFM technique is that a
uniform current flows near the surface of the specimen. If a surface breaking crack is
present, the current is disturbed. A magnetic field associated with the electrical field
and the magnetic field disturbances (associated with the electrical current disturbances)
can be measured using magnetic field sensors. Usually, the current in ACFM method is
induced rather than injected. In this paper the current is injected into the sample
through electrodes in longitudinal direction of the sample (Fig. 3). Therefore we used
the term Injected Current Field Measurement (ICFM) instead of ACFM. In order to
find an optimal method various configurations were considered and tested.
First, Direct Current ICFM (DC-ICFM) technique was evaluated. The DC-ICFM
method involves passing a constant current ( 3A ) through the specimen volume to be
inspected. In the next method, an alternating current was injected into the specimen. In
order to improve reliability of defect sizing procedure a multi-frequency excitation was
utilized (IMAX =150 mA, the frequency range was from 400 Hz to 8 kHz). In all cases
the resulting magnetic field (a tangential component) was measured by high sensitivity
GMR sensors (manufactured by NVE). Two sensors were applied: the extremely
sensitive AAH-series GMR magnetometer and the ABH-series GMR gradiometer. The
ABH sensor enables sensing of small gradients in large and small magnetic fields. The
ability to detect only magnetic gradients allows low sensitivity to external sources of
uniform magnetic field. Unfortunately, the experiment shows that the gradiometer was
useless in this specific application. The gradient of magnetic field disturbances caused
by the crack in nonmagnetic materials was too small to generate sufficient signal in the
sensor. The magnetic field was measured by moving the sensor in x (longitudinal) and
y-direction in steps of 0.5 mm.
3. Experiment Description and Results of Measurements
The purpose of the experiment was to evaluate the changes of the material’s structure
during the fatigue loading and no measurements of artificial notches were done in the
final tests. All experiments were carried out for a planar specimen made of aluminum
alloy (Fig. 3). The minimum yield strength of this material was around 245 MPa. The
sample was tensile deformed in the longitudinal direction under cyclic stress using
cyclic loading machine. The testing time was divided into loading periods. Each period
had 30000 of loading cycles. After each loading period the specimen was examined by
the eddy current method. Figure 4 shows the signals obtained after each loading period.
Electrode Thickness: 3 mm Electrode

45 mm
35 mm
Measurement area
30 mm
255 mm
Figure 3. Shape of the specimen and photo of computer controlled cyclic loading machine.
After 1 loading period (AL_1) After 6 loading periods (AL_6)

y [mm]
After 2 loading periods (AL_2) After 7 loading periods (AL_7)

y [mm]

y [mm]

y [mm]

y [mm]
x [mm] x [mm]
Figure. 4 Results of eddy current measurements carried out after each loading period. Testing frequency f =
1.6 kHz. A spatial derivative of the signal with respect to x-coordinate (wU(x,y)/(wx)) is plotted.
It is possible to observe that the measured signals have strong disturbances

concentrated in the one place, since very early stage of the fatigue process. Finally, in
this place the crack aroused and the experiment was stopped. The sample achieved this
way was named AL_10 (Fig. 4). A very detailed inspection of the sample was carried
out using all available methods. These tests allow us to:
x find an optimal testing parameters (i.e. excitation frequency),
x detect all the cracks and also other minor changes of material properties caused by
the fatigue.
3.1 Radiographic Examination of the Test Sample
In order to achieve precise information about the cracks existing in the sample
photography of the surface was taken (Fig. 5). Next, the X-ray examination was carried
out. The achieved radiograph is shown in Fig. 6.
These figures provide us detailed information (“ground truth”) about the position
and structure of the crack. One can observe that beside surface braking crack there is
also a hidden part of this crack visible in the radiograph (Fig. 6).
Figure. 5 Photo of the tests sample and enlargement of the crack area
Figure. 6 Radiograph of the tests sample and enlargement of the crack area
3.2 DC Injected Current Field Measurement Method (DC-ICFM)
In case of Direct Current ICFM (DC-ICFM) technique a constant current (3A) was
passed into the specimen through the electrodes attached to the end of the sample (Fig.
3). At first, the tangential component of the magnetic field near the surface of the
sample was measured by the gradiometer (ABH001). The gradient of the induced
magnetic field is relatively small even around the crack and therefore we achieved a
very low level of the output signal from gradiometer (Fig. 7). The signal was corrupted
by noises and even if the crack can be detected the results are insufficient in practical
applications. Higher probability of detection was achieved in case of measurements
carried out using a very high sensitivity (18 mV/V Oe) GMR magnetometer (AAH002).
The results shown in Fig. 8 prove that it is possible to detect the crack and the influence
of the sample’s edges can be reduced after numerical calculation of the signal gradient
(Fig. 8 and Fig. 9). Figure 9 shows signal achieved for the reference unloaded sample.
From comparison of signals derivative presented in Fig. 8 and Fig. 9 one can observe
that the signal amplitude due to the crack presence is about 6 times higher than in case
of edge influence. The noises are visible but the SNR is more than 40dB. The problem
with external interfering fields may arise in the industrial environment and in case of
DC field it will be very difficult to resolve it. Therefore Alternating Current ICFM
method was considered in the next section.
a) b)
y [mm]
x [mm] x [mm]
Figure. 7 Signals obtained from DC-ICFM method and GMR gradiometer used for magnetic field
measuring; sample AL_10; a) signal; b) derivative of the signal with respect to x-coordinate
a) b)
y [mm]
x [mm] x [mm]
Figure. 8 Signals obtained from DC-ICFM method and GMR magnetometer used for magnetic field
measuring; sample AL_10; a) signal; b) derivative of the signal with respect to x-coordinate
a) b)
y [mm]
x [mm] x [mm]
Figure. 9 Signals obtained from DC-ICFM method and GMR magnetometer used for magnetic field
measuring; unloaded sample AL_0; a) signal; b) derivative of the signal with respect to x-coordinate
f [Hz]
x [mm]
Figure. 10 Spectrogram obtained from MF-ICFM method and GMR magnetometer used for magnetic field
measuring; sample AL_10; y = -14 mm;
a) b)
y [mm]
x [mm] x [mm]
Figure. 11 Signals obtained from MF-ICFM method and GMR magnetometer used for magnetic field
measuring; sample AL_10; testing frequency f = 3 kHz; a) signal; b) derivative of the signal with respect to x-
coordinate
3.3 Multi-frequency Injected Current Field Measurement Method (MF-ICFM)
The high sensitivity GMR magnetometer AAH002 was utilized to measure the
magnetic field in case of the Multi-Frequency ICFM method (MF-ICFM). The
frequency range of the AC current injected into the sample was decided after
preliminary tests. Figure 10 shows the spectrogram [2] achieved during the linear scan
along the x-axis (longitudinal direction of the sample) over the crack. It is possible to
observe a different frequency characteristic in case of the crack (-20< x <8) and an edge
(10< x <20). After gradient filtering (Fig. 11) the signal changes caused by the edge
nearly disappears, while the signal caused by the crack are significantly enhanced.
Thanks to the applied FFT algorithm the achieved noise level is very low (SNR>50 dB).
3.4 Multi-frequency Eddy Current Method (MF-EC)
The MF-EC method with spectrogram was proved [1,2,3] to be very reliable and
informative NDE technique. The results shown in this paper confirm that it can be also
successfully applied to evaluate materials with fatigue degradation. Figure 12 shows
the spectrogram achieved during the linear scan along the x-axis over the cracked area.
f [Hz]
x [mm]
Figure. 12 Spectrogram obtained from MF-EC method; sample AL_10; y =-8 mm;
a) b)
y [mm]
x [mm] x [mm]
Figure. 13 Signals obtained from MF-EC method; testing frequency f = 1.6 kHz; sample AL_10
a) signal; b) derivative of the signal with respect to x-coordinate
a) b)
y [mm]
x [mm] x [mm]
Figure. 14 Signals obtained from MF-EC method; testing frequency f = 1.6 kHz; unloaded sample AL_0
a) signal; b) derivative of the signal with respect to x-coordinate
a) b)
y [mm]
x [mm] x [mm]
Figure. 15 Signals obtained from MF-EC method; testing frequency f = 1.6 kHz; selected part of the sample
AL_10; a) signal; b) derivative of the signal with respect to x-coordinate
It is possible to observe that also in the case of fatigue degradation the multi-
frequency excitation and spectrogram methods offer opportunity to distinguish signals
of cracks and edges. The noise level is similar to the level achieved using MF-ICFM
method. The disadvantage of he MF-EC method is a stronger influences of edges (Fig.
14) and a lower spatial resolution (a blurring effect). Signals shown in Fig. 15 prove
that this technique is more sensitive than MF-ICFM method and enables to detect also
the subsurface minor materials’ degradation (-5< x <10).
Acknowledgment
This work was supported in part by the State Committee for Scientific Research,
Poland, under the Grant no: 3T10A 017 30 (2006-2009).
References
[1] T. Chady et al., Proceedings of XIII International Symp. on Theoretical Electrical Engineering, Lviv
(2005), p. 333-336.
[2] T. Chady, M. Enokizono, Eddy current testing of stainless plates by using matrix sensor, Proceedings of
the 7th Magnetodynamics Conference, April 24, 1998, pp.107-110.
[3] T. Chady, P. Lopato, R. Sikora, M. Komorowski, High Sensitivity Differential Eddy Current Transducer,
12th ISEM Conference 12-14 Sept. 2005, Bad Gastein, Austria, pp. 308-309
IOS Press, 2007
Evaluation of Circumferential Crack

Location in Pipes by Electromagnetic
Waves
Kavoos ABBASIa,1, Satoshi ITOa , Hidetoshi HASHIZUMEa, Kazuhisa YUKIa
a
Dept. of Quantum Science and Energy Engineering, Tohoku University, Sendai, Japan
Abstract. Crack detection is one of the most important issues in large

component such as nuclear power plant. A NDT method using electromagnetic
waves has strong possibility for crack detection in large pipes. In this study,
TM01-mode of electromagnetic wave is used for detecting a circumferential
crack and time of flight (TOF) of the electromagnetic wave in straight pipe with
crack is evaluated to determine the crack location.
Keywords. Nondestructive testing, Electromagnetic wave, circumferential crack,
inspected pipe, Time of flight (TOF)
1. Introduction
Pipes carrying fluids are frequently used in many applications such as power plants,
gas and oil transportation, etc. Many techniques are available to detect cracks in such
components. The conventional methods are ultrasonic testing (UT), eddy-current based
method testing (ECT), X-ray, electric impedance technique, etc. For example, the UT
method with various kinds of advanced probes can precisely determine the size of
crack [1]. However these methods require full scale scanning of component and for
large component these method are costly and very long inspection time is needed [2-4].
This fact motivates to develop alternative method. The NDT method using microwave
has availability to detect crack and determine crack location [5].
In a previous study, experimental results showed that the electromagnetic wave of
circular TM01-mode has possibility to detect circumferential crack, and response of
electromagnetic wave to the crack at frequency range including cutoff frequency of
straight pipe was confirmed [6].
In the present paper, a new experimental analysis is performed to determine crack
location for crack with different location in straight pipes.
2. Theoretical background
Since this study is carried out to detect a circumferential crack in a straight pipe, the
circular TM-mode which is suitable for detection of circumferential crack should be
1Corresponding Author: Kavoos Abbasi, Tohoku University, Aramaki-Aza-Aoba 6-6-01-2, Sendai, Japan;
E-mail: abbasi@karma.qse.tohoku.ac.jp
118 K. Abbasi et al. / Evaluation of Circumferential Crack Location in Pipes
propagated in straight pipe with crack. The TM-modes are characterized by fields with
Hz =0 while Ez z 0 where z indicates the direction of propagation (in case of cylindrical
pipe, z corresponds to axial direction of the pipe). When an electromagnetic wave is
propagated in the waveguide, the surface current is produced in the inner surface of the
waveguide and flows in the direction of the propagation. When this surface current
flows in the straight pipe with circumferential crack, the crack prevents the surface
current from flowing. Consequently some parts of the incident wave are reflected. The
reflected wave has information relating to the crack.
Group velocity of the wave (vg), which is important parameter for calculating TOF,
is given by the following equation in terms of frequency.
2
c § fc ·
v 1 ¨ ¸ (1)
g P H ¨ f ¸
R R © ¹
where c,PR,HR and fc is the light velocity, relative permeability, relative permittivity and
cutoff frequency respectively.
A schematic diagram of the experimental setup is given in Figure 1. The

electromagnetic wave, which is generated by a network analyzer, passes through the
mode converter via the coaxial line. The mode converter is formed by joining the
rectangular waveguide to the circular waveguide and converts the rectangular
TE10-mode to the circular TM01-mode which is suitable for detection of circumferential
cracks.
Figure 2 shows the electric field of TE10-mode and TM01-mode generated in the
mode converter. The experimental system shown in Figure 1 has two mode converters
and each mode converter can be connected to one port of the network analyzer. This
setup makes it possible to measure the characteristics of both the reflected and
transmitted waves. The TM01-mode wave in the system can be resonated by moving the
plunger inside the circular waveguide with the following equation:
m
l Og . (2)
2
where “l” is the distance between plunger surface and center of rectangular waveguide
and m is an integer.
Similar pipes, one with crack and the other without crack are used as inspected
pipe in this experiment. The inspected pipes are made of SUS-304, with length of 1200
mm, inner diameter of 34 mm and outer diameter of 38 mm. The open circumferential
crack whose width is 0.3 mm and depth of 2 mm (Thickness of the inspected pipe) is
made by the wire cutting method on one half of the circumference of the pipe which is
shown in Figure 3(a). The frequency range is chosen based on the experimental results
[7] and the frequency band of the rectangular wave guide.
K. Abbasi et al. / Evaluation of Circumferential Crack Location in Pipes 119
Figure 1. Experimental system for detection of circumferential crack in

the straight pipe
110.9 mm Length, l
Plunger
59.5 mm
Circular TM01-mode
Electric field
Rectangular TE10-mode
Figure 2. Distribution of electric field in mode converter
(a) (b)
Figure 3. (a) Inspected pipe with circumferential crack and (b) Pipe with circumferential
crack when aluminum foil is inserted into the crack
Since the purpose of this study is to determine time of flight, which can be used to
predict crack location, it is necessary to evaluate the signals in time domain mode by
getting Inverse Fast Fourier Transform (IFFT) of the signal in the frequency domain.
4. Methods and Results
4.1 Estimation of Time of Flight (TOF)
Firstly, in order to show accuracy of calculated TOF, aluminum foil is inserted into the
crack (Figure 3(b)) to produce a large reflection at the location of crack. The results are
shown in Figure 4 and Figure 5 for a crack location of 400 mm and 800 mm
respectively. Horizontal axis indicates plunger position and vertical axis indicates time
of flight (TOF). In these figures, difference of reflection coefficient ('S) between the
case without crack and the case with aluminum foil are presented as black and white
images. Color bar of the each image indicates the difference of reflection coefficients.
As the figures show, the differences of reflection coefficients are clearly changed at
specific time. This time indicates the TOF of reflected wave from the crack.
By knowing the group velocity in each part of experimental system (waveguides)
at given maximum frequency, the TOF for each crack location can be calculated. The
total path for traveling wave is shown in Figure 1 as A-B-C-B-A. The white line in
Figure 4 and Figure 5 shows the TOF which is obtained by the calculation. As the
figures indicate, there is good agreement between experimental result and calculation.
This fact indicates the accuracy of the calculations.
4.2 Evaluation of Crack Location
Next, crack location is evaluated by measuring TOF of reflected wave. The results of
two frequency range and two plunger positions are presented. Figure 6 displays the
difference of reflection coefficient between the case without crack and the case with
crack when the plunger is located at 90 mm and frequency range is 5 GHz ~ 15 GHz.
The calculated and predicted TOFs are specified by the solid and dashed arrows in each
figure respectively. With regard to the largest peak in each figure that appears at time
15 15
0.02 0.02
Time [nsec]
Time [nsec]
0.015 0.015
10 10
0.01 Calculated time 0.01

5 Calculated time 5
0.005 0.005
0 0
100 120 140 160 180 100 120 140 160 180
Plunger position [mm] Plunger position [mm]
Figure 4. Difference of reflection coefficient, Figure 5. Difference of reflection coefficient,
for a crack located at 400 mm for a crack located at 800 mm
larger than 15 ns, the TOF of crack is the point when difference of reflection coefficient
starts to increase before these peaks. By comparing these figures it can be seen that, the
response of electromagnetic wave to the crack located at 400 mm is relatively high to
measure and TOF can be predicted easily. In some figures several candidates can be
considered as predicted TOF of the crack. As the figures show, several small peaks also
appear with time region between 6 ns and less than TOF of the crack. These peaks that
are delineated by circle or ellipse in each figure are produced due to connection of the
inspected pipe and tapered waveguide in Figure 1.
Calculated time= 8.43 ns Calculated time= 11.4 ns
(a) (b)
Figure 6. Difference of reflection coefficient, for crack located at (a) 400 mm and (b) 800 mm ,
when plunger position = 90 mm and frequency range is 5 ~ 15 GHz.
(a) (b)
Figure 7. Difference of reflection coefficient, for crack located at (a) 400 mm and (b) 800 mm,
Figure 7 shows the result, when plunger position is 90mm and frequency range is 5
GHz ~12 GHz. By comparing these results with those in Figure 6, it can be seen that by
reducing frequency range some undesired signals till 5 ns are reduced considerably.
The next results are obtained when plunger position is changed to 134 mm. Figure
8 shows the result for two mentioned crack location when frequency range is 5 GHz ~
15 GHz and plunger is located at 134 mm. In comparison with Figure 6, there is less
possibility to determine TOF of each crack location. By choosing frequency range of
Calculated time= 9 ns Calculated time= 12 ns
(a) (b)
(a) (b)
Table 1. Comparison of calculated and predicted TOF
Crack Plunger Frequency Calculated Predicted Predicted crack

location position Band TOF [ns] TOF [ns] location [mm]
[mm] [mm] [GHz]
400 90 5 ~ 15 8.43 8.1 360
400 90 5 ~12 8.70 8.0 320
400 134 5 ~15 9.00 8.0 275
400 134 5 ~12 9.29 8.6, 13 320, 860
800 90 5 ~ 15 11.40 8.5, 12.5 409, 945
800 90 5 ~12 11.93 8,12.4,15.5 312, 858, 1240
800 134 5 ~15 12.0 9, 14.0 410, 1050
800 134 5 ~12 12.52 11.2, 13.0 650, 860
5GHz~12GHz and plunger position of 134 mm, Figure 9(a) and Figure 9(b) are
obtained for crack located at 400 mm and 800 mm respectively.
As seen from these figures, almost the entire undesired signal is canceled. In
comparison with Figure 7(a), the reflected signal for crack location of 400 mm is
reduced in Figure 9(a). In each figure two candidates can be considered as the predicted
TOF and one of these candidates in each figure has good agreement with calculation.
The result shows that, the response of electromagnetic waves to the crack strongly
depends on the frequency range and plunger position. By changing plunger position
and frequency range it is possible to resonate electromagnetic waves inside the pipe
with crack and consequently to make high reflection at the location of crack.
The predicted TOF and predicted crack location by the above result are
summarized in the Table 1. By comparing these data, it is concluded that by choosing
proper frequency range and plunger position it is possible to determine crack location
by this method.
5. Conclusions
Through this study, the following conclusions can be drawn.

1) The possibility of detecting a circumferential crack in straight pipes by
electromagnetic waves has been experimentally verified.
2) Crack location can be determined by measuring TOF of reflected wave.
3) In this method reflection coefficient of reflected wave strongly depends on the
location of the plunger. By changing plunger position, best conditions for measuring
TOF can be obtained.
4) The response of electromagnetic wave for a crack located at a small distance is
larger than for a crack at larger distance.
5) Additional signal processing technique is needed to reduce noise and errors in the
future investigation. More accurate result could be obtained with this technique.
References
[1] K. Sugawara, H. Hashizume, S. Kitagima., Development of NDT method using electromagnetic waves,
JSAEM Studies in Applied Electromagnetic and Mechanics 10 (2001), 313-316.

[2] AD. Dimarogoans, Vibration of cracked structures: a state of the art review.Engng Fract Mech 55 (1996),
831-857.
[3] S. Doebling, C. Farrar, M. Prime, A summary review of vibration-based damage identification methods,
Shock Vib Digest 30 (1998), 91-105
[4] OS. Salawu, Detection of structural damage through changes in frequency, a review. Engng Struct., 19
(1997), 718-723.
[5] H. Hashizume, S. Kitajima, T. Shibata, Y. Uchigaki and K. Ogura, Fundamental study on NDT method
based on electromagnetic waves, ENDE2003, Saclay, Studies in Applied Electromagnetic and Mechanics
24 (2003), 263-270.
[6] H. Hashizume, T. Shibata and K. Yuki , Crack detection method using electromagnetic waves,
International Journal of Applied Electromagnetics and Mechanics 20 (2004), 171-178
[7] K. Abbasi, S. Ito, H. Hashizume, K. Youki, Crack detection by using electromagnetic waves, EPRI 5th
International conference on NDE in relation to Structural Integrity For Nuclear And Pressurized
Components 2006.
IOS Press, 2007
Inspection of Cement Based Materials

Using Microwaves
Kavitha ARUNACHALAM 1, Vikram R. MELAPUDI, Lalita UDPA and Satish S. UDPA
Department of Electrical and Computer Engineering
Michigan State University
East Lansing, Michigan 48824, USA
Abstract. Microwave nondestructive evaluation techniques proposed for civil

structures and cement based materials offer distinct advantages in that they are
non-radioactive and provide good penetration, excellent contrast between steel
reinforcement material and concrete, and insensitivity to ambient temperature.
This paper presents a far field frequency domain reflection coefficient
measurement setup for non-invasive inspection of cement-based materials at
microwave frequencies. Imaging results obtained in the X-band frequency for
planar mortar and concrete samples in the presence of steel reinforcement bar are
discussed.
Keywords. Microwaves, NDE, cement-based materials, reflection Coefficients
1. Introduction
Corrosion in reinforcement material inside concrete structures is influenced by cement

type, mixture proportions, curing conditions, age and environmental conditions. Thus,
structures such as roads, bridges, historical buildings and other public concrete
structures require periodic assessment for structural health and maintenance operations
[1]. Several non-destructive methods employing microwaves have been proposed for
monitoring degradation in concrete structures. Of these, the far field techniques are
favorable as they are contact free, relatively easy for modeling studies and are more
suitable for on-site measurement. Far-field techniques are also contact-free methods
that deal only with the dominant excitation mode, do not require surface preparation
and offer larger area of coverage and hence are advantageous compared to the near
field techniques [2]. Microwave far field techniques for imaging cement-based
materials employing SAR based approach have been described in [3-5]. Studies
relating the microwave far field reflection coefficients of cement based materials to the
moisture content, curing and moisture proportions have been reported in [2].
Experimental and numerical simulations on the feasibility of using far field

reflection coefficient measurements for imaging cement-based specimens in the
presence of metallic scatterers are presented in [6]. In this paper, additional
1
Corresponding Author: Kavitha Arunachalam, Department of Electrical and Computer Engineering,
Michigan State University, East Lansing, Michigan 48824, USA; E-mail: kavithaa@msu.edu
126 K. Arunachalam et al. / Inspection of Cement Based Materials Using Microwaves
experimental study conducted on mortar sample to investigate the capability of the

proposed imaging technique in resolving adjacent steel reinforcement bars is presented.
2. Far Field Reflection Measurement
Let a planar dielectric slab of thickness d be illuminated by a time harmonic plane

wave in free space at normal incidence. Due to impedance mismatch, the incident plane
wave is partially reflected and partially transmitted at the air-dielectric interface and the
transmitted wave undergoes multiple reflections inside the dielectric slab. The reflected
and transmitted waves are characterized by the reflection and transmission coefficients
of the dielectric slab as,
Er Et
*d (Z ) , Td (Z ) (1)
Ei Ei
where Er, Et are the reflected and transmitted fields after the incident plane wave, Ei
impinges on the dielectric slab. In (1), *d and Td are the reflection and transmission
coefficients of the dielectric slab. The reflection and transmission properties of a
homogeneous scatterer are related to the intrinsic impedance of the material [7]. For
normal incidence, the frequency domain reflection coefficient in (1) can be expressed
as [7],
§ 1 e j 2T ·
*d *12 ¨¨ 2 j 2T
¸¸
© 1 *12 e ¹ (2)
K 2 K1
*12 , T E2 d
K 2 K1
In (2), K1, K2 are the intrinsic impedances of free space and dielectric slab
respectively and E2 is the wave number inside the dielectric slab.
For a heterogeneous material like concrete, the reflection and transmission

property are related to the bulk permittivity and permeability. These bulk material
properties depend on the macroscopic behavior of the material content and vary as the
material degrades. Thus, the reflection and transmission coefficients of concrete
structure could be used to derive information about the material condition and strength
[2], [8]. Reflection measurement requires access to only side of the concrete and is
appealing from a practical point of view. Hence, the feasibility of using far field
frequency domain reflection coefficients for nondestructive evaluation of cement-based
specimens is investigated in this paper.
Figure 1 illustrates the far field reflection measurement set up. The measurement set up
consists of an X-band horn antenna connected to a vector network analyzer (VNA) and
positioned in the far field of the sample at normal incidence. The non-ideal behavior of
the microwave components and reflections from the surroundings introduces loss in the
K. Arunachalam et al. / Inspection of Cement Based Materials Using Microwaves 127
measurements. Prior to measurements, frequency is swept from 8 to 12 GHz and loss in

the VNA is compensated using calibration standard loads. The reflections in the
external components connected to the VNA such as the coaxial to rectangular
waveguide adapter, horn antenna and the surrounding environment are compensated
using measurements from an aluminum plate and the background. Assuming a linear
system model, the system transfer function is estimated by setting the reflection
coefficient of the aluminum plate to -1 in the entire X-band [6]. The estimated system
transfer function is used to compensate the loss introduced by the instrument, external
microwave components and background in the reflection measurements.
Figure 1 Experimental setup.
The experimental procedure was validated using two different planar dielectric
samples with known electrical property. Swept frequency measurements of the far field
reflection coefficients recorded over 8-12 GHz for the dielectric test samples were
calibrated using the system transfer function [6]. Figure 2 shows the comparison
between calibrated data and theoretical *d computed using (2) for the test samples. The
good agreement between calibrated measurements and analytical reflection coefficients
validates the experimental setup and procedure.
4. Results
Planar mortar (60x60x4 cm) sample composed of fine aggregate, portland cement and
water was cast for experimental studies. Far field X-band reflection coefficient of the
planar mortar sample was measured at normal incidence in the presence of two 60 cm
long 1.27 cm diameter steel reinforcement bar (rebar). The two rebars were placed 2.54
cm apart, in contact to the rear side of the mortar sample and in parallel to the
polarization of the incident plane wave as shown in Figure 3(a). The difference in the
reflection coefficient magnitude in Figure 3(b) shows the presence of the rebars. It can
also be observed that the resolution is better at higher frequencies.
Figure 2 Comparison between ī dtheory and ī experiment

d for test samples (a) sample A (b) sample B.
Figure 3 (a) Planar mortar sample with two rebars behind (b) Difference image of reflection coefficient for
mortar sample with two rebars.
K. Arunachalam et al. / Inspection of Cement Based Materials Using Microwaves 129
Figure 4 (a) Centrally located rebar inside heterogeneous planar concrete (b)Difference image of reflection
coefficient.
The ability to image the presence of a rebar inside a concrete sample was tested by
casting a planar concrete sample (60x60x5 cm) with a centrally located 1.27 cm
diameter rebar. Concrete mixture used for bridge decks was used to cast the planar
concrete sample with a rebar parallel to the polarization of the incident plane wave.
Figure 4(a) shows the plan view of the planar concrete sample used in the experiment.
Far field reflection measurements were recorded for the planar concrete sample with
rebar in the X-band. For the concrete sample, differencing technique was applied using
the reflection measurements of the same sample as in reality only one set of data is
available for interpretation. A first order forward difference equation of the form,
*d ( x0 , yi , z0 , f j ) *d ( x0 , yi 1 , z0 , f j ) *d ( x0 , yi , z0 , f j ) (3)
was applied to the calibrated far field reflection measurements. In (3), x0 and z0 refer to
antenna location in XZ plane as illustrated in Figure 1; subscripts j and i refer to
discretization in excitation frequency and antenna position along y-axis respectively
with 'f=80MHz and 'y=10mm. The difference image for the concrete sample with
steel rebar is shown in Figure 4 (b). The presence and location of the steel rebar is
clearly evident in the difference image. With advanced post processing algorithms the
precise location and dimension of the rebars can be estimated.
5. Conclusions
Far field reflection coefficient measurement setup for inspection of cement-based

materials is presented. Experimental setup and data calibration procedure were
validated using two different test samples with known dielectric property in the X-band.
The agreement between the calibrated measurements and analytical prediction validates
the data acquisition and analysis procedures followed in this investigation. The ability
to image rebars in cement-based materials using frequency domain reflection
coefficients was tested using a mortar sample with two rebars on the rear side.
Experimentation of the imaging technique on concrete specimen with a steel rebar
indicates the feasibility and robustness of the proposed technique for heterogeneous
cement-based materials. Imaging results obtained for mortar and concrete samples
appear promising and demonstrate the feasibility of using microwave far field
reflection coefficients for noninvasive inspection. The proposed concrete imaging
system will be evaluated for imaging complex concrete structures with multiple rebars
in the presence of voids, delaminations and cracks.
References
[1] Handbook on Nondestructive Testing of Concrete. Editors. V. M. Malhotra, and N. J. Carino, Second ed.
CRC Press, 2004.
[2] S. N. Kharkovsky, F. Akay, U. C. Hasar, and C. D. Atis, Measurement and monitoring of microwave
reflection and transmission properties of cement-based specimens, IEEE Instr. Meas. 51(2002), 1210-
1218.
[3] M. Pieraccini, G. Luzi, D. Mecatti, L. Noferini, and C. Atzeni, A highfrequency penetrating radar for
masonry investigation. 9th International GPR Conference (2002), Santa Barbara, CA.
[4] M. Pieraccini, G. Luzi, D. Mecatti, L. Noferini, and C. Atzeni, A microwave radar technique for dynamic
testing of large structures, IEEE Trans. Microwave Theory Tech. 51(2003), 1603–1609.
[5] Y. J. Kim, L. Jofre, D. F. Flaviis, and Q. M. Feng, Microwave reflection tomographic array for damage
detection of civil structures, IEEE Trans. Antennas Propag. 51(2003), 3022–3032.
[6] K. Arunachalam, V. R. Melapudi, L. Udpa, and S. S. Udpa, Microwave NDT of Cement-based Materials
using Far Field Reflection Coefficients, NDT &E Int. 39(2006), 585-593.
[7] Advanced Engineering Electromagnetics. A. B. Constantine, New York: Wiley, 1989.
[8] K. J. Bois, A. D. Benally, and R. Zoughi, Microwave near-field reflection property analysis of concrete
for material content determination, IEEE Trans. Instrum. Meas. 49(2000), 49–55.
IOS Press, 2007
Defect Profiling Using Multi-Frequency

Eddy Current Data from Steam Generator
Tubes
Kavitha ARUNACHALAMa, Oseghale UDUEBHOa, 1, Ameet JOSHIa,
Shiva ARUN KUMARa, Lalita UDPAa, Pradeep RAMUHALLIa,
Satish S. UDPAa and James BENSON b
a
Department of Electrical & Computer Engineering, Michigan State University,
East Lansing, MI, USA
b
Electric Power Research Institute, Palo Alto, CA, USA
Abstract. Steam generator tubes in nuclear power plants are continuously exposed to
harsh environmental conditions including high temperatures, pressures, fluid flow rates
and material interactions resulting in various types of degradation mechanisms.
Consequently they need to be inspected periodically for cracks and leaks. Multi-frequency
eddy current technique is one of the widely used Non-Destructive Evaluation (NDE)
techniques for steam generator tube inspection in nuclear power industry. The
multi-frequency technique enables nullifying the effect of extraneous discontinuities via
multi-frequency mixing and improves defect identification, classification and
characterization of the eddy current data. In this paper, different defect characterization
algorithms developed for the multi-frequency eddy current data obtained from the heat
exchange tubes is discussed. A comprehensive comparison of the performance of the
different characterization algorithms is also presented.
1. Introduction
Heat exchange tubes are used in nuclear power stations to transfer heat from the
primary loop to the pressurized water circulating on the outside to produce steam,
which is used to run the turbines. It is critical that the primary coolant, which is
radioactive, does not leak into the secondary side. The steam generator tubes once in
service are continuously exposed high temperatures, pressures, fluid flow rates and
material interactions resulting in various types of degradations.
NDE techniques are widely used in nuclear power industry to detect and
characterize surface and sub-surface defects in heat exchange tubes. Of the widely used
NDE techniques, multi-frequency eddy current technique is one of the most techniques
for steam generator tube inspection in nuclear power industry. The multi-frequency
eddy current technique is advantageous in that they allow simultaneous data acquisition
resulting in faster inspection speeds, enables separation of discontinuities that yield
dissimilar signals at different frequencies and nullifying the effect of extraneous
discontinuities via multi-frequency mixing and improves defect detection,
interpretation and defect size profiling in the presence of artifacts. Due to “skin effect”,
eddy current signals from defects changes with frequency [1]. In effect, this means that
1
Corresponding Author: Oseghale Uduebho, Department of Electrical and Computer Engineering,
Michigan State University, East Lansing, Michigan 48824, USA; E-mail: uduebhoo@msu.edu
132 K. Arunachalam et al. / Defect Profiling Using Multi-Frequency Eddy Current Data
multi frequency eddy current response signals have more information that can be
analyzed to extract relevant features and utilized to yield more accurate profiling results.
Besides defect detection and classification accurate structural evaluation of the tube
condition is also important. If detailed flaw dimensions (i.e., depth, length) can be
accurately measured, then calculations of the tube structural integrity can be performed
to determine the need for in-situ pressure testing. This paper presents different defect
characterization algorithms developed for the multi-frequency eddy current rotating
coil probe data. In this paper, the algorithms developed for multi-frequency defect
profiling include the industry standard calibration curve, alternative curve fitting
technique and radial basis neural networks. A comprehensive comparison of the
different defect characterization algorithms is also presented for performance
evaluation.
2. Defect Characterization
Defect characterization is the estimation of depth profile of the detected defect. Several
factors contribute to distortion of the measured eddy current signal. One example is the
limitation of the inspection system relative to resolution of the flaw segments that make
up the entire flaw length. Another factor is that the probe speed changes during the
inspection process which introduces errors in the collected data. Additive noise
generated during the scan due to presence of contaminants and surface roughness can
also introduce noise. Furthermore, when an analog signal is sampled to generate a
digital signal, quantization errors are introduced. This can lead to additional distortion
of the signal. All these issues make defect characterization in steam generator tubes a
very challenging task. The different characterization algorithms implemented for defect
profiling are explained in this section.
3. Industry Standard Calibration Curve
One of the earliest and most widely used approaches is the calibration method. A
calibration curve of defect depth versus phase, amplitude or magnitude is constructed
using known defects signals in a calibration standard tube with machined defects of
varying depths, orientation at different tube locations. Figure 1 shows typical
calibration standard curves at different frequencies constructed for defect profiling. For
a given defect signal, the equivalent length, width and depth of the defect are predicted
using the calibration curve along with interpolation methods. For instance, a calibration
curve that relates the phase, amplitude or magnitude of the eddy current signal with the
depth of the defect can be represented by a linear relationship of the form [2],
ba
f b f a > f c f a @ (1)
ca
where a, b and c refer to magnitude and/or phase value of the calibrated eddy current
signal. The phase and magnitude computed from flaws in the calibration standard tube
are used to construct the function f. This method is currently used in industry to
generate estimates of the one-dimensional size of a defect.
K. Arunachalam et al. / Defect Profiling Using Multi-Frequency Eddy Current Data 133
(a)
(b)
Fig 1 Calibration standard curves (a) Phase (b) Magnitude
4. Alternative Curve Fitting
In practice, eddy current signals from calibration standard machined defects are not
representative of the defect signals acquired from the in service heat exchange tubes.
This is because the damage mechanisms that arise due to stress, corrosion, pitting,
sludge piles are not captured completely by the machined defects with precise depths.
Thus the calibration standard signals require a transformation to compensate for signal
variation between defects in the calibration standard and in service heat exchange tubes.
A modified magnitude calibration curve is constructed such that the error in using the
calibration defect signals for profiling is minimized. This method generate estimates of
depth profile of a defect using an expression of the form,
log e ( flawdepth ) a 0 a1 log e ( flawmagnitude ) (2)
where the coefficients a0 and a1 are obtained by fitting a polynomial for the magnitude
of flaw signals in the calibration standard curve. Figure 2 shows a magnitude curve
constructed using the logarithmic mapping.
Fig 2 Magnitude calibration curve using equation (2).
5. Radial Basis Neural Networks
An alternate approach is to use a learning algorithm that implements function

approximation using radial basis function neural network (RBFNN). Mathematically,
this approach yields a highly nonlinear mapping from an input vector space on to an
output vector space. The preference for this network architecture is due to the
simplicity of its implementation and rapid convergence characteristics for small
training data size. A typical architecture of RBFNN is shown in Figure 3. The
input-output transformation equation for the RBFNN can be expressed as [3],
P
y ¦w
i 1
i f ( x t i ,V i ) (3)
where x is the input vector of dimension N , y is output vector of dimension M ,

t i is the i th basis center, w i is the weight vector of dimension M
th
corresponding to the i center and f ( x t i , V i ) is the scalar and radially
symmetric basis function with spread V i . A total of P centers (or nodes in hidden
layer) are used in the basis function expansion. The weights, bias and centers and
spread V i are computed during the training process using a training data set.
Input Layer Hidden Layer Output Layer
1 ti 1 wi 1
Output vector y
Input vector x
N P M
Fig 3 Architecture of RBFNN
The operation of RBFNN is typically divided into two distinct phases: (1) Training
phase and (2) Testing phase. In training phase the inputs and desired outputs both are
known. For defect profiling, radial basis function (RBF) of the form,
1

2 Vi
f ( x t i ,V i ) ( x ti 1) (4)
is chosen and the basis centers ( t i ) and their spreads or radii ( V i )are determined using
unsupervised clustering algorithms on the input data. The weights ( w i ) are estimated
with the help of input-output training data pairs and appropriate regularization methods
[4]. In the test phase the output values for any given unknown input vector are
predicted using the trained parameters of the RBFNN in (3).
6. Results and Discussions
This section describes the implementation of the proposed RBFNN approach for depth
profiling and presents a comparison of the performance of the method with that of the
calibration approaches. The Examination Technique Specification Sheet (ETSS)
maintains a database of tube defects and their associated depth profiles as determined
by metallographic procedures. The ETSS - provided by EPRI - was used to develop
algorithms for characterizing defect depth profiles that are consistent with
metallographic results.
6.1. Flaw Length Estimation
Flaw length estimation is an integral part of the defect characterization algorithms. The
flaw length, as determined by industry, is computed by evaluating the magnitude and
phase of each line signal within the region of interest (ROI) and performing a phase
and magnitude thresholding of any extraneous line signal in accordance with the
respective calibration curves. On the other hand, an adaptive threshold scheme that
relies on the histogram of the rotating probe coil (RPC) signal magnitude has been
implemented in this paper for estimating the flaw length in the neural network
predicted depth profile. An empirical relationship for the magnitude threshold (K) to
the ratio of the signal magnitude in the histogram (J) is derived as,
K D exp{O [J J 0 ]} % (5)
where the constants D, O and J0 are determined from the available profiling data.
During defect characterization, RPC signals that lie within K% of the maximum
magnitude are used for defect depth profiling.
6.2. Training the Neural Network
Seven calibrated Westinghouse Laboratory outer diameter (OD) flaw samples of Tube
Support Plate (TSP) extraction and non-constant depth were used for training the
neural network and the algorithm for the implementation as shown in Figure 4. Using
the axial scale and MET- magnitude correlation methods, the axial length of the ROI is
determined. The peak magnitude and phase at multiple frequencies and phase spread
(maximum difference in phase values for all frequencies) for each line scan is
computed and stored in a feature matrix for training the network alongside the
corresponding depths values obtained by metallographic (MET) analysis. For training
the network, the feature vectors for the seven sample flaws are systematically
aggregated and mapped unto the corresponding MET depths. A more robust neural
network is obtained by mapping three spatially consecutive line scans/feature vectors
onto the depth value at each position in the flaw.
Figure 4 Schematic of the overall approach using RBFNN
6.3. Testing with Examination Technique Specification Sheet (ETSS) Data
A second data set of pulled tubes with metallographic profile information was used to
evaluate the performance of the approach. The defect length was estimated and line
scans along the defect region were used to calculate the feature vectors. Figure 5(a)-(c)
shows the profiling results obtained for three ETSS flaws using the classical magnitude
calibration curve approach (CC), and the two RBF neural networks, RBF1 (one feature
vector per depth) and RBF2 (three feature vector per depth).
(a)
(b)
(c)
Fig. 5 Comparison of metallographic flaw profiles (MET) with profiles generated by different algorithms.
The initial results in Figure 5 indicate that the RBF network provides more accurate
profiles than the calibration curve techniques. The performance of RBF network on the
axial OD defects in the test dataset is summarized in Table 1. The neural network
performance is significantly better than that obtained using conventional methods
largely due to the fact that is utilizes all information available from multiple frequency
inspections.
Table 1 Flaw depth and length estimation on the test dataset
Flaw Length
Maximum % Through Wall
(Inches)
Flaw
Log
MET* RBF 1 RBF 2 Magnitude MET* Estimated
Magnitude
1 47.00 51.42 42.07 44.00 43.90 1.4 1.51
2 43.00 53.45 48.38 39.00 38.68 1.31 1.27
3 60.00 49.23 56.80 26.00 28.53 0.53 0.48
4 77.00 57.39 64.17 62.00 60.14 0.73 0.65
5 63.00 76.85 65.12 22.53 9.00 0.85 0.99
6 60.00 58.72 56.92 63.00 65.23 0.52 0.42
7 50.00 47.79 55.40 18.00 27.27 0.60 0.34
8 35.80 34.47 31.15 44.00 46.25 0.11 0.14
* MET - Metallographic Data
7. Conclusions
Defect characterization algorithms based on calibration curves and RBF networks have
been developed for estimating the depth profiles of axial outer diameter defect
indications. Comparisons of different characterization algorithms reveal that the
performance of the RBF networks is significantly better than the industry standard
calibration curve approaches. The improvement in the performance of the RBF network
is solely due to the fact that the multi-frequency features that capture the defect
behavior in the impedance plane are used intelligently to map the eddy current data to
the flaw depth profile.
References
[1] S. S. Udpa, L. Udpa, Wiley Encyclopedia of Electrical and Electronics Engineering, edited by John G.
Webster, vol. 6, 1999, pp 149-163.
[2] Rotating Probe Eddy Current Data Analysis, Technical Report, EPRI, Palo Alto, CA: 2004.
[3] Simon Haykin, "Neural Networks A Comprehensive Foundation", Prentice Hall, Upper Saddle River, New
Jersey 07458, 1994, pp.Chapter 7, pp.236-281.
[4] Tikhnov A.N and V.Y. Arsenin, "Solutions of Ill-posed Problems", Washington, DC, 1977.
IOS Press, 2007
Electromagnetic Reading of
Laser Scribed Logistic Markers on
Metallic Components
Szabolcs GYIMÓTHY a,1, József PÁVÓ a, Imre KISS a, Antal GASPARICS b,
Zoltán KALINCSÁK a, Imre SEBESTYÉN a, Gábor VÉRTESY b,
János TAKÁCS a and Hajime TSUBOI c
a
Budapest University of Technology and Economics, Hungary
b
Research Institute for Technical Physics and Materials Sciences, Hungary
c
Fukuyama University, Japan
Abstract. An application of electromagnetic nondestructive evaluation is

presented in the paper: Laser scribed markers used as logistic codes on metallic
components are reconstructed from the signal of an eddy current probe. The
authors developed a computational model of the reading out process by which both
the code specification and the probe can be optimized. The model is applied here
for the optimization of marker density. The reconstruction of barcode using
inverse filtering approach is also demonstrated.
Keywords. eddy current testing, barcode, deconvolution, inverse filtering
1. Introduction
Specific laser marking technique of the surface of metallic components has been
reported in [1]. This marking was originally used for the detection of thermal induced
stresses on the surface of rails. In recent times its application for bar-coding is being
considered. This specific barcode can be used for the identification of metallic parts in
production logistics, during the automated assembly, and also for tracing the
manufactured product. The laser scribed barcode exhibits high endurance and thermal
stability compared to other marking techniques. Moreover, it can be read out by using
electromagnetic principles in a fast, non-destructing and contact free operation, and
even from under protective coating (e.g. paint). The automotive industry is one of the
main targeted utilizers of the laser marking technique.
The measurable effect of the marking stems from the rapid heating and cooling
process due to the laser treating, causing local phase transformations in the metal, as
well as modifications in the stress field and the domain structure in the vicinity of the
surface. These structural modifications affect the local electromagnetic properties like
conductivity and permeability. Therefore an eddy-current sensor will produce some
time varying signal while scanning above the marking.
Figure 1 illustrates the process of laser scribing. Markers providing good
signal/noise relation during the reading out can be achieved for example by CO2 laser
irradiation with the following parameters: laser power: P 100 300 W; diameter of
1
Corresponding Author: Szabolcs Gyimóthy, Budapest University of Technology and Economics,
Egry J. u. 18, 1521 Budapest, Hungary; E-mail: gyimothy@evtsz.bme.hu
140 S. Gyimóthy et al. / Electromagnetic Reading of Laser Scribed Logistic Markers
Figure 1. Illustration of the process of laser scribing
laser spot: d 0.5 2.0 mm; movement speed: v 1200 mm/min. Figure 2 shows the
photomicrograph image on the cross-section of the transformed zone due to laser-metal
interaction in high carbon (0.6 C%) rail steel (left), and in low carbon (0.1 C%) cold
rolled structural steel (right), respectively. The latter image contains some graphical
enhancement, because the structural changes are not visual enough in this case.
Although some parameters of the laser scribing process itself have already been
fixed [1], there are still several unspecified parameters of the complete marking
technique, which have to be optimized towards the efficient reading out and
reconstruction of the code. These are among others the type and geometry of the sensor
probe, the coding system used for the barcode, as well as the geometry specification of
the marking (e.g. width, depth, density), which latter may depend on both the spatial
resolution of the probe and the applied coding system.
In order to facilitate the optimization, the authors have developed a computational
model of the reading out process based on the finite element method. Using this model
the optimal distance between adjacent markings has been estimated. Finally, barcode
reconstruction from measured signal has been demonstrated by using an inverse
filtering approach. The results associated with the three outlined topics are presented in
this paper.
2. Numerical Model for the Simulation of the Reading Out Process
Figure 3 provides two schematic views of the measurement configuration used for
reading out the barcode. The Fluxset sensor [3] is applied for measuring the magnetic
field above the conducting object. The magnetic field is induced by an exciting coil
(inner diameter: 4 mm; outer diameter: 8.4 mm; height: 4.7 mm; lift-off: 0.25 mm)
Figure 2. Photomicrograph images taken on the cross section of laser marked metallic components. Left:
high carbon rail steel (0.6 C%); right: low carbon cold rolled structural steel plate (0.1 C%)
S. Gyimóthy et al. / Electromagnetic Reading of Laser Scribed Logistic Markers 141
Figure 3. Typical measurement configuration used for the reading out
above the sensor. The orientation of the sensor core coincides with the scanning
direction of the probe (x-direction), which is perpendicular to the marker lines. The
investigated steel plate specimen (size: 100mm×100mm; thickness: 0.6 mm) is made of
low carbon steel (C < 0.1 %, Mn = 0.7 %, Si = 0.3 %, Ti = 0.06 %) having 0.01 mm
iron-zinc coating on both sides. The applied frequency in the measurement is 20 kHz.
2.1. Simplified Finite Element Model
A finite element model of the reading out process has been developed (Figure 4). The
model applies the T-ĳ formulation combined with hexahedral edge elements [4].
Instead of modeling the complex microstructure in the laser irradiated zone, a
macroscopic equivalent model has been defined. Using this approach the measured data
can be reproduced with sufficient accuracy, while still describing the overall physical
picture well. The geometry of the cross-section of the marker is simplified to the
rectangular shape (width: 0.5 mm; depth: 0.2 mm) for the following two reasons. First,
the non-uniform graded mesh around complex geometries would influence the
calculated results in a way that a kind of “false signal” component appears in the output
of the probe. Second, when using the mesh on Figure 4, which has uniform
x-subdivision, one can easily modify the barcode without changing the mesh, by just
adjusting the material properties. The length of each marker line is 30 mm. The region
of the marking is considered as a homogeneous linear material in the model.
2.2. Fitting of the Model Parameters to Measured Data
Electric conductivity and magnetic permeability of the marked region are parameters of
the model, which have to be set towards achieving the best fit between the simulated
output signal and the real measured data. Although the Fluxset sensor is intended to be
used for code reading in the long run, particularly for the identification of the model
parameters a classical test measurement has been chosen. The impedance change of a
coil is measured above the sheet. We selected the impedance probe for this purpose
because by using this, it is easier to get reliable quantitative data both in experiment
and in theory. The fitting of electromagnetic parameters is carried out in two steps: first
the parameters of the base material are determined, and then the parameters of the
marked region are finely tuned.
In the first step, the impedance of the exciting coil was measured above a non-
marked sheet, and was compared to its impedance measured in the absence of the sheet.
At the same time, these impedance change values were calculated for several
conductivity-permeability pairs, using the computer code Teddy [5] based on semi-
Figure 4. Part of the finite element model of the measurement used for reading out the barcode. The
conducting plate, three markers and the half of the exciting coil are plotted. The surrounding air is discretized
as well, but not shown here.
analytical method [6]. The comparison of measured and calculated results indicated
that the ratio of the two parameters must be about P r / V 30 m/MS . Also it became
evident from the simulation that the influence of the thin iron-zinc coating on these
results is negligible. Finally we chose the conductivity value, V 4 MS/m from the
literature, which results P r 120 for the permeability parameter of the base material.
In the second step of the model calibration the impedance change of the exciting
coil was recorded while it scanned above one single marker far from the edges of the
plate. At the same time several impedance scans were simulated with systematically
varied parameters of the marked zone in the above described finite element model. The
best fit to the measurements has been found with slightly lower conductivity and higher
permeability values ( V 2 3 MS/m , and P r 400 500 ) in the marked region than
in the base material. We are well aware, that the expected behavior is opposite, as
permeability usually decreases after rapid heating and cooling. In order to resolve this
contradiction, further analysis of the complex microstructure in the metal, as well as the
refinement of the numerical model would be required. Nevertheless, the current model
can still be used as one which is reasonable from the point of view of the measurements.
3. Prediction of Optimal Marker Density Using the Numerical Model
The information density in the barcode can be increased by lowering the distance between
the marker lines. On the other hand, the signals generated by the markers tend to overlap
for shorter distances, as it is witnessed by the measured curves presented in Figure 5, and
the inflectional points or peaks originally relating to marker positions loose their meaning.
Therefore finding the optimum distance between adjacent markers is an important
practical problem. The fundamental question is whether or not the overlapped signals
can be separated somehow, that is, whether, or to which extent, the superposition
principle for marker signals holds. This “linearity” feature has been analyzed by using
the above described finite element model.
Figure 5. Measured curves of the Fluxset sensor for double marker lines of various separation distances
(marker positions are x = ±2.5 mm, x = ±1.5 mm and x = ±1 mm, respectively).
In the investigated problem, the Fluxset sensor measures the weighted average of
the x-component of the magnetic flux density taken along the sensor core [7]. This
weighted average is evaluated from the numerical model. Note that although the
measurement principle of the Fluxset sensor is different, the model parameters obtained by
the impedance measurements (as described in Section 2) can still be used, because the
eddy current fields are practically the same. Figure 6 confirms the superposition principle
for two marker lines located at x = ±0.5 mm. That is, if we shift the output signal
measured for one standing alone marker (thin solid line) with +0.5 mm and –0.5 mm,
respectively, and add them up, we get the dashed line, which is a good approximation of
the signal measured for the two markers (thick solid line). Note, that for smaller distances
the transformed regions tend to fuse, and superposition may not work any longer.
Figure 6. Illustration of the superposition principle for two marker lines located at x = ±0.5 mm.
4. Barcode Reconstruction Using Inverse Spatial Filtering
The superposition of marker signals can be described by convolution, thus the

reconstruction can be carried out by deconvolution. One of the simplest and most
elegant methods of deconvolution is inverse filtering [8][9], a common solution in DSP,
which can be summarized briefly for the current application as follows. A digital
coding of the information scribed by the laser is considered. Assuming that the same
laser lines (markers) are scribed or not scribed in equidistant positions depending on
whether a digital 0 or 1 is supposed to be coded. It has been proved by the developed
numerical model that the output signal y (x ) of the probe can be formulated as a sum
of individual marker signals (see Section 3):
N
y( x) ¦ a j y0 ( x x j ) , (1)
j 1
where y0 ( x ) stands for the output signal generated by one single marker located at the
origin, and x j is the center of the j-th marker location; aj =0 or a j =1 is considered if a
marker is present or not present at the location x j , respectively. Let us define the
spatial Fourier transform (FT) and its inverse for the g (x ) and G (ik ) functions pairs as,
f
1 f
F ^ g ( x )` ³ g( x) e
ikx
F 1^G (ik )` G (ik ) e ikx dk . (2)
2S ³f
G ( ik ) dx ; g( x)
f
Taking the FT of (1) we get,

N
ik x j
Y ( ik ) ¦ a j Y0 (ik ) e . (3)
j 1
Now assume that y0 ( x ) is obtained as the convolution of two signals, that is in

the spectral domain we are allowed to write,
Y0 (ik ) H (ik ) W0 (ik ) . (4)
Note that one of the functions H and W0 can be defined arbitrarily. Let us define the
corresponding spatial function,
w0 ( x ) F 1^W0 (ik ) `, (5)
so that it allows to determine the actual values of aj in a simple way from the following
type of combined function,
N
u( x ) ¦ a j w0 ( x x j ) . (6)
j 1
This can be done easily if, for example, the w0 functions do not overlap. In the same
time, let the bandwidth of W0 as small as possible. The transfer function of the filter to
be used for the reconstruction of the barcode will benefit from this latter property of
W0 . Writing (4) into (3) yields,
N
ik x j
Y ( ik ) H (ik ) ¦ a j W0 (ik ) e . (7)
j 1
After dividing by H and taking the inverse FT we get,
¦ a j F 1 ^W0 (ik ) e `
1 ½ N
ik x j
N
F 1 ® Y ( ik ) ¾ ¦ a j w0 ( x x j ) u( x ) . (8)
¯ H ( ik ) ¿ j 1 j 1
This tells us that if we process the signal measured by the ECT sensor with a filter
having the transfer function, 1 / H (ik ) , we will get a function u( x ) from which it is
easy to decide whether a given marker line is present or not (i.e. whether aj is 1 or 0).
Inverse filtering can be realized by simple and cheap FIR-filters, and may be
carried out on-the-fly. Note however, that the method requires the knowledge of the
output signal y0 ( x ) that is generated by one single marker. As this signal depends on
both the material properties and the applied coating of the measured component, it
cannot be “hard coded” into the reading out device. Rather, some isolated “leading
marker” should be scribed before the barcode, from which the signal y0 ( x ) is to be
obtained. One serious setback of inverse filtering is that it tends to amplify high
frequency noise. This property can be improved by the proper selection of the
w0 function. However, if this property would inhibit the reading out, then more
sophisticated methods based on optimization must be used for the deconvolution [10].
Inverse filtering is demonstrated here for two markers at 1 mm distance from each
other. The output signal of the probe was simulated on the numerical model described
above. In order to suppress the unnecessary high frequency components of y ( x )
during the inverse filtering, the function w0 ( x ) was defined as a Blackman window [9]
of 2 mm width, which has relatively low bandwidth. The simulated output signal
together with the reconstructed signal representing the two markers are plotted in
Figure 7. It is apparent that because of the overlapping of the impulses, the peaks
shifted away from their expected positions, but the result can still be decoded if marker
distances are limited to be multiples of some base unit (e.g. 1 mm).
5. Conclusions
Methods of computational electromagnetism are used for the development of an

electromagnetic reading out system for laser scribed logistic markers of metallic
components. Simulation of the measured signal based on FEM calculations is
successfully applied for the identification of the material parameter variation due to
irradiation with the laser beam. Bar code system consisting of identical lines is
proposed for the coding of information. The application of this coding requires the
separation of the overlapping ECT signal due to the parallel code lines. This separation
is made by inverse filtering. It is proved both numerically and experimentally that the
signal of several bars can be assumed as the superposition of signals due to the
individual bars. In other words there is practically no electromagnetic interaction
between the individual bars. Based on this linearity, a method is outlined for the
identification of the transfer function of a filter that can be used for the reconstruction
of the bar code.
Figure 7. The result of inverse filtering applied to the signal of a double marker (c.f. Figure 6). The markers
are approximately localized by the two high peaks.
Acknowledgments
This work supported by the project GVOP -3.1.1.-2004-05-0452/3.0 of the Hungarian

National Office for Research and Technology, by the “Bolyai János” Research
Fellowship of the Hungarian Academy of Sciences, by the Hungarian Research and
Technology Innovation Fund under grant JAP 17/02, and by the Hungarian Scientific
Research Fund under grant T-049389. The authors also express their special thanks to
Dr Oszkár Bíró for making available the finite element software EleFAnT-3D
developed at TU-Graz, IGTE; as well as for Dr Theodoros P. Theodoulidis for offering
his software Teddy for the required computations.
References
[1] Z. Kalincsák, J. Takács, G. Vértesy and A. Gasparics, “The optimisation of laser marking signals for
eddy current detecting of marks”, Proc. Laser Assisted Net Shape Engineering 4, Erlangen (Germany),
Sept. 21-24, 2004, vol. 1, pp. 535-544.
[2] I. Mészáros, “Micromagnetic measurements and their applications”, Materials Science Forum, vols. 414-
415, 2003, pp. 275-280.
[3] G. Vértesy, A. Gasparics, J. SzöllĘsy, “High sensitivity magnetic field sensor”, Sensors and actuators,
vol. 85, 2000, pp.202-208.
[4] O. Bíró, “Edge element formulations of eddy current problems”, Computer Methods in Applied
Mechanics and Engineering, vol. 160, 1999, pp. 391-405.
[5] T. P. Theodoulidis, M. K. Kotouzas, “Eddy current testing simulation on a personal computer”, Roma
2000 NDT World Conference.
[6] T. P. Theodoulidis, E. E. Kriezis, “Impedance evaluation of rectangular coils for eddy current testing of
planar media”, NDT & E International, 2002, vol. 35, pp.407-414.
[7] J. Pávó, A. Gasparics, I. Sebestyén, G. Vértesy, “Calibration of Fluxset sensors for the measurement of
spatially strongly inhomogeneous magnetic fields”, Sensors and Actuators A: Physical, vol. 110, no. 1-3,
2004, pp.105-111.
[8] J. P. Wikswo, Jr, “The Magnetic Inverse Problem for NDE”, in SQUID Sensors: Fundamentals,
Fabrication and Applications (ed. H. Weinstock), Kluwer Academic Publishers, The Netherlands, 1996,
pp. 629-695.
[9] S. W. Smith, The Scientist and Engineer’s Guide to Digital Signal Processing, California Technical
Publishing, San Diego, 1999, pp. 297-310. [www.dspguide.com]
[10] S. Esedoglu, “Blind deconvolution of bar code signals”, Inverse Problems, vol. 20, 2004, pp. 121-135.
IOS Press, 2007
Design of a Remote Field Eddy Current

Probe Dedicated for Inspection of
a Magnetic Tube from its Outer Surface
Tomas MAREK a, Daniela GOMBARSKA a, Ladislav JANOUSEK a, 1,
Klara CAPOVA a, Noritaka YUSA b, Kenzo MIYA b
a
University of Zilina, Faculty of Electrical Engineering
Department of Electromagnetic and Biomedical Engineering
Univerzitna 1, 010 26 Zilina, Slovak Republic
b
IIU, Imon Ikenohata Bldg. 7F, 2-7-17 Ikenohata, Taito-ku, Tokyo 110-0008, Japan
Abstract. The paper deals with the design of a special remote field eddy current
probe dedicated for inspections of a magnetic tube from its outer surface. A simple
configuration of the probe with one exciting coil and one pick-up coil is
considered. The coils are shielded with magnetic material to gain the remote field
effect in the given configuration. Results of numerical investigations prove the
effectiveness of the probe.
Keywords. Remote field eddy current testing, probe, shield, magnetic tube,
inspection from the outer surface
1. Introduction
Non-destructive testing (NDT) is utilized for examination of structural components that

might cause malfunction of a system with high economical and/or ecological impacts.
The remote field eddy current testing (RFECT) is an NDT method used especially for
the inspection of magnetic tubes [1]. The RFECT assures almost equal sensitivity to the
inner (ID) and the outer (OD) defects because the magnetic flux penetrating through a
tube wall is detected.
The RFECT has been mainly used for the inspection of a tube wall from the inner
surface of the tube. However, not all tubes can be accessed from their inside; they must
be inspected from the outer surface. It is reported that there is a significant difference
between the inspection of a tube from the inner surface and the outer one; it is quite
difficult to gain the remote field effect in the later case [2]. Several studies have
proposed to use shielding to realize a remote field effect when probes are situated
outside of a tube. Whereas successful results were reported [2], the studies consider
only nonmagnetic tubes. In reality, many tubes that must be periodically inspected are
made from magnetic materials [3].
The paper presents design of an RFECT probe dedicated for inspections of a
magnetic tube from its outer surface. Numerical simulations of electromagnetic field
1
Corresponding Author: Ladislav Janousek, Department of Electromagnetic and Biomedical
Engineering, Faculty of Electrical Engineering, University of Zilina, Univerzitna 1, 010 26 Zilina,
Slovak Republic; E-mail: janousek@fel.utc.sk
148 T. Marek et al. / Design of a Remote Field Eddy Current Probe
distribution using the finite element method are performed to simulate the inspection of
a tubular specimen with a defect of a variable depth and width. The probe is shielded to
achieve the remote field effect from the outside of the tube. Dimensions of the probe
are adjusted in such a way that a distance between an exciting coil and a pick-up one is
minimized to gain higher level of the signal while still keeping the RFEC effect.
2. Probe Design
2.1. Target Definition
This study considers the inspection of a magnetic tube with an outer diameter of
500mm and a wall thickness of 10 mm[3]. The electromagnetic parameters of the
tube material are σ = 1 MS/m and μr = 100. Whole circumferential wall thinning is
used to model a defect arising from the inner or the outer surfaces of the tube.
2.2. Layout of the Probe
The main goal is to reach the RFEC effect with a probe situated from the outside of the
tube. A shielded outer circumferential probe is utilized in the paper. There are many
variable parameters concerning the design of a probe, i.e. arrangements of coils,
dimensions of coils and distance between them, materials, configuration and
dimensions of shield, and finally the exciting frequency. Numerical simulations have
been carried out to find out a proper design of the probe.
The considered tube as well as the probe is axis symmetrical and thus it is possible
to dismiss one dimension of the problem. Accordingly, a two-dimensional finite
element code has been used for the numerical simulations.
Preliminary results of simulations indicated that a complex arrangement of the
coils does not bring reasonable results for sake of the RFEC effect. Therefore, simple
configuration with one exciting coil and one pick-up coil is chosen. Such configuration
is also beneficial from the probe dimensions point of view.
Figure 1. Configuration of the RFECT probe with one merged shield

Several configurations of the probe shield have been studied, e.g. separate shields
for each coil, compound shield layered from different materials, etc. As the complex
shield does not bring any advantage, the probe with one merged monolithic shield
covering both the coils is chosen for further investigation. In such case, the distance
T. Marek et al. / Design of a Remote Field Eddy Current Probe 149
between the coils can be shorter and the fabrication of the probe is simpler. The
proposed configuration is shown in Fig. 1.
Preliminary numerical simulations of the electromagnetic field distribution inside
and around the tube revealed that electromagnetic parameters of shielding material
(conductivity and relative permeability) and their ratio have significant influence in
achievement of the RFEC effect in the concerned problem. Several materials have been
used in numerical simulations to find a suitable one. Based on the findings, Cobalt is
used for this purpose. The electromagnetic properties of Cobalt are: σ = 16 MS/m and
μr = 68.
Dimensions of the probe are adjusted by numerical means.
2.3. Dimensions of the Probe
The distance between the exciting coil and the pick-up coil and the dimensions of the
shield are adjusted to obtain reliable behaviour of the probe (Fig. 1). The driving
frequency is set along with the probe dimensions. Three parameters are changed, i.e.
the distance between the coils, a width of the defect and the exciting frequency to find
the proper dimensions of the probe. Table 1 summarizes the changes of the three
parameters. The ID and OD cracks with depths of 20 % and 50 % of the material
thickness are considered. The gained results are evaluated in a way that there should be
minimum difference in phase and in amplitude between the signals of ID and OD
defects with the same depth and width.
Table 1. Variables used in the probe design process

Variable Interval Step
Coil distance L 70mm-90mm 5mm
Defect width wc 5mm-50mm 5mm
Frequency f 100Hz-400Hz 100Hz
Table 2. Dimensions of the probe

Probe parameter Dimensions
Cobalt shield ws = 130 mm, hs = 15 mm
The coils distance D = 80 mm
The coil dimensions wec = wpc = 3 mm, hepc = 2 mm
Exciting frequency f = 300 Hz
Lift-off L = 0.5 mm
It should be noted that the dimensions of the coils (width, height) have been set in
advance as they do not influence the required behavior of the probe. The increasing
distance between the coils as well as the increasing exciting frequency reduce the
amplitude of the pick-up signal. Thus, it is preferable to adjust both the parameters as
low as possible to obtain a higher level of the detected signal. However, certain
limitations have to be taken into account for the probe design. The distance between the
coils influences maximum width of the crack for which the signals of ID and OD with
a same depth are close. Outer dimensions of the shield, made of Cobalt, should be
adjusted according to the selected distance between the coils to minimize unwanted
edge signals. The preliminary numerical simulations showed that a distance between
the coil and the shield edge should be at least 25 mm to assure that the edge effect does
not influence the detected signal. The exciting frequency influences the amplitude level
of the pick-up signal as well as the phase distinction between the signals of cracks with
different depths. Therefore, the value of the frequency is adjusted along with the other
parameters of the probe to gain good properties of the probe. Table 2 summarizes the
parameters of the proposed RFECT probe.
The proposed probe is used for the inspection of the considered tube. Numerical
results of the inspection are presented in the next section.
3. Simulation Results
The magnetic tube with parameters given in section 2.1 is inspected by numerical
means. The whole wall thinning of variable depth ranging from 0 % to 100 % of the
tube thickness and of variable width is used to model the ID and OD defects. The
RFECT probe shown in Fig. 1 is used for the inspection. The dimensions of the probe
as well as the exciting frequency are given in Tab. 2.
Figure 2 displays Lissajous plots of the crack signals obtained with the proposed
probe for the ID and the OD cracks with a depth of 20 % of the tube thickness and with
widths of 5 and 40 mm, respectively.
0.8
ID 20% ID 20%
0.3 OD 20% OD 20%
0.6
0.2
0.4
imaginary part [mV]
imaginary part [mV]
0.1 0.2
0 0
-0.2
-0.1
-0.4
-0.2
-0.6
-0.3
-0.8
-0.3 -0.2 -0.1 0 0.1 0.2 0.3 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
real part [mV] real part [mV]
a. b.
Figure 2. Lissajous plot of the pick-up signal, f = 300 Hz, D = 80 mm, ID&OD = 20%: a. wc = 5 mm;
b. wc = 40 mm
Similar results for the ID and the OD cracks with a depth of 50 % of the tube
thickness are shown in Fig. 3. It can be seen that the signals of the ID crack and the OD
crack with the same depth are close to each other.
Dependences of the crack signal amplitude and the phase on the crack depth for ID
and OD cracks with depths ranging from 10 to 100 % of the tube thickness and with
a constant width of 20 mm are shown in Fig. 4. The figure shows that ID signals are as
clear as OD ones that confirms the effectiveness of proposed RFECT probe.
T. Marek et al. / Design of a Remote Field Eddy Current Probe 151
1 ID 50% ID 50%
OD 50% OD 50%
2
0.5
1
imaginary part [mV]

imaginary part [mV]
0 0
-1
-0.5
-2
-1
-1 -0.5 0 0.5 1 -2 -1 0 1 2
real part [mV] real part [mV]
a. b.
Figure 3. Lissajous plot of the pick-up signal, f = 300 Hz, D = 80 mm, ID&OD = 50%: a. wc = 5 mm;
b. wc = 40 mm
100 40 5.5 -30

ID amp ID amp
OD amp OD amp
ID phs 5 ID phase
20
OD phs OD phase -35
4.5
0
4 -40
10
log amplitude [mV]
-20
amplitude [mV]
3.5
phase [deg]
phase [deg]
-45
-40 3
-50
2.5
-60
1
2 -55
-80
1.5
-100 -60
1
0.1 -120 0.5 -65

10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40 45 50
crack depth [%] crack width [mm]
Figure 4. Amplitude of the pick-up signal Figure 5. Amplitude of the pick-up signal and its
(log scale) and its phase depending on the crack phase depending on the crack width; crack depth
depth; crack width is wc = 20 mm is ID&OD 50%
The numerical results prove the applicability of the probe with merged Cobalt
shield in the RFECT inspection of the magnetic tube. It was already mentioned that the
distance between the coils along with the width of the defect affect the difference
between the ID and OD signals. Figure 5 shows dependences of the crack signal
amplitude and the phase on the crack width for the ID and the OD cracks with a depth
of 50 % of the tube thickness and with a variable width 0 – 50 mm. It can be observed
that when the crack is wider than 40 mm, the difference between the signals of the ID
and the OD cracks starts to become significant.
The robustness of the probe against fluctuations in the properties of the tube
material and also of the shielding material has been investigated. The lift-off between
the probe and the tube has also been changed to see its effect on the possibility to keep
the RFECT effect. It can be stated that the proposed RFECT probe is quite robust
concerning the changes in the material properties of the tube and of the shield within
considered ranges (change of conductivity and relative permeability ±10 %). Influence
of the variations in the materials’ parameters on the calculated pick-up signals is
negligible. However, larger lift-off causes the features of the probe to significantly
deteriorate, because the remote field effect is lost when the lift-off exceeds 1 mm.
4. Conclusion
The aim of the study was to enhance remote field eddy current testing performed from
the outside of a magnetic tube. Design of a remote field eddy current probe was
considered in the paper. In order to obtain the remote field effect, it was necessary to
use the probe with appropriate shield. Among several considered configurations of the
probe, the simple reflection type probe (1 exciter – 1 pickup) was chosen for the study
as others do not bring any significant advantages; moreover, more complex signals are
obtained when a multiple coil design is used.
The configuration of the merged shield covering both the coils was proposed here.
Numerical simulations were used to examine properties of several shielding materials;
Cobalt was found to be a suitable one.
The distance between the coils as well as the exciting frequency were adjusted by
numerical means to gain good sensitivity of the probe to the inner and the outer defects
in the tube. Numerical results proved the effectiveness of the proposed probe. The
probe is quite robust against fluctuation in material properties of the tube as well as of
the shield; however, increased lift-off causes loss of the remote field effect.
Acknowledgment
This work has been partially supported by a grant VEGA No. 1/2053/05 “Design and
Organization of Electromagnetic and Acoustic Methods and Tools for Material
Nondestructive Testing” of the Slovak Ministry of Education.
References
[1] Yushi Sun, An introduction to electromagnetic nondestructive testing. Non-linear electromagnetic

systems, IOS Press, 1998, pp.145-152.
[2] Young-Kil Shin, Achievement of RFECT effects in the nuclear fuel rod inspection by using shielded
encircling coils. Electromagnetic nondestructive evaluation (VI), IOS Press, 2002, pp.83-90.
[3] METI press release,
http://www2.jnes.go.jp/atom-db/en/trouble/individ/power/l/l048091/index.html.
IOS Press, 2007
Application of Thermoelectric Power

Measurement to Nondestructive Testing
Shinsuke YAMANAKA a,1, Yasuhiro KAWAGUCHI b, Toshihiro OHTANI c
and Ken KUROSAKI a
a
Division of Sustainable Energy and Environmental Engineering,
Graduate School of Engineering, Osaka University,
2-1 Yamadaoka, Suita, Osaka 565-0871, Japan
b
Maintenance Department, Reprocessing Plant, Reprocessing Business Division,
Japan Nuclear Fuel Limited,
4-108, Aza Okitsuke, Oaza Obuchi Rokkasyo-mura, Kamikita-gun, Aomori 039-3212
Japan
c
Materials Laboratory, Ebara Research Co., Ltd.,
4-2-1 Hon-Fujisawa, Fujisawa, Kanagawa 251-8502, Japan
Abstract. Thermal aging of cast duplex stainless steels and creep damage of a
chrome steel were evaluated from thermoelectric power measurements.
Relationships between the thermoelectric power and thermal aging and/or creep
damage were studied.
Keywords. thermoelectric power, duplex stainless steel, thermal aging, chrome

steel, creep damage
1. Introduction
Cast duplex stainless steel is frequently used in the main coolant pipes and reactor
coolant pump casings of pressurized water reactor (PWR) type nuclear power plants
because of its excellent material strength, toughness, and superior corrosion resistance.
However, during long periods of operation at high temperature (285-325°C), thermal
aging occurs and the toughness decreases. Therefore, it is necessary to develop a
nondestructive inspection technique for evaluating degree of the deterioration.
We have studied the feasibility of using thermoelectric power (TEP) measurement
as a nondestructive inspection technique. Our previous studies reveal that the TEP
measurement can be an excellent indicator to detect the degree of the thermal aging of
cast duplex stainless steels [1, 2]. In addition, we have tried to detect the creep damage
of chrome steel from the TEP measurement. In the present paper, we review the typical
results concerning to the TEP measurement of the thermally aged cast duplex stainless
steels as well as the creep damaged chrome steel.
1
Corresponding Author: Graduate School of Engineering, Osaka University, 2-1 Yamadaoka, Suita,
Osaka 565-0871, Japan; E-mail: yamanaka@nucl.eng.osaka-u.ac.jp
154 S. Yamanaka et al. / Application of Thermoelectric Power Measurement to Nondestructive Testing
2. TEP Measurement of Thermally Aged Cast Duplex Stainless Steels
We prepared three kinds of cast duplex stainless steels with different ferrite content.
The ferrite content and chemical composition of each phase in the samples are
summarized in Table 1. The sample symbol implies the ferrite content. The samples
were heated at around 1100 °C for more than 6 hours and then cooled in water under
426 °C within 5 minutes. The optical micrograph of F23 sample is shown in Fig. 1. In
order to accelerate the thermal aging, the reference samples were aged at 400°C in air
for 100, 300, 1000, 3000 and 10000 hours.
Figure 2 shows the TEP measurement system used for the thermally aged cast
duplex stainless steels [2]. The system can be utilized for measuring the TEP of field
materials in nuclear power plants. Both the hot tip and the cold touch are made of
copper. The shape of the hot tip is a truncated cone with a plane (0.6 mm in diameter).
The hot tip is heated at 40 °C using a heating element. The temperature of the cold
touch is not controlled, being almost equal to room temperature and maintained at the
same temperature through the measurement. Therefore, the temperature deference
between the hot and cold electrode is maintained at the same level through all the
measurements. The TEP measurements were performed at 20 points for each specimen
by moving the hot tip in 0.5 mm increments.
Table 1. Chemical composition of each phase in cast duplex stainless steels determined by SEM–EDX
analysis (at.%)
Symbol Ferrite Phase Fe Cr Ni Si Mo Mn

content
(vol.%)
F8 7.3 Ferrite 63.55 25.32 5.54 1.05 3.65 0.89
Austenite 65.82 20.03 10.00 0.94 2.20 1.01
F15 14.6 Ferrite 64.67 24.62 5.17 1.35 3.39 0.80
Austenite 66.98 19.64 9.15 1.24 2.02 0.97
F23 21.3 Ferrite 63.78 25.52 5.57 1.67 2.91 0.55
Austenite 66.28 20.13 9.46 1.40 1.84 0.89
Austenite Phase
Ferrite Phase 100 Pm
Figure 1. Optical micrograph of cast duplex stainless steel (sample F23).

S. Yamanaka et al. / Application of Thermoelectric Power Measurement to Nondestructive Testing 155
(a)
Portable computer Instrumentation rack
Measurement cell
(b)
Figure 2. Appearance (a) and schematic view (b) of the TEP measurement system used for the cast duplex
stainless steels.
Figure 3 shows the TEP of the thermally aged cast duplex stainless steels as a
function of the aging time. The TEP values are negative and increase with increasing
aging time. Specifically, the TEP values of the starting samples (unaged samples) are -
2.65 PVK-1 (sample F8), -2.49 PVK-1 (sample F15), and -2.35 PVK-1 (sample F23).
The TEP values of the 10000 hours thermally aged samples are -2.39 PVK-1 (sample
F8), -2.16 PVK-1 (sample F15), and -1.73 PVK-1 (sample F23). In addition, there is a
clear correlation between the TEP and the ferrite content; that is a cast duplex stainless
steel containing a low amount of ferrite content exhibits large absolute TEP values.
From these results, it can be confirmed that the TEP is an excellent indicator of the
degree of thermal aging of cast duplex stainless steels. In other words, the degree of the
thermal aging can be detected by measuring the TEP of the materials.
Figure 4 shows a relationship between the TEP and micro Vickers hardness of the
thermally aged cast duplex stainless steels. It is observed that there is a clear
relationship between the TEP and micro Vickers hardness. The large TEP values
correspond to the large micro Vickers hardness. Therefore, the TEP measurement can
be an effective method for evaluating degree of the thermal aging as well as
degradation of the mechanical properties of cast duplex stainless steels.
Thermoelectric power, TEP (PVK )

-1
-1.5
Ferrite content: 7.3 %
-2.0
-2.5
-3.0
10 100 1000 10000
Aging time (h)
Figure 3. Thermoelectric power (TEP) vs. aging time of thermally aged cast duplex stainless steels.
260
Micro Vickers hardness (HV)
240 Ferrite content: 14.6 %

220
200
180
160
140
-2.8 -2.6 -2.4 -2.2 -2.0 -1.8 -1.6
-1
Figure 4. Relationship between thermoelectric power (TEP) and micro Vickers hardness.
3. TEP Measurement of a Creep Damaged Chrome Steel
We provided a Cr-Mo-V steel (SNB16) for the creep test and TEP measurement. The
chemical composition of the chrome steel is summarized in Table 2. The samples were
heated at 1010 °C for 2 hours, followed by air-cooled and maintained at 950 °C for 2
hours. And then, the samples were oil-quenched, reheated at 690 °C for 6 hours, and
air-cooled. The creep test was performed at 650 °C under the pressure of 25 MPa.
Figure 5 show a relationship between the creep time (hours) and creep strain (%),
in which a reasonable result is confirmed; that is the creep strain increases with
increase the creep time.
Table 2. Chemical composition of the chrome steel, Cr-Mo-V (SNB16), (wt.%).
Sample C Si Mn P S Cr Mo V Fe
SNB16 0.420 0.290 0.66 0.016 0.009 1.090 0.51 0.28 Balance
45
Time Srain
40
(hour) (%)
35 a 0 0
㪚㫉㪼㪼㫇㩷㫊㫋㫉㪸㫀㫅㩷㩿㩼㪀
30 b 368.9 4.14
c 620.0 6.66
25
d 760.8 6.46
20 e 1001.0 12.03
15 f 1560.7 16.06
g 1485.3 23.29
10
h 1200.5 29.43
5 i 2129.0 35.29
0 j 1953.9 38.57
0 500 1000 1500 2000 2500
㪚㫉㪼㪼㫇㩷㫋㫀㫄㪼㩷㩿㪿㫆㫌㫉㪀
Figure 5. Relationship between creep time and creep strain of the chrome steel. The creep test was
performed at 650 °C under the pressure of 25 MPa.
Electrode
Sample
Thermocouple
Small heater Electrode Small heater
Before sample setting After sample setting
Figure 6. TEP measurement system (ZEM 1, ULVAC Ltd.) used for the creep damaged chrome steel sample.

-1
8.5
(a)
8.0
7.5
7.0
6.5
6.0
0 500 1000 1500 2000 2500
Creep time (hour)
-1
8.5
8.0
(b)
7.5
7.0
6.5
6.0
0 10 20 30 40
Creep strain (%)
Figure 7. Thermoelectric power (TEP) of chrome steel (Cr-Mo-V steel: SNB16) as a function of the creep
conditions; (a) creep time, (b) creep strain.
Figure 6 shows an appearance of the TEP measurement system (ZEM 1, ULVAC

Ltd.) used for the creep damaged chrome steel. The sample is a rectangular-shaped
with the size of approximately 2 mm x 2 mm x 14 mm. The bottom of the sample is
heated by a small heater and the temperature gradient ('T) is detected by the
thermocouple attached to the surface of the sample. The voltage between one-side
wires of the thermocouple ('V) is measured, and the thermopower (TEP) is calculated
from a relation TEP='V/'T. The measurement was performed 21 times per sample.
The base temperature and temperature gradient were set as approximately 50 °C and
10 °C, respectively
Figure 7 shows the TEP of the chrome steel (Cr-Mo-V steel: SNB16) as a function
of the creep conditions; (a) creep time, (b) creep strain. In these figures, the average
TEP value and the standard deviation obtained from the 21 times measurements are
plotted. It is clearly observed that the TEP values are positive and increase with
increasing the creep time or creep strain. Specifically, the TEP value of the starting
sample (before the creep test) is 6.48±0.17 PVK-1 and that of the creep damaged
sample (2129 hours) is 7.74±0.15 PVK-1. In addition, a broad peak is observed at the
creep time of around 1000 h which corresponds to the creep strain of around 12 %. The
lattice defects such as dislocations arising from the creep tests may affect the TEP of
the chrome steel. Although the mechanism of this phenomenon is still unclear, it can be
said that the TEP measurement can detect the creep damage of the chrome steel.
In order to understand the relationship between the TEP and creep damage of the
chrome steel, we investigated a result of the creep damage evaluation from an
electromagnetic acoustic resonance (EMAR) analysis performed for the same samples
[3]. EMAR is an emerging ultrasonic spectroscopy technique for nondestructive and
noncontact materials characterization, relying on the use of electromagnetic-acoustic
transducers (EMATs) and the superheterodyne circuitry for processing the received
reverberation signals excited by long radio-frequency (RF) bursts [4,5].
Figure 8 shows the relationship between the attenuation coefficient and creep
strain for the creep damaged chrome steel. The attenuation coefficient increases,
showing a peak at the creep strain of around 3 %, and then decreases, showing a
minimum near the creep strain of around 5 %, and finally increases. From the SEM and
TEM observations, the behavior has been interpreted in terms of dislocation mobility
and restructuring. A similar peak is observed in the TEP vs. creep strain plot, but the
peak position and sharpness differ from those in the attenuation coefficient vs. creep
strain plot. Although a theoretical discussion has not been carried out at this moment, it
is supposed that there is a relationship between the TEP and attenuation coefficient viz.
dislocations of the creep damaged chrome steel.
Attenuation coefficient, D (Ps )
-1
0.01
1E-3
0 10 20 30 40
Creep strain (%)
Figure 8. Relationship between attenuation coefficient (D) and creep strain for the creep damaged chrome
steel [3]. The measurements were performed for the same samples as those used in the TEP measurements.
The attenuation discussed here is measured by exciting the test-specimen in one of its resonance frequencies,
stimulating a reverberation, and measuring the exponential decay of the ring down curve.
4. Summary
Thermal aging of cast duplex stainless steels and creep damage of a chrome steel were
evaluated though the thermoelectric power measurements. The TEP values of the
thermally aged cast duplex stainless steels were negative and increase with increasing
the aging time. In addition, there was a clear relationship between the TEP and micro
Vickers hardness of the thermally aged cast duplex stainless steels. The TEP values of
the creep damaged chrome steel were positive and increase with increasing the creep
time or creep strain, in which a peak was observed in the TEP vs. creep conditions plot
at the creep time and strain around 1000 hours and 12 %, respectively. A similar peak
has been observed in the attenuation coefficient vs. creep strain plot collected through
the EMAR analysis performed for the same samples, and the behavior has been
interpreted in terms of dislocation mobility and restructuring. Now, we are trying to
evaluate the electronic structure of iron with or without the lattice defects from first
principle simulations. The relationship between the TEP and lattice defects such as
dislocations will be clarified in near future. Nevertheless, the TEP measurement can
detect degree of thermal aging of the cast duplex stainless steels as well as creep
damage of the chrome steel. It can be concluded that the TEP measurement can be an
effective nondestructive analytical technique for evaluating material degradation.
References
[1] Y. Kawaguchi and S. Yamanaka, J. Japan Inst. Metals 66 (2002), 377-383.

[2] Y. Kawaguchi and S. Yamanaka, J. Alloys and Compd. 336 (2002), 301-314.
[3] T. Ohtani, H. Ogi and M. Hirao, Acta Materialia 54 (2006), 2705-2713.
[4] M. Hirao and H. Ogi, Ultrasonics 35 (1997), 413-421.
[5] H. Ogi, M. Hirao and T. Honda, J. Acoust. Soc. Am., 98 (1995), 458-464.
NDE by Magnetism and Magnetics
IOS Press, 2007
Accurate Detection of
Material Degradation of Stainless Steel
by ECT Sensor
Tokuo TERAMOTO1
Graduate School of Systems and Information Engineering,
University of Tsukuba, Japan
Abstract. Non-magnetic phase in austenitic stainless steel is often changed into

ferromagnetic phase by stress induced martensitic transformation, which is
accompanied by substantial plastic deformation. In order to detect the phase
transformation and to accurately evaluate the material degradation by
nondestructive method, an ECT sensor made of three tandem coils was used. The
tensile test at low and room temperatures showed that the measurement by this
sensor agreed well with the distributions of plastic stain, saturation magnetization
and Vickers hardness.
Keywords. ECT, Stress induced martensite, Magnetization, Plastic strain, Hardness
1. Introduction
Although the austenitic stainless steel used as structural material is originally a

semi-stable non-magnetic body, the applied stress often causes the martensitic
transformation especially at very low temperature. This transformation amount of
martensite, which is a ferromagnetic substance, depends upon residual plastic
deformation. Recently, applying the magnetic fluxgate sensor to the austenitic stainless
steel, several authors succeeded in clarifying the relationship between the leakage
magnetic flux density and the behavior of fatigue crack propagation[1], the bending
fatigue damage[2] and the creep behavior at high temperature[3]. Then, it was proved
that the sensor output strongly correlated with the material degradation and the
magnetization of residual martensite. This study aims at accurately evaluating the
material degradation such as plastic strain, hardness and amount of martensite in
stainless steel with the use of a simple ECT sensor.
2. Experimental
2.1. ECT Sensor
As shown in Figure 1, an ECT sensor made of three tandem coils was produced such
that the central coil was arranged as an excitation coil and two outside coils were used
1
Corresponding Author: Tokuo Teramoto, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki
305-8573, Japan; E-mail: teramoto@kz.tsukuba.ac.jp
164 T. Teramoto / Accurate Detection of Material Degradation of Stainless Steel by ECT Sensor
as a detection coil of which the two output voltages were electronically compensated as
long as this sensor was away from any electric conductive body. The output was
amplified in an appropriate degree by a lock-in amplifier. Then, the measurement by
this sensor was conducted with high sensitivity. Each coil consisted of a small pan-cake
type coil of 50 turns, 4.9mm in outer diameter, 2.5mm in inner diameter and 2mm in
thickness. The excitation frequency was always set at 100kHz and the lift-off was set at
1mm. The size of this ECT sensor was expected to be small enough to accurately detect
the distribution of material degradation such as plastic strain, hardness and martensite
content. The output of the ECT sensor corresponds to the real part in terms of lock-in
amplifier output.
2.2. Specimen
The material used is a SUS304 stainless steel, whose chemical composition is shown in
Table 1. Although a SUS304 stainless steel usually shows ductile and non-magnetic
properties, it can become brittle and ferromagnetic by stress induced martensitic
transformation because of unstable austenitic phase around and below room
temperature[4]. Figure 2 shows the standard and the modified tensile test specimen
where the center part is 2mm thick in both specimens. The top specimen was produced
to realize in-situ measurement by ECT sensor when loaded to an extent and unloaded at
300K. This specimen yields uniform plastic strain in specimen center part. In-situ
measurement may be appropriate even if the sensor position moves during test. In-situ
measurement, however, is very unstable at 77K because the electromagnetic properties
Table 1. Chemical composition of SUS304 (wt%)
Material C Si Mn P S Ni Cr
SUS304 0.04 0.4 1.13 0.027 0.003 8.23 18.0
Figure 1. Schematic diagram of measurement equipment with ECT sensor

T. Teramoto / Accurate Detection of Material Degradation of Stainless Steel by ECT Sensor 165

5

˓

5

˓
Figure 2. Tensile test specimen (top) and modified tensile test specimen (bottom) [unit in mm]
of coil changes in a great degree. Then, using standard tensile specimen, in-situ
measurement was conducted only at 300K. On the other hand, the bottom specimen
was produced to conduct the measurement along the longitudinal direction after loaded
at 77K and 300K and to investigate the distribution of the ECT output with plastic
strain. The modified tensile specimen was used to investigate the relation between the
ECT output and the plastic strain after test. Since the plastic strain changes
monotonically along the longitudinal direction of the specimen, it is convenient to
compare the ECT output with the plastic strain.
3. Results and Discussion
3.1. Tensile Test at 300K
First, the tensile test was carried out at 300K with several loading and unloading
processes after the ECT sensor was mounted on the specimen center surface. The
output of the sensor was measured continuously during the test. Figure 3 shows the
variation of applied stress and ECT output with strain at 300K. The ECT output is
almost constant until applied strain reaches 5 to 10% where the specimen is obviously
yielded. Namely, the variation of the microscopic structure by yielding does not affect
the electromagnetic parameter of this material in the early stage. The ECT output,
however, increases with increasing strain after this stage. This behavior corresponds
with stress-strain relation according to magnetism change by the increase of martensitic
phase. The martensitic phase was found in the intensively strained body even at room
temperature by measuring the magnetization with the use of VSM. In addition, MFM
observation was made on 2% and 8% strained specimen as shown in Figure 4(a) and
(b), respectively. The magnetic domain is found relatively clear at 8% strain while it is
obscure at 2% strain. Although a clear evidence of martensitic transformation can not
be drawn from these MFM images, the appearance of magnetic domain may be
correlated with martensitic transformation. Accordingly, the martensitic transformation
is possibly initiated at 5 to 10% applied strain. Next, the ECT output increases even at
700 -150
Pick-up voltage Vx [mV]

600 Stress
Vx
-200
500
Stress [MPa]
400
-250
300
200
-300
100
0 -350
0 5 10 15 20 25 30 35
Strain [%]
Figure 3. Variation of stress and pick-up voltage with strain

(a) (b)
Figure 4. MFM images of (a) 2% and (b) 8% strained specimen, respectively
unloading process. This behavior may be attributed to variation of electric conductivity

accompanied by elastic strain release due to unloading.
3.2. Modified Tensile Test at 77K
Using schematically modified test specimen, the tensile load was applied at 77K until
the development of martensitic transformation was anticipated to an extent. After the
tensile test, the output of the sensor was measured at room temperature on the specimen
surface. Figure 5 shows the distribution of ECT output and residual plastic strain along
15
-100
d=6mm
Plastic strain [%]

-150 10
-200
Pick-up voltage
Plastic strain
-250 5
d=2mm
-300
0
0 5 10 15 20 25
Distance from center [mm]
Figure 5. Distribution of pick-up voltage and plastic strain along the longitudinal direction
50
-100
Magnetization [emu/g]
d=6mm 40
-150
30
-200
Pick-up voltage
Magnetization 20
-250 d=2mm
10
-300
0
0 5 10 15 20 25
Distance from center [mm]
Figure 6. Distribution of pick-up voltage and magnetization along the longitudinal direction
the longitudinal direction of the specimen. The ‘d’ in this figure means the applied
displacement between two loading pins of the specimen in the test. The plastic strain
was numerically obtained by using a finite element analysis code. The
amplitude-locus-curve of ECT output agrees well with that of plastic strain. Figure 6
shows the distribution of ECT output and the value of the saturation magnetization
along the longitudinal direction of the specimen. Using VSM, the magnetization was
Volume fraction of martensite [%] 30

77K
25 300K
20
15
10
5
0
-400 -300 -200 -100 0
Pick-up voltage V x [mV]
Figure 7. Relation between pick-up voltage and volume fraction of martensite
50
77K
40 300K
Plastic strain [%]
30
20
10
0
-400 -300 -200 -100 0
Pick-up voltage V x [mV]
Figure 8. Relation between pick-up voltage and plastic strain
measured from small pieces cut from the modified tensile specimen in the longitudinal
direction. In this case, the tendency of ECT output also agrees well with that of
magnetization. Namely, the agreement means that the plastic strain quantitatively
corresponds with the martensite content at 77K and that the extent of martensitic
transformation can be accurately detected by this ECT sensor.
3.3. Prediction of Material Degradation
Figure 7 shows the relations between ECT output and volume fraction of martensite at
77K and 300K, respectively. Using Slater-Pauling curve, the saturation magnetization
of SUS304 was estimated to be 168 emu/g when the austenitic phase is fully
transformed to martensitic phase[5]. In the present study, the volume fraction of
martensite was obtained by normalizing magnetization by this value. The ECT output
is almost linearly proportional to the volume fraction of martensite at both temperatures.
However, there is an obvious difference in this relation at 77K and 300K. Figure 8
shows the relations between ECT output and plastic strain at 77K and 300K,
respectively. It is found that the ECT output is linearly proportional to plastic strain at
77K and that the martensitic phase begins to come out just after yielding. On the other
hand, the ECT output is almost constant at 300K until plastic strain reaches about 8%
and then begins to linearly increase with increasing plastic strain. The plastic strain
over 8% may produce martesitic transformation at 300K because the ECT output
begins to change at this strain amount. Certainly, large plastic strain yields some
amount of martensite but causes the reduction of electric conductivity at 300K. The
sensor output is relatively small because of lower conductivity and larger skin depth at
300K. On the other hand, small plastic strain yields more martensitic transformation at
77K and then the sensor output becomes large because of higher permeability and
conductivity. It is also found that the Vickers hardness is almost linearly proportional to
ECT output at 77K and 300K, respectively. Especially the hardness is correlated with
martensite content at 77K while it is not attributed to martensite content but to large
plastic strain at 300K.
4. Conclusion
At low and room temperatures, it is possible to accurately detect the material

degradation of SUS304 stainless steel with the use of this nondestructive ECT sensor.
Especially, a unique relationship between the ECT output and the material degradation
such as martensite content, plastic strain and hardness, is obtained at low temperature.
On the other hand, the ECT output is not only dependent upon martensite content but
change of electromagnetic properties due to large plastic deformation at room
temperature.
References
[1] Y. Nakasone et al., Non-destructive Detection of Damage in an Austenitic Stainless Steel SUS 304 by the
Use of Martensitic Transformation, Int. J. of Applied Electromagnetics and Mechanics 15 (2001/2002),
309-313.
[2] M. Oka et al., Evaluation of the Amount of Fatigue Damage in Austenitic Stainless Steel by the Leakage
Magnetic Flux, Proc. of the 14th MAGDA Conf., Gifu, 2005, pp. 170-175.
[3] Y. Nagae and K. Aoto, A Study on the Detection of Creep Damage in Type 304 Stainless Steel Based on
Natural Magnetization, Int. J. of Applied Electromagnetics and Mechanics 15 (2001/2002), 295-300.
[4] R.P. Read and C.J. Gunter, TMA-AIME 230, 1964, p.1713.
[5] JSAEM-R-0005, 2001.
IOS Press, 2007
Micromagnetic Characterization of
Thermal Degradation in Cu-rich Alloys and
Results of Neutron-Irradiation
Madalina PIRLOGa,1, Iris ALTPETERa, Gerd DOBMANNa, Gerhard HÜBSCHENa,
Melanie KOPPa, Klaus SZIELASKOa
a
Fraunhofer-Institute for non-destructive testing (IZFP), Saarbrücken, Germany
Abstract. The Cu-rich German martensitic/bainitic structural steel WB36 (15

NiCuMoNb 5, 1.6368) is in service in fossil power plants as well as in pipes and
pressurizer vessels of German nuclear power plants. That material – when exposed
at elevated temperatures in between 280°C - 350°C – shows the effect of
precipitation hardening by Cu-rich precipitates and a shift 'FATT = 70°C can be
observed. Micromagnetic measurements based on multiple linear regression or
pattern recognition algorithms and taking into account different micromagnetic
quantities were applied in order to characterize the material degradation. In
parallel a different approach was employed to evaluate in the same way alloys
based on pure iron with different contents of Cu. Because Cu-rich precipitates play
an essential role when nuclear pressure vessel material is exposed to neutron
irradiation, micromagnetic measuring quantities are also of interest to characterize
these materials. Therefore the paper reports experiments performed in the hot cells
of AREVA NP in Germany.
Keywords. Thermal ageing, neutrons embrittlement, non-destructive, micro-

magnetic
1. Introduction
Non-destructive material characterization techniques have traditionally been employed

to detect, classify and to size defects in materials. However in the last two decades a
significant amount of effort has been invested, to develop NDT technique which can
reliably characterize materials in terms of properties describing the fitness for use. In
case of the power plants components such as pressure vessels and pipes the fitness for
use under mechanical loads is characterized in term of the determination of mechanical
properties like mechanical hardness, yield and tensile strength, toughness, fracture
appearance transition temperature, fatigue strength or usage factor. Except hardness
tests which are weakly invasive, all of these parameters can be determined by using
destructive tests on special standardized samples. Such a procedure cannot be
performed on components in service and is therefore restricted to quality checks during
manufacturing where enough representative material is available. Procedures based on
less destructive material sampling and weakly influencing the integrity of the
1
Corespondent author: Dr.-Ing. Madalina Pirlog, Fraunhofer-Institute for non-destructive testing
(IZFP), University of Saarland, Saarbrücken, 66123, Germany Email: madalina.pirlog@izfp.fraunhofer.de
M. Pirlog et al. / Micromagnetic Characterization of Thermal Degradation 171
component are not yet validated or standardized and are still the objective of
investigations. Therefore, there is a need for the development of such non-destructive
testing techniques.
2. Characterization of the thermal degradation effects in Cu-rich alloys
Smallest changes in the materials state e.g. the change of the precipitation of coherent
Cu particles induced by the thermal ageing sensitively affect the magnetic domain
structure.
2.1. Thermal degradation of WB 36 steel
In the typical “as delivered” state of WB 36, half of the contained Cu is already
precipitated, while the other half remains in solid solution. After long term service
exposure above 320 °C, damage was observed due to further precipitation of Cu; an
increase in yield strength 'Vy = +150 MPa and a shift of the fracture appearance
transition temperature 'FATT = +70 °C can be measured. Small angle neutron
scattering measurements revealed the fact that the mechanical properties changes are
caused by Cu precipitates ranging from 1 to 3 nm in size [1]. The particles are coherent,
have a bcc structure that induces a high level of compressive residual stress in their
vicinity, balanced by tensile stresses in the environmental matrix. On a set of
approximately 70 round samples (80 mm in length, diameter 6 mm) of WB 36, thermal
service exposure was simulated in an accelerated manner through long-term annealing
at 400 °C. A U-shaped electromagnet was used to excite an alternating magnetic field
along the longitudinal axis of the sample. A disc-shaped pickup coil and a temperature
stabilized hall probe were used to record Barkhausen noise events and magnetic field
strength, respectively. The Barkhausen noise signal was amplified by 60 dB and
bandpass-filtered to a range of 5 to 200 kHz. All signals were digitized using common
data acquisition hardware. The Barkhausen noise signal was then digitally refiltered for
separate analysis of its different frequency components. Characteristic scalar quantities
[2] were extracted from the envelope of the Barkhausen noise signal as a function of
the applied magnetic field strength. Moreover, an upper harmonics analysis [3] of the
magnetic field strength signal was performed and characteristic quantities derived. As
changes in conductivity may be expected due to copper precipitation, a simplified eddy
current analysis procedure was performed based on the relationship between magnetic
field strength and exciting voltage of the electromagnetic coil. The scalar result
quantities of all three methods (Barkhausen noise, upper harmonics and eddy current
analysis) are combined to a vector which characterizes the material condition.
In the case of WB 36 steel, only few of the measured quantities seem to correlate
well with the copper precipitation state at first, when comparing them across all
samples mentioned above. Therefore, a preliminary experiment was performed in order
to identify the optimum measurement parameters and most significant quantities for the
detection of copper precipitation in WB 36. Instead of comparing the electromagnetic
properties across several samples, the changes in the electromagnetic properties of
single samples were recorded in several stages of the simulated service exposure. The
annealing was therefore interrupted in regular intervals where electromagnetic tests
were performed. All measurements were done using a fixed set of different
172 M. Pirlog et al. / Micromagnetic Characterization of Thermal Degradation
magnetization frequencies and magnetic field amplitudes in order to create a

comprehensive database of material behaviour. A total 6 samples underwent this
procedure for statistical coverage, and reference samples were kept for verification.
As a result of these experiments, it was found that eddy current measurements
allow the detection of copper precipitates whilst remaining insensitive to most
disturbing influences. Fig. 1 shows how the coil impedance (expressed in terms of its
relative magnitude and phase) performed as a function of the simulated service
exposure duration for initially recovery-annealed WB 36. Changes in the eddy current
impedance represent changes in the conductivity and the permeability of the material.
The electrical conductivity is proportional to the concentration of the free electrons and
their velocity, which decreases with rising defect density. Therefore the electrical
conductivity decreases. It was also shown that the increase of the Cu precipitates
density causes a decrease of the relative magnetic permeability. In the initial phase of
the experiment, some scattering of the measured values was observed. Measurements
on reference samples have shown that these deviations from the subsequent material
behaviour relate to the polishing procedure which was required for additional Vickers
hardness tests.
400 0.4
magnitude
phase
200 0.2
relative magnitude [μV]
relative phase [°]
0 0
-200 -0.2
-400 -0.4
0 250 500 750 1000
duration of service simulation at 400 °C [h]
Figure 1. Eddy current quantities magnitude and phase as a function of service exposure duration for initially
recovery-annealed (3h / 600 °C) WB 36 steel
Fraunhofer-IZFP uses the so-called 3MA method (“Micromagnetic Multi-Parame-

ter Microstructure and Stress Analysis”) in order to solve the inverse problem of target
quantity prediction from a limited set of calibration data [4]. In this case, a specialized
pattern recognition algorithm [5] based on nearest neighbour search was used to obtain
approximate values of the Vickers hardness (HV 5) from several quantities, including
the eddy current quantities mentioned above. The optimum parameters which were
found in the preliminary experiments were used throughout this measurement. Fig. 2
shows the resulting correlation of actual and predicted hardness values.
215
Calibration
Test
predicted HV 5 (3MA)
195
175
155
155 175 195 215
actual HV 5
Figure 2. Predicted Vickers hardness for WB 36 versus actual Vickers hardness
2.2. Thermal degradation of Fe-Cu alloys
The micromagnetic test methods have a high potential to detect change of the MRS
because they sensitively react to the changes of the domain wall configuration [6]. The
Cu precipitates coherently embedded in a ferritic matrix induce two different kinds of
MRS: coherence tensile MRS of IIIrd order and thermally-induced compressive MRS of
IInd order. The coherence MRS of the IIIrd order appear when the lattice parameter of
the IInd phase particles coherently embedded in the matrix and the lattice parameter of
the surrounding matrix are different. The thermally-induced MRS of IInd order arise at
the interface between different material phases because of their different thermal
expansion coefficient. In case of the Fe-Cu system the bigger lattice parameter and
thermal expansion coefficient of Cu compared to Fe lead to coherence tensile MRS of
IIIrd order and thermally-induced compressive RS of IInd order respectively. In order to
detect the precipitation-induced MRS, Barkhausen noise measurements under
superimposed tensile load stress were performed [7]. The samples were magnetized in
the longitudinal uniaxial load stress direction. Due to that fact by means of that
approach the detected residual stresses are those in the direction of external
magnetization. That means that a change of the residual stress state in the perpendicular
direction caused by the applied load will not be detected. The Barkhausen noise signal
was recorded by using two differential air-core coils to separate the influence of the
energizing magnetic field. The magnetic Barkhausen noise was triggered by an
alternating magnetic field (Hmax = 5 A/cm, fE = 40 Hz) applied to the sample using an
U-shape electromagnet.
The noise signal was recorded as induced voltage, appropriately filtered, rectified,
amplified and displayed as function of the tangential field strength. The resulting so-
called Barkhausen noise profile curve was evaluated by a computer with respect to the
maxima and their respective magnetic field strengths. Fig. 3 left- hand side, shows
schematically the maximum of the Barkhausen noise amplitude (MMAX) obtained
during one hysteresis cycle recorded as a function of the load-induced stress, V. Fe-Cu
samples with a variation of Cu content from 0.65 to 2.1 wt.% have been manufactured
and investigated. By means of a suitable heat treatment (solution annealing at
1000 °C/2h, quenching into water and thermal ageing at 550 °C variation of the amount
of coherent Cu particles of 1.42 Vol.% was induced. The thermal ageing time was
determined by means of hardness measurements performed during thermal ageing in 10
minutes steps. The thermal aging time for each sample must be shorter than the thermal
aging time corresponding to the hardness maximum. Those results were validated by
SANS-measurements. In case of the Fe-Cu alloys the MMAX(V) curves do not reach any
maximum before break under load, i.e. the curve of Figure 3 right-hand side is strongly
shifted to the right. Therefore the position of the local minimum (VMMIN) is evaluated
(Fig. 3 right-hand side).
0.5
Fe-0.65 wt.% Cu
MMAX Fe-2.1 wt.% Cu
RS-free 0.4
MMAX [V]
tensile RS
0.3
compressive RS 0.2
_ +
'V § 50 MPa
VMMIN 0 V [MPa] 0.1
VMMIN
0 75 150
V [MPa]
Fig. 3b: Schematical (left-hand side) and experimentally determined (right-hand side) tensile load
dependence of maximum Barkhausen noise amplitude
X-Ray measurements performed on two Fe-Cu samples containing 0.65 and 2.1
wt.% Cu in the “as-quenched” state showed that both samples contain compressive RS
of Ist order with values of -210 MPa and -290 MPa respectively. X-Ray measurements
performed after thermal ageing documented that both samples still contain compressive
RS but with smaller values, of -120 MPa and -180 MPa, respectively. Those
measurements confirmed the fact that the behaviour of the MMAX(V) curves is mainly
influenced by quenching. However, the curve is a superimposition of RS induced by
quenching (largest effect) with the coherent tensile RS of IIIrd order and thermally
induced compressive RS of IInd order. By subtracting the compressive residual stresses
induced by quenching and taking into account that the thermally-induced MRS of IInd
order are negative too, values of the coherence MRS of IIIrd order higher than 90 MPa
and higher than 110 MPa for the Fe-Cu samples containing 0.65 and 2.1 wt.% Cu
respectively (Fig. 4) were obtained [8] by evaluating the shift in the minimum ('MRS).
The minimum of the MMAX(V) curves indicates the load stress value which is required
by the Bloch walls to break away from the lattice defects as pinning points. Therefore
the position of that minimum is a measure for the lattice defects density. An increase of
the lattice defects density and an increase of the MRS order causes a shift of the
MMAX(V) curve to higher tensile load stresses because the Bloch walls need more
energy to break away from the lattice defects. This means that the shift ('V) of the
MMAX(V) curve (Fig. 3, right-hand side) indicates the in-crease of the precipitation-
induced MRS. Because the thermally-induced RS of IInd order are expected to be very
small, their influence can be neglected and it can be assumed that the shift of the
MMAX(V) curve represents the change of the coherence tensile MRS of IIIrd order. It is
well known that the coherence MRS increase with increasing density of Cu precipitates.
A mathematical relation between the coherent MRS and the density of precipitates is
unknown. For simplification the MRS in the others Fe-Cu samples were determined by
interpolation (Fig. 4). The variation of the MRS between the both Fe-Cu alloys is
bigger than 20 MPa, which confirm the value obtained by means of the micromagnetic
procedure, if we consider the error band of the X-Ray measurement of about 25 MPa
(Fig. 4).
Figure 4. Comparison between the MRS determined by means of the micromagnetic

and by means of the X-ray methods
3. Characterization of the neutron irradiation induced embrittlement
Depending on the specific design of a pressure vessel – which is different in the

different countries of the world – the pressure vessel material in nuclear power plants
(NPP) exposed to neutron flux is in a range between 5x1018 n/cm2 in 32 years at 288 °C
in Germany and 8x1019 n/cm2 in 14 years at 254 °C, for instance, in France. The energy
input of the neutrons is directly producing lattice defects like vacancies and indirectly
by stimulating the precipitation of coherent Cu-rich particles. As in case of the
thermally-induced embrittlement of the steel WB 36 these are also in the 3 nm diameter
range and coherently embedded in the bcc lattice. Both, the vacancies and the
precipitates reduce the toughness of the material, which can be characterized by a
reduction of the Charpy energy and a shift in the fracture appearance transition
temperature to higher temperatures. In practice the material degradation is
characterized in surveillance programs by using standardized Charpy V-notch
specimen and tensile test specimen made of the pressure vessel steel and its weldments.
The specimens are exposed in special radiation chambers near the NPP core at a higher
fluence than at the i.d. surface of the pressure vessel wall. These specimens from time
to time are removed from the chambers and used for destructive tests. In order to
assume a higher nuclear safety in between two subsequent destructive tests one would
like to have many non-destructive tests and the nd-technology should also be
developed to an in service inspection method to be applied at the pressure vessel inner
surface. In order to characterize the neutron irradiation-induced embrittlement samples
(10*10*55 mm) made of 20 MnMoNi 55 steel used in the German reactor pressure
vessel in non-irradiated state as well as after irradiation of 3.78x1018 n/cm2,
7.66x1018 n/cm2 and 1.05x1019 n/cm2 neutron flux have been investigated. As in case of
the thermally-induced embrittlement of the steel WB 36, it was observed that a suitable
measuring quantity for the characterization the neutron irradiation-induced
embrittlement is the eddy current impedance magnitude which decreases with the
fluence (Fig. 5).
1.098 10
1.092
'T41 [°C]
6
Vmag [V]
4
1.086
1.08 0
0 4 8 12 0 3 6 9 12
18 2 18 2
Fluence*10 [n/cm ] Fluence*10 [n/cm ]
Figure 5. Correlation between the eddy current Figure 6. Dependency of the fracture appearance
impedance magnitude and the neutron flux transition temperature on the neutron fluence
IZFP has applied 3MA-approaches to calibrate regression models. The

measurements were performed in the same way like in case of the steel WB 36. One
part of each specimen set was used to calibrate and the other part – independently
selected – was taken to test the model. In this case, a specialized pattern recognition
algorithm [5] based on nearest neighbour search was used to obtain approximate values
of the shift of the fracture appearance transition temperature, which is a measure for the
embrittlement from the measured quantities (Fig. 6).
10.5
Calibration
Test
7.5
'T41 (3MA) [°C]
4.5
1.5
-1.5
-1.5 1.5 4.5 7.5 10.5
'T41 (Charpy-Test) [°C]
Figure 7. Prediction of the shift of the fracture appearance transition temperature by 3MA
It was demonstrated that by means of the 3MA-procedure the prediction of the

shift of the fracture appearance transition temperature is possible (Fig. 7). By using a
regression analysis algorithm a correlation coefficient of 89.3% and a residual standard
deviation of 0.35 °C is received. By testing the calibration with independently selected
test specimens a standard error of 1.2 °C (residual standard deviation to the destructive
test values) was obtained.
4. Conclusions
In the present research work the nd-characterization of the material degradation was
discussed on the basis of experiences with: precipitation-induced embrittlement in
WB 36 steel and Fe-Cu alloys and neutron-irradiation induced embrittlement in the
20 MnMoNi 55 steel used in the German reactor pressure vessel. It was shown that the
3MA approach using pattern recognition methods based on magnetic Barkhausen noise,
upper harmonics analysis and eddy current analysis data allows with high accuracy a
non-destructive prediction of the Vickers hardness and of the shift of the fracture
appearance transition temperature, which are measures to characterize embrittlement.
The present study shows also that eddy current impedance measurements represent a
suitable method to characterize the thermally- as well as the neutron irradiation-
induced embrittlement.
The present research results show the suitability of the micro-magnetic method for
the non-destructive evaluation of Cu precipitation-induced MRS. By means of a
micromagnetic non destructive procedure based on the tensile stress dependency of the
maximum Barkhausen noise amplitude the change of MRS caused by an increase of the
coherent Cu precipitates volume fraction in Fe-Cu based alloys was measured.
References
[1], D. Willer, G. Zies, D. Kuppler et al, Service-Induced Changes of the Properties of Copper-Containing
Fer-ritic Pressure-Vessel and Piping Steels. GRS Reactor Safety Research – Project No 150 1087. Final
Report avail-able at GRS, Cologne, (2001)
[2] I. Altpeter, Spannungsmessung und Zementitgehaltsbestimmung in Eisenwerkstoffen mittels dynamischer
magnetischer und magnetoelastischer Messgrößen. Dissertation at Saarland University, Saarbrücken, (1990)
[3] H. Pitsch, Die Entwicklung und Erprobung der Oberwellenanalyse im Zeitsignal der magnetischen
Tangentialfeldstärke als neues Modul des 3MA-Ansatzes (Mikromagnetische Multiparameter
Mikrostruktur und Spannungsanalyse). Dissertation, Saarland University, Saarbrücken, Germany, (1989)
[4] G. Dobmann and P. Höller, “Non-destructive determination of material properties and stresses”, 10th Int
Conf NDE in the nuclear and pressure vessels industries, Glasgow, ASM International, (1990)
[5] R. Tschuncky: Entwicklung eines Mustererkennungs- und Klassifikationsmoduls für die indirekte
Charakterisierung von Werkstoffeigenschaften, Diploma Thesis, Saarland University, Germany, 2004
[6] E. Kneller, Ferromagnetismus, Springer Verlag, (1962)
[7] I. Altpeter, R. Becking, R. Kern, M. Kröning, S. Hartmann, Mikromagnetische Ermittlung von thermisch
induzierten Eigenspannungen in Stählen und weißem Gusseisen, in: Deutsche Forschungsgemeinschaft,
Eigenspannungen und Verzug durch Wärmeeinwirkung, Forschungsbericht, Wiley-Vch, (1999), 407/426
[8] Pirlog, M., Szielasko, K., Altpeter, I., Dobmann, G., Kröning, M.: Micro-Magnetic Evaluation of
Residual Stresses of the 2nd and 3rd Order, in VDI Berichte Nr. 1899 (2005), 355-365, ISBN 3-18-
09189-3
[9] Altpeter, I., Szielasko, K., Dobmann, G.: “Optimization and Assessment of electromagnetic testing
methods for the detection of property changes in power plant components, caused by service-induced
copper precipitation”. Final report to GRS reactor safety research project no. 150 1269, available at GRS
Forschungsbetreuung, Köln, Germany or at IZFP, report no. 060119-TW (2006)
IOS Press, 2007
Challenges in Quantifying Barkhausen

Noise in Electrical Steels
Anthony J MOSES 1, Harshad V PATEL and P I WILLIAMS
Wolfson Centre for Magnetics, School of Engineering, Cardiff University, Wales UK
Abstract. The origin and interpretation of (a.c.) magnetic Barkhausen effect is still
not fully understood. In this paper Barkhausen noise is measured by two methods,
surface probes and enwrapping flux-sensing search coils. They are shown to
follow similar trends with varying a.c. flux density and frequency in electrical steel.
It is shown that the rate of Barkhausen noise data capture above 100 kHz at 50 Hz
magnetizing frequency does not appreciably affect data interpretation. Various
methods of signal analysis are compared and correlations of with static hysteresis
loss have been demonstrated. Trends are found also in the influence of factors such
as texture and microstructure under different magnetizing conditions in electrical
steels although localized Barkhausen noise varies widely from point to point
within grains and near grain boundaries. It is concluded that Barkhausen noise
measurement can be a significant tool for analyzing physical and microstructural
factors in electrical steels but there are challenges in the interpretation of results at
present.
Keywords. Barkhausen Noise, Electrical Steel, Hysteresis Loss
1. Introduction
The magnetic Barkhausen effect was discovered at the beginning of the last century but
its origin and interpretation is still not fully understood. It is generally accepted to be
due to microscopic discontinuities in domain wall motion due to the presence of defects
[1]. Various methods of measurement based on detecting induced voltages caused by
Barkhausen jumps within a material have been developed. Definite trends are found in
the influence of factors such as texture and microstructure on Barkhausen noise (BN)
under different magnetizing conditions in electrical steels and some direct connection
with hysteresis loss appears to exist [2]. However, quantitative correlation of methods
of BN signal analysis is not well documented.
1
Corresponding Author: Anthony J Moses, Wolfson Centre for Magnetics, School of Engineering,
Cardiff University, Wales, UK, CF24 4AA; E-mail: mosesaj@Cardiff.ac.uk
A.J. Moses et al. / Challenges in Quantifying Barkhausen Noise in Electrical Steels 179
2. Measurement Techniques and Analysis
Epstein strip samples of electrical steels were tested at frequencies up to 100 Hz and
sinusoidal peak flux density up to 1.5 T using the magnetization method described
previously [3]. Figure 1 schematically shows two types of sensor arrangement, which
have been compared, namely enwrapping coils and surface probes. In both methods,
the output voltage must be filtered to remove unwanted noise and Faraday emfs of non
BN origin before analyzing. A band pass filter was used so that signals in the range 3.5
kHz to 100 kHz were detected at magnetizing frequencies in the range 5 Hz to 100 Hz.
The component of emf induced in coils, enwrapping the complete width of a single
strip, originating from the BN has been detected and analyzed. The main challenge of
this type of measurement is that the BN component is typically some 2000 times
smaller than the predominate 50 Hz Faraday secondary emf in the coil. Also high order
harmonics of the filtered fundamental component may still be present due to non-
sinusoidal voltage components caused by grain to grain flux distortion that cannot be
eliminated even using advanced flux control systems such as described in [3]. The BN
detected in a single 80 turn enwrapping coil was compared with that detected from a
double coil arrangement. The latter consisted of two 80 turn coils separated by 5 mm
and connected in series opposition. The predominant 50 Hz Faraday emf is cancelled
out by the double coil because of its differential output leaving only the BN component.
The BN component is only partially cancelled out since each coil picks up a different
Barkhausen contribution.
The second method involved the use of a 1000 turn and 500 turn ferrite core
detection coil 8 mm in length placed on the surface of the strip. Sensors supplied by
Stresstech Oy with cores of length 9.5 mm, 13 mm and 24 mm were tested. The core
diameter was 3 mm and the coil position relative to the sample surface remained
unchanged as shown in figure 1(c).
Because of the stochastic nature of BN, a variety of statistical methods can be used
to analyse the signals using both detection methods. The sum of the amplitudes of BN
peaks is convenient but does not count simultaneously occurring peaks or take account
of the contribution of peaks of different magnitudes whereas measurement of the power
spectrum does contain such information. Kurtosis is a parameter that describes the
peakedness of a BN distribution relative to a normal distribution. In this investigation
we focussed our attention on correlating RMS BN noise with hysteresis loss. RMS BN
was chosen because its value is fairly stable from cycle to cycle. The RMS value is also
very useful for determining amplitude as a function of time particularly for cyclic
phenomena such as BN which has both positive and negatives amplitudes per cycle. It
also allows better comparison with the work of other investigators that in the main have
also measured RMS BN. However we would expect to draw similar conclusions if the
BN signal were to be analysed in terms of the other parameters.
The BN evaluated from the outputs of single and double coils as well as the three
sizes of ferrite cored sensors were measured with dummy cardboard and aluminium
samples in an attempt to quantify and compare the inherent measurement error. As the
180 A.J. Moses et al. / Challenges in Quantifying Barkhausen Noise in Electrical Steels
magnetizing field at the sensor position was increased from zero to 1200 A/m at 50 Hz
the output of all sensors apart from the double enwrapping coil initially increased
rapidly and then levelled out. The output from the enwrapping coils was lowest and the
24 mm ferrite cored sensor had the highest induced noise but its RMS value was 93 %
less than that of the typical measured BN output of a steel sample at 1.5 T, 50 Hz.
There was no obvious difference between noise outputs when the aluminium and
cardboard samples were present. The random variation of BN measurements is partly
due to non repeatability of domain wall motion from cycle to cycle [4].
PC with DAQ card PC with DAQ card

Search Coil Outputs and LabVIEW Search Coil Outputs and LabVIEW
Sample software package Sample software package
Search Coil Carrier Search Coil Carrier
(a) (b)
Ferrite PC with DAQ card

Search Coil Outputs and LabVIEW
Sample software package
Search Coil Carrier Pick up Coil
(c)
Fig. 1. Schematic diagrams of sensors. (a) Single enwrapping search coil (b) Double
enwrapping search coil wound in series opposition (c) Ferrite cored induction coil
Figure 2 compares the outputs of a single search coil, double search coils wound in
series opposition and ferrite cored surface sensor with 500 & 1000 turn pick up coils
with 13 mm ferrite core length for a sample of commercial 0.27 mm thick grain
oriented 3 % silicon steel magnetized from 0.1 T to 1.8 T at 50 Hz. Obviously the
magnitude of the output signal depends on the sensor characteristics as well as the BN
events so the results are normalized as shown. It is interesting that the trends with the
ferrite cored surface probe are very similar to those of the enwrapping coils although
they are oriented perpendicular to each other. However, it can be seen that the change
of BN RMS value with flux density detected by the ferrite cored sensor with 1000 turn
pick up coil is highest and that of the double search coil is the lowest. The ferrite cored
sensor outputs increase rapidly at high flux density indicating that the different sensors
are effectively picking up different BN events. It is important to note that increasing the
number of turns in the ferrite cored sensor from 500 to 1000 turns does not simply
double the BN measurement. In fact it can be seen that the increase is more than twice
approximately 3.3 times more, this contribution maybe due to the area enclosed by the
coil is larger, and signals are amplified with more turns hence data acquisition card can
measure more BN jump signals. There maybe other factors that increase the BN signal,
further investigation is needed.
In the case of the double coil, some events will be lost in the subtraction process
but even here, it is interesting to note that as the separation between the coils is
changed the magnitude of the output changes. For a coil separation distance of 5 mm to
40 mm the BN RMS signal increases around 140 %.
1.6E-3
Ferrite Core
1.4E-3 1000 turn coil
1.2E-3
1.0E-3
BN RMS [V]
800.0E-6
600.0E-6
Ferrite Core
500 turn coil
400.0E-6
200.0E-6 Single Coil
Double Coil
000.0E+0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
B [T]
Fig. 2. Variation of normalized BN RMS outputs of the sensors with flux density at 50
Hz in grain oriented 3 % silicon steel (ferrite length 13 mm in both ferrite sensors)
Based mainly on early d.c. measurements Barkhausen jumps were thought to occur
very rapidly and therefore be largely independent of magnetizing frequency [5]. Our
measurements however indicate that there is a degree of dependency on magnetizing
frequency. This may be in part confirmation of the formation of microscopic eddy
currents due to domain wall dynamics at the scale of microstructural features.
To investigate the effect of the data acquisition rate on the BN output the number
of sampling points per cycle was varied from 128 to 2048 for magnetizing frequency of
50 Hz. Figure 3 shows that as the sampling frequency increases the BN RMS value
increases but eventually levels out showing that, at 50 Hz magnetizing frequency,
sampling at 2048 points per cycle is sufficient for picking up most BN events.
160.0E-6
140.0E-6 1.4 T
120.0E-6
1.0 T
100.0E-6
BN RMS [V]
80.0E-6 0.5 T
60.0E-6
40.0E-6
20.0E-6
000.0E+0
0 20 40 60 80 100
Sampling Frequency [kHz]
Fig. 3. Variation of RMS BN signal at 50 Hz magnetization with sampling frequency at

peak flux densities of 0.5 T, 1 T and 1.4 T (Double coil sensor)
3. BN and Hysteresis
The coating applied to the surface of grain oriented electrical steel not only provides
interlaminar insulation in assembled cores but it induces a beneficial stress into the
steel, which reduces losses and stress sensitivity of the steel [6]. The BN before and
after chemical removal of the coating was measured using the ferrite cored sensor at 4
points on the surface of commercial 0.27 mm thick grain oriented 3 % silicon steel at
50 Hz in the range 0.3 T to 1.4 T (The sample was annealed after coating removal). The
measurement points were chosen with the aid of a surface magnetic domain viewer.
Points were chosen over grain boundaries or in the centre of well oriented grains. The
uncertainty of RMS BN signals was up to 5 % over the measurement range. Coating
removal caused domains to become wider and the power loss increased by 4 % at 1.4 T.
No correlation was found between sensor position and BN output. However, the
average RMS BN increased by up to 25 % as can be seen in figure 4.
Removal of the coating and the subsequent annealing has resulted in 180o domain
refinement leading to the observed widening of domain wall spacing. This is well
known for causing increases in losses. Since changes in BN under applied stress are
primarily attributable to changes in 180o domain wall population it is not unexpected to
also see an increase in RMS BN. However, the relatively large change in BN (25 %)
compared with power loss (4%) is not so easily explained and other factors need to be
investigated further before definitive conclusions can be made.
700E-6
600E-6
Decoated
500E-6
BN RMS [V]
400E-6
Coated
300E-6
200E-6
100E-6
000E+0
0.4 0.6 0.8 1.0 1.2 1.4 1.6
B [T]
Fig. 4. Variation of RMS BN with flux density in grain oriented 3 % Silicon steel
before and after coating removal (ferrite sensor of 13 mm length)
Some correlation has previously been found between BN under AC magnetization

and the static hysteresis component of power loss [2]. In the present investigation, the
static hysteresis loss of non-oriented electrical steels with silicon contents between
0.2 % and 6.5 % was estimated. This was achieved by measuring (using the double coil
sensor) the total loss over the frequency range 5 Hz to 200 Hz at 1.5 T and
extrapolating the loss per cycle curve to zero frequency in the conventional way. In all
the materials tested the RMS BN at 50 Hz rises with static hysteresis as shown in figure
5 confirming the general trend found previously occurs in a wider range of materials . It
is recognized today that loss separation in this way is artificial, however the result does
indicate that a close relationship between BN and hysteresis under a.c. magnetization
does exist.
100E-6
90E-6
80E-6
70E-6
BN RMS (V)
60E-6
50E-6
0.23%
40E-6 0.3%
30E-6 1.3%
1.8%
20E-6 3%
6.5%
10E-6
5.5%
000E+0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
3
PH (W/m )
Fig. 5. Variation of RMS BN with calculated static hysteresis loss in non-oriented

electrical steels with different silicon contents at 50 Hz (Double coil sensor)
Interesting variations in BN were observed using a ferrite cored sensor placed on

the surface of individual grains of grain oriented 3% silicon iron. Close agreement was
found between BN measured at different points within a single grain at 50 Hz
magnetization, however a comparison between different grains of similar size and
orientation showed a variation of up to 50% in the BN signal. The RMS BN detected
close to grain boundaries in the same material was 40 % lower than that seen in some
of the well oriented grains. Problems in interpreting such results arise because the BN
noise sensing region is not sufficiently localized. Also the BN is probably strongly
related to the domain wall activity at a particular location which itself is known to vary
significantly from grain to grain. Therefore, future work should attempt to correlate the
two phenomena.
4. Conclusion
Similar trends in the variation of BN with flux density and magnetization frequency in
electrical steels can be detected using enwrapping coils or surface probes as detectors.
The method of analysis of a.c. BN signals does not require more than 2048 data capture
points per cycle at 50 Hz. Correlation can be found between changes in the state of
electrical steel such as that brought about by coating removal. Localized variation in
BN has not been definitively quantitatively attributed to structural variation such as
grain boundaries or resulting domain activity and improved sensor design is necessary
to investigate such phenomena.
Acknowledgment
This work was supported by EPSRC Grant EP/C518616. The authors would also like
to thank Stresstech Oy for providing the ferrite cored sensor.
References
[1] D. Jiles, Introduction to Magnetism and Magnetic Materials, Chapman and Hall, NY, 1991
[2] A. J. Moses, F. J. G. Landgraf, K. Hartmann and T. Yonamine, Correlation between angular
dependence of A.C. Barkhausen noise and hysteresis loss in a non-oriented electrical steel,
Stahleisen (2004) 215 – 219
[3] H. V. Patel, S. Zurek, T. Meydan, D. C. Jiles, L. Li, A new adaptive automated feedback system for
Barkhausen signal measurement, Journal of Sensors and Actuators A: Physical, Volume 129, Issues 1-2,
Pages 112-117, 24 May 2006
[4] A. J. Moses, P.I. Williams and O. Hoshtanar, A novel instrument for real time dynamic domain
observation in bulk and micromagnetic materials, IEEE Trans. Mag. 41 (10), (2005) 3736-3738
[5] G. Bertotti, Hysteresis in magnetism: for physicists, materials scientists and engineers, London:
Academic, 1998
[6] A. J. Moses, Electrical Steels – Past, present and future developments, IEE Proc. 137 Pt.A (5) (1990)
2333-2345
IOS Press, 2007
Magnetic Adaptive Testing: Influence of

Experimental Conditions
Gábor VÉRTESY a,1, Tetsuya UCHIMOTO b, Toshiyuki TAKAGI b and Ivan TOMÁŠ c
a
Research Institute for Technical Physics and Materials Science, Budapest, Hungary
b
Institute of Fluid Science, Tohoku University, Sendai, Japan
c
Institute of Physics, Academy of Sciences of the Czech Republic, Praha, Czech Republic
Abstract. Influence of the magnetizing yoke on the calculated magnetic

descriptors of Magnetic Adaptive Testing was studied by measuring gray cast iron
samples. It was found that the material and shape of the yoke has no influence on
the relative values of the descriptors.
Keywords. Magnetic Adaptive Testing, Magnetic hysteresis measurements
1. Introduction
Magnetic measurements are frequently used for characterization of changes in the

structure of ferromagnetic materials, because magnetization processes are closely
related to the microstructure of the materials. This fact also makes magnetic
measurements an obvious candidate for non-destructive testing, for detection and
characterization of any defects in materials and manufactured products made of such
materials [1,2,3]. Structural non-magnetic properties of ferromagnetic materials have
been non-destructively tested using traditional magnetic hysteresis measurement
methods for a long time with fair success. A number of techniques have been suggested,
developed and currently used in industry, see e.g. [4,5]. They are mostly based on
detection of structural variations via the classical macroscopic parameters of hysteresis
loops.
An alternative, more sensitive and more experimentally friendly approach to this
topic was considered recently in [6] and [7], based on magnetic minor loops
measurement. In [6] the method of Magnetic Adaptive Testing (MAT) was presented,
which introduced general magnetic descriptors to diverse variations in non-magnetic
properties of ferromagnetic materials, optimally adapted to the just investigated
property and material. In [7] the sets of minor hysteresis loops were scrutinized, and
sensitive descriptors of plastic deformation of the material (independent on the minor
loops amplitudes) were identified.
In this work MAT is applied on gray cast iron samples. Influence of the
experimental conditions, mainly the application of different magnetizing yokes on the
1
Corresponding Author: Gábor Vértesy, Hungarian Academy of Sciences, Research Institute for Technical
Physics and Materials Science, H-1525 Budapest, P.O.Box 49, Hungary, E-mail: vertesyg@mfa.kfki.hu
G. Vértesy et al. / Magnetic Adaptive Testing: Influence of Experimental Conditions 187
evaluated magnetic descriptors is studied.
2. Experiments
Three series (FC150, FC200 and FC250) of gray cast iron specimens were investigated.
Chemical compositions of the series were different. Table 1 shows the chemical
composition of the samples. In each series one sample was in as-cast condition, one
sample was heated up again and cooled in air and one was heated up and cooled in a
furnace. Hardness of the samples (HB) was measured by the Rockwell hardness test
method. Results can be seen in Table 2.
Table 1 Chemical composition of the investigated specimens
Sample C% Si% Mn% P% S% Cr% Ti%

FC150 3.77 2.78 0.78 0.025 0.015 0.029 0.015
FC200 3.36 2.15 0.69 0.018 0.010 0.014 0.011
FC250 3.13 1.66 0.72 0.017 0.002 0.038 0.010
Table 2 Hardness of the investigated specimens
Hardness HB
as Cast furnace cooling air cooling
FC150 89 89 135
FC200 178 101 206
FC250 206 126 212
A specially designed Permeameter [8] with a magnetizing yoke was used for
measurement of families of minor loops differential permeability of the magnetic
circuit. The block-scheme of the device and the sketch of the yoke with a sample can
be seen in Figure 1. Two different yokes were used for the measurements. The ferrite
yoke, made of M2TN-B type soft ferrite material, was 16 mm long, 11 mm high, and
the cross section of its legs was 5x6 mm. The iron yoke was made of a C-shaped
laminated Fe-Si core and the corresponding size was 27, 26 and 8x10 mm, respectively.
The magnetizing coil wound on the ferrite yoke gets a triangular waveform current
with step-wise increasing amplitudes and with a fixed slope magnitude in all the
triangles. This produces time-variation of the effective field, ha(t), in the magnetizing
circuit and a signal is induced in the pick-up coil. As long as ha(t) sweeps linearly with
time, the voltage signal U(ha,hb), in the pick-up coil is proportional to the differential
permeability, P(ha,hb), of the magnetic circuit
P (ha , hb ) const * U (ha , hb ) const * wB(ha , hb ) / wha * wha / wt
The Permeameter works under full control of a PC computer, which sends appropriate
control information to the function generator, and collects the measured data. An input-
188 G. Vértesy et al. / Magnetic Adaptive Testing: Influence of Experimental Conditions
output data acquisition card accomplishes the measurement. The computer registers
data-files for each measured family of the minor “permeability” loops, corresponding
to each measured sample. They contain detailed information about all the pre-selected
parameters of the voltage signal induced in the pick-up coil. The step of the magnetic
circuit effective field amplitudes was 'hb=36 A/m, and the magnetizing current rate of
change was r5 A/s.
Figure 1. Block-scheme of the Permeameter and sketch of the yoke
The described regime of the Permeameter yields a characteristic signal in the pick-
up coil; typical examples of its shape are shown in Figure 2. The signal values start at
the origin of the plot (the magnetic circuit was demagnetized before the measurement),
then increases into positive values (up to the positive starting field amplitude, +1*'hb).
Then it drops down into negative values as the applied field changes the direction of its
rate, then proceeds in the negative values until the negative starting field amplitude -
1*'hb is reached, changes its rate direction and polarity again, raises up to +2*'hb, etc.
Pick-up coil signal [arb.units]
Pick-up coil signal [arb.units]
0,2 0,2
(a) Iron yoke (b)

Ferrite yoke
0,1 0,1
0,0 0,0
-0,1 -0,1
-0,2 -0,2
-0,75 -0,50 -0,25 0,00 0,25 0,50 0,75 -0,75 -0,50 -0,25 0,00 0,25 0,50 0,75
Magnetizing current, I [A] Magnetizing current, I [A]
Figure 2. The typical signal registered for one family of the triangular variations of the magnetizing
current, obtained on the same sample by using ferrite(a) and iron (b) yokes
The experimental raw data are processed by a data-evaluation program, which

divides the originally continuous data of each measured sample into a family of
individual permeability half-loops. Then the family, either of the top half-loops or the
bottom half-loops or their average is chosen for the next processing stage. The program
filters experimental noise and interpolates the experimental data into a regular square
grid of elements, Pij { P(hai,hbj), of a P-matrix with a pre-selected field-step. The
consecutive series of P-matrices, each taken for one sample with a value of the strain, H,
of the consecutive series of the more-and-more deformed material, describes the
magnetic reflection of the material plastic deformation.
The matrices are processed by a matrix-evaluation program, which normalizes

them by a chosen reference matrix, and arranges all the mutually corresponding
elements Pij of all the evaluated P-matrices into a Pij(HB) table. Each Pij(HB)-column of
the table numerically represents one Pij(HB) matrix element. These matrix elements are
used for high sensitive characterization of the investigated material.
The scheme used for optimization of parameters obtained from the evaluated
matrices is discussed detailed in [9]. This reference describes how the most sensitive
and at the same time the most reliable descriptor (matrix element) can be picked up
from the big data pool. The optimal Pij(HB)-descriptor of the investigated samples were
determined according to this procedure.
3. Results
The signals induced in the pick-up coils, measured on one of samples (the same piece;
sample FC200, air cooled) are shown in Figure 2, for the ferrite and for the iron yoke.
It is clearly seen that the recorded signals are very different. This reflects the different
behaviour of the whole magnetic circuit (yoke, air gap, sample).
1,8
ferrite yoke
iron yoke
P m atrix elements
1,6
1,4
furnace cooling
1,2
1,0 as cast
air cooling
0,8
80 100 120 140 160 180 200 220
Hardness
Figure 3. Calculated optimal Pij(HB) -matrix elements for the ferrite and for the iron yokes as functions
of hardness (sample FC200)
The above detailed calculation was carried out on all the investigated samples. The
obtained optimal Pij(HB)-descriptors (P-matrix elements) are shown for the sample
series FC200 in Figure 3, for sample series FC250 in Figure 4 and for sample series
FC150 in Figure 5, respectively. In all cases the measurement and the corresponding
evaluation were made for both types of yokes.
1,7
ferrite yoke
1,6 iron yoke
P matrix elements
1,5
furnace cooling
1,4
1,3
as cast
1,2
1,1
1,0
air cooling
0,9
120 140 160 180 200 220
Hardness
1,7
1,6 furnace cooling ferrite yoke

iron yoke
1,5
P matrix elements
1,4
1,3
1,2 as cast
1,1 air cooling
1,0
90 100 110 120 130 140

Hardness
4. Discussion
Magnetic measurements were carried out with the aid of magnetic yokes. The
consecutive series of P-matrices describe the magnetic reflection of the material
modification very well. As it was proved in previous works [6,8,9], by applying the
Magnetic Adaptive Testing method, the relatively small difference between the
magnetic characteristics of the investigated sample series can be determined much
more sensitively, than by the conventional methods.
Another advantageous and independent outcome of the tested method is the

confirmation, that without magnetic saturation of the samples, measuring a series of
minor loops and performing MAT method on the obtained data-pool, reliable and
sensitive parameters can be determined. Moreover, the relative measurement can be
done with a ferromagnetic yoke attached to the sample, and the yoke does not even
have to be large and very special. The method does not give absolute values of the
traditional magnetic quantities, because of the non-uniform magnetic circuit and of the
not-reached magnetic saturation, but evidently it is able to serve as a powerful tool for
comparative measurements, and for detection of changes, which occur in structure of
the inspected samples during their lifetime or during a period of their heavy-duty
service.
Even if quality of the magnetic contact between the samples and the yoke was
assumed to be stable, the magnetic circuits were certainly non-uniform and the
magnetic values obtained from each measurement were rather effective magnetic
parameters of the circuit than real magnetic parameters of the samples.
As can be seen in Figures 3-5, independently of the applied yoke very similar
dependences were found between hardness and the magnetic descriptors. This fact
shows that by application of MAT, the samples’ characteristics are determined, and the
remaining part of the magnetic circuit – if kept constant at each measurement – has no
basic influence on the relative value of the calculated quantities. The feature of the
magnetic circuit has an influence on the absolute value of the measured signal, but
considering that the whole measurement is relative, and only the results, measured at
the same conditions are compared, this fact has no influence on applicability of the
method.
In certain cases the measurements, performed by the ferrite yoke give better
sensitivity, in other case iron yoke does the same. This feature of the measurements is
not understood yet, further study is necessary in this field. However, the difference is
not significant, so it can be concluded that any of the investigated yoke is suitable for
sensitive and reliable measurements. In case of sample FC150 the as cast and furnace
cooled samples have the same value of hardness, in spite of the fact that – according to
our measurements – they are different magnetically. This fact needs further study, the
possible error of hardness measurements cannot be excluded, too. But it has no
influence for the main message of the present work.
The most important difference between ferrite and iron yokes is the saturation
magnetization of their material. Ferrite yoke is saturated at about 300 mT, while the
saturation magnetization of iron yoke is much higher. It means that the observed virtual
saturation, which is suggested by Figure 2/a is the magnetic saturation of the yoke, and
not the whole magnetic circuit.
It should be emphasized that these descriptors show the relative modification of

the material behaviour within the same series of the measured samples. These
modifications in the present series of samples are connected with the different way of
sample preparation. Due to different cooling rate, magnetic parameters became
different, as seen clearly in the figures. The interpretation of the changes of magnetic
parameters as a function of sample preparation (structure variation of the cast iron) is
not the subject of the present paper, it will be discussed elsewhere.
From the same calculated data pool the inverse values of matrix elements can be
picked up, as well, which results just the opposite relationship between the independent
parameter (hardness) and P-matrix elements, but the influence (or better to say the lack
of influence) of the experimental condition on the MAT parameters remain the same.
5. Conclusions
The influence of the applied magnetizing yoke was investigated on the nondestructive
magnetic characterization of gray cast iron samples when Magnetic Adaptive Testing
method was used. It was found that exactly the same relationship with very similar
sensitivity was obtained, regardless of the actual type of yoke.
The results obtained are considered to be very important, because if different

relative dependencies within the same sample series would have been experienced by
application of different yokes, the reliability and applicability of the whole MAT
became questionable. In other words: the obtained relationship is not influenced by the
experimental conditions (how the samples are magnetized), it reflects only nature of the
measured samples.
Acknowledgments
The financial support by the Hungarian Scientific Research Fund (K-62466), by the
Research and Technology Innovation Fund of the Hungarian Government in the frame
of the Japanese-Hungarian Bilateral Intergovernmental S&T Cooperation (JAP 17/02),
by the Academy of Science of the Czech Republic (project No.1QS100100508) and by
the Grant Agency of the Czech Republic (project No.102/06/0866) is appreciated.
References
[1] M.J. Johnson, C.C.H.Lo, B.Zhu, H.Cao, D.C.Jiles, J. Nondestruct. Eval. 20 (2000), 11.
[2] D.C. Jiles, Magnetic methods in nondestructive testing, K.H.J.Buschow et al., Ed., Encyclopedia of
Materials Science and Technology, Elsevier Press, Oxford, (2001), 6021.
[3] I. Mészáros, Materials Science Forum, 473-474 (2005), 231-236.
[4] D.C. Jiles, NDT International, 21 (1988), 311-319.
[5] J. Blitz, Electrical and magnetic methods of nondestructive testing, Adam Hilger IOP Publishing, Ltd.,
Bristol, 1991.
[6] I.Tomáš, J.Magn.Magn. Mat., 268 (2004), 178-185.
[7] S.Takahashi, L.Zhang, T.Ueda, J.Phys.: Condens.Matter, 15 (2003), 7997-8002.
[8] I. Tomáš, O.Perevertov., in: JSAEM Studies in Applied Electromagnetics and Mechanics 9, ed. T.
Takagi and M. Ueasaka, (IOS Press, Amsterdam, 2001), 5-15.
[9] G. Vértesy, I. Tomáš, I. Mészáros, J. Magn. Magn. Mat., 310 (2007), 76-82.
IOS Press, 2007
Magnetic and Acoustic Barkhausen Noise

for the Characterisation of Tensile
Deformation and Stresses in Steel
Gui Yun TIAN a, 1, John WILSON a, Jiri KEPRT b
a
Newcastle University, UK
b
Brno University of Technology, Kolejni 2906, Brno 612 00, Czech Republic
Abstract. This paper introduces a new method of pulsed acoustic Barkhausen

noise for stress detection and material characterisation. Pulsed excitation can
provide wide spectral components and time information for non-destructive
evaluation and material characterisation. After introducing the system set-up,
several experiments including stress variation and residual stress orientation are
outlined, in comparison with magnetic field intensity measurement. Different
Barkhausen noise and magnetic acoustic emission characterisation methods such
as Root Mean Square (RMS), absolute energy and event count are discussed and a
new quantitative method for pulsed acoustic Barkausen noise and magnetic flux
leakage is investigated for stress detection and material characterisation.
Keywords. Acoustic emission, permeability, Acoustic Barkhausen noise (ABN),

Magnetic Barkhausen noise (MBN), NDT&E, pulsed excitation
1. Introduction
Non-Destructive Evaluation (NDE) is used to assess the integrity of a system or

component without compromising its performance. Detailed defect sizing and
characterisation has become the major objective of much NDE work underway today.
To address this challenge, the NDE community has turned to novel techniques and a
combination of multiple mode inspections and computer-aided data analyses. In many
of the non-destructive techniques used for material evaluation, such as acoustics,
magnetic particle, eddy current, magnetic flux leakage methods or their integration [1-
3], surface and sub-surface flaws in the material are detected by measuring the induced
or leakage fields as a result of the interaction of an excitation electromagnetic/magnetic
field with the flaws.
An electromagnetic pulse contains many frequency components, so pulse sources
typically produce significantly more peak power than single frequency sources using a
simple control circuit. The advantage of a pulsed excitation source is that it can provide
time information along with rich frequency components, which can deliver depth
information for defects and materials based on skin effects. Pulse techniques have been
used for pulsed eddy current and pulsed magnetic flux leakage for non-destructive
1
Corresponding Author: G Y Tian, School of Electrical, Electronic and Computer Engineering,
Newcastle University, UK, NE1 7RU Email: g.y.tian@ncl.ac.uk
194 G.Y. Tian et al. / Magnetic and Acoustic Barkhausen Noise
testing and evaluation [2, 3] with the help of advanced signal processing.
In a ferromagnetic material, Barkhausen noise is generated by the discontinuous
movement of irreversible domains walls. This movement can be induced by applying a
time varying magnetic field across the sample. This noise can be detected in the form
of acoustic noise (electrical energy) known as acoustic Barkhausen noise (ABN) or
magneto-acoustic emission (MAE) and in the form of voltage pulses which are induced
in a coil placed near the surface of the material, called magnetic Barkhausen noise
(MBN). MBN results from the reversible and irreversible displacement of 180o and
non-180o domain walls, or by abrupt rotation of domain magnetisation vectors at higher
magnetic fields; ABN is only caused by the discontinuous motion of non-180o walls, or
the irreversible rotation of domains through angles other than 180o. ABN and MBN are
highly correlated [4]. Therefore, MBN is a useful NDE technique and many researchers
have studied MBN for iron-based materials [4,5]. The mechanism of MBN, however,
has not been clearly understood, especially the relationship between MBN and
micromagnetic states such as grain size, residual stress, dislocations, voids, etc.
2. Probe Design and Characterisation of Signals
Figure 1 shows the design of the dual acoustic emission / magnetic permeability probe.
The probe consists of a ferrite core wound with a 200 turn excitation coil supplied with
current by a power amplifier supplied with a 10Hz waveform, wound onto one leg of
the ferrite core is a 100 turn pick-up coil. 45mm to the left of the core is a piezoelectric
ABN sensor; this has a 1kg weight applied to it during the tests and petroleum jelly is
used to insure effective acoustic coupling between the sensor and the material surface.
Figure 1. System design and associated signals

The ABN sensor and the pickup coil are interfaced to a data acquisition card via
signal processing electronics. The ABN sensor used in the tests is a R15I-AST from
Physical Acoustics Corporation (PAC); the sensor is a resonant type with a frequency
range of 50 kHz – 200 kHz. This frequency range was deemed acceptable by previous
testing with a broadband sensor. Data acquisition was via an Adlink 2010 4-channel
data acquisition card, using a sampling frequency of 2MHz. Data recording was
performed in LabView and data processing in Matlab.
G.Y. Tian et al. / Magnetic and Acoustic Barkhausen Noise 195
(a) (b)
Figure 2. Pulsed excitation current and ABN signal (a) and excitation current and integrated pickup coil
voltage (b)
Figure 2 shows some typical signals from the ABN sensor and the pickup coil,
along with the current applied to the excitation coil using pulsed excitation. Figure 2a
shows the excitation current along with the signal from the ABN sensor. It can be seen
from the plot that the ABN sensor output is greatest where the excitation current is
exhibiting its greatest rate of change and the minimum ABN amplitude comes where
the gradient of the excitation signal is at a minimum.
The voltage drop over the pickup coil indicates the rate of change of the field
through the coil with respect to time, and is given by:
§ dB ·
VCOIL kn¨ ¸ (1)
© dt ¹
Where VCOIL is the potential drop over the pickup coil, B is the magnetic field through
the coil, n is the number of turns in the coil and k is a constant associated with the coil
characteristics. So integration of the coil voltage gives a signal proportional to the field
through the coil. The coil voltage integral is shown in figure 2b, where the difference in
the form of the two curves shows the effect of the hysteresis of the material and the
high level of magnetism where the excitation reaches zero indicates the remnant
magnetisation of the material.
As the coil voltage indicates the field intensity round magnetic circuit formed by
the core and the material under test, measurement of the coil voltage can be used to
indicate variations in the magnetic permeability of the material. It has been shown
previously that variations in magnetic permeability can be used to indicate both applied
and residual stresses in ferromagnetic materials [6].
Once the ABN signal exceeds a certain level (the threshold), it is classed as an
acoustic event. Each repetition of the excitation waveform evokes hundreds of these
events, each of which corresponds to “noise” picked up by the transducer. Several
parameters can be used to quantify each of these events, including area – this is a
measurement of the area contained between the peaks of the event and the threshold
and is also referred to as signal energy and the event count – the number of times that
the signal rises and crosses the threshold during each event. The overall RMS
amplitude is also used to quantify the signal. A single Matlab program was developed
for the work to process the signals acquired during the tests and output several
parameters, including peak to peak and RMS values of the coil voltage, pickup coil
integral and RMS voltage and signal energy and event count for the ABN signal.
3. Active Stress Measurement
3.1. Barkhausen Acoustic Emission Analysis
The test shown in figure 3a was set up to study the effect on the acoustic emission
signal and coil voltage of the application of stress. A steel sample measuring 230mm x
30mm x 2mm was placed in a Hounsfield material test machine with the test system
shown in figure 1 attached. The force applied to the sample was increased from 0kN to
16kN in 1kN steps, corresponding to a maximum stress of 267MPa and a maximum
nominal strain of 8.5%, well into the plastic region for the material. After each 1kN
step, both sine wave and pulsed excitation were applied to the sample and the
corresponding ABN signals and coil voltages recorded.
(a) (b)
Figure 3. Active stress test set-up (a) and stress/strain plot for the material (b)
Before the test was carried out, the stress / strain curve for the material (figure 3b)
was calculated using an identical test sample, cut from the same piece of steel. The
stress / strain curve shows that the yield point for the material is around 150MPa of
stress at 0.8% strain. So stresses below this point will be referred to as elastic and
stresses above this point will be referred to as plastic.
Figures 4a and 4b shows plots of the event count (the number of times the ABN
signal crosses a particular threshold) using sine and pulsed excitation for applied stress.
It can be seen from the plots that the event count is most sensitive to stresses in the
elastic region, below 150MPa. As applied stress approaches the yield point, the event
count for both sine and pulsed excitation starts to level out, but the pulsed excitation
exhibits the most linear correlation with applied stress up the 100 MPa point.
Figures 4c and 4d show plots of the integral of the pickup coil voltage in volts x
seconds (VS) for sine and pulsed excitation divided by the RMS excitation power
(watts) for the two excitation waveforms to show the sensitivity of the two methods
with respect to excitation power. It can be seen from the plots that the permeability of
the material is most sensitive to stresses in the plastic region, giving a fairly linear
correlation with measured strain, especially above the yield point of 0.8%.
(a) (b)
(c) (d)
Figure 4. Event count for applied stress for sine excitation (a) and pulsed excitation (b), Peak to peak integral
of coil voltage / excitation power (watts) plotted against % nominal strain for pulsed (a) and sine (b)
excitation
The sensitivity of the integral pick-up coil signal (VS) with respect to the extension
of the material under test (%strain) and the power supplied to the excitation coil
(Watts) was calculated, for pulsed and sine excitation signals. The result of the
calculation yielded 494 u 10 6 VS / Watt / % Strain for pulsed excitation and
6
164 u 10 VS / Watt / % Strain for sine excitation; three times greater sensitivity for pulsed
excitation, giving a considerable power saving for the pulsed excitation for comparable
results.
3.2. Barkhausen Noise
The test outlined in section 3.1 was repeated using an identical sample cut from the
same sheet of steel. The ABN sensor was replaced by a GMR sensor positioned
equidistant between the feet of the ferrite core (figure 3a). The signal from the sensor
was sampled at a frequency 2MHz and 0.3 kHz – 38 kHz band-pass filtered using a
Matlab program to extract the Barkhausen noise from the sensor signal. The RMS
value for each reading was then computed.
Figure 5a shows a typical filtered signal from the GMR sensor. It can be seen from
the plot that, as with the acoustic emission signal, the greatest amplitude of Barkhausen
noise comes where the excitation signal is exhibiting its greatest rate of change. Figure
5b shows the RMS amplitude of the Barkhausen noise signal using sine excitation.
Examination of figure shows that after an initial increase in amplitude, the MBN signal
decreases all the way to the end of the test, rapidly in the elastic region and then less
steeply in the plastic region.
(a) (b)
Figure 5. a) Barkhausen noise signal with excitation current, b) RMS Barkhausen noise amplitude for
applied nominal strain
3.3. Discussion
Stress dependence of domain wall activity is complex and heavily influenced by

microstructural factors such as material phases and grain size. This is especially true for
MBN measurement which has been shown to increase or decrease in response to
applied tensile stress dependant the microstructure of the material under inspection,
with MBN activity increasing with applied tensile stress for martensitic structures and
decreasing with applied tensile stress in cementite structures [7]. In contrast ABN is in
general less sensitive to microstructure, with the application of tensile stress to steel
almost always causing a decrease in ABN activity.
The results from both the magneto-acoustic emission and the Barkhausen noise
tests confirm that for this particular material under these test conditions, Barkhausen
noise testing for active stress detection is most suited to the measurement of stresses in
the elastic region, with both ABN and MBN showing greater variation in response to
elastic stresses. The differences in sensitivity in the elastic and plastic regions are due
to the fact that the mechanisms which give rise to domain wall activity are different in
each region. So changes due to elastic and plastic stresses must be considered as
separate phenomena.
In the elastic region, deformation causes changes in the inter-atomic spacing in the
material and corresponding changes in domain configuration. This has a direct effect
on magneto-elastic energy and thus the Barkhausen activity [8]. Once the plastic region
is reached, the macroscopic elastic strain remains fairly constant and other mechanisms
take over, the most significant of these being the introduction of dislocations to the
material acting as pinning sites to domain wall motion. In the plastic deformation
region, at each stain value, the substructure changes in addition to the change in stress.
The influence of substructure dominates, which reduces the mean free path and hence
the overall emission is reduced and the correlation is not held. In the elastic region, the
easy magnetisation direction rotates towards the tensile axis without any structural
changes and hence a good correlation with applied stress is maintained.
Another factor which influences domain wall motion is the intensity of the applied
magnetic field. An increase in the maximum applied magnetic field has been shown to
invoke greater ABN and MBN activity as well as a shift of the maximum ABN and
MBN activity to a higher stress level [5], thus changing the sensitivities of the
techniques to different sections of the stress strain curve. In this case maximum
sensitivity is apparent in the elastic region. This effect is thought to be due in some part
to the application of fixed maximum excitation amplitude in the tests [4]. Although the
applied field is initially sufficient to drive the material into saturation, as pinning sites
increase in number the mean free path for pinning sites reduces [9] and the energy
required to overcome these barriers to domain wall movement increases, thus
modifying the hysteresis curve for the material and increasing the level of applied
magnetic field required to drive the material into saturation. So although the potential
Barkhausen energy and therefore the sensitivity of Barkhausen techniques to applied
stress is higher in the plastic region, through greater release of energy from domains
overcoming pinning sites, the actual measured Barkhausen noise reduces because not
as many domains are able to overcome pinning sites with the supplied excitation field.
4. Determination of Residual Stress Orientation in Pre-stressed Steel Sample
The test shown in figure 6a was set up to investigate the correlation between residual
stress orientation, permeability and ABN in a steel sample previously exposed to
plastic tensile stress. A steel sample which had been exposed to 8.5% nominal strain in
the tests outlined in the previous section was used, along with a miniature version of
the developed probe. After the load had been removed, readings from the ABN sensor
and the signal from the pickup coil were taken at 22.5° increments as the ferrite core
was rotated by 360°.
(a)
(b) (c)
Figure 6. a) Pre-stressed steel test set up, b) Change in peak to peak coil voltage, c) Change in ABN event
count in 22.5° increments around a 360° scan for pre-stressed steel sample.
Figures 6b and 6c show the change in peak to peak coil voltage and ABN event
count around the full 360° scan of the material, with the residual stress orientated
horizontally. It can be seen from figure 6b that the coil voltage is at a maximum at 90°
with respect to residual stress and from figure 6c that the event count is greatest parallel
to residual stress.
Plastic deformation causes dislocations, leading to a complete reorganisation of

the material domain structure and increased magnetic anisotropy, with the easy
magnetic direction of domain walls rotating towards the tensile stress direction leading
to increased permeability parallel to tensile stress and at 90° to compressive stress. It
has been shown in previous studies that the application of plastic tensile stress to a
material usually results in the generation of compressive residual stress on release of
applied stress [1]. This corresponds to figure 6c, where the signal from the pickup coil
indicates increased permeability at 90° to the previously applied stress, thus indicating
the presence of compressive residual stress. The two sets of results are complementary,
in that the maximum displacement of non-180° domain walls will take place at 90° to
the easy magnetisation direction.
5. Conclusions
The work shows that the pulsed ABN system can offer substantial improvements to
traditional single frequency ABN systems where depth information is of primary
concern. The nature of the pulsed signal also means that magnetic saturation can be
achieved with a much lower RMS current than with single frequency excitation, so the
system could offer improvement where power consumption is of importance, i.e.
portable test devices. Fusion of the signals from the ABN sensor and the pickup coil
could provide data for stress measurement from the elastic region right up to fracture,
with the ABN sensor providing data in the elastic region and the pickup coil in the
plastic region.
Acknowledgments - The authors would like to thank EPSRC for funding the research
and Dr T. Jayakumar at IGCAR, India for some useful comments and suggestion.
References
[1] A. Dhar, L. Clapham and D. L. Atherton, Influence of uniaxial plastic deformation on magnetic
Barkhausen noise in steel, NDT and E Int., 34(8), Dec. 2001, pp. 507-514.
[2] A Sophian, G Y Tian, S Zairi, Pulsed Magnetic Flux Leakage probe for crack detection and
characterisation, Sens. Actuators, A, 125(2), 10 Jan. 2006, pp. 186-191.
[3] A Sophian, R S Edwards, G Y Tian and S Dixon, Dual-probe methods using pulsed eddy currents and
electromagnetic acoustic transducers for NDT inspection, INSIGHT, 47(6), Jun. 2005, pp. 341-345.
[4] D O'Sullivan, M Cotterell, D A Tanner and I Mészáros, Characterisation of ferritic stainless steel by
Barkhausen techniques, NDT and E Int., 37(6), Sep. 2004, pp. 489-496.
[5] B Augustyniak, Correlation between acoustic emission and magnetic and mechanical Barkhausen
effects, J. Magn. Magn. Mater., 196-197, 1999, pp. 799-801.
[6] J M Makar and B K Tanner, The effect of plastic deformation and residual stress on the permeability
and magnetostriction of steels, J. Magn. Magn. Mater., 222(3), Dec. 2000, pp. 291-304.
[7] D J Buttle, C B Scruby, G A D Briggs, J P Jakubovics, The Measurement of Stress in Steels of Varying
Microstructure by Magnetoacoustic and Barkhausen Emission, Proc. R. Soc. London, Ser. A, 414(1847),
Dec. 1987, pp. 469-497.
[8] C G Stefanita, D L Atherton and L Clapham, Plastic versus elastic deformation effects on magnetic
Barkhausen noise in steel, Acta Mater., 48(13), Aug. 2000, pp. 3545-3551.
[9] S Vaidyanathan, V Moorthy, P Kalyanasundaram, T Jayakumar, and B Raj , Effect of Different Stages
of Tensile Deformation on Micromagnetic Parameters in High-Strength, Low-Alloy Steel, Metall.
Mater. Trans. A, 30A, Aug. 1999, pp. 2067-2072.
IOS Press, 2007
Electromagnetic Acoustic Resonance to

Assess Creep Damage in a Martensitic
Stainless Steel
Toshihiro OHTANI
Materials Lab., Ebara Research Co. LTD., Japan
Abstract. The microstructure evolution of a martensitic stainless steel, JIS-

SUS403, subjected to tensile creep at 873K has been studies by monitoring of
shear-wave attenuation and velocity using electromagnetic acoustic resonance
(EMAR). The study revealed an attenuation peak at around 20 % of and a
minimum value at 50 % of the creep life, independent of the applied stress. This
novel phenomenon is interpreted as a result of microstructural changes, especially
dislocation structure. This interpretation is supported by TEM observations of
dislocation structure. The relationship between attenuation change and
microstructure evolution can be explained with the string model. The study results
have suggested that EMAR possesses the potential to assess the progress of creep
damage and predict the remaining creep life of metals.
Keywords. Creep damage, Ultrasonic attenuation, Martensitic stainless steel, Non-

contacting evaluation, dislocation.
1. Introduction
Structural metals are subject to aging from fatigue, creep, corrosion, and their
combination. Exposure to elevated temperatures promotes creep. Aged metals lose
toughness, or the ability to absorb energy for stresses above the yield point. They can
not endure the occasional high load without fracturing. In-service degradation by creep
is one of the most critical factors determining the structural integrity of the elevated-
temperature components in power plants, chemical plants and oil refineries the world
over [1]. For instance, fossil-plants have been in operation for such long durations that
the critical components have exceeded the design life of 30 to 40 years. They have
undergone progressive damage as time has proceeded. In order to save energy and meet
recent regulatory standards for CO2 emission, as well as to improve thermal efficiency,
the steam pressures and operating temperatures in the components have been increased.
As a consequence, material degradation has been accelerating. Economic and
environmental concerns that prohibit the construction of new plants increase the
severity of this problem. Therefore, in-service assessment of the state of damage is
important for ensuring safe operation, predicting remaining life, and promoting life-
extension programs [2, 3]. For this assessment, a non-destructive technology enabling
the evaluation of the current state of materials and the prediction of their remaining life
Corresponding Author: Toshihiro Ohtani, Ebara Research Co. LTD., 4-2-1 Hon-Fujisawa, Fujisawa, 251-
8502, Japan; E-mail:ohtani.toshihiro@er.ebara.com
202 T. Ohtani / Electromagnetic Acoustic Resonance to Assess Creep Damage
has long been sought [1]. It is essential that the technique should provide simple and
quick measurement to cope with a large number of objects and give accurate
information about microstructural change.
This study describes the changes in ultrasonic attenuation and velocity and
microstructural evolution during creep tests on a martensitic stainless steel (JIS-
SUS403) as observed using EMAR (electromagnetic acoustic resonance) [4]. This steel
is widely used as a high temperature material in power plants and chemical plants. It
has high strength at elevated temperatures (up to 923 K) and also provides thermal
expansion and good resistance to corrosion [5]. EMAR is a combination of the resonant
technique and a non-contacting electromagnetic acoustic transducer (EMAT) [4].
Incorporation of EMATs in resonant measurement greatly contributes to the
improvement of the weak coupling efficiency of EMATs. The attenuation measurement
is inherently free from the energy loss associated with contact transduction, thus
providing a pure measurement of attenuation in a metal sample.
2. Sample and EMAT
The material was from a commercial plate of martensitic stainless steel, JIS-SUS403. It
was heated at 1253 K for 2 h, water-quenched, heated at 1023 K for 2 h, and tempered.
A specimen for the creep test was machined; the gauge sections appeared as plate
shapes of 5 mm thickness, 18 mm width and 35 mm length [6]. The mechanical
properties of the plates at room temperature were as follows: 0.2% proof stress at 562
MPa, an ultimate tensile strength at 720 MPa and elongation of 25 %. The chemical
composition is given in Table 1.
A shear-wave EMAT with a 10 mm x 10 mm active area was used. It was
comprised of an elongated spiral coil and a pair of permanent magnets in directions
opposite to the specimen surface. It generated and received the shear waves with the
magnetostrictive effect of ferromagnetic materials in a non-contacting manner [4]. The
polarized shear waves traveled back and forth in the thickness direction. For more
details, see Refs [7, 8].
Resonant frequencies and attenuation coefficients for thickness resonant modes
were measured with a superheterodyne spectrometer [4]. The resonant peaks appear at
equal frequency intervals and at each resonant frequency the attenuation coefficient
was measured by the free-decay method, leading to the frequency dependence of
attenuation [4].
3. Experiments
Creep tests were carried out at 873 K in air for three stresses of 120, 140 and 160 MPa
with lever-type equipment. During the creep test, creep loading was interrupted and
the samples were cooled in the furnace. After measuring ultrasonic properties (the
attenuation coefficients and velocity), the creep test was restarted. This procedure was
repeated every 20, or 50 h until rupture. A series of crept samples with different strains
was obtained. To distinguish the effects of thermal history alone, the unstressed
samples along side the stressed samples was examined.
T. Ohtani / Electromagnetic Acoustic Resonance to Assess Creep Damage 203
Table 1. Chemical composition of SUS403 (wt. %).
C Si Mn P S Ni Cr Fe
0.120 0.300 0.40 0.033 0.018 0.200 11.67 Bal
To investigate the relationship between the microstructure and ultrasonic responses

through the creep life, the microstructures of specimens were observed with an optical
microscope (OM), a scanning electron microscope (SEM) and a transmission electron
microscope (TEM). To obtain the microstructure observation samples, the modified Tҏҏ
projection [9] and a rupture parameter, PD [10] were applied. The sample preparation
procedure is shown in detail elsewhere [6].These micrographs were scanned into the
computer for further analysis.
4. Results
The resonant frequencies in the 1 to 8 MHz range and their attenuations during the
creep test were measured. Figure 1(a) shows the typical relationship between the
attenuation coefficientҏ D, velocity change, 'V/Vi (Vi: initial velocity), creep strain, and
life fraction t/tr (creep time/rupture life) as creep advances. The rupture life was 990.6 h.
The attenuation coefficient increases, showing a peak at t/tr =0.2, then decreases,
showing a minimum near t/tr =0.5, and finally increases until rupture. The velocity
gradually decreased and showed a local minimum at the attenuation peak, then slightly
increased until the attenuation minimum, and finally quickly decreased to the rupture.
The total decrease in velocity was about 0.3 %. The creep strain, however,
monotonously increased until the rupture.
coefficient, D [Ps ]
-1
:D
:'V/Vi
0.05
Attenuation
(b) 14 th mode
'V/Vi [%]
0.008
0.00
0.006
873 K, 0 MPa
0.004 -0.05
0 500 1000
Time [h]
0 500 1000
25
(a) 14 th mode 873 K, 120 MPa
coefficient,D [Ps ]
-1
0.03 20
Strain [%]
'V/Vi [%]
Attenuation
0.0
15
0.02 :D
:'V/Vi 10
:Strain
0.01 5
-0.5
0.00 0
0.0 0.2 0.4 0.6 0.8 1.0
Life fraction, t/tr
Figure 1. Relationship between attenuation coefficient, D, velocity change, 'V/Vi, creep strain and life
fraction, t/tr, at the 14th resonant mode (around 4.5 MHz) [(a) 120 MPa, 873 K] and reference sample [(b)
stress free, 873 K, until 1000 h]. The rupture life of crept sample was 990.6 h. The shear wave polarization
was parallel to the stress direction.
0.05 14th mode :120 MPa
coefficient, D [Ps ]
:120 MPa
-1
0.04 873 K
:140 MPa
Attenuation
:160 MPa
0.03
0.02
0.01
0.00
0.0 0.2 0.4 0.6 0.8 1.0
Life fraction, t/tr
Figure 2. Relationship between D at the 14th resonant modes and life fraction (120, 140 and 160 MPa, 873
K). The shear wave polarization is parallel to the stress direction. Rupture life tr= 990.6 h and 1077.3 h for
120 MPa, tr =1040.4 h for 140 MPa and tr =356.5 h for 160 MPa.
These trends were commonly observed in the other resonant modes and the other
direction of polarization. Attenuation always showed a peak at around t/tr = 0.2 and a
minimum near t/tr = 0.5, independent of the stress for creep as shown in Fig. 2. The
heat treatment only caused insignificant changes in the attenuation and velocity as
shown in Fig. 1(b), indicating that the changes in the ultrasonic properties were caused
solely by creep.
5. Discussion
Possible factors contributing to the attenuation coefficient change in a MHz frequency

range [11] are as follows:
1) Grain scattering,
2) Scatterings caused by precipitation, and
3) Dislocation damping
Effects of 1) and 2) were examined with the scattering theory in the Rayleigh region
[11]. It has been reported that they cause only negligible change in the attenuation
coefficient [6]. Therefore, only dislocation damping can explain the observed acoustic
response.
5.1 Dislocation Damping
Dislocations vibrate in response to ultrasonic stress with a phase lag because of

viscosity and dissipate the energy of the ultrasonic waves. This anelastic mechanism
also lowers the ultrasonic velocity. Dislocation lines are pinned by point defects,
precipitates, and other dislocations. These pinning points act as nodes of vibration of
elastic strings. From the string model by Granato and Lücke [12] for dislocation
damping, the following equation for a lower frequency range than the dislocation-
segment resonant frequency has been derived:
D = A1 /̓L 4 f 2 (1),

where A1 is a positive constant that depends on the shear modulus, Poisson’s ratio,
specific damping constant, and Burger’s vector. / is the dislocation density, and L is
the dislocation length. According to this model, D is proportional to the density /
multiplied by the fourth power of the segment length of the “effective” dislocations,
which are mobile and can vibrate with the low-stress ultrasonic waves. Not all
dislocations interact with ultrasonic waves. These interacting dislocations were defined
as mobile dislocations, which cause ultrasonic attenuation. Dislocations piling up
against grain boundaries or sub-boundaries cannot vibrate and contribute to D.
Dislocation
Cell
structure
Lath
1Pm
1Pm
(a) t/tr=0 (b) t/tr=0.20
Re-crystallization
Sub-grain
1Pm 1Pm
(c) t/tr=0.52 (d) t/tr=0.71
Figure 3. Transmission electron micrographs of crept specimens at t/tr = 0, 0.20, 0.52, and 0.71 (120 MPa,
873 K).
To demonstrate the possibility of dislocation damping, the dislocation structure of

specimens was observed with TEM. Figure 3 shows the TEM micrographs of a crept
sample at 120 MPa. Figure 3 (a) shows the dislocation structure before creep (t/tr = 0),
and Fig.3 (b) through (d), at t/tr = 0.20, 0.52, and 0.71. The direction of stress was in
the longitudinal direction of the photograph. In Fig. 3 (a), the lath structure covered the
entire area. Many dislocations were observed within the lath. Figure 3 (b) shows the
microstructure at t/tr = 0.20, when D shows a peak (Fig. 1). The lath structures still
accounted for the majority of the whole area. The width of the lath structure was larger
than that in Fig. 3 (a). The dislocation density within the lath structure increased. In a
small fraction of the area, cells and sub-grains were observed. Their boundaries were
not clear. Figure 3 (c) shows the microstructure at t/tr = 0.52, when D was at its
minimum. Numerous sub-grains were observed. They were larger in size and their
boundaries were clearer than in Fig. 3 (b). The dislocation density within the sub-grains
was low. Figure 3 (d) shows the microstructure at t/tr = 0.71, when D increased again.
The entire area was covered by the sub-grain structure. The boundaries of the sub-
grains were clearer than those in Fig. 3 (c). The size was larger than in Fig. 3 (c), and
the density was about the same as in Fig. 3 (c).
Figure 4 shows the results obtained by analyzing the TEM microstructures stored
in the computer. For the following determination, at least five TEM micrographs from
different locations on the thin foil were used. Figure 4 (a) shows the dislocation density
/1 and the average dislocation length L1. The density of dislocations was measured
with a method proposed by Keh and Weissmann [13]. Only the dislocations pinned by
carbides in grains and other dislocations, as well as movable dislocations, were
considered. As shown in Fig. 4(a), /1 increased until t/tr = 0.2 and then decreased until
rupture. On the other hand, L1 increased until t/tr = 0.2, and it remained almost constant
within the range of t/tr = 0.2 to 0.5. Afterward, L1 increased with an increase in t/tr. The
change of the dislocation structure in Fig. 3 and the change in free dislocation density
in Fig. 4(a) can be found in past reports. [13-16].
㩷
Life fraction, t/tr
/1[x10 m ] coefficient, D[Ps ]
0.0 0.2 0.4 0.6 0.8 1.0

-1
-1
Dislocation density, Attenuation
10 (b) 14th mode
-2
10
:Calculated D
-3
10 㩷 :Measured D
2 (a) :Density
0.6
length, L1[Pm]
:Length
Dislocation
-2
0.4
14
0.2
0
0.0 0.2 0.4 0.6 0.8 1.0
Life fraction, t/tr
Figure 4. (a): Change of dislocation density and length as creep progresses (120 MPa, 873K), (b):
Comparison between calculated and measured attenuation coefficient in the 14th resonant mode (120 MPa,
873 K).
The attenuation coefficient was given by substituting L1 and /1 for L and / of

Equation (1). Shown in Fig. 4 (b) was the comparison with the calculated and measured
attenuation coefficients of the 14th resonant mode at 120 MPa, based on an assumption
that /1 v / and, L1 v L as shown in Fig. 4 (b). The constant A1 was determined so that
the measured and calculated Ds were consistent with each other before creep.
Theoretically, / and L are the density and length of dislocations that vibrate with
ultrasonic waves (on the order of 0.1 nm or less amplitude), and they are not exactly
equal to L1 and /1. It is impossible to identify which dislocations in the TEM image
actually oscillated with the ultrasonic waves. In general, point defects were absorbed
and pinned on dislocation lines. The distance between the point defects corresponded to
L, which means that L1 exceeded L. The interaction of the point defects pinning
dislocations was considered. It was assumed that longer dislocation lines would
provide longer distances between the point defects. In Fig. 4(b), the calculated and
measured Ds show similar behavior. Similar to the case of austenite stainless steels,
2.25% Cr-1% Mo and Cr-Mo-V steels and Ni-based super-alloy [6, 17-19], it is proved
that this assumption is valid, and the attenuation change results from energy absorption
caused by dislocations. Figure 4(b) shows, however, that the absolute value of the
calculated attenuation coefficient is different from that of the measured one. This
difference arises from the difference between L1 and L since L affects D, as shown in
Equation (1). Thus, the attenuation mainly changes from dislocation restructuring.
In addition, Granato and Lücke [12] derived the following equation for velocity:
'V/V0 = -A2 / L2 (2),
where A2 is a positive constant and V0 =(G/U)1/2 which is the dislocation-independent

velocity, which would ideally be determined at such a high frequency that the
dislocation cannot oscillate in response to the elastic waves. G is the shear modulus and
Uis the density of the material. Velocity is proportional to the dislocation density /
multiplied by the square of the dislocation length L. The increases in / and L lead to
the decrease in velocity. At the point showing the peak of the attenuation coefficient in
Fig. 1, the velocity is altered from a decrease to a slight increase. The velocity rapidly
decreases while D quickly increases (t/tr > 0.5). It appears to be the strongest proof that
the velocity change is caused by dislocation vibration. The attenuation peak was
observed in the creep progression on austenite stainless steels, 2.25%Cr-1%Mo, Cr-
Mo-V steels and Ni-based super-alloy [6, 17-19]. Shown in Table 2 is t/tr, where the
attenuation shows the peak in these materials. These data show the universality of the
EMAR method. In addition, the method is capable of detecting the microstructural
evolution, dislocation mobility, and transition of the dislocation structure during creep
progression.
Table 2. Life fraction at attenuation peak for various materials.

2.25Cr-1Mo Cr-Mo-V
Material SUS 403 Type 304 Type 316L Waspaloy
steel steel
Life fraction, t/t r ,
20-30 % 30-40% 60-70% 50-60% 25-30% 35-40%
at attenuation peak
䇭 Stress (MPa) 120, 140 and 160 100, 110 and 120 100,110 and 120 45 and 65 25,35,45 and 55 140, 150 and 160
Temperature (K) 873 973 973 923 923 1073
6. Conclusion
Creep damage in a martensitic stainless steel (JIS-SUS403) at 873 K in air was

evaluated through ultrasonic attenuation measured with the EMAR method.
Attenuation showed a peak at around 20% and a minimum value at 50 % of the creep
life, independent of the applied stress, which is interpreted as resulting from
microstructural changes, especially, dislocation mobility and restructuring. This result
is supported by TEM observations. The relationship between the changes in ultrasonic
attenuation and velocity and the microstructural change can be explained with the
Granato-Lücke string model. This technique has the potential to assess the damage
advance and to predict the creep life of metals. The EMAT which is most suitable for a
real component will be studied as a future work
References
[1] R. Viswanathan, Damage mechanism and life assessment of high temperature components, ASM
International, Ohio, 1989.
[2] G. Dobmann, M. Kroning, W. Theiner, H. Willems and U. Fiedler, Nondestructive characterization of
materials (ultrasonic and micromagnetic techniques) for strength and toughness prediction and
detection of early creep damage, Nucl. Eng. Design 157 (1995), 137-158.
[3] P. Auerkari, J. Salonen, Accuracy requirement for life assessment, Int. J. Pressure Vessel Piping 39
(1989), 135-144.
[4] M. Hirao, and H. Ogi, EMATs for Science and Industry: Nondestructive Ultrasonic Measurements,
Kluwar Academic Publishers, Boston, 2003.
[5] J. Pesicka, R. Kuzel, A. Dronhofer and G. Eggler, The evolution of dislocation density during heat
treatment and creep of tempered martensitic ferritic steels, Acta Materialia 51 (2003), 4847-4862.
[6] T. Ohtani, H. Ogi and M. Hirao, Creep-induced microstructural changes in 304-type austenitic stainless
steel, Trans. ASME J. Eng. Mat. Tech. 128 (2006), 234-242.
[7] B. W. Maxfield and C. M. Fortuko, The design and use of electromagnetic acoustic wave transducers
(EMATs), Mater. Eval. 11 (1983), 1399-1408.
[8] R. B. Thompson, In: R. N. Thurston and A. D. Pierce A D, eds. Physical Acoustics, XIX. Academic
Press, New York, 1990.
[9] K. Maruyama, C. Harada and H. Oikawa, A strain-time equation applicable up to tertiary creep stage, J.
Soc. Mater. Sci. Jpn. 34 (1985), 1289-1295.
[10] K. Maruyama and H. Oikawa, An extrapolation procedure of creep data for St Determination: with
special reference to Cr-Mo-V steel, Trans. ASME. J. Pressure Vessel Tech. 109 (1987), 142-146.
[11] R. C. Truell, C. Elbaum, B. B. Chick, Ultrasonic methods in Solid State Physics, Academic Press, New
York, 1969.
[12] A. Granato and K. Lücke, Theory of mechanical damping due to dislocations, J. Appl. Phys. 27 (1956),
583-593.
[13] A. S. Keh and S. Weissmann, In: G. Thomas, J. Washburn, eds. Electron Microscopy and Strength of
Crystals, Interscience, New York, 1963.
[14] C. R. Barrett, W. D. Nix and O. D. Sherby, The influence of strain and grain size on the creep
substructure of Fe-3Si, Trans. ASM. 59 (1966), 3-15.
[15] A. Orlova and J. Cadek, Some substructure aspects of high-temperature creep in metals, Phil. Mag. 28
(1973), 891-899.
[16] T. Hasegawa, Y. Ikeuchi and S. Karashima, Internal stress and dislocation structure during sigmoidal
transient creep of a copper-16 at.-% Aluminum alloy, Metal. Sci. J. 6 (1972), 78-82.
[17] T. Ohtani, H. Ogi and M. Hirao, Change of ultrasonic attenuation and microstructure evolution during
creep of 2.25%Cr-1%Mo steels, Metall. Mater. Trans. A 36A (2005), 2967-2977.
[18] T. Ohtani, H. Ogi and M. Hirao, Evolution of microstructure and acoustic damping during creep of a
Cr-Mo-V ferritic steel, Acta Materialia 54 (2006), 2705-2713.
[19] T. Ohtani, H. Ogi and M. Hirao, Acoustic damping characterization and microstructure evolution of Ni-
based super-alloy during creep, Int. J. Solids Struct. 42 (2005), 2911-2928.
IOS Press, 2007
NDE Method Using Minor Hysteresis Loops

in Ferromagnetic Materials
Seiki TAKAHASHI 1, Satoru KOBAYASHI, Yasuhiro KAMADA,
Hiroaki KIKUCHI, and Katsuyuki ARA
NDE and Science Research Center, Faculty of Engineering, Iwate University, Morioka
020-8551, Japan
Abstract. Minor hysteresis loops of plastically deformed Fe single crystals, polycrystals and low
carbon steel have been analyzed. We propose an analysis method of minor loops where coefficients
sensitive to lattice defects are obtained from relations between field-dependent minor-loop
parameters. The coefficients are independent of magnetic field amplitude of minor loops and
magnetic field, and can be obtained with low magnetic fields less than a 1 kA/m. These
characters are useful for nondestructive evaluation of material degradation in ferromagnetic
materials.
Keywords. Minor hysteresis loops, dislocations, nondestructive testing, Fe, low carbon steel
1. Introduction
Nondestructive evaluation (NDE) of material degradation of the pressure vessel in nuclear

reactor is one of the most urgent priorities for the safe operation of such plants. The
deterioration of the pressure vessel determines the lifetime of the nuclear reactor and is
currently evaluated by impact tests of Charpy test pieces which are made of the individual
vessel base and weld material and preinstalled in the reactor. However, Charpy tests need a
large number of test pieces and the shortage of test samples becomes problematic especially
when liftime extension is planned.
The structure-sensitive magnetic properties are useful physical quantities to obtain
information on lattice defects.[1,2] It was revealed that coercive force of the major loop
increases in proportion to the square root of dislocation density. However, magnetic fields
larger than 10 kA/m are necessary to measure the major loop in general. On the other hand,
minor hysteresis loops, which can be obtained with lower magnetic field, offer several
advantages compared with the major loop. By measuring minor loops with various magnetic
field amplitudes, it would be possible to obtain much information on lattice defects in
ferromagnetic materials with high sensitivity.[3] However, the analysis of minor loops has
not been exploited as a NDE technique so far, because minor-loop properties strongly
1
Corresponding Author: NDE&Science Research Center, Faculty of Engineering, Iwate University, 4-3-5
Ueda, Morioka 020-8551, Japan; Phone:+81-19-621-6431; Email: seiki.t@iwate-u.ac.jp
210 S. Takahashi et al. / NDE Method Using Minor Hysteresis Loops in Ferromagnetic Materials
depend on external parameters and it is difficult to relate magnetic properties with lattice
defects quantitatively.
In this paper, we present a method for analyzing minor hysteresis loops by using
results of Fe single crystals, Fe polycrystals and low carbon steel, where the above
disadvantages are drastically reduced and field-independent structure-sensitive properties
can be obtained.[4-6]
2. Experiment
The Fe single crystals, Fe polycrystals, and A533B steel were plastically deformed in
tension. Sheets of Fe single crystals with 1 mm in thickness have surface plane of (001) and
were deformed along [100]. A533B, a low carbon steel (0.18 wt% C, 0.15 wt% Si, 1.50
wt% Mn, 0.03 wt% Cu, 0.66 wt% Ni, 0.56 wt% Mo, balance Fe) is used in nuclear reactor
pressure vessels. The grain size of A533B steel, determined using metallurgical microscopy,
is about 10 Pm. After unloading, the samples were shaped into picture frames for Fe single
crystals or rings for Fe polycrystals and A533B steel for minor-loop measurements. The
exciting and detecting coils were wound on these samples. Minor hysteresis loops with
various magnetic field amplitude Ha up to 1.6 kA/m for Fe single crystals and polycrystals
and up to 8 kA/m for A533B steel, were measured at room temperature using a flux meter.
Before measuring each minor loop, the samples were demagnetized. Magnetic properties of
the major loops for Fe single crystals and polycrystals, and for A533B steel were obtained
with Ha = 1.6 kA/m and Ha = 8 kA/m, respectively.
3. Experimental Results
Figure 1. Set of minor loops for Fe single crystals with true stress of (a) 0 MPa, (b) 95 MPa, and (c) 142 MPa.
For clarity, only representative loops are shown.
S. Takahashi et al. / NDE Method Using Minor Hysteresis Loops in Ferromagnetic Materials 211
As representative data, we will show results of minor loops for Fe single crystals. Figure 1
shows a set of minor hysteresis loops for Fe single crystals with and without tensile
deformation. It is clearly seen that tensile stress drastically changes the shape of minor
loops. For analysis of minor loops, several magnetic properties of minor loops are
introduced as shown in Fig. 2: minor-loop coercive force Hc*, minor-loop magnetization
Ma*, minor-loop hysteresis loss WF*, minor-loop remanence MR*, minor-loop remanence
work WR*, and three minor-loop susceptibilities FH*, FR*, and Fa* at magnetic field H =
M
Ma*
Fa*
FR* +MR*
WR* FH*
-Hc*
0 +Hc* +Ha H
WF*
MR*
Figure 2. Magnetic parameters of a minor loop.
㪈㪅㪏㪐㪇
(a) (b)
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫄㪸㪾㫅㪼㫋㫀㫑㪸㫋㫀㫆㫅㩷P㪇㪤㪸㪁㩷㩿㪫㪀
㪈㪅㪍㪏㪇
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㪺㫆㪼㫉㪺㫀㫍㪼㩷㪽㫆㫉㪺㪼㩷㪟㪺㪁㩷㩷㩿㪘㪆㫄㪀
㪈㪅㪋㪎㪇
True stress
0 MPa 54 MPa
㪈㪅㪉㪍㪇
95 MPa 135 MPa
142 MPa
㪈㪅㪇 0.5 㪌㪇
㪇㪅㪏 0.4 㪋㪇
0.3
㪇㪅㪍㪊㪇
0.2
㪇㪅㪋 0.1
㪉㪇
True stress
㪇㪅㪉 0.0 㪈㪇
0 20 40 60 80 100 0 MPa 54 MPa
Ha (A/m) 95 MPa 135 MPa 142 MPa
㪇㪅㪇㪇
㪇㪅㪇㪇㪅㪉㪇㪅㪋㪇㪅㪍㪇㪅㪏㪈㪅㪇㪈㪅㪉㪈㪅㪋㪈㪅㪍㪇㪅㪇㪇㪅㪉㪇㪅㪋㪇㪅㪍㪇㪅㪏㪈㪅㪇㪈㪅㪉㪈㪅㪋㪈㪅㪍
㪤㪸㪾㫅㪼㫋㫀㪺㩷㪽㫀㪼㫃㪻㩷㪸㫄㫇㫃㫀㫋㫌㪻㪼㩷㪟㪸㩷㩷㩿㪈㪇㪊㩷㪘㪆㫄㪀㪤㪸㪾㫅㪼㫋㫀㪺㩷㪽㫀㪼㫃㪻㩷㪸㫄㫇㫃㫀㫋㫌㪻㪼㩷㪟㪸㩷㩷㩿㪈㪇㪊㩷㪘㪆㫄㪀
Figure 3. (a) Minor-loop magnetization Ma* and (b) minor-loop coercive force Hc*, as a function of magnetic
field amplitude Ha for Fe single crystals with and without tensile deformation.
Hc*, 0, and Ha, respectively. WR* is the area enclosed by a minor loop in the second
quadrant. These parameters correspond to those of the major loop when Ha is large enough
to saturate the samples.
Figures 3(a) and 3(b) show Ma* and Hc* as a function of Ha, respectively, for Fe single
crystals with and without tensile deformation. It is clearly seen that Ma* decreases with
increasing true stress while Hc* drastically increases. The curves of Fig. 3(a) correspond to
the virgin magnetization curves and can be divided conveniently into four stages as follows.
In the first stage, the Bloch wall moves reversibly and the slope at low fields gives the
initial susceptibility. The irreversible displacement of the Bloch wall contributes mainly to
the magnetization in the second stage and the magnetization shows the steepest slope. The
gentle slope following the steepest section corresponds to the third stage. Finally, in the
fourth stage, the magnetic domain walls sweep out of the sample and the magnetization
proceeds only by the rotation of magnetic moments. For the present analysis of minor loops,
magnetization process both in first and second stages is very important.
For all deformed Fe single crystals, minor loops have a similar figure in the second
stage. This can be confirmed by comparing WR* and MR* with WF* and Ma*, respectively.
As shown in Fig. 4, both WF*-WR* and Ma*-MR* curves can be plotted on a straight line for
all samples except for the large values; WR*/WF* a 1/6 and MR*/Ma*a 3/4. The former
relation means that the area of hysteresis work is one-third of the total area of a minor loop
in the second stage. These results show that the shape of minor loops is independent of Ha
as well as true stress, being indicative of the shape similarity of the Bloch wall potential in
the second stage. Such similarity was observed also in Fe polycrystals and A533B steel
after tensile deformation, whereas a slope of the relations is slightly different from that of
Fe single crystals; WR*/WF* a 1/8 and MR*/Ma*a 2/3 for Fe polycrystals and WR*/WF* a
㪎㪇㪇㪈㪅㪏
True stress True stress
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㪿㫐㫊㫋㪼㫉㪼㫊㫀㫊㩷㫃㫆㫊㫊㩷㪮㪝㪁㩷㩷㩿㪡㪆㫄㪊㪀
0 MPa 0 MPa
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫄㪸㪾㫅㪼㫋㫀㫑㪸㫋㫀㫆㫅㩷P㪇㪤㪸㪁㩷㩿㪫㪀
㪈㪅㪍
㪍㪇㪇 54 MPa 54 MPa
95 MPa 95 MPa
135 MPa 㪈㪅㪋 135 MPa
㪌㪇㪇 142 MPa 142 MPa
㪈㪅㪉
㪋㪇㪇㪈㪅㪇
㪊㪇㪇㪇㪅㪏
㪇㪅㪍
㪉㪇㪇
㪮㪩㪁㪆㪮㪝㪁a㪈㪆㪍㪤㪩㪁㪆㪤㪸㪁㪔㪊㪆㪋
㪇㪅㪋
㪈㪇㪇㪇㪅㪉
㩿㪸㪀㩿㪹㪀
㪇㪇㪅㪇
㪇㪈㪇㪉㪇㪊㪇㪋㪇㪌㪇㪍㪇㪇㪅㪇㪇㪅㪉㪇㪅㪋㪇㪅㪍㪇㪅㪏㪈㪅㪇㪈㪅㪉
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷㫎㫆㫉㫂㩷㪮㪩㪁㩷㩷㩿㪡㪆㫄㪊㪀㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷P㪇㪤㪩㪁㩷㩿㪫㪀
Figure 4. Relations (a) between WF* and WR* and (b) between Ma* and MR* for Fe single crystals with and
without tensile deformation.
1/8 and MR*/Ma*a 3/5 for A533B steel.

All minor-loop parameters depend on the external parameter Ha as shown in Fig. 3 and
it is difficult to extract the intrinsic physical properties from the Ha dependence of
㪈㪇㪊㪈㪇㪉
True stress
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷㫎㫆㫉㫂㩷㪮㪩㪁㩷㩷㩿㪡㪆㫄㪊㪀
True stress
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㪿㫐㫊㫋㪼㫉㪼㫊㫀㫊㩷㫃㫆㫊㫊㩷㪮㪝㪁㩷㩷㩿㪡㪆㫄㪊㪀
0 MPa 0 MPa
54 MPa 54 MPa
95 MPa 㪈㪇㪈
㪈㪇㪉 95 MPa
135 MPa 135 MPa
142 MPa 142 MPa
㪈㪇㪇
㪈㪇㪈
㪈㪇㪄㪈
㪇
㪈㪇
㪈㪇㪄㪉
㪈㪇㪄㪈
㪈㪇㪄㪊
㩿㪸㪀㩿㪹㪀
㪈㪇㪄㪉㪈㪇㪄㪋
㪈㪇㪄㪊㪈㪇㪄㪉㪈㪇㪄㪈㪈㪇㪇㪈㪇㪄㪋㪈㪇㪄㪊㪈㪇㪄㪉㪈㪇㪄㪈㪈㪇㪇
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫄㪸㪾㫅㪼㫋㫀㫑㪸㫋㫀㫆㫅㩷P㪇㪤㪸㪁㩷㩿㪫㪀㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷P㪇㪤㪩㪁㩷㩿㪫㪀
Figure 5. The relations (a) between WF* and Ma* and (b) between WR* and MR* for Fe single crystals with and
without tensile deformation. The solid lines through the data show least-squares fits.
㪈㪇㪉
True stress True stress
0 MPa 0 MPa
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㪺㫆㪼㫉㪺㫀㫍㪼㩷㪽㫆㫉㪺㪼㩷㪟㪺㪁㩷㩷㩿㪘㪆㫄㪀
54 MPa 54 MPa
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫊㫌㫊㪺㪼㫇㫋㫀㪹㫀㫃㫀㫋㫐㩷F㪩㪁㩷
95 MPa
135 MPa
㪈㪇㪋 95 MPa
135 MPa
142 MPa 142 MPa
㪈㪇㪈
㪈㪇㪊
㪈㪇㪉
㪈㪇㪇
㩿㪸㪀㩿㪹㪀
㪈㪇㪄㪋㪈㪇㪄㪊㪈㪇㪄㪉㪈㪇㪄㪈㪈㪇㪇㪈㪇㪄㪋㪈㪇㪄㪊㪈㪇㪄㪉㪈㪇㪄㪈㪈㪇㪇
㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷P㪇㪤㪩㪁㩷㩿㪫㪀㪤㫀㫅㫆㫉㪄㫃㫆㫆㫇㩷㫉㪼㫄㪸㫅㪼㫅㪺㪼㩷P㪇㪤㪩㪁㩷㩿㪫㪀
Figure 6. The relations (a) between Hc* and MR* and (b) between FR* and MR* for Fe single crystals with and
without tensile deformation. The solid lines through the data show least-squares fits.
minor-loop parameters. Therefore, it is important to find new magnetic properties that are
represented by the internal physical properties.
Figures 5(a) and 5(b) show WF*-Ma* and WR*-MR* curves, respectively, for Fe single
crystals with and without tensile deformation. In the second stage, these curves show nearly
straight lines in double logarithmic scale and can be represented by a simple equation given
by
nF
§ M* ·
¸ , (1)
* 0
WF W ¨ a
¨ Ms
F
¸
© ¹
and
nR
§ M* ·
WR *
W ¨ R
R
0
¸ , (2)
¨ MR ¸
© ¹
where WF0 and WR0 are minor-loop coefficients independent of Ha and sensitive to lattice
defects. Ms and MR are the saturation magnetization and remanence of the major loop,
respectively. nF a 3/2 and nR a 3/2, being independent of true stress and ferromagnetic
materials. Note that the relation of Eq. (1) was discovered by Steinmetz about a century ago
and is known as Steinmetz law where nF a 1.6.[7]
As shown in Fig. 6(a), we also found that Hc* shows a relation with MR* in the second
stage, expressed as
nc
§ M* ·
Hc * 0
H ¨ R
c ¸ , (3)
¨ MR ¸
© ¹
where Hc0 is a coefficient and nc a0.45.

Similarly, as is seen in the relation between FR* and MR* in Fig. 6(b), three simple
relations about minor-loop susceptibilities were found as below;
1 H c* , (4)
Rc* Rc0 exp(b )
F H* Hc
nT
§ M R* · , (5)
F R* F T0 ¨ ¸
© MR ¹
ns
§ M a* · , (6)
F a* Fs0 ¨ ¸
© Ms ¹
6
10 16 10
Coefficient FT0 Coefficient HC0
(a) 0 (b)
Coefficient F S 14 Coefficient WF0
5
Initial susceptibility Fi Coercive force HC 8
10
Hc0, Hc(102 A/m)

12
WF0(103 J/m3)
F6F5FK
Fe polycrystal
10
6
Fe polycrystal
4 A533B steel
10
8
Fe single crystal
6 4
Fe single crystal
3
10
4 A533B steel
2
2
2
10
0 0
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
True stress(MPa) True stress(MPa)
Figure 7. The true stress dependence (a) of Fi, FT0 and FS0, and (b) Hc, WF0 and Hc0, for Fe single crystals,
polycrystals and A533B steel. The solid lines through the data are guide to eyes.
where Rc0, FT0 and Fs0 are coefficients which are related with lattice defects. b, nT and ns are
constants which depend on the kind of materials, and have a value of 2-6, 0.1-0.5 and
0.5-1.0, respectively.
The six simple relations between field-dependent minor-loop parameters, Eqs. (1)-(6),
were found to be valid also for Fe polycrystals and A533B steel, and the dependence of
minor-loop coefficients on true stress was obtained for all samples investigated.
Figures 7(a) and 7(b) show the true stress dependence of minor-loop coefficients FT0,
FS , WF0 and Hc0. To compare with traditional structure-sensitive properties of the major
0
loop, initial susceptibility Fi and coercive force Hc are also shown. As shown in Fig. 7(a),
FT0, Fs0 and Fi decrease with the increase of true stress for all deformed samples, though the
sensitivity of these coefficients is low in A533B steel. This is due to the fact that A533B
steel contains many lattice defects even at zero true stress and the increase of dislocation
density by tensile deformation does not influence the Bloch wall mobility significantly. The
minor-loop coefficients WF0 and Hc0, and Hc increase with increasing true stress for all
samples as shown in Fig. 7(b).
4. Discussion
The six minor-loop coefficients, WF0, WR0, Hc0, Rc0, FT0, and Fs0 were obtained from simple
relations between field-dependent minor-loop parameters. They are intrinsic physical
quantities that are independent of magnetic field amplitude as well as the magnetic field,
and are sensitive to lattice defects such as dislocations. The present method using set of
minor hysteresis loops has several advantages of nondestructive evaluation of degradation

compared with the traditional method using major loop. One of them is the very small
magnetic field required for minor-loop measurements. For instance, a maximum field
amplitude used for minor-loop analysis is typically less than 20% of that for major-loop
analysis. This small measurement field is an important characteristic for the design of the
equipment.
The minor loops were measured with increasing Ha, step by step, and the smaller step
may bring about higher reliability of minor-loop coefficients. Nevertheless, using a
similarity rule of minor loops, the amount of time for measurements and analysis can be
drastically reduced without loss of reliability.
In the present study, the picture frame samples cut from deformed tensile test pieces
were used for the detailed analysis of minor loops. For NDE of pressure vessel in nuclear
reactor, however, we need to measure magnetic properties of Charpy impact test pieces
nondestructively. It has been shown by recent magnetic measurements using magnetic
yokes that our analysis method can be applied for NDE of such test pieces and the detailed
results will appear elsewhere.
We need further investigation of lattice defects due to neutron irradiation using present
minor-loop analysis. The relation between Cu precipitates and minor-loop coefficients are
under investigation for samples irradiated by neutrons.
Acknowledgments
This research was supported by a Grant-in-Aid for Scientific Research (S), No. 141020345,
from the Ministry of Education, Culture, Sports, Science and Technology of Japan.
References
[1] H. Kronmüller and M. Fähnle, Micromagnetism and the microstructure of ferromagnetic solids, Cambridge
University press, 2003.
[2] H. Träuble: Magnetism and Metallurgy. Edited by Berkowitz A E and Kneller. E (Academic, New York
1969) Chap XIII 621.
[3] I. Tomás, J. Magn. Magn. Mater. 268 (2004) 178-185.
[4] S. Takahashi, T. Ueda and L. Zhang, J. Phys. Soc. Jpn. 73 (2004) 239-244.
[5] S. Takahashi and L. Zhang, J. Phys. Soc. Jpn. 73 (2004) 1567-1575.
[6] S. Takahashi, L. Zhang, S. Kobayashi, Y. Kamada, H. Kikuchi, and K. Ara, J. Appl. Phys. 98 (2005)
033909/1-8.
[7] C. P. Steinmetz, Trans. A. I. E. E. 9 (1892) 3-51.
.
IOS Press, 2007
Investigation of Neutron Radiation Effects on

Fe Model Alloys by Minor-loop Analysis
Satoru KOBAYASHI1 , Hiroaki KIKUCHI, Seiki TAKAHASHI,
Katsuyuki ARA, and Yasuhiro KAMADA
NDE and Science Research Center, Faculty of Engineering, Iwate University, Morioka
020-8551, Japan
Abstract. We have measured minor hysteresis loops of neutron-irradiated Fe model alloys. The change
of minor-loop properties with neutron fluence, which depends on the copper content was detected. The
results were explained as a compensation of internal stress in dislocations by copper precipitates around
the dislocations.
Keywords. Neutron irradiation, minor hysteresis loop, steel, Fe alloy, copper precipitates
1. Introduction
The nondestructive evaluation of irradiation embrittlement of pressure vessels in

nuclear-power plants has became an urgent matter of study because deterioration of the
pressure vessels determines the lifetime of the nuclear reactors. At present, the irradiation
embrittlement of pressure vessels is evaluated using the Charpy impact test. This
destructive testing technique exploits the increase in ductile-brittle transition temperature
(DBTT). The diminishing stock of Charpy impact test samples preinstalled in the reactors is
an urgent issue.
The change of mechanical properties due to neutron irradiation has been studied for
reactor pressure vessel materials and Fe model alloys.[1] The neutron irradiation induces
several kinds of lattice defects such as vacancies, interstitial atoms, voids, dislocation loops
and copper precipitates. Some of the lattice defects make the ductility of materials low and
make the brittleness increase. The structure of copper precipitates has been studied by the
atom-probe field ion microscope,[2] positron annihilation,[3] and small-angle neutron
scattering[4] because copper precipitation is the primary cause of brittleness in nuclear
reactor pressure vessels (NRPVs). It has been shown that copper precipitates of 2 - 3 nm in
size, which nucleate and grow due to irradiation, make materials brittle.[5,6] However, the
nucleation mechanism which takes into effect of dislocations has not been studied in detail,
although the NRPVs have high dislocation density even in the initial state. Dislocations
1
Corresponding Author: NDE&Science Research Center, Faculty of Engineering, Iwate University, 4-3-5 Ueda,
Morioka 020-8551, Japan; Phone:+81-19-621-6350; Email: koba@iwate-u.ac.jp
218 S. Kobayashi et al. / Investigation of Neutron Radiation Effects on Fe Model Alloys
would exert a great influence on the nucleation of lattice defects, because there exists
elastic interaction between copper precipitates and dislocations.
The magnetic method using minor hysteresis loops offers several advantages to get
information on lattice defects.[7-9] It has been shown for cold rolled low carbon steel that
minor-loop properties exhibit good correlation with mechanical properties such as Vickers
hardness and DBTT.[10] In addition, the magnetic method is nondestructive and can be
used with low magnetic fields (less than a 1 kA/m). However, reports on magnetic
properties of irradiated materials are very few and irradiation effects on magnetic properties
are not fully understood yet.
In this paper, we have systematically investigated minor-loop properties of neutron
irradiated Fe model alloys and tried to explain the results in view of copper precipitation in
the presence of dislocations.
2. Experiment
Neutron-irradiated tensile test samples were prepared by Professor Odette’s group at the
University of California, Santa Barbara (UCSB). The samples have dimensions of 24 u 5 u
0.5 mm3 and their chemical compositions are listed in Table I, where the samples are
labeled as OV15-20. Before neutron irradiation, the samples were solution treated for 17 h
at 775 qC and then quenched at room temperature by forced helium gas.
Table I. Chemical compositions of measuring samples. (wt %)
Sample Cu Ni Mn Mo Si
OV15 0.9 1.6 1.0
OV16 0.5 1.6 1.0
OV17 0.2 1.6 1.6
OV18 1.6 1.6 0.25
OV19 1.6 1.6 0.5
OV20 1.6 0.5 0.5
Table II. Conditions of neutron irradiation.
t t
Capsule Fluence ) Flux I Flux regime Effective Fluence ) eff
19 -2 12 -2 -1 19 -2
(10 n·cm ) (10 n·cm ·s ) (10 n·cm )
T29 0.44 0.89 high 0.44
T30 0.02 0.07 low 0.07
T31 0.06 0.72 high 0.07
T32 0.02 0.26 Medium 0.04

S. Kobayashi et al. / Investigation of Neutron Radiation Effects on Fe Model Alloys 219
We examined four irradiation conditions with various neutron fluence )t up to 0.44 ×

10 n cm-2, and with neutron flux in the three different regimes as listed in Table II; 0.07 ×
19
1012 n cm-2s-1 (low), 0.26 × 1012 n cm-2s-1 (medium), and 0.72 - 0.89 × 1012 n cm-2s-1 (high).
The irradiation temperature was 290 qC.
Minor hysteresis loops with various magnetic field amplitude Ha up to 6 kA/m were
measured at room temperature using an apparatus for neutron-irradiated small tensile test
samples. The apparatus includes a closed magnetic circuit with a sample sandwiched
between two magnetic yokes made of Fe-3wt% Si steel. The magnetic field with the
frequency of 1 Hz was applied along the long axis of the sample. The magnetic properties
were typically averaged over 5 and 2 samples for unirradiated and irradiated cases,
respectively.
To analyze minor loops, we introduce several minor-loop parameters as shown in Fig.
1: minor-loop coercive force Hc*, minor-loop magnetization Ma*, minor-loop hysteresis
loss WF*, minor-loop remanence MR*, minor-loop remanence work WR*, and three
minor-loop susceptibilities FH*, FR*, and Fa* at magnetic field H = Hc*, 0, and Ha,
respectively. WR* is the area enclosed by a minor loop in the second quadrant.
These minor-loop parameters strongly depend on Ha and it is difficult to extract
intrinsic physical properties from minor loops. Recently, however, we found six simple
relations between the field-dependent minor-loop parameters, where minor-loop
coefficients independent of Ha and sensitive to lattice defects were obtained[7-9]. In this
study, we will pay attention to three relations given by
nF
§ M* ·
¸¸ , (1)
* 0
WF W ¨ a
¨ Ms
F
© ¹
M
Ma*
Fa*
FR* +MR*
WR* FH*
-Hc*
0 +Hc* +Ha H
WF*
MR*
Figure 1. Magnetic parameters of a minor loop.

nR
§ M* ·
WR * 0
W ¨ R ¸ , (2)
R
¨ MR ¸
© ¹
nc
§ M* ·
H c* H c0 ¨ R ¸ . (3)
¨ MR ¸
© ¹
Here, WF0, WR0, and Hc0 are minor-loop coefficients, and Ms and MR are the saturation
magnetization and remanence of the major loop, respectively. The exponents of the power
laws, nF, nR, and nC are a 3/2, a 3/2, and a0.45, respectively, which are independent of
kinds of lattice defects and ferromagnetic materials. The relation of Eq. (1) with nF a 1.6 is
known as Steinmetz law.[11] These minor-loop coefficients were usually obtained with low
magnetic fields (less than 1 kA/m) and are proportional to coercive force of the major loop.
In this study, these coefficients were used to obtain intrinsic physical properties reflecting
lattice defects in irradiated materials.
3. Experimental Results
Figures 2(a)-2(d) show hysteresis loops for OV15, OV16, OV17, and OV19 samples,
respectively, before and after neutron irradiation, obtained with the maximum magnetic
field amplitude of Ha = 6 kA/m. For clarity, only the results obtained for high flux
conditions are shown. Note that although the amplitude of Ha = 6 kA/m is not enough to
saturate the samples, Hc* is almost saturated and equals to coercive force. For all measuring
samples, changes in the shape due to neutron irradiation are not so large. However, we
found that for OV15, OV16, and OV17 samples, Hc* becomes small at a sufficiently high
neutron fluence and its behavior with fluence strongly depends on the chemical
composition. On the other hand, no significant change in Hc* was detected for OV18,
OV19 and OV20 samples within limit of experimental accuracy.
Figure 3(a) shows minor-loop coefficients WF0 and WR0, as a function of neutron
fluence for OV15 sample with various neutron flux. Note that since Hc0 obtained from Eq.
(3) shows a similar fluence dependence to WF0 and WR0, the data for Hc0 are not shown.
Both coefficients for medium and low flux have different from each other despite the same
neutron fluence of )t = 0.02 × 1019 n cm-2. Such difference was also seen for OV16 and
OV17 samples. To compare the minor-loop coefficients obtained for different neutron flux,
an effective neutron fluence )teff is introduced [12]:
1/ 2
§I ·
) teff )t ¨ r ¸¸ , (4)
Ï
© ¹
where I is a neutron flux and Ir is a reference flux and 0.89 × 1012 n cm-2s-1 in this study.
This equation is based on a vacancy plus self-interstitial-atom recombination rate
controlling mechanism and is used to explain the dependence of irradiation hardening on
neutron flux; pre-plateau region of irradiation hardening is shifted to higher )t with
increasing neutron flux.
Figure 2. M-H loops for (a) OV15, (b) OV16, (c) OV17, and (d) OV19 samples with various neutron fluence,
measured with Ha = 6 kA/m. Only the data for high flux are shown.
Figures 3(b)-3(d) show the dependence of WF0 and WR0 on )teff for OV15, OV16, and
OV17 samples, respectively. One can see that both WF0 and WR0 show smooth curves in this
case. This implies that both magnetic and mechanical properties have the underlying
mechanism of irradiation effects. Both WF0 and WR0 sharply increase at low fluence and
show a maximum around )teff = 0.05 × 1019 n cm-2, followed by a gradual decrease for
4200 300 4200 300

(a) OV15 (b) OV15
4000 280 4000 280
3800 260
WR0 (J/m3)
3800 260
WF0 (J/m3)
WR0 (J/m3)
WF0 (J/m3)
3600 240 3600 240
3400 220
3400 220
3200 200
3200 200
3000 180
3000 180 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
t 19 -2
19 -2 Effective neutron fluence ) eff (10 cm )
Neutron fluence )t(10 cm )
3800 260 5500 420
3700
(c) OV16 (d) OV17 400
250
3600 5000 380
240
WR0 (J/m3)
WR0 (J/m3)
WF0 (J/m3)
360
WF0 (J/m3)
3500
340
3400 230 4500
320
3300
220
300
3200 4000
210 280
3100
260
3000 200 3500
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
t 19 -2 t 19 -2
Effective neutron fluence ) eff (10 cm ) Effective neutron fluence ) eff (10 cm )
Figure 3. WF0 and WR0 as functions of neutron fluence )t for (a) OV15 sample, and of effective neutron fluence
)teff for (b) OV15, (c) OV16, and (d) OV17 samples. The solid circles, open circles, and triangles denote the data
for high, medium, and low flux, respectively.
OV15 and OV16 samples, while WF0 and WR0 monotonically decrease with fluence for
OV17 sample. In contrast, for OV18, OV19, and OV20 samples, no significant change in
minor-loop coefficients was detected within an experimental accuracy.(not shown) These
results indicate that the copper content influences the behavior of minor-loop coefficients at
low neutron fluence.
4. Discussion
For OV15, OV16, and OV17 samples including copper, minor-loop coefficients become
smaller at a sufficiently high neutron fluence. This indicates that internal stress decreases
during neutron irradiation. This result is in contrast to our expectation that the coefficients
will increase with neutron fluence because irradiation defects act as obstacles to the
movement of Bloch wall. To explain this, we introduce an idea that copper precipitates
grow up in the vicinity of dislocations during neutron irradiation. Copper precipitates have
stress field in themselves and will make minor-loop coefficients increase if copper
precipitates form in the matrix. However, the stress field is reduced and minor-loop
coefficients will decrease when copper precipitates exist in the vicinity of dislocations.
Edge dislocations include the compressive and repulsive stress field and copper precipitates
will gather around the dislocations in order to reduce the elastic energy. Consequently, the
copper precipitates compensate the stress field of dislocations, resulting in a decrease in
minor-loop coefficients as was observed experimentally. The radiation defects fixed around
dislocations strongly disturb the dislocation movement and will make mechanical
properties increase. Direct observation using transmission electron microscopy is planned
to confirm this explanation.
5. Conclusions
Minor hysteresis loops of neutron-irradiated Fe model alloys have been measured to

investigate the influence of radiation defects on minor-loop properties. It was revealed that
the behavior of minor-loop coefficients at low neutron fluence depends on copper content
and the coefficients become small at a sufficiently high neutron fluence, whereas no
significant change was detected for alloys with no copper content. This indicates that
copper precipitates induce the changes in magnetic properties. The decrease of the
coefficients with neutron fluence may be explained as due to compensation of internal
stress of dislocations by copper precipitates grown around the dislocations.
Acknowledgments
This research project has been conducted under the research contract with the Japan
Nuclear Safety Organization (JNES). We thank Professor G. R. Odette of UCSB for

allowing us to measure neutron irradiated samples, and Dr. T. Yamamoto, J. Smith, and Dr.
D. Klingensmith of UCSB for their technical supports for treating irradiated samples.
References
[1] J. Koutský and J. Koík, in Radiation Damage of Structural Materials, (Elsevier Science Publishers, 1994).
[2] P. Othen, M. Jenkins, G. Smith, and W. Phythian, Phil. Mag. Lett. 64 (1991) 383.
[3] S. Pizzini, K. Roberts, W. Phythian, C. English, and G. Greaves, Phil. Mag. Lett. 61 (1990) 223.
[4] M. K. Miller, B. D. Wirth and G. R. Odette, Mater. Sci. Eng. A 353 (2003) 133.
[5] A.Youle and B. Ralph, J. Met. Sci. 6 (1972) 149.
[6] S. Goodman, S. Brenner, and J. R. Low, Jr., Metall. Trans. 4 (1973) 2371.
[7] S. Takahashi, L. Zhang, S. Kobayashi, Y. Kamada, H. Kikuchi, K. Ara, J. Appl. Phys. 98 (2005) 033909.
[8] S. Takahashi, T. Ueda and L. Zhang, J. Phys. Soc. Jpn. 73 (2004) 239.
[9] S. Takahashi and L. Zhang, J. Phys. Soc. Jpn. 73 (2004) 1567.
[10] S. Takahashi, S. Kobayashi, H. Kikuchi, Y. Kamada, J. Appl. Phys. 100(2006) 113908.
[11] C. P. Steinmetz, Trans. A. I. E. E. 9 (1892) 3.
[12] G. R. Odette, T. Yamamoto, D. Klingensmith, Philos. Mag. 85 (2005) 779.
IOS Press, 2007
Numerical Analysis for Non-Destructive

Evaluation of Hardening Steel Taking into
Account Measured Magnetic Properties
Depending on Depth
Yuichiro KAI a,1, Yuji TSUCHIDA b and Masato ENOKIZONO b
a
JSPS Research Fellow, Oita University, 700 Dannoharu, Oita, 870-1192, JAPAN
b
Faculty of Engineering , 700 Dannoharu, Oita, 870-1192, JAPAN
Abstract. We will examine a non-destructive evaluation method of hardening

depth and hardness using numerical analysis. In our numerical analysis, the
magnetic properties of carbon steel at different hardening depth are assumed by
using the measured results from the cut-out ring type specimens. We try to
evaluate the hardening depth and the hardness by calculating the magnetic flux
density and the magnetic field strength by the two-dimensional finite-element
method taking into account the measured magnetic properties.
Keywords. Carbon steel, Hardening depth, Magnetic property, Numerical analysis
1. Introduction
Parts of machines and structures are hardened by using induction heating. The case
hardening can increase the hardness of steel. The different strength for the parts is
required by controlling the hardening depth and the hardness. However, it is difficult
to control the hardening depth and the hardness. If the case hardening is applied
effectively to the steels, it is very important to evaluate the hardening depth and the
hardness.
We have proposed a non-destructive evaluation system of the hardening depth and
the hardness by using an electromagnetic method. In previous works, the difference of
the magnetic property depending on depth is clarified by measuring the magnetic
properties of cut-out ring specimens [1, 2]. And we suggest to evaluate the hardening
depth and the hardness by measuring the change of the magnetic properties depending
on the depth. At first, we try to detect the change of the eddy current inside the
hardening carbon steel by using a pick-up coil. In this case, we proposed to employ a
multi-frequency excitation and spectrogram (MFES) method [3, 4]. Seen from the
results, it was clear that the pick-up voltage changes as a function of the different
hardening conditions. However, we suppose that it is important to obtain not only the
change of the pick-up voltage but also the plenty of information inside the hardening
steel since the several magnetic properties and mechanical properties change by using
1
Corresponding Author: JSPS Research Fellow, Materials Science and Production Engineering,
Graduate School of Engineering, Oita University, 700 Dannoharu, Oita 870-1192, Japan; E-mail:
ykai@cc.oita-u.ac.jp
226 Y. Kai et al. / Numerical Analysis for Non-Destructive Evaluation of Hardening Steel
the case hardening. Therefore, We suggest to evaluate the hardening depth and the
hardness by measuring magnetic flux density B and the magnetic field strength H from
outside instead of measuring the pick-up voltage. It is possible to obtain the plenty of
information inside the hardening steel, which mean the permeability, the coercive force
and the residual magnetic flux density. Therefore, we suppose that the measurement of
the B and the H is an effective method to evaluate the hardening depth and the hardness.
In the other institution, the non-destructive evaluation method for measuring the
change of the material has been reported by measuring the B and the H [5]. We try to
develop magnetic sensor to measure the B and the H and we obtain the magnetic
properties depending on the depth by changing the frequency to control the penetration
depth of the magnetic flux. We have examined the hardening depth and the hardness
by measuring the B and the H for the steels of the different hardening temperature. It is
clear that the measured B and the measured H change by the different hardening
temperature. However, we could not clarify the relationship between the measured
magnetic properties and the depth. Because it is difficult to estimate the magnetic
properties depending on depth. In the numerical analysis, it is possible to take into
account the magnetic properties depending on depth. Therefore, we propose the
hardening model, which takes into account the measured magnetic properties
depending on depth.
In this paper, we will examine the non-destructive evaluation method of the
hardening depth and the hardness by using the numerical hardening model. First, the
definition of the hardening model is given by considering the measured magnetic
properties. Then, the B and the H at the surface of the hardening steel are examined by
using the proposed hardening model and the two-dimensional finite-element method.
2. Magnetic Property of Hardening Carbon Steel depending on Depth
Figure 1 shows the measured magnetic properties of the hardening carbon steel
depending on depth. The measured hysteresis loops change as shown in Figure 1. In
particular, the measured hysteresis loops at the surface and the boundary change very
much in comparison with the interior. So, the changes of the measured magnetic
properties have to be clarified depending on the depth by measuring cut-out ring type
specimens. We try to evaluate the hardening depth and the hardness by measuring the
B and the H from the surface. Additionally we propose to obtain information
depending on the depth by changing the frequency.
In the numerical analysis, we try to examine whether the changes of the magnetic
䎔䎔
䎓䎑䎛䎓䎑䎛
䎓䎑䎙䎓䎑䎙
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏏䏘䏛䎃䏇䏈䏑䏖䏌䏗䏜䎃䎥䎃䎾䎷䏀
䎓䎑䎗䎓䎑䎗
䎓䎑䎕䎓䎑䎕
䎓䎓
䎐䎓䎑䎕䎙䎓䎓䎐䎓䎑䎕
䎐䎓䎑䎗䎐䎓䎑䎗䎔
hardness [HV]
䎐䎓䎑䎙䎓䎑䎛
䎘䎓䎓䎐䎓䎑䎙
䎓䎑䎙
䎐䎓䎑䎛
䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎾䎫䎹䏀
䎐䎓䎑䎛
䎐䎔䎐䎔䎓䎑䎗
䎐䎗䎓䎓䎓䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎗䎓䎓䎓䎗䎓䎓䎐䎗䎓䎓䎓䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎗䎓䎓䎓䎓䎑䎕
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎾䎤䎒䏐䏀䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎾䎤䎒䏐䏀
䎓
Surface 䎖䎓䎓 Interior-1 䎐䎓䎑䎕
䎐䎓䎑䎗
䎔䎔
䎐䎓䎑䎙
䎕䎓䎓
Vickers
䎓䎑䎛䎓䎑䎛
䎐䎓䎑䎛
䎓䎑䎙
䎓䎑䎙
䎐䎔
䎓䎑䎗䎓䎑䎗
䎔䎓䎓䎐䎗䎓䎓䎓䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎗䎓䎓䎓
䎓䎑䎕䎓䎑䎕䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎾䎤䎒䏐䏀
䎐䎓䎑䎕
䎓
䎓䎐䎓䎑䎕
䎓
Interior-2
䎓䎕䎗䎙䎛䎔䎓
䎐䎓䎑䎗
䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀䎐䎓䎑䎗
䎐䎓䎑䎙 Depth from Surface [mm] 䎐䎓䎑䎙
䎐䎓䎑䎛䎐䎓䎑䎛
䎐䎔䎐䎔
䎐䎗䎓䎓䎓䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎗䎓䎓䎓䎐䎗䎓䎓䎓䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎗䎓䎓䎓
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎾䎤䎒䏐䏀䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎾䎤䎒䏐䏀
Boundary Interior-3
Figure 1. Magnetic properties of hardening carbon steel depending on depth (Hardening depth : 2 mm).
Y. Kai et al. / Numerical Analysis for Non-Destructive Evaluation of Hardening Steel 227
properties depending the hardening depth and the hardness are obtained by measuring
the B and H from the surface.
3. Expression Method of Measured Magnetic Property
The approximation method of Potter and Schmulian is employed to consider the

measured magnetic properties [6]. The relationships between magnetization M and
field H are approximated by using the hyperbolic curve function. As an advantage of
this method, the hysteresis loop is expressed by giving the coercive force, the residual
magnetization and the saturation magnetization in the measured magnetic property.
The relationship between H and M is given by the following equation,
ª ° § H H sgn D · ½°º
M M S «sgn D D ®1 tanh ¨ C tanh 1 S ¸ ¾» , (1)
«¬ °¯ © HC ¹ °¿»¼
where Ms is the saturation magnetization of the major loop, Hc is the coercive force of
the major loop, S is the squareness ratio, D is the curve parameter, respectively. The
range of Dis -1<D<1. If D isr the major loop is expressed by the equation(1).
Additionally, it is possible to express magnetic properties of the minor loop by
recalculating the D. The equation for the recalculation of the D is written as follows,
ª ° § H C H m sgn D · °½º ° § H C H m sgn D · °½

D ' « 2sgn D D ®1 tanh ¨ tanh 1 S ¸ ¾» ®1 tanh ¨ tanh 1 S ¸ ¾
«¬ °¯ © HC ¹ °¿»¼ °¯ © HC ¹ °¿
(2)
where D’ is the recalculated D.
4. Formulation for Numerical Analysis
Figure 2 shows the numerical analysis model of the investigation. In the case of the
actual measurement, the magnetic properties of the hardening steel are measured under
the condition of three-dimension as shown Figure 2. In this paper, we suppose that the
two-dimensional numerical analysis can obtain enough result from the view point of
our purpose, which consider the magnetic properties depending depth and evaluate the
B and the H from the surface.
The two-dimensional governing equation including the magnetization and eddy
current is written as follows,
§ w2 A w2 A · § wM y wM x ·
Q¨ 2
2 ¸ Q 0 ¨ ¸ J0 Je 0, (3)
© wx wy ¹ © wx wy ¹
wA wI ½
Je V ® ¾ , (4)
¯ wt wz ¿
Figure 2. Numerical analysis model.

where A is the z-direction component of the magnetic vector potential, I is the electric
scalar potential, Mx, My are the x, y-direction component of the magnetization, J is the
exciting current density, Je is the eddy current densty, Iz is the z-direction
component of the gradI, V is the conductivity, Q is the magnetic reluctivity and Q is
the vacuum reluctivity, respectively. The terminal voltage method is applied to
calculate the voltage of the excitation coil and the B coil. In this case J is written as
follows,
WI
J0 , (5)
SW
where W is the winding number of coil, I is the current of the coil and Sw is the section
area of coil, respectively. To treat the unknown current value, the second Kirchhoff’s
law of the is written as follows,
d
V , (6)
dt ³C
Ad" RI
where R is the resistance of the coil, V is the applied voltage and C is the path of
integration through the winding of coil, respectively.
Next, the law of the conservation of electric charge is defined as follows,
³ J e dS 0. (7)
Sc
Equation (4) is substituted into the equation (7), which yields,
wI 1 wA
0, (8)
wz Sc ³ S c wt
dS
where Sc is the section area of conductor.

In the numerical analysis, A is calculated by using the equations (3), (6) and (8).
Additionally, the non-linear calculation of the measured magnetic properties is
considered by using the Newton-Raphson method.
Table 1. Numerical analysis condition.

Number of node 9225
Number of element 18056
Frequencyẅ[Hz] 10,20,25,50
Winding number of 37
exciting coil [turns]
Winding number of 10
B coil [turns]
Conductivity [S/m] 7 ᶣ 106
Figure 3. Structure of magnetic sensor and hardening model.
䎙䎓䎓䎃䎔䎛䎓䎓䎙䎓䎓䎃䎔䎑䎕䎙䎓䎓䎃䎔䎑䎘
䎶䏄䏗䏘䏕䏄䏗䏌䏒䏑䎃䏐䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䎃䎃䎾䎷䏀
䎘䎓䎓䎔䎘䎓䎓䎘䎓䎓䎔䎘䎓䎓䎔䎑䎕䎘
䎵䏈䏖䏌䏇䏘䏄䏏䎃䏐䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䎃䎃䎾䎷䏀
䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎃䎃䎾䎫䎹䏀
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䎃䎃䎾䎤䎒䏐䏀
䏖
䏕
䎗䎓䎓䎔䎕䎓䎓䎗䎓䎓䎓䎑䎛䎗䎓䎓䎔
䏆
䎖䎓䎓䎜䎓䎓䎖䎓䎓䎓䎑䎙䎖䎓䎓䎓䎑䎚䎘
䎕䎓䎓䎙䎓䎓䎕䎓䎓䎓䎑䎗䎕䎓䎓䎓䎑䎘
䎔䎓䎓䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎖䎓䎓䎔䎓䎓䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎓䎑䎕䎔䎓䎓䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎓䎑䎕䎘
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䏆䎵䏈䏖䏌䏇䏘䏄䏏䎃䎰䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䏕䎶䏄䏗䏘䏕䏄䏗䏌䏒䏑䎃䎰䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䏖

䎓䎃䎓䎓䎃䎓䎓䎃䎓
䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓
䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀
(a) Hardening depth : 2 mm(H.D.2, Measurement data)

䎙䎓䎓䎃䎔䎛䎓䎓䎙䎓䎓䎃䎔䎑䎕䎙䎓䎓䎃䎔䎑䎘
䎶䏄䏗䏘䏕䏄䏗䏌䏒䏑䎃䏐䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䎃䎃䎾䎷䏀
䎘䎓䎓䎔䎘䎓䎓䎘䎓䎓䎔䎘䎓䎓䎔䎑䎕䎘
䎵䏈䏖䏌䏇䏘䏄䏏䎃䏐䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䎃䎃䎾䎷䏀
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䎃䎃䎾䎤䎒䏐䏀
䏖
䏕
䎗䎓䎓䎔䎕䎓䎓䎗䎓䎓䎓䎑䎛䎗䎓䎓䎔

䏆
䎖䎓䎓䎜䎓䎓䎖䎓䎓䎓䎑䎙䎖䎓䎓䎓䎑䎚䎘
䎕䎓䎓䎙䎓䎓䎕䎓䎓䎓䎑䎗䎕䎓䎓䎓䎑䎘
䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙
䎔䎓䎓䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎖䎓䎓䎔䎓䎓䎹䏌䏆䏎䏈䏕䏖䎃䏋䏄䏕䏇䏑䏈䏖䏖䎃䎫䏙䎓䎑䎕䎔䎓䎓䎓䎑䎕䎘
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䏆䎵䏈䏖䏌䏇䏘䏄䏏䎃䎰䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䏕䎶䏄䏗䏘䏕䏄䏗䏌䏒䏑䎃䎰䏄䏊䏑䏈䏗䏌䏝䏄䏗䏌䏒䏑䎃䎰䏖

䎓䎃䎓䎓䎃䎓䎓䎃䎓
䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓䎓䎔䎕䎖䎗䎘䎙䎚䎛䎜䎔䎓
䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀䎧䏈䏓䏗䏋䎃䏉䏕䏒䏐䎃䏖䏘䏕䏉䏄䏆䏈䎃䎾䏐䏐䏀
(b) Hardening depth : 4 mm(H.D.4)

Figure 4. Magnetic property of difference hardening depth.
5. Hardening Model and Numerical Analysis Condition
Figure 3 shows the structure of magnetic sensor and the hardening model. The
magnetic sensor is set on the surface of the hardening steel. The specimen is excited by
the exciting coil, and the closed magnetic path is made between the magnetic sensor
and the specimen. In the numerical analysis, B in the B coil is controlled to be B = 0.3
T. H is calculated by using the continuous condition of the tangential component of the
H. Because the H inside the steel and the H of the near-surface of the steel is
approximately equal. The hysteresis loop is calculated by the relationship between the
calculated B and the calculated H. The effective magnetic properties to evaluate the
hardening depth and the hardness are examined by using the calculated hysteresis loops.
The Hc, the Mr and the Ms of the measured magnetic properties are gave for every 1-
mm depth to express the different hardening depth and the hardness as shown in Figure
3. Figure 4 shows the Hc, the Mr and the Ms depending on the depth. The Hc, the Mr
and the Ms of the hardening depth 4 mm are created on the base of the measured
magnetic property of the hardening depth 2 mm. Table 1 shows the analysis condition.
The winding number of the excitation coil is set 37 turns, and that of the B coil is set 10
turns. In this numerical analysis, the conductor is treated as constant. We try to obtain
the change of the magnetic properties depending on the depth by changing the
N.H.D. H.D.2 H.D.4

(a) f =10 Hz
N.H.D. H.D.2 H.D.4

(b) f =100 Hz
Figure 5. Distribution of magnetic flux line depending on depth.
䎓䎑䎗䎃
䎓䎑䎗䎃
䎓䎑䎖䎓䎑䎖
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏏䏘䏛䎃䏇䏈䏑䏖䏌䏗䏜䎃䎥䎃䎃䎾䎷䏀
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏏䏘䏛䎃䏇䏈䏑䏖䏌䏗䏜䎃䎥䎃䎃䎾䎷䏀
䎓䎑䎕䎓䎑䎕
䎓䎑䎔䎓䎑䎔
䎓䎓
䎐䎓䎑䎔䎐䎓䎑䎔
䎐䎓䎑䎕䎐䎓䎑䎕
䎱䎑䎫䎑䎧䎱䎑䎫䎑䎧
䎐䎓䎑䎖䎫䎑䎧䎑䎕䎐䎓䎑䎖䎫䎑䎧䎑䎕
䎫䎑䎧䎑䎗䎫䎑䎧䎑䎗
䎐䎓䎑䎗䎃䎐䎓䎑䎗䎃
䎐䎖䎓䎓䎓䎐䎕䎓䎓䎓䎐䎔䎓䎓䎓䎓䎔䎓䎓䎓䎕䎓䎓䎓䎖䎓䎓䎓䎐䎛䎓䎓䎓䎐䎙䎓䎓䎓䎐䎗䎓䎓䎓䎐䎕䎓䎓䎓䎓䎕䎓䎓䎓䎗䎓䎓䎓䎙䎓䎓䎓䎛䎓䎓䎓
䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎃䎾䎤䎒䏐䏀䎰䏄䏊䏑䏈䏗䏌䏆䎃䏉䏌䏈䏏䏇䎃䎫䎃䎃䎾䎤䎒䏐䏀
(a) frequency f = 10 Hz (b) frequency f = 100 Hz
Figure 6. Calculated hysteresis loop depending on hardening depth and hardness
frequency as shown in Table 1.
6. Numerical Analysis Results
Figure 5 shows the distributions of the magnetic flux. The distributions of the
magnetic flux differ from each other depending on depth and frequency. The magnetic
flux at f = 10 Hz penetrates deeply in comparison with the one at f = 100 Hz because
of the skin effect. At the same frequency, it is clear that the penetration of the magnetic
flux changes depending on the different hardening depth. As the hardening is deep, the
magnetic flux penetrates deeply. In the case of high frequency, it is difficult to obtain
the difference of the H.D.2 and the H.D.4 because the magnetic flux concentrates at the
surface of the hardening steel. However, it is impossible to measure the change of
magnetic flux line. Therefore, we try to evaluate the hardening depth and the hardness
by calculating the B and the H from the surface. The correlation of the hardening depth
and the hardness are examined by calculating the hysteresis loop from the relationship
between the B and the H.
Figure 6 shows the calculated hysteresis loop. At the low frequency, the
calculated hysteresis loop changes depending on the hardening depth. In the case of
the high frequency, the difference of the hysteresis loop of the H.D.2 and the H.D.4 is
䎔䎗䎓䎃䎙䎓䎓䎓䎃
䎱䎑䎫䎑䎧
䎔䎕䎓䎫䎑䎧䎑䎃䎝䎃䎕䎃䏐䏐
䎘䎓䎓䎓
䎵䏈䏏䏄䏗䏌䏙䏈䎃䏓䏈䏕䏐䏈䏄䏅䏌䏏䏌䏗䏜䎃P䏖
䎫䎑䎧䎑䎃䎝䎃䎗䎃䏐䏐
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䏆
䎔䎓䎓
䎗䎓䎓䎓
䎛䎓
䎖䎓䎓䎓
䎙䎓
䎕䎓䎓䎓
䎗䎓
䎔䎓䎓䎓䎱䎑䎫䎑䎧
䎕䎓䎫䎑䎧䎑䎃䎝䎃䎕䎃䏐䏐
䎫䎑䎧䎑䎃䎝䎃䎗䎃䏐䏐
䎓䎃䎓䎃
䎓䎕䎓䎗䎓䎙䎓䎛䎓䎔䎓䎓䎓䎕䎓䎗䎓䎙䎓䎛䎓䎔䎓䎓
䎩䏕䏈䏔䏘䏈䏑䏆䏜䎃䏉䎃䎃䎾䎫䏝䏀䎩䏕䏈䏔䏘䏈䏑䏆䏜䎃䏉䎃䎃䎾䎫䏝䏀
(a) Relative permeability (b) Coercive force
Figure 7. Magnetic property v.s. frequency depending on the hardening depth and hardness
䎔䎕䎓䎓
䎦䏒䏈䏕䏆䏌䏙䏈䎃䏉䏒䏕䏆䏈䎃䎫䏆䎃䎾䎤䎒䏐䏀
䎔䎔䎓䎓
䎔䎓䎓䎓
䎜䎓䎓
䎛䎓䎓
䎚䎓䎓
䎙䎓䎓
䎓䎔䎕䎖䎗䎘
䎫䏄䏕䏇䏈䏑䏌䏑䏊䎃䏇䏈䏓䏗䏋䎃䎃䎾䏐䏐䏀
Figure 8. Relationship between hardening depth and coercive force.
small because the magnetic flux concentrates on the surface of the hardening steel.
Next, the magnetic properties depending on the hardening depth and the hardness are
examined in detail by using the calculated hysteresis.
Figure 7 shows the magnetic property v.s. the frequency. Figure 7(a) shows the
relative permeability Ps v.s. frequency. Ps is calculated by using the maximum
magnetic flux density Bm and the maximum magnetic field strength Hm. At the low
frequency, the difference of Ps is large since the magnetic flux penetrates deeply inside
the hardening steel. Figure 7(b) shows Hc v.s. frequency. At the low frequency, the
difference of the Hc becomes large depending on the hardening depth and the hardness.
On the other hand, the difference of the H.D.2 and the H.D.4 becomes small by
increasing the frequency because the magnetic flux density becomes high near the
surface of the hardening steel. Therefore, it is very difficult to evaluate the hardening
depth and hardness. In these results, the difference of the magnetic properties
depending on the hardening depth and hardness is larger when the frequency is lower.
We try to estimate the coercive force under the D.C. field from the frequency
characteristics by noting the change of the Hc. As the estimation method, the Hc under
the D.C. field is calculated by using a polynomial approximation as shown in Figure
7(b). Figure 8 shows the hardening depth v.s. the estimated Hc. It is clear that the
estimated Hc increases depending on the depth since the high area of the Hc become
large by increasing the hardening depth. As mentioned previously, it is possible to
evaluate the hardening depth and the hardness by estimating the Hc under the D.C. field.
In this hardening model, it is clear that the calculated magnetic properties between
the every hardening steel do not change v.s. frequency since the measured magnetic
properties under the D.C field are used for the hardening model. However, we suppose
that the estimation method of the magnetic properties under D.C. field is an effective
method to evaluate the hardening depth and the hardness. The non-destructive
evaluation of the hardening depth and the hardness enables by measuring the B and the
H on the surface of the hardening steel.
7. Conclusions
In this paper, we try to evaluate the hardening depth and the hardness by measuring the
B and the H from the surface. The magnetic properties depending on the depth are
considered by using the measured magnetic properties of the cut-out ring specimens in
the numerical analysis. The conclusions are as follows,
(1) In the numerical analysis, the measured magnetic properties of the cut-out ring
specimens are used to consider the magnetic property depending on depth. It is
clear that the distributions of the magnetic flux differ from each other depending
on hardening depth and hardness.
(2) The magnetic properties at the different hardening depth are evaluated by
calculating the B from the B coil and the H on the surface of the hardening steel.
The calculated hysteresis loops change depending on hardening depth and
hardness. The Ps and the Hc depending on hardening depth and hardness are
examined by using the calculated hysteresis. The correlation can be obtained by
estimating the Hc under the D.C. field. As the results, it is possible to evaluate the
hardening depth and the hardness by using the estimation method of the magnetic
property under the D.C. field.
The non-destructive evaluation method in this paper will be applied into the steel of the
other heat treatment as well as that of the difference hardening depth and hardness.
References
[1] Yuichiro KAI, Yuji TSUCHIDA and Masato Enokizono, Magnetic Properties of Hardened Carbon Steel
by Cutout Ring Type Specimen, Journal of the Japan Society of Applied Electromagnetics and
Mechanics, Vol.14, No.1 (2006), 145-150.
[2] Marcus JOHNSON, Chester LO, Scott HENTSCHER and Emily KINSER, Analysis of Conductivity and
Permeability Profiles in Hardness Steel, Studies in Applied Electromagnetics and Mechanics,
Electromagnetic Nondestructive Evaluation (IX), IOS Press, (2005), 135-142.
[3] Y. Kai, Y. Tsuchida, and M. Enokizono, Non-destructive Evaluating of Case Hardening by Measuring
Magnetic Properties, Studies in Applied Electromagnetics and Mechanics, Electromagnetic
Nondestructive Evaluation (IX), IOS Press, (2005), 143-150.
[4] Y. Kai, Y. Tsuchida, M. Enokizono and S. Nagata, Evaluation of Case Hardening Depth by the MFES
Method, Studies in Applied Electromagnetics and Mechanics 24, Electromagnetic Nondestructive
Evaluation (VIII), IOS Press, (2004), 153-158.
[5] Hiroaki Kikuchi, Akinori Takahashi, Lefu Zhang, Katsuyuki Ara, Yasuhiro Kamada, and Seiki
Takahashi, NDE for Magnetic Material by Minor Loop Method Using Magnetic Yoke Probe, Studies in
Applied Electromagnetics and Mechanics 24, Electromagnetic Nondestructive Evaluation (IX), IOS
Press, (2005), 119-125.
[6] ROBERT I. POTTER and ROBERT J. SCHMULIAN, Self-Consistently Computed Magnetization
Patterns in Thin Magnetic Recording Media, IEEE TRANSACTION ON MAGNETICS, VOL. MAG-7,
NO. 4, (1997).
IOS Press, 2007
Development of Metal Detection System

for Reuse of Dismantled Wood from
Houses
Tomoharu YASUTAKEa, 1, Tomasz CHADYb,
Yuji TSUCHIDAa and Masato ENOKIZONOa
a
Faculty of Engineering, Oita University, 700 Dannoharu, Oita, 870-1192, Japan
b
Szczecin University of Technology, Department of Electrical Engineering, Poland
Abstract. This paper describes the development of a metal detection system that is
able to identify the exact position and the shape of metal pieces embedded in
lumber dismantled from houses. A high sensitivity transducer consisting of several
differentially connected pick-up coils is developed for the detection of metal
pieces. Measurements showing the ability of the system to detect metal pieces are
prosecuted. Effects to improve the performance using a three-dimensional
boundary element model are discussed.
Keywords. Differential coil, detection of metallic pieces, dismantled woods from

houses
1. Introduction
A lot of used wood is discarded when houses are dismantled in Japan. Lumber can be
recycled from the dismantled wood from houses by bonding pieces of the used wood
after cutting them to pieces. Recycled lumber bonded from dismantled wood compare
favorably with new timber because of the high strength and stiffness characteristics.
However, metallic pieces such as nails are often contained in the dismantled wood. In
this case, an expensive blade of a cutting machine can be damaged during the process
of cutting the dismantled wood. Therefore, the dismantled wood cannot be used for
recycling unless all of the metallic pieces are removed. At present, metallic pieces in
the dismantled wood can be found by using a conventional metal detector. However, it
is difficult to identify the exact position and shape of the metal pieces by the metal
detectors. Therefore, an excessive amount of the dismantled wood is cut off including
the metal detection area, thereby discouraging the recycling of dismantled wood. In the
current situation, only a small amount of the dismantled wood is recycled for wooden
boards, green materials and so on, and not for other types of lumber even though it is
worth doing so. Almost all of the dismantled wood from houses is burnt now. If the
system that is able to detect the exact position of metallic pieces is developed, we can
reuse a lot of the dismantled wood[1].
In this paper, we propose a highly sensitive metal detection system consisting of
several differential coils and differential amplifiers. Variations in the differential
1
Corresponding Author: Tomoharu Yasutake, Graduate School of Engineering, Oita University, 700
Dannoharu, Oita 870-1192, Japan; E-mail: yasutake@mag.eee.oita-u.ac.jp
234 T. Yasutake et al. / Development of Metal Detection System for Reuse of Dismantled Wood
voltage are examined by changing the nail’s position and angle. However, several
problems were identified following the construction of a prototype unit. Therefore,
numerical models employing the three-dimensional boundary element method were
used to improve and optimize the design.
2. Measurement System
A prototype unit consisting of several differential coils was developed to detect

metallic pieces in wood. Fig. 1 shows the schematic view of the detection system.
The excitation coil was set up at the center. The pick up coils were set up in symmetric
positions with respect to the reference coils. The pick-up and reference coils are each
connected differentially. The measuring area lies between the excitation coil and the
pick-up coils as shown in Fig. 1. No voltage is obtained by the pairs of the coils if there
are no metallic pieces in the measuring area because they are connected differentially.
The coils generate a signal even if there are small metallic pieces in the measuring area.
The metallic piece’s position and shape are identified by using the variations of the
differential voltage because they vary depending on the position of the metallic pieces.
Fig. 2 shows the proposed measurement system. The voltage is applied to the
excitation coil by a signal generator through a power amplifier as shown in Fig. 1. The
excitation coil has 1500 turns, and the diameter of the excitation coil wire is 1.0 mm.
All the pick-up coils and the reference coils are identical each containing 5000 turns
and the wire diameter 0.042 mm. The excitation frequency of the signal generator is
selected as 250 Hz. The picked-up voltages from the differential coils are amplified by
differential amplifiers, and filtered using low pass filters. The differential voltages from
the five coils are transferred to a computer after being digitized by an A/D converter[2].
A small nail was selected a metallic piece in the wood. The length of the nail is
31.5mm, and the diameter is 1.88 mm. The measurement region is from x = -50 mm to
50 mm, from y = 0 mm to y = 100 mm and from z = 0 mm to z = 40 mm as shown in
Fig. 1. The zero point of the each axis is at the center of the excitation coil as shown in
Fig. 1. The angle of the nail is also varied from 0° to 90° every 45° toward the y -
direction to simulate various conditions of the metallic pieces.
Exciting coil
Power Function
Pick-up amplifier generator
coils
Output
S1 S1’
S2 S2’
S3 S3’ Synchronizing
Reference signal
S4 S4’
coils
S5 S5’
Differential
amplifier and
BPF A/D Personal
converter computer

Figure 1. Schematic view of detection system Figure 2. Measurement system
T. Yasutake et al. / Development of Metal Detection System for Reuse of Dismantled Wood 235
3. Results and Discussions
The variation of the differential signal as a function of positions and angles of the nail
are shown in Fig.3, Fig.4 and Fig. 5. Fig. 3 shows the variation of the signal as a
function of the nail position. As shown in Fig. 3, we can confirm that the signal
amplitude increased as the nail approaches the pick-up coils. This is because the nail
disturbs the magnetic field when it is close to the pick- up coils. Fig. 4 shows the
variation in the signal amplitude as a function of the nail’s position in the Z - direction.
It is also clear from Fig. 4 that the differential signal become larger as the position of
the nail moves lower. As shown in Fig. 5, the variations of the differential signals
become small when the angle of the nail is 90 degrees toward the y - axis. As the
results show, we can measure the different differential signals depending on the
positions and the angles of the nail. As can be found from the results in Fig. 3 and Fig.
4, if the angles of the nail are known, we can identify the position of the nail
approximately. Large variations of the differential signal indicate that the nail is near
the pick-up coils. As shown in Fig. 5, it is also possible to identify the angles of the nail
from the variations of the differential signals if the position of the nail is known.
For the next step, we propose using a neural network to identify the position and
the angle of the nail automatically. At first, we trained the neural network using
Figure 3. Variations of differential voltages. ( Z=0mm and angle of nail is 0 degree )
Figure 4. Variations of differential voltages. ( Y=0mm and angle of nail is 0 degree )

Figure 5. Variations of differential voltages. ( Z=0mm and Y=0mm )
Table. 1 Identified results from trained neural network

Y-axis [mm] Z-axis [mm]
Nail’s angle Objective Identified Objective Identified
values results values results
0q 100 100.20 0 12.09
0q 100 101.66 20 19.39
0q 55 58.28 0 0.03
90q 100 103.21 0 2.25
90q 100 101.51 20 17.98
90q 55 75.04 0 6.23
data which was measured by changing the positions and the angles of the nail. So far
the trained neural network can recognize the position if that data was used as training
data. After training the neural network, new measured data was applied to the trained
neural network. Table 1 shows examples of results obtained from the trained neural
network. As shown in Table. 1, we can confirm that the trained neural network
sometimes gives the position with some error depending on the position of the nail.
Two issues relating to the current system are apparent. The first one is that the data
used to train the neural network is not sufficiently diverse due to the symmetry of the
positions and the angles of the nail. The second issue is that all the variations of the
differential voltages become very small when the nail moves to certain positions and
angles; so getting repeatable results is not so easy. Therefore, the neural network must
to be trained with data that avoids the above two circumstances.
When the nail was moved in the measuring area, all pick-up voltages vary
similarly because the coils are set up in the two dimensional plane. If the angles of the
nail are vertical to the magnetic flux, the variations of the magnetic flux become small.
Therefore, it is difficult to detect the nail by the differential coils. We conclude that the
coils should be placed in a three-dimensional setting. It is assumed that the variations
of the pick-up voltages by allocating the coils in the three-dimension offer much
advantage compared with the coils allocated in the two-dimension. It is also seen that
we can detect the nail by locating the coils in the direction perpendicular to the
magnetic flux in the case that the variations of the magnetic flex are small. Then, we
proposed and evaluated the second system by using the three-dimensional boundary
element method.
ᵣᶖᶁᶇᶒᵿᶒᶇᶍᶌᴾᶁᶍᶇᶊ
ᵱᵓ’
ᵱᵓ
ᵱᵒ’
ᵱᵒ
ᵱᵏ’ ᵫᵿᶅᶌᶃᶒᶇᶁᴾᶋᵿᶒᶃᶐᶇᵿᶊ
ᵱᵐ’ ᵆᵓᵎᵊᵎᵇ ᵫᶍᶔᶇᶌᶅ
ᵱᵑ’ ᵆᵓᵎᵊᵏᵎᵎᵇ
ᵱᵑ
ᵆᵋᵓᵎᵊᵎᵇ ᵱᵐ
ᵆᵋᵓᵎᵊᵏᵎᵎᵇ ᵱᵏ
ᵸ ᵶ
ᵷ
Figure 6. Three-dimensional setting of the pair coils
4. Evaluation by Using Three-dimensional Boundary Element Method
Fig. 6 shows the second proposed system consisting of coils set in the three-dimension.
In order to examine and optimize the performance of the second system shown in Fig.
6, a three dimensional numerical simulation is carried out. We use the boundary
element method to model the geometry. The finite element method is time consuming
to model our system. From these advantages, we evaluate our second system using
the three-dimensional boundary element method.
4.1. Formulation of Three-dimensional Boundary Element Method
We consider the electromagnetic static case as shown in Fig. 6. The internal area of the
magnetic material is shown by V1 and the external area including the coils is shown by
V2. By applying the Green’s theorem, a boundary integral equation can be obtained for
the analytical region shown in Fig. 6.
C fi A fi ³ G n A

f ds ³ G
u n u A f ds ³ G Q f ds , in V1 (1)
Sf Sf Sf
Cai A ai ³
Sa
G n A ds ³ G

a
Sa

u n u A a ds ³
Sa
G Q a ds ³ P J 0 G dV
V
, in V2 (2)
A, J0 and P are the magnetic vector potential, the magnetizing current density, and
permeability, respectively. G* the Green’s function of the three-dimension Laplace
equation, n the unit normal vector on the boundary. Q is written by,
Q u A u n , (3)
Q is the tangential component of the magnetic flux density at unknown boundary

points. (1) and (2) can be expressed in the matrix form as,
> H @Â f ` >G @^Q f ` , in V1 (4)
> H @^ A a ` >G @^Qa ` ^P` , in V2 (5)

[H] and [G] are the coefficient matrices and {P} is the column matrix related to the
current. From the continuity condition of the magnetic vector potential on the boundary
and the tangential component of the magnetic flux density, the coupled condition is
expressed as follows,
P1 1
Â ` Â ` , ^Q `
f a f ^Qa ` ^Qa ` , (6)
P2 D
μ1 and μ2 are the magnetic permeability in the V1 and V2. The matrix to solve is
expressed as follows,
ª > H @ >G @ º °Â f `½° 0 ½. (7)

« »® ¾ ® ¾
¬ > H @ D >G @¼ ¯°^Q f `¿° ¯^P`¿
By calculating of (7), we are able to get the {A} and {Q}. And, the following equation
(8) can calculate the magnetic flux density of a given point i in the analytical area[3].
Bi rotA i (8)
4.2. Numerical Analysis Results
Fig. 7 shows the numerical analysis model. Details of the numerical model are shown
in Table 2. The current density of the excitation coil is 0.7×106 [A/m2], and the relative
permeability of the magnetic material is 1000. The magnetic flux density that is
calculated by (8) allows calculation of the induced voltages in the position of the
pick-up coils and reference coils. The position of the magnetic material (assumed as a
metallic piece) was moved every 4 [mm] to the X-axis direction, and the induced
voltages are calculated. The area of the magnetic material is as same as the previous
measurement area. The induced voltages in all the coils are identical at the position of
symmetry.
Figure 7. Analysis model
Table 2. Numerical analysis conditions

Node 386
Magnetic material Element 768
Magnetic permeability 10004Ǹ10-7
Node 144
Element 720
Excitation coil
Magnetic permeability 4Ǹ10-7
Current density 0.7106 [A/m2]
Fig. 8 shows voltage as the magnetic material is rotated toward the y - direction
and the moving in the Y-direction with Z=0 mm. As shown in Fig. 8, we confirm that
the variations of the differential voltages from Sensor1, Sensor2 and Sensor3 become
larger as the magnetic material approaches the pick-up coils. The variations of the
differential voltages from Sensor4 become small when Y=60 mm. In Sensor5, the
variations of the differential voltage become larger as the magnetic material approaches
the excitation coil.
Fig. 9 shows the effect of the height of the moving magnetic material in the case
where the angles of the magnetic material are 0 degree toward the y-direction and move
in the Z-direction with Y=0 mm. It is also clear from Fig. 9 that the variations of the
differential voltage from Sensor1, Sensor2 and Sensor3 become lower as the magnetic
material’s position becomes higher. On the other hand, in the case of Sensor4 and
Sensor5, the variations of the differential signal become larger as the height of the
moving magnetic material move higher.
Fig. 10 shows the effect of the magnetic material’s angle in the case where magnetic
material’s position is Y = 100 mm and Z = 0 mm. In the case of Sensor1, Sensor2 and
Sensor3, the variations of the differential voltages become smaller when the angles of
the magnetic material are 90 degrees toward the y-axis. And, in Sensor4 and Sensor5,
the differential voltages did not vary as much when the magnetic material was orientate
at 90 degrees.
Figure 8. Variations of differential voltages. ( Z=0mm and angle of nail is 0 degree )
Figure 9. Variations of differential voltages. ( Y=0mm and angle of nail is 0 degree )

Figure 10. Variations of differential voltages. ( Z=0mm and Y=0mm )
As can be seen from Fig. 8 to Fig. 10, we got much more variations in the
measured differential voltages relative to the previous prototype system shown in Fig. 1.
The variations of the differential voltages in the previous system shown in Fig. 1 were
small in some positions and angles of the nail. Moreover, the characteristics of the
pick-up signals are much more different depending on the magnetic material’s
positions and angles. It can be said that the second system is offers higher performance
in regard to the position of the metallic piece and angles due to the use of the neural
network. In future, we will optimize the location of the coils using a suitable numerical
analysis tool.
5. Conclusions
In this paper, we describe a prototype system developed for detecting nails embedded
in wood. A three-dimensional boundary element model was employed to improve the
design. The numerical analysis proved that the coils located in the third dimension
were useful for identification of the metallic piece’s positions. In the future, we will
improve the design even further using numerical analysis tools.
References
[1] T. Yasutake, T. Chady, Y. Tsuchida, and M. Enokizono, Study on an exploratory system for metal pieces
using differential coils, Papers of Techical Meeting on Magnetics, IEE Japan, MAG-05-174~196,
pp.143-150, 2005.
[2] T. Chady, M. Enokizono, Y. Tsucida, and T. Yasutake, Identification of Three-dimensional Distribution
of Metal Particles Using Electromagnetic Tomography System, Proc, vol. 4, 4th Japanese-Mediterranean
Workshop on Applied Electromagnetic Engineering for Magnetic, Superconducting and Nano Materials,
pp.123-124, 2005.
[3] M. Enokizono, Boundary Element Method Taking Into Account External Power Source, IEEE
TRANSACTIONS ON MAGNETICS, Vol. 24, No. 4, July 1988.
Inverse Problem and Benchmark
IOS Press, 2007
Nondestructive Evaluation for Material

Degradation of Steel Sample using Minor
Hysteresis Loop Observations
1
Fumio KOJIMA and Ryou NISHIYAMA
Graduate School of Science and Technology, Kobe University
Abstract. New assessment strategies involving the analysis of minor hysteresis loop
for characterizing degradation of magnetic materials are presented. Alternating cur-
rents at low frequencies are applied to a probe with magnetic yoke and the response
can be measured using a pick-up coil attached to the probe. The material degra-
dation can be characterized by a set of curvatures of minor hysteresis loops. An
inspection model is described by a nonlinear Euler equation in conjunction with
unknown parameters of the curvatures. An efficient inverse scheme is proposed for
identifying a magnetic parameter related to material degradation. The effectiveness
of the proposed method is shown by simulation experiments.
1. Introduction
Assessments of material degradation of nuclear power plants are critical for lifetime
elongation and safe operation. During operation, structural materials in these plants suf-
fer damage due to high stress, heat and neutron radiation. The changes of microstruc-
ture cause change in mechanical properties, such as hardness, yield stress, etc. The mi-
crostructural chages also affects magnetic properties and the measurement of the associ-
ated magnetic parameters provide a basis for monitoring degradation levels before gen-
erating macroscopic cracking. It is well known that the mechanism caused by the strain
field due to dislocations influences the magnetization curve through magneto-elastic cou-
pling. Taking into account this fact, our previous efforts were directed to the inverse anal-
ysis of the so-called “c” parameter related to the magnetization curve [1,2]. Recently,
Takahashi and his co-authors [3] have shown that changes in the minor hysteresis loops
of ferromagnetic materials are related to material degradation. These observations can
be used for the on-site inspection of power plants since minor loop observations do not
require high magnetization excitation. In this paper, we propose a computational method
for estimating material sensitive magnetic parameters by casting it as a nonlinear elec-
tromagnetic inverse problem.
1 CorrespondingAuthor: Fumio Kojima, Graduate School of Science and Technology, Kobe University, 1-1
Rokkodai-cho, Nada-ku, Kobe, 657-8501 Japan; E-mail: kojima@cs.kobe-u.ac.jp
244 F. Kojima and R. Nishiyama / Nondestructive Evaluation for Material Degradation of Steel Sample
1.5
B χa*
Ba*
1
χrev*
χR*
Magnetic flux density B [T]
0.5
+BR*
0
1
Ha Ha Ha3
2 4
Ha WR* χ*
H
Hc*
-0.5 0 +Hc* +Ha
H
-1 WF*
-1.5
-1500 -1000 -500 0 500 1000 1500
-BR*
Magnetic field H [A/m]
Figure 1. Set of minor hysteresis loops. Figure 2. Set of parameters for each minor hys-
teresis loop.
2. Minor Loop and Degradation Parameter Set
A set of minor loops can be constructed by changing the maximum applied magnetic
field. Figure 1 depicts a set of hysteresis curves obtained by using this procedure. Each
minor loop is characterized by a set of parameters related to the major hysteresis loop.
Let Hc∗ and χ∗H be pseudo-coercive force at each minor loop and magnetic permeabil-
ity at H = Hc∗ . Given the maximum magnetic field H a AC ˛ let χ∗a , χ∗rev and Ba∗ be the
permeability at H = Ha during magnetization and for demagnetization, and the mag-
∗
netic flux density at H = H a , respectively. Let B R be the magnetic residual flux den-
∗
sity at each minor loop and let χ R be the magnetic permeability at H = 0. Furthermore
we define the pseudo-hysteresis loss and the pseudo-residual flux loss by W F∗ and WR∗ ,
respectively. Figure 2 illustrates the set of parameters for each minor hysteresis loop.
It was experimentally found that those parameters are quite sensitive with respect
to material degradation of magnetic samples [4]. From these experimental findings, the
maximum applied magnetic field is divided into three sub-intervals as listed in Table
1. Each minor loop curve can be characterized by a set of parameters associated with
material degradation of sample materials. Our objective is to identify magnetic sensitive
parameters
q = {Ba , Hc , BR , χR , χH , χa , χrev } (1)
through magnetization and demagnetization procedures at low, middle, and high regions
of the applied magnetic field. To characterize the degradation map using the estimated
parameter vector q, we define

Ba∗ − Ba− χ∗a − χ+
a
Deg = 0.5 + (2)
Ba+ − Ba− a − χa
χ+ −
where Ba∗ and χ∗a denote the magnetic flux density and magnetic permeability at the
maximum applied magnetic field H a for samples to be evaluated, while B a− ,χ− +
a ,Ba , and
+
χa imply the corresponding parameters for the reference material with low degradation
(−) and for the another reference material with high level degradation (+), respectively.
F. Kojima and R. Nishiyama / Nondestructive Evaluation for Material Degradation of Steel Sample 245
1.5
Pseudo magnetization Ba
Magnetic flux density Ba*

Magnetic susceptibility ǿ
Maximum magnetic field Ha
1
Table 1. Range of maximum magnetic field.
Range Ha− [A/m] Ha+ [A/m]

0.5
Low 720 830
Middle 900 1010
High 1080 1190 0
0 500 1000 1500
Figure 3. Positions of Ba∗ and χ∗a .
3. Mathematical Model of Inspection
In this section, the mathematical modeling of our inspection is considered in accordance

with the set of minor hysteresis loops. Let Ω be the sufficiently large domain related to
our inspection in two dimensions,
Ω = {x = (x1 , x2 )|w− ≤ x1 ≤ w+ , h− ≤ x2 ≤ h+ }.
Let ht be the thickness of the material sample to be inspected and the sample region is
defined by
Ωm = {x = (x1 , x2 )|w− ≤ x1 ≤ w+ , 0 ≤ x2 ≤ ht } ⊂ Ω.
Thus the inspection area is assumed to be given by
Ωi = {x = (x1 , x2 )|wi− ≤ x1 ≤ wi+ , 0 ≤ x2 ≤ ht } ⊂ Ωm .
The schematic measuring set-up of our inspection is illustrated in Fig. 4. The magnetic
vector potential A in two dimensions is governed by an Euler equation of the form [2]:

∂ ∂A ∂ ∂A
ν + ν = −Jin Ω (3)
∂x1 ∂x1 ∂x2 ∂x2
with the boundary conditions
∂A
= 0 on ΓN A=0 on ΓD (4)
∂n
where
ΓN = {x = (x1 , x2 )|x1 = w± , h− ≤ x2 ≤ h+ }
ΓD = {x = (x1 , x2 )|w− ≤ x2 ≤ w+ , x2 = h± }.
In Eq.(3), ν denotes the magnetic reluctivity given by

x2
Dirichlet boundary 㰰D h+
㱅
Transmitter coil
Neumann Magnetic yoke
boundary
㰰N 㰰N
ht Pick-up coil
㱅m 㱅i
Sample material x1
w- wi - 0 wi + w+
㰰D h-
Figure 4. Inspection procedures and our testing environments.
⎧
⎪
⎨ν(|∇ × A| )
2
in Ωi
1
ν = = νyoke (|∇ × A|2 ) in yoke (5)
μ ⎪
⎩
1/μ0 Otherwise
where νyoke is the magnetic reluctivity of the yoke. Then the corresponding measure-
ments are made by

∂A
Y (ν) = C − (6)
∂x1
where C denoted the effective area of the detecting coil. The minor hysteresis loop is then
represented by
{H(Ji ), B(A(Ji ))}i∈ΩI (7)
where Ji implies a set of the increasing magnetic field amplitudes corresponding to the
minor hysteresis loops at the pick-up coil. To solve the system (3) with (4), the magnetic
minor loop is approximated by series of cubic B-spline functions. Let I H = [Hc∗ , Ha ] be
a domain of magnetic field for a minor loop approximation. Then the knot sequence on
IH of the cubic spline functions is given by
# $
ΔN = H̄− = H̄0N ≤ H̄1N ≤ · · · ≤ H̄N
N
= H̄+ .
Then each hysteresis loop on I H is approximated by

N +1
BΔN (q; H) = αN N N N
i (q; Δ )βi (H; Δ ) (8)
i=−1
where {βiN }N +1
i=−1 denotes the sequence of B-spline function with the knot sequence Δ
N
([5]) and where q is the material sensitive parameters defined by (1). Figure 5 depicts the
1.8 1
1.6 0.9
0.8

1.4
0.7
1.2
0.6
1
0.5
0.8
0.4
0.6
0.3
0.4
0.2
A
Approxim
mattted B-H curvve
m
0.2 pproxima ed B-H cur
mennttal B-H ccurvve
m
Experim 0.1 Ex erime al B-H cur e
0
0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
Magnetic Field H [A/m] Magnetic Field H [A/m]
(a) 0 [MPa] (b) 518 [MPa]

Figure 5. Experimental data and the approximated B-H curves using B-splines.
experimental data and the approximated B-H curves at the low, middle, and high levels
of applied magnetic fields H a . The magnetic reluctivity ν in (3) can be represented as
1 H(J(Ha ))
ν(q, J(Ha ); H) = =
μ B(q; J(Ha ))
⎧
⎪
⎨νΔN (q, J(Ha ), |∇ × A| ; H) in Ωi
2
= νyoke (|∇ × A|2 ) in yoke (9)

⎪
⎩
1/μ0 Otherwise
where J(Ha ) implies applied current density related to the minor loop for the maximum
applied field Ha (|H| ≤ Ha ). Thus the parameter-to-output mapping can be represented
as follows:
{q, J(Ha )} =⇒ {Y (ν(q, J(Ha ), ΔN )}
Figure 6 illustrates the examples of the magnetic reluctivities with the laboratory exper-
imental data as shown in Table 2 for the two minor loops. The forward problem is to
implement the model output (6) with the constraint (3) to (9) corresponding to the ap-
propriate minor loop with material sensitive parameters q related to the approximated
B-H curves. To implement those, we use the nonlinear finite element analysis in two
dimensions. Since the reluctivity ν is state dependent, an iterative solver using Newton-
Raphson method is adopted to the forward analysis (See [2] for more details).
4. Formulation of Inverse Analysis
Our inverse problem is to evaluate material sensitive magnetic parameter q from the
Np
measurements by the pick-up coil. Namely, given the measured data {{ Ỹi }i=1 }j corre-
Np
sponding to the magnetization and demagnetization processes {{J i }i=1 }j at the location
index j, the problem is to seek the minimum solution q = q ∗ of

Np
Ej (q) = min |Y (ν(q, Jij (Ha ), ΔN ) − Y˜ij |2 (10)
a∈Q
i=1
Table 2. Parameter settings of some typical B-H relations in the laboratory experiments.
(a) Low range
Ba Hc BR χH χR χa χrev
0MPa 1.31e-00 2.64e+02 8.68e-01 2.85e-03 6.65e-04 1.65e-03 3.10e-04
260MPa 9.20e-01 2.70e+02 4.98e-01 6.55e-04 2.09e-03 1.36e-03 3.55e-04
518MPa 5.76e-01 2.69e+02 2.20e-01 1.00e-03 4.81e-04 1.18e-03 3.55e-04
(b) High range

Ba Hc BR χH χR χa χrev
0MPa 1.67e-00 3.03e+02 1.11e-00 3.28e-03 6.55e-04 1.18e-03 2.69e-04
260MPa 1.29e+00 3.20e+02 7.40e-01 2.09e-03 6.50e+04 1.00e-03 2.70e-04
518MPa 9.67e-01 3.61e+02 3.41e-01 1.01e-03 6.33e-04 9.03e-04 3.55e-04
ǵ 3000 0MPa (Low range)

ǵ 3000
0MPa (High range)
260MPa (Low range)
2500 260MPa (High range)
518MPa (Low range) 2500 518MPa (High range)
2000
2000
1500
1500
1000 1000
500 500
0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
B [T ] B [T ]
(a) Low range (b) High range

Figure 6. ν function associated with the approximated B-H curves.
where Q denotes the appropriate admissible parameter class. This formulation allows us
to use the various kinds of optimization techniques. In this paper, we use the Tabu-search
algorithm. Tabu-search algorithm involves constructing from a current solution q(i), a
update solution q(j), and checking whether one should stop there or initiate another step.
Neighborhood search methods and some information related to the exploration process
(Tabu-list) play important roles in the algorithm. Detailed discussions are presented in
[6,7], etc.
5. Computational Experiments
Throughout the computational experiments, we solve the nonlinear inspection problem

with the “true” degradation parameter q true , added the Gaussian random sequence as
noise disturbance and used the result as simulation data. At each inspection area, the set
of degradation parameters is divided into the five subdivisions as shown in Fig. 7 and the
scanning strategies are illustrated in Fig. 8. Figure 9 shows the B-H curve of the magnetic
yoke which was experimentally determined in the laboratory. The unknown parameter
vector q(l,m) given by (1) takes different values at the position (l = 1, 2, 3, 4, 5) on the
Scanning lines Scanning area
x2 x
3
x1
Figure 7. Parameter setting in the experiments. Figure 8. Scanning strategies in the experiments.
2.5
Magnetic flux density B [T ]
1.5
0.5
0
0 1000 2000 3000 4000 5000
Figure 9. B-H curve of magnetic yoke.
Table 3. True and estimated parameters in the computational experiments.

(a) True Parameter values
position 1 position 2 position 3 position 4 position 5
line 1 0 0 0.5 0.2 0
line 2 0 0.5 0.2 0 0
line 3 0.5 1 0.5 0 0
line 4 0 0.5 0 0 0
(b) Estimated results (5% noise)
position 1 position 2 position 3 position 4 position 5
line 1 0.01 0.12 0.43 0.15 0.11
line 2 0.21 0.38 0.10 0.12 0.08
line 3 0.43 0.94 0.62 0.11 0.09
line 4 0.17 0.52 0.10 0.18 0.12
scanning line (m = 1, 2, 3, 4). To solve this, we seek the optimal solution of (10) at each
sensor location (l, m). Then the distribution of the estimated degradation parameter de-
fined by (2) is listed in Table 3. Figure 10 depicts the degradation map (5×4) correspond-
ing to the true parameter {q true }(l,m)=(1,1) and estimated {q∗ }(l,m)=(1,1) , respectively.
(5,4) (5,4)
Deg Deg
Deg Deg
1 1
1 0.8 1 0.8
0.8 0.6 0.8 0.6
0.6 0.6
0.4 0.4
0.4 0.4
0.2 0.2 0.2
0.2
0 0 0 0
e
ine 5 lin 5
4 gl 4 4
ing 4
3 nnin 3 ann
Sc
a 3 Sc 3
2 2 2
2
1 1 1 1
x3 0
x3 0
x1 x1
(a) True degradation map (b) Estimated map (5% Noise)

Figure 10. Degradation map in the computational experiments.
6. Concluding Remarks
New assessment strategies for a material degradation were proposed by using magnetic
inverse problems. The material degradation was characterized by analyzing a set of mi-
nor hysteresis loops at the low, middle, and high level of the applied magnetic fields. The
inspection model was given by the nonlinear Euler equation with unknown magnetic pa-
rameters of the curvatures related to minor hysteresis curves. The Tabu-search algorithm
was effectively adopted to solve the inverse problem. The effectiveness of the proposed
method was demonstrated via computational experiments. Our current study is directed
to evaluating the estimation procedure to experimental data.
Acknowledgements
This study was supported in part by the Grant-in-Aid for Scientific Research No.
17560373 by the Japan Society for the Promotion of Science and was also supported in
part by the JFE 21st Century Foundation.
References
[1] F. Kojima, K. Ara, and S. Takahashi, Identification of material degradation for nonlinear electromagnetic
problem, Electromagnetic Nondestructive Evaluation (VIII), Studies in Applied Electromagnetics and
Mechanics, Vol. 24 IOS Press (2004) pp. 85-92.
[2] F. Kojima and T. Fujioka, Quantitative evaluation of material degradation parameters using nonlinear
magnetic inverse problems, International Journal of Applied Electromagnetics and Mechanics, (2007)
to appear
[3] S. Takahashi and L. Zhang, Minor hysteresis loop in Fe metal and alloys, J. Phys. Soc. Japan, Vol. 73,
No. 6 (2004) pp. 1567-1575.
[4] H. Kikuchi, K. Ara, N. Ebine, Y. Sakai, Y. Kamada, and S. Takahashi, A probe using a magnetic yoke
for NDE of ferromagnetic steels, Electromagnetic Nondestructive Evaluation (VIII), Studies in Applied
Electromagnetics and Mechanics, Vol. 24 IOS Press (2004) pp. 146-152.
[5] C. de Boor, On calculating with B-splines, J. Approx. Theory, Vol. 6 (1972) pp. 50-62.
[6] F. Glover, Tabu search, Part I, ORSA Journal on Computing, Vol. 1 (1989) pp. 190-206.
[7] F. Glover, Tabu search, Part II, ORSA Journal on Computing, Vol. 2 (1990) pp. 4-32.
IOS Press, 2007
Identification of Defects from ECT Signals

Using Linear Discriminant Function
Weiying CHENG a, 1, Shigeru KANEMOTO b, Ichiro KOMURA a
a
NDE Center, Japan Power Engineering and Inspection Corporation,
Yokohama, Japan
b
School of Computer Science and Engineering, the University of Aizu,
Aizuwakamatsu, Japan
Abstract. Statistical pattern recognition method is applied to eddy current testing

(ECT) based defect identification in this work. Spatially distributed ECT signals
are converted into a multi-dimensional space vector using data embedding method
and the Fisher’s discriminant is then applied to find the linear projection of the
multi-dimensional data that best distinguish defect and non-defect signals. A set of
potential defects are clearly distinguished from welding noise by utilizing the
discriminant function derived from supervised learning. Furthermore, the
correlation between the probability of detection and the dimension of constructed
vector space is investigated. Both the false and miss detection probability decrease
with the increase of vector dimension. The reliability of a discriminant method is
significantly enhanced by increasing the dimension of vector space.
Keywords. Eddy current testing, defect identification, Linear discriminate
function, False/miss probability
1. Introduction
Eddy current testing (ECT) is a low cost, high speed non-destructive inspection (NDI)
method to detect surface breaking defects, and widely used for non-destructive testing
and evaluation of metallic materials, such as those used in nuclear, aerospace, high
pressure engineering system, etc. A typical time-varying ECT signal is represented by a
complex quantity, namely, impedance or voltage, by analogy with the AC circuit
impedance analysis in engineering. The complex quantity contains real and imaginary
components, which are respectively in-phase and 90 degrees out of phase with the
excitation current. ECT signal is analyzed in time-domain (strip-chart) or in complex
plane [1-3]. The identification and quantitative evaluation of defects depend on the
interpretation of waveforms in strip-chart or locus curves in complex plane, and
significantly rely on the skill and experience of an inspector. Furthermore, the presence
of surface unevenness, variation of electro-magnetic properties, edge-effect, change of
lift-off, etc., make the problem even more complicated. Identification of defects in
weldment is a typical problem with such complexity. Signals taken in or close to weld
beads and heat affect zones are mixed with noises from surface roughness, variations of
electrical conductivity and magnetic permeability of base metal and welding.
1
Corresponding Author: Weiying Cheng, NDE Center, Japan Power Engineering and Inspection
Corporation, Benten-cho 14-1, Yokohama, Kanagawa 230-0044, Japan; E-mail: cheng-
weiying@japeic.or.jp
252 W. Cheng et al. / Identification of Defects from ECT Signals Using Linear Discriminant Function
Sophisticated and automatic defect identification is preferred.

Pattern recognition is within the area of machine learning. It is well studied and
applied in text recognition, speech recognition, image analysis, etc., [4-6]. It is
sophisticated, and the algorithm itself can also be quantitatively evaluated based on the
probability of classification. In this work, the above-mentioned defect identification
problem is studied using a linear discriminant function, based on statistical pattern
recognition method. Spatially distributed ECT signals are converted into a multi-
dimensional space vector using data embedding method and the correlation between
the dimension of data and the probability of false/miss identification is investigated.
The defect identification procedure is briefly described in section 2, topics summarized
in the procedure are described in detail in sections 3 and 4, and conclusion is given in
section 5.
2. Procedure of Defect Identification
When applying pattern recognition method to defect identification, a system should be

instructed by defect and noise signals in advance. Discriminant function and the
threshold value to decide the acceptance or rejection of defects are calculated based on
the identification of supervised defects. Thereafter, this discriminant function and
threshold value will be applied to the identification of other defects with similar noise
and signal patterns.
The process of defect identification is divided into four parts in detail and
summarized in Figure 1:
The pre-processing part consists of data acquisition, data cleaning, and data
normalization steps. Data acquisition involves selection of sensors and testing
conditions, and data acquisition/storage. Disturbing noises in raw ECT data are
eliminated as much as possible in data cleaning step. Signals are normalized so that
they will not be affected by scale or unit.
Data construction/Feature selection involves construction of data space vector and
selection of characteristic features for efficient defect identification.
A discriminant operator is applied to supervised signals in supervised learning.
The data constructed in multi-dimensional vector space are projected to a one-
dimensional space and the discriminant boundary to classify defect and non-defect is
decided.
The discriminant function and threshold value derived from supervised learning are
then applied to the identification
of other potential defects with
Pre-Processing similar characteristics as the
supervised ones.
Feature selection/Data construction
3. Identification of
Supervised learning--Identification of supervise defects,
Supervised Defect
discrimination function and threshold value obtained
SUS304 pipes (10 inches outer-
Applying--Identification of other defects
diameter, 12mm thickness) are
circumferentially welded by
Inconel 182. The welding
Figure 1. Flow chart of defect identification. beads are about 8mm and
W. Cheng et al. / Identification of Defects from ECT Signals Using Linear Discriminant Function 253
15mm wide respectively on the inner and outer surfaces of the pipe. The base metal is
non-magnetic while the welding is magnetic. Defects in or close to the welding line are
significantly affected by the variation of electro-magnetic property of base metal and
welding part. Besides, since the weld beads on the inner side of the pipe were not
removed completely, there are noises due to the unevenness of surface and tilt of probe.
A 2mm deep, 10mm long, 0.3mm wide circumferential EDM slit fabricated along
a welding line is selected as supervised defect. A 2.3mm outer diameter pancake probe
(excitation frequency 400 kHz) scans on the pipe’s inner surface in a 30mm×30mm
(30mm along the pipe’s circumferential and axial directions, respectively) area. The
scanning pitch is 0.5mm. ECT measurement is balanced at base metal, that is, the ECT
signal is set to 0 on flaw-free base metal.
3.1 Pre-processing
ECT signal Y changes with sampling position (i,j), and can be expressed by:
Y(i, j ) [ zre (i, j ), zim (i, j )] , (1)
where Z re and Z im are the real and imaginary components of ECT signal. The
complex quantity Z re (i, j ) Z im (i, j ) j of the supervised signal is plotted in Figure 2(a),
in which the phase angle and the amplitude of the complex value at sampling point (i,j)
are indicated by the direction and length of an arrow originated from point (i,j). The C-
scan image of the signal’s real component is depicted in Figure 2 (b). It is impossible
to identify the defect from Figures 2(a) and 2(b) directly.
䎖䎓䎑䎖䎓䎑
(a) (c)
䎕䎓䎑䎕䎓䎑
䏄䏛䏌䏄䏏䎋䏐䏐䎌
䎔䎓䎑䎔䎓䎑
䎔䎓䎑䎕䎓䎑䎖䎓䎑䎔䎓䎑䎕䎓䎑䎖䎓䎑

䏆䏌䏕䏆䏘䏐䏉䏈䏕䏈䏑䏗䏌䏄䏏䎋䏐䏐䎌䏆䏌䏕䏆䏘䏐䏉䏈䏕䏈䏑䏗䏌䏄䏏䎋䏐䏐䎌
䎖䎓䎑䎖䎓䎑
(b) (d)
䎕䎓䎑䎕䎓䎑
䎔䎓䎑䎔䎓䎑
䎔䎓䎑䎕䎓䎑䎖䎓䎑䎔䎓䎑䎕䎓䎑䎖䎓䎑

䏆䏌䏕䏆䏘䏐䏉䏈䏕䏈䏑䏗䏌䏄䏏䎋䏐䏐䎌䏆䏌䏕䏆䏘䏐䏉䏈䏕䏈䏑䏗䏌䏄䏏䎋䏐䏐䎌
Figure 2. ECT signals of supervised defect.

It is noticed that the welding line is in circumferential direction and the welding
noise keeps almost unchanged circumferentially. This noise can be eliminated to some
extent by using a space filter [7], that is, subtracting circumferentially a series of
signals along an axial line far away from defect. The after-filtering ECT signals are
then normalized by taking the means of signals to 0, and the variance to 1. The
complex quantities of the after processing signals are depicted in Figure 2(c), and the
contour of the signal’s real component is presented in Figure 2(d). An indistinct outline
of defect is delineated in Figure 2(d). However, to realize automatic defect
identification and evaluate the identification quantitatively, an identification based on
the pattern recognition is adopted hereafter.
3.2. Construction of Multi-dimensional Data
One arrow in Figures 2(a) or 2(c) stands for ECT signal at one sampling point only,
while the locus patterns represent ECT signals along neighboring sampling points. In
this study, a new multi-dimensional state space is constructed based on the analogy
with the Taken’s embedded theory [8]. And by embedding signals at neighboring
sampling points of (i,j), i and j are position indexes along circumferential and axial
axes, respectively
Z(i, j ) [ zre (i n, j m), zim (i n, j m)] (n N㨪 N , m M㨪 M ) , (2)
a 2×(2N+1)×(2M+1) multi-dimensional vector space is constructed and the locus curve
patterns in region (i-N, j-M) to (i+N, j+M) are explored. For example, when M=0, and
N>0, the locus along the circumferential direction is investigated, and when N=0, and
M>0, the locus along the axial direction is investigated.
3.3. Identification of Supervised Defect with Linear Discriminant Operator
A linear discriminant operator referred to as ‘Fisher’s Discriminant’ is applied in this

study. The Fisher’s linear discriminant is a classification method that projects high
dimensional data onto a line and performs classification in this one-dimensional scalar
space.
In this study, the defect and noise classes are defined respectively by,
1) class of defect signal ȦS 㧦㨧Z(i,j) 㨨㧔i,j㧕in defect region㨩
2) class of noise signal ȦN 㧦 {Z(i,j) 㨨㧔i,j㧕out of the defect region}
where the subscripts S and N stand for defect signal and noise, respectively. The
optimized linear discriminant function to distinguish these two classes is,
g (Z(i, j )) wT Z(i, j ) W0 , (3)
where w is a vector representing an axis to make the projection from multi-dimensional
space to one-dimensional space, W0 is an optional scalar value which decides the
original point of discriminant function along the projection axis, and g is a scalar value
in the projected one-dimensional space.
According to the Fisher’s discriminant method, the projection maximizes the
distance between the means of two classes while minimizing the variance within each
class [ 9]. w is calculated by
w ^Ȉ S Ȉ N `1{m S m N } , (4)
where m S and m N are means of the multi-dimensional vectors of defect and noise
classes, Ȉ S and Ȉ N are metrics of variance of the two classes respectively. If W0 is
calculated by
(V N2 m S V S2m N )
W0 wT , (5)
(V S2 V N2 )
where V S2 and V N2 are variance of the two classes which can be calculated by
V S2 wT Ȉ S w , V N2 wT Ȉ N w , then the two classes can be simply classified by the
plus or minus signs of the discriminant function g, that is,
g (Z(i, j )) ! 0 : Z(i, j ) Z S
. (6)
g (Z(i, j )) 0 : Z(i, j ) Z N
The value of g can be represented in a contour graph. By comparing the region of
the supervised defect and that of g>0, a threshold value to decide the acceptance or
rejection of defect can be decided. It should be noticed that this is an approach aiming
to identify defects from noise, the ‘defect region’ defined here is the region of defect
signal, not the exact physical region of a defect.
A supervised learning is carried out using the algorithm mentioned above. The
ECT signals described in Figure 2(c) are utilized, and the supervised defect is defined
in a region surrounded by the solid line in Figure 2(d). Signals inside and out of this
region are assigned to defect and noise classes, respectively. A 22-dimensional vector
space (N =0, M=5 in Eq. (2)) is constructed. The supervised defect is identified using
Eqs. (3)~(6). The value of discriminant function g is calculated by Eq. (3) and
described in Figure 3(a). Only the positive value, that is, g ! 0 is plotted in Figure 3(b).
Although there is very strong welding noise as indicated in Figure 2, the notch is
clearly extracted and indicated in Figure 3(b). The length of image g consistent with the
true defect length when the threshold value is set to 0.4 u g max . The over-threshold
region is shown in Figure 3(c).
By applying the statistics identification method, the probability of defect
identification can be evaluated quantitatively. The classification boundary is indicated
by vertical dash-dot line in Figure 4. Figure 4(a) shows the probability of defect
identification by utilizing a 22 dimensional vector data (N=0, M=5, both real and
imaginary components of ECT signals are utilized). The horizontal axis labels the
discriminant function value g(Z), and the vertical axis labels the probability of
identification. The identification probability of defect and noise are denoted by (×) and
circle ( ٤ ), respectively. Normalized histograms of defect and noise classes are
constructed based on the means and variances of the distribution of each class, and
(a) (b) (c)
Figure 3. Projection of multi-dimensional data to one-dimensional space.

䎓䎑䎔䎕
represented by solid and broken lines
䏇䏈䏉䏈䏆䏗 respectively.
䏑䏒䏌䏖䏈
䎓䎑䎔 (a) The correlation between the
dimension of vector space and the
䎓䎑䎓䎛 probability of identification is
䎳䏕䏒䏅䏄䏅䏌䏏䏌䏗䏜
investigated as well. The probability

䎓䎑䎓䎙
of identification by using a two
䎓䎑䎓䎗 dimensional vector (N=0, M=0, real
and imaginary components of ECT
䎓䎑䎓䎕 signal are utilized) is presented in
Figure 4(b). This identification
䎓 corresponds to the identification of
䎐䎘䎓䎐䎗䎓䎐䎖䎓䎐䎕䎓䎐䎔䎓䎓䎔䎓䎕䎓䎖䎓
䎧䏌䏖䏆䏕䏌䏐䏌䏑䏄䏑䏗䎃䎧䏌䏖䏗䏄䏑䏆䏈 utilizing real and imaginary
components at each sampling point.
䎓䎑䎓䎚
䏇䏈䏉䏈䏆䏗 Comparing with the identification of
䏑䏒䏌䏖䏈
䎓䎑䎓䎙 (b) using 22-dimensional vector space in
䎓䎑䎓䎘
Figure 4(a), the identification
probability is much lower, while false
䎓䎑䎓䎗 alarm and miss alarm probability

䎓䎑䎓䎖
increases significantly. The
comparison clearly demonstrates that
䎓䎑䎓䎕 defect identification capability is
䎓䎑䎓䎔
enhanced by using signals in multi-
dimensional space vector.
䎓
䎐䎔䎘䎐䎔䎓䎐䎘䎓䎘䎔䎓䎔䎘 Since the defect region is
䎧䏌䏖䏆䏕䏌䏐䏌䏑䏄䏑䏗䎃䎧䏌䏖䏗䏄䏑䏆䏈 indicated to some extent in the after-
filtering C-scan image of signal’s real
䎓䎑䎓䎛
component in Figure 2(d), a
䎓䎑䎓䎚
(c) 䏇䏈䏉䏈䏆䏗
comparison of defect identification by
䎓䎑䎓䎙䏑䏒䏌䏖䏈 utilizing only the real component of
signal to that of utilizing both the real
䎓䎑䎓䎘
and imaginary components of signal is

䎓䎑䎓䎗 made. Figure 5(c) shows the
䎓䎑䎓䎖 probability of identification by
䎓䎑䎓䎕
utilizing an 11-dimensional (N=0,
M=5, and only real component of
䎓䎑䎓䎔
signal is utilized) vector space.
䎓
䎐䎕䎓䎐䎔䎓䎓䎔䎓䎕䎓
Comparing with Figure 4(a), it is clear
䎧䏌䏖䏆䏕䏌䏐䏌䏑䏄䏑䏗䎃䎧䏌䏖䏗䏄䏑䏆䏈 that the probability of defect
identification by using both real and
Figure 4. Comparison of defect identification imaginary component is higher than
probability by using signals of different that of using only the real component
dimensions.
(a) 22 dimension data, N=0, M=5, both real
of signal. A sophisticated automatic
and imaginary components are utilized. identification based on the multi-
(b) 2 dimensional data, N=0, M=0, both real dimensional space vector constructed
and imaginary components are utilized. by both real and imaginary
(c) 11 dimensional data, N=0, M=5, only
components of ECT signals is more
real component is utilized.
reliable.
Table 1 Probability of defect identification.

Condition for identification Probability (%)
Dimension N M Component False alarm Miss alarm
22 0 5 Re.,Im. 0.1 0.04
2 0 0 Re., Im.. 1.4 5.0
11 0 5 Re. 0.46 0.58
3 0 1 Re. 6.9 7.3
False alarm probability and miss alarm probability are measures of defect
identification capability of a classification method. The false alarm and miss alarm
probabilities are calculated using normal distribution and listed in Table 1. An
additional case of using data constructed in a 3-dimensional vector space (N=0, M=1,
only real component of signal is utilized) are presented in Table 1 also. Comparing
with the 14% error probability (total of false alarm and miss alarm probabilities) by
utilizing data constructed in 3-dimensional vector space, the error probability reduced
to less than 0.2% by utilizing 22-dimensional signals. The results in Table 1 and Figure
4 show that:
x Defect identification is improved by utilizing a multi-dimensional vector space.
x Defect identification is enhanced by utilizing both the real and imaginary
components of signals.
4. Identification of Other Defects
The discriminant function obtained from the supervised learning, that is, the axis to
make the data projection, w, the original point to decide the hyper-plane to separate the
defect and noise region W0 , and the threshold value for defect region clarification are
applied to other ECT data for the identification of other defects with similar noise and
defect characteristics. Since the comparison of identification probability shows that the
22- dimensional vector space has the highest identification probability and lowest error
alarm probability, 22-dimensional data are constructed and utilized for defect
identification hereafter.
Figures 5 and 6 show the identification of circumferential 10mm long cracks
fabricated close to and in the circumferential welding bead, respectively. An indistinct
outline of crack is illustrated in the contour graph of the real component of signals in
Figure 5, and the projection gives a distinct crack shape. On the other hand, the crack in
Figure 6 can not be identified without the discrimination process. The cracks identified
are 8mm long, which are 2mm shorter than the true crack lengths, the existence and the
location of crack are identified, however.
5. Discussion
A statistical pattern recognition paradigm has been utilized for ECT based defect
identification in this work. By embedding ECT signals into a multi-dimensional vector
space and using the Fisher’s discriminant, objective and automatic defects
identification is realized and the probability of defect identification can be
quantitatively evaluated as well. The quantitative evaluation of false alarm probability
and miss alarm probability shows that the capability of identification is significantly
improved by utilizing both the real and imaginary components of signals in a multi-
dimensional vector space. However, some points should be addressed:
Figure 5. Identification of a crack close to welding bead.
Figure 6. Identification of a crack located in welding bead.
x This approach is only valid to defects which have similar noise and signal
patterns as those in supervised learning.
x Due to the limitation of measurement data, only one defect is used in supervised
learning in this study. Since the ECT signal changes with the defect’s dimension,
especially the defect depth, practically, signals covering all the possible signal
patterns should be included in the supervised learning, and the calibration or
normalization of signal should be made based on the whole database set in prior.
x What we identified here is the region of defect signal, not the region of a physical
defect.
References
[1] Dogandzic, A., Eua-anant, N., and Zhang B., Defect detection using hidden markov random fields,
Review of Progress in Quantitative Nondestructive Evaluation, Vol. 24 (2005), 704-711.
[2] Zavaljevski, N., Bakhtiari, S., Miron, A., Kupperman, D.S., Automatic algorithms for eddy current
array probes for steam generator inspection, Review of Progress in Quantitative Nondestructive
Evaluation, Vol. 24 (2005), 728-735.
[3] Shin, B., Ramuhalli, P., Udpa, L., and Udpa, S., Independent component analysis for enhanced feature
extraction from NDE, Review of Progress in Quantitative Nondestructive Evaluation, Vol. 23 (2004), 597-604.
[4] C.C. Chibelushi, F. Deravi, and J.S.D. Mason, "A review of speech-based bimodal recognition", IEEE
Transactions on Multimedia, Vol. 4, no. 1 (2002), 23-37.
[5] J. Zhang and T. Tan, "Brief review of invariant texture analysis methods", Pattern Recognition, vol. 35,
no. 3 (2002), 735-747.
[6] M.-H. Yang, D.J. Kriegman and N. Ahuja, "Detecting faces in images: a survey", IEEE Transactions
on Pattern Analysis and Machine Intelligence, Vol. 24, no. 1 (2002), 34-58.
[7] Weiying CHENG, Ichiro KOMURA, Mitsuharu SHIWA, Shigeru KANEMOTO, Eddy Current
Examination of Fatigue Cracks in Inconel Welds, Transactions of the ASME, Journal of Pressure
Vessel Technology, Vol. 129 (2007), 169-174.
[8] Kantz, H., Schreiber, T., Nonlinear time series analysis.
[9] Ishii, K., Ueda, S., Maeda, H., and Murase, Y., Pattern Recognition (in Japanese), 1998, Ohamsha.
IOS Press, 2007
Sizing of Volumetric Stress Corrosion

Crack from Eddy Current Testing Signals
with Consideration of Crack Width
Zhenmao CHEN a,1 Noritaka YUSA b and Kenzo MIYA b
a
MOE Key Laboratory for Strength and Vibration, Xi’an Jiaotong University, China
b
International Institute of Universality, Tokyo, Japan
Abstract. In this paper, the influence of crack width on the reconstruction of Stress
Corrosion Crack (SCC) from Eddy Current Testing (ECT) signals is investigated.
The features of the simulated ECT signals due to cracks of different widths and
conductivities are extracted and compared with those of the measured SCC signals.
A strategy, which treats the crack width and the other crack parameters separately,
is introduced to cope with the effect of crack width. The feature parameters of 2D
ECT signals from a pluspoint sensor are applied to predict the crack width and con-
ductivity, before the other detailed parameters are reconstructed in a conventional
way. To implement the proposed strategy, the fast forward solver developed by au-
thors is updated for asymmetric problem and cracks of arbitrary width. It is demon-
strated that the new strategy can improve sizing result for artificial SCC though it
is still not satisfactory for a deeper SCC.
Keywords. ECT Inversion, Crack width, Crack conductivity, 2D signal simulation
1. Introduction
For optimizing the maintenance of a large mechanical system such as a nuclear power
plant, quantitative evaluation on crack propagation becomes more important nowadays
[1]. However, sizing of a Stress Corrosion Crack (SCC), the major concern in the de-
fects initiated in the key structural components of a nuclear power plant, is still difficult
because of its complicated microstructure [2]. Recently, it is found that Eddy Current
Testing (ECT) is applicable in the sizing of some natural cracks [3,4]. For a volumetric
SCC, however, there is a large error - usually less estimate, between the result of ECT
inversion and the true sizes especially if the crack is deep. The major reason of this er-
ror is considered due to volumetric and conductive properties of SCC [5,6]. The crack
conductivity has been introduced to the signal simulation and crack reconstruction by
several researchers in order to improve the precision. For some cases, however, the sizing
result is still poor even treating the crack region as a conductive zone.
In this paper, effects of both crack width and conductivity on the reconstruction
of crack profile are investigated numerically at first. Several testpieces with artificial
1 Corresponding Author: Zhenmao Chen, MOE Key Laboratory for Strength and Vibration,
School of Aerospace, Xi’an Jiaotong University, 28 West Xianning Road, Xi’an, China, E-mail:
chenzm@mail.xjtu.edu.cn
260 Z. Chen et al. / Sizing of Volumetric Stress Corrosion Crack from Eddy Current Testing Signals
SCC are fabricated and inspected (C-scan) by using a pluspoint sensor. The features of
simulated signals from cracks of different widths and conductivities are compared with
that of the measured signals. The results show that the crack width significantly affects
the signal of C scan in case of a volumetric crack. The crack width has to be considered
in the crack inversion in order to get better sizing precision.
In addition, a scheme is proposed in this paper that treats the crack width effect
based on signal features (such as the distribution) of 2D signals, while the other crack
parameters are reconstructed by using signals just over the crack line. The pluspoint sen-
sor is selected to take advantage of its relative higher sensitivity on the crack width and
conductivity. The efficiency of the strategy is investigated by reconstructing an artificial
SCC of volumetric type from measured signals.
2. Effect of Crack Width on ECT Signals
In paper [7], numerical results shown that ECT signals depend on the crack width when
the crack region is conductive. The conventional viewpoint that the crack signal is inde-
pendent of the crack width is found only valid for the cracks of 0 conductivity. For the
cracks with nonvanishing conductivity, the signal amplitude depends on the crack width
almost linearly.
On the other hand, the crack width also affect the distribution of crack signals espe-
cially for a pluspoint probe. In fact, the crack signal of the coil in parallel with the crack
surface increases with the enlarging crack width that results in a smaller probe output
for crack of large width. In Fig.1 (a) and (b), measured C scan signals of an SCC and
a fatigue crack are depicted respectively. From the results given in paper [8], the ma-
jor differences of an SCC and a fatigue crack are the crack conductivity and the crack
width. From the signals at line x=0 (the signal at a scanning line perpendicular to the
crack surface and passing the crack center), one can find that the signal value just over
the crack is not the biggest while the signal of fatigue crack does. These results show
that the distribution of C scan signals also contains important information on the crack
status.
In order to clarify the correlation between the C scan signal and the crack width,
crack conductivity and other profile parameters, parameter survey on C scan signals is
conducted by developing a 2D fast signal simulator.
Absolute value (mv)
10
0.20
8
0.15 6
4
0.10 2
30 0
0.05
-2
20 -4
0
10
30
0 20
-20 10
-15 -10 y (mm) -20
-10 -15 0
-5 -10 -10
0 -20 -5
0
5 5 -20 Y(mm)
x (mm) 10 10
15
20 -30 X(mm) 15
20
(a) Stress corrosion crack (b) Fatigue crack

Figure 1. Comparison of ECT signals from an SCC and a fatigue crack
Z. Chen et al. / Sizing of Volumetric Stress Corrosion Crack from Eddy Current Testing Signals 261
2.1. Upgrade of the Fast Forward Solver for 2D Signal Simulation
For inversion of C scan signals, a 2D fast forward simulation tool is necessary. The
database type fast forward solver proposed by authors [7,9] is applicable but only for
cracks of given width. In case of an unknown crack width, the method has to be modified
as the databases have to be established a priori based on a fixed crack width.
On the other hand, the problem geometry becomes asymmetry when the probe does
not move just over the crack line. Therefore, full problem region has to be considered to
establish the database for the fast forward solver. This results in a large requirement on
computer resource.
In this work, methods to solve these problems are developed as follows:
1) Strategy for arbitrary crack width
The basic idea of this strategy is to calculate the signal of a crack in arbitrary width
by interpolating the signals from cracks of given widths, which can be calculated based
on the fast solver given in [7]. In practice, this can be realized by establishing n sets of
databases in prior for cracks of width W i = W1 + (i − 1)(W2 − W1 )/(n − 1), i =
1, 2, ..., n if the selected range of crack width is [W 1 , W2 ]. The signal S(W ) of crack in
width W can be calculated by interpolating crack signal S Wi and SWi+1 (Wi < W <
Wi+1 ) that have been computed by using databases of selected crack width.
2) Establishment of asymmetric database by symmetric simulations
For an asymmetric problem, the databases can be established by using information
of two symmetric problems. In fact, the field due to a unit source at node n calculated
by the symmetric FEM-BEM code is a superposition of field due to this unit source and
another unit source applied at its symmetric node N − n. In symmetric case the unit
source at node N − n is a positive unit while it becomes negative for the antisymmetric
case. Therefore, the field due to the unit source can be calculated by making average on
the field of symmetric and antisymmetric cases. This consideration is also valid for the
unflawed databases of the excitation and pickup coils.
In the strategy above, it is important to be pointed out that the potential values in
the databases have to be calculated and stored in a very high precision. Otherwise, the
variation component in the potential can be cut because the scalar potentials due to a unit
source have a huge base value (for instance over 1.5*e20).
Based on these considerations, the fast forward analysis code is upgraded and the
databases are established for over ten crack widths ranging from 0.1 mm to 2 mm. Figure
1(a) shows an example of crack signal for a pluspoint probe calculated by using new
fast forward solver and the FEM-BEM code. The results show that the new code can
give a better precision with smaller computer resource. Figure 1(b) gives an example of
simulated 2D signal for a 3 mm depth crack, a pluspoint probe and 100 kHz excitation
frequency. This 2D signal is calculated in less than 5 minutes of CPU time with a PC of
3.0 GHz.
2.2. Investigation of Dependence of 2D Signals on Crack Parameters
1) The correlation of the crack signal and crack conductivity

By calculating ECT signals due to a shallow crack (0.8 mm depth, 10 mm length and
0.4 mm width) and a relative deep crack (4 mm depth, 10 mm length and 0.4mm width)
for crack conductivity ranging from 0 to 50% of the base material (pluspoint sensor,
100kHz), the following properties are obtained.
2.0
FEM-BEM Code Absolute singal (mV)
Fast Solver
1.5
1.8
1.0 1.6
1.4
1.2
1
Imaginary Part (mV)
0.5
0.8
0
0.6
0.4 30
0.2
-0.5 0 20
10
-1.0
0
y (mm)
-1.5 -20 -10
-15
-10
-5 -20
-2.0
0 5
-2.0 -1.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 10 15 -30
Real part (mV) x (mm) 20
(a) Comparison with FEM-BEM code (b) An example of 2D simulation results

Figure 2. 2D signal results of the asymmetric code
• For a deep crack, the crack signals decrease significantly with the increasing crack
conductivity. However, the signals of a shallow crack do not change due to the
crack conductivity as rapid as the case of deep crack. For instance, 10% conduc-
tivity increment only results in an 1/5 signal reduction for a shallow crack but
decrease 2/3 time for the deep crack case.
• The amplitude and distribution of crack signals due to the shallow and the deep
crack become similar when the crack conductivity is big. If the crack conductivity
is over 50%, the signals of the deep crack and the shallow crack are almost the
same (Fig.3 for example. The other results are not shown here due to the page
limit).
absolute signal (mV) absolute signal (mV)
0.05 0.06
0.045
0.04 0.05
0.035 0.04
0.03
0.025 0.03
0.02
0.015 0.02
0.01 0.01
0.005
0 0
30 30
20 20
10 10
-20 0 -20 0
-15 -15
-10 -10 y (mm) -10 -10 y (mm)
-5 -5
0 0
5 -20 5 -20
x (mm) 10 x (mm) 10
15 15
20-30 20-30
(a) 0.8 mm crack depth (b) 4.0 mm crack depth

Figure 3. Comparison of crack signals for case of large crack conductivity
From these results, it is no difficult to understand that the depth of a deep SCC is
much more difficult to be predicted than the case of a shallow one.
2) The effect of the crack depth on the crack signals

In this case, the signals of nonconductive crack and cracks of 10% conductivity are
calculated for crack depth from 0.4 mm to 8 mm. From the results, it is found that:
• The amplitudes of crack signals saturates at a much smaller crack depth for a
conductive crack than that for a nonconductive one. The saturation depth of the
nonconductive crack is about 8 mm while it is 3 mm for the 10% conductivity
crack. This result also show the difficulty of crack reconstruction for a deep SCC.
• When crack depth is small, there is a signal valley over the crack line for both
signals of the nonconductive and conductive crack. On the other hand, the signal
distribution of the conductive deep crack still show valley shape over the crack
line, but has a signal peak over the crack line for the nonconductive deep crack
case. (see Fig.4)
absolute signal (mV) absolute signal (mV)
0.3 1.8
1.6
0.25
1.4
0.2 1.2
1
0.15
0.8
0.1 0.6
0.4
0.05
0.2
0 0
30 30
20 20
10 10
-20 0 -20 0
-15 -15
-10 -10 y (mm) -10 -10 y (mm)
-5 -5
0 0
5 -20 5 -20
x (mm) 10 x (mm) 10
15 15
20-30 20-30
(a) 10% conductivity (b) nonconductive

Figure 4. Comparison of deep crack signals for different conductivity
3) The effects of the crack width on the crack signals

To investigate the effect of crack width on the signal distribution, the signals of a
deep and a shallow crack are calculated for crack width of 0.1 mm, 0.2 mm, 0.4 mm, 0.6
mm, 0.8 mm 1.0 mm, 1.4 mm and 1.8 mm respectively. From the simulation results, the
following conclusions are observed.
• For the conductive crack of conductivity 10%, the signals of the shallow crack
show almost the same distribution for each crack width. However, the signal dis-
tribution for a deep crack has different feature for a narrow crack and a volumetric
one. If the crack width is large, the signal distribution is similar with the shallow
crack case, i.e., a signal valley appears, while it has a signal peak just over the
crack line for a deep and narrow crack.
• For a conductive crack, the crack signals increases with the crack width signifi-
cantly. However, the crack signals do not show much change in case of noncon-
ductive cracks.
3. Reconstruction Scheme of a Volumetric SCC with Consideration of the Crack

Width Effect [3,4,10,11]
From the numerical investigation, it is verified that both the crack width and the crack
conductivity give significant effects on the crack signals. Therefore, both factors have to
be considered in the enhancement of SCC signal inversion.
Up to now, the reconstruction of an SCC is mainly carried out based on conductive
crack models and a fixed crack width. These approaches are feasible for several kinds of
practical SCC. However, it is also found that a large error may occur in the reconstruction
of a relative deep SCC. In this work, we try to improve the error by taking crack width
into the inverse analysis also.
However, as the crack width, the crack conductivity and the crack size belongs to
different parameter categories, to reconstruct them at the same time with a deterministic
optimization method is difficult. To overcome this difficulty, we propose following two-
step inversion scheme to predict crack depth and length by using the distribution of crack
signal.
The procedure of the proposed inversion strategy is as follows:
1) Estimate the crack width and conductivity by using simple searching approach,
i.e., calculating the 2D crack signals for each combination of selected crack width and
conductivity values and to find the proper set of crack width and conductivity by com-
paring the distribution of the simulated and measured 2D signals. In this procedure, the
crack length is considered long enough and the depth is taken as the value of a noncon-
ductive crack which gives a signal amplitude near the measured ones.
2) Based on the measured and the simulated crack signals, extract the crack parallel
1D signal that contains the biggest signal peak point.
3) Reconstruct the other crack sizes from the extracted 1D signal by using the inverse
analysis method given in [7]. The asymmetric forward analysis code has to be employed
in this case as the symmetric property is not valid when the probe does not move just
over the crack surface.
4) Go back step 1, 2 and 3 but with new crack depth and length.
5) Terminating the iteration if selected iteration number is reached or the residual
error cannot be improved further.
4. Numerical Validation
To investigate the efficiency of the proposed strategy, reconstructions of an artificial SCC

in an Inconel 600 alloy testpiece are carried out. The testpiece is 8 mm in thickness, 200
mm in length and 100 mm in width. The SCC is introduced to the testpiece by using
a jig of 3 point bending and the solution of polythionate acid. The crack is inspected
by a pluspoint sensor of the same size used in the numerical analysis described in the
previous sections. Figure 1(a) shows the amplitude distribution of the measured signal
for an excitation frequency of 100kHz. The signal is calibrated by using a signal of an
EDM notch of 3 mm depth.
In order to decide the crack width and crack conductivity, a lot of forward simula-
tions are performed for various crack width and conductivity. In these simulations, the
crack depth and length are taken as 0.8 mm and 10 mm respectively. From the numerical
results, it is found that a crack of 0.4 mm width and 2% σ 0 conductivity gives most sim-
ilar signal with the measured one. Figure 4 show a comparison of the simulated and the
measured signals (real part). One can find that both the distribution and the amplitude
are near each other. The imaginary part of the signal shows same agreement, i.e., the
phase property of the signal is also reasonable. Therefore, we chose the width and the
conductivity of the crack as 0.4 mm and 2% of the base material in the further simulation.
As shown in Fig.5, the inspection signal at the line just over the crack is relative
small. To use a signal of better S/N ratio in the reconstruction of other crack parameters,
the crack signal at a line 2 mm away from the crack line is selected. After 30 iterations,
the reconstructed crack profiles are 11.5 mm in length and 0.58 mm in depth. Further
iteration does not give much improvement on the residual error. The comparison of the
measured signal and the simulated signal from the crack of reconstructed profile is shown
in Fig.5. Because the crack was assumed in a rectangular shape, the first half and the
later half of the simulated signal are symmetric. The measured shape, however, has a
Imaginary (mV)
Imaginary (mV)
0.15
0.1 0.15
0.05 30 0.1
30
0.05
0 20 0 20
-0.05
10 -0.05
-0.1 10
-0.1
-0.15
y(mm)
0 -0.15 0 y (mm)
-10 -10
-20 -20
-20 -15 -20 -15
-10 -5 0 -10 -5
5 10 -30 0 5 10 -30
15 20 15 20
x (mm) x (mm)
(a) simulated (b) measured

Figure 5. Comparison of 2d signal distribution of simulated and measured signal for crack width 0.4 mm and
conductivity 2%, (real part)
bigger peak at one side of the crack, i.e., the rectangular assumption results in a smaller
prediction of crack depth.
To reduce this effect, a signal is generated based on the later half of the measured
signal (the half with bigger signal peak) by using symmetric operation. The reconstructed
crack depth based on this signal should be better at the position of the maximum signal
point. Figure 5(b) shows a comparison of the measured signal (with symmetric treatment)
and the reconstructed ones. In this case, the reconstructed crack depth is 0.64 mm and the
length is about 11.0 mm. The reconstructed signal gives a better agreement. To estimate
the depth at the shallow part of the crack, another new signal is also generated based
on the symmetric operation and the first half of the measured signal. In this case, the
reconstructed crack depth and length are 0.53 mm and 12.2 mm respectively. Better
residual error is obtained again.
From the result given above, we conclude that the crack is in a slope shape and the
maximum depth is about 0.64 mm and the length about 12.2 mm. The maximum depth
predicted using 0.2 mm crack width and 0 conductivity is less than 0.3 mm. This means
that the new method can give some improvement on the problem of less estimate.
After destructive testing, however, it is found that the maximum depth of the true
crack is about 5 mm, i.e., the new strategy still gives much small depth estimation though
the difference is improved. The major reason is considered that the excitation frequency
applied in the measurement is too high (the working frequency of used pluspoint sen-
sor is over 100kHz). As the skin depth of the material at 100 kHz is about 1.5 mm, to
reconstruct a crack much deeper than skin depth is difficult.
However, even for the crack greatly less estimated, the reconstructed crack signals
are agree well with the measured ones. This means that additional efforts have to be taken
to overcome the difficulty in the reconstruction of a deep and volumetric SCC.
5. Conclusions
In this paper, it is found through numerical analysis that both the crack width and con-
ductivity have large effects on the SCC signals. The crack width has to be taken into
account in the crack sizing procedure.
An inversion strategy, which treats both the crack width and conductivity as vari-
ables, is proposed to improve sizing result for volumetric SCC. By reconstructing a mea-
0.15
0.15 Reconstructed (Real)
Reconstructed Reconstructed (Imag)
Reconstructed (Imag) Measured (Real)
Measured (Real) Measured (Imag)
Measured (Imag)
0.1
0.1
0.05
0.05
Signal (mV)
Signal (mV)
0
0
-0.05
-0.05
-0.1
-0.1
-0.15
-0.15 -30 -20 -10 0 10 20 30
-30 -20 -10 0 10 20 30
x (mm) x (mm)
(a) using original signal (b) using generated signal

Figure 6. Comparison of the measured signal with the simulated one due to crack of reconstructed profile
sured SCC signal, it is demonstrated that the proposed strategy can give some improve-
ment on the reconstruction precision. However, as the excitation frequency applied in the
experiment is relative high, and the true crack is very deep, the reconstructed crack depth
is not in good agreement with the measured crack profile. More efforts are still necessary
for inversion of deep SCC in the future.
Acknowledgment
This work was supported in part by the National Natural Science Foundation and
the National Basic Research Program of China through Grant No.50677049 and
No.2006CB601202.
References
[1] ASME, 2001 boiler and pressure vessel code, section XI, Rules for in-service inspection of nuclear
power plant components, July, 2001.
[2] W.Cheng and et al., Depth sizing of partial contact SCC from ECT signals, NDT&E Int., Vol.39, 374-
383, 2006.
[3] N.Yusa, Z.Chen and K.Miya, Sizing of stress corrosion cracks in piping of austenitic stainless steel from
eddy current NDT signals, Nondestructive Evaluation, Vol.20, 103-114, 2005.
[4] Z.Chen, K.Aoto and K.Miya, Reconstruction of cracks with physical closure from signals for eddy
current testing, IEEE Trans.Mag., Vol.36, 1018-1022, 2000.
[5] K.Oshima,M.Hashimato, Research on numerical analysis modeling of SCC on eddy current testing,
J.JSAEM, Vol.10, 384-388, 2002.
[6] L.Janousek, Z.Chen, N.Yusa, K.Miya, A novel nondestructive method for distinguishing between fatigue
and stress corrosion cracks using electromagnetic induction, Proc.13th ICONE, Beijing, Vol.1, p72,
2005.
[7] Z.Chen, M.Rebican, N.Yusa and K.Miya, Fast simulation of ECT signal due to a conductive crack of
arbitrary width, IEEE Trans. Mag., Vol.42, No.4, 683-686, 2006.
[8] Z. Chen, L. Janousek, N. Yusa and K. Miya, A nondestructive strategy for the distinction of natural
fatigue and stress corrosion cracks based on signals from eddy current testing, ASME PVT, 2007 (ac-
cepted).
[9] T.Takagi, et al, Numerical evaluation of correlation between crack size and eddy current testing using a
very fast simulator, IEEE Trans. Mag., Vol.34, No.5, 2581-2584, 1998.
[10] Y.Li, L.Udpa and SS.Udpa, Three dimensional defect reconstruction for eddy current NDE signals using
a genetic local search algorithm, IEEE Trans. Mag., Vol.40, No.2, 410-417, 2004.
[11] S.Norton and J.Bowler, Theory of ECT inversion, J.Appl.Phys., Vol.73, 501-512, 1993.
IOS Press, 2007
Reconstruction of Fatigue Cracks Using

Benchmark Eddy Currents Signals
Maxim MOROZOV a, 1, Guglielmo RUBINACCI b, Salvatore VENTRE c and
Fabio VILLONEc
a
CREATE Consortium, Naples, Italy
b
Ass. EURATOM/ENEA/CREATE, DIEL,
Università degli Studi di Napoli Federico II, Italy
c
Ass. EURATOM/ENEA/CREATE,DAEIMI, Università degli Studi di Cassino, Italy
Abstract. This paper presents fast numerical reconstruction of fatigue cracks in

Inconel600 plates using experimental data of benchmark eddy current
measurements. The cracks are treated as thin surfaces having certain conductivity
and therefore enabling some electric current flowing through them. The
computational technique applied for the crack reconstruction is based on an
integral formulation of the eddy current problem.
Keywords. Eddy currents, nondestructive evaluation, fatigue cracks, numerical

simulation.
1. Introduction
This paper concerns Electromagnetic Non-Destructive Evaluation (ENDE) of real

defects arisen in common construction material on the basis of measured signals
obtained with commercial Eddy Current (EC) instrumentation. The main objective of
the present work has been to validate an original crack reconstruction method based on
the representation of a crack as a two-dimensional surface when applied to real defects.
The numeric method is based on an integral formulation in terms of a two-component
current density vector potential expanded over edge-elements [1]. The method concerns
numerical simulation of the response due to a crack and the solution of the inverse
problem. The exploitation of superposition and a proper choice of the current density
degrees of freedom gives rise to a very efficient numerical implementation.
Furthermore, the fatigue cracks are treated as surfaces having certain conductivity and
therefore enabling some electric current flowing through them [2,3], in contrast to
usual assumption of a crack being perfectly insulating [4].
Experimental data comprising eddy current responses to fatigue cracking (FC) and
stress corrosion cracking (SCC) defects in Inconel600 plates, as well as crack profiles
found by destructive metallographic examination, have been offered as a benchmark to
the scientific community by a research team of International Institute of Universality,
Tokyo, Japan [5]. The data are available for various EC probes. In this paper we
primarily focus on reconstruction of FC cracks from the benchmark study on the basis
1
Corresponding Author: Maxim Morozov, Create Consortium, Via Claudio, 21 – 80125, Napoli, Italy;
E-mail: morozov@unicas.it
268 M. Morozov et al. / Reconstruction of Fatigue Cracks Using Benchmark Eddy Currents Signals
of signals obtained with an absolute type pancake coil [5]. SCC cracks due to their
nature should be treated as volumetric defects and will be studied separately with
application of a special approach reported in [6].
2. Numerical Method
2.1. Thin Insulating Crack
The present method consists in an integral formulation of the eddy currents problem in
terms of a two-component electric vector potential [1]. This approach has a number of
advantages as follows. Using an integral formulation allows us to discretise only the
conducting domains where the eddy currents are induced, automatically enforcing
regularity conditions at infinity. The introduction of the electric vector potential T,
such that the eddy currents density is J=∇×T, ensures that J is solenoidal. The choice
of the two-component gauge minimizes the number of discrete unknowns required. The
equations to be solved are the standard eddy current equations in the frequency domain.
The electric field is:
E = −∂A/∂t−∇ϕ (1)
where ϕ is the scalar electric potential and A is the magnetic vector potential given by:
μ0 J(x' , t )
A(x, t ) =
4π ∫ x − x'
dV ' + A0 (x, t ) (2)
V
c
where μ0 is the magnetic permeability of the vacuum, Vc is the conducting domain and
A0 is the contribution of the external current density. From the numerical point of view,
the formulation is solved using finite elements: a mesh of Vc is given, and an edge
element basis functions Nk is introduced for T:
T = ∑ Ik Nk ⇒ J = ∑ Ik ∇ × Nk (3)
k k
On the one hand, the choice of edge elements allows us to enforce the right
continuity conditions of the various electromagnetic quantities; on the other hand, their
properties are fully exploited both for the gauge and boundary conditions imposition.
The electric constitutive equation is imposed in weak form using Galerkin
approach:
∫ Vc
∇ × N k ⋅ (ηJ + jω A)dV = 0 ∀N k (4)
where η is the resistivity. The term involving the electric scalar potential gives no
contribution thanks to the solenoidality of the test function. Using (2) we have:
(R+jωL) I = U (5)
M. Morozov et al. / Reconstruction of Fatigue Cracks Using Benchmark Eddy Currents Signals 269
where I = {Ik}, U = {Uk} and
μ0 ∇ × N i ( x) ⋅ ∇ × N j ( x' )
Lij =
4π ∫∫
Vc Vc x − x'
dV dV ' (6)
R ij = ∫ ∇ × Ni ⋅η∇ × N j dV (7)
Vc
Ui = − ∫ ∇ × Ni ⋅ jω A 0 dV (8)
Vc
Supposing that the crack has a negligible thickness, it can be schematised as a

surface (not necessarily planar), discretised via a set of finite element facets (defect
pixels), where the normal component of the current density must vanish. In order to
reduce the computational load, and exploiting linearity, we use superposition: the total
current density is the sum of the solution computed in absence of crack (unperturbed
solution J0) plus the perturbation δJ due to the presence of the defect.
J = J 0 + δJ (9)
In particular, on the insulating crack surface, since the total current density normal
component must be zero, we impose that
J ⋅ n = 0 ⇒ δJ ⋅ n = −J0 ⋅ n, (10)
where n is the normal to the crack. This approach offers the great advantage that J0 can
be calculated either analytically, or numerically using the scheme described above on a
mesh that does not depend on the crack geometry. Conversely, when solving for δJ the
mesh must account for the crack only, so that the mesh refinement is required only
close to the crack, regardless of the position of the exciting source. Due to the
properties of edge elements, the set δ G of perturbation currents crossing the crack
facets (that must be equal the unperturbed currents −G0) can be written as [4]:
δG = P δI (11)
where δI are the coefficients of the expansion of δJ in terms of edge elements, and P is
a (m,n) sub-matrix of the edge-facet incidence matrix with coefficients 0, +1 or -1.
The degrees of freedom of the edge element expansions are in fact related to the
line integrals of T along the edges, and the circulation of T along a closed line gives
the total current (flux of ∇×T) linked with the line. We then make a change of variables
[4]:
δI = K δX + P+ δG (12)
where K is a (n,n-m) matrix given by an orthonormal basis for the null space of P, P+ is
the pseudo-inverse of P, and δX is a new set of unknowns, providing no net current
flowing through the crack. Galerkin’s procedure in terms of these new variables yields:
KTZK δX = KTZ P+ G0 (13)
solving which we have δX, and hence δI from (12).

Knowing δ I it is possible to compute the reaction field. In particular, the
impedance change of the excitation coil is given by
δZ = −U0TδI/Is2 (14)
where U0 is the applied voltage and Is is the impressed current.
2.2. Treatment of Thin Conducting Cracks
If the cracked region is partly conducting, some current can flow across the crack, the
measured electromagnetic signal is weakened, and the detection becomes more difficult.
A crack with leakage can be modeled as a region with a different resistivity, i.e., a
resistivity higher than in the conducting specimen (but not infinite). The integral
formulation has also been applied to the treatment of this case in [2,3]. As in the
previous cases, having defined the unperturbed problem (no defect is present), the
perturbation is defined by superposition as
⎧ η δ J + (ηc − η0 )J 0 inVc
δ E = − jωδA − ∇δϕ = ⎨ c (15)
⎩η0δ J inV0
where V0 is the conducting body with resistivity η0, Vc is the planar crack of width dc,
filled by a material with resistivity ηc>η0. It is assumed that V0∪ Vc is simply
connected. The weak form of the problem in terms of the electric vector potential leads
to the following numerical formulation in the limit dc→0 and ηc→+∞, in such a way
that dc ηc remains finite [2]:
(jωL + R+ Rc) δI = −Rc I0 (16)
where δI = {δIk}, I0 = { I0k } and
R ij = ∫ ∇ × Ti ⋅η0∇ × Tj dV , R cij = d cηc ∫ (∇ × Ti ⋅ nˆ )(∇ × Tj ⋅ nˆ ) dV (17)

V0 Σd
The inverse problem is solved on the basis of a priori information on the crack
region by iterating the forward analysis until minimising its error with respect to EC
signals.
3. Experimental Data and Samples
The studied EC test specimens represent Inconel600 plates with fatigue cracks [5]. The
electric conductivity of Inconel600 is assumed to be 1 MS/m. The dimensions of the
specimens (mm) are given in Figure 1. FC were produced into the plates by cyclic
three-point bending. FC fabrication conditions are given in Table 1. EC signals due to

the FC were obtained with an absolute type pancake coil probe, shown in Figure 2. The
excitation frequency of EC testing with the pancake coil was 100 kHz and its lift-off
above the surface of a sample was 1 mm. After EC testing crack profiles were found by
destructive metallographic examination. Since by conditions of the benchmark test
neither the excitation current flowing through the coil, nor the phase shift and
amplification of the measured signals are known, the numerical results have been
calibrated with respect to an artificial notch produced by Electrical Discharge
Machining (EDM). The EDM notch has a rectangular profile of 10 mm in length, 0.3
mm in width, and 5.0 mm in depth (see Figure 3). The calibration process consists in
finding a magnitude scaling factor and an appropriate phase shift at which the numeric
simulation result is in agreement with the measured signal for the EDM notch. Then the
FC cracks are numerically reconstructed maintaining the scale factor and phase shift
found by calibration.
Table 1. Fatigue crack fabrication conditions
Specimen Max. loading (t) Min. loading (t) Cycles Crack Length(mm)
TP07 2.5 0.5 60,000 18
TP10 3.0 0.5 30,000 16
TP19 2.5 0.5 120,000 22
TP21 2.7 0.5 87,000 32
Figure 1. Specimen layout Figure 2. Absolute pancake coil
4. Results and Discussion
Figure 3 shows the finite elements mesh (3378 elements, 6320 active edges),
illustrating the discretisation of the solution region, and the profile of the EDM notch.
Black and white facets denote the crack search region, with black facets corresponding
to undamaged conducting material and white facets corresponding to the defect (EDM
notch) with zero conductivity. The calibration signal (represented as the real and
imaginary components) due to the EDM notch is shown in Figure 4, with a magnitude
scaling factor and an appropriate phase shift being applied to the simulation result in
order to bring it to agreement with the measured signal. Certain error occurs partly due
to the low-pass filtering applied to the measured signal. The magnitude scale factor and
phase shift found by calibration are maintained when numerically reconstructing the
FC cracks. However, in contrast to completely non-conducting EDM notch, natural
fatigue cracks might have certain conductivity and therefore enable some electric
current flow across their surface. The conductivity of fatigue cracks might exist either
owing to material relaxation after removing loading, when sides of the crack come to
partial electrical contact with each other (which is most likely for the very small FC
cracks), or owing to deposits of conductive materials inside a crack. Consequently,

partial conductivity of a crack, denoted by grey facets, should be introduced when
reconstructing fatigue cracks. Dependence of a crack signal on the crack resistivity is
shown in Figure 5, with the response values being normalised to the absolute value of
the crack signal when the crack resistivity is infinite. The crack’s thickness-resistivity
product dcηc = 5.E-9 Ωm2 has been optimised for the thinnest FC crack TP07 whose
low EC signal indicates current leakage (Figures 8-9). In the case of a large crack
partial conductivity occurs on the boundary of a FC crack profile where the width of a
crack becomes very small. Metallographic profiles of various fatigue cracks and the
respective numerically reconstructed profiles, as well as comparison of the
corresponding measured and simulated eddy current signals (represented as the real and
imaginary components) are given in Figures 6-13. The reconstruction of crack profiles
as well as identification of the partially conductive facets has been done by minimising
error between numerical and experimental signals due to a fatigue crack.
5. Conclusions and Outlook
A numerical method has been presented for simulating EC signals of partially

conductive thin cracks in conductive materials. The method enables real-time accurate
reconstruction of fatigue cracks on the basis of measured signals obtained with
conventional EC instrumentation.
The future development will address:
• implementation of a genetic algorithm for automated crack reconstruction;
• reconstruction of defects using signals obtained with other EC probes, for
instance uniform EC probe [5];
• reconstruction of SCC cracks as 3D volumetric defects with finite
conductivity.
Figure 3. Profile of the EDM notch superimposed on the finite elements mesh: grey elements represent the
finite elements mesh; black and white facets denote the crack search region: black facets correspond to
undamaged material; white facets correspond to the defect (EDM notch)
Figure 4. Measured and simulated eddy current signal obtained by a pancake coil due to an EDM notch
Figure 5. Dependence of crack signal on crack resistivity (signals are normalised to the absolute value
of the crack signal when the crack resistivity is infinite)
Figure 6. Fatigue crack profile in specimen TP10: white and grey facets represent numerically reconstructed
crack; white facets have zero conductivity; grey facets enable electric current flow through them (dcηc=5.E-9
Ωm2 ); black facets represent undamaged material; dashed line represents metallographic profile of the crack
Figure 7. Measured and simulated EC signal obtained by a pancake coil due to fatigue crack in TP10
Figure 8. Same as Figure 6, but for specimen TP07
Acknowledgments
The experimental data have been kindly provided by Dr. Noritaka Yusa of International
Institute of Universality, Tokyo, Japan.

References
[1] R. Albanese, G. Rubinacci, A. Tamburrino, F. Villone, “Phenomenological approaches based on an

integral formulation for forward and inverse problems in eddy current testing”, Int. J. of Applied
Electromagnetics and Mechanics, Vol. 12, No. 3-4/2000, pp. 115-137
[2] F. Villone, “Simulation of Thin Cracks with Finite Resistivity in Eddy Current Testing”, IEEE Trans.
Mag., Vol. 36, No. 4, July 2000, pp. 1706-1709
[3] F. Villone and N. Harfield, “Simulation of the effects of current leakage across thin cracks”,
Electromagnetic Non-destructive Evaluation (IV), S. Udpa et al. (Eds.), IOS Press, 2000.
[4] R. Albanese, G. Rubinacci, F. Villone, “An integral computational model for crack simulation and
detection via eddy currents”, J. of Comp. Phys., Vol. 152, pp. 736-755 (1999)
[5] N. Yusa, L. Janousek, Z. Chen, K. Miya. “Diagnostics of stress corrosion and fatigue cracks using
benchmark signals”, Materials Letters 59 (2005), 3656-3659
[6] M. Morozov, G.Rubinacci, A.Tamburrino, S.Ventre, “Numerical Models of Volumetric Insulating
Cracks in Eddy-Current Testing With Experimental Validation”, IEEE Trans. Mag., Vol. 42, No. 5,
May 2006, pp. 1568-1576
IOS Press, 2007
2D Axisymmetric ECT Simulation of the

World Federation’s First Eddy Current
Benchmark Problem
Ovidiu MIHALACHE a,1, Masashi UEDA a, Takuya YAMASHITA a
a
Fast Breeder Reactor Research and Development Center,
Japan Atomic Energy Agency, Tsuruga, Japan
Abstract. This paper describes the numerical results of the first eddy current
(ECT) benchmark problem proposed by the World Federation of Nondestructive
Evaluation Centers in the year 2000. The ECT problem, consisting in predicting
the signal from axisymmetric defects in an Inconel tube with ferromagnetic
support plate, is solved using two simulation softwares: a 2 dimensional
axisymmetric code (2D ECT) developed in-house and the commercial software
FEMLAB, both codes being based on the Finite Element Method (FEM). The
simulations compare the numerical results obtained with the two codes for several
eddy current coils and support plate configurations and frequencies ranging from 1
to 200 kHz. Numerical results from both codes agree very well only when
carefully attention is paid to details related to boundary limits, finite element mesh
structure, and interface conditions.
Keywords. Eddy current, Benchmark, Finite Element, 2D axisymmetric code
1. Introduction
Several eddy currents (ECT) benchmarks were proposed starting from the year 2000
by the World Federation of Nondestructive Evaluation Centers (WFNDEC) [1] in
order to offer a clear basis to compare and verify the numerical results obtained with
different simulations codes. Using these benchmarks the main difficulties arising in
accurate numerical ECT simulations and their limitations can be show up, opening new
research activities.
The present paper addresses the WFNDEC’s first eddy current benchmark
problem. The numerical simulations are conducted using two ECT codes, both based
on the finite element method (FEM). The first one was developed in-house [2] and it is
a 2-dimensional axisymmetric ECT code (2D ECT), using both harmonic and time-
transient approximation, and based on the magnetic vector potential A. The second
code is a commercial FEM package, named FEMLAB® [3], and also based on the
magnetic vector-scalar potential A-V formulation.
1
Corresponding Author: Ovidiu Mihalache, Japan Atomic Energy Agency, Advanced Nuclear System
Research and Development Directorate, Fast Breeder Reactor Research and Development Center, 1 Shiraki,
Tsuruga-shi, Fukui-ken, 919-1279, Japan, E-mail: mihalache.ovidiu@jaea.go.jp
276 O. Mihalache et al. / 2D Axisymmetric ECT Simulation
The results of the present benchmark were also presented in the past by two
different researchers: a Polish group [4] and a group from Iowa State
University/Michigan State University [5] revealing some artifacts and small
discrepancies between results. In the paper are analyzed the main simulation conditions
to accurate model the benchmark problem. The results are presented in a similar way
with the data published by the other researchers groups in order to easy compare them.
2. Calculation of Coil Impedance in the 2D-ECT Axisymmetric FEM Code
The FEM model of the two-dimensional axisymmetric 2D-ECT code is based on the
Maxwell electromagnetic equation using A-V formulation with electric scalar potential
V set to zero:
§1 · wA
u ¨¨ u A ¸¸ J s (t ) V ( ) (1)
©P ¹ wt
where Js is the source excitation current density, V is the material electrical
conductivity while P is its relative magnetic permeability. The code is based on the
nodal implementation, calculating the values of the magnetic vector potential A at the
nods of the finite elements cell.
The eddy current signal in the pick-up coils is calculated taking into account the
time variation of magnetic flux density:
wB
e .m . f .
³ wt dS
S
, (2)
where S=S n, S is the coil surface, n is normal at the surface S and C is the contour of
the surface S.
In the two dimensional axisymmetric approximation, the coil impedance Z is
derived in the FEM model by integrating Equation 2 over the coil surface, surface
which was divided in n1 u n2 cells:
j 2SZ N 0
Z
n1n2 I 0 ¦ ³³ rAT (r, z)drdz
k S (k )
(3)
where S(k) is the k-cell area, I0 is the current running trough the wire, N0 is the number
of turns in the coil.
3. Description of the World Federation 1st Eddy Current Benchmark Problem
The WFNDEC’s first eddy current benchmark problem is presented in Figure 1. It

consists of the simulation of the eddy current response from an axisymmetric
circumferential defect in an Inconel 600 tube and under a support plate (SP) made of
ferromagnetic steel. The two identical air-cored coils (each of them containing 1000
turns with a 10 mA AC excitation current passing through the wire), used both as eddy
current excitation and detection system, are connected differentially and move along an
infinite long tube.
The frequency of the excitation system is: 1, 10, 100 and 200 kHz.
O. Mihalache et al. / 2D Axisymmetric ECT Simulation 277
Figure 1. Configuration of the World Federation 1st eddy current benchmark for axisymmetric defect.
The tube and SP electrical conductivity are 106 S/m while their relative magnetic
permeabilities are 1 and 1000, respectively. The 100 value of relative SP permeability
was also considered in the simulations to compare the results with the data presented
by other researchers. The parameters of the system change by varying the distance
between ECT coils or the gap between SP and tube, as is shown in Figure 1. The defect
is located on the outer tube surface and centered under SP.
4. Numerical Simulations of the Benchmark Problem
Numerical simulations of the benchmark problem were conducted using two

axisymmetric FEM codes: a professional one, FEMLAB, and another one developed
in-house and named 2D ECT code. In all simulations, the tube length was 200 mm,
with the defect and SP located in the middle area of tube. In order to avoid in
simulations the edge effects from the external air boundary, the tube was supposed to
be symmetrically surrounded by a cylindrical air model 300 mm long and 200 mm in
diameter. During simulations, the magnetic vector potential was set to zero on the
external air boundary points while the symmetry condition was imposed at the edge
defined by zero radius. The coils move inside the tube, scanning the SP and defect area
in 120 steps each of them 0.25 mm apart.
The mesh structure used in numerical simulations is the following: FEMLAB uses
an automatic mesh generator at each step, while the 2D-ECT code use a single mesh
with a highly regular pattern and fine tuned to cover all the steps. Both meshes consist
in around 130,000 quadratic triangular elements.
a) b)
Figure 2. Axisymmetric defect in an Inconel tube under a ferromagnetic (P=1000)
support plate. Simulations of the magnetic vector potential A contours plot (real part) using:
a) FEMLAB code; b) 2D ECT code.
The same mesh is used to simulate the eddy currents inside the tube and SP for all
frequencies. Both codes run on a the same Pentium Xeon PC machined, in 1 GB of
memory, running time being around 1.5 hours for FEMLAB and 2 minutes for 2D-
ECT to simulate a complete impedance signal trajectory.
Numerical simulations were performed for various distances of the air gap
between tube and support plate, coil geometries and excitation frequencies. In order to
facilitate the display of the signal impedance trajectory, the result for every frequency
was normalized with respect to the maximum amplitude for each complex impedance.
Also, since the coil scan is symmetric, only half of the impedance of the trajectory is
displayed corresponding to the movement of coils system from outside of SP to the
middle point of SP. Because SP signal has an intricate shape, the signal impedance
phase was defined as the phase of the point in the impedance trajectory lobe
corresponding to the maximum signal amplitude.
In Figure 2 is shown a comparison between numerical simulations of the magnetic
vector potential contour lines (real part) using FEMLAB and 2D-ECT codes. The SP
gap and the distance between coils is 1 mm. The directions of currents flowing in the
two coils are oppositely to each other and the excitation frequency is 100 kHz. Due the
high relative magnetic permeability (P=1000) of SP, eddy currents slightly penetrate
SP. The distribution of magnetic vector potential is very similar, using both codes,
validating the correctness of the simulations. The small artifacts in the contour lines,
near SP boundary, in FEMLAB simulation are due to the mesh size irregularities near
SP interface which could not be corrected using the FEMLAB automatic mesh
generator.
In Figure 3 are presented the impedance trajectories in the absence of the support
plate for various distances between coils: 0, 0.5 and 1 mm. It can be seen that the
agreement between FEMLAB and 2D-ECT simulations is very good at all excitations
frequencies.
In Figures 4 and 5 are shown the comparison between FEMLAB and 2D-ECT
simulations when the SP relative magnetic permeability is P=100 and the distance
between SP and tube is 0 and 0.5 mm. It was found that the differences between results
are very small in all cases, even if the FEMLAB results were affected by the mesh
errors, since the mesh structure changed at every step. Also, the phase of the signal
change with the frequency and is less affected by the gap between coils or the SP gap.
Figure 3. Comparison between 2D-ECT(continuous line) and FEMLAB (circle continuous

line) in the absence of support plate for coil gap distance: 0, 0.5 and 1 mm (from the left to
the right picture).

line), with support plate (relative magnetic permeability P=100), SP gap=0 mm and for coil
gap distance: 0, 0.5 and 1 mm (from the left to the right picture).

line), with support plate (relative magnetic permeability P=100), SP gap=0.5 mm and for
coil gap distance: 0, 0.5 and 1 mm (from the left to the right picture).

line), with support plate (relative magnetic permeability P=1000), SP gap=0 mm and for

line), with support plate (relative magnetic permeability P=1000), SP gap=0.5 mm and for
In Figures 6 and 7 are shown the same comparison, as in Figure 4 and 5, between
FEMLAB and 2D-ECT simulations of the impedance lobby trajectories but when the
SP relative magnetic permeability was set to P=1000. Even in this case the agreements
between simulations are very good but very small artifacts are visible at higher
frequencies 100 and 200 kHz. The main reason is that at higher frequency the eddy
current penetration in SP decreases and the structure mesh and mesh size near SP edge
becomes more important in the SP signal simulation.
Table 1. Amplitudes A and phase M (degrees) of the defect impedance in the absence of the SP.
Coil Code 1 kHz 10 kHz 100 kHz 200 kHz
gap name A[ǡ] M A[ǡ] M A[ǡ] M A[ǡ] M
[mm]
0 FEMLAB 0.0016 177.4 0.1478 153.4 3.499 58.24 5.361 13.80
2D ECT 0.0016 177.7 0.1478 153.4 3.499 58.24 5.361 13.78
0.1 FEMLAB 0.0017 177.2 0.1545 153.4 3.643 55.46 5.566 13.76
0.5 FEMLAB 0.0019 177.2 0.1798 153.3 4.179 55.25 6.289 13.38
2D ECT 0.0019 177.7 0.1798 153.4 4.177 55.22 6.288 13.33
1 FEMLAB 0.0023 177.6 0.210 152.7 4.736 54.81 6.996 10.42
2D ECT 0.0023 177.7 0.210 152.7 4.735 54.79 6.996 10.38
Table 2. Amplitudes A and phase M (degrees) of the defect impedance in the presence of the SP (P=100).
SP Coil Code 1 kHz 10 kHz 100 kHz 200 kHz
gap gap name A[ǡ] M A[ǡ] M A[ǡ] M A[ǡ] M
[mm] [mm]
0 FEMLAB 1.657 84.10 13.39 55.26 25.00 -47.05 17.43 -95.36
2D ECT 1.656 84.11 13.39 55.27 24.97 -47.18 17.31 -95.81
0.1 FEMLAB 1.733 84.09 13.99 55.21 26.10 -47.02 18.29 -95.38
0 0.5 FEMLAB 2.023 84.04 16.31 54.94 29.85 -48.38 20.65 -97.23
2D ECT 2.022 84.05 16.30 54.95 29.78 -48.56 20.46 -97.61
1 FEMLAB 2.361 83.97 18.97 54.55 33.85 -50.34 23.06 -99.42
2D ECT 2.359 83.98 18.96 54.56 33.82 -50.22 23.02 -99.82
0 1.242 83.75 9.901 53.79 17.32 -52.33 11.63 -102.4
0.5 2D ECT 1.521 83.73 12.13 53.53 20.86 -53.42 13.85 -104.1
0.5 1 1.785 83.67 14.19 53.21 23.95 -54.76 15.78 -105.86
0 0.957 83.42 7.533 52.31 12.28 -57.99 7.989 -109.9
0.5 2D ECT 1.176 83.37 9.249 52.06 14.90 -58.86 9.613 -111.4
1 1 1.384 83.31 10.88 51.79 17.25 -59.95 11.03 -113.3
Table 3. Amplitudes A and phase M (degrees) of the defect impedance in the presence of the SP (P=1000).
SP Coil Code 1 kHz 10 kHz 100 kHz 200 kHz
gap gap name A[ǡ] M A[ǡ] M A[ǡ] M A[ǡ] M
[mm] [mm]
0 FEMLAB 1.745 86.26 15.51 60.48 33.05 -40.71 24.24 -84.52
2D ECT 1.743 86.26 15.52 60.34 32.43 -39.73 23.62 -84.79
0.1 FEMLAB 1.824 86.25 16.22 60.55 35.06 -41.07 25.51 -88.1
0 0.5 FEMLAB 2.130 86.22 18.92 60.31 39.80 -41.76 28.92 -84.93
2D ECT 2.128 86.23 18.93 60.14 38.85 -42.07 27.94 -86.01
1 FEMLAB 2.486 86.19 22.04 60.11 45.46 -41.34 33.10 -88.13
2D ECT 2.484 86.19 22.07 59.90 44.57 -43.23 31.81 -88.02
0 1.311 86.14 11.62 59.86 23.48 -42.70 16.84 -88.27
0.5 2D ECT 1.608 86.10 14.24 59.37 28.33 -43.51 20.17 -89.14
0.5 1 1.888 86.07 16.70 59.18 32.64 -44.53 23.03 -90.16
0 1.015 86.00 8.953 58.95 17.26 -47.21 11.98 -93.54
0.5 2D ECT 1.248 85.97 11.00 58.80 20.97 -47.82 14.44 -94.43
1 1 1.468 85.94 12.93 58.38 24.33 -48.55 16.60 -95.49
Figure 8. The error distribution between numerical simulations of impedance amplitude

and impedance phase using the two codes FEMLAB and 2D ECT.
By refining the mesh near SP-air interface and in the SP-tube gap, both codes
could provide very similar and reliable results. In order to have a quantitative
comparison of the impedances signal amplitude and phases, their values are listed in
Tables 1, 2 and 3 for both codes in the presence or absence of the support plate. As it
can be seen, the values are very similar with the ones calculated by others researchers.
The distribution of differences in the numerical simulations of the signal using the
two codes is illustrated in Figure 8.
As the support plate relative magnetic permeability increases, the errors between
simulations using the two codes increases but the maximum difference in the amplitude
do not exceed 6%, even at 200 kHz. Also, the differences in the signal phases is very
small, less than 2.5 degrees.
5. Discussions on the Accuracy of Numerical Simulations
Accurate simulations of the eddy current signal of defects under SP, using both codes
FEMLAB and 2D-ECT, could be obtained only in special conditions, which are
explained in the following.
First, it was found that as the defect size is smaller there is a greater effect of the
external boundary conditions to the impedance signal trajectory. The simulated defects
signal did not change their shapes when the dimensions of air-cylinder domain,
symmetrically surrounding the tube and SP, decreased from a 500 mm radius to 100
mm radius and from a 2000 mm length to 300 mm length. Further reduction of the
domain size had a significant effect on the signal shape, even when very fine meshes
and highly interpolated elements were used in the numerical simulations.
Second, the tube length greatly influences the signal trajectory, due to tube edge
effect. The minimum tube length was found to be 200 mm.
Third, for high relative magnetic permeability of SP, even if the boundary of air-
cylinder surrounding the tube is far from the tube and SP (500 mm), accurate
simulations of SP signal could be obtained only for a fine and regular small size mesh
near SP-air interface.
Fourth, linear triangular elements, in both codes, could not simulate well the
impedance signal trajectory in all cases, especially in the presence of SP, even for a
very fine mesh structure around SP-air interface and a dense mesh with up to a 500,000
triangular elements. In Figure 9 it is presented the mesh structure (500,000 linear

triangles) around coil, defect and SP for FEMLAB and 2D-ECT code. In Figure 10 it is
shown the comparison between the impedance trajectory signal calculated with both
quadratic and linear elements using FEMLAB and 2D ECT code. It can be seen that
higher accuracy is obtained when a regular pattern mesh is used with linear triangle
cells.
Figure 9. Structure of mesh (500,000 linear triangles) for FEMLAB (left) and 2D-ECT code (right)
a) b) c)
Figure 10. Comparison between simulations using 130,000 quadratic triangular elements
(a) and 500,000 linear triangular elements in FEMLAB (b) and 2D-ECT code (c).
Simulations were performed in the presence of SP (P=100), SP gap=0.0 mm and coil gap
distance: 1 mm
6. Conclusions
The WFNDEC’s first eddy current benchmark was simulated using two finite elements
codes 2D-ECT (developed in-house) and FEMLAB (commercial code) showing
excellent agreements between simulations results. The maximum error in calculating
signal trajectory with the two codes was less than 6%, the very good agreement being
obtained by carefully monitoring the mesh structure and mesh size near SP-air
interface and the size of boundary domain around tube.
References
[1] World Federation of Nondestructive Evaluation Centers, “http://www.wfndec.org/Default.htm”, 2000.
[2] O Mihalache, “Advanced Remote Field Computational Analysis of Steam Generators Tubes”, Studies in
Applied Electromagnetics and Mechanics, Electromagnetic Nondestructive Evaluation 26, (2006)
pp.220-227.
[3] FEMLAB manual, Comsol Inc., 2005.
[4] R. Sikora, R. Palka, Review of Progress in QNDE, 21, (2001) pp.1909.
[5] Y. Tian, Y. Li, Z. Zeng, L. Udpa, S. S. Udpa,”Simulation of the World Federation’s First Eddy Current
Benchmark Problem ” Review of Progress in QNDE, 23, (2003) pp.1560-1566.
IOS Press, 2007
Author Index
Abbasi, K. 117 Lefevre, A. 13
Abbate, C. 75 Lesselier, D. 49
Altpeter, I. 170 Łopato, P. 109
Ara, K. 209, 217 Marek, T. 147
Arun Kumar, S. 131 Maurice, L. 49
Arunachalam, K. 125, 131 Melapudi, V.R. 125
Bayani, H. 57 Mihalache, O. 275
Benson, J. 131 Miya, K. 83, 91, 147, 259
Berthiau, G. 13 Morozov, M. 75, 267
Bowler, J. 41 Moses, A.J. 4, 178
Capova, K. 147 Nair, N. 67
Chady, T. 109, 233 Nicolas, A. 49
Chen, Z. 259 Nishiyama, R. 243
Cheng, W. 251 Ohtani, T. 153, 201
Dobmann, G. 170 Paillard, S. 25
Doirat, V. 13 Patel, H.V. 178
Enokizono, M. 109, 225, 233 Pávó, J. 49, 139
Fouladgar, J. 13 Perrin, S. 91
Gasparics, A. 139 Pichenot, G. 25
Gombarska, D. 147 Pinassaud, J. 33
Gyimóthy, S. 139 Pirlog, M. 170
Hashizume, H. 117 Poulakis, N. 41
Hübschen, G. 170 Prémel, D. 49
Ito, S. 117 Psuj, G. 109
Janousek, L. 83, 147 Ramuhalli, P. 131
Jiles, D.C. 4 Rubinacci, G. 33, 75, 267
Joshi, A. 131 Sasada, I. 57
Joubert, P.-Y. 33 Sebestyén, I. 139
Kai, Y. 225 Sikora, R. 109
Kalincsák, Z. 139 Szielasko, K. 170
Kamada, Y. 209, 217 Takács, J. 139
Kanemoto, S. 251 Takagi, T. 99, 186
Kawaguchi, Y. 153 Takahashi, S. v, 209, 217
Keprt, J. 193 Takahashi, Y. 99
Kikuchi, H. v, 209, 217 Tamburrino, A. 33, 67, 75
Kiss, I. 139 Teramoto, T. 163
Kobayashi, S. 209, 217 Theodoulidis, T. 41, 57
Kojima, F. 243 Tian, G.Y. 193
Komura, I. 251 Tian, H. 99
Kopp, M. 170 Tomáš, I. 186
Kurokawa, M. 3 Tsuboi, H. 139
Kurosaki, K. 153 Tsuchida, Y. 109, 225, 233
Lambert, M. 25 Uchimoto, T. 99, 186
284
Udpa, L. 67, 125, 131 Williams, P.I. 178

Udpa, S.S. 67, 125, 131 Wilson, J. 193
Uduebho, O. 131 Yamanaka, S. 153
Ueda, M. 275 Yamashita, T. 275
Ventre, S. 33, 75, 267 Yasutake, T. 233
Vértesy, G. 9, 139, 186 Yuki, K. 117
Villone, F. 267 Yusa, N. 83, 91, 147, 259
Voillaume, H. 25

S. Takahashi - Electromagnetic Nondestructive Evaluation (X) (Studies in Applied Electromagnetics and Mechanics) - IOS Press (2007)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

S. Takahashi - Electromagnetic Nondestructive Evaluation (X) (Studies in Applied Electromagnetics and Mechanics) - IOS Press (2007)

Uploaded by

Copyright:

Available Formats

ELECTROMAGNETIC NONDESTRUCTIVE

Amsterdam • Berlin • Oxford • Tokyo • Washington, DC

Distributor in the UK and Ireland Distributor in the USA and Canada

PRINTED IN THE NETHERLANDS

referees listed below, especially regional editors, Professors G. Rubinacci, T. Takagi

Seiki Takahashi Hiroaki Kikuchi

Prof. Luming Li Prof. Seiki Takahashi

Dr. Ovidiu Mihalache Prof. Tokuo Teramoto

Dr. Kenzo Miya Prof. Theodoros Theodoulidis

Engineering, Michigan State University, Environmental Engineering, Graduate

Inspection Experience of Steam Generator Tubes with Intelligent ECT Probe 3

ECT Modeling and Simulation

Skin and Proximity Effects in ECNDT Sensors 13

Eddy Current Testing and Technique

Experimental Extraction of Time-of-Flight from Eddy Current Test Data 67

Industrial Applications and New Methods

Evaluation of Fatigue Loaded Conducting Structures Using Selected

Application of Thermoelectric Power Measurement to Nondestructive Testing 153

NDE by Magnetism and Magnetics

Accurate Detection of Material Degradation of Stainless Steel by ECT Sensor 163

Inverse Problem and Benchmark

Nondestructive Evaluation for Material Degradation of Steel Sample Using

Author Index 283

Inspection Experience of Steam Generator

(summarized by reviewer, K. Ara)

Origin, Measurement and Application of

1. Origin of Barkhausen Noise

concluded reversals were occurring in volumes in the nanometer range [4] It is

The discontinuities were attributed to rotation of magnetization within a domain

Some correlation has been found between BN measured in electrical steel

2. Measurement of Barkhausen Noise

Recent improvements in digital technology, particularly with data acquisition cards,

An alternative method for Barkhausen measurement is to use a wound ferrite core

A new initiative: Universal Network for

Magnetic measurements are suitable for characterization of changes in structure of

Degradation of Gas Pipelines, Railways, Bridges and Other Steel Constructions

Standardization of Magnetic Properties

Skin and Proximity Effects in ECNDT

Abstract. A semi-analytical forward model based on the electromagnetic coupled

The CCM is used to solve axisymmetric conﬁgurations. The sensor is constituted of a

Only sector discretization

Layer and sector discretization

Lift-oﬀ Load (σ, μ)

Figure 1. System discretization.

Figure 2. Equivalent circuit for two discretizations.

Wire 1 Wire 2 Wire Nw

RB1,2 LB1,2 RB2,2 LB2,2 RBNw ,2 LBNw ,2

Figure 3. Equivalent electrical scheme.

2.1. Skin Effect in a Cylindrical Wire

Normalized current density |Q(r)/Q |

0.5 0.5 300 kHz

0.3 0.3 500 kHz

Figure 4. Outer Radius R = 0,5mm. Figure 5. Outer Radius R = 0,25mm.

2.2. Proximity Effect in a Multi Coil Sensor

Figure 6. Circular coil inducing the magnetic potential vector dA in P.

inductance M s is calculated by integrating (7) with respect to r. The "surface" S d in (8)

Ibob = Sdi,k Qi,k (11)

(Rsi,1 + j ωMsi,1 )Qi,1 − Rsi,k Qi,k − j ωMsp,q Qp,q = 0 (13)

(a) Layer and sector discretization. (b) Only sector discretization.

Table 1. Reduce discretization results and comparison with the experiment.

3.1. Introduction of a Ferritic Core

(μ0 K’(M )) + n × dB’ =− n × dB0 (16)

Figure 8. Inductor + Ferritic core contributions to magnetic induction.

³uT k (KJ jZA)dV 0 Tk (2)