Anda di halaman 1dari 497

ELECTROMAGNETIC NONDESTRUCTIVE

EVALUATION (XIV)
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 35
Previously published in this series:

Vol. 34. S. Wiak and E. Napieralska Juszczak (Eds.), Computer Field Models of
Electromagnetic Devices
Vol. 33. J. Knopp, M. Blodgett, B. Wincheski and N. Bowler (Eds.), Electromagnetic
Nondestructive Evaluation (XIII)
Vol. 32. Y. K. Shin, H. B. Lee and S. J. Song (Eds.), Electromagnetic Nondestructive
Evaluation (XII)
Vol. 31. A. Tamburrino, Y. Melikhov, Z. Chen and L. Udpa (Eds.), Electromagnetic
Nondestructive Evaluation (XI)
Vol. 30. S. Wiak, A. Krawczyk and I. Dolezel (Eds.), Advanced Computer Techniques in
Applied Electromagnetics
Vol. 29. A. Krawczyk, R. Kubacki, S. Wiak and C. Lemos Antunes (Eds.), Electromagnetic
Field, Health and Environment Proceedings of EHE07
Vol. 28. S. Takahashi and H. Kikuchi (Eds.), Electromagnetic Nondestructive Evaluation (X)
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. Prmel and D. Lesselier (Eds.), Electromagnetic Nondestructive
Evaluation (VIII)
Vol. 23. F. Kojima, T. Takagi, S.S. Udpa and J. Pv (Eds.), Electromagnetic Nondestructive
Evaluation (VI)
Vol. 22. A. Krawczyk and S. Wiak (Eds.), Electromagnetic Fields in Electrical Engineering
Vol. 21. J. Pv, G. Vrtesy, T. Takagi and S.S. Udpa (Eds.), Electromagnetic Nondestructive
Evaluation (V)
Vol. 20. Z. Haznadar and . tih, Electromagnetic Fields, Waves and Numerical Methods
Vol. 19. J.S. Yang and G.A. Maugin (Eds.), Mechanics of Electromagnetic Materials and
Structures

Volumes 1 6 were published by Elsevier Science under the series title Elsevier Studies in
Applied Electromagnetics in Materials.

ISSN 1383 7281 (print)


ISSN 1879 8322 (online)
Electromagnetic Nondestructive
Evaluation (XIV)

Edited by
Tomasz Chady
West Pomeranian University of Technology, Szczecin, Poland

Stanisaw Gratkowski
West Pomeranian University of Technology, Szczecin, Poland

Toshiyuki Takagi
Tohoku University, Japan
and
Satish S. Udpa
Michigan State University, USA

Amsterdam Berlin Tokyo Washington, DC


2011 The authors and IOS Press.

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 60750 749 9 (print)


ISBN 978 1 60750 750 5 (online)
Library of Congress Control Number: 2011928887

Publisher
IOS Press BV
Nieuwe Hemweg 6B
1013 BG Amsterdam
Netherlands
fax: +31 20 687 0019
e mail: order@iospress.nl

Distributor in the USA and Canada


IOS Press, Inc.
4502 Rachael Manor Drive
Fairfax, VA 22032
USA
fax: +1 703 323 3668
e mail: iosbooks@iospress.com

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 (XIV) v
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.

Preface
This volume contains selected papers from the fifteenth International Workshop on
Electromagnetic Nondestructive Evaluation, which was held in Szczecin, Poland, from
June 13 to 16, 2010.
Previous ENDE Workshops have been held in: London, United Kingdom (1995);
Tokyo, Japan (1996); Reggio Calabria, Italy (1997); Chatou, France (1998); Des
Moines, United States (1999); Budapest, Hungary (2000); Kobe, Japan (2001); Saar-
brcken, Germany (2002); Paris, France (2003); East Lansing, United States (2004);
Iwate, Japan (2006); Cardiff, United Kingdom (2007); Seoul, Korea (2008); Dayton,
United States (2009).
The aim of the workshop, organized by the West Pomeranian University of Tech-
nology, Szczecin, Poland and the Japanese Society of Maintenology, was to bring to-
gether scientists from universities and research institutions conducting in-depth re-
search into the basics of electromagnetic non-destructive evaluation (ENDE) on the one
hand, and engineers presenting practical problems and industrial applications on the
other.
Ninety nine participants from eleven European countries and from Algeria, Austra-
lia, Brazil, China, India, Japan, Korea, and the United States, were officially registered.
Eighty papers were presented in all, among them five invited papers, namely:
1. D.C. Jiles, .P. Mierczak, Y. Melikhov, Detection of Surface Condition in
Ground Steel Components Using Magnetic Barkhausen Measurements,
2. S. Honda, Pipe Wall Thickness Inspection with Current Driven Thermal
Method,
3. J.S. Knopp, M. Blodgett, J. Calzada, E. Lindgren, C. Buynak, J. Aldrin, Com-
putational Methods in ENDE: Revolutionary Capability for the Sustainment of
Aerospace Systems in the 21st Century,
4. Z. Chen, T. Hwang, L. Wang, S. Tian, N. Yusa, Investigation on the Features
of the Electric Conductivity Around a Stress Corrosion Crack,
5. S.C. Chan, R. Grimberg, J.A. Hejase, Z. Zeng, P. Lekeakatakunju, L. Udpa,
S.S. Udpa, Development of a Nonlinear Eddy Current Technique for Estimat-
ing Case Hardening Depths.
Short versions of all the contributions have been published in the Book of Ab-
stracts, and reviewed and accordingly revised full papers have been accepted and are
now included in this volume: Electromagnetic Non-Destructive Evaluation (XIV) pub-
lished by IOS Press in the series Studies in Applied Electromagnetics and Mechanics.
In closing, we would like to thank the authors, session chairs, and reviewers for
conscientiously executing their duties to maintain the high scientific quality of the pa-
pers published in this volume. We believe that the readership of this book will find the
included papers interesting and inspiring.

T. Chady, S. Gratkowski, T. Takagi, S.S. Udpa


Co-Editors
This page intentionally left blank
vii

ENDE 2010 Szczecin Poland


Conference Organization
The 15th International Workshop on Electromagnetic Nondestructive Evaluation,
Szczecin, Poland, June 1316, 2010.

Organized by
West Pomeranian University of Technology, Szczecin, Poland
in Cooperation with
Japan Society of Maintenology, Tokyo, Japan

Co-Sponsors
Japan Society of Maintenology, Tokyo, Japan
Oddzia Szczeciski Stowarzyszenia Elektrykw Polskich, Szczecin, Poland
Federacja Stowarzysze Naukowo-Technicznych NOT w Szczecinie, Poland
ZAPOL grupa reklamowa, Szczecin, Poland
Technika Obliczeniowa, Krakw, Poland

Standing Committee
F. Kojima Japan
I. Altpeter Germany
J. Bowler USA
N. Bowler USA
T. Chady Poland
Z. Chen China
D. Jiles UK
J. Knopp USA
D. Lesselier France
K. Miya Japan
G.Z. Ni China
J. Pavo Hungary
G. Pichenot France
G. Rubinacci Italy
S.J. Song Korea
T. Takagi Japan
A. Tamburrino Italy
L. Udpa USA
S.S. Udpa USA
viii

List of Referees and Scientific Committee


P. Baniukiewicz L. Janousek K. Stawicki
C. Bardel T. Jayakumar G. Steffes
H. Brauer H. Kikuchi O. Stupakov
J.R. Bowler F. Kojima T. Takagi
K. Capova D. Lesselier A. Tamburrino
T. Chady P. Lekeaka-Takunju T. Uchimoto
S.C. Chan N. Mahapatra L. Udpa
Z. Chen A. Nishimizu S.S. Udpa
W. Chlewicki D. Premel G. Vertesy
Y. Goto J. Pavo T. Yamamoto
S. Gratkowski G. Psuj N. Yusa
R. Grimberg J.M.A. Rebello M. Zikowski
D. He M. Shiwa
H. Huang M. Smetana

Organizing Committee
Honorary Member: Rector ZUT Wodzimierz Kiernoycki
Chairman: Tomasz Chady
Co-Chairman: Stanisaw Gratkowski
Ryszard Sikora
Stefan Domek
Andrzej Brykalski
Members:
Piotr Baniukiewicz
Wojciech Chlewicki
Pawe Frankowski
Justyna Joczyk
Jacek Kowalczyk
Krzysztof Kujawski
Pawe Lesiecki
Przemysaw opato
Lech Napieraa
Marzena Olszewska
Grzegorz Psuj
Tomasz Pietrusewicz
Krzysztof Stawicki
Barbara Szymanik
Marcin Zikowski
ix

ENDE 2010 Szczecin Poland


List of Participants
Yuji Akiyama Iris Altpeter
Kanagawa Institute of Technology, Fraunhofer-Institut IZFP, Saarbrcken
Yokohama Campus E3.1
3-19-25 Shimosueyoshi, Tsurumu-ku D-66123 Saarbrcken
230-0012 Yokohama Germany
Japan iris.altpeter@izfp.fraunhofer.de
akiyama.sml@gmail.com

Piotr Baniukiewicz B.P.C. Rao


Department of Electrical Engineering NDE Division, Metallurgy and Materials
West Pomeranian University of Group,
Technology, Szczecin Indira Ghandi Centre Atomic Research
ul. Sikorskiego 37 603102 Kalpakkam
70-313 Szczecin India
Poland bpcrao@igcar.gov.in
piotr.baniukiewicz@zut.edu.pl

Bilicz Sndor Hartmut Brauer


Budapest University of Technology Ilmenau Univeristy of Technology
and Economics Helmholtzplatz 2, P.O.Box 100565
Egry Jzsef t 18 98684 Ilmenau
1521 Budapest Germany
Hungary hartmut.brauer@tu-ilmenau.de
bilicz@evt.bme.hu

Andrzej Brykalski Stephen Burke


Department of Electrical Engineering DSTO
West Pomeranian University of 506 Lorimer St
Technology, Szczecin VIC 3207 Fishermans Bend
ul. Sikorskiego 37 Australia
70-313 Szczecin steve.burke@dsto.defence.gov.au
Poland
Andrzej.Brykalski@zut.edu.pl

Flavio Calvano Juan Calzada


Universit degli Studi di Napoli Federico II Air Force Research Laboratory
Via Claudio 21 2230 Tenth Street
80125 Naples 45433 WPAFB
Italy USA
flavio.calvano@unina.it juan.calzada@wpafb.af.mil
x

Klra pov Tomasz Chady


DEBE, Faculty of Electrical Engineering, Department of Electrical Engineering
University of ilina West Pomeranian University of
Univerzitn 1 Technology, Szczecin
SK-01026 ilina ul. Sikorskiego 37
Slovak Republik 70-313 Szczecin
capova@fel.uniza.sk Poland
tchady@zut.edu.pl

Zhenmao Chen Wojciech Chlewicki


Key Laboratory for Strength and Department of Electrical Engineering
Vibration, School of Aerospace, West Pomeranian University of
Xian Jiaotong University Technology, Szczecin
28 West Xianning Road ul. Sikorskiego 37
710049 Xi-an 70-313 Szczecin
China Poland
chenzm@mail.xjtu.edu.cn wojciech.chlewicki@zut.edu.pl

Georgios Christofi Thanh Long Cung


Department of Physics, SATIE ENS Cachan
University of Cyprus 61 avenue du prsident Wilson
Building: 02,20537 94230 Cachan
1676 Nicosia France
Cyprus cung@satie.ens-cachan.fr
ph05gc1@ucy.ac.cy

Rmi Douvenot Dagmar Faktorov


L2S-DRE University of ilina
(CNRS-SUPELEC-UNIV.PARIS SUD) Univerzitn 1
Joliot Curie 3 SK-01026 ilina
91192 Gif-sur-Yvette Slovak Republik
France faktor@fel.uniza.sk
remi.douvenot@lss.supelec.fr

Mouloud Feliachi Emna Amira Fnaiech


IREENA CEA List
CRTT, Bd de lUniversit Centre Saclay; Bt 611 PC120
44600 Saint Nazaire 91191 Gif-sur-Yvette
France France
mouloud.feliachi@univ-nantes.fr emna-amira.fnaiech@cea.fr

Bryan Foos Pawe Frankowski


First Principles Inc. Department of Electrical Engineering
800 Harman Ave West Pomeranian University of
45419-3432 Oakwood Technology, Szczecin
USA ul. Sikorskiego 37
foosbc@yahoo.com 70-313 Szczecin
Poland
dmode101@interia.pl
xi

Konstanty Marek Gawrylczyk Stanisaw Gratkowski


Department of Electrical Engineering Department of Electrical Engineering
West Pomeranian University of West Pomeranian University of
Technology, Szczecin Technology, Szczecin
ul. Sikorskiego 37 ul. Sikorskiego 37
70-313 Szczecin 70-313 Szczecin
Poland Poland
kmg@zut.edu.pl sgratkowski@zut.edu.pl

Raimond Grimberg Gyimthy Szabolcs


National Institute of Research and Budapest University of Technology
Development for Technical Physics and Economics
47 D.Mangeron Goldman Gy. tr 3
700050 Iasi 1111 Budapest
Romania Hungary
grimberg@phys-iasi.ro gyimothy@evt.bme.hu

Ammar Hamel He Dongfeng


A/Mira University, Bejaa National Institute for Materials Science
Route de Targa Ouzemour Sengen 1-2-1
06000 Bejaa 305-0047 Tsukuba
Algeria Japan
hamelkane@yahoo.fr he.dongfeng@nims.go.jp

Tommy Henrisson Susanne Hillmann


L2S-DRE Fraunhofer IZFP, Dresden
(CNRS-SUPELEC-UNIV.PARIS SUD) Maria-Reiche-Str. 2
Joliot Curie 3 01109 Dresden
91192 Gif-sur-Yvette Germany
France susanne.hillmann@izpf-d.fraunhofer.de
henriksson@lss.supelec.fr

Satoshi Honda Huang Haiying


Faculty of Science & Technology, University of Texas at Arlington
Keio University 500 W. First Street, WH211
Hiyoshi 3-14-1, Kohoku 76019 Arlington
223-8522 Yokohama USA
Japan huang@uta.edu
honda@appi.keio.ac.jp

Moayyed Hussain Ladislav Janouek


Benet Laboratoris DEBE, Faculty of Electrical
Watervliet Arsel Engineering, University of ilina
12189 Watervliet Univerzitn 1
USA SK-01026 ilina
mhussain@nycap.rr.com Slovak Republik
janousek@fel.uniza.sk
xii

David Jiles Hiroaki Kikuchi


Cardiff University Iwate University
CF24 3AA Cardiff 4-3-5 Ueda
United Kingdom 020-8551 Morioka
jilesd@cardiff.ac.uk Japan
hkiku@iwate-u.ac.jp

Kim Jungmin Kiss Imre


Chosun University Budapest University of Technology
Seosukdong 375 and Economics
501-759 Gwangju Egry Jzsef t 18
Korea 1111 Budapest
ekanus@naver.com Hungary
kiss@evt.bme.hu

Jeremy Knopp Satoru Kobayashi


AFRL/RXLP Iwate University
WPAFB, OH, 45433 4-3-5 Ueda
jeremy.knopp@wpafb.af.mil 020-8551 Morioka
Japan
koba@iwate-u.ac.jp

Fumio Kojima Eugeniusz Kornatowski


Kobe University Department of Electrical Engineering
Rokkodai 1-1, Nada West Pomeranian University of
657-8501 Kobe Technology, Szczecin
Japan ul. Sikorskiego 37
kojima@koala.kobe-u.ac.jp 70-313 Szczecin
Poland
KORN@zut.edu.pl

Jacek Kowalczyk Marc Kreutzbruck


Department of Electrical Engineering BAM Federal Institute for Materials
West Pomeranian University of Research and Testing
Technology, Szczecin Unter den Eichen 87
ul. Sikorskiego 37 12205 Berlin
70-313 Szczecin Germany
Poland marc.kreutzbruck@bam.de
sjk@zut.edu.pl

Marc Lambert Brian Larson


L2S-DRE Iowa State University
(CNRS-SUPELEC-UNIV.PARIS SUD) 175 ASC II
Joliot Curie 3 50011 Ames
91192 Gif-sur-Yvette USA
France blarson@iastate.edu
marc.lambert@lss.supelec.fr
xiii

Lee Jinyi Dominique Lesselier


Chosun University L2S-DRE
Seosukdong 375 (CNRS-SUPELEC-UNIV.PARIS SUD)
501-759 Gwangju Joliot Curie 3
Korea 91192 Gif-sur-Yvette
jinyilee@chosun.ac.kr France
lesselier@lss.supelec.fr
Przemysaw opato Lim Tekoing
Department of Electrical Engineering CEA Saclay
West Pomeranian University of Centre Saclay; Bt 611 PC120
Technology, Szczecin 91191 Gif-sur-Yvette
ul. Sikorskiego 37 France
70-313 Szczecin tekoing.lim@cea.fr
Poland
plopato@zut.edu.pl
Steve Mahaut Kenzo Miya
CEA Saclay Professor-emeritus of the University of
Centre Saclay; Bt 611 PC12D Tokyo
91191 Gif-sur-Yvette President of Japan Society of
France Maintenology (JSM)
steve.mahaut@cea.fr http://www.jsm.or.jp/jsm/en/index.html
Editor in Chief of E-Journal
of Advanced Maintenance (EJAM)
http://www.jsm.or.jp/ejam
miya@jsm.or.jp
Hassane Mohellebi Lech Napieraa
Electrical Engineering Laboratory Department of Electrical Engineering
Tizi-Ouzou University West Pomeranian University of
BP17RP Technology, Szczecin
15000 Tizi-Ouzou ul. Sikorskiego 37
Algeria 70-313 Szczecin
mohellebi@yahoo.fr Poland
lech.napierala@gmail.pl
Yoshihiro Nishimura Youngmin Park
National Institute of Advanced Industrial Chosun University
Science and Technology Seosukdong 375
1-2-1 Namiki 501-759 Gwangju
305-8564 Tsukuba Korea
Japan ekanus@naver.com
nishimura.yoshihiro@aist.go.jp
Jozsef Pavo Cuixiang Pei
Budapest University of Technology and Uesaka-Demachi Laboratary,
Economics The University of Tokyo
Goldman Gy. tr 3 2-11-16, Yayoi, Bunkyoku
1111 Budapest 113-0032 Tokyo
Hungary Japan
pavo@evt.bme.hu peicx@nuclear.jp
xiv

Matthias Pelkner Tomasz Pietrusewicz


BAM Federal Institute for Materials Department of Electrical Engineering
Research and Testing West Pomeranian University of
Unter den Eichen 87 Technology, Szczecin
12205 Berlin ul. Sikorskiego 37
Germany 70-313 Szczecin
matthias.pelkner@bam.de Poland
tp@zut.edu.pl

Denis Prmel Grzegorz Psuj


CEA List Department of Electrical Engineering
Centre Saclay; Bt 611 PC120 West Pomeranian University of
91191 Gif-sur-Yvette Technology, Szczecin
France ul. Sikorskiego 37
denis.premel@cea.fr 70-313 Szczecin
Poland
gpsuj@zut.edu.pl

Joao Rebello Verena Reimund


Federal University of Rio de Janeiro BAM Federal Institute for Materials
Ilha do Fundao CT Room F-210 Research and Testing
21941-972 Rio de Janeiro Unter den Eichen 87
Brasil 12205 Berlin
jmarcos@metalmat.ufrj.br Germany
verena.reimund@bam.de

Maciej Roskosz Guglielmo Rubinacci


Silesian University of Technology Universit degli Studi di Napoli
ul. Akademicka 2A Federico II
44-100 Gliwice Via Claudio 21
Poland 80125 Naples
maciej.roskosz@polsl.pl Italy
rubinacci@unina.it

Yasutomo Sakai Adriana Savin


Tohoku University National Institute of Research and
6-6-01-2, aramaki aza aoba, aoba ward Development for Technical Physics
980-8579 Sendai 47 D. Mangeron
Japan 700050 Iasi
ysaka@karma.qse.tohoku.ac.jp Romania
asavin@phys-iasi.ro

Hans-Peter Schmidt Karl Schmidt


UAS Amberg Weiden (HAW) Evisive, Inc.,
Kaiser Wilhelm Ring 23 8867 Highland Road,
D-92224 Amberg #378, Baton Rouge, Louisiana, 70808
Germany USA
h.schmidt@haw-aw.de karl@evisive.com
xv

Jan Sikora Ryszard Sikora


Lublin University of Technology Department of Electrical Engineering
Electrical Engineering and West Pomeranian University of
Computer Science Faculty Technology, Szczecin
20-618 Lublin ul. Sikorskiego 37
38A Nadbystrzycka Str. 70-313 Szczecin
Poland Poland
sik59@wp.pl rs@zut.edu.pl

Christian Seibold Kisu Shin


UAS Amberg Weiden (HAW) Korea Nationa Defense University
Kaiser Wilhelm Ring 23 205 Susaek-dong
D-92224 Amberg 122-875 Seoul
Germany Korea
c.seibold@haw-aw.de ksshin@kndu.ac.kr

Marek Smaga Milan Smetana


University of Kaiserlautern DEBE, Faculty of Electrical
Gottlieb-Daimler-Strae Engineering,
67663 Kaiserlautern University of ilina
Germany Univerzitn 1
smaga@mv.uni-kl.de SK-01026 ilina
Slovak Republik
smetana@fel.uniza.sk

Thierry Sollier Krzysztof Stawicki


IRSN Department of Electrical Engineering
31, av. de la Division Leclerc BP 17 West Pomeranian University of
92262 Fontenay-aux-Roses Technology, Szczecin
France ul. Sikorskiego 37
thierry.sollier@irsn.fr 70-313 Szczecin
Poland
ks@zut.edu.pl

Tatiana Strapov Barbara Szymanik


DEBE, Faculty of Electrical Engineering, Department of Electrical Engineering
University of ilina West Pomeranian University of
Univerzitn 1 Technology, Szczecin
SK-01026 ilina ul. Sikorskiego 37
Slovak Republik 70-313 Szczecin
strapacova@fel.uniza.sk Poland
szymanik@zut.edu.pl

Toshiyuki Takagi Antonello Tamburrino


Tohoku University University of Cassino
Katahira 2-1-1, Aoba-ku V.G. Di Biasio, 43
980-8577 Sendai 03043 Cassino
Japan Italy
takagi@ifs.tohoku.ac.jp tamburrino@unicas.it
xvi

Theodoros Theodoulidis Tetsuya Uchimoto


University of Western Macedonia Tohoku University
Bakola & Sialvera 2-1-1 Katahira, Aoba-ku, Sendai, Miyagi
50100 Kozani 980-8577 Sendai
Greece Japan
theodoul@uowm.gr uchimoto@ifs.tohoku.ac.jp
Lalita Udpa Satish S. Udpa
Michigan State University Michigan State University
2100C Engineering Building 3410 Engineering Building
48824-1226 East Lansing 48824-1226 East Lansing
USA USA
udpal@egr.msu.edu udpa@egr.msu.edu
Elmar van den Elzen Vertesy Gabor
Lismar Engineering BV Research institute for Technical Physics
Ambachtenstraat 55 and Materials Science
1191JM Ouderkerk a.d. Amstel Konkoly Thege Mikls ut 29-33
Netherlands H-1121 Budapest
c.m.j.vandenelzen@lismar.com Hungary
vertesyg@mfa.kfki.hu
Xie Shejuan Katsuhiko Yamaguchi
Tohoku University Fukushima University
2-1-1 Katahira, Aoba-ku, Sendai, Miyagi 1 Kanayagawa
980-8577 Sendai 960-1296 Fukushima
Japan Japan
xie@wert.ifs.tohoku.ac.jp yama@sss.fukushima-u.ac.jp
Toshihiro Yamamoto Noritaka Yusa
JAPEIC Tohoku University
14-1 Benten-cho, Tsurumi-ku 6-6 Aramaki, Aza Aoba Aoba, Sendai
230-0044 Yokohama 980-8579 Miyagi
Japan Japan
toshihiro@japeic.or.jp noritaka.yusa@qse.tohoku.ac.jp
Marcin Zikowski Chiara Zorni
Department of Electrical Engineering CEA Saclay
West Pomeranian University of Centre Saclay; Bt 611 PC120
Technology, Szczecin 91191 Gif-sur-Yvette
ul. Sikorskiego 37 France
70-313 Szczecin chara.zorni@cea.fr
Poland
marcin.ziolkowski@zut.edu.pl
Marek Zikowski
Ilmenau Univeristy of Technology
Helmholtzplatz 2, P.O.Box 100565
98684 Ilmenau
Germany
marek.ziolkowski@tu-ilmenau.de
xvii

Contents
Preface v
T. Chady, S. Gratkowski, T. Takagi and S.S. Udpa
Conference Organization vii
List of Participants ix

Keynote Lecture

Pipe Wall Thickness Inspection with Current Driven Thermal Method 3


Satoshi Honda

Modeling and Inverse Problems

Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection 13


G. Yang, Z. Zeng, L. Udpa and S.S. Udpa
Non-Iterative MUSIC-Type Algorithm for Eddy-Current Nondestructive
Evaluation of Metal Plates 22
Tommy Henriksson, Marc Lambert and Dominique Lesselier
Identification of Defects in 3D Space Using Computer Radiography System 30
Wojciech Chlewicki, Piotr Baniukiewicz, Tomasz Chady and
Andrzej Brykalski
A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole 36
S.K. Burke
Magnetic Response Field of Spherical Defects Within Conductive Components 44
M. Kreutzbruck, H.-M. Thomas, R. Casperson, V. Reimund and M. Pelkner
Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking from
Eddy-Current Signals 53
Ladislav Janousek, Milan Smetana and Marcel Alman
Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition 61
Konstanty M. Gawrylczyk
Metamodel as Input of an Optimization Algorithm for Solving an Inverse
Eddy-Current Testing Problem 71
Rmi Douvenot, Marc Lambert and Dominique Lesselier
Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations.
Application to Non Destructive Testing on Complex 3D Geometries 79
Tekoing Lim, Gregoire Pichenot and Marc Bonnet
Computation of the Magnetostatic Field in Nonlinear Media via the Integral
Equation Formalism: Application to the Characterization of Magnetic Flux
Leakage NDT System 87
Emna Amira Fnaiech, Denis Prmel, Claude Marchand and
Bernard Bisiaux
Efficient Computation of Eddy Current Crack Signals 96
Theodoros Theodoulidis
xviii

Experimental Validation of a Fast Non-Iterative Imaging Algorithm for Eddy


Current Tomography 103
Salvatore Ventre, Flavio Calvano, Guglielmo Rubinacci and
Antonello Tamburrino
Numerical Evaluation of Microwave Testing for Pipe Thinning 111
Yasutomo Sakai, Noritaka Yusa and Hidetoshi Hasizume
Parallelization of Crack Signal Calculation Using CUDA 117
Imre Kiss, Jzsef Pv and Szabolcs Gyimthy
Electromagnetic-Acoustic Transducers Modeling in Time Domain Using
the Finite Element Method 125
Marcin Ziolkowski and Stanislaw Gratkowski
Eddy Current Testing of Ferromagnetic Materials: Modelling of Flaws in a
Planar Medium 134
Chiara Zorni, Christophe Reboud, Marc Lambert and
Jean-Marc Decitre
Efficient Propagation of Uncertainty in Simulations via the Probabilistic
Collocation Method 141
Jeremy S. Knopp, John C. Aldrin and Mark P. Blodgett
Computation of the Magnetic Field Due to a Defect Embedded in a Planar
Stratified Media: Application to AC Field Measurement Techniques 149
Denis Prmel and Grgoire Pichenot
Automatic Detection and Identification of Image Quality Indicators in
Radiograms 157
Piotr Baniukiewicz and Ryszard Sikora
Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media 164
Cuixiang Pei and Kazuyuki Demachi
BEMLAB Universal, Open Source, Boundary Element Method Library
Applied in Micro-Electro-Mechanical Systems 173
Pawe Wieleba and Jan Sikora
Use of Half-Analytical Method for the Detection of Defects in Diet Pulses 183
Hassane Mohellebi, Ferroudja Bouali and Mouloud Feliachi
Imperialist Competitive Algorithm Applied to Eddy Current Nondestructive
Evaluation 192
Ammar Hamel, Hassane Mohellebi and Mouloud Feliachi

Transducers and Techniques

Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power


Plants Using Magnetic Sensor 203
Shejuan Xie, Toshihiro Yamamoto, Toshiyuki Takagi and
Tetsuya Uchimoto
Developing ECT System with AMR Sensor for Combustion Chamber 211
Dongfeng He, Mitsuhara Shiwa, Jianping Jia, Hisashi Yamawaki,
Ichizo Uetake, Junji Takatsubo, Shinichi Moriya and Koichi Okita
Flux Leakage Measurements for Defect Characterization Using NDT Adapted
GMR Sensors 217
Matthias Pelkner, Andreas Neubauer, Mark Blome, Verena Reimund,
Hans-Martin Thomas and Marc Kreutzbruck
xix

Evaluation of EMAT Signals Using Magnetostrictives for Imaging 225


Yoshihiro Nishimura, Akira Sasamoto and Takayuki Suzuki
Pulsed Eddy Current Testing Application to Defect Evaluation 233
Milan Smetana, Ladislav Janousek, Klara Capova and
Maria Michniakova
An ECT Probe with Widely Spaced Coils for Local Wall Thinning in Nuclear
Power Plants 241
Toshihiro Yamamoto, Tetsuya Uchimoto and Toshiyuki Takagi
PLL Based Eddy Current Measuring System for Inspection of Outer Flaw in
Titanium Alloy Plate 249
Tomasz Chady, Jacek Kowalczyk, Leon Nawos-Wysocki, Grzegorz Psuj
and Ireneusz Spychalski
Giant Magneto-Resistive Sensor Based Magnetic Flux Leakage Technique for
Inspection of Track Ropes 256
W. Sharatchandra Singh, B.P.C. Rao, S. Mahadevan, T. Jayakumar and
Baldev Raj

Application of Electromagnetic Nondestructive Techniques

Pipe Wall Thickness Measurement by Electro-Magnetic Acoustic Transducer


Using Band Exciting Method 267
Daigo Kosaka, Fumio Kojima and Kosuke Umetani
Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring
in Nuclear Power Plants 275
Hiroaki Kikuchi, Isamu Shimizu, Katsuyuki Ara, Yasuhiro Kamada and
Satoru Kobayashi
Stress Corrosion Cracks Evaluation in 316 Austenitic Stainless Steel Plate 282
Tomasz Chady, Jacek Kowalczyk and Grzegorz Psuj
Multi-Frequency Eddy Current NDE of the Distance Between Parts in
Aeronautical Assemblies 288
Thanh Long Cung, Pierre Yves Joubert and Eric Vourch
Artificial Heart Valve Testing Using Electromagnetic Method 296
Tatiana Strapacova, Klara Capova, Ladislav Janousek and Milan Smetana
Conceptual Design of an Industrial System for Automatic Radiogram Analysis 302
Ryszard Sikora, Tomasz Chady, Piotr Baniukiewicz, Marcin Caryk,
Przemysaw opato, Lech Napieraa, Tomasz Pietrusewicz and
Grzegorz Psuj

Material Characterization

Quantification of Sigma Phase Precipitation by Magnetic Non Destructive


Testing 311
Joo M.A. Rebello, Rodrigo Sacramento, Maria C.L. Areiza and
K. Santos de Assis
Nondestructive Characterization of Neutron Induced Embrittlement in Nuclear
Pressure Vessel Steel Microstructure by Using Electromagnetic Testing 322
I. Altpeter, G. Dobmann, G. Hbschen, M. Kopp and R. Tschuncky
xx

In-Line Thin Film Characterization Using Eddy Current Techniques 330


Susanne Hillmann, Marcus Klein and Henning Heuer
Feasibility of Stress State Assessment on the Grounds of Measurements of
the Strength of the Residual Magnetic Field of Ferromagnetics 339
Maciej Roskosz
Magnetic Characterization of Material Degradation Using Dynamical Minor
Loops 347
Satoru Kobayashi, Shinichi Tsukidate, Hiroyuki Okazaki,
Yasuhiro Kamada, Hiroaki Kikuchi and Toshihiro Ohtani
Local Magnetic Properties and Magnetic Particle Distribution Due to Cr
Depletion in Sensitized Ni Based Alloy 355
K. Yamaguchi, K. Suzuki, T. Takase, T. Takara, O. Nittono, T. Uchimoto
and T. Takagi
Metamaterials in Electromagnetic Nondestructive Evaluation 362
Adriana Savin, Rozina Steigmann, Alina Bruma, Nicoleta Iftimie and
Raimond Grimberg
Evaluation of Plastic Deformation in Steels by Magnetic Hysteresis
Measurements 371
Gbor Vrtesy, Shuzo Ueda, Tetsuya Uchimoto, Toshiyuki Takagi,
Ivan Tom and Zofia Vrtesy

High Frequency Techniques and Others

GPR for UXO Recognition 381


Raimond Grimberg, Adriana Savin, Nicoleta Iftimie, Sorin Leitoiu and
Aurel Andreescu
Application of Microwave Interferometry in Complex Engineered Dielectric
Materials 389
Karl Schmidt and Jack Little
Detection of Concealed, Graphite Containing Frescos, Using Microwave
Enhanced Infrared Thermography 402
Pawe Lesiecki and Barbara Szymanik
Solid Materials Complex Permittivity and Inhomogenities Determination in
High Frequency Range 410
Dagmar Faktorov
Terahertz Time Domain Inspection of Composite Coatings for Corrosion
Protection 417
Przemyslaw Lopato, Tomasz Chady and Joao M.A. Rebello
Simulation Assisted Diagnostic of Switching Arcs 425
Hans-Peter Schmidt, Michael Anheuser and Sylvio Kosse
Studies on Analysis Technique of Commutation Phenomena for Cleaners
Universal Motor Using the State Variable Method 432
Yuji Akiyama and Yuta Niwa
Propositions for the Analysis of Commutation Phenomena of Universal
Motors Using the State Function Method 439
Yuji Akiyama and Yuta Niwa
xxi

Detection of CFRPs Degradation Through Electromagnetic Procedures 447


Bogdan Serghiac, Daniel Petrica Salavastru, Paul Doru Barsanescu,
Adriana Savin, Rozina Steigmann and Raimond Grimberg
Microwave Antenna Sensors for Fatigue Crack Monitoring Under Lap-Joints 456
Justin Erdmann and Haiying Huang
Subject Index 467
Author Index 471
This page intentionally left blank
Keynote Lecture
This page intentionally left blank
Electromagnetic Nondestructive Evaluation (XIV) 3
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-3

Pipe Wall Thickness Inspection with


Current Driven Thermal Method
Satoshi HONDA a,1
a
Faculty of Science & Engineering, Keio University, Japan

Abstract. Electric potential dierence(EPD) method is widely used for


diagnosing and monitoring ow conduits. Theoretical analysis and the
exact solution of the electrical potential eld in pipe wall have been
given, and 2D approximation is eective for small pipe wall thickness
there. This study proposes to use temperature distribution driven by
Joule heat with the electric current other than electrical potential dier-
ence. Under thin wall approximation, 2D electric and thermal problems
are analyzed and exact solutions are given.
Keywords. Nondestructive test, pipe wall thinning, forward analysis,
electric potential dierence

Introduction

Electric potential dierence(EPD) method is widely used for diagnosing and mon-
itoring ow conduits [1]. It applies large electric currents between a pair of elec-
trodes on pipe wall surface, and the electrical potentials are measured with mul-
tiple potential electrodes. Since the localized pipe wall thinning or a wall crack
deforms the potential eld, we can monitor and/or diagnose pipe wall condition
under test.
The method is also required to evaluate pipe wall thinning in larger span. We
have studied the prospect of the method to evaluate pipe wall thinning. Based
on the work on non-intrusive resistance thermometer for fast breeder reactor[2,3],
theoretical analysis and the exact solution of the electrical potential eld in pipe
wall have been proposed[4], and 2D approximation is eective for small pipe wall
thickness.
We propose in the paper to use temperature distribution driven by Joule heat
with the electric current other than electrical potential dierence. Under thin
wall approximation, 2D electric and thermal problems are analyzed and exact
solutions are given.

1 Corresponding Author: Satoshi Honda, Faculty of Science & Technology, Keio University,

Hiyoshi 3-14-1, Kohoku, Yokohama 223-8522, Japan; Email: honda@appi.keio.ac.jp


4 S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method

1. Analysis of Wall Potential Problem for EPD Method

1.1. 3D Analysis

Let the electric potential inside the uid be 1 (, , z), and the one inside the
pipe wall, 2 (, , z), where cylindrical coordinates are adopted with pipe axis as
z coordinate. Electrical current is fed between a pair of rectangular electrodes of
bwbw on the outer surface (b, 0, 0). If we let 1 and 2 be electrical conductivities
of the uid and the pipe wall, the following potential problem is formulated.

2 = 0, a<<b (1)
1 = 0, <a (2)

2  i0
= rect(; w) rect(z; bw), (3)
=b 2
  
2  1  2 
1 |rho=a = 2 |=a 2 , 1 = 2 , (4)
=a =a =a

where rect(; ) denotes the rectangular shape of the electrodes



1, |x| w/2
rect(x; w)  (5)
0, |x| > w/2

and i0 , the driving current density. In order to solve the problem, we introduce
the potential distribution 0 for the bulk cylinder of the radius b:

0 = 0, <b (6)

0  i0
 = rect(; w) rect(z; bw), (7)
=b 2

This potential 0 is obtained by using Greens function G(, , z;  ,  , z  ) corre-


sponding to this Neumann problem which satises,

1
G = (  ) (  )(z z  ), (8)


G 
=0 (9)
  =b

The Greens function is given as,

   

m Km (b)
G(, , z;  ,  , z  ) = cos m( 
) Km ( 
)  (b) m
I ( 
)
m=0
2 2 0 Im

Im () cos (z z  ) d
(10)
S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method 5

where Im () and Km () denote m-th order modied Bessel functions of rst and
second kinds, respectively, and m , Neumann factor. This Greens function gives
the potential 0 for the simplied model:

i0
0 = G(, , z; b,  , z  ) rect( ; w) rect(z  ; bw)bd dz 
S 2
 

mw


i0 m Im () bw
= cos m sinc  (b)
cos z sinc d,
2 m=0 2 2 2 0 bIm 2
(11)

Due to the linearity of the original potential problem, we assume the following
solution:

2 = 0 + 2
 
m
= 0 + cos m [dm Km () + cm Im ()] cos z d, (12)
m=0
2 2 0

 
m
1 = cos m fm Im () cos z d (13)
m=0
2 2 0

Since 0 has already accounted for the current feeding electrodes, the boundary
condition for the potential 2 reads as follows:

2 
 = 0. (14)

=b

From three boundary conditions (4) and (14), we obtain three independent equa-
tions for the coecients cm , dm , fm of series expansions, which evaluate the coef-
cients.
When we put a pair of current feeding electrodes on = 0, , the exact form
of the potential distribution is,
 1   
Im () Am ()
2 = cos m  (b)
+ Wm () cos z d (15)
2 0 bIm Bm ()
m:odd
   

 1 Km (b)
Am ()  Im (a) + Im (a) Km ()  Im () (16)
b b Im (b)
 
Im (a)    
Bm ()   + [Im (b)Km (a) Im (a)Km (b)]
Im (a)
 
+ [Im (a)Km (b) Im (b)Km (a)] (17)

1.2. Application to Evaluation of Pipe Wall Thinning

For the convenience, let us normalize the variables here in order to evaluate surface
potential, as  b,  a/b.. In the limit case of the conductivities being
6 S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method

0 1  2 and of very thin pipe wall, the potential distribution on the wall
reduces to,
 mw 
z
i0  2 w
2 (b, , z) = cos msinc Q m () sinc cos d
2 b 2 2 0 2 b
m:odd
(18)
Im () I  () 1
Qm ()  
2m    () K  ()I  ()
(19)
Im () Im () Km ()Im m m

ba
Since  = 1 is small, dropping higher order term of in Taylor series
b
expansion of each function of Qm gives the following simplication.
 
Im () Im () Im () 1
Qm () =  ()
 ()
 ()I  () K  ()I  ()]
Im 2 Im [Km m m m
1 a 1
= = 2 (20)
2 2
( + m ) b a + m2

At the reduction to this nal form, we have adopted formulas for modied Bessel
functions. From Eq.(20), the surface potential distribution is also reduced to,

a i0 2   mw
2 (b, , z) = 2 cos m sinc
b a b 2 2
m:odd
  w z
1
sinc cos d (21)
0 2 + m2 2 b

Let the pipe wall thickness be t  b a. If we make the size w of the current
feeding electrode innitesimally small, sinc(, w) approaches to unity. Therefore,
in this limit case surface potential can be evaluated as follows.
 z
i0 2  1
2 (b, , z) = 2 cos m 2 2
cos d
2 t 0 +m b
m:odd
z
1 i0 cosh cos
= log  zb (22)
4 2 t cosh + cos
b
The denominator, 2 t, implies that we can model thinning pipe wall as the region
with reduced conductivity. Consequently, reduced 2-D problem can be analyzed
in the complex plane,

Z = X + iY, z X, b Y (23)

The following conformal mapping




Z
= exp , Z = b log (24)
b
S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method 7

 

/ 

 


          

z/b

Figure 1. Complex potential evaluation for symmetric excitation Fs (Z).

transforms cylindrical surface of pipe, 0 2, to whole complex plane.


And hence, 2D potential distributions on the circular cylinder can be easily ob-
tained by analytic functions in whole plane as complex potentials. Since current
feeding electrodes are represented by electric charges with corresponding polari-
ties, log( 1) reproduces 2D approximation of 3D exact solution to wall potential
distributions in Eq.(22). Fig.1 shows current eld streamlines on the pipe wall,
where a pair of current electrodes are set at (4b, 0) : (4b, 0), and the correspond-
ing analytic function is


Z
sinh 4b
b
Fs (Z) = log
(25)
Z
sinh + 4b
b

From the gure, we conclude that the current distribution is uniform in the region
far from the current feeding electrodes.

2. Thinning Pipe Evaluation with Joule Heating

Joule heat generated by the driving current varies with wall pipe thickness, or
conductivity. Thus the resulting temperature distribution changes with the wall
thickness. In order to analyze the resulting temperature distribution, Joule heat
distribution is modeled beforehand.
We model the thinning wall region as small circle with the radius of a at the
origin in whole 2D plane far from the current feeding electrodes. Then Joule heat
distribution is obtained from the following 2D potential problem:

1 = 0, a<r (26)
2 = 0, r<a (27)
 
1  2 
1 |r=a = 2 |r=a , 1 = 2 , (28)
r r=a r r=a

where the same symbols 1 , 2 as in the previous section are used for describing
the electric potentials on the innite 2D plate. As in the case of 3D analysis, we
8 S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method

expand the potentials in series of ( ar )n , cos n, sin n, and the boundary condi-
tions determine the coecients of the series expansion to give the following simple
solutions.
1 a2
1 = E0 r cos E0 cos + c0 , ra (29)
1+ r
2
2 = E0 r cos + c0 , r<a (30)
1+
The resulting Joule heat reads,

P1 = j 1 E 1 = 1 (grad 1 ) (grad 1 )

2 4
1 a2 1 a
= 1 E0 1
2
2 cos 2 + , (31)
1 + r2 1+ r4
4
P2 = j 2 E 2 = 2 (grad 2 ) (grad 2 ) = 1 E02 , (32)
(1 + )2
This Joule heat generates the temperature distribution u according to the
following heat transfer equation:
u
= 2 u 2 u + P (33)
t


a2 2 a
4
P = H(r a)P0 2 2 cos 2 + 4 ,
2
(34)
r r

where H() denotes Heaviside function and  (1 )/(1 + ). In order to avoid


divergent solution, the second term in the right hand side of Eq.(33) is introduced
and the origin of the temperature is shifted to simplify the third term by linearity.
The Greens function satisfying the following dierential equation


1 2
2 G(r, t ; r  , t ) = (r r  ) (t t ) (35)
t 2
is given by


1 r2 2rr cos(  ) + r2
G = H(t t ) exp 2
(t t
) . (36)
4(t t ) 42 (t t )
By using this Greens function, we obtain the temperature distribution as the
three-fold integration with respect to t ,  , r , as follows:
 t 2 
u(r, , t) = G(r, t ; r  , t ) P r dr d  dt
0 0 0
 t 

2
P0  r2 2rr cos(  ) + r2
= exp (t t )
2
0 0 a 4(t t ) 42 (t t )


a2 a4
2 2 2 cos 2 + 2 4 r dr d  dt (37)
r r
S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method 9

We start the evaluation of the outermost integration with respect to t . The


factor relating to the variable t is picked up and dened as T (t):

 t

1 
T (t) = exp (t t )
2
dt (38)
0 4(t t )2 t t
1
 (r2 2rr cos(  ) + r2 )
42

Since the stationary state is of main interest, we evaluate the integration with t
to innity and change the variable p = t t :



1 1 1
T () = exp 2 p dp = 2K0 (2 )
420 p p 4 2


1  2   2
= K0 r 2rr cos( ) + r . (39)
2

In order to evaluate the remaining integrals, let us normalize the variables:


 
r = , r = , a = a. Then the integration comes to,

 2  
P0
u (r, )  u(r, , ) = K 0 2 2 cos(  ) + 2
0 a 2 2


a2 2 a
4
 2 2 cos 2 +  4  d d 
2 
(40)

Since the addition theorem holds for modied Bessel function:




 

Kn ()Jn ( ) ein( ) (|| | |)

n=
K0 2 2 cos(  ) + 2 =




Kn ( )Jn () ein( ) (| | ||)

n=
(41)

integration with respect to  remains only for n = 0, 2. And nally, after the
evaluation of the last integral with respect to  , we obtain,



1 1
u (r, ) = P 0 2
aJ0 () K 1 (a) + a 3
K 0 (
)d 
2 a  3
1
P0 2a2 cos 2J2 ()K1 (a), (r < a) (42)
2
10 S. Honda / Pipe Wall Thickness Inspection with Current Driven Thermal Method


1 2 3 1  
u (r, ) = 2 P0 aJ0 () K1 () + a K0 ( )d
 3
1
P0 2a2 cos 2J2 ()K1 ()
2


1 1  
+ 2 P0 K0 () aJ1 (a) J1 () + a
2 4
J0 ( )d
a  3


1 1 1
2 P0 2a cos 2K2 ()
2
J1 (a) J1 () , (r a) (43)
a

An example shape of the Joule heat distribution is shown in Fig.2(a), and the
corresponding temperature distribution, in Fig.2(b).

Figure 2. Stationary temperature and heat distributions are numerically evaluated from Eqs.(42)
& (43).(a)Joule heat distribution. (b)3D plot. First quadrant is shown due to the symmetry.

3. Conclusion

We have presented the analytical solution for temperature distribution driven by


Joule heat over the pipe wall. This exact solution of the forward problem with
the local wall thinning is eective and applicable to wide range diagnosis of pipe
wall anomaly. It, however, should be concidered the eect of mass reduction to
heat conduction, which was neglected in the formulation.

References

[1] A. M. Pritchard and P. Webb Use of the FSM Technique in the Laboratory to Measure
Corrosion Inhibitor Performance in Multiphase Flow Proc. CORROSION/98, paper no.8
[2] H. Yamasaki, S. Honda, M. Ueda, A. Endou, M. Fueki A Novel Non-intrusive Resistance
Thermometer for Fast Breeder Reactor Proc. SICE2003 Annual Conf. Fukui pp 519 /
523(2003)
[3] H. Yamasaki, S. Honda, M. Ueda, A. Endou, M. Fueki A Novel Non-intrusive Resistance
Thermometer for Fast Breeder Reactor Trans. SICE, vol.43, No.9, pp 756 / 761(2007)
[4] S.Honda Thinning Pipe Wall Evaluation by using Electric Potential Dierence Method
Proc. SICE2008 Annual Conf. Tokyo pp 520 / 529(2008)
Modeling and Inverse Problems
This page intentionally left blank
Electromagnetic Nondestructive Evaluation (XIV) 13
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-13

Modeling and Signal Processing Sensor


Tilt in Eddy Current-GMR Inspection
G. YANG1,Z. ZENG 2,L. UDPA1*, and S. S. UDPA1
1
Department of Electrical & Computer Engineering, Michigan State University, East
Lansing, Michigan 48824, U.S.A2 Department of Aeronautics, Xiamen University,
Xiamen, Fujian 361005, China

Abstract. In any NDE application, the availability of a theoretical model is very


useful in understanding the capability of the system since it can enable the
visualization of field/flaw interaction for different flaw and sensor geometries. A
numerical model capable of simulating inspection of multi layered structures using
eddy current excitation and magneto resistive sensing is developed to predict 2D
image data of rivet locations. The model is used to study particularly the effect of
sensor tilt on the probability of detection (POD) of defects at rivet sites. The
modeling of sensor tilt during inspection around rivet sites has to take into account
finite source coil and meshing of the finite source coil at each scan position. This
paper presents a finite element (FE) model formulation using reduced magnetic
vector potential and electric scalar potential in a way that does not require meshing
source coils. In addition to simulating the effect of sensor tilt on the signal, this
paper also describes features in the signal that are invariant to sensor tilt but
sensitive to defect parameters.

Keywords. Eddy current, Giant magnetoresistive sensor, sensor tilt, multi layered
structure

Introduction

The detection of hidden cracks around fastener sites in multi-layered structures poses a
major challenge in the aviation industry [1]. Although, eddy current testing methods
have been used extensively in this application, the skin depth or penetration depth of
the fields has limited their use in detecting 2nd and 3rd layer defects under fastener
heads. With the development of magnetoresistive (MR) sensors, eddy current
inspection system combined with MR sensors is finding increasing use in detecting
directly the magnetic field associated with the induced currents. An eddy current - giant
magnetoresistive (GMR) sensor system has been developed and shown to be effective
in detecting subsurface cracks in a multilayer structure. A schematic of the sensor, with
linear multiline currents and GMR sensor on the line of symmetry, is shown in Figure 1.
In the presence of fasteners and/or cracks, a non-zero normal magnetic field is
produced at the center of the source coil and is picked up by the GMR sensor. The
GMR signal comprises measurement of normal component of magnetic flux density
associated with induced eddy currents in the sample [2].

*
Corresponding Author: udpal@egr.msu.edu
14 G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection

Figure 1. The EC GMR sensor: (a) the excitation coil with a GMR sensor and the tested sample; (b)
schematic of operation of sensor for crack detection.
In a practical inspection scenario, variations in several experimental parameters
such as lift off, top and bottom layer conductivity, edge effect can alter the measured
signal and hence the POD of a subsurface defect. This paper presents the modeling of
defect signals in the presence of sensor tilt during the measurement process. Further
the signals generated are used to derive discriminatory features that enhance the
detection probability of subsurface defects in the presence of sensor tilt. The detection
of detects based on the asymmetry of two lobes of the rivet signal using model-based
methods has been reported in [3-4]. In these studies, the source current is modeled as
an infinite coil, obviating the need for modeling the scanning operation which is a time
intensive process. However modeling of sensor tilt is rendered complex because of the
need to model a finite source coil and the re-meshing at each scan position of the
sensor.
This paper presents the application of a finite element formulation using reduced
magnetic vector potential and electric scalar potential [5-6] for simulating the multi-
line coil motion across the sample without re-meshing the source coil which results in
significant computational savings. The feature based defect detection and classification
using multi-component signal features that are insensitive to sensor-tilt effect validate
the EC - GMR sensor system for the application considered.

1. Numerical Model and Sensor-tilt Parametric Study

1.1 System and Test Geometry

A multi-line coil producing uniform eddy current across the test rivet structure,
sketched schematically in Fig. 1 (a), is employed as the current source. In the absence
G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection 15

of any discontinuity, the symmetry of the coil geometry results in zero normal
component of magnetic field at the line of symmetry on the center of the source coil. In
the presence of a rivet and/or crack, the uniform magnetic field is distorted to yield the
nonzero normal component response. A GMR sensor is placed at the center location to
measure this normal flux density Bz due to the perturbation in induced current. The
analysis of nonzero signal measured by the GMR sensor can be used to detect and
characterize cracks around rivet sites.
The sample consisting of three aluminum layers riveted together is studied in this
paper. The rivets are drilled through three layers and cracks of different sizes are
machined in the 3rd layer plate around rivet sites. A single frequency excitation current
is applied to the multi-line coil and the changes in magnetic field are collected by the
GMR sensor.

1.2 Reduced Vector Potential Formulation based Numerical Model

Finite element formulation of 3-D eddy current testing problems is typically based on
magnetic vector potential A and electric scalar potential V. However this formulation
requires generating a mesh for the excitation foil/coil together with the test sample and
the domain has to be remeshed for each coil position. This can result in excessive
computation time and error. Finite element model based on a formulation using the
reduced magnetic vector potential and the electric scalar potential
( Ar V Ar formulation) has been proposed to simulate the EC-GMR inspection
signals [5-6]. The advantage of this formulation is that the source coil is not meshed
and hence there is no need for re-meshing the source current during scanning. The
mesh and system matrix remain unchanged with different coil motion. The
implementation of the system matrix pre-conditioner is performed only once during the
entire scan.
The formulation model defines the reduced magnetic vector potential by the
decomposition of the magnetic flux density B into two parts:
B = Bs + Br = 0 H s + Br (1)
where Bs and Hs are the flux density and field density due to the source in free space,
and Br is the flux density due to induced current or magnetization. Considering the field
equations together:

Bs = 0 , Hs = Js , Br = B Bs = 0 ,
A s = Bs , A r = Br (2)

The magnetic vector potential A is correspondingly decomposed into As and Ar


where As is the vector potential in air due to excitation current and Ar is the vector
potential due to induced and/or magnetization currents. Generally, Hs and As can be
evaluated by Biot-Savarts law. Hs is calculated by

1 r r
Hs = J s (r) d (3)
4
r r
3
16 G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection

The magnetic vector potential As is:

0 J s (r)
As =
4 r r
d (4)

where is the volume of the current source; 0 is the free-space permeability. r and r
denote the coordinates of observation and source points, respectively.
Correspondingly, the governing equations using this formulation can be written as:

2 A r + j A r + V = (1 vr ) H s j A s
( jA r + V ) = jA s (5)

Note that the weak form of (5) is written as

(N jA + N V )d + N [ j (A

i r i

i r + A s ) V ] n d
= N j A d
i s (6)

where N i ( i= 1,2, N with N the number of nodes) are shape functions, the
solution domain, conductor surface, and n the unit outward normal vector of . By
neglecting the surface integral in (6), we implicitly set the normal component of
induced currents on conductor surface to 0.
The cost of the new formulation is that besides H s , H s in presence of magnetic
materials and A s must be evaluated. However, evaluating these analytical values takes
much less time than that required for performing the incomplete factorization in each
coil position. The model has been validated by comparing model predicted signal and
experimental signal as shown in Figure 2. The Figure 2 (a) presents the model
validation in terms of defective rivet detection. The geometry of this sample is two
layers (6mm and 4mm thickness) with a 2nd layer radial crack of length 5mm at a rivet
site. The 1D signal measured in a scan across the rivet in X direction is plotted and the
same trend and asymmetry is observed in both simulated and experimental signals. The
Figure 2 (b) presents validation of the model for a cylindrical geometry; air core
pancake probe inspection of a tube with an OD axial notch at 100KHz single frequency
excitation.
G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection 17

0.9

0.8

0.7

0.6

(a) 0.5

0.4

0.3

0.2

0.1 Numerical 1D scan


Experimental 1D scan
0
-6 -4 -2 0 2 4 6

(b)

Figure 2. Model validation: comparison of 1D model simulated signal (red) with experimental measurement
(blue) (a) 3rd layer crack in multilayer sample (b) OD axial notch in a tube.

2. Simulation Results and Feature-based Crack Detection

2.1 Sensor-tilt Parametric Study

The FE model using reduced vector potential formulation was used to conduct a
systematic study of the effect of increasing sensor tilt on the defect signal. Three layer
Aluminum sample, each layer of 1 mm thickness, with rivets and 3rd layer sub-surface
defect was modeled. Two radial defects of 3mm length and 0.1mm width and two
different depths of 20%, and 30% wall thickness were modeled with simulation
frequency of 2kHz. The sensor was moved in a linear scan across the rivet and defect
and the GMR signals measuring the normal and tangential components were generated.
In order to separate the effect of sensor tilt and defect, four different cases were studied,
namely, i) no crack ; no sensor-tilt, ii) no crack ; with sensor-tilt, iii) crack; no sensor-
tilt and iv) crack ; with sensor-tilt. The tilt parameter was varied from 00 to 50 for two
different defects.
18 G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection

The Figure 3 shows the geometry of sensor tilt, where the sensor measures both
normal and tangential components in contrast to measuring only normal component
when there is no tilt. Hence, if two components of the magnetic field were measured,
the magnitude of the tangential component will contain information about sensor tilt
and the normal component will contain defect information. Both components of the
magnetic fields were measured and studied. The signals ( Bz and Bx components) due
0
to 20% depth subsurface defect with 1 sensor-tilt are shown in Figure 4. The real part
of Bx component is seen to be sensitive to defect but insensitive to sensor tilt, whereas
the imaginary part of Bz component is seen to be more sensitive to sensor tilt. This
offers the potential of detecting defects in the presence of sensor tilt using two
component signals.

Figure 3. Field measurement with tilted GMR sensor.

2.2 Feature Analysis and Crack Detection

As presented in the Figure 4, the real part of Bx component and imaginary part of
Bz component carry discriminatory information about presence of defect and sensor tilt.
In particular, the peak magnitude and slope of peak to peak variation of these signal
components have been selected as useful features for classification. The peak
magnitude of real part of Bx component is extracted as feature F1 and the slope of peak
to peak variation of imaginary part of Bz component represents feature F2. These
features are illustrated in Figure 5. The two defined features were calculated from the
two defect signals with 10 sensor tilt shown in Figure 4. The features are plotted in the
feature space in Figure 6. It is seen that the 4 cases considered ((i) no crack ; no sensor-
tilt, ii) no crack ; with sensor-tilt, iii) crack; no sensor-tilt and iv) crack ; with sensor-
tilt ) are clearly separated in the feature space. Similarly the signals from the two
defects (depth 20%) and (depth 30%) with 30 sensor tilt were calculated using the
simulation model. The features from these signals are calculated and plotted in Figure 7.
It is seen that as the sensor tilt increases, the 4 cases described in section 2.1 are still
distinguishable but the separation is lesser that what seen in Figure 6.
G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection 19

Figure 4. Field measured by the tilted GMR sensor: B z and B x components.

Figure 5. The extracted features based on Bz and B x components.

Figure 6. Features derived from signals in Figure 4 plotted in the two dimensional feature space.
20 G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection

Figure 7. Features derived from simulated signals with 3 0 sensor tilt , plotted in the two dimensional feature
space.

3. Conclusion

An efficient finite element model using reduced magnetic vector potential formulation
is presented. The model was validated and used for studying the effect of varying
degrees of sensor tilt on EC-GMR signals. Features in GMR output signals were
extracted and their ability to distinguish defect signals from sensor-tilt signals were
demonstrated for two different tilt angle values. More extensive validation of the
approach on experimental signals is under way. A more powerful approach involving
invariance transformation using independent component analysis (ICA) is also being
investigated for transforming the measured signal so that the transformed signal is
invariant to sensor tilt. A sensor system measuring two components of the magnetic
field will be built to for experimental validation of the approach.

References

[1] Aging of U.S. Air Force Aircraft, National Research Council Final Report, Publication NMAB 488 2,
National Academy Press, Washington, D.C., 1997
[2] N. V. Nair, V. R. Melapudi, H. R. Jiminez, X. Liu, Y. Deng, Z. Zeng, L. Udpa, T. J. Moran, and S. S.
Udpa, "A GMR based eddy current system for NDE of aircraft structures," IEEE Transactions on
Magnetics, vol. 42, no. 10, 2006, pp. 3312 3314
[3] J. C. Aldrin, J. S. Knopp, "Method for crack characterization with noise invariance for eddy current
inspection of fastener sites, " Review of Progress in Quantitative Nondestructive Evaluation, vol. 25,
2006, pp. 315 322
[4] X. Liu, Y. Deng, Z. Zeng, L. Udpa, and J. S. Knopp, "Model based inversion technique of GMR signal
using element free Galerkin method," Proceedings of the Annual Review of Progress in Applied
Computational Electromagnetics, 2008, pp. 221 226
G. Yang et al. / Modeling and Signal Processing Sensor Tilt in Eddy Current-GMR Inspection 21

[5] Z. Zeng, X. Liu, Y. Deng, and L. Udpa, "Reduced magnetic vector potential and electric scalar potential
formulation for eddy current modeling," Przeglad Elektrotechniczny, vol. 83, no. 6, 2007, pp. 35 37
[6] Z. Zeng, L. Udpa, S. S. Udpa, and M. S. C. Chan, "Reduced magnetic vector potential formulation in the
finite element analysis of eddy current nondestructive testing," IEEE Transactions on Magnetics, vol.
45, no. 3, 2009, pp. 964 967
22 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-22

Non-Iterative MUSIC-Type Algorithm for


Eddy-Current Nondestructive Evaluation
of Metal Plates
Tommy HENRIKSSON a , Marc LAMBERT a and Dominique LESSELIER a,1
a
Dpartement de recherche en lectromagntisme,
Laboratoire des signaux et systmes (CNRS - SUPELEC - Univ Paris-Sud),
3, rue Joliot-Curie, 91192 Gif-sur-Yvette cedex, France

Abstract. A non iterative MUSIC type imaging algorithm for time harmonic
eddy current testing of a damaged conductive plate is considered. The method is
based (i) on a rst order asymptotic formulation of the secondary magnetic eld
observed outside the plate and induced by volumetric defects within it, which is
valid when those are small enough with respect to the skin depth at the frequency of
operation, and (ii) on the singular value decomposition of the multistatic response
matrix that can be collected using an appropriate source and sensor array. Herein,
the asymptotic eld formulation is validated by comparing the results with those
provided by means of the CIVA platform in a realistic case. Further, the singular
value pattern and MUSIC images are investigated, for a small spherical void placed
at different locations inside a plate. A Gaussian distributed noise is added to the
asymptotic data during the initial investigation, to alleviate inverse crime at least
in part (more complex arrangements of source/sensor arrays and defects will be
considered with external data in a later stage). Yet, these preliminary results as ex
hibited indicate the potential of the method to locate small defects in conductive
materials in a typical eddy current testing conguration
Keywords. Eddy current testing, non iterative inversion, MUSIC type algorithm,
Detection of small defects

Introduction

Eddy-Current (EC) Non-Destructive Testing (NDT) of metal plates is of interest in a


wide range of applications such as quality control in production lines or in-service in-
spection of industrial facilities. Both tube and plate congurations have been studied
earlier, by using different iterative optimization processes [1,2]. Yet, these solutions are
computation-heavy and time-consuming, and in many practical situations the processing
time certainly is a critical criterion, thus a quicker non-iterative method might be more
appropriate, even at the loss of some accuracy in the mapping of the anomalies within
the tested workpieces. So, an alternative method could be to solve the inverse problem
1 Corresponding Author: Dpartement de recherche en lectromagntisme, Laboratoire des signaux et
systmes (CNRS SUPELEC Univ Paris Sud), 3 rue Joliot Curie, 91192 Gif sur Yvette cedex, France; E
mail: lesselier@lss.supelec.fr
T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm 23

by interpolation within an off-line database as is discussed in [3]. Indeed, in general,


only the position, and a gross shape of the defect (with the useful distinction between
volumetric and crack-like defects also) is of real relevance to EC-NDT. In this case a lin-
earized non-iterative method could be suitable, since implemented with a quasi-real-time
functionality, provided that it yields satisfactory results upon position, in particular.
In the present study a non-iterative MUSIC (MUltiple SIgnal Classication)-type
imaging algorithm is investigated for the detection of small void defects inside a con-
ducting plate. The method has been presented theoretically in the general case of di-
electric and/or magnetic volumetric inclusions in free space and full three-dimensional
Maxwells equations in [4], and it has in particular been successfully developed for the
detection of one or several small defects within a dielectric (and possibly lossy) half-
space, in the MHz region [5]. However, to the knowledge of the authors this is the rst
time that the technique is employed in the low-frequency region (kHz) for EC-NDT, here
to image small voids inside a conducting plate.
The paper is organized as follows. The setup with the bobbin array, the conducting
plate and the void defect is described in section 1. In section 2 the formulation is given
for this particular case. In section 3 preliminary results are shown under the form of
validation of the asymptotic formulation of the secondary magnetic eld by comparing it
with the results provided by the simulation tools of the CIVA platform [6], and MUSIC
images for a defect at different positions are given for illustration. Some conclusions and
perspectives follow.

1. Source-sensor arrangement

For illustration of the feasibility of a MUSIC-type algorithm in the EC domain, a compact


bobbin array is considered in free space (region 1, 1 = 0 ), and operated at a frequency
of 100 kHz, as is sketched in Figure 1. It is placed parallel with and above a conducting
plate (region 2, Inconel 600, 2 = 1.02 106 S m1 ), wherein the skin-depth is s =
1.58 mm at 100 kHz. To simplify, the plate is assumed non-magnetic (2 = 0 ) and
linear isotropic with complex permittivity 2 = 0 + i2 as the time-dependency eit is
considered in practice, the dielectric permittivity can be neglected. Also, in the present
study, the plate is considered thick enough to be seen as a half-space.
The bobbin array as chosen consists of 7 7 transmitting/receiving small coils with
inner and outer radii bn = s /8 ( 0.20 mm), cn = s /6.5 ( 0.25 mm), respectively,
and a height H = 0.1 mm. The separation between each coil is s /2 ( 0.8 mm) along
both the x- and y-axis, and the lift-off of the array is 0.1 mm. During inspection each coil
n is transmitting one at the time while all coils including the n one are acting as sensors.
The defect within the metal is taken as a small spherical void (j = 0 ) with a diameter
of aj = s /10 (0.16 mm) and we consider that it is a priori located inside a 21 21 21
cell cubic search domain of s s s extending right under the centre of the bobbin
array.
All dimensions of coils and defect are small with respect to the skin-depth s . Within
this constraint the bobbin array can be suitably approximated as an array of magnetic
dipoles. Then, the impedance variation due to the defect is equivalent to the variation of
(n)
the z-polarized magnetic eld Hs (r) z at the center of the coil located at r (again it
might be the centre of the emitting coil as well as the centre of all other non-emitting
24 T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm

Figure 1. The arrangement for the eddy current MUSIC type algorithm investigation is sketched: a 7 7
(or 3 5 inside the dark rectangle) bobbin array above a non magnetic conducting plate affected by a small
spherical defect (with radius aj ) inside the search domain somewhere under the array.

(n)
coils). Similarly, the primary electric eld E0 (xj ) inside the search region may be com-
puted by using the appropriate dyadic Greens function Gem t (xj , r) of the conguration.
The small size of the defect keeps the asymptotic formulation of the secondary magnetic
eld valid, as is seen next. Otherwise, it could be argued that the overall dimensions of
the array are not that large (side of 3s ); yet number and location of coils as hypothesized
remain realistic in view of the technology whilst most of the anomalous eld due to the
assumed defect under it appears to be collected (in practice also an array will be moved
step by step in either direction to cover a wider zone).

2. Mathematical formulation

The MUSIC algorithm herein is based upon a rst-order asymptotic formulation of the
secondary magnetic eld outside the plate induced by volumetric defects within it, which
is valid when those are small enough with respect to the skin-depth at the frequency of
operation, and on the Singular Value Decomposition (SVD) of the so-called Multistatic
Response (MSR) matrix collected using, e.g., the bobbin array.The method is briey
described below for the particular case of ECT. A detailed description of the MUSIC
algorithm in the small-defect framework for the full Maxwells equations is found for the
general case in [4], with [5] as an illustration in a stratied-space case similar with ours
save made in the propagative (microwave) regime here only diffusion occurs.

2.1. Asymptotic formula secondary magnetic eld

Considering P small spherical voids centered at xj and with radii aj  s , the asymp-
(n)
totic formulation of the secondary magnetic eld Hs (r) may be formed as
T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm 25

P 


(n) j (n)
Hs(n) (r) =H (n)
(r) H0 (r) = i2 Gme
r (r, xj ) Mj , Vj E0 (xj ) ,
j=1
2
(1)
where n is the transmitter index (centered at rn ). Letting I be the identity tensor, the
polarization tensor Mj for such void spherical defects (j = 0 ) is given by


j 3(j 2 ) 4a3j
Mj , Vj = |Vj |I, where |Vj | = . (2)
2 22 + j 3

The Gmer (r, xj ) term is the dyadic Greens function between the defect and the
receiving bobbin array; it satises the following differential equation

Gme 2 me
r (r, xj ) k Gr (r, xj ) = I(r xj ), (3)

considering the defect as an electric impulse current, with usual magnetic-type boundary
conditions, k free-space (air) wavenumber.
Since the radius of each bobbin array element is bn  s the dyadic Greens func-
tion Gem
t (xj , rn ) between the n transmitting coil of the array placed at rn and the
th
(n)
defect can be used to compute the primary magnetic eld E0 (xj ) as

(n)
E0 (xj ) = Gem
t (xj , rn ) z, (4)

where the differential equation satised for a magnetic impulse current source is

Gem 2 em
t (xj , rn ) k Gt (xj , rn ) = I(xj rn ), (5)

with usual electric-type boundary conditions.

2.2. MUSIC imaging

From the secondary magnetic eld the Multistatic Response matrix A is given as


P
A= G me em
r (xj ) Mj G t (xj ), (6)
j=1

where the matrices G me em


r (xj ), a M 3 matrix, and G t (xj ), a 3 N matrix, are con-
structed from the dyadic Greens functions as
me
G r (xj ) = k 2 [Gme me
r (r1 , xj ) z, . . . , Gr (rM , xj ) z] ,
t

(7)

G em 2 em em
t (xj ) = k [Gt (xj , r1 ) z, . . . , Gt (xj , rN ) z] .
t

M and N are the numbers of receiving and transmitting array elements, respectively
(they are not necessarily equal), t denoting the transpose. Since the MSR matrix has the
dimension N M is it possible to apply the Singular Value Decomposition (SVD)
26 T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm

A = UV , (8)

where denotes transpose conjugation. The number of nonzero singular values s in ,


for the j th defect, depends upon the rank of the matrices G me em
r (xj ) and G t (xj ). In our
case, to each defect, will correspond only two singular values since only magnetic
elds are collected, referring to [4] for further mathematical analysis as numerically
illustrated later on in section 3.2. By the rst s columns of U and V (Us , Vs ) the
orthogonal projection onto the left and right signal space can be dened as

Pr = (I Us Us ), (9)

Pt = (I Vs Vs ). (10)

When M = N , the imaging function W(x) is given by

1 1
W(x) = + , (11)
||Pr (G me
r (x) a)||2 ||Pt (G em
t (x) a)||2

where a = (1, 1, 1) [4]. Other values of a could be considered, taking different eld
components into account, how to achieve the best choice of a needing further analysis.

3. Results

The results presented next validate the framework of analysis considering only one spher-
ical void inside the plate in account with the material shown in section 1.

3.1. Numerical Validation - CIVA

The asymptotic formulation of the secondary magnetic eld is validated by compar-


ing the results with those achieved via modeling tools of the CIVA platform [6]. The
impedance variations due to a defect of volume Vj are obtained by a proper calculation
of the volume integral

1 (m) (n)
Znm = E0 (r) Jj (r)dr, (12)
In Im Vj

letting In and Im the currents in the emitting coil n and the receiving coil m (consid-
(m)
ered as a ctitious emitter), E0 (r) the primary eld due to the coil m and Jj (r) the
volumetric eddy-current in Vj due to the source n [2]. Since the radius of the bobbin
element bn  s , the z-polarized magnetic eld from the asymptotic
 formulation should

(n)
be equivalent to the CIVA-computed impedance variation, Hs (rm ) z Znm .
To verify this, a smaller 3 5 bobbin array (depicted in Figure 1) is used with the void
(a = s /10) at the centre of the search region (s /2 depth). The spherical void is mod-
eled as a cube discretized by 5 5 5 cells in the CIVA model, which gives a RMS error
less than 0.2% on the computed results, compared to using a higher number of cells.
T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm 27

The comparison in magnitude between the z-polarized magnetic eld from the
asymptotic formulation and the impedance variation from CIVA is illustrated in Figure 2,
6 different coils in the bobbin array transmitting (specied as bobbin 1 3 and 8 10 in
Figure 1). The asymptotic results () t very well with the CIVA results () in all cases
except when the transmitting coil is right over the void (Figure 2 (d)). In this case indeed,
(8)
the primary eld E0 (x1 ) is very small at the center of the defect. Since the asymptotic
formulations are only using this central value, the result is quite different from the one
yielded by carrying out the volume integral. However, except this peculiar point, it ap-
pears that the asymptotic formulation and the more brute-force approach via a CIVA tool
provide very close results.

3.2. Construction of the images - MUSIC

Once the asymptotic formulation has been validated, it can thus be used to compute the
MSR matrix A for the 7 7 bobbin array. To minimize the inverse crime a Gaussian
distributed complex noise is added onto the data (SNR = 10 dB) as

Anoise = A + , (13)

where is a M N matrix lled with normal distributed pseudorandom complex values,


with an average of zero, and the variance dened as

  |A|2

 10 log10 10
= MN . (14)
2

The same void is to be retrieved at two different depths (s /4 and 3s /4). In Fig-
ure 3 both the singular value distribution and the MUSIC image are depicted, for both
positions. As expected, the void is associated with two nonzero singular values (much
larger than all others, which should ideally be zero, and for all what matters very small
compared to the two main ones), where the amplitudes are decreasing with the depth
of the defect. However in this investigation the SNR of the data has been kept constant
between both positions, which might not be so realistic.
The MUSIC images are iso-surface plots of W(x) with an iso-level of 70%. The
position of the void appears well retrieved for both locations inside the search region,
which is also the case (examples not shown) everywhere inside the search region. How-
ever, with increasing depth the reconstruction along the z-axis gets more blurred (less
powerful signal is acquired, and the aperture angle under which the defect is seen is
getting smaller and smaller).

4. Conclusion and future work

A non-iterative MUSIC-type imaging algorithm has been introduced in the framework


of EC-NDT of a conductive plate. The asymptotic eld formulation on which it is based
has been validated by comparing asymptotic results with those provided by simulation
tools of the CIVA platform where the interaction is simulated from a full eld integral
28 T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm

4 4
x 10 x 10
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
2 4 6 8 10 12 14 2 4 6 8 10 12 14
(a) (b)

4 4
x 10 x 10
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
2 4 6 8 10 12 14 2 4 6 8 10 12 14
(c) (d)

4 4
x 10 x 10
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
2 4 6 8 10 12 14 2 4 6 8 10 12 14
(e) (f)

Figure 2. Comparison in magnitude between the variation of impedance computed via CIVA ( ) and the
z polarized magnetic eld from the asymptotic formulation with a computed primary eld E0 using Gem t (),
along the 15 (3 5) receiving coils when transmitting from (a) coil 1, (b) coil 2, (c) coil 3, (d) coil 8, (e) coil
9, (f) coil 10.

formulation using a method of moments. Preliminary results for a single small spherical
void have been presented, as illustration of the good potential of the method. The position
of the void is indeed successfully retrieved along all three axes, with high accuracy even
in presence of strong noise, whatever its location within the plate.
Future work should involve more complex arrangements of the bobbin array and
several small defects, and also multi-frequency data. The inverse crime should be fur-
ther minimized by constructing the MSR matrix from CIVA data. More complex defects
T. Henriksson et al. / Non-Iterative MUSIC-Type Algorithm 29

Figure 3. Singular value distribution and images for two positions of the void (a s /10), with the 7 7
bobbin array setup. (left) at (0.3, 0, 0.3)s and (right) at (0.3, 0.3, 0.3)s . The exact voids are displayed
as transparent spheres as well.

should be investigated in order to see whether their size can be approached to an ex-
tent via a determination of their polarization tensor, main emphasis being however on
the investigation of crack-like defects, for which the theoretical machinery might still be
incomplete (which does not preclude useful numerical experimentation).

References

[1] D. Dos Reis, M. Lambert, and D. Lesselier. Eddy current evaluation of three dimensional defects in a
metal plate. Inverse Problems, 18:1857 1871, 2002.
[2] J. F. P. J. Abascal, M. Lambert, D. Lesselier, and O. Dorn. 3 D eddy current imaging of metal tubes by
gradient based, controlled evolution of level sets. IEEE Trans. Magn., 44:4721 4729, 2008.
[3] S. Bilicz, M. Lambert, and S. Gyimthy. Kriging based generation of optimal databases as forward and
inverse surrogate models. Inverse Problems, 26(7):074012, July 2010.
[4] H. Ammari, E. Iakovleva, D. Lesselier, and G. Perrusson. Music type electromagnetic imaging of a
collection of small three dimensional inclusions. SIAM J. Sci. Comput., 29:674 709, 2007.
[5] S. Gdoura, D. Lesselier, P. C. Chaumet, and G. Perrusson. Imaging of a small dielectic sphere buried in
a half space. ESAIM:Proceedings, 26:123 134, 2009.
[6] CIVA: State of the art simulation platform for NDE. http://www civa.cea.fr.
30 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-30

Identification of Defects in 3D Space Using


Computer Radiography System
Wojciech CHLEWICKI1, Piotr BANIUKIEWICZ, Tomasz CHADY and Andrzej
BRYKALSKI
Department of Electrical and Computer Engineering, Faculty of Electrical
Engineering, West Pomeranian University of Technology, al. Piastow 17, 70-310
Szczecin, Poland

Abstract. A radiographic system for identification of defects in three-dimensional


space has been built and tested. It is a cost effective alternative to sophisticated
fully three-dimensional tomography systems. We present initial results of using the
extended radiographic system for the images of the phantom made of wood and
pieces of copper.

Keywords. Radiography, computed tomography

Introduction

Fully three-dimensional tomographic X-ray inspection of an object requires projections


to be taken around 360 degrees. This is not possible in many circumstances due to time
and accessibility restrictions and can only be obtained using sophisticated computed
tomography (CT) systems. A much simpler (and cheaper) way for doing radiographic
inspection is using digital radiography systems. However, the output of digital
radiography is a two dimensional image which is a projection of three-dimensional
objects.
In nondestructive testing this could be a strong limitation. Firstly, the information
about the depth is lost. Secondly, the defects at different depth may overlap, i.e. they
may appear as one defect in the radiographic projection. Thirdly, planar defects can be
positioned parallel to irradiating X-rays and hence would not be detected in the image.
Lastly, there are strong requirements on maximal time of acquisition and reconstruction,
which limits the use of CT systems.
In this paper we present initial results of using the extended radiographic system
for identification of defects in three-dimensional space. The system is a cost effective
alternative to sophisticated fully three-dimensional tomography systems. The initial
results of recovering the position in three dimensions are presented. The abilities and
limitations of the system are discussed.

1
Corresponding author. E-mail address: chlewi@zut.edu.pl
W. Chlewicki et al. / Identication of Defects in 3D Space Using Computer Radiography System 31

1. System description

1.1. Image acquisition

Projection data have been acquired using the system presented in Figure 1. The system
consists of a X-ray source CP1201, a digital detector Shad-o-Box1024, a manual
positioning system, a frame grabber and a PC workstation. In the near future the
manual positioning system will be replaced with motorized one.
The acquisition of radiographic projections has been performed along a linear
trajectory. The X-ray source is moved above the object whose position is fixed in space.
Such an image acquisition scheme is an equivalent to limited angle tomography which
was shown by Zhou et al. [1] and confirmed by Gondrom et al. [2]. Their systems,
however, are characterized by a linear translation of the object instead of a motion of
the detector like in our system.

1.2. Image reconstruction

Approximate three-dimensional image reconstruction from projections has been carried


out using iterative algorithms. An adaptation of Algebraic Reconstruction Techniques
(ART) [3] has been implemented in order to perform reconstruction from projections. It
takes into account the linear source detector trajectory. ART is an iterative-type image
reconstruction algorithm that has been proven superior to the widely used Filtered
Backprojection algorithm in case of limited angle tomography [4]. The ART algorithm
consists of a series of successive projections and backprojections (of the correction) of
the reconstructed image. Each step within the algorithm is detailed in the scheme
depicted in Figure 2. As a result, we obtained a volume which possibly fits all acquired
X-ray projections. The correction formula of ART is presented below

N
pi winvn( k )
v(jk +1) = v (jk ) + n =1
N
wij
w
n =1
2
in
(1)

where vj is a reconstructed volume element, pi is a projection element, win is an element


of the system matrix and is a relaxation coefficient.
The algorithm has been implemented in 64-bit Visual C++. The three-dimensional
projection mechanism is using tri-linear interpolation. The size of the reconstructed
volume has been set to 1024x1024x1024. Our experiments showed that the number of
2 iterations is sufficient.
32 W. Chlewicki et al. / Identication of Defects in 3D Space Using Computer Radiography System

Figure 1. The radiographic system.

Figure 2. Iteration scheme of the image reconstruction algorithm.

1.3. Depth identification

The root mean square (RMS) contrast has been used to calculate the contrast of the
Region of Interest (ROI) containing the inclusion for all slices. It is defined as the
standard deviation of the pixel intensities. Maximum of the Crms stands for the inclusion
being in focus:
W. Chlewicki et al. / Identication of Defects in 3D Space Using Computer Radiography System 33

N 1 M 1

( I (i , j ) I (i , j ) )
1 2
C RMS = (2)
MN i=0 j =0

where: Crms is the root mean square contrast function, M and N stand for width and
height of the ROI respectively and I is a map of pixel's intensity of the ROI. The RMS
contrast does not depend upon the spatial frequency content or the spatial distribution
of contrast in the image.

2. Experiments

2.1. Phantom

The phantom is made of wood as a three-slice structure, two inclusions of copper


representing the defects (material, inhomogenity). Copper was put inside holes drilled
in two wooden slices. Dimensions of the phantom are 50 mm x 50mm x 50 mm.

2.2. Data acquisition

Seven projections of size (1024x1024) have been acquired in the manner described in
the above. The voltage of the X-ray tube has been set to 80kV, the current being
equated to 1 mA. The integration time of the detector has been set to 4s. During the
acquisition process the source is moved along the linear trajectory, the range of linear
translation of the detector and the object being (-28cm, 28cm). In such a way, 7
radiographic projections have been taken at different source positions.

2.3. Results of reconstruction

The acquired projection has been input for the reconstruction of the 1024x1024x1024
volume. Some of the significant slices for depth identification are displayed in Figure 3.
Left hand side inclusion is in focus in slice number 271 whereas in the other
crossections it appears blurred. The right hand inclusion is "in focus" in slice number
146. This proves the abilities of the system to localize the depth of the occurrence of
certain objects within the investigated volume. In Figure 4 the process of depth
identification is visually presented.

Figure 3. Crossections of number 101, 121, 146, 201, 271 and 406 (from left to right) within the
reconstructed volume. The maximum of sharpness for each inclusion is clearly visible making it possible to
localize the depth of each inclusion.
34 W. Chlewicki et al. / Identication of Defects in 3D Space Using Computer Radiography System

Figure 4. Results of depth identification.

3. Conclusions

The ability of the system to localize the depth of the occurrence of objects within the
investigated volume has been experimentally confirmed. It can be concluded that the
diagnostic value of digital radiography systems could be significantly improved via
only minor extensions of the systems themselves. Further work is required in order to
assess the impact of changes of the linear translation range on the quality of the pseudo
reconstructed volume [5].

Acknowledgments

The work was conducted in the framework of the research project Identification of
structural heterogeneities in mechanical systems, supported by the Polish Ministry of
Science and Higher Education. Grant no. 1141/B/T02/2009/36 (2009-2012).
W. Chlewicki et al. / Identication of Defects in 3D Space Using Computer Radiography System 35

References

[1] J. Zhou, M. Maisl, H. Reiter, W. Arnold, Computed laminography for material testing. Appl. Phys. Lett.
68, 3500.
[2] Gondrom S., Zhou J., Maisl M., Reiter H., Kroning M., Arnold W.: X-ray computed laminography
approach of computed tomography for applications with limited access. Nuclear Engineering and Design
190, 1999, pp. 141-147.
[3] Kak A., Slaney M.: Principles of Computerized Tomographic Imaging. Philadelphia: SIAM, 2001.
[4] A. H. Andersen, Algebraic Reconstruction in CT from Limited Views, IEEE Trans. Med. Img., vol. 8,
no.1, 1989, pp. 50-55.
[5] B. Li, G. B. Avinash, R. Uppaluri, J. W. Eberhard, B.E.H. Claus, The impact of acquisition angular
range on the z-resolution of *radiographic tomosynthesis, International Congres Series 1268, 2004, pp.
13-18.
36 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-36

A Thin-Skin Model for Eddy-Current NDE


of Cracks in a Borehole
S. K. BURKE 1
Defence Science and Technology Organisation, Melbourne, Australia

Abstract. Eddy-current induction by a coaxial circular air-cored coil in a cracked


borehole is considered in the limit of small electromagnetic skin depth. A closed
form expression is derived for the change in coil impedance Z due to a long axial
crack with constant depth and opening based on an extension of the Lewis
extended surface impedance boundary conditions to a cylindrical geometry. The
theoretical predictions are compared with experimental measurement of Z for a
series of boreholes containing wire-cut slots in 2024 Al alloy test specimens. The
model predictions were in good agreement with experiment for smaller slot depths,
with the accuracy tending to decrease as the slot depth increased. In all cases, the
predictions were superior to those based on a simple Born approximation.

Keywords. Eddy-current NDE, thin-skin theory, bolthole cracks

Introduction

Fastener holes in aircraft act as stress concentrators and are common sites for fatigue
cracking. If left undetected, the growth of such fatigue cracks can lead to catastrophic
structural failure. Considerable effort has been devoted to the development of
nondestructive evaluation (NDE) techniques which can detect cracks in fastener holes
with high reliability. Along with ultrasonic testing, eddy-current NDE is foremost
among these techniques.
The theory underpinning eddy-current NDE for cracks in cylindrical geometries
such as fastener holes continues to be developed, including recent approaches based on
boundary element methods [1]. However, unlike planar geometries, for which there are
highly effective analytical approximations valid in the limit of small electromagnetic
skin depth [2-4], there are currently no similar highly developed thin-skin models for
cylindrical geometries which can be used to complement finite-element and boundary
element methods.
In this paper, a thin-skin model for eddy-current NDE of cracks in a fastener hole
is presented based on an extension of the Lewis extended surface impedance boundary
conditions [4] to a cylindrical geometry. In Section 1, a closed-form expression for the
change in coil impedance Z is presented for the simple case of a coaxial bobbin coil in
a borehole containing a long crack with uniform depth and opening. The coil is
assumed to be sufficiently remote from the ends of the hole that the hole can be treated

1
Corresponding Author. S K Burke, Maritime Platforms Division, DSTO, 506 Lorimer St., Fishermans
Bend, Vic 3207 Australia. Email: steve.burke@defence.gov.au
S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole 37

as a borehole, i.e. infinite in length. The theoretical predictions are compared with
experimental results in Sections 23 and the significance of the work is discussed.

1. Theory

The configuration is shown schematically in Fig. 1 where a cylindrical air-cored coil


carrying an alternating current I exp(i t) is centred coaxially in a borehole containing
a longitudinal crack. The coil, of length 2h, is wound with N turns and has inner and
outer radii r1 and r2 respectively. The borehole has a radius a and the length of the hole
is assumed to be infinite. The crack has a uniform depth b, uniform opening u and is
infinite in length. The line of the crack is parallel to the z-axis at = 0 with the crack
faces in the xz-plane. The parent metal is assumed to have an isotropic electrical
conductivity and linear magnetic permeability =r0 where r is the relative
permeability and 0 is the permeability of free space. As usual in eddy-current NDE,
the frequency is assumed to be sufficiently low that the quasistatic approximation can
be used.
In the following, the problem is formulated in terms of current-free regions using
the magnetic scalar potential where the magnetic field is related to via

H = (1)

and satisfies the three-dimensional Laplace equation

2 = 0. (2)

In the limit of small skin depth ( = [2/()] ) the normal component Hn of the free-
space magnetic field at the surface of an uncracked conductor satisfies the surface
impedance boundary condition [6]

x y

Figure 1. Eddy-current induction by a coaxial source coil in a borehole containing a long rectangular slot.
38 S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole

H n
= ( n) H n (3)
n

where n is the outward normal to the conductor surface and = (1+i)/(r). From
Eqs.(1) and (3) the surface impedance boundary condition for an uncracked borehole
using the coordinate system shown in Fig.1 is

2 1
= ( + ) , r = a. (4)
r 2
r r

Clearly Eq.(4) is not valid in the presence of a crack and must be modified. By
considering the flux leaving the mouth of the crack, Lewis [4] derived an extended
surface impedance boundary condition for a planar conductor containing a crack.
Extending this approach to cylindrical geometries, the extended surface impedance
boundary condition for the cracked borehole in Fig.1 can be written as,

2 ( e ) 1 ( e ) 2
+ ( + ) = H n( i ) ( ), r = a, (5)
r 2 r r a

and links the fields within the crack (the interior fields, denoted by the superscript i)
and the fields within the borehole (the exterior fields denoted by the superscript e). In
deriving Eq.(5), and in the following sections, it is initially assumed that the crack has
zero opening. The effect of crack opening is introduced in Section 1.4.

1.1. Exterior fields fields within the borehole

The exterior scalar potential can be regarded as the superposition

( e ) = ( s ) + ( r ) , a r r2 , (6)

of the known source potential (s) of the isolated coil in free space [7] and the potential
(r) arising from induced currents in the cracked conductor.
A representation for the potential (r) can be constructed by solving Eq.(2) in
cylindrical coordinates using the method of separation of variables. If the origin of the
polar coordinate system is chosen to lie at the center of the coil, then


( r ) (r , , z ) = sin( z ) ( r ) (r , ; ) d , r a, (7)
0


( r ) (r , ; ) = c0 I 0 ( r ) + 2 cn I n ( r ) cos(n ), (8)
n =1

where In are modified Bessel functions and the carat denotes the Fourier sine transform
S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole 39


f (r , ; ) = sin( z ) f (r , , z ) dz . (9)
0

Note also that (r) is an odd function of z and an even function of . The unknown
coefficients cn = cn() are to be determined by application of the extended surface
impedance boundary condition Eq.(5) once a suitable representation for the potential
within the crack has been derived.

1.2. Interior fields fields within the crack

It is convenient to describe the fields within the crack using the Cartesian coordinate
system in Fig.1. In the thin-skin limit (i) satisfies the 2D Laplace equation [4]

2 (i ) 2 (i )
+ = 0, a x a + b. (10)
x 2 z 2

The boundary conditions on (i) are as follows:


Continuity condition on the tangential component of H at the crack mouth,

( i ) ( e )
= , y = 0, x = a. (11)
z z

Thin-skin boundary condition on Hn at the crack tip [23,5],

( i )
= 0, x = a + b. (12)
x

The solution of the boundary value problem defined by Eqs.(10) (12) can be solved
through the use of Fourier transforms [9]. The resulting Fourier sine transform of the
interior potential takes the form

( i ) ( x; ) = ( e ) (a, 0; ) cosh[ (a + b x)] / cosh( b), a x a + b. (13)

1.3. Solution

Closed form solutions for can now be derived using the representations obtained in
sections 1.11.2. The extended surface impedance boundary condition Eq.(5) linking
the exterior and interior potentials can be written in terms of sine Fourier transforms as

( e ) 2 ( e )
a 2 + 2 a 2 ( e ) = 2aH n(i ) ( ), r = a, (14)
r 2

where the term involving the second derivative with respect to r has been eliminated
using Eq.(2). From Eqs. (1), (9) and (13) it follows that
40 S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole

H n( i ) ( x = a; ) = ( e ) (a, 0; ) tanh[ b]. (15)

Substituting Eqs.(6), (8) and (15) into Eq.(14) and simplifying gives the condition


s + c0 0 + cn n cos(n ) =
n =1

(16)
2 a tanh( b) [ ( s ) + c0 I 0 ( a) + 2 cn n cos(n )] ( ),
n =1

where s, n and ( s ) are the known functions,

( s ) (r , 0; ) = nd I sin( h) K 0 ( r ) I (r1 , r2 ), (17)

( s ) (r , 0; )
s = 2 a 2 ( s ) (a, 0; ) a 2 ,
r r =a (18)
n = a I n ( a ) (n + a ) I n ( a).
2 2 2 2

Note that in Eq.(17), nd = N/[2h (r2 r1)] denotes the coil turn density and

r2
I ( r1 , r2 ) = x I1 ( x) dx. (19)
r1

It is useful first to consider the solution to Eq.(16) in the absence of the crack
b = 0. In this case c0 = s/0 and cn = 0 for n 0 as the fields are axisymmetric. Hence,
from Eqs.(6) and (8), the Fourier sine transform of the scalar potential on the surface of
the uncracked borehole in the thin-skin limit is

( u ) (a, 0; ) = ( s ) (a, 0; ) s I 0 ( a ) / 0 . (20)

It can be shown that the thin skin expression Eq.(20) agrees with the exact expression
[7] for an uncracked borehole to order 1/ as required.
In the presence of the crack, Eq.(16) can be solved for the unknown coefficients cn
by multiplying both sides of the equation by cos(m) where m = 0,1,2 , integrating
over the interval and applying the cosine orthogonality conditions. Solving
the resultant simultaneous equations and using Eq.(20) gives the final coefficients

s a tanh( b)
c0 = + ( u ) (a, 0; ),
0 0 [1 + tanh( b) G ( )] (21)
cn = ( s + c0 0 ) / n , n 1,

where the function G() is the infinite sum


S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole 41

a

G=
I 0 ( a ) / 0 + 2
n =1
I n ( a ) / n ,

(22)

1
= ,
n = q I n ( ) / I n ( ) + n 2 + 2

and the arguments q = a and = a are dimensionless. Evaluation of the slowly


convergent infinite sum G can be performed efficiently for small or large arguments
using asymptotic expansions and, for more general arguments, by subtracting out the
known series for digamma functions and summing the more rapidly convergent
difference numerically. In the limit of large a it can be shown that Eq.(22) reduces to
the expression derived by Lewis [4] for the planar case.
Substituting Eq.(21) into Eqs.(6)(8), and making use of Eq.(20), gives the
following closed-form expression for the sine Fourier transform of the magnetic scalar
potential at the mouth of the crack,

tanh( b) G ( )
( e ) (a, 0; ) = ( u ) (a, 0; ) [1 ]. (23)
1 + tanh( b) G ( )

Combining Eq.(1) and Eq.(23) and using the properties of the Fourier transform, the
tangential component of magnetic field at the mouth of the crack can be written

2 cos( z ) tanh( b) G ( ) (u )

H z ( a, 0, z ) = H z( u ) (a, 0, z ) (a, 0; ) d , (24)
0 1 + tanh( b) G ( )

where the first term is the tangential field for an uncracked borehole and the second
term is the field scattered by the crack. Note for simple thin-skin models [5] the
scattering term in Eq.(24) is omitted: an approach termed the Born approximation.

1.4. Coil impedance change due to the crack

An expression for Z in the thin-skin limit due to a crack with finite opening u
can now be obtained using the knowledge of the fields obtained in the proceeding
subsections. The derivation closely parallels that in previous work for the related case
of an infinitely long rectangular slot in a half-space [2,3,5,9] and relies on the
application of surface impedance boundary conditions to the general Auld Z formula
[5] for a defect in a conductor. The higher-order Kahn effect contributions to Z are
also included by assuming that the curved corner fields take the same form [2] as for a
90 edge. Following this approach, it can be shown that

40 g
I 2 Z =
0 1 + tanh( b) G ( )
[ ( u ) (a, 0; )] 2 d , (25)

where the coefficients take the form

g = i u g f + (1 + i) r ( g f 12 u g 0 ) + 12 r 2 g k ,
42 S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole

g f = tanh( b),
g 0 = 2 [1 sech( b)], (26)
g k = 4 (1 + ) / + sech( b).
2 2

2. Experiment

Experiments were performed to test the validity of the theory presented in Section 1.
The experiments were carried out using a series of air-cored coils and a set of 10 mm
diameter bolthole specimens containing simulated cracks. The coil parameters are
given in Table 1.
The test specimens were prepared by drilling and reaming a series of through holes
in thick 2024 Al alloy plate. The electrical resistivity of the plate was 6.00.05 cm
and the plate thickness (hole length) was 24.6 mm. The simulated cracks were
introduced into the reamed holes by electro-discharge machining a slot using a thin
wire electrode. The specimen parameters are given in Table 2.
Coil impedance measurements were made over the frequency range 5kHz400kHz
using a low frequency impedance analyser (HP-4192A). The change in coil impedance
due to the slot Z was determined by measuring the difference in impedance when the
coil is centered in a cracked bolthole and when the coil is centered in an uncracked
reference hole with the same diameter. The impedance data were corrected for the
effects of stray capacitance using a procedure described previously [8].

Table 1. Coil parameters


Inner radius Outer radius Length Turns Inductance Resonant frequency
r1 (mm) r2 (mm) 2h (mm) N L0 (H) fR (kHz)
3.845 4.43 7.805 408 809.5 770
Table 2. Slot parameters and designations. The borehole radius a = 5.00 0.005 in an 2024 Al-alloy block
Slot BB15 Slot BB13 Slot BB2 Slot BB3 Slot BB5 Slot BB4
depth b (mm) 1.33 0.02 3.02 0.02 4.04 0.02 6.01 0.04 8.04 0.08 10.05 0.05
opening u (mm) 0.32 0.30 0.29 0.30 0.31 0.31

3. Results and Discussion

In Fig. 2, the experimental values of Z are compared with the theoretical predictions
obtained using Eq. (25). The results are plotted as a function of skin depth (rather than
frequency) to illustrate the structure of the data clearly. Also shown in Fig. 2 are the
predictions of the Born approximation, for which the scattering term in Eq.(25) is
neglected by setting G()=0.
As shown in Fig.2a, the calculated values of the coil inductance change L are
in good agreement with experiment for small slot depths (b 6 mm ) but less so for
larger depths where the scattering from the crack is greater. Nevertheless, even for
these deeper slots (up to b = 10 mm examined here), good agreement is obtained at the
smallest skin depths within the range. In all cases, the model predictions are superior to
those of the Born approximation for which the scattering is neglected entirely.
S.K. Burke / A Thin-Skin Model for Eddy-Current NDE of Cracks in a Borehole 43

12
45
Experiment
Calculated b=10 mm
40 10
Born Experiment
35 Calculated
Inductance Change (H)

Resistance Change ()
8 Born
30
b (mm) 4 mm
25 6
10
20
8 4
15 6 1.3 mm
4 2
10
3 (a) (b)
5 1.3 0
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Skin Depth (mm) Skin Depth (mm)
Figure 2. Change in (a) coil inductance and (b) coil resistance due to the slot as a function of skin depth.

The corresponding results for the coil resistance change are shown in Fig.2b. The
agreement between the model predictions and experiment is good for all slots at small
skin depth but becomes markedly poorer with increasing skin depth. This is likely due
to the assumption in Eq.(25) that the planar Kahn effect term (which contributes to
R to a greater extent that L) can be adopted for a cylindrical geometry. The model
predictions for R overall are again superior to those based on the Born approximation.
Although the experiments were performed using Al alloys, the model is applicable
to magnetic steels for which r 1. Future experimental work on magnetic materials
would be valuable, particularly as this provides access to significantly smaller skin
depths. The analysis can also be extended to more complex defect types, such as a pair
of longitudinal cracks located on opposite sides of the borehole, and to more complex
coil configurations, e.g. where the coil axis is normal to the borehole surface.
The present model, in which the scattering is incorporated using extended surface
impedance boundary conditions, is superior to the Born approximation and represents
the most complete thin-skin theory for eddy-current NDE of cracks in boreholes to date.
It does not, however, achieve the typical accuracy obtained for planar geometries for
which the scattering from the crack has been treated more completely by Harfield and
Bowler [2-3]. Further theoretical work is clearly required but the extension of the
Harfield-Bowler approach to cylindrical geometries does not appear straightforward.

References

[1] J.R. Bowler and T.P. Theodoulidis, IEEE. Trans. Magn. 45 (2009), 10121015.
[2] N. Harfield and J.R. Bowler, J.Appl.Phys. 82: (1997), 45904603.
[3] J. R. Bowler and N. Harfield, IEEE. Trans. Magn. 34 (1998) 515523.
[4] A.M. Lewis, J.Phys.D: Appl. Phys. 25 (1992), 31926.
[5] B.A. Auld and J.C. Moulder, J.Nondestr.Eval. 18 (1999), 336.
[6] A. Nethe, R. Quast and H. Stahlmann, IEEE.Trans.Magn. 34 (1998), 33313334.
[7] C.V. Dodd, W.E. Deeds and J. W. Luquire, Int.J.Nondestr.Test. 1 (1969), 233293.
[8] D.J. Harrison, L.D. Jones and S.K. Burke, J.Nondestr.Eval. 15 (1996), 2134.
[9] S.K.Burke, J.Appl.Phys., 76 (1994), 30723080.
44 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-44

Magnetic Response Field of Spherical


Defects within Conductive Components
M. KREUTZBRUCK and H.-M. THOMAS, R. CASPERSON, V. REIMUND and
M. PELKNER
BAM Federal Institute for Materials Research and Testing, 12205 Berlin, Germany

Abstract. The determination of magnetic distortion fields caused by inclusions


hidden in a conductive matrix using homogeneous current flow needs to be
addressed in multiple tasks of electromagnetic non destructive testing and
materials science. This includes a series of testing problems such as the detection
of tantalum inclusions hidden in niobium plates, metal inclusion in a nonmetallic
base material or porosity in aluminum laser welds. Unfortunately, straightforward
tools for an estimation of the defect response fields above the sample using
pertinent detection concepts are still missing. In this study the Finite Element
Method (FEM) was used for modeling spherically shaped defects and an analytical
expression developed for the strength of the response field including the
conductivity of the defect and matrix, the sensor to inclusion separation and the
defect size. Finally, the results also can be useful for Eddy Current Testing
problems, by taking the skin effect into consideration.

Keywords. Electromagnetic Testing, Spherical Defects, Finite Element Method

Introduction

When applying an electrical current flow within a given material using induction or
injection techniques, variation of its spatial electrical conductivity distribution will
impact the distribution of the electrical currents and their corresponding magnetic fields
within and outside the tested sample, respectively. Significant conductivity changes can
be observed in case of defect regions within the homogeneous material structure such
as cracks, voids, shrink holes and inclusions. Among the common electromagnetic
Nondestructive Testing methods Magnetic Particle Testing (MPT) has the lions share
in routine testing applications. In comparison to Eddy Current Testing (ECT) MPT can
be regarded as a pure surface inspection technique, whereas ECT provides a view at a
significant depth of several mm below the samples surface, when using adapted testing
equipment.
The first use of MR sensors and other sensitive field sensors in NDT, like the
SQUID [1-3], in the mid-90s opened the door for a new NDT research branch. An
AMR sensor in ECT-probes was already investigated at the Federal Institute for
Materials Research and Testing in 1995 [4]. It was the advent of a new magnetic micro
systems generation after the discovery of the Giant Magneto Resistance effect (GMR)
in 1988 that paved the way for new detection concepts in ECT as well as in flux
leakage testing. This boom in GMR-research was also supported by the Nobel Prize in
Physics in 2007 for the discovery of the GMR effect by Peter Grnberg and Albert
Fert. Currently GMR sensors are being increasingly proposed to detect small defects
M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects 45

with a remarkable signal-to-noise ratio and high spatial resolution [5-8]. The results of
this work meet this new research approach, in which the magnetic flux density B is
detected directly by a field sensor instead of an induction coil which measures the
corresponding field derivative with respect to time B/t. There exist a number of
inclusion-related NDT problems such as tantalum inclusions hidden in niobium plates
used for superconducting resonators or testing superconducting cables with their
complex structure of micro filaments. The latter also involves the estimate of their size
and position within the wires cross-section, which is vital for a reliable quality check.
The problem posed by inclusions also needs to be addressed when testing aluminum
laser welds involving spherical shaped and non-conducting pores representing a
limiting case of the inclusion problem.
A remarkable amount of theoretical work has been done in electromagnetic testing
by the Eddy Current Testing community. Several approaches were devised by Bowler
et al. to calculate cracks, voids and other inhomogeneities and their impact on a sensing
coil [9]. However, these works are mainly concerned with zero conductivity defects
and the changes in the detection coils impedance, and do not address the magnetic flux
density.

1. Field calculations

The most standard FEM codes for ECT problems use the so called [A, V-A]-
method, in which vector potential A and scalar potential V has to be found by solving
the differential equation system [19, 20] for each node in the FEM mesh. From both of
these quantities the electrical current density and the corresponding magnetic flux
density can then be calculated.
For the generation of a homogeneous current flow both an ultra-low frequency
Eddy Current (EC) model and an injection model we used as well. The latter uses two
potential boundary conditions at the left and the right edge of the sample applying a
voltage and a corresponding homogeneous current flow. The EC method and the
injection method yield the same results in most cases. It turned out that the numerical
accuracy of the solver for alternating fields was somewhat higher than that of the dc-
solver. For the calculation of the response field at great distances it was therefore
preferred to calculate using ultra-low frequencies of MHz to simulate the dc-case.

Figure 1. left. The surface mesh of a metal plate showing the surface current density induced by a flat
circular excitation coil. Right: Model of inclusions with different sizes (50 m 800 m in dia.), incorporated
at the center of the niobium plate.
46 M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects

For the calculation we modeled a 4 mm-thick plate with a size of 50 mm  50 mm.


Inclusions with a diameter ranging from 50 m to 800 m (see Fig. 1, right) were
incorporated into the plate at depths between 1 mm and 3 mm. For the EC excitation
we used either a circular coil (see Fig. 1, left) or a symmetric differential (not shown
here) coil to achieve better homogeneity.
The node density in the FEM mesh is distinctly enhanced at the spheres locations.
This serves to adapt the distance between two nodes to the field gradients generated by
the inclusions. The element size of a 50-m-sized inclusion is in the order of 4 m,
whereas the edge elements of the metal plate yield dimensions of nearly 1 mm. For
better visualization in Fig. 1, right, only the surface mesh of the tantalum spheres is
represented, neglecting the spheres connection to the volume mesh of the matrix. The
depth at which the inclusion is located can easily be shifted in the model.
In the following we discuss the electrical and magnetic field variation generated
by an inclusion with a conductivity i within a conductive host medium of conductivity
M, carrying a uniform current density j0. For simplification we used the absolute value
of all spatial current components ( j j  j  j ) for the current density, whereas we
2 2 2
0 x y z

calculated only the vertical component of the magnetic flux density Bz, which usually is
then detected by a field sensor. For inclusions showing a higher conductivity than that
of the matrix, the currents will be focused into the inclusion, resulting in higher current
densities at the front and back of the inclusion where the current flow direction is
parallel to the normal vector of the inclusion surface. This is shown in Fig. 2, left or the
case of a tantalum inclusion hidden in a niobium matrix.
For low conductivity inclusions the situation is the opposite. Here the current is
forced to flow around the inclusion and a minimum current density can be observed at
the front and back (abutting face) with respect to the direction of the homogeneous
current flow. The maximum of the distortion current strength occurs at the inclusion-
host interface along the inclusions cross-section, which is aligned perpendicular to the
current direction and runs through the center of the inclusion. If it is assumed that the
currents run along the y-direction, this interface circle would pass through the
following points in Fig. 2, left: (-0.4, 0, 0), (0, 0, 0.4), (-0.4, 0, 0), (0, 0, -0.4). In the

Figure 2. left: Eddy current distortion of a 2 mm deep circular tantalum inclusion (0.8 mm in dia.) located in
a planar niobium sheet. Eddy current distortion in the x y plane. The current flows along the y axis. The
conductivity of tantalum is somewhat higher than that of niobium, leading to increased distortion currents at
the front side along the y axis at (0, 0.4, 0) and (0, 0.4, 0) and to reduced current densities at (0.4, 0, 0) and
( 0.4, 0, 0). Right: Current density j (jx+jy+jz)0 5 in the vicinity of the inclusion at the x/y plane at z 0 (cut
plane through the inclusions centre). Uniform current flow passes along x axis. Zero conductivity inclusion
(200 m in dia.) hosted in a Ti matrix ( 2.34 MS/m).
M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects 47

Figure 3. Current density j (jx+jy+jz)0 5 in the vicinity of the inclusion (200 m in dia.) along the x axis at
z 0 (line through the inclusions centre). Uniform current flow passes along x axis. Left: upper curve (blue,
dashed) represents Ti inclusion ( 2.34 MS/m) hosted in a Cu matrix ( 59.6 MS/m) and lower curve
(red, solid) represents air inclusion hosted in an Al matrix ( 37.7 MS/m). Right: upper curve (red)
represents Cu inclusion ( 59.6 MS/m) hosted in a Ti matrix ( 2.34 MS/m) and lower curve (blue)
represents air inclusion hosted in a Ti matrix.

case of zero-conductivity inclusions, the enhancement of the current density is about


50% of the matrix current density, regardless of the inclusions size. Figures 2, right
and Fig. 3 show that the presence of an inclusion affects the current distribution only in
the close proximity of the inclusion. If the distance between the sensor and the
inclusions surface is equal to the inclusions diameter (this is equal to three times the
inclusion radius beginning from the inclusion center), the distortion currents are
already attenuated to only 3 % of the maximum perturbation currents at the inclusions
surface. Therefore the uniform current distribution only experiences a significant
variation near the inclusion-host interface. For distances larger than the inclusions
diameter, the deviations due to perturbation currents can be neglected. These results are
in accordance with the work of G. Sepulveda et al. who also showed that the distortion
of the uniform current distribution can be neglected for distances which are greater than
three times the inclusion radius [21].
Figure 3 shows a line scan of the current distribution along the y-axis for diverse
inclusion-matrix material combinations. Note that because of the shape of the inductor,
the generated current flow was not exactly homogeneous, showing a slight linear rise
along the path in Fig. 3, right. Because of the inclusions low conductivity, the current
density outside the inclusion is higher than inside. In the case of a Ti inclusion hosted
in a Cu matrix a small amount of current still flows through the titanium. On the other
hand, if the inclusion conductivity is higher than that of the host, the currents will flow
into the inclusion and Fig. 3, right indicates that this effect is somewhat higher. The Cu
inclusion located within a Ti matrix shows significantly higher current densities
compared to its air counterpart.
48 M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects

2. Response for varied conductivity of inclusion and host

We now take a look at the magnetic fields generated by the distortion currents and
quantitatively determine how different inclusion conductivity and different host
materials affect the current distribution and the corresponding magnetic field response.
As materials for host and inclusion, Cu, Ti, Al and air with the following conductivity
values have been used: titanium ( = 2,34 MS/m), niobium ( = 6,93 MS/m), and
aluminum ( = 37,7 MS/m).

5
I < I > M
2.5
Ta
B z [T]

0
-2.5
-5
Nb-matrix, 6.93 MS/m
-7.5
Ti-matrix, 2.34 MS/m
10
3 4 5 6 7 8 9
10 10 10 10 10 10 10
conductivity inclusion [S/m]

Figure 4. left: Strength of the Magnetic field variation above the sample as a function of the inclusions
conductivity. Two different host conductivities were used (black: Ti, 2.34 MS/m, grey: Nb, 6.93 MS/m).
Inclusion diameter: 200m, Sensor to inclusion separation: 5 mm. Each inclusion is located 2 mm below the
surface. Solid lines represent the results of the fit function using Eq. 3. Right: Example of the field
distribution for low and high conductivity inclusions.
In Figure 4 the maximum field magnitude (Bz, peak-to-peak amplitude) is shown
for different compositions as a function of inclusion conductivity. In the case of
inclusion conductivity smaller than that of the tested material, the induced currents
which otherwise would pass undisturbed through the volume of the inclusion are
perturbed and have to flow around the inclusion. The corresponding vertical
component of the magnetic field is displayed in Fig. 4, top right (the effect of low-
inclusion conductivity is shown to the left from the zero point). The situation is
opposite when the inclusion conductivity is greater than that of the tested material.
Then the currents flow from the host into the inclusion, leading to a change in the
direction of circulation. The circular perturbation currents that once flowed clockwise
now flow anti-clockwise (Fig. 4 right, on the right from the zero point and Fig. 4, right
bottom). Consequently, the sign of the magnetic field is changed from positive field
values to negative ones. When the conductivity of the inclusion and matrix are the
same, this will obviously result in a vanishing response field and no perturbation
currents will occur (zero point, Fig. 4, left).
One can observe that for very low inclusion conductivity the defect response does
not change significantly if the conductivity of the inclusion is further reduced. The
same saturation effect occurs for very high inclusion conductivity. By comparing the
field values for very low and very high defect conductivities one can observe an
anomaly, already seen when dealing with the current distribution. In the limit of high
M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects 49

inclusion conductivity one finds a field response exactly twice as high as that of an
inclusion with negligible conductivity. We define  as the ratio between the inclusions
conductivity I and matrix conductivity M as follows:
I
(1)
M
The numerical data for any M in the limit at infinity and zero conductivity show
the following relation:

lim Bz


 2 (2)
lim Bz

0
The response field Bz of an inclusion with excellent electrical transport properties,
say silver, in a low conductivity metal matrix will be higher than that of an air inclusion
(pore) located in the same matrix carrying the same homogenous current density. The
reason for these phenomena can be explained by the limited inclusion volume. The
defect volume is limited with respect to the much larger volume of the host. Assuming
that all currents are pushed out of the inclusion, it is not possible to increase the
response field by lowering the inclusion conductivity. It is therefore reasonable to
assume that for the virtually infinite volume of the matrix significantly more current
can be passed through the inclusion if its conductivity is high enough. Of course, there
also exists an optimum path for the currents flowing from the host into the inclusion
and back again. The simulation reveals that this geometrical condition causes a
response field which is a factor of 2 larger than that of the zero conductivity case. A fit
of the numerical data shows that
  M  1 
Bz ~ I   (3)
 I  2 M   2 
When using the FEM results (dots in Fig. 4, left) we found the error of this
analytical expression (solid line in Fig. 4, left) to be below 2%. Note that these results
are independent of the inclusions size or the distance of the sensor above the samples
surface.

3. Sensor-to-Inclusion Separation and Inclusion Size

In Figure 5, left, the magnetic field is shown as a function of the sensor-to-inclusion


separation z, in which z describes the distance between the centre of the inclusion and
the field detecting area of the sensor. In addition, for better visualization of the fall-off
behavior, Fig. 5, right shows the ratio of B(z/2)/B(z), which is denoted by  in the
following sections. The simulation was carried out for inclusions with sizes ranging
between r = 100 m and r = 800 m. The sensor-to-inclusion separation is divided into
two regimes both of which show a different fall-off characteristic of 1/z. Starting in
the very near field regime at the inclusions centre we see that  is distinctly below 2.
By approaching the inclusions surface  converges to 2.
By exceeding the interface of the inclusion and the matrix, the fall-off
characteristic shows the quadratic behavior in which  = 2 because of the quantity
B(z/2),  = 4  = 2) that is not achieved at the surface at once. It is only obtained when
50 M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects

distances are at least twice the inclusions radius. Closer examination of the data shows
that the interface of the two regimes  < 2 and  = 2 is located exactly at the inclusion-
host interface. The simulations also confirm that Bz as function of the inclusion radius r
scales exactly with a r law. This also can be seen in Fig. 5, left, where the magnetic
field response is distinctly increased in the presence of larger inclusions. Furthermore,
the graph shows that a larger inclusion diameter will shift the regime of  < 2 towards
higher values. Independent of the inclusion size, the  = 2 regime invariably is entered
at the interface of inclusion and host.

Figure 5. Fall off characteristics calculated for different inclusion diameters. Magnetic field variation
as a function of the sensor to inclusion separation (left) and ratio of Bz(z/2)/ Bz(z) as function of the
dipole to sensor separation (right). Inclusion with zero conductivity (air) located in a titanium matrix
( 2.34 MS/m).

4. Analytical Expression for the Response Field Bz

Size dependence and spacing between sensor and inclusion are in good accordance
with the results of G. Sepulveda et al. [21]. They used spheroidal shaped cracks with
variable conductivity to analytically calculate the magnetic field that would be
measured by a magneto sensor. Due to the complexity of the problem, the solution is
given only for the dc-problem and is a first-order approximation. The general solution
is proportional to the Legendre functions of the first kind and consists of odd Legendre
polynomials. One can use several boundary conditions to determine the coefficients,
but due to the limited number of boundary conditions, only first terms of the
polynomial expansion can be used. The spheroidal approach is hereby a kind of
approximation, which makes an analytical calculation of the current distribution
possible. However, when turning to the magnetic field, unfortunately no unique
mathematical expression can be derived, leading also to finite element approaches.
Thus no expression of the field strength for different inclusion conductivity was
presented.
Our FEM results are in good agreement with the current and field distribution of
G. Sepulveda, who also found that the magnitude of the far field distances much
further than the prolate spheroids diameter falls off as 1/R2 at any given direction.
Implementing the findings of sections 3 and 4 we now present a simple analytical
expression for varying conductivity in the inclusion and the matrix, which fits the
numerical data extremely well for all types of isotropic materials:
M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects 51

0 r3  1 
Bz pp
  j0    (4)
1.354 z2   2 

To provide a broad data set, we calculated the peak-to-peak magnetic flux density
Bz pp for a number of different inclusions. The sensor-to-inclusion separation was varied
in the regime ranging from z = 2 mm to z = 6 mm. The radius of the inclusions was
varied between 50 m and 400 m. Finally for the inclusion conductivity we used
values between 0 MS/m and 59.9 MS/m (Cu) and for the conducting host we used
values ranging from the conductivity of Ti to that of Cu. As a result, Bz pp varies
between 30 pT and 7 T, covering almost 6 orders of magnitude. Despite this high
dynamic field range, the constant k = 0/1.354  9.28 10-7 Tm/A only varies in a range
of a few percent (see Fig. 6).

FIGURE 6. Variation of k for different inclusion types. Left axis: absolute value of k. Right axis: Relative
error assuming k to be 9.28 10 7. Ti:  2.34 MS/m, Al:  37.7 MS/m, Cu:  59.9 MS/m. Sample 1 4 :
i 0 , M Ti , zi 3,5,4,5 mm, r in m; sample 6 9: i 0, M Al , zi 3,4,3,5 mm, r in m; sample 11
14: i 0, M Ti , zi 2,3,4,5 mm, r 50 m; sample 16 18: i 0, M Ti , zi 2,3,4 mm, r 100 m;
sample 20 22: i 0, M Ti , zi 3,4,6 mm, r 200 m; sample 24 26: i 0, M Ti , zi 2,4,5 mm, r
400 m; sample 28 32: i in MS/m, M Ti , zi 5 [mm], r 100 m; sample 33 36: i, M , zi 4,5,4,5 mm,
r 100 m.

5. Conclusion

In routine non-destructive testing applications the estimation of the detectability of


different material defects is a common task in choosing the proper testing method. The
signal strength generated by inhomogeneities within the material being tested should be
beyond the noise level either caused by the electronic noise of the testing device or the
intrinsic noise sources of the sample itself, such as local variation of the conductivity or
permeability due to the materials microstructure. The analytical expression derived
from the FEM-results will facilitate the estimation of the signal strength caused by
spherical inclusions and requires the current density, sensor-to-inclusion separation,
inclusion size the conductivity of matrix and inclusion. Finally, the question should be
52 M. Kreutzbruck et al. / Magnetic Response Field of Spherical Defects

addressed as to how these results can be transferred to alternative defect types like
porosities and cracks. We found in further investigations that the qualitative behaviour
of the magnetic field does not change significantly for more elongated flaws. Even for
pure rectangular cracks with high aspect ratio the magnetic field above the samples
surface can be interpreted as being generated by a dipole-like source with its 1/z
characteristics - a result also found by Sepulveda. However, for the absolute signal
strength we found a stronger influence of the cross-section and the current direction
compared to the influence of the crack width. This relation still has to be further
investigated. Meanwhile there exist a series of NDT activities, with which the proposed
analytical expression can be confirmed by experimental data [2,8]. A systematic
validation of these theoretical results will be a future task.

References

[1] M. v. Kreutzbruck, J. Trll, M. Mck, C. Heiden and Y. Zhang Trans. Appl. Supercond.7, 3279, IEEE,
1997.
[2] M. Mueck, C. Welzel, F. Gruhl, M. v Kreutzbruck, A. Farr, F. Schoelz, Physica, C 368 96 99 (2002).
[3] H. J. Krause and M. v. Kreutzbruck, Physica C 368 70 79 (2002).
[4] M. Stoppel, D. Filbert, H. M. Thomas, Research report at Technical University Berlin and Federal
Institute of Materials Research and Testing, May 1995.
[5] C. H. Smith, R. W. Schneider, T. Dogaru, S. T. Smith, Rev. Prog. in QNDE, 22, 419 426, (2003).
[6] B. Wincheski, M. Namkung, Rev. Prog. Quant. Nondestr. Eval. 509, 465 (2000).
[7] K. Allweins, M. v. Kreutzbruck, G. Lembke, J. Appl. Phys. 97, 10Q102 (2005).
[8] M. Kreutzbruck, H. Bernau, K. Allweins, TM-TECHNISCHES MESSEN , 75 (9) p. 477 484 (2008).
[9] J. R. Bowler, J. Appl. Phys. 75 (12) 8128 (1994).
[10] J. R. Bowler, S.A. Jenkins, L.D.Sabbagh, and H.A. Sabbath, J. Appl. Phys. 70 (3) 1107 (1991).
[11] N. Harfield, Y.Yoshida and J. R. Bowler, J. Appl. Phys. 80 (7) 4090 (1996).
[12] A.P.Raiche , J.H. Coggon, Geophys J. R. Astron. Soc. 42, 1035 (1975).
[13] J. R. Bowler, J. Appl. Phys. 86 (11) 6494 (1999).
[14] S. M. Nair and J.H. Rose, J. Appl. Phys. 70 (4) 1924 (1991).
[15] C.T. Tai, Dyadic Geenss Functions in Electromagnetic Theory (Intex, Scranton, 1971).
[16] P.E. Wannamaker, G.W. Hohmann, and W.A. SanFilipo, Geophysics 49, 60 (1984).
[17] B. J. Roth, N. G. Sepulveda, and J.P Wikswo, Jr., J. Appl. Phys. 65 (1) 361 (1989).
[18] R.E. Beissner, J. Appl. Phys. 60 (1) 352 (1986).
[19] A. Kost, Numerische Methoden in der Berechnung elektromag. Felder, Springer Verlag (Berlin) (1994).
[20] S.C. Brenner, L.R. Scott, The Mathematical Theory of Finite Element Methods, Springer Verlag, 2002.
[21] N. G. Sepulveda, J. P. Wikswo, Jr., J. Appl. Phys. 79 (4) 2122 (1996).
Electromagnetic Nondestructive Evaluation (XIV) 53
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-53

Decreasing Uncertainty in Size Estimation


of Stress Corrosion Cracking from Eddy-
Current Signals
Ladislav JANOUSEK 1, Milan SMETANA, Marcel ALMAN
Department of Electromagnetic and Biomedical Engineering, Faculty of Electrical
Engineering, University of Zilina, Univerzitna 1, 010 26 Zilina, Slovak Republic

Abstract. The paper deals with uncertainty in depth estimation of partially


conductive cracks from eddy current testing signals. Numerical simulations are
carried out to investigate influences of crack parameters on the response signals.
Three parameters of the crack, i.e. its depth, width and conductivity are changed to
simulate variety of possible scenarios. Standard pancake probe is used for the
inspection at first. The presented numerical results confirm that the inverse
problem is highly ill posed when partially conductive cracks, as stress corrosion
cracking, are concerned. Accordingly, a new approach in the response signal
sensing is proposed. The idea is to pick up more information provided by the eddy
currents. Thus, three coils positioned perpendicularly to each other are employed
to sense all the three spatial components of the perturbation electromagnetic field.
The gained results reveal that it is possible to considerably decrease the
uncertainty in depth estimation of partially conductive cracks by employing the
proposed idea.

Keywords. non destructive evaluation, eddy currents, partially conductive cracks,


depth estimation, uncertainty

Introduction

System health monitoring (SHM) and condition based maintenance (CBM) are of high
interest nowadays in order to sustain safety, reliability and quality of various processes.
Non-destructive evaluation plays accordingly the key role in SHM and CBM where not
only reliable detection but also precise estimation of dimensions of a detected anomaly
is required. Enhancing non-destructive evaluation methods is therefore very important
for accomplishing their challenging missions.
Different physical principles are utilised for the non-destructive inspection and
evaluation of materials. Eddy current testing (ECT) is one of the widely utilized
electromagnetic methods. It originates from the electromagnetic induction phenomena
and its principle underlies in the interaction of induced eddy currents with structure of
an examined body. Perturbations in eddy currents distribution due to a presence of an
anomaly are detected. Indisputable advantages of the method such as high inspection
speed, high sensitivity for surface breaking cracks, versatility account for gradually

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.uniza.sk
54 L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking

increasing employment of the method in practical applications, especially in nuclear,


petrochemical and aviation industries. However, interpretation of the ECT signals is
rather difficult as the inverse problem is ill-posed [1].
The progress in powerful computers has allowed development of automated
procedures to evaluate dimensions of a detected anomaly in ECT. In general, two
approaches are utilized for the purpose, deterministic and stochastic ones [2-6]. Usually,
one dimensional signal gained by scanning just above a crack along its length is taken
as an input to the evaluation procedure. Mostly, three variables of the crack are
estimated, its depth, length and a position of its centre, while a profile, a width and the
electromagnetic properties of the defect have to be adjusted in advance. Satisfactory
results are reported by several groups for evaluation of artificial slits [3], even of
several close parallel notches [6]. However, evaluation of real cracks, especially stress
corrosion cracking (SCC) from ECT response signals remains still a challenge [3], [7].
SCCs are quite different comparing to artificial slits or even to other types of real
defects. Cross sections of SCC frequently show branched structure and group of cracks
usually occur in what is known as a colony. The local opening of SCC is usually very
small, e.g. tens of micrometers; however a damaged region itself is much broader. SCC
contains many unbroken ligaments both in depth and opening directions what makes
SCC partially conductive and in this case its width has to be also considered [8]. It
means that additional variables should be taken into account for evaluation of a
detected SCC what considerably increases ill-posedness of the inverse problem [1].
Thus, many unsatisfactory results are reported when the automated procedures
originally developed for non-conductive cracks are employed in the evaluation of SCC
[3], [7]. It is stated that one of the possible reasons is lack of sufficient information [3].
The authors propose a new idea in sensing eddy currents response signals in order
to decrease uncertainty in SCC evaluation. The vector lines of the eddy-current density
are closed themselves, because the divergence of the eddy-current density vector equals
to zero in an inspected material. Thus, any electromagnetic anomaly at present
influences the path of eddy-currents in quite complicated manner and creates
perturbations in the resulting electromagnetic field. Even, many ECT probes have been
developed over past decades reflecting special demands of particular applications, they
are usually designed in such a way that only one component of the perturbation field is
sensed. However, curved paths of eddy-currents provide more information. The
authors proposal is to sense all the three spatial components of the perturbation field to
gain three separate signals of a same crack. Two-dimensional scan over a cracked
region is required in this case. Numerical investigations are carried out to prove
effectiveness of the proposal. Varieties of non-conductive as well as partially
conductive cracks are modelled to study their influence on the perturbation field. The
paper presents the gained results and their evaluation.

1. Numerical Model

A plate specimen having the electromagnetic parameters of a stainless steel SUS316L


is inspected in this study. The specimen has a thickness of t = 10 mm, a conductivity of
= 1.4 MS/m and a relative permeability of r = 1. A surface breaking crack appears in
the middle of the plate. It is modelled as the cuboid having different electromagnetic
properties from the base material. Configuration of the plate with the crack is shown in
Figure 1.
L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking 55

y SUS316L
wc
x

lc
c
dc

= 1.4 MS/m
r= 1
10

Figure 1. Configuration of plate specimen with crack

The numerical calculations are performed using the edge-element code. Three
parameters of the crack are varied in other to simulate various structures of real cracks.
The crack depth dc is changed from 1 mm up to 10 mm with a step of 1 mm, its width
wc is adjusted to five values ranging from 0.2 mm until 1.0 mm with a step of 0.2 mm
and the conductivity of cracked region c is set to 0, 1, 2, 5, 10% of the base material
conductivity. Length of the crack is kept at a constant value of lc = 10 mm for all the
cases. Frequency of the harmonic driving signal is adjusted to f = 10 kHz. An ECT
probe scans 1 mm above the plate surface. Influences of the crack parameters on the
ECT response signals are studied. The results are presented and discussed in the
following sections.

2. Standard Approach

The standard self-inductance absolute pancake probe is employed for the inspection of
the specimen with crack at first. The circular coil is positioned normally regarding the
surface of plate. Its axis is parallel to the z-axis of the coordinate system shown in
Figure 1. Inner diameter of the coil is 4 mm, the outer one has a value of 8 mm and a
height of the winding is 2 mm.
05 04
1 mm 1 mm
3 mm 3 mm
5 mm 03 5 mm
04
02
absolute value [mV]

imaginary [mV]

01
03

02
01

02
01
03

0 04
20 15 10 5 0 5 10 15 20 04 03 02 01 0 01 02 03 04
probe position [mm] real [mV]
a) absolute value vs. probe position b) signal in complex plane
Figure 2. Crack response signal influence of crack depth
56 L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking

The inspection is realised in a standard way; one dimensional scanning with the
probe over the crack centre along its length is performed. The response signal is
calculated for the various parameters of the crack. Figure 2 shows the response signal
for the non-conductive crack with depths of dc = 1, 3, 5 mm and a width of
wc = 0.2 mm. Similar results are displayed in Figure 3. The crack has a depth of
dc = 3 mm, a width of wc = 0.2 mm and conductivities of c = 0, 5 and 10% of the base
material conductivity. Influence of the crack width on the response signal for partially
conductive crack with a conductivity of c = 5% of the base material conductivity and a
depth of dc = 3 mm is shown in Figure 4. These results are gained for the crack widths
of wc = 0.2, 0.6, 1.0 mm.
04 04
0% 0%
5% 5%
10% 03 10%

03 02
absolute value [mV]

imaginary [mV]
01

02 0

01

01 02

03

0 04
20 15 10 5 0 5 10 15 20 04 03 02 01 0 01 02 03 04
probe position [mm] real [mV]
a) absolute value vs. probe position b) signal in complex plane
Figure 3. Crack response signal influence of crack conductivity

04 04
0 2 mm 0 2 mm
0 6 mm 0 6 mm
1 0 mm 03 1 0 mm

03 02
absolute value [mV]

imaginary [mV]

01

02 0

01

01 02

03

0 04
20 15 10 5 0 5 10 15 20 04 03 02 01 0 01 02 03 04
probe position [mm] real [mV]
a) absolute value vs. probe position b) signal in complex plane
Figure 4. Crack response signal influence of crack width

It can be seen that the response signal is the complex variable of all the considered
crack parameters. Figure 5 demonstrates quite high uncertainty in depth evaluation of
partially conductive cracks. In this case only the maximum points of the response
signals are plotted for the crack with three depths of dc = 1, 3, 5 mm and all the
considered widths and conductivities. As it can be observed the crack signals of the
crack with the three different depths are highly overlapped and thus under- or over-
estimation of the crack depth can occur with high probability. Especially, the
underestimation can bring serious consequences in many applications.
L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking 57

0
1 mm
3 mm
5 mm
01

imaginary [mV]
02

03

04

05
0 01 02 03 04 05
real [mV]

Figure 5. Overlap of response signals due to partial conductivity of cracked region

The reason of the high uncertainty in depth estimation of partially conductive


cracks using standard ECT probes and procedures is lack of sufficient information. The
next section introduces a new idea for the response signal sensing to increase
information content of the signal.

3. Proposal of New Approach

The vector lines of the eddy-current density must be closed themselves in an inspected
material. Thus, any electromagnetic anomaly at present influences the path of eddy-
currents in quite complicated manner and creates perturbations in the resulting
electromagnetic field. The previous section clearly showed that the degree of
uncertainty in depth evaluation of a partially conductive crack is quite high when only
one component of the perturbation field is sensed; however, the eddy currents provide
more information.
The authors propose sensing all the three spatial components of the perturbation
electromagnetic field to increase the information rate of ECT response signals.
Numerical results are provided to support the proposal. The same coil as in the
previous section drives the eddy currents in the specimen. Three circular coils of same
dimensions, i.e. the outer diameter equals to 4 mm, the inner one is set to 2 mm and the
height of winding is 0.5 mm, sense the response signals independently. The coils are
positioned perpendicularly to each other to sense all the three spatial components of the
perturbation field. All the coils are centred at one point. The numerical simulations are
performed under similar conditions as described in the previous section; however two
dimensional scanning over the cracked region is realized. The response signal gained
with the detecting coil oriented along the x-axis according to the coordinate system
shown in Figure 1 is denoted as the x- component. Accordingly, the response signals
sensed with the detection coils oriented along the y-axis and z-axis are labelled as the
y- component and the z- component, respectively.
Figure 6 displays the response signals for the non-conductive crack with a width of
wc = 0.2 mm and a depth of dc = 3 mm. As it can be seen, each component provides
some information about the inspected defect; however, x- and y- components have
lower level than the z- component.
58 L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking

absolute value [mV] absolute value [mV]

0.05
0.03 0.04
0.02 0.03
0.02
0.01
0.01
0 0
20 20
15 15
10 10
5 5
20 15 0 20 15 0
10 5 5 y [mm] 10 5 5 y [mm]
0 10 0 10
5 10 15 5 10 15
x [mm] 15 20 20 x [mm] 15 20 20

a) x- component b) y- component
absolute value [mV]

0.3

0.2

0.1

0
20
15
10
5
20 15 0
10 5 y [mm]
5 0 10
5 15
x [mm] 10 15 20
20

c) z- component
Figure 6. Crack response signal gained using proposed detector

Particular numerical results are presented in Figure 7. Only maximum points of the
sensed signals are shown in the plots. Signals of the x- and the y- components are
magnified by factor 5 to be able to display them in a same plot together with the z-
component because of the reason mentioned above. Figure 7a) presents influence of the
crack depth on the response signals amplitudes. Only signals amplitudes of the non-
conductive crack with a width of wc = 0.2 mm and depths dc ranging from 0 to 9 mm
with a step of 1 mm are depicted. Figure 7b) displays the results for various values of
the crack conductivities; c = 0, 1, 2, 5, 10% of the base material conductivity. The
other parameters of the crack are adjusted to the following values wc = 0.2 mm,
dc = 3 mm in this case. The influence of the crack opening on the crack signals
amplitudes is shown in Figure 7c). The partially conductive crack with a relative
conductivity of c = 5% and a depth of dc = 3 mm is considered here. Its width wc is
varied from 0.2 mm up to 1.0 mm with a step of 0.2 mm.
It can be observed from the results shown in Figure 7 that each component behaves
in slightly different manner when changing parameters of the crack, i.e. its depth, width
and conductivity.
It seems from Figure 7a) that the x- component is more sensitive to the crack depth
comparing to the other two components. It even does not saturate so quickly with
increasing depth of the near-side surface crack. The reason can be explained from the
nature of eddy current lines changes due to a presence of crack. The x- component is
especially formed by the perturbation field arising from eddy currents flowing around
tips of the crack that is higher than in the other two axes.
On the other side, Figures 7b) and 7c) show that the conductivity and the width of
crack influences more the y- component than the two other components as a sensitivity
L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking 59

of the y- axis coil to the perturbation field arising from eddy currents flowing through a
partially conductive crack is highest comparing to the sensitivity of the pick-up coils in
the other two axes.
It can be concluded that sensing all of the three spatial components of the
perturbation field can increase information rate of the ECT response signals. This is
especially advantageous when partially conductive cracks, like SCCs, can appear in an
inspected structure and dimensions of the cracked region need to be evaluated. Degree
of uncertainty in depth estimation of such cracks can be reduced by employing the
proposed approach.

x component 0
x component
0.4 y component y component
z component 0 05 z component

01
0.2 c
imaginary [mV]

imaginary [mV]
0 15
dc
0 02

0 25
0.2
03

0 35
0.4

04
0.4 0.2 0 0.2 0.4 0 0 05 01 0 15 02 0 25 03 0 35 04
real [mV] real [mV]
a) influence of crack depth b) influence of crack conductivity
0

0.05

0.1
wc
imaginary [mV]

0.15

0.2

0.25

0.3

0.35 x component
y component
z component
0.4
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
real [mV]

c) influence of crack width


Figure 7. Amplitude of crack response signal gained using proposed detector

Conclusion

The paper focused on degree of uncertainty in depth evaluation of a detected crack


from eddy current testing signals. The investigations were done from numerically
calculated signals. Non-conductive as well as partially conductive cracks were
concerned here. Three parameters of a crack were varied, i.e. its depth, conductivity
and width. Standard pancake probe was employed for the inspection at first. The
presented results demonstrated that the inverse problem is highly ill-posed when only
one component of the perturbation field is picked-up for a single scanning line just
60 L. Janousek et al. / Decreasing Uncertainty in Size Estimation of Stress Corrosion Cracking

above a crack along its length. Consequently a new inspection approach was proposed
in the paper. It comes from the fact that more information about a detected crack can be
gained by sensing all of the three spatial components of the perturbation
electromagnetic field. Two dimensional scanning is required in this case. It was shown
that the response signals sensed with three coils that are oriented perpendicularly to
each other behave in slightly different way when changing the parameters of crack.
This is advantageous as degree of uncertainty in depth evaluation especially of partially
conductive cracks, as stress corrosion cracking, can be reduced in this way. Further
work is going to be concentrated on experimental verification.

Acknowledgements

This work was supported by the Slovak Research and Development Agency under the
contract No. APVV-0194-07. This work was also supported by a grant of the Slovak
Grant Agency VEGA, project No. 1/0308/08.

References

[1] N. Yusa, H. Huang, K. Miya, Numerical evaluation of the ill posedness of eddy current problems to size
real cracks, NDT&E International 40 (2007), 185 191.
[2] G. Rubinacci, A. Tamburino, S. Ventre, Fast numerical techniques for electromagnetic nondestructive
evaluation, Nondestructive Testing and Evaluation 24 (2009), 165 194.
[3] N. Yusa, Development of computational inversion techniques to size cracks from eddy current signals,
Nondestructive testing and evaluation 24 (2009), 39 52.
[4] Z. Chen, N. Yusa, K. Miya, Some advances in numerical analysis techniques for quantitative
electromagnetic nondestructive evaluation, Nondestructive testing and evaluation 24 (2009), 69 102.
[5] B.P.C. Rao, An artificial neural network for eddy current testing of austenitic stainless steel welds,
NDT&E International 35 (2002), 393 398.
[6] M. Rebican, Z. Chen, N. Yusa, L. Janousek, K. Miya, Shape reconstruction of multiple cracks from ECT
signals by means of a stochastic method, IEEE Transactions on Magnetics 42 (2006), 1079 1082.
[7] N. Yusa, L. Janousek, M. Rebican, Z. Chen, K. Miya, N. Dohi, N. Chigusa, Y. Matsumoto, Caution
when applying eddy current inversion to stress corrosion cracking, Nuclear Engineering and Design
236 (2006), 211 221.
[8] N. Yusa, K. Miya, Discussion on the equivalent conductivity and resistance of stress corrosion cracks in
eddy current simulations, NDT&E International 42 (2009), 9 15.
Electromagnetic Nondestructive Evaluation (XIV) 61
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-61

Semi-Discrete Time-Domain Sensitivity


Analysis For Cracks Recognition

Konstanty M. GAWRYLCZYK
West Pomeranian University of Technology in Szczecin, Poland
Department of Electrotechnology and Diagnostics

Abstract. Semi discrete method is known since 80s of XXth century. The method
provides analytical solution in time, so the time stepping may be omitted.
Comparing to usual finite elements method in time, this method seems not to be
numerically effective, because produced matrices are dense. Because of this reason,
it was rather rarely used. However to carry out sensitivity analysis with adjoint
models [1] solutions in forward and backward time have to be obtained. Both time
points should coincide. For space discretization finite elements are used, as usual.
The semi discrete method allows to determine analytically the continuous solution
for any given time of analysis. In this paper evaluation of this method for different
kinds of excitation shapes is presented. The sensitivity analysis is applied for
inverse task of recognition of conductivity distribution in non destructive testing
of materials using eddy currents.

Keywords. Time domain field analysis, sensitivity analysis, cracks recognition.

Introduction

The proposed semi-discrete method allows obtaining time-domain solution without


time-stepping. For space discretization triangular finite elements of first order are used.
The proposed method is used for algorithms of non-destructive testing (NDT) using
eddy-currents. The excitation currents used by NDT in time domain have usually the
shape of rectangular impulse or single sinusoidal impulse. The numerical solutions for
both current shapes are listed below. For identification of crack shape the iterative
algorithm is given, based on measurement of coil voltage moving over the crack and
sensitivity analysis. To calculate sensitivity the Tellegens method is applied. The
derivation of sensitivity equations in time domain one can find in autors earlier works,
e.g. [1], while the sensitivity equations for frequency domain in [4]. The semi-discrete
method delivers analytical and continuous solution for any given time of analysis,
which has a form of exponential functions. In order to obtain an analytical formula, the
integration of the sensitivity equation is necessary. The paper finds possible solutions
of this problem, either the application of Zassenhaus formula or improvement of
commutational properties of two matrices. The resultant gradient information may be
used for solving inverse problems, such as identification of material conductivity
distributions.
62 K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition

1. FE-TS analysis

Electromagnetic field diffusion into conducting region for two-dimensional models


may be described by vector magnetic potential A:

sA
2 A  0 i t
 f t
. (1)
st

Approximating (1) with finite elements and applying time-dependent elements the
solution with well known Finite Element Time Stepping method is obtained, leading
to the following system of linear equations:


K M A ti 1

t
(2)
M
 f ti 1
1
f ti
 1
K A ti
,
t

with: K stiffness matrix, M mass matrix, t time step, i - time index and
0,5    1 defines the differential scheme of the time stepping method. The effective
solution of (2) is as long possible, as the time step t remains constant. In other case,
matrix decomposition should be renewed.
If there is the need to calculate the sensitivities of measured voltage versus
material parameters, adjoint models are used [1], which appear in Tellegen theorem.
The quantities relating to adjoint models are denoted by the index (+). Time moments
from the analysis of adjoint model should coincide with that of original model. Because
the adjoint model is analyzed applying reverse time, fulfillment of this condition may
lead to excessive time steps and very long computational time. This is the reason, why
the semi-discrete method has been elaborated. It allows to calculate the solution for any
desired time moments, not only equidistantly placed.

2. Semi-discrete finite element analysis

The inhomogeneous diffusion equation to be solved has the form of

A(t)
[ K ]{A(t)} + [ M ] = { f (t)}. (3)
t

The solution is calculated as superposition of transient (t) component and steady-


state solution (s) of A = As + At . Steady-state solution depends on the shape of
excitation f(t). Transient component results from homogeneous equation:

sAt (t)
<K >\At (t)^ <M > 0 , (4)
st

which solution is
K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition 63

\At (t)^ \C^exp t< M > <K >


, with: \C^constant of integration.
1
(5)

The constant C depends on initial value A(0), which is zero in our case, and also
on the excitation shape. The solutions for different excitations f are shown below.

3. Unit-step excitation

However the current exciting a coil can not take a form of unit-step 1(t), it is handy
approximation of the real shape of current. The steady-state response of the magnetic
vector potential As in this case has the form:

1 1
\As (t)^< K > \ f (t)^<K > 1 t
. (6)

The constant of integration C can be evaluated from the initial condition

\As (0)^ \At (0)^\A(0)^0 (7)

and the semi-discrete solution for vector magnetic potential takes the form:

{ A (t )} = ([1] exp ( t [ M ] [ K ]) ) [ K ] {1(t)} .


1 1
(8)

Because of mass matrix inversion, (8) cannot be applied for the whole region.
Outside the conducting region the mass matrix contains zeroes on main diagonal and
for this reason becomes singular. The region has to be divided our into a conductive
part 1 and a non-conductive part 2:

s
K K12 A1 t
M11 0 A1 t
0
11 st  . (9)
K 21 K 22 A2 t
0 0 f2
0

It is assumed that excitation currents are located only in a non-conducting part of


the region. Then the following matrix equation for the conductive region is obtained:

<K11 ><K12 ><K22 > <K21>


\A1 (t)^ <M11>\sA1 (t)/st^
1

1
<K12 ><K22 > \ f2 (t)^, or: <Kc >\A1 (t)^ <M11>\sA1 (t)/st^\ fc (t)^. (10)

The similarity of (10) to (3) allows for exploitation of solution (8). The correctness
of described method was proved using very simple 2D-model (Fig. 1). The model
consists of conducting region, air and linear excitation current. The material parameters
are given on the picture.
64 K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition

Figure 1. 2D model for proof of semi discrete method

Figure 2. Comparison of A in node 1 for ti


Comparison of magnetic vector potential Ai (i - means the index of time moment)
calculated using traditional FE-TS method and semi-discrete method is shown in
Fig. 2. The comparison done for other nodes seems similarly.

4. Harmonic excitation

For harmonic excitation there is the following inhomogeneous differential equation:

sA(t)
<K >\A(t)^ <M > \I m sin(t)^ . (11)
st

The steady-state response As may be evaluated in frequency-domain using standard


FEM:
K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition 65

1
\As^ <K > j<M >
\^,
(12)
1
\As (t)^imag <K > j<M >
\^exp(jt) .

The constant of integration C for transient component At (Eqn. 5) will be similarly


evaluated, as described by Eqn.7 and assuming zero initial condition:

1

\C^imag <K > j<M >


\^ imag As
. (13)

Therefore, the semi-discrete solution for transient magnetic vector potential in


presence of harmonic excitation equals:

\A(t)^imag As exp(jt)
imag As
exp t<M > <K >
.
1
(14)

Obtained formula was tested using the same model as this shown in Fig.1. In Fig.3
there are solutions obtained with FE-TS, with proposed semi-discrete method and the
steady-state shape of potential for node number 1. i means the time-moment index.

Figure 3. Comparison of magnetic vector potential in node 1 for harmonic excitation

5. Excitation with the rectangular impulse

The solution for rectangular impulse was achieved as superposition of solutions for two
unit-step impulses:

(<1>et<M11>1<Kc > ) K 1 1(t) if t b T ,


< > \ ^
\A t
^  t-T
<M11>1<Kc > t<Mc 11>1<Kc > 1 (15)
(e e )<Kc > \1(t)^ otherwise.
66 K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition

6. Single sinusoidal pulse

The solution for single, sinusoidal current pulse was derived in the same manner, using
superposition of two solutions given by Eqn.(14):

1
t<M > <K >
imag As e jt  As
e if tbT ,
\A(t)^ (16)

1 1
 tT
< M > <K >
imag As
e  As
e < > < > otherwise.
t M K

The comparison done for test models reveals good conformity of solutions
obtained using semi-discrete method versus this from classical FE-time-stepping.

7. Semi-discrete sensitivity analysis

To perform inverse task basing on sensitivity analysis there has to be calculated


sensitivity of voltage induced in the coil versus conductivity in finite elements , it
means the left hand side of equation (17). The term on the right-hand side of
sensitivity equation (17) denotes the sensitivity components coming from finite
elements. This sensitivity equation for time domain was derived in [1]. (+) denotes
value related to adjoint model.

T T

( J 0 ( ) E ( t ) ) dV dt = E ( t )  E ( ) dV dt.
+ +
(17)
0V 0V

Integration of the product of electric intensity vectors over the finite elements of
first order requires the Gaussian quadrature of the second order. Because Gaussian
points are located on the element borders in this case, the products containing
combinations of nodal values should be calculated. To obtain all necessary products,
the following matrix R is defined:

( ) {i }
T T
R = E + ( ) E (t ) dt = [ M11 ] exp t [ K c ][ M11 ]
T 1 1
c
0 0
(18)
( exp ( [ K ][ M ] ) {i })
T
[ M11 ]
1 1
c 11
+
c dt

where
 is the size of finite element. The matrix R can be rewritten as follows:

( ) {i }{i }
T
R = [ M11 ] exp t [ K c ][ M11 ]
1 1 + T
0 c c
(19)
[ M11 ] exp ( [ K ][ M ] ) dt,
1 1
c 11
K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition 67

1
because the term [ M11 ] exp [ Kc ][ M11 ] ( 1
) is symmetric. A drawback of the
{ }
T
established definition is that R is singular (it contains a singular term [ I c ] = {ic } ic+ ).
For better numerical efficiency, analytical integration would be essential. Since simple
disentangling of exponential functions in (19) is impossible, two following methods for
this purpose are proposed.

8. The Zassenhaus formula

The Zassenhaus formula [3] is a version of the Baker-Campbell-Hausdorff formula. It


can be written involving commutator nestings:

1
exp( X ) Y exp( X ) = Y + [ X ,Y ] + [ X , [ X ,Y ]] + 1 X , [ X , [ X ,Y ]] + ... ,
2! 3! (20)
where: [ X ,Y ] = X Y Y X .

In our case:

X = t [ K c ][ M11 ] and Y = {ic } {ic+ } [ M11 ] = [ I c ] [ M11 ] .


1 T 1 1
(21)

If 1/M is the norm of matrices X and Y, an error of the Zassenhaus approximation


is of the order O(1/M 2). It means that sensitivity calculations based on the Zassenhaus
formula converge only for short times t, dependent on the norm of matrix [Kc][M11]-1.
In practical calculations the convergence is obtained at the beginning of the transient
process only.

9. Improvement of matrix commutations

By applying the Zassenhaus formula for two good commuting matrices, the result is

exp( X ) Y exp( X ) = Y , assuming: [ X ,Y ] = 0. (22)

Because the identity matrix commutes well with all other matrices, the identity
matrix [1] multiplied by an arbitrarily large coefficient C is added and subtracted:

( )
T
R = [ M11 ] exp t [ K c ][ M11 ]
1 1
0
(23)
(
{ I c } [ M11 ] + C [1] C [1]) exp ( [ K ] [ M ] ) dt.
1 1
c 11

The components in above integral can be separated as follows:


68 K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition

( ) exp ( LA) exp ( [ K ][ M ] ) dt


T
R = [ M11 ] exp t [ K c ] [ M11 ]
1 1 1
c 11
0

[ M ] exp ( t [ K ] [ M ] ) C [1]
T 1 1
11 c 11
0 (24)
exp (t [ K ][ M ] ) exp ( T [ K ] [ M ] ) dt,
1 1
c 11 c 11

(
where: LA = ln {I c } [ M11 ] + C [1] .
1
)
Now, the exponents in above formula can be disentangled and integrated:

( ( )
R = T [ M11 ] exp T [ K c ][ M11 ] + LA C exp -T [ K c ][ M11 ]
1 1
( 1
)). (25)

This formula allows to calculate sensitivity components for the right-hand side of
sensitivity equation (17).

10. Recognition of conductivity distribution

The dependence of measurement coil (shown in Figs. 4, 5) voltage on the variation of


electric conductivity inside finite elements is represented by the sensitivity matrix
[S(t)] for consecutive time steps i :

V (t1 ) S1 (t1 ) S2 (t1 ) S3 (t1 ) .... S j (t1 ) 1


V (t ) S (t ) S (t ) S (t ) .... S j (t2 ) 2
2 1 2 2 2 3 2
V (t3 ) = S1 (t3 ) S2 (t3 ) S3 (t3 ) .... S j (t3 ) . 3 (26)

.... .... .... .... .... .... ....
V (ti ) S1 (ti ) S2 (ti ) S3 (ti ) .... S j (ti ) j

with : j - number of finite elements in the search region,


i - number of time steps of electromagnetic field evaluation.

The terms of sensitivity matrix are derived from sensitivity equation (17). V
means the differences between the measured and actually simulated coil voltage.
Solving (26) as linear equations system the corrections to conductivity in finite
elements  are obtained. It can be called quasi-Gauss-Newton algorithm. The process
is non-linear and must be repeated iteratively. In the case of poor convergence
truncated singular value decomposition of the sensitivity matrix is applied, as described
in [1]. The process starts with initial configuration, in our case with a conductivity
distribution without cracks. The search regions were applied with number of elements j
lower then the number of time steps i, so the equations system (26) was over-
determined. The excitation current had the shape of unit-step, so the solution (8) was
exploited. After each iteration the conductivity in search region was corrected with .
For the next iteration the sensitivity S was evaluated anew.
When the coil moves over the crack, number of equations i consists of number of
time steps multiplied by the number of coil positions (10 steps in examples below).
K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition 69

The examples below (Figs. 4, 5) show conductivity distribution inside tube wall
estimated after 15 iterations. The input data for assumed crack shape were taken not
from measurement, but from simulation utilizing the FEM-model with coarser
discretization. It gave the similar effect as a presence of noise provided by real
measurement. The original flaw had non-zero conductivity of = 0,1[MS/m]. The size
of test flaws is shown in Figs.4 and 5. The solutions were carried out using modified
vector potential u = Ar and cylindrical symmetry.

Figure 4. Rectangular flaw shape and conductivity distribution obtained after 15 iterations.

Figure 5. Skew flaw and conductivity distribution obtained after 15 iterations.


70 K.M. Gawrylczyk / Semi-Discrete Time-Domain Sensitivity Analysis for Cracks Recognition

Conclusions

The drawback of numerical identification of flaws basing on sensitivity analysis is high


computation cost. Comparison of the semi-discrete method efficiency to classical FE-
TS shows, that despite of high demand for memory, the described method may
compete in relation to finite elements with the time stepping. Although in this work
only two-dimensional examples are presented, conclusions drawn above are directly
applicable to solution of three-dimensional models. The results of sensitivity evaluation
using semi-discrete method were validated comparing it to usual finite elements with
time stepping.
We obtained good convergence for simulated voltages of measurement coil. It has
to be pointed, that the convergence of numerical identification algorithm described in
this paper depends strongly on precise measurement of coil voltage (its time function).
The measurement has been simulated using finite element model with coarse
discretization. In the case of application of real measurement data the convergence of
identification process would be worse.
In future works the described method will be combined with space mapping
technique, described in [5] and successfully applied to defect characterization in [6].

References

[1] K. M. Gawrylczyk, M. Kugler, Time domain sensitivity analysis of electromagnetic quantities utilizing
FEM for the identification of material conductivity distributions, COMPEL, 25 (2006), 589 598.
[2] K. M. Gawrylczyk, M. Kugler, Semi discrete time domain sensitivity analysis of electromagnetic field,
COMPEL, 28 (2009), 1138 1348.
[3] R. Sridhar, R. Jagannathan, On the q analogues of the Zassenhaus formula for disentangling exponential
operators, Journal of Computational and Applied Mathematics 160 (2003), 297 305.
[4] D. N. Dyck, D. A. Lowther, E. M. Freeman, A Method of Computing the Sensitivity of Electromagnetic
Quantities to Changes in Materials and Sources, IEEE Transactions on Magnetics, Vol. 30, No. 5
(1994), 3415 3418.
[5] J. W. Bandler, Q. S. Cheng, S. A. Dakroury, A. S. Mohamed, M. H. Bakr, K. Madsen, J. Sondergaard,
Space mapping: the state of the art, IEEE Transactions on Microwave Theory and Techniques, vol.52, 1
(2004), 337 361.
[6] R. K. Amineh, S. Koziel, N. K. Nikolova, J. W. Bandler, J. P. Reilly, A space mapping methodology for
defect characterization from magnetic flux leakage measurement, IEEE Transactions on magnetics, vol
44, 8 (2008), 2058 2065.
Electromagnetic Nondestructive Evaluation (XIV) 71
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-71

Metamodel as Input of an Optimization


Algorithm for Solving an Inverse
Eddy-Current Testing Problem
Rmi DOUVENOT a,1 , Marc LAMBERT a and Dominique LESSELIER a
a
Dpartement de Recherche en lectromagntisme,
Laboratoire des Signaux et Systmes UMR 8506 (CNRS - SUPELEC - Univ Paris-Sud),
91192 Gif-sur-Yvette, France

Abstract A method to characterize defects by processing eddy current testing


(ECT) signals is presented. It works in a short time for generic defects and work
pieces. To retrieve the dimensions of a defect, a metamodel based Particle Swarm
Optimization (PSO) is considered. Indeed, using a metamodel as an input of a
stochastic method enables to signicantly speed it up. The metamodel is generated
using an adaptive database generation. Both tube and plate congurations, corre
sponding with either simulated or laboratory controlled measured data are consid
ered as illustration. Good accuracy and satisfactory speed of the method are exhib
ited, additional information besides the inversion results being provided as well,
which highlights possibly indeterminate cases, making the method useful also for
decision analysis.
Keywords. Eddy current testing, Metamodel based inversion, Particle swarm
optimization, Decision analysis

Introduction

Eddy-current testing (ECT) is a well-known non-destructive technique for defect detec-


tion in metal parts. Yet ECT signals can be used also to characterize a defect, beyond
mere detection. In the framework of the present study, a (quasi-)real-time method is
aimed at. This implies that there is strong a priori information about the tested workpiece
and the type of defects to be characterized. Previous efforts to reach this goal have fo-
cused on learning algorithms, with encouraging results as seen in [1,2]. To improve the
learning step, adaptive databases have been introduced [3].
Here, the adaptive databases are employed to create a metamodel, which is used as
an input of a Particle Swarm Optimization (PSO). A fast stochastic inverse method is
then obtained. It is put to the test on a number of cases, involving synthetically generated
data and laboratory-controlled measured data, in both tube and plate congurations.
To create the metamodel, an adaptive database is generated, refer to [4]. This
database in practice concentrates a higher number of points where the output variations
are the fastest ones, with respect to those of the inputs. In so doing, all the variations of
1 Corresponding author e mail: remi.douvenot@lss.supelec.fr
72 R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm

the outputs with relatively few points can be encompassed. This being made, a classical
radial basis function (RBF) interpolation [5] with a thin plate spline kernel function is
applied. The adaptive database interpolated by RBF is the metamodel that is used to feed
the stochastic inverse method.
The performances of the metamodel-based inversion method are illustrated for two
congurations, each corresponding to one benchmark: a circular tube workpiece affected
by groove defects (two parameters are to be retrieved) and a plate with parallelepiped de-
fects (three parameters are to be retrieved). For each conguration, the method is applied
on many simulated signals and a couple of measured data extracted from the benchmarks.

1. The metamodel-based PSO method

1.1. Forward problem solver

Let us assume that tests are performed on generic pieces. Constitutive materials and ge-
ometrical dimensions of both workpieces and probes are known, at a given frequency of
operation. The input parameters are the ones describing the defect inside the workpiece.
Then the response of the awed piece to the probe is computed via an integral volume
formulation (see for example [6]), and a variation of impedance is obtained as an output.
The CIVA simulation software [7] is used for this purpose. If P are the input parame-
ters and y is the measured or simulated output impedance, we denote y = L(P). Other
methods could obviously be preferred for carrying out the forward simulations. For our
purpose, the forward simulations are only used to generate the database.

1.2. Generation of the adaptive database

The aim of the adaptive database is to get, at the end, input parameter sets with equivalent
simulated output impedance variations, which are such that there exists a higher density
of inputs where the output variations are faster.
The database is created iteratively in order to achieve this goal. This adaptive
database generation is described in some more detail in [4]. First, a coarse mesh is gen-
erated. Then, at each edge of each simplex of the mesh, simulations are compared to
linearly interpolated data. If the interpolation error , dened below, is larger than a
prescribed stopping criterion, a node is added in the middle of the longest edge of the
simplex. An example of a 2D mesh is shown in Figure 1. The interpolation error is
expressed as

||y y||22
= (1)
||y||22

where y and y are the exact and the interpolated data, respectively.
Once all calculations have been performed, the database is a set (Pi , yi )i=1, ,N
where N is the size of the database, each vector P being of dimension D (number of
parameters to describe a defect), and each vector y being of dimension M (number of
measurement points for one conguration).
The generation of the adaptive database certainly is a crucial step for metamodel-
based optimization. Depending upon the formulation of the interpolation error, different
R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm 73

Parameter 2
00 1
Parameter 1
Figure 1. Example of a 2D adaptive database.

behaviors of the interpolated data can be observed (a better accuracy on low or high
amplitude signals for instance) [8].

1.3. Interpolation of the database

The aim of interpolating the database is to create a metamodel. The operator L is ap-
proached via an expression depending on the data in the database, this involving little
computational effort compared to a forward ECT simulation.
Once the database has been created, an RBF interpolation is used to create the meta-
model. That is, when an input P is given, the output y can be approximated by the
quantity


N
y = L(P) = wi K(P, Pi ), (2)
i=1

where K is a so-called kernel function, and where the weights wi are computed dur-
ing a training step. In the following, the thin plate spline kernel function [9], which is
commonly used for interpolation, is chosen (Eq. 3). However, other kernels could be
preferred without stringent differences being expected.
The thin plate spline kernel can be expressed as

P1 P2 22 ln (P1 P2 2 ) if P1 =
 P2
K(P1 , P2 ) = (3)
0 if P1 = P2

1.4. Stochastic optimization

The stochastic optimization is carried out by Balanced PSO [10] as a variant of the clas-
sical PSO method [11]. It is a metaheuristic based on swarm intelligence. Its principle
is that particles moving in the input space while sharing information can nd the global
minimum.
Having an impedance data y (the objective function), particles (usually about 20
[10]) are positioned in the input space with given initial speeds. For each one, a cost
function is calculated, and the particles are moved towards likelier positions. The parti-
cles share information with some of them (generally 3 or 4) to know which particle has
obtained the minimum cost function. Thus, the updated speeds of the particles depend
74 R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm

on their current speeds, weighted by an inertia w, on their best personal position, called
personal best p, and on the best position known by the particle, called social best s.
At iteration k, and along each dimension d [1, . . . , D], the positions x of the
particles are updated according to the formula:

vd (k + 1) = w vd (k) + r1 c1 (pd (k) xd (k)) + r2 c2 (sd (k) xd (k))


(4)
xd (k + 1) = xd (k) + vd (k + 1)

where r1 and r2 are uniformly chosen in [0,1] (independently on each dimension). w, c1 ,


and c2 are constants chosen following [11] (w = 0.7 or 1, randomly; c1 = c2 1.193).
The method is known to be relevant for solving nonlinear inverse problems [12], and
its balanced version, which is consisting in sending a particle around a local minimum
at each iteration, improves the compromise between exploration of the input space and
exploitation of the interesting areas [10].
Stochastic methods as PSO are expensive since they require many simulations of the
forward problem. With a metamodel to replace the simulation, the cost is signicantly
reduced and the method becomes relevant in view of almost real-time applications. Yet,
consequently, the accuracy of the optimization is directly related to the accuracy of the
metamodel.
One of the advantages of stochastic methods is also that the many cost function
evaluations to be performed can be used in order to localize areas of large likelihood,
which is useful to obtain a decision analysis tool.

2. Results on benchmark data

Results are discussed for two standard ECT congurations: a tube and a plate.

2.1. Tube workpiece

As rst conguration, a circular cylindrical non-magnetic tube has been chosen. It is


considered of innite extent along the tube axis. Its internal diameter is 19.68 mm and
the wall thickness is 1.27 mm. The test probe is made of a 70-wrapping emitting bobbin
of internal diameter 15.66 mm, thickness 0.67 mm, and length 2 mm, and of an identical
receiving bobbin which is placed at 0.5 mm of the emitting one. The impedance variation
is collected on a distance of 19 mm. This conguration is extracted from the 2008 ECT
benchmark that is discussed in [13].
To successfuly apply the presented inversion method, strong a priori information is
required. Here, the defect is assumed as being a circular notch with varying length and
depth. Its depth is taken between 10 and 90 % of the wall thickness, and its length is
taken between 0.1 and 10 mm. The defect can either open at the internal surface or the
external surface of the tube.
First, two databases are generated, one database per type of defect. Then, two meta-
models are generated. The inversion method is tested on 400 test cases randomly selected
(defect parameters) and synthetically generated (impedance variations): 200 internal de-
fects and 200 external defects. A metamodel-based PSO inversion is carried out on each
generated impedance variation. The inverted parameters are the likeliest ones obtained
with this inversion. The type of defect (internal or external) is assumed to be known, and
R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm 75

90 90

80 80

70 70
100 - Depth (%)

60

Depth (%)
60

50 50

40 40

30 30
PSO likelihood
20 PSO parameters 20
True parameters
10 10
2 4 6 8 2 4 6 8
Length (mm) Length (mm)
(a) GE40 (b) GI10

Figure 2. Results of the metamodel based PSO inversion on measured data GE40 and GI10.

the inverse problem reduces to a regression on two continuous parameters. The normal-
ized quadratic error q = (Pd Pd )2 /Pd2 is computed for each test case and is averaged
on the 200 test cases.
For the internal defects, the average normalized quadratic error q is 1.5 % for the
depth and 1.9 % for the length. For the external ducts, the error q is 0.6 % for the depth
and 0.5 % for the length. Since the metamodel is used as the input of the optimization
algorithm, these results are obtained in less than 2 seconds on a 3 GHz personal computer.
Two specic results are detailed. They are results on real data corresponding to this
conguration and given in [13]. Measurements are avaiable for two defects: GE40 is an
external defect with a 40 % depth, and GI10 is an internal defect with a 10 % depth.
Both have the same length of 1 mm. Figure 2 shows the results on these two sets of
measured data. Darker areas correspond to the likeliest parameter values, the lighter ones
to less likely parameter values. The approximation of the likelihood is obtained using
the computations carried out during the inversion. The circles correspond with the true
parameters of the defects, and the squares give the likeliest ones retrieved by PSO.
The approximation of the likelihood gives all the sets of parameters for which the
signal due to the reconstructed defect is close to the measured one (the darkest area).
Since different defects can lead to similar signals, comparing the measured signal with
the one due to the more likely defect is not relevant for the purpose of making a decision
analysis tool.
The results obtained by the metamodel-based PSO inversion with this congura-
tion appear very good in a reasonable time. The result obtained on GI10 is particularly
promising because it is a very low signal, harder to invert. Comparisons of this method
with support vector machines on this data can be found in [8].

2.2. Slab workpiece

As second conguration, a non-magnetic plate workpiece is considered, from the


COFREND benchmark [14]. Its width and length are considered innite, and its thick-
ness is 1.55 mm. The probe is a 328-wrapping bobbin of internal diameter 2 mm, thick-
76 R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm

80 80
100 - Depth (%)

100 - Depth (%)


60 60

40 40

20 20

0 0
5 10 15 0.2 0.4 0.6 0.8 1
Length (mm) Width (mm)

14

12
PSO likelihood 10
Length (mm)
PSO parameters
True parameters 8

6
Figure 3. Results of the metamodel based PSO inver
4
sion on measured data F5 (depth 100 %, length 2 mm,
and width 0.1 mm). 2

0.2 0.4 0.6 0.8 1


Width (mm)

ness 1.25 mm, and length 2 mm. The probe scans a 8 mm per 12 mm surface. The defect
position is assumed to be known, at the center of the scanned zone.
The defect is a rectangular notch with varying length, width, and depth. Its depth is
considered between 10 and 100 % of the tube thickness, its length is taken between 0.5
and 15 mm, and its width between 0.1 and 1 mm.
The average normalized quadratic error q is 0.6 % for the depth, 1.0 % for the width,
and 3.2 % for the length of the defect. These results are obtained with an average time of
5.5 s on the same computer as in section 2.1.
So, in general, the accuracy of the inversion is good. Two measurements correspond-
ing to this benchmark have been processed. F5 corresponds to a defect of depth 100 %,
length 2 mm, and width 0.1 mm. F7 corresponds to a defect of depth 40 %, length 10
mm, and width 0.1 mm. The two inverted results are displayed in Figures 3 and 4, re-
spectively. Both results are quite similar: the inverted parameters are acceptable yet not
perfectly retrieved. However, the true parameters are inside the likeliest areas. In par-
ticular, it is difcult to precisely decide upon the depth of crack F5 or upon the width
of the crack F7. These examples show how undetermined cases can be identied by the
inversion method.
If other inversions were to be carried out on the same data, and since a stochastic
inversion is non repeatable, slightly different results could be found (minimum of the
cost function). However, the likeliest area should be approximately the same. So, the
inversion tool is appropriate for decision analysis.
R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm 77

80 80
100 - Depth (%)

100 - Depth (%)


60 60

40 40

20 20

0 0
5 10 15 0.2 0.4 0.6 0.8 1
Length (mm) Width (mm)

14

12
PSO likelihood 10
Length (mm)

PSO parameters
True parameters 8

6
Figure 4. Results of the metamodel based PSO inver
4
sion on measured data F7 (depth 40 %, length 10 mm,
and width 0.1 mm). 2

0.2 0.4 0.6 0.8 1


Width (mm)

3. Conclusion

A metamodel-based optimization as a means to perform defect characterization by pro-


cessing ECT signals has been presented. Interesting results on both simulated and mea-
sured data have been shown. This method requires strong a priori information (work-
piece and global shape of the defect), but it can be applied to various problems. Defects
in a tube and a plate have been retrieved successfully. Moreover, the time necessary to
invert the data is reasonable and rather promising for on-site use. Let us note that this
tool is also able to retrieve the local minima of the cost function, which makes it suited
to decision analysis.
A representation of the data in a lower dimensional space (by principal component
analysis for instance) could speed up the RBF interpolation. More forward simulations
could then be performed in the stochastic method and the inversion could be improved.
Indeed, the more time devoted to the stochastic inversion, the more accurate the inver-
sion. But, the true challenge is to apply this type of methods to problems involving, say,
5 input parameters or more. Indeed, as the dimension is increased, the generation of an
accurate database requires more and more forward simulations, possibly far too many for
realistic use. In this case, other means in metamodel generations should be considered
[15].
78 R. Douvenot et al. / Metamodel as Input of an Optimization Algorithm

Acknowledgements

The authors thank CEA (Commissariat lnergie Atomique) and the COFREND work-
ing group Eddy current modelization for providing the measured benchmark data.

References

[1] L. Udpa and S. Udpa. Eddy current defect characterization using neural networks. Mater. Eval., 48:342
347, 1990.
[2] A. Bernieri, L. Ferrigno, M. Laracca, and M. Molinara. Crack shape reconstruction in eddy current
testing using machine learning systems for regression. IEEE Trans. Instrum. Meas., 57:1958 1968,
2008.
[3] J. Pv and S. Gyimthy. Adaptive inversion database for electromagnetic nondestructive evaluation.
NDT & E Int., 40:192 202, April 2007.
[4] G. Franceschini, M. Lambert, and D. Lesselier. Generation of adaptive database results synthesis.
Technical report, Laboratoire des Signaux et Systmes, January 2009.
[5] M. D. Buhmann. Radial Basis Functions: Theory and Implementation. Cambridge University Press,
Cambridge, 2003.
[6] J. R. Bowler, S. A. Jenkins, Sabbagh L. D., and Sabbagh H. A. Eddy current probe impedance due to
volumetric aw. J. Appl. Phys., 75:8128 8137, 1991.
[7] CIVA: State of the art simulation platform for NDE. available at http://www civa.cea.fr, accessed the
8th of June 2010.
[8] R. Douvenot, M. Lambert, and D. Lesselier. Particle optimization with metamodel for crack character
ization. In Proceedings of the 2010 URSI Electromagnetic Theory Symposium (EMTS 2010), 2010. 4
pp., to appear.
[9] F. L. Bookstein. Principal warps: Thin Plate Splines and the decomposition of deformations. IEEE
Trans. Pattern Anal. Mach. Intell., 11:567 585, 1989.
[10] M. Clerc. The mythical balance or when Particle Swarm Optimisation does not exploit. Unpublished, 6
pp., available at http://clerc.maurice.free.fr/pso/Balanced_PSO/Balanced_PSO.pdf, 2008.
[11] M. Clerc. Particle Swarm Optimization. ISTE, London, 2006.
[12] P. Rocca, M. Benedetti, M. Donelli, D. Franceschini, and A. Massa. Evolutionary optimization as
applied to inverse scattering problems. Inverse Problems, 25:123003, 2009. 41pp.
[13] C. Reboud, G. Pichenot, and S. Mahaut. 2008 ECT benchmark results: Modeling with CIVA of 3D
aws responses in planar and cylindrical workpieces. In 35th Annual Review of Progress in Quantitative
Nondestructive Evaluation (QNDE - 2008), Chicago, 20 25 july 2008. 8 pp.
[14] COFREND benchmark. available at http://www.cofrend.com/modelisationcf/GT_CF_COFREND_
GB. html, accessed the 8th of June 2010.
[15] R. R. Barton. Simulation optimization using metamodels. In Proceedings of the 2009 Winter Simulation
Conference, pages 230 238, Austin, TX, December 13 16 2009.
Electromagnetic Nondestructive Evaluation (XIV) 79
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-79

Fast Multipole Method for 3D


Electromagnetic Boundary Integral
Equations. Application to Non Destructive
Testing on Complex 3D Geometries
Tekoing LIM a Gregoire PICHENOT a and Marc BONNET b
a
CEA-List, 91191 Gif-sur-Yvette, France
b
LMS, Ecole Polytechnique, 91120 Palaiseau, France

Abstract. The motivation of this work is to develop a computational treatment of


eddy current simulation applicable to complex 3D geometries. To this end, an ap-
proach based on surface integral equations and involving the homogeneous full-
space Greens function and dyad is developed. A Galerkin variant of the method of
moments, using bilinear basis functions, is employed for discretizing the surface in-
tegral equations, leading to a linear system. Solving the linear system using a direct
matrix solver is prohibitively expensive in computing time and memory resources.
This difculty is circumvented by using an iterative solver (typically GMRES) to-
gether with the Fast Multipole Method (FMM). The response of an eddy current
probe is nally expressed with the Auld reciprocity theorem.
Keywords. Fast Multipole Method, Surface Integral Equations, Method of Moments,
Singularity

Introduction

The response of an eddy current probe can currently be modelled in the CIVA non de-
structive testing (NDT) simulation platform (www-civa.cea.fr) by eddy current technique
(ECT) simulation for planar and cylindrical geometries. The developed ECT simulation
models are mainly based on the volume integral method using the dyadic Green formal-
ism, for which semi-analytical representations are known for canonical geometries. The
motivation of this work is to develop a treatment addressing complex 3D geometries (see
Figure 1(a)), which responds to real industrial interest. It aims at modelling the response
of an eddy current probe on complex, homogeneous and non-magnetic 3D geometries
affected by a aw (see Figure 1(b)).

1. The proposed approach

As a semi-analytical representation of the Green dyadic is unknown for complex 3D ge-


ometries, an alternative approach is developed. This approach involves the homogeneous
80 T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations

(a) (b)

Figure 1. (a) Imaginary part of the surface density current J for f = 105 Hz and = 106 S/m (results for
1062 edges) and (b) a complex 3D geometry affected by a aw.

full-space Green function gki (r, r ) and dyad Gki (r, r ) where ki is the wavenumber in
vacuum (i = 0) or in the medium (i = 1). It is based on transmission conditions on the
surface of the object S given by

n0 E0 = n1 E1 and n0 H0 = n1 H1 on S, (1)

where a eld A is read A0 in vacuum, A1 in the medium (n1 = n0 is the normal on


S oriented toward the medium). Moreover it is convenient to decompose elds into their
incident and scattered parts according to

Ai Ainc
i = Ascat
i on S. (2)

Let M(r) and J(r) be the surface magnetic and electric densities of current dened by

M(r) = M1 (r) = n1 E1 (r) and J(r) = J1 (r) = n1 H1 (r),

the electric and magnetic scattered elds are given on S by the integral formulations

1
Escat (r) = n i M i (r) +  gki (r, r ) Mi (r )dr
i
2 S 
jk0 Gki (r, r ) Ji (r )dr
 S (3)
1
Hscat (r) = Ji (r) ni  gki (r, r ) Ji (r )dr
i
2 S 
i
jk0 Gki (r, r ) Mi (r )dr
i S

with the Green function



 e+jki |rr |
gki (r, r ) = ,
4|r r |

and the Green dyad


T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations 81

Figure 2. Surface discretization

1
Gki (r, r ) = [I + ]gki (r, r ).
ki2

We then insert (2) and (3) in the transmission conditions (1), and solve the problem for
M and J by employing a Galerkin variant of the method of moments [1].

1.1. Surface discritization

A surface discretization with bilinear quadrilateral nite elements is used (see Figure 2).
The surface currents are expressed as:


Nquad
M(r)  Mq (rq )
q=1

where Mq is the value corresponding to the q-th quadrilateral and Nquad the number of
quadrilateral on S. On each element, the surface currents are expressed as:

4

Mq (rq ) = q
Mm fm (rq (u, v))
m=1

q
where Mm are the current values corresponding to the m-th local edge of each element
and fm (rq ) are the basis functions described as

1
f1 (rq (u, v)) = (1 + u)aqu (u, v)
JSq
1
f2 (rq (u, v)) = q (1 u)aqu (u, v)
JS
1
f3 (r (u, v)) = q (1 + v)aqv (u, v)
q
JS
1
f4 (r (u, v)) = q (1 v)aqv (u, v),
q
JS

au (u, v) and av (u, v) are units vectors expressed as


82 T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations

r(u, v) r(u, v)
au (u, v) = , av (u, v) = ,
u v

and where the surface Jacobian JS (u, v) is given by

JS = |au av |.

1.2. Solving the linear system

A Galerkin variant of the method of moments, using bilinear basis functions, is employed
for discretizing the surface integral equations, leading to the linear system, with matrix
Z and primary eld Minc (r) and Jinc (r).

M bM
Z = (4)
J bJ

Accurate 3D simulations of eddy currents may lead to a matrix Z of huge size (higher
than the capacity of a standard PC). Consequently, solving the linear system using a
direct matrix solver is prohibitively expensive in computing time and memory resources.
This difculty is circumvented with the use of an iterative solver (typically GMRES) to
nd the surface currents M(r) and J(r) by minimizing
 
 
 M bM 
 Z J 
 J b 

with the use of the Fast Multipole Method (FMM) [2], which accelerates the compu-
tation of products of the matrix Z with the surface currents [M(r) , J(r)] and which
avoid the explicit evaluation of the matrix Z. The response of an eddy current probe is
nally expressed with the Auld reciprocity theorem, which involves the electric current
M(r) and the magnetic current J(r). The FMM has been generally used for problems
with dielectric objects without attenuation. In this work, we are dealing with low fre-
quency electromagnetic problem and conductive regions, which are typical conditions
under which ECT are applied, for the application of FMM.

2. Fast Multipole Method

2.1. FMM algorithm

The FMM is based on the subdivision of the region of space containing S into cubic
clusters. The double integrals arising from the Galerkin discretization are then divided
into near interactions (involving pairs of adjacent clusters) and far interactions (involving
pairs of non-adjacent clusters). The former are computed using classic integral equation
methods, while the latter are computed using a multipole expansion of gk1 (r, r ) and
Gk1 (r, r ). On the basis of preliminary numerical tests, it has been found that the most
suitable FMM formulation in the medium (k1 ) for ECT application is the decomposition
in diagonal form given by
T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations 83

Figure 3. Fast Multipole principle



ejk1 R jk1 
g(r, r ) = = ejk1 sOr T L (s, O O)ejk1 sr O ds, (5)
4R 16 2 |s|=1

where R = |r r |, O and O are the centers of two non-adjacent clusters and the

transfer function T L (s, O O) is dened in terms of the Legendre polynomials Pl and
the spherical Hankel functions of the rst kind hl by
 (1)
T L (s, O O) = 
(2l + 1)j l hl (k1 |O O|)Pl (s O  O) (6)
0lL

2.1.1. Far clusters


The interactions between two clusters are schematically depicted in Figure 3. Starting
from the source cluster centered at O , the information on each sources element r
are concentrated at O by the operator F (r O ). Then informations dened at O are
transferred to the center O using the transfer function T L (s, O O). Finally, F (Or) dis-
patches informations from O to each element r in the observation cluster centered at O.

2.1.2. Near clusters


As the FMM is not valid for adjacent clusters, computation of the interactions in this
case need to be done explicitly. We need to pay attention of the singular behaviour of
Greens function involved in the treatment of near interactions. The Green function has
a singularity for r = r . But this divergence disappears while integrating the Green
function on the surface of the object, which is properly managed with adapted methods.
The Green function is rst split into a singular part and a regular part like
 
 1 jk1 k12 jk13 2 k14 3
gk1 (r, r ) = + + + R R R + .
4R
  4 2!4 3!4 4!4 (7)
singular part
  
regular part
84 T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations

Table 1. Gain of computation time of FMMs LF/HF compared to the explicit calculation for different values
of the wavenumber used in NDT applications
k = 525+525j k = 1664+1664j k = 3441+3441j k = 5256+5256j
FMM LF 50 x faster 8 x faster - -
FMM HF 50 x faster 20 x faster 10 x faster 4 x faster

The regular part is integrated using classical methods and integrations involving the sin-
gular part are performed using Duffy transformations [3].

2.2. Most suitable multipole expansion for NDT

Different multipole expansions exist, called High Frequency (HF) or diagonal form, and
Low Frequency (LF). As low frequency occurs in NDT applications, one might think
that the LF expansion is the most suitable multipole expansion. However, FMM has been
generally used for problems with a real wavenumber (dielectric object without attenu-
ation) whereas problems involve a complex wavenumber in NDT applications, due to
the conductivity of the object. Preliminary numerical tests have been done to evaluate
computing time with explicit calculation and FMMs HF/LF in NDT.
Comparison of computation time is shown for two far clusters with 100 elements
in each one. So 10,000 interactions have to be computed in explicit calculation and 201
interactions for FMM. Even if less interactions is needed with FMM, the time spent to
calculate the integration in Equation 5 has to be evaluated to compare the methods.
One can remark in Table 1 that the gain of computation time of FMM LF compared
to the explicit calculation is only signicant for low value of wavenumber. On the other
hand, the gain of computation time of FMM HF is signicant for almost all values of
wavenumber involved in NDT applications. So even if in NDT applications, we are deal-
ing with low frequency electromagnetic problem, the most suitable multipole expansion
for NDT is the expansion called HF (in the medium). It is possible to explain theoretically
this result by analysing the wavenumber expression with and without attenuation.
wavenumber expression with attenuation:

|k 2 | = | 2 0 (0 + j/)| 0

wavenumber expression without attenuation:

k 2 = 2 0  0

For example, in NDT applications using low frequency, for = 0.7 MS/m and f =
100kHz, we have

|k 2 | = 5.5 105 .

But in comparison, without attenuation, we need f = 35GHz to get the same value of
|k 2 |. Thats why HF expansion is the most suitable multipole expansion for NDT (low
frequency electromagnetic problem but with conductive regions). A rst study has been
done to dene the conditions under which the calculation FMM works properly in the
NDT applications. For example, gure 4 shows the accuracy of the FMM HF expansion
as a function of the truncation level L and the integer Q (such that the integration over
the unit sphere appearing in (5) is effected using (Q + 1)(2Q + 1) quadrature points),
T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations 85

Figure 4. Accuracy of the FMM HF expansion based on L and Q for k1 = 1662 + 1662jm1 and centers
of far clusters distance of 2a with a = 102 m

for k1 = 1662 + 1662jm1 and centers of far clusters distance of 2a with a = 102 m.
The recommended truncation level is

Lopt = 0.55(real(L) + imag(L))



with L = 3k1 a + 7.5 log10 ( 3k1 a + ), with Q then chosen as

Qopt = 1.5Lopt .

3. Preliminary Validation Tests

Preliminary tests have been done on planar geometries (see Figure 5) and curved ge-
ometries (see Figure 6 depicting the computed distribution of the current density J on a
cylindrical sample). The response of an eddy current probe in the presence of a complex
object affected or not by a aw, leads to the computation of the variation of impedance
Z of the probe in the presence of the object:

Z = Z0 + Z,

where Z0 is the impedance of the probe in free space (i.e. in the absence of the object).

Figure 5. Conguration for the validation tests on planar geometries


86 T. Lim et al. / Fast Multipole Method for 3D Electromagnetic Boundary Integral Equations

Figure 6. Computed distribution of current density J on a cylindrical sample

For canonical geometries, for example a plate, it is possible to calculate this variation
of impedance with the Dodd and Deeds approach [4] or by using the Auld reciprocity
theorem [5] once the linear system (4) is solved for J and M:

I 2 Z = J E0 M H0 ds. (8)
S

The equivalence of the two approaches has been numericaly checked by expressing
surface currents J and M and primary eld E0 and H0 on the surface of the object with
Dodd and Deeds calculation. Therefore, Equation 8 can be used since the linear system
(Equation 4) has been resolved.

Conclusion and perspectives

In this paper, we report a modelling study aiming at simulating the response of an eddy
current probe on complex, homogeneous and non-magnetic 3D geometry affected by a
aw. The Galerkin variant of the method of moments is employed for the surface integral
equations with bilinear basis functions to solve the linear system (Eq. 4) by using an
iterative solver together with the Fast Multipole Method. Preliminary validation tests
have been done on planar geometries. Numerical validations on canonical geometries
are in progress by comparison with CIVAs results. Ongoing work concerns also the
extension of the approach in order to take into account a contrast of conductivity inside
the defect.

References

[1] B. C. Usner, K. Sertel, M. A. Carr and J. L. Volakis, Generalized Volume-Surface Integral Equation for
Modeling Inhomogeneities With High Contrast Composite Structures, IEEE Trans. Antennas Propag.,
Vol. 54, No. 1, Jan. 2006
[2] G. Sylvand, La mthode multiple rapide en lectromagntisme: Performances, Paralllisation, Appli-
cations, PhD Student, Ecole Nationale des Ponts et Chausses, Juin 2002.
[3] J. DEla, L. Battaglia and M. A.Storti, Full Nulmerical Quadrature in Galerkin Boundary element
methods, CIMEC, INTEC, CONICET, 2008.
[4] C.V Dodd, W.E. Deeds and J.W. Luquire, Integral Solutions to some Eddy Current Problems, Interna-
tional Journal of Nondestructive Testing, Vol.1, pp. 29-90, 1969.
[5] B.A. Auld, F.G. Muennemann and M. Riaziat, Quantitative Modelling of Flaw Responses in Eddy Cur-
rent Testing, Nondestructive Testing, Vol. 7, 1984.
Electromagnetic Nondestructive Evaluation (XIV) 87
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-87

Computation of the magnetostatic eld in


nonlinear media via the integral equation
formalism:
Application to the characterization of
Magnetic Flux Leakage NDT system
Emna Amira FNAIECH a Denis PRMEL a,1 , Claude MARCHAND b and
Bernard BISIAUX c
a
CEA, LIST, Laboratoire Simulation et Modlisation, PC 120, F-91191, Gif sur Yvette
Cdex, France
b
Laboratoire de Gnie Electrique de Paris (LGEP), CNRS-Suplec (UMR 8507),
Gif-sur-Yvette, France
c
VRA - Vallourec Research Aulnoye, Aulnoye-Aymeries, France

Abstract. This paper deals with the development of a fast numerical semi-
analytical model dedicated to the simulation of the non-destructive testing of fer-
romagnetic tubes by using the magnetic ux leakage method(MFL). Taking into
account the characteristic of the ferromagnetic material, the B-H curve, the numer-
ical model is based on the generalized boundary element method which implies
the evaluation of two magnetic ctitious scalar densities. Integral equations are dis-
cretized via the Galerkins method and the particularity of this paper lies in the
implementation of high order basis functions for the interpolation of the scalar un-
knowns. Simulated data provided by the numerical model are compared to FE data
in order to show the validity of the numerical approach.
Keywords. NDT by magnetic ux leakage, Magnetostatic regime, Nonlinear
media, Galerkins method, High order basis functions.

Introduction

The detection of aws in steel pipes by using the magnetic ux leakage technique con-
sists in magnetizing the pipe and due to the presence of a defect, a fraction of the ux
lines leaks from the pipes wall in the air inside or outside the pipe. This leakage ux
is detected by a magnetic sensor or an induction coil. In the Vallourec group, a world
leader manufacturer of steel pipes, most of NDT units use MFL for testing ferromag-
netic pipes. In order to improve the performances of aw detection, specically for inner
aws, the CEA LIST and the Vallourec Research Aulnoye (VRA) are aiming to develop
1 Corresponding Author: Denis Prmel, CEA LIST; E-mail: denis.premel@cea.fr
88 E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media

a fast 3D numerical model for the simulation of MFL systems. In a rst step, the frame-
work addresses an experimental system specically dedicated to the detection of longi-
tudinal aws. A simplied 2D geometry derived from the second magnetic MFL bench-
mark problem proposed by the World Federation of NDE centers [1] has been previously
studied [2] considering a linear regime. This means that at low magnetization, the ferro-
magnetic material behaves like a linear material. In this case, the semi-analytical model
based on the boundary element method (BEM) has been implemented and numerically
validated. This preliminary work could give satisfaction for linear model but we truly
aim to improve the capabilities of detection for different kinds of defects. For the detec-
tion of external defect, a low level of magnetization is usually sufcient but in order to
detect an inner defect or buried defects, we have observed that it is preferable to increase
signicantly the level of magnetization [2]. In this case, the main difculty in the compu-
tation of the magnetostatic eld within the pipe of ferromagnetic material lies in the non-
linearity of the B-H curve characterizing the material constituting the magnetizing cir-
cuit and the pipe. Each material can also be characterized by its own specic B-H curve.
The framework of this present work consists in extending the rst semi-analytical model
for computing the magnetostatic eld in the pipe considering the non-linear regime. For
solving a non-linear magnetostatic problem, the generalized boundary element method
(BEMG) is widely used since the statement of the problem is determined by two equiv-
alent scalar potentials: the surface charge density and the volume charge density [3,4,5].
This numerical approach has the capability to compute the response of an arbitrary de-
fect in the pipe. Nevertheless, this paper is focused on the numerical validation of the
semi-analytical model in the non-linear regime and without any aw.

1. Integral Equation Formalism

Let us consider some nite domain occupying a nite volume V , bounded by a surface
S and made of a ferromagnetic material. The non-linear problem is considered: the B-H
curve associated to the material is introduced in the model. r (r) stands for the rela-
tive permeability. Some encircling coils generate a source magnetic eld Hs which can
be calculated by using Biot-Savart law [6]. The total magnetic eld is splitted into the
sum of the source eld and the unknown induced eld produced by ferromagnetic re-
gions: H(r) = Hs (r) + Hi (r). The total eld is related to the magnetization vector
M(r) = (r 1) H(r). The resolution of the modeling problem amounts to determine
the induced eld Hi (r) produced by ctitious magnetic sources. The estimation of the
surface and volume magnetic charges is performed by using the extended boundary el-
ement method implying boundary (surface in 3D) elements and surface (volume in 3D)
elements. This approach is briey described in the following section.
The development, based on Maxwells equations and satisfying the condition
B = 0, leads to a system of two equations combining a surface integral equation and
a volume integral equation [4]. The forward problem consists in evaluating two scalar
functions: the surface charge density denoted by (r) = M(r) n(r) and the volume
charge density denoted by (r) = M(r). When solving the non-linear case, the
surface charge density does not vary signicantly compared to the linear case, so the
convergence of the iterative process cannot be reached. To overcome this problem, it is
necessary to solve the forward problem in two steps. The rst one consists in assuming
E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media 89

that the magnetic permeability is innite in such a way that the volume charges density
disappears. In this case, the surface charge density denoted by must be computed as
follows:

1
(r) + n G(r, r ) (r ) ds = n(r) Hs (r) (1)
2 S

with the normal vector n(r) pointing outwards from the volume V , G(r, r ) denotes the
Greens function in the free space and n G is dened by n G(r, r ) = n(r)G(r, r ).
Then, in the non-linear case, we consider that the total surface charge density results
from the addition of (r) and a correction term such as:

(r) = (r) + (r) (2)

A new system of two coupled integral equations leads to the computation of the differ-
ence surface charge and the volume charges density :

(r)
+ n G(r, r ) (r ) ds +
2(r) S

(r) 1
n G(r, r )(r ) d = (r) (3)
V 2(r)

(r) + ln(r (r)) G(r, r )((r ) + (r )) ds +
S

ln(r (r)) G(r, r ) (r ) d = lnr (r) Hs (r) (4)
V

r (r)1
is dened by (r) = r (r)+1 . Then, the total induced magnetic eld is given by:
 
Hi (r) = G(r, r ) (r ) ds + G(r, r ) (r ) d (5)
S V

The two coupled equations (4) and the observation equation (5) need to be discretized in
order to solve numerically the forward problem.

2. 2D Implementation and discretization

2.1. Geometric discretization

Before starting to solve the 3D problem, we chose to study the 2D model. The rst step
involves approximating the geometry by an appropriate mesh which leads to the dis-
cretization of the integral equations. Each contour of the piece is divided into curvilinear
elements and each surface is approximated by a number of quadrilateral elements (with
8 or 9 nodes). So, the coordinates of any point on each curvilinear segment (quadrilateral
element) are dened from the coordinates of the nodes using the shape functions:
90 E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media

n
 n

x= xi Ni (), y= yi Ni () for curvilinear elements (6)
i=1 i=1
nq nq
 
x= xi Ni (, ), y= yi Ni (, ) for quadrilateral elements (7)
i=1 i=1

where Ni (), Nqi () are the shape functions usually used in Finite Element Methods. n
and nq are respectively the number of nodes associated to each curvilinear element and
to each quadrilateral element.

2.2. The choice of basis functions

The unknowns of the problem, the surface charge density and the volume charge density
translated into the 2D problem, are interpolated by using a set of basis functions. Two
interpolation techniques are implemented: the rst one uses isoparametric elements and
the second one is based on higher-order basis functions.

2.2.1. Isoparametric elements


The discretization of integral equations by using isoparametric elements involves using
similar functions for the approximation of the geometry and for the interpolation of the
unknowns. The two unknown scalar functions are expressed as:
n nq
 
(r) = i Ni (), (r) = i Ni (, ) (8)
i=1 i=1

For this description, the number of unknowns i (i ) corresponds to the number of nodes
dening the contour (the surface).

2.2.2. Higher-order basis functions


The second interpolation technique is based on the use of basis functions of high order.
The two unknowns of the problem are expressed in series of Legendre polynomials of
higher order whith the following development:
M M q Nq
  
(r) = m Cm Pm (), (r) = mn Cm Pm ()Cn Pn () (9)
m=0 m=0 n=0

where M , (Mq and Nq ) are respectively the order of Legendre polynomials used for the
description of the density on an elementary curvilinear element and for the description
of the density on one elementary quadrilateral element. Pm (), Pn () are respectively
the Legendre polynomials and the Legendre polynomials modied [7], Cm and Cn are
some scaling factors. The total number of unknowns (the number of coefcients m ) is
equal to (M + 1) Ne where Ne is the number of curvilinear elements describing the
contour of the ferromagnetic region. The number of unknowns (the number of coef-
cients mn ) is equal to (Mq + 1) (Nq + 1) Neq where Neq is the number of elements
dening the surface of the ferromagnetic domain.
These two techniques of interpolation are compared in the following in order to test
their efciency and we expect to reduce the computation time and the memory space
with high order basis functions.
E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media 91

2.3. The matrix system

To avoid the difculty to dene the surface charge density at any point of the mesh, we
have used the Galerkin method. The matrix system is nally obtained by:
    
A A b
= (10)
A A g

with


fi (r) 
A = fi (r) + fj (r)n G(r, r )dk de (11)
e 2(r) k

 

A = fi (r) fqj (r)n G(r, r )dSk de (12)
e Sk

 
A = fqi (r) fj (r)lnr (r) G(r, r )dk dSe (13)
Se k



A = fqi (r) fqi (r) + fqj (r)lnr (r) G(r, r )dSk dSe (14)
Se Sk

  
(r) 1
b= fi (r) (r)de (15)
e 2(r)



g= fqi (r) lnr (r) Hs (r) + (r )lnr (r) G(r, r )dk dSe
Se k
(16)
where f and fq denote respectively the basis functions associated to contours and the ba-
sis functions associated to surfaces, e and k denote respectively the indices of the obser-
vation element and the source element. If we use isoparametric elements, the shape basis
functions fi and fj are associated to the nodes indexed respectively i and j. If we use
higher order basis functions, the indexes i and j stand for the order of Legendre polyno-
mial series. Due to the fact that the magnetic relative permeability depends on the local
intensity of the magnetic eld, this matrix system is solved iteratively. At each iteration,
the magnitude of the magnetic eld is evaluated in each quadrilateral element consider-
ing a number of observation points. Then, the gradient of the logarithm of the magnetic
relative permeability is deduced numerically by using spline interpolation functions.
92 E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media

3. Numerical validation

3.1. System

The implemented numerical model can be rstly validated by considering a simplied


geometry without any geometric corner. Let us consider a ferromagnetic pipe which is
magnetized by two coils driven by a constant current. The material of the tube being
ferromagnetic, the B-H curve of the ST37 steel, represented in Figure 1(b), has been in-
troduced into the iterative process. A 2D view of this simplied geometry is displayed in
Figure 1(a). The geometry is assumed to be innite along the transverse axis perpendic-
ular to the plane of the gure. Two kinds of simulated experiments are conducted. For

(a) System studied (b) Tube ST37

Figure 1. Conguration and ferromagnetic material used

the rst one, we aim to validate the numerical results provided by the semi-analytical
model and other simulated ones obtained by using a nite element software. The second
numerical experiment aims at highlighting the interest of the use of higher order basis
functions rather than the use of shape basis functions of second order.

3.2. A comparison between IEM - FEM data

The mesh of the tube contains a total number of elements equal to Neq = 200. The
inner radius of the tube is Rin = 48 mm and the outer radius of the tube is Rout = 88
mm. The current density driving the two coils is xed at Js = 1.1 105 A/m2 . This level
of magnetization allows one to go beyond the linear part of the B-H curve. Keep in
mind that in this numerical experiment, the unknown scalar functions are interpolated by
using isoparametric elements. The radial component and the tangential component of the
magnetic eld are computed on a circumferential line, outside the tube with a distance
of 2 mm from the outer surface. The angle of each computation point varies from 45 to
135 . All simulated data are compared in Figure 2(a) and 2(b).
Figure 3(a) displays a cartography of the intensity of the magnetic ux density inside
the tube. Figure 3(b) extracts some specic values on a radial line at = 60 from
R = 47.5 mm to R = 87.5 mm.
The good agreement between simulated data shows the numerical validity of the
semi-analytical proposed model. In the following, we look for the improvement of the
accuracy of the numerical model by using higher order basis functions.
E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media 93

(a) Hr (b) H

Figure 2. The radial and the tangential components of the magnetic eld provided by FEM and IEM.

(a) A cartography of |B| (b) An extracted radial line at = 60

Figure 3. Magnitude of the magnetic ux density inside the tube.

3.3. The interest of higher-order basis functions

In any geometry which contains a geometric corner, we expect to observe very huge
variations of the eld at the vicinity of the corner [8]. In such a situation, it is required
to rene the mesh around the corner but in this case, the number of unknowns can grow
up strongly. This seems to be very harmful for the computational time and the mem-
ory space. In order to give a better approximation of theses variations, we preferably
choose to use higher order basis functions, keeping in mind that the geometry remains
well approximated with a coarse mesh. In order to show the interest in the increasing of
order, a coarse mesh with Neq = 75 is applied for both interpolations techniques (shape
functions of second order and high order functions). Some reference numerical results
are nevertheless obtained with a ne mesh with shape functions of second order. Fig-
ure 4(a) and Figure 4(b) show the radial and the tangential components of the magnetic
eld along a radial line inside the tube and located at = 60 .
These numerical results show really the interest of high order basis function. Indeed,
this interpolation method provides better capabilities for representing the local variations
of the eld without the need of a rene mesh.
94 E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media

(a) Hr (b) H

Figure 4. The radial and the tangential components of the magnetic eld: A comparison between numerical
results obtained by shape functions of second order and by higher order basis functions.

4. Conclusions and perspectives

This paper is focused on the development of a 2D numerical model for solving a spe-
cic non-linear magnetostatic problem dedicated to NDT MFL systems. First numerical
results show a good agreement between IE data provided by the semi-analytical model
and FE data. Two semi analytical numerical models based on two different techniques
of interpolation have been implemented. Thus, the advantage of using high order basis
functions is clearly illustrated. It seems to be necessary to build up a ne mesh for ap-
proximating with a good accuracy the complex geometry and to ensure a good evalua-
tion of the unknown scalar densities, implicitly the intensity of the magnetic eld, it is
interesting to obtain a better description of the unknowns with high order basis functions.
Thanks to this numerical approach, any B-H curve can be introduced since we solve the
non linear problem by an iterative procedure. In order to pursue this work, specic nu-
merical problems due to numerical singularities remain to overcome. Moreover, we hope
to give a better approximation of more complex geometries without using a ne mesh
but by using Non-Uniform Rational Basis Splines NURBS. In close future work, we
aim to simulate the second magnetic MFL benchmark problem, a specic NDT system
proposed by the World Federation of NDE centers [1].

References

[1] Z. Zeng, Y. Tian and S. Udpa, "Finite Element Modeling of the World Federations Second MFL Bench-
mark Problem", Review of Progress in Quantitative Nondestructive Evaluation, 23(700), (2004), 1553
1559.
[2] EA. Fnaiech, D. Prmel, C. Marchand and B. Bisiaux, "A fast numerical model for predicting magnetic
ux leakage signals due to defects in a steel pipe by using an integral equation approach", Proceedings
of the 8th International Symposium on Electric and Magnetic Fields (EMF2009)
[3] B. Krstajic, Z. Andelic, S. Milojkovic, S. Babic, and S. Salon, "Nonlinear 3D magnetostatic eld calcu-
lation by the integral equation method with surface and volume magnetic charges". IEEE Transactions
on Magnetics, 28(2), (1992),10881091.
[4] W. Haa, A. Buchau, F. Groh and W.M. Rucker, "Efcient integral equation method for the solution of
3D Magnetostatic problems". IEEE Transactions on Magnetics, 41, (2005), 14081411.
[5] J. McWhirter, J. Oravec, and R. Haack, "Computation of magnetostatic elds in three-dimensions based
on fredholm integral equations", IEEE Transactions on Magnetics, 18, (1982), 373378.
E.A. Fnaiech et al. / Computation of the Magnetostatic Field in Nonlinear Media 95

[6] I.R. Ciric, "Surface source models and formulas for the magnetic eld of polygonal cross section con-
ductors". IEEE Transactions on Magnetics, 24(6), (1988), 31323134.
[7] E. Jorgensen, J L. Volakis, P. Meincke, "Higher order hierarchical legendre basis functions for electro-
magnetic modeling", IEEE Transactions on Antennas and Propagation, 52(11),(2004), 29852995.
[8] F. Groh, W. Haa, A. Buchau and W.M. Rucker, "Field strength compuation at edges in nonlinear mag-
netisatics", The international Journal for Computation and Mathematics in Electrical and Electronic
Engineering, 23(3),(2004),662669.
96 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-96

Efcient computation of eddy current


crack signals
Theodoros THEODOULIDIS 1
University of Western Macedonia, Department of Mechanical Engineering, Greece

Abstract. In simulations of eddy current inspections using the boundary element


method (BEM) is is often the case that the calculation of the system matrix car-
ries the largest computational cost. This matrix calculation typically makes use of a
Greens function expressed as a Fourier integral. The present study has aimed to re-
duce the computational cost by adopting an alternative approach for the rapid com-
putation of the Greens function in the spatial domain. This is possible by employ-
ing Sommerfeld integral expressions and by approximating them in terms of dis-
crete complex images. Theoretical results are compared to both benchmark exper-
imental data for a semi-elliptical crack and to numerical data from a nite element
commercial package. Agreement is excellent over a wide frequency range.
Keywords. Eddy current testing, crack modelling, Sommerfeld Integrals, Discrete
Complex Image Method, Boundary Element Method

Introduction

In the application of the Volume Integral Method (VIM) for the simulation of eddy cur-
rent defect inspections we employ the following integral equation for the calculation of
the electric eld in the presence of a void defect embedded in a multilayered conductive
half-space [1]

E(r) = E(0) (r) + i0 G(r|r ) p(r )dr (1)
V0

where E(r) is the electric eld with the defect present, E(0) (r) is the electric eld with
the defect absent (incident eld), G(r|r ) is the electric-electric dyadic Greens function
dened as the eld response to a unit point source (electric dipole), p(r ) is the unknown
electric dipole density, V0 is the defect volume and primed and non-primed position
vectors refer to source and eld points respectively.
When, however, the defect has the form of a crack, the very narrow crack opening
can make the use of (1) somewhat problematic because a dense discretization is required
across the gap in order to get accurate results. The situation is even worse in the limit of
a vanishing gap which describes the ideal crack case, when the crack is represented by
just a surface impenetrable to the eddy current ow. In such a case of a narrow or ideal
1 Corresponding Author: Theodoros Theodoulidis, University of Western Macedonia, Department of

Mechanical Engineering, Bakola & Sialvera, 50100 Kozani, Greece; E-mail: theodoul@uowm.gr.
T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals 97

crack, instead of using (1), we can replace it with a simpler integral equation if we take
into account the very fact that at the crack surface the normal component of the eddy
current density vanishes. After imposing this condition we obtain [2]

Jn(0) (r ) = k 2 Gnn (r |r )p(r )dS  (2)
S0

where S0 is the crack surface and subscript n denotes the direction normal to the crack
surface. Since the discretization of the crack is reduced from a volume to a surface, the
BEM is used instead of VIM. The model based on the solution of (2) has been recently
developed to achieve high computational efciency. In [3], where a literature review of
research work regarding the model is presented, further improvements are suggested on
the original numerical scheme in [2] regarding computation of the incident eld, Greens
function computation and singularity treatment. In [4], an alternative approach is sug-
gested for the Greens functions computation and for the simulation of cracks with shapes
other than rectangular. During all these developments, the computation of the Greens
function, i.e. the Greens function matrix ll time was identied as the bottleneck in the
calculations. In the present work, we study the subject from a different perspective and
compute these functions in the spatial domain. This is possible by employing the Dis-
crete Complex Image Method (DCIM) which in turn is based on the efcient evaluation
of the Sommerfeld integrals. The method employs the Generalized Pencil of Function
(GPOF) method to approximate the Sommerfeld integral expression for the kernel with
a series of complex exponential functions and hence cast them in closed form [5], [6].
This is similar to the method presented in [4] but with a basic difference that will be
explained later. Results are provided for a semi-elliptical crack and they are compared to
benchmark experimental results. The same conguration is also modelled with a numeri-
cal method (a 3D-FEM commercial package) in order to identify the differences between
the two methods.

1. Greens function

Consider Figure 1 which shows a cylindrical coil above a narrow slot in a homogeneous
conductive and non-magnetic half-space. The coil is excited by a time harmonic current
varying as the real part of I exp(it). The crack surface is discretized in small rectan-
gular elements and a constant dipole density is assumed over each element. The constant
dipole density is equivalent to representing the dipole distribution with pulse functions.
The integral equation is then transformed to a linear system of equations by imposing the
vanishing current constraint at the middle points of the elements, which is the denition
of point matching and is equivalent to the use of delta functions as testing functions. The
solution of the integral equation (2) yields the dipole density on the crack surface, which
acts as a secondary eddy current source. This source determines the coil impedance due
to the crack. By using a reciprocity relation, we can calculate directly the crack signal
(impedance change due to the crack) through [2]

1
Z = 2 Jx(0) p dV0 (3)
I
V0
98 T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals

Coil

y
x

Crack x

Figure 1. Coil above a narrow crack in a conductive half-space.

where is the half-space conductivity. In this paper we are interested in the calculation
of the term Gxx used in (2) by nding the x-component of the electric eld produced
in the half-space by an elementary electric dipole oriented also in the x-direction. The
Greens function consists of two parts.

Gxx = G(s) (r)


xx + Gxx (4)
(s) (r)
where Gxx is the part that arises in the unbounded space and Gxx is the part arising from
the air-conductor interface. The calculation of (4) has to be done for each combination
of source and eld cells that comprise the grid of the discretized crack. Hence, in each
case Gxx accounts for the interaction between the cells. The source part is given directly
in the spatial domain by


2 eikR
G(s) = k 2
+ (5)
xx
x2 4R

with R = (x x )2 + (y y  )2 + (z z  )2 , while the expression for the x-
component of the reected electric eld produced by an electric dipole, in Cartesian
coordinates, is as follows [3], [7]
  

1 e(z+z ) 2 2  
G(r) = k v u ei[u(xx )+v(yy )] dudv (6)
2 2
xx
4 2 2 +


with = u2 + v 2 and = 2 k 2 . Both expressions (5) and (6) have to be inte-
grated over the source volume, i.e. over x , y  , z  to determine the matrix used in nding
a numerical solution. The advantage of (6) is that it allows analytical integration over the
source cell dimensions but it has the disadvantage that a double improper integral has to
be computed for each matrix element. The method proposed in [3] was to approximate
(6) by Fourier series.
T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals 99

1.1. Discrete Complex Image Method

Nevertheless, an extensive search in the literature of Greens function computations


shows that it can be computed rapidly and accurately in the spatial domain. In this case
we use a cylindrical coordinate system (, , z) centered on the eld point and express
(r)
Gxx in the following form [3], [7], which uses Sommerfeld type integrals.


k2 1 
G(r) = e(z+z ) J0 () d
xx
4 +
0



cos2 2 2 
+ e(z+z ) J0 () d
4
0



sin2 cos2 2 2 
+ e(z+z ) J1 () d (7)
4
0

where Jn denotes the n order Bessel function. Eq.(7) has to be integrated over the source
cell volume. In particular, the integration over the z-dimension of the source cell, i.e.
over z  , can be done analytically. Assuming a cell of z extent we obtain
  

k2 1 z 
G(r)
xx dz

= sinh e(z+z ) J0 () d
2 + 2
0

 

cos2 2 2 z 
+ sinh e(z+z ) J0 () d (8)
2 2
0

 

sin2 cos2 2 2 z 
) 1
+ sinh e(z+z J1 () d
2 2
0

Now we seek an analytical calculation of the above integrals. For that we utilize the
following identities

z eikR
e J0 () d = (9)
R
0


1 z eikR eikz
e J1 () d = (10)
ik
0

with z 0. By approximating the bracketed terms with exponentials of the complex


variable


1 z
sinh = c1,n es1,n (11)
+ 2 n
100 T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals



2 2 z
sinh = c2,n es2,n (12)
2 n

we end up with a closed form expression for the reected part of the Greens function

k2  eikR1,n cos2  eikR2,n
G(r) 
xx dz c1,n + c2,n
2 n R1,n 2 n
R2,n

sin2 cos2  eikR2,n eikz2,n


+ c2,n (13)
2 n
ik
!
where zj,n = z + z  + sj,n and Rj,n = 2 + zj,n 2 . The reason for calling the method

Discrete Complex Image is now obvious. The Greens function consists of images lo-
cated at a complex distance Rj,n (compare to (5)). Eq.(13) has to be integrated over the
x and y-directions. This can be done numerically or even analytically if rst the term
eikR2,n /Rn is Taylor expanded as an m-th degree polynomial in Rn . Such an approach
is followed in [8] and leads to considerable time saving. The error introduced with such
an expansion is given by a closed-form expression depending on the approximating poly-
nomial degree and the ratio of the cell dimensions and thus it is controlled by the number
of terms in the Taylor series. In our case, having set x = y = 0 (eld point
 at the center
of the cylindrical coordinate system), we make the substitution = x 2 + y  2 and
= arctan (y  /x ) and integrate numerically (13) in order to obtain Gxx for (2).
The approximation in (11)-(12) is possible with the GPOF algorithm [5]. Note that
the algorithm has to be performed in each equation just once since it does not depend
on the specic distance between the source and the eld cells. In [4], the GPOF algo-
rithm was used for approximating the Sommerfeld integral kernel with exponentials of
the real variable. Such an approach results in an expression that can be evaluated ana-
lytically over the volume of the source cell but it requires multiple GPOFs for each z + z 
combinations. This is the basic difference of the present approach and the one in [4].

2. Results

Here we present results for a benchmark problem described in [9]. A coil is moved
above and along a semi-elliptical crack and the impedance change due to the crack (real
and imaginary part) is recorded with respect to the coil position. The semi-elliptical
notch has length 22.1 mm, max depth 8.61 mm and width 0.33 mm. The host material
is Aluminum with conductivity 22.5 MS/m. The cylindrical coil has inner radius 2.51
mm, outer radius 7.38 mm, height 4.99 mm, 4000 wire-turns and lift-off 0.491mm. The
signals are presented for 16 frequencies that cover a wide range from the very low to the
very high. In order to present them all in a common diagram, they are all normalized to
the isolated coil inductive reactance X0 = L0 with a theoretical value of L0 =100.47
mH and an experimental one L0 =100.45 mH. The term low and high frequency is
dened by the crack depth to skin depth ratio. For example at 50 kHz this ratio is 18.15
and the case is dened as a thin-skin limit case or a high frequency case.
Fig.2 compares the normalized complex impedance plane crack signals obtained
from the BEM model to experimental data and also to numerical results from a 3D-FEM
T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals 101

0.035

0.03 f [Hz]
250
356
0.025 507
721
1027
1462
0.02
2081
2963
4218
0
X/X

0.015 6006
8550
12172
0.01 17329
24670
35121
50000
0.005

0.005
0.02 0.015 0.01 0.005 0 0.005 0.01
R/X 0

Figure 2. Impedance plane crack signals for the 16 frequencies shown on the right of the diagram. Comparison
of theoretical BEM (lines) to experimental (dots) and FEM (circles) results.

calculation [10]. Excellent agreement is observed at all frequencies. Computation time


was about 11 hours for the FEM (31 coil positions and 16 frequencies from 250 Hz to
50 kHz) and 20 seconds for the BEM with a quad-core I7 processor, 16 GB memory
computer. Computation times for the different parts of the calculation were provided
in [4]. The computation time for FEM increases linearly with the coil positions in contrast
to BEM where the time does not depend on this factor. The BEM computation time on
the other hand depends on the frequency since higher frequencies require a denser grid
for the discretized crack. For the FEM the same grid was used for every frequency with
good results.

3. Conclusions

We have presented a method for rapidly computing the Greens function matrix, which is
the slowest part in the computation of crack signals via method of moments. Impedance
predictions compare very well to experimental measurements as well as to FEM calcu-
102 T. Theodoulidis / Efcient Computation of Eddy Current Crack Signals

lations. There are many ways to further extend the scope of the model. For example, it
can incorporate various coil types or coils under conditions of tilt [11], since their effect
enters through only the incident eld.

Acknowledgements

The work is funded by CEA LIST in the framework of the collaborative project CIVA-
MONT 2012.

References

[1] S. Paillard, G. Pichenot., M. Lambert and H. Voillaume, Eddy current modelling for inspection of riveted
structures in aeronautics, Electromagnetic Nondestructive Evaluation (X), S. Takahashi and H. Kikuchi
(Eds.), IOS Press, (2007), 2532.
[2] J.R. Bowler, Eddy current interaction with an ideal crack. I. The forward problem, J. Appl. Phys. 75
(1994), 81288137.
[3] T. Theodoulidis, N. Poulakis and A. Dragogias, Rapid computation of eddy current signals from narrow
cracks, NDT&E Int. 43 (2010), 1319.
[4] T. Theodoulidis, Developments in efciently modelling eddy current testing of narrow cracks, NDT&E
Int. 43 (2010), 591598.
[5] Xing Lingling, Rapid calculation of eddy current eld Greens function using the matrix pencil method,
NDT&E Int. 42 (2009), 8591.
[6] M.I. Aksun, A robust approach for the derivation of closed-form Greens functions, IEEE Trans. Mi-
crowave Theory and Techniques 44 (1996), 651658.
[7] A. Banos, A dipole radiation in the presence of a conducting half space, Pergamon Press, 1966.
[8] L. Alatan, M.I. Aksun, K. Mahadevan and M.T. Birand, Analytical evaluation of the MoM matrix ele-
ments, IEEE Trans. Microwave Theory and Techniques 44 (1996), 519525.
[9] D.J. Harrison, L.D. Jones and S.K. Burke, Benchmark problems for defect size and shape determination
in eddy-current nondestructive evaluation, J. NDE 15 (1996), 2134.
[10] Comsol 3.5a Users Guide, 2009.
[11] T.P. Theodoulidis and E.E. Kriezis, Eddy current canonical problems (with applications to nondestruc-
tive evaluation, TechScience Press, 2006.
Electromagnetic Nondestructive Evaluation (XIV) 103
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-103

Experimental Validation of a Fast


Non-Iterative Imaging Algorithm for
Eddy Current Tomography
Salvatore VENTREa, Flavio CALVANOb, Guglielmo RUBINACCIb
and Antonello TAMBURRINOa,1
a
Ass. EURATOM/ENEA/CREATE, DAEIMI, Universit di Cassino, Italy
b
Ass. EURATOM/ENEA/CREATE, DIEL, Universit di Napoli Federico II, Italy

Abstract. In this work for the first time we validate experimentally a fast (real
time) non iterative imaging method for eddy current tomography of conductive
materials. The presented method is based on the monotonicity principle applied in
the low frequency limit where the skin depth is great with respect to the relevant
dimension of the problem under test. The experimental tests prove the
effectiveness of the monotonicity based imaging method. Moreover, we prove that
the inversion can be carried out without resorting the direct problem solver.

Keywords. Eddy currents, non iterative imaging algorithms, non destructive


evaluation, monotonicity, ECT signal inversions.

Introduction

In eddy-current testing a prescribed time-varying current flowing in the excitation


coils induces eddy currents in the conductive specimen under inspection. The electrical
resistivity of the specimen affects the induced eddy currents density and, consequently,
the voltage at the receptive coils. The measured data consist of measurements of the
impedance matrix (self and mutual impedances) between the coils of the array.
In recent years, numerical methods for solving the forward or direct problem, i.e.,

Figure 1. The planar surface to be investigated (specimen) together with a probe made by an array of seven
coils and a rectangular defect.

1
Corresponding Author: Ass. EURATOM/ENEA/CREATE, DAEIMI, Universit di Cassino, v. G. Di Biasio,
43 Cassino 03043 Italy; e mail: tamburrino@unicas.it.
104 S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm

the problem of predicting the measured quantities for a given electrical resistivity and
source current, have been extensively studied. The main applications of the direct
problem are: (i) the computer aided prediction of the performances of existing probes,
(ii) the design of ad-hoc optimized probes and (iii) the solution of the inverse problem
through model based inversion methods. Numerical formulations based on finite
element method have been studied and developed by the authors for both volumetric
[1] and zero thickness inclusions [2].
A key role in eddy current tomography is the reconstruction of the resistivity
profile of the specimen under test, starting from the measurement data, through the
solution of an inverse problem. The identification of the conductivity distribution
inside a material is seriously affected by the inherently ill-posed and nonlinear nature
of the eddy currents inverse problem (see [3]-[5] for mathematical issues). First and
foremost, the ill-posedness translates into low sensitivity measurements, i.e. different
resistivity distributions produce similar measurements and, in addition, the nonlinearity
requires sophisticated inversion algorithms.
The imaging methods can be divided in iterative and non-iterative (direct).
Usually the iterative algorithms are based on the Gauss-Newton method. Among the
iterative methods it is worth mentioning quadratic and higher order approximation of
the forward operator, deterministic and stochastic algorithms, pre-calculated database,
statistical methods, Total Variation regularization, Level Set regularization, etc. (see
[6]-[24] and references therein).
As main drawbacks, the iterative methods require the solution of the direct
problem for several assigned tentative shapes of the inclusion and this can be very
expensive in terms of computational cost. Moreover, the convergence cannot be
guaranteed.
On the other hand non-iterative methods have attracted a lot of interest because
they provide a test for evaluating if a point of the domain (or a subregion) is part or not
of the anomaly, regardless other points (or subregions). The test is very cheap from the
computational viewpoint because no iterations are necessary.
In this paper we present the first experimental validation of the fast and non-
iterative method based on the monotonicity property of the real part of the measured
impedance matrix. This method was initially developed by the authors for elliptic
problems such as electrical resistance tomography [25], and later extended to parabolic
problems such as the eddy current testing in both low frequency (large skin depth) [26]
and high frequency (skin depth negligible with respect to the relevant dimension) [27-
28].
Numerical simulations of the direct problem have been exploited to check
numerically if a given probe was able to guarantee appropriate reconstructions. The
monotonicity imaging method has been applied to identify the resistivity profile of
several benchmarks.
The paper is organized as follows. The monotonicity is proven for a proper
unknown-data mapping in Sections 1, the inversion method is presented in Section 2,
an experimental validation of the monotonicity imaging method is presented in Section
3 and finally in Section 4 the conclusions are drawn.

1. Monotonicity principle

In this section we briefly describe the monotonicity principle for the real part of
the measured impedance matrix (self and mutual impedances between pairs of coils of
S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm 105

the array shown in figure 1). In the low frequency limit, the impedance matrix admits
the following expansion:
D = R0 + jL0 + PD + j PD + O( )
Zcoil 2 (2) 3 (3) 4
(1)

that is valid as long as the frequency is enough low so that the 4th and higher order
terms in (1) are negligible. We mention that all matrices appearing in (1) are real and
symmetric.
Matrix PD(2 ) , which contributes to the real part of the impedance matrix, plays a key
role because of its monotonicity property [26]. In particular it is possible to show that if
the resistivity of the considered material increases pointwise, then the ohmic power
dissipated in the conductor decreases and consequently PD(2 ) decreases in the sense that
will be clarified in the following. For two-phase materials the monotonicity can be
stated as [26]:

V1 V2 in D PV(2 ) PV(22 ) (2)


1

where Vk D represents the domain occupied by the k-th anomaly (resistivity a )


that is hosted in the background material having resistivity b < a . In (2) PV(12 ) PV(22 )
means that matrix PV(12 ) PV(22 ) is positive semi-definite, i.e. all its eigenvalues are non-
negative.

2. Monotonicity imaging method

Equation (2) forms the basis for the method to solve the inverse problem (see [26]).
Let V be the (unknown) subset of D where the resistivity is a (the resistivity in D\V is
b). The inverse problem consists in retrieving V.
Let us consider a generic (and known) test domain k. From (2) it follows that:

P(2) PD(2 ) false k D , (3)


k

and, thus, by checking (3) for different test anomalies k it is possible to


estimate/retrieve V. It is worth noting that the algorithm can be applied to an arbitrary
number of anomalies with unknown shape and topology.

Figure 2. The conductive domain D subdivided in elementary regions together with an anomaly V (grey
pixels) and a test region k (black pixel).
106 S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm

(2 )
In order to test (3) we need to compute the eigenvalues of P(2k) PD for checking if
this symmetric matrix is positive semi-definite or not. However, only a noisy version
~
PD(2 ) = PD(2 ) + E , PD(2 ) being the noiseless matrix and E the noise matrix, can be
measured and it can be processed as described in [26]. Similarly, the test matrices P(2k)
may be affected by either measurement errors (if measured) or numerical errors (if
~ ~ (2 )
numerically computed). Therefore, the eigenvalues of P(2k) PD can be different
(2 )
from the eigenvalues of P(2k) PD and this may alter in an unpredictable way the sign
of the eigenvalues with smaller magnitude, i.e. this may alter the reconstruction.
~ ~ (2 )
In order to deal with this situation, we quantify how much the matrix P(2k) PD is
positive semi-definite through the so-called sign index sk defined as:


j =1
k, j

s k = M
(4)
k , j
j =1

~ ~ (2 )
where k , j is the j-th eigenvalues of the matrix P(2k) PD and M is the number of the
available eigenvalues. We notice that sk is closer to 1 when almost all relevant
~ ~ (2 )
eigenvalues are non-negative (sk=1 when P(2k) PD is rigorously positive semi-
definite).
In addition, from perturbation theory [29], it follows that the eigenvalues that may
change their sign due to the presence of noise, are only those having a magnitude
smaller than the Euclidean norm of the noise. When the magnitude of the noise is
known or estimated, these eigenvalues can be removed from the summations in (4).
Finally, we highlight that the dimension (number of rows and columns) of the
~ ~ (2 )
matrix P(2k) PD is small because it is equal to the number of coils of the array that,
usually, is made by few elements (few tens). Therefore, a very low computational cost
(2 )
(computation of the eigenvalues of P(2k) PD ) is required to compute (4) for
performing test (3).

3. Experimental results

The experimental tests we describe in the following consist of an array made by two
coils mounted in a fully automated imaging system composed by a scanning system, a
LCR meter and a PC controlling the whole acquisition and inversion process (figure 3).
The specimen under test is a printed circuit boards (PCB, thickness 1.6mm) presenting
copper islands , having different size and shapes, to be imaged. The copper islands
(thickness 35mm) are union of 5mm5mm elementary domains. These elementary
domains are taken as the k during the imaging process.
The sensor, that has been designed through extensive numerical simulations, is
made by two pancake coils. The first coil (internal diameter=5mm, external
diameter=10.5mm, height=6.5mm, number of turns=700) contains internally the
S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm 107

second coil (internal diameter=1mm, external diameter=4mm, height=3mm, number of


turns=180). Simulated inversions have provided that this array gives no reconstruction
errors for a noise level not greater than 50 m. This threshold has been achieved
during the experimental test that, therefore, provide error-free reconstructions.
The excitation frequency is 30kHz, chosen to satisfy the condition of 4th and higher
order terms in (1) negligible. At same time, this frequency guarantees an appropriate
magnitude of the measured signal and a noise level smaller than the 50 m threshold.
The monotonicity algorithm processes the experimental data providing the
reconstruction, in 0.1ms per pixel.
It is worth noting that the second order moment for a test domain ( P(2k) ) has been
experimentally measured on a 5mm5mm copper island. Moreover, thanks to the
translational invariance of the problem, it has been sufficient to perform the
measurement of P(2k) only for a single test domain located on the top side and for a
~
single test domain located on the bottom side. The related measurements are P(2k), bottom
~
and P(2k), top .

3.1. First case

The first test is a printed circuit board where the copper forms the letters IP
(Inverse Problem) as shown in figure 4 (left).
The inversion, that is shown in Figure 4 (right) is error free. This is because the
experimental noise level is smaller than the threshold (50 m) found by the
aforementioned numerical simulations involving the inversion of synthetic data.

Figure 3. The two coils composing the array (left). The smaller coil is inserted into the bigger one. Block
diagram of the measurement system (right).
108 S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm

Figure 4. The specimen under test (left) and its reconstruction (right). The white pixels represent the
conductive pixels. The pixel dimensions are 5mm5mm.

3.2. Second case

The second test is a double-face printed circuit board (see Figure 5). It consists of a
double sided PCB (copper, thickness 35m) as shown in figure 5 (top). The
measurements are taken from only the top side of the PCB.
In this case the imaging algorithm requires the test domains on both sides on the PCB.
The test domains on a side of the PCB provide the reconstruction for the related side.
For instance, the test domains on the top side provide an error free reconstruction of the
top side (figure 5, left column). On the other hand, the test domain in the bottom layer
provide as reconstruction the union of the pixels in the top and bottom sides (figure 5,
right column). This unexpected result can be easily explained by considering that it
~ ~
results P(2k), bottom P(2k), top and, therefore, it is trivial to prove that the test domains
related to the bottom side provide a reconstruction that is the union of the pixels from
the top and the bottom sides.

4. Conclusions

A fast inversion method for inverting eddy-current testing data has been
experimentally validated for the first time. Numerical simulation of the whole imaging
process (not reported here for the sake of brevity) provide noise level threshold of
about 50 m such that for noise level below such threshold the reconstruction is error-
free. The time required to form the image in a single pixel is about 0.1ms. The imaging
algorithm is fully non-linear and, therefore, can treat arbitrary shapes and topologies.
Moreover, the measured data have been processed by means of pre-measured data and
without resorting to the numerical solution of the direct problem that, as well known, is
time-consuming.
S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm 109

Acknowledgements

This work was supported in part by the Association EURATOM/ENEA/CREATE,


Italy.

(a) (b)

(c) (d)

Figure 5. The specimen under test: the top side (a) directly under the probe, and the bottom layer (b).
Reconstruction obtained by exploiting the test domain located on the top side (c) and reconstruction obtained
by using the test domain located on the bottom side (d). For this latter inset the white pixels represent the
pixels of the bottom side whereas the grey pixels represent the pixels of the top side.

References

[1] M. Morozov, G. Rubinacci, A. Tamburrino, and S. Ventre, Numerical Models with Experimental
Validation of Volumetric Insulating Defects in Eddy Current Testing, IEEE Trans.on Magnetics, vol.
42, pp. 1568 1576, 2006.
[2] R. Albanese, G. Rubinacci, and F. Villone, An integral computational model for defect simulation and
detection via eddy currents, J. Comp. Phys., vol.152, pp. 736 755, 1999.
[3] D. Colton and L. Paivarinta, The uniqueness of a solution to an inverse scattering problem for
electromagnetic waves, Arch. Ration. Mech. Anal., vol. 119, pp. 59 70, 1992.
[4] V. Isakov, Uniqueness and stability in multidimensional inverse problems, Inverse Problems, pp. 579
621, 1993.
[5] M. Yamamoto, A mathematical aspect of inverse problems for non stationary Maxwells equations, Int.
J. Appl. Electromagn. Mech., pp. 77 98, 1997.
110 S. Ventre et al. / Experimental Validation of a Fast Non-Iterative Imaging Algorithm

[6] W. Q. Yang (Guest Editor), Special feature: Electrical Tomography Industrial and Medical,
Measurement, Science and Technology, vol. 21, 2001.
[7] R. A. Williams and Beck M S (eds), Process Tomography: Principles, Techniques and Applications,
Oxford: Butterworth-Heinemann, 1995.
[8] M. Soleimani and A. Tamburrino, Shape reconstruction in magnetic induction tomography using
multifrequency data, Int. Jour. of Information and System Sciences. vol. 2, pp. 343 353, 2006.
[9] W. Yin and A. J. Peyton, A planar EMT system for the detection of faults on thin metallic plates,
Measurement Science and Technology vol 17, pp. 2130 2135, 2006.
[10] A. Pirani M. Ricci R. Specogna A. Tamburrino and F. Trevisan, Multi frequency identification of
defects in conducting media, Inverse Problems, vol. 24, 2008.
[11] A. Tamburrino, S. Ventre and G. Rubinacci, Electrical resistance tomography: complementarity and
quadratic models, Inverse Problem, vol. 16, pp. 1585 1618, 2000.
[12] R. Pierri and A. Tamburrino, On the local minima problem in conductivity imaging via a quadratic
approach, Inverse Problems, vol. 13, pp. 1547 68, 1997.
[13] G. Rubinacci, A. Tamburrino and S. Ventre, A differential formulation based on a perturbative
approach to solve the ECT inverse problem, Comput. Methods Appl. Mech. Eng. vol. 169 pp. 407 24,
1999.
[14] D. Dos Reis, M. Lambert and D. Lesselier, Non destructive evaluation of 3 D voids in a metal plate,
Inverse Problems, vol. 18, pp. 1857 71, 2002.
[15] D. Premel and A. Baussard, Eddy current evaluation of three dimensional flaws in flat conductive
materials using a Bayesian approach, Inverse Problems, vol. 18, pp. 1873 89, 2002.
[16] J. R. Bowler, Thin skin eddy current inversion for the determination of crack shapes, Inverse Problems,
vol. 18, pp. 1891 905, 2002.
[17] B. Luong and F. Santosa, Quantitative imaging of corrosion in plates by eddy current methods, SIAM J.
Appl. Math., vol. 58, pp. 1509 31, 1998.
[18] V. Monebhurrun, B. Duchne and D. Lesselier, Three dimensional inversion of eddy current data for
nondestructive evaluation of steam generator tubes, Inverse Problems, vol. 14, pp. 707 24, 1998.
[19] Z. Badics, J. Pv, H. Komatsu, S. Kojima, Y. Matsumoto and K. Aoki, Fast flaw reconstruction from
3D eddy current data, IEEE Trans. Magn., vol. 34, pp. 2823 2828, 1998.
[20] Li Yue, L. Udpa and S.S. Udpa, Three dimensional defect reconstruction from eddy current NDE
signals using a genetic local search algorithm, IEEE Trans. Magn., vol. 40, pp. 410 417, 2004.
[21] W. Yin, S. J. Dickinson and A. J. Peyton, Imaging the continuous conductivity profile within layered
metal structures using inductance spectroscopy, IEEE Sensors J., vol 5, pp. 161 166, 2005.
[22] S. M. Nair and J. H. Rose, Reconstruction of three dimensional conductivity variations from eddy
current (electromagnetic induction) data, Inverse Problems, vol. 6 pp. 1007 30, 1990.
[23] O. Dorn and D. Lesselier, Level set methods for inverse scattering some recent developments, Inverse
Problems, vol. 25, pp. 67 131.
[24] Q. N. Jin, On the iteratively regularized Gauss Newton method for solving nonlinear ill posed
problems, Math. Comput., vol. 69, pp. 1603 23, 2000.
[25] A. Tamburrino and G. Rubinacci, A new non iterative inversion method for electrical impedance
tomography, Inverse Problems, pp. 1809 1829, 2002.
[26] A. Tamburrino and G. Rubinacci, Fast Methods for Quantitative Eddy Current Tomography of
Conductive Materials, IEEE Trans. Magn., vol. 42, pp. 2017 2028, 2006.
[27] G. Rubinacci, A. Tamburrino, S. Ventre, Eddy current imaging of surface breaking defects by using
monotonicity based methods, ACES Journal, vol.23, no. 1, pp. 46 52, 2008.
[28] A. Tamburrino, S. Ventre, G. Rubinacci, Recent developments of a Monotonicity Imaging Method for
Magnetic Induction Tomography, Inverse Problems, vol. 26, no. 7, pp. 074016, 2010.
[29] Golub G H and Van Loan C F 1996 Matrix Computations (Johns Hopkins Studies in Mathematical)
Sciences 3rd Edition, Johns Hopkins University Press, Baltimore, USA.
Electromagnetic Nondestructive Evaluation (XIV) 111
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-111

Numerical Evaluation of
Microwave Testing for Pipe Thinning

Yasutomo SAKAI, Noritaka YUSA, Hidetoshi HASIZUME
Department of Quantum Science and Energy Engineering, Tohoku University

Abstract. This study evaluates the applicability of using microwaves for the
inspection of pipe wall thinning by using numerical simulations and experiments.
The configuration considered in this study is an axisymmetric one consisting of a
straight brass tube and coaxial cables. After the agreement between
three-dimensional numerical simulations and experimental result is confirmed,
experiments to locate wall thinning are carried out. Clear signals due to the wall
thinning are obtained. The time of flight of microwaves agrees with the
theoretical prediction, which demonstrates the possibility of locating wall
thinning from microwave signals.

Key words. Electromagnetic wave, piping detection, wall thinning, FEM

1. Introduction
Rapid inspection for piping systems used in chemical or power plants is one of the
most important issues from the viewpoint of reducing maintenance costs. Conventional
methods such as ultrasonic and eddy current testing provide local information, which
results in a long inspection period for large piping systems. Due to this fact, another
nondestructive testing method for evaluating a large area at once is highly demanded.
Currently, to solve these problems, guided waves are mainly studied[1][2][3].
Guided waves propagate in pipe walls longitudinally, which enables inspection of a
broad area. However, there are some problems, such as the reflections at the
connections of pipes. Another approach is to utilize microwaves[4][5][6][7]. The basic
concept of this method is to regard the pipes as a waveguide and to detect defects from
transmission and reflection of microwaves propagating inside the pipes. Since
microwaves propagate through the free space surrounded by the metal pipes with little
attenuation, connections of pipe do not affect signals so significantly.
Recent studies have shown through numerical simulations that the presence of
wall thinning affects transmitted energy of microwave, and analyzing the signals in the
frequency domain enables one to evaluate the shape of wall thinning[8]. The studies,
however, include several problems from the viewpoint of practical applications. That is,
the model was quite simplified. In addition, the locations of wall thinning were
assumed to be known in advance and only the intensity of the spectrum of measured
signals were analyzed. Therefore, the purpose of this study is to evaluate the
applicability of microwave nondestructive inspection to the detection and evaluation of
wall thinning in more practical configurations. This study conducts three-dimensional
numerical simulations and experimental verifications. Obtained signals are analyzed in
time domain to correlate the time of flight of the signal with the location of wall
thinning.
112 Y. Sakai et al. / Numerical Evaluation of Microwave Testing for Pipe Thinning

2. Materials and Methods

The principle of the microwave testing is to evaluate wall thinning from the transmitted
and reflected energy of microwave as mentioned above. The time of flight (TOF) of
reflection, which is calculated from the group velocity of the microwaves, enables one
to locate discontinuities such as wall thinning.
Fig. 1 shows the configuration used in both the numerical simulation and the
experiment. It consists of a brass tube modeling the piping, and coaxial cables and
connectors to inject microwaves into the tube. The coaxial cables are extended into the
pipes with a length of 5 mm to suppress the microwave reflections occurring at the end
of the cable. Microwaves are generated and the reflected energies are measured at port
1. The microwaves propagate as TEM mode inside the coaxial cables, and are
transformed into TE and TM modes inside the tube.
This study considered two lengths of a pipe, L=120 and 1120 mm, which are
termed as the short and the long range model hereafter. The short range model is
evaluated both in numerical simulations and experiments, to confirm the validity of the
method as well as to evaluate the behavior of microwaves propagating inside the tube.
The long range model is only dealt with by the experiments due to limitations of
computational resources needed to consider the model. Table 1 shows the parameters of
the two models. The numerical simulations shortened the length of coaxial cables, Lc,
into 20 mm to decrease computational resources.
The experiment system is composed of semi-rigid cables (K-118 Anritsu),
microwave connectors (K-101, Anritsu) several brass pipes and brass plates. The brass
plates and the semi-rigid cable are soldered. A short brass tube with a larger diameter is
utilized to simulate wall thinnings. That is, actually the brass tube shown in Fig. 1
consists of three tubes connected with flanges. Microwaves are generated and
measured by a network analyzer (Agilent PNA, E8363B). Flexible coaxial cables are
connected from the network analyzer to semi-rigid cables. Data is obtained in
frequency domain and transformed to time domain using inverse Fourier
transformation.
The numerical simulations are conducted using commercial software COMSOL
Multiphysics and its RF Module. The governing equation adopted is

( )
rot r1 rotE k 02 r j E = 0 ,
0
(1)

where
k 0 = 0 0 2 .
2
(2)
Subscript 0 denotes physical quantities in vacuum. Tube wall was modeled as a
conductor with a zero thickness and impedance boundary condition,
0 r
j 0 r n H + E ( E n )n = 0 , (3)
0 r j
is imposed on the surface of the tube wall. In the analysis, tetrahedral quadratic vector
elements are employed. Table 2 summarizes parameters used in the simulations. It
should be noted that the numerical simulations are three-dimensional ones, whereas the
configurations are axisymmetric. The cut-off frequencies of TE11, TM01 and TM02
modes in the pipe are 9.25, 12.1, and 19.3 GHz, respectively.

Y. Sakai et al. / Numerical Evaluation of Microwave Testing for Pipe Thinning 113

3. Results
Fig. 2 shows comparison of the numerical and experimental results of the amplitude of
the reflected energy in the short range model. The figures show good agreement, which


Figure 1 Analysis and experimental system (unit mm)

Table 1 Details of models


short range model long range model
L[mm] 120 1120
LC[mm] 20(simulation),235(experiment) 235
LD[mm] 40 290,790

Table 2 Parameter of Numerical simulations


tube wall air teflon
[MS/m] 6 0 0
r [-] 1 1 1
r [-] 1 1 3

 

 

 
$PSOLWXGH> @

$PSOLWXGH> @

 

 

 

 
       
)UHTXHQF\>*+]@ )UHTXHQF\>*+]@

Figure 2. Comparison between simulations (left) and experiments (right)

validates both the simulations and the experiments. The small differences between the
two results are mainly caused by the flexible cables not considered in the numerical
simulations. Other numerical simulations confirmed that TM01 is dominant inside the
pipe, which are not shown in this paper.
Fig. 3(a) shows the results of numerical simulations in the short range model.
As reflections at the connection part are relatively large, the measured data did not
show significant difference. The signals without any defects are subtracted from those
114 Y. Sakai et al. / Numerical Evaluation of Microwave Testing for Pipe Thinning

with defects to extract and to evaluate the effect of the defects quantitatively. The
subtracted signals are presented in Fig. 3 (b), which shows the reflections caused by
wall thinning. Fig. 4 shows the results of experiments in the short range model. The

0.40 0.40
QV QV
0.30 0.30
0.20 0 20

amplitude[-]
amplitude[-]

0.10 0.10
0.00 0.00
-0.10 -0.10
-0.20 -0 20
-0.30 -0.30
-0.40 -0.40
0.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0
time[ns] time[ns]

(a) Raw data of pipes without wall thinning(right: non wall thinning, left: wall thinning depth 2mm)

0.10 0.10
0.54ns 0.61ns 0.54ns 0.61ns

0.05 0.05
amplitude[-]
amplitude[-]

0.00 0.00

-0.05 -0.05

-0.10 -0.10
0.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 3.0 4.0
time[ns] time[ns]
(b) differences(left: depth 2 mm, right: depth 1 mm)
Figure 3 Reflected microwaves in the short range model, obtained by the numerical simulations

0.20 0.20
2.34ns 2.41ns 2.41ns 2.34ns
0.15 0.15
0.10 0.10
amplitude[-]

amplitude[-]

0.05 0.05
0.00 0.00
-0.05 -0.05
-0.10 -0.10
-0.15 -0.15
-0.20 -0.20
0.0 1.0 2.0 3.0 4.0 5.0 0.0 1.0 2.0 3.0 4.0 5.0
time[ns] time[ns]

Figure 4 Reflected microwaves in the short range model, obtained by the experiments
(left depth 2 mm, right depth 1 mm)
Y. Sakai et al. / Numerical Evaluation of Microwave Testing for Pipe Thinning 115

Table 3 Calculated TOF in short range testing


numerical simulation experiment
Entrance of pipe 0.23 ns 2.03 ns
Near edge of wall thinning (25GHz) 0.54 ns 2.34 ns
Near edge of wall thinning (17GHz) 0.61 ns 2.41 ns

0.20 0.20
0.15 QV QV 0.15 QV QV

0.10 0.10
amplitude[-]

amplitude[-]
0.05 0.05
0.00 0.00
-0.05 -0.05
-0.10 -0.10
-0.15 -0.15
-0.20 -0.20
0 2 4 6 8 10 12 0 2 4 6 8 10 12
time[ns]

time[ns]
Figure 5 Reflection of wall thinning (LD=290mm, LD=790mm, both depth 2mm)

Table 4 Calculated TOF in long range testing


290 mm 790 mm
Entrance of pipe 2.03 ns 2.03 ns
Near edge of wall thinning (25GHz) 4.25 ns 8.08 ns
Near edge of wall thinning (17GHz) 4.78 ns 9.52 ns

signals shown in the figure are also the difference between the models with and without
defects. Table 3 shows the calculated TOF of reflections due to the three discontinuities.
front edge of wall thinning and the connection part. The TOF of the reflections are
evaluated by using the group velocity of two frequencies, 25 and 17 GHz, since the 25
GHz is the maximum frequency considered in the experiment and 17 GHz provides
maximum transmitted energy. It should be noted that since the length of coaxial cable
of experiment is much longer than that of numerical simulations, TOF of experiment is
also larger than that of numerical simulations. Fig. 3(b) and Fig. 4 confirm that the
initial rises of measured signals coincide with the TOF of reflections due to the near
edge of wall thinning shown in Table 3. These results show the possibility of detecting
wall thinning. Signals corresponding to the far edge of wall thinnings are unclear
because of the large reflection due to the end of pipe. The results obtained in the short
range model also imply that the method would have difficulty in detecting defects
located too close to microwave ports..
On the basis of the results above, experiments in the long range model were
carried out. The results of the experiments and calculated TOF are shown in Fig. 5 and
Table 4, respectively. It should be noted that the signals are not subtracted ones. The
strongest signals near 2 ns are the reflection at the connection part. The next signals
come from the near edge of wall thinning, and their TOF are between the two TOFs
calculated using 25 GHz and 17 GHz, whereas dispersion of the microwaves makes
signals unclear. Analyzing modes of electromagnetic waves propagating inside the pipe
more quantitatively will lead to more accurate localization of the wall thinning from
measured signals.
116 Y. Sakai et al. / Numerical Evaluation of Microwave Testing for Pipe Thinning

4. Conclusion
In this study, microwave testing for piping systems was evaluated both by
three-dimensional finite element simulations and experiments. Signals obtained in
frequency domain were converted into time domain to locate wall thinnings from their
time of flight. After the validity was confirmed in a short range model, practical
experiments using 1120mm long brass pipe were carried out. The results confirmed the
validity of the method.

Acknowledgement
This work was supported by the Tohoku University Global COE Program on World
Center of Education and Research for Trans-disciplinary Flow Dynamics

Reference
[1] D. N. Alleyne, M. J. S. Lowe, P. Gawley, The reflection of guided waves from circumferential notches in
pipes, Journal of Applied Mechanics 65 (1998), 635-641.
[2] P. D. Wilcox, A rapid signal processing technique to remove the effect of dispersion from guided wave
signals, Ultrasonics, Ferroelectrics, and Frequency Control 50 (2003), 419-427.
[3] Joon-Hyun Lee, Seung-Joon Lee, Application of laser-generated guided wave for evaluation of
corrosion in carbon steel pipe, NDT&E International 42 (2009), 222-227.
[4] Kavoos Abbasi , Study of Microwave Nondestructive Technique to Detect Crack and Predict Its
Location in Piping System, Tohoku University Doctoral Dissertation, (2007).
[5] Y. Ju, L. Liu, M. Ishikawa. Quantitative evaluation of wall thinning of metal pipes by microwaves.
Materials Science Forum 614 (2009), 111-116.
[6] K. Sugawara, Development of inspection method for the defect in pipes using microwave, Tohoku
University Master thesis, (2002), (in Japanese)
[7] T. Shibata, H Hashizume, S. Kitajima, K. Ogura, Experimental study on NDT method using
electromagnetic waves, Journal of Materials Processing Technology 161 (2005), 348-352.
[8] Y. Sakai, N. Yusa, S. Ito, H. Hashizume, Numerical analysis of microwave NDT applied to piping
inspection, Proceeding of The 13th Asia-pacific Conference on Non-destructive Testing, (2009), 230.
Electromagnetic Nondestructive Evaluation (XIV) 117
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-117

Parallelization of Crack Signal


Calculation Using CUDA
Imre KISS 1 , Jzsef PV and Szabolcs GYIMTHY
Budapest University of Technology and Economics,
Department of Broadband Infocommunications and Electromagnetic Theory

Abstract. A calculation method for the simulation of the Eddy Current Testing
(ECT) signal of multiple cracks is implemented on a general purpose video card.
NVIDIAs parallel computing solution, the Compute Unied Device Architecture
(CUDA) was used as a shell for the implementation on both hardware and software
sides. Possibilities inherent in this technology are shown through an example and a
detailed implementation focusing on data parallel execution is discussed. Different
realizations of the same algorithm are inspected and a comparison from the view of
efciency is given.

Keywords. CUDA, GPGPU, Multiple cracks, Surface integral crack model

Introduction

The mathematical model of several different types of defects is given in the form of inte-
gral equations. The numerical solutions of such integral equations might be obtained us-
ing the Method of Moments. The main computational task in such solutions is the evalu-
ation of the kernel of the integral equation at numerous locations (either spatial or spatial
frequency locations) inside the integration domain. Most part of such calculations in a
given location can be carried out independently of the calculations related to other loca-
tions. Consequently the kernel can be evaluated simultaneously in the desired locations.
As a result, parallel computation of this can be easily designed.
A number of approaches are known for accelerating an application by paralleliz-
ing it. Among others, one may use supercomputers for high performance computing, or
could build up a Beowulf cluster from ordinary machines. The application of a target
hardware for solving time consuming tasks is also a possibility, like for example the
project GRAPE (GRAvity piPE) for n-body computations [6]. It is easy to see that all of
these solutions have drawbacks, especially from the nancial point of view.
In this paper a parallel computing method is described aiming to give an extremely
fast evaluation of the signal of a thin crack using a CUDA capable Graphic Processing
Unit (GPU).

1 Corresponding Author: Imre Kiss, Budapest University of Technology and Economics, Goldmann Gyrgy

Tr 3, 1111 Budapest, Hungary; E mail: kiss@evt.bme.hu.


118 I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA

1. Compute Unied Device Architecture

Since graphics hardwares reached notable power, there always have been attempts to use
their computational capacity for high-performance computations. One of the main causes
preventing the spreading of programs utilizing graphics cards were the complexity how
these target devices could be programmed. GPUs were primarily designed for opera-
tion on graphical objects, hence long standing programming languages (e.g. OpenGL,
DirectX) were hard to be used for solving general, non-graphical related problems.
NVIDIA recently introduced its CUDA (Compute Unied Device Architecture)
technology [1] which issued in a general purpose graphics hardware. The new design
unied the work-ow from shader levels to a general purpose core (the term unied is
originated here), which serves the needs of both traditional graphical and general purpose
utilizations. Consequently, a hardware design consisting of hundreds of processor cores,
and capable to execute thousands of threads simultaneously have become available, that
can be programmed in a general purpose programming language (a slightly modied
version of the regular C language).

1.1. Achievements on parallel computation with CUDA

Efforts have already been made at different elds of engineering to utilize gain from the
application of GPU based parallel computation. Application elds dealing with large,
independent data structures are the primary targets of the utilization. For example in
molecular dynamics, the movement of numerous particles perfectly ts into the main
idea behind parallelization by treating each particle in a separate thread. In numerical
algebra, GEMM routines in the BLAS library give an example on parallel dense matrix
multiplication. Many other references could be enumerated from the elds of medical
imaging, uid dynamics, seismic imaging. A good summary on the application elds can
be found in [5].

2. Signal calculation using the multiple thin crack model

2.1. Integral equation model for the calculation of the signal of multiple thin cracks

Assume the arrangement shown in Figure 1 where C innitesimally thin cracks are lo-
cated close to each other in a conducting specimen. The surfaces and the surface normal
vectors of the cracks are Sc and nc (c = 1, 2, . . . , C), respectively.
ECT probe

0 , 0 n1
n3
n2
... nC

S1 S2 S3 ... SC
Surface Cracks
Vs : Specimen Surface Cracks

Figure 1. Eddy current probe above a conductor specimen containing innitesimally thin planar cracks.
I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA 119

The eddy current signal of several innitesimally thin planar cracks can be calculated
by using the extension of the integral equation model of the single thin crack [2]. Each
crack is represented by a surface current density pc (c = 1, 2, . . . , C) concentrating on
the surface Sc of the crack. The unknown pc functions can be obtained by the solution of
the following system of integral equations [3]:


C 
0 = Ecn
i
(rc ) j0 gkc (rc |r  )pk (r  ) dr 
k=1
k=c Sk

j0 lim gcc (r|r  )pc (r  ) dr  , rc Sc , c = 1, 2, . . . , C, (1)
r 
 rc
Sc

where Ecni
(rc ) is the so-called incident eld, that is the normal component of the elec-
tric eld due to the interaction of the probe and the specimen without the presence of
the cracks. gkc (r|r  ) = nc Ge (r|r  ) nk (k, c = 1, 2, . . . , C), where Ge (r|r  ) is
the electric-electric Greens function assuming the geometry of the given specimen. rc
denotes the vector approaching the location vector rc from both sides of the surface Sc .
Knowing the solution of (1) the ECT signal, that is the impedance variation of the
probe, Z, is calculated as,
C 
1 
Z = 2 i
Ecn (r)pc (r) dr, (2)
I c=1
Sc

where I is the magnitude of the exciting current.

2.2. Discretization of the integral equation

In this part we will show that the most time-consuming calculations during the solution
of the integral equation (1) can be done parallelly if the method of moments is used
for the discretization of (1). For the sake of simplicity assume that the only one crack
(C = 1) is located in a plate specimen. The surface of the plate is in the xy plane of the
Descartes coordinate system and the crack is in the x = 0 plane (n1 = x, where x is the
unit vector of the x coordinate direction).
Assume that the surface current density is approximated by a properly chosen func-
tion series as,


M 
N
p1 (y, z) pmn fym (y)fzn (z) y, z S1 , (3)
m=1 n=1

and the integral equation is tested by the following testing functions:

tkl (y, z) (k = 1, 2, . . . , K; l = 1, 2, . . . , L) y, z S1 . (4)

For example, pulse approximating (fym (y), fzn (z)), and testing (tkl (y, z)) functions are
used in [2], linear approximating and testing functions are applied in [3], while sinusoidal
type global approximating and pulse testing functions are used in [4].
120 I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA

As a result, the discretized integral equation will be:


M 
N 
pmn [j0
m=1 n=1 S1


lim g11 (x, y, z|x  = 0, y  , z  )fym (y  )fzn (z  ) dy  dz  tkl (y, z) dydz =
xx
S1
 k = 1, 2, . . . , K,
= i
E1n (x = 0, y, z)tkl (y, z) dydz; (5)
l = 1, 2, . . . , L.
S1

Consequently the following linear equation system is to be solved:

A p = b, (6)

where
$ %
A (k1)L+l,(m1)N +n is the normal component of the electromagnetic eld due
to a surface current distribution described by fym (y)fzn (z) and tested by the test-
ing functions tkl (y, z),
(b)(k1)L+l is the incident eld tested by the testing function tkl (y, z),
$ %
p (m1)N +n = pmn is the unknown coefcients of the surface current distribu-
tion.
One can see that the most time-consuming task, the calculation of the elements of A
can be calculated independently from each other. This is the reason why the paralleliza-
tion of the outlined solution can really decrease the calculation time needed to obtain the
crack signal. Note also that the same conclusion can be derived if multiple cracks are
considered as the solution of (1).

3. CUDA parallel programming model

A CUDA program mainly consists of two parts. There is a so called host program,
roughly speaking the function of which is to initialize the GPU for the computation while
itself runs on the CPU only. The program running on the device (GPU) is usually called
the kernel. As the kernel can operate only on data stored in the on-board memory of the
device, all data must be transferred to the device prior to the computation. Similarly, the
host program cannot directly access the on-board device memory, hence the result of the
computations must be transferred back the computers main memory.
These transfers are often act as the bottleneck of the whole computation. If the
amount of data to be transferred is large, then due to the relatively slow PCI-E bus sys-
tem, even with an excessively fast GPU the overall wall-time would increase, and the
parallel execution would be pointless.
I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA 121

3.1. Memory handling

Since the available memory on todays average GPUs is limited in 1-2 GB, one may think
that this memory constraint is just preventing any practical application, as time consume
often arises from the need of manipulating large data sets. To overcome on this problem,
one must not use of long standing algorithms: algorithms that do not need the whole
data set to be stored in the memory all at once are ideally suited for this task. An other
approach could be the decomposition of the original problem to smaller ones that t into
the memory constraint.
The overall speed-up strongly depends on both data transfer and computation times.
A commonly applied technique is to overlap data transfer with computation, which can
be approached from CPU and/or GPU side. Namely, if a so-called asynchronous data
transfer is applied, the control is immediately returned to the initializer of the transfer.
Since latest graphic cards are capable to overlap kernel execution and memory transfers,
partitioning the GPU memory to non-overlapping parts, the kernel may operate on one
part, while the other(s) could be simultaneously loaded. As a result, one could keep both
the CPU and the GPU busy, signicantly decreasing the inertia of the data transfer.

3.2. Hardware design

Output of the compiler regarding the GPU code is in a pseudo-assembly language (PTX),
which is generally understandable by any CUDA capable GPU. Compilation of this code
to the actual hardware instruction set is taking place at execution time by the device
driver, therefore portability and backward-compatibility is also ensured.
CUDA applies a Single Instruction Multiple Thread (SIMT) execution model, which
is a variant on the Single Instruction Multiple Data (SIMD) architecture. At any given
clock cycle, each streaming processor executes the same instruction, but operates on
different data without the need of packing and unpacking the original data like in the
case of SIMD (SSE2). Due to the SIMD architecture, thread divergence in a given core
causes threads being in an inactive path to be executed, but not evaluated: all threads
walk through all paths, signicantly increasing the execution time. One may also note
that thread divergence is only critical if occurs on numerous cores. A detailed overview
on the topic can be found in [7].

3.3. Design of parallelization

The original program computing the ECT signal of multiple cracks (based on the for-
mulations in Section 2.2) was written for a single core CPU. Most of the calculations
are carried out analytically in the spatial frequency domain where the singularity of the
kernel can be easily handled. Elements of matrix A in (6) are obtained in kn km points
of the spatial frequency domain followed by an analytical integration in the z direction.
Spatial boundaries are determined adaptively, the kn km sampling points are distributed
uniformly within the studied frequency domain. The main concept of the implementation
can be found in [3] and [4], while this paper will focus on the design of parallelization.
Generally, the rst step during the design of parallelization is to obtain the execu-
tion times for the different parts of the program. In the given case, the assembly of the
matrix elements is performed in subroutine 4, which requires about 45% of the total ex-
122 I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA

ecution time. The rest of the work is carried out in 3 other subroutines, where certain
coefcients of the basis and testing functions related to different spatial positions are pre-
calculated. These 3 subroutines requires about 15% of the total computation time each.
Certain initialization process and post-processing is neglected for clarity. Fig. 2.a shows
the schematic owchart of the original program.
After careful inspection of the program code, subroutine 4 is found the only part
which can be efciently parallelized, since each matrix element can be computed in-
dependently of the others. In the case of subroutine 1-3, numerous conditional state-
ments are evaluated, resulting in divergent execution paths, which prevents the efcient
execution on the SIMD architecture. Regretting this fact, these subroutines were also
re-formulated for data parallel execution, but obtained wall-times conrmed theoretical
considerations.

3.4. Concepts of parallelization

Subroutine 4 will be parallelized and outsourced to the GPU while the rest of the pro-
gram will be executed on the CPU. The execution path for the original single core CPU
program is shown in Fig. 2.a: serial execution of subroutine 1-4 is carried out organized
into a loop, in each cycle of which a spatial frequency component is computed. Note that
the dots correspond to the neglected initialization procedure.

     

   



   
    

      




  




 

Figure 2. Execution paths of the CPU and the two different CUDA models.

To replace subroutine 4, a CUDA kernel was carefully designed, performing the par-
allel computation over the different spatial frequency sampling points. Since the avail-
able memory on the GPU is less than that is required to store all the precomputed coef-
cients, the problem must be partitioned to smaller parts. Each partition contains a subset
of the coefcients belonging to several spatial frequency points, the kernel will act on.
Based on our investigations, the GPU performs the computation over the sampling
points signicantly less time (with about 2 magnitudes) than the computation of the
coefcients takes on the host side. As a weird consequence, it become more important
how the data is transferred to the GPU, as it is comparable to CPU computation time.
Two CUDA programs were designed and implemented to investigate the effect of
the different transfer methods. These differ only in the way how the coefcients are
I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA 123

transferred to the GPU, while executing the same kernel. Both programs compute the
required coefcients on the host side, enclosed in a loop containing subroutine 1-3. In
Fig. 2.b, a semi-serial program model is shown. After a certain number of coefcients are
computed in the host loop, a single asynchronous memory transfer occurs, followed by
the invocation of the kernel. This enables the host loop (CPU) to keep computing during
the time consuming transfers. The size of the subset to be transferred is equal to the total
available memory on the GPU.
The second program is shown in Fig. 2.c, and is based on the idea that memory trans-
fer time can only be hidden if the CPU is busy with the computation of coefcients during
the transfers. In order to achieve such a case, precomputed coefcients are transferred in
smaller sets. The size of a subset to be transferred was determined to take insignicantly
small time, hence its amount is limited in several MBs. In the given example, a data-set
corresponding to km sampling points is chosen for such purpose. The kernel is invoked
on the transferred data when the memory of the GPU is full up after numerous transfers.

4. Results

The CPU and the two CUDA programs were executed on a HP XW8600 workstation
with a total of 32 GB memory, 2 Quad-Core Intel Xeon X5450 processors and an
NVIDIA GTX 480 video card.

20 GB
10 GB
5 GB
2 GB
1 GB
transferred data

a)
7 MB
3 MB
1 MB
transfer in each cycle
total amount of transfers

80
CUDA semi serial
Sub_4 run time [s]

60
CUDA overlapping
b) CPU
40

20

0
6
speed-up []

5
c) 4
3
2

1 200 400 600 800 1000 1200 1400 1600 1800 2000

km = kn = N

Figure 3. a) amount of data for the overlapping transfer method (note that values on the y axis are shown in
logarithmic scale); b) execution times of subroutine 4; c) measured relative speed up.
124 I. Kiss et al. / Parallelization of Crack Signal Calculation Using CUDA

Fig. 3.a shows the amount of data transferred in each cycle, as well as for the total
problem in the overlapping case for different number of sampling points (note that the
total amount of GPU memory is only 1.5 GB for the given hardware). Run-times with
respect to subroutine 4 are shown in Fig. 3.b. Based on the amount of data transferred to
the GPU (Fig. 3.a) and calculating with an average of 1 GB/s transfer rate, it can clearly
be seen that a signicant portion of time is spent on data transfer, rather than on (kernel)
computation.
The semi-serial CUDA computational model performed a bit worse than that of the
CPU. The second CUDA model, applying overlapping data transfer and CPU computa-
tion, performed signicantly better, preceding both the CPU and the semi-serial CUDA
model. It must be emphasized that the kernel function itself is the same in both program
models, only the applied data transfer technique varies. The results indicate that the ex-
ecution time of the kernel of the implemented algorithm is negligible compared to data
transfer time. The acceleration achieved by the overlapping CUDA model is shown in
Fig. 3.c.

5. Conclusion

In this paper the possibility of using CUDA for re-coding already existing single proces-
sor algorithms/programs was discussed. In the presented example by parallelizing only a
fraction (subroutine 4 constitutes only about 10%) of the whole source code, a remark-
able speed-up could be achieved.
Drawbacks of the required memory transfers between host and device memory are
shown together with some commonly applied techniques, that could be used to hide the
inertia of the transfer. Application of overlapping memory transfers with CPU computa-
tions shown remarkable performance gain compared to the semi-serial data processing
model. Although it would be possible to overlap the data transfers with kernel executions
as well, in the present case due to the negligible time of kernel execution, similar results
would be obtained. For problems having more intensive (time consuming) computations
in the kernel, a trade-off in the amount of transferred data must be found to achieve
optimal overall performance.

References

[1] Online available: www.nvidia.com/cuda


[2] J.R. Bowler, Eddy current interaction with an ideal crack I. The forward problem, J. Appl. Phys., 75 (11)
(1994), 8128 8137.
[3] J. Pavo, Reconstruction of Group of Cracks in Plate Specimens Using ECT Impedance Data, in Elec-
tromagnetic Nondestructive Evaluation (IV), S. S. Udpa, T. Takagi, J. Pavo and R. Albanese, Eds., IOS
Press, Amsterdam, 2000, 204 211.
[4] J. Pavo, D. Lesselier, Calculation of eddy current probe signal with global approximation, IEEE Trans-
action on Magnetics, 42 (4) (2006), 1419 1422.
[5] A Garland, M.; Le Grand, S.; Nickolls, J.; Anderson, J.; Hardwick, J.; Morton, S.; Phillips, E.; Zhang,
Yao & Volkov, V., Parallel Computing Experiences with CUDA, IEEE_M_MICR, 28 (2008), 13 27.
[6] Makino, J., & Taiji, M., Special Purpose Computers for Scientic Simulations The GRAPE systems
(Chichester: John Wiley and Sons), 1998.
[7] T. Halfhill, Parallel Processing with CUDA, Microprocessor Journal, (2008).
Electromagnetic Nondestructive Evaluation (XIV) 125
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-125


      

      

   
  !"#$%#&'%" (  ' 
) *+%#&'%"
'HSDUWPHQWRI(OHFWULFDODQG&RPSXWHU(QJLQHHULQJ)DFXOW\RI(OHFWULFDO


(QJLQHHULQJ:HVW3RPHUDQLDQ8QLYHUVLW\RI7HFKQRORJ\6LNRUVNLHJR 
6]F]HFLQ3RODQG

$EVWUDFW "  ,-     . , 


    
  / 0 1       . 
 .  ,
 .  
  , 2
  3 '   4     4 
  
,
  ,   1 , 5    ,   5 

    3    .    ,1
 - 5 ,    
,   .  3 "  ,,  )   55   ,  5  
5  5
      3 6  

 )
   5 ) -,  5   2 
). 7 3 3 8  8  
       $ 9
7 ,
     
  ,
 3 8
- 

- 1   :   1 
  3

.H\ZRUGV 6    . .
 7
      7

 ). 7   

7
  
 
-

,QWURGXFWLRQ

6   . .
  /6 0 4   )
-      - 5 
,  5 .      
; ;7 ,,
 7 ,
      -, 
5 )
  3 '   ,   

-  -     , .  5


5
 7 ,  7 2
;   .  <,
3 # 5   55 . 7
5  
 .
-
)  ,  4   
  7 )    
        7 , ,
5    ). = ,,    3

) 5 4  - 
 4  1  
2     5      -, 
5 , 5      )- 5  
  3 '   4 7

- 

 $ + 
  
- /$+0   

-  

 
).      1-  1  5 , 2
    
, 

-   ,
 =  ,, = 5 3  
-   5  5    ) 
 ,                 = )

 ; 3 # 
55  1    5 7  ,  
 
  5  ,,  ,
  1
  3 "  
)-  -  , 5   - ,  )     

(
9 ,  >   !
;);7  ,   5
 
 9,   7
 
- 5
 
 7 &  8  .  - 5  
-7 ';;  ?@7 @A ?(?
'2 2 7 8
7  32
;);B2 3 3,

126 M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain

   . 4        5  5   7 1 .    5 


  
    
-     <  5   - ,   - 
 
-  ,    5  1C    3    , 2

4  4    
,
 /

- 
4   0 1 )  
   ,  = 5       
  -   , 
 
 1 5  ,   ,  = 5  1
     3
D - ,1
   1  
2  4    ,   .   E(7
:F3
 
 .  7 )  .      .  7    ,
- 
 
  ,
5 4 1        5 
 
5  
- )   ,  ,  3   - 5 
 ).    
  

 
  
 5
  5     1    
,,    1    5 
   ,  3       
)  .       
    

 ,  7 3 3 $  2 5 7
 2  5 7      . 5  / 
 ) ; ,
 
- 
 5 5    
0   
- 


    
  / 0 E?GHF3  ,     )  .        5
7
-   7    

 
 
    
 .  -3 " ,,  
.  
, 2
 55 7 
  -    
- )  
,    1 , 3
           
 .  . -,  5 1
;    
 ). 3
"  ,,  ) )  
  
    55   ,  5  
5  5
       ,
    7
  ,
 7 3 3 8  8        $ 9

3          

)       2 

- /'I0
,
2  ). 7 )   
 , .    .  ) . - 5  5 
E(G@F3 '  ,      7 
- $  2 5   )

1
  3  1    
 . 1   ,
-   ,
- 
 


-3 '  5  )

) 5 
    5- ,  
 ).
 7 )  ,,    ,

 
3 "  5  7  ;)

 ,,     , 3

1XPHULFDOPRGHOGHVFULSWLRQ

 

-   5  
)        <    ,  
  5     1   5
3

-         7   ,     )  


     5
    5 )     5     
,, 3    $  2 5 )    .
  
 5  )

;) <,  E?F>

) - u% /(0

)  > - J -     - .    ,  7 % J      5


<   -
.  5  ,    3
M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain 127

330(0$7

 5        )  3(    5


  ,


 . 
-     8  8     /8803 ;    
  ] ,  5  -     - .  4
  2 7 -] K A7 
,   5  5 ) ;  5>

)[ K -\%]7 )\ K G -[%]7 ) ] K - [% \ G - \% [ /:0

)LJXUH  )  8  8     


  ,

  

 ,  3

 
 <   1-      5 1 -
 E:F>

:SIW
, A :SIW (   7 5 W d 1 I
, 1 /?0
A7 5 W ! 1 I

" 3:  , 5  <         5 4  - ,   1 


)7  
5   7  , .
-3

)LJXUH # 
5 <    
7      5 4  - , L I M(HAA I27 ,A :A 7 1 H
-
3
128 M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain

 ,
  -7    


    5  <    
 1 
 3 

    
 7
 
   
,  .
1  2   1
(7 :  ?7  , .
-3

7DEOH
 
,       


 
5HPDQHQWIOX[GHQVLW\RIPDJQHW%U ( EF
&XUUHQW, :A EF
6SHFLPHQPDJQHWLFUHODWLYHSHUPHDELOLW\U ( EF
6SHFLPHQFRQGXFWLYLW\ H3@N(A@ E'OF

7DEOH   
,       


 
<RXQJ
VPRGXOXV( :N(A(( E8F
3RLVVRQ
VUDWLR A3??
'HQVLW\ @MHA E;O?F

7DEOH *   
       


 
3ODWHGLPHQVLRQV (<( EF
3ODWHWKLFNQHVV (A EF
&RLOVHPLD[HV P < (3H E F
0DJQHWGLPHQVLRQV H < H < (A EF
/LIWRII A3H EF

   

 7  5 
   1  9

,- 
5 )  1   3     . 1   
2       .
,  
3
 5   .    1  5       5
     1-
,    3 6 < 7   1   ,   5   - 
-7
)    5 ) , >
       5
    
). 4    7  
 ,
 3
     ,  .
1  ,
 1-  $  2 5 3   ?H )  , 
,
    
< ,
-  W K A3(H  /   ,     5 
 ). 0  [7 \  ]
 7  , .
-3

)LJXUH ,
    [      ,     5 
 ).  W A3(H 3
M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain 129

)LJXUH ,
    \      ,     5 
 ).  W A3(H 3

)LJXUH ,
    ]      ,     5 
 ).  W A3(H 3

  ,    5 7

-  1    55   ).   ,, 3
Q
;   5 7 1        ). 7 <  
 
 

 3 
 
).  )   5      ). 3

"    1   1    55   -    5   


 1 C    
 ).
 3    5 4   7   
 1  L ) . 7 5 . -     55   -   

-     EHF3

0HDQGHU /LQH&RLO(0$7

 .  5 88     ,1


   ,
-   
5 4   3 '   ,1
  1 .  1-     -, 5 'I). 
1     
1   5
 /1-  ,  0   ,  =
5    
 
3 "    'I). 5 4  -   

 1- 
  
 , 3 '       1  )  3 P3
130 M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain

)LJXUH  )    


 
  , ,      

 ,  3

  @R )  , 


,
     < ,
-  W K A3(H  / 
 ,     5 
 ). 0  [7 \  ]  7  , .
-3  , 
   

    
;   , .  3

)LJXUH ,
    [      ,     5 
 ).  W A3(H 3

)LJXUH ,
    \      ,     5 
 ).  W A3(H 3
M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain 131

)LJXUH ,
    ]      ,     5 
 ).  W A3(H 3

"     7 
    
 ) ).  3 6 < ,  )

) 
,
- ,
,  7 )     5-  ,  
 ).   5

 
3

7UDQVLHQWDQDO\VLV

&   ,   


- 
 
   ,,  / 3 ?GH7
@GR03  1  5      -,  ,     
    ).
5 4  -3 ' 5      , . )
    , . 3

 ) 
   2 
).       

-  ,, 
    . 7    )   
    ). =
 5   3   ,     1     5   5  
5 
-   
   3   5 7     
   


  5   
 ,
   ). 
  ,
 3 & ,,  ,,
-  )
 
   5     5  ,
   < 
,
 /  7   0   ,  . ,
 /5 4  -7 ). 1 0 E(F3

" 3 (A  :  ,
-  
  1  ) 5 < ,
- \  ]
,
    ;  
 [</ ? 9   07 ;  5 88 3
 ,     

    ) 1
)>
x   , W K HN(A@ EF7 <  W< K SN(AS EF
x    , [ K A3A:A? EF
132 M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain

)LJXUH :   
 5 \ ,
     / 
5 0  ] ,
     /   0 5
88 7 5 4  - I M(3H E;I2F7 1 H -
L ,11
,  . ;   1
 ;
 3

#   ,
   3 (A ,11
,  . ,  . 1  

;  / 1
 ;
 03 T   ;)
 7 5   ,  1 
)  
-  

        5- 

 ,  
 ).  3

&RQFOXVLRQVDQGIXUWKHUZRUN

 ) -,  5   2 
).  ,
    7  
,
 . 1  ,     
   ,, 3  5 -,  4   5
. 
- ,
2  ,     
  ,

3   
 .
   1 )        5
  -    1 1   1-
   5  )      
     5
 / 33 5 ,
 0    
 
3     ,    1    5
<1

,
  1
    55   -  
  ,     
5 4  -3 "
1    55   ).   ,, 3 "    
   5- 

     
    5 3  ,,   )  1   

). = :    ,1
   
 ).  3 I) . 7   4 
 5   .    )  
-   ,   
 
-  
 3
"   < );    )

5   5

-   5-  ).  3 


  1   5   

 / 3 (A0 )

1 ,  )  , 


. ,
  1   5 
-  
<, 3

$FNQRZOHGJPHQW

 ); ) ,,  1-  ,  9 -= @   );


8 7  '19  7 63 :::(@S3
M. Ziolkowski and S. Gratkowski / EMATs Modeling in Time Domain 133

5HIHUHQFHV

E(F '3 '7 3 +   I3 *7 *  &.    '     6 9 

        6  
,,  5   4 #,  7 < ,
5  8  5  9#'#$ 95   :AAR7 
3
E:F 3 !
;);7 +3  C7 3'3 
-7 "3 +54  3 Q>  .
,  5 6.
' +

 *  &. /'+*&0  4 5 $  8
 ", 7 T"6 :AA@ 95   7
(M :A ' , 1  :AA@7 *
)7   %3
E?F I33 7
  
  > 8 ,
7 ,   ,,
  7 8- 

   7 D
3 (S7 &383   +363   37    8 7 6 ) U;7 (R@R3
ESF 3 %;7 3 %C7 I3 U   %3  7   '-  5 8, &


   
      7 Q
5 .    7 D
3 7
Q, '  - 5  
- :A(AO:A((3
EHF 3 I  I3 #7 (0$76IRU6FLHQFHDQG,QGXVWU\7 %
)     81
 7 :AA?3
EPF U3 !7 33 I ;  Q3$3 + 7 6.
  5 9 ;    *   '   8 7
+ . ) 5 8   V   . 6   . .
 7 D
3 7    "   5 8- 7
:AAP3
E@F 3 '  '3 -7 6  

- 5 
   1-  7 + . ) 5
8   V   . 6   . .
 7 D
3 7    "   5 8- 7 :AA(3
134 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-134

Eddy Current Testing of Ferromagnetic


Materials: Modelling of Flaws in a Planar
Medium
Chiara ZORNI a,1 , Christophe REBOUD a Marc LAMBERT b and
Jean-Marc DECITRE a
a
CEA, LIST, Dpartement Imagerie Simulation pour le Contrle,
F-91191 Gif-sur-Yvette, France
b
Dpartement de Recherche en lectromagntisme,
Laboratoire des Signaux et Systmes UMR 8506 (CNRS - SUPELEC - Univ Paris-Sud),
91192 Gif-sur-Yvette, France

Abstract. Eddy Current Testing (ECT) is a standard technique in industry for the
detection of surface breaking aws in ferromagnetic materials such as steels. In this
context, simulation tools can be used to improve the understanding of experimental
signals, optimize the design of sensors or evaluate the performance of ECT proce-
dures. CEA has developed for many years semi-analytical models embedded into
the simulation platform CIVA [1] dedicated to non-destructive testing. Following
a previous work carried out at the laboratory in the case of one aw located in a
cylindrical ferromagnetic part, the developments presented herein address the case
of a aw located inside a planar and ferromagnetic medium. The theoretical ap-
proach will be presented, as well as comparisons between simulation data obtained
from the literature.
Keywords. Integral formulation, Dyadic Green Functions, ferromagnetism, cracks,
eddy current testing

Introduction

Eddy current testing (ECT) is a standard technique in industry for the detection of surface
breaking aws in ferromagnetic materials such as steels. In this context, simulation tools
can be used to improve the understanding of experimental signals, optimize the design
of sensors or evaluate the performance of ECT procedures. CEA has developed for many
years semi-analytical models embedded into the simulation platform CIVA [1] dedicated
to non-destructive testing.
Following a previous work [2] carried out at the laboratory in the case of one aw
located in a cylindrical ferromagnetic part, the developments presented herein address
the case of one aw located inside a planar [3] and ferromagnetic medium. Simulation
results are obtained through the application of the Volume Integral Method (VIM) [4].
This approach has proved its efciency when considering canonical geometries, mainly
1 Corresponding Author E-mail: Chiara.Zorni@cea.fr
C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium 135

Figure 1. Conguration of the ECT problem.

due to the fact that, in those cases, analytical expressions of dyadic Green operators are
available in the spectral domain. While only one integral equation, involving either the
electric or the magnetic eld, is needed to describe the non-magnetic case completely, in
the ferromagnetic case two coupled integral equations have to be solved.
Therefore, when considering the ECT of a single aw, a system of two integro-
differential equations is derived from Maxwell equations. The numerical resolution of
the system is carried out using the classical Galerkin variant of the Method of Moments
[5]. Finally, the probe response is calculated by application of the Lorentz reciprocity
theorem [6].
The resolution has been generalized to the ECT simulation of M aws located in a
planar stratied medium. The theoretical approach will be presented, as well as compar-
isons between simulation results and measured data obtained from the literature [7].

1. Theoretical formulation

Let us consider the general conguration described in Figure 1: a planar, stratied and
ferromagnetic medium consisting of N linear, isotropic and homogeneous layers. The
rst and the last layers are two air half-space of permeability 0 and permittivity 0
and a time harmonic source with implied time-dependence ejt is placed in the rst
layer (with = 2f the angular frequency). The others k = 2, . . . , N 1 layers are
characterized by their conductivity k , their permeability k = 0 kr (kr, being the k th
layer relative permeability) and the complex permittivity k = 0 + jk 1 . One or
more 3-D bounded  material(s) is contained in l = n, . . . , m layers and is characterized
m
by its volume = l=n l and its conductivity ld and permeability dl .
The approach consists in calculating the electric and magnetic elds, E and H re-
spectively, in the volume of the defect without considering the defect contribution.
Then, the variations of the elds induced by the presence of the defect are computed
in . The volume integral formulation is derived by applying the Greens theorem [4]
to the Maxwells equations and involves the dyadic Greens functions for a multilay-
ered medium [3]. Let us introduce the ctitious electric and magnetic total (Eq. (1)) and
incident (Eq. (2)) currents as:

Jk (r) = k (r) Ek (r) , Mk (r) = k (r) Hk (r) (1)

J0k (r) = k (r) E0k (r) , M0k (r) = k (r) H0k (r) (2)
136 C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium

where k (r) = k kd (r) and k (r) = k dk (r) are the electric contrast and
the magnetic contrast functions respectively and Ek (r) and Hk (r) the total and E0k (r)
and H0k (r) the incident electric and magnetic elds in the k th -layer. A system of two
integro-differential equations gives the expression of the electric and magnetic elds in
the k th -layer as:

m 

Jk (r) = J0k (r) jk k (r) Gee  
kl (r, r ) Jl (r ) dr


l=n l

m 

jk (r) Gem  
kl (r, r ) Ml (r ) dr

(3)
l=n l

m 

Mk (r) = M0k (r) k (r) Gme  
kl (r, r ) Jl (r ) dr


l=n l

m 

2
k k (r) Gmm   
kl (r, r ) Ml (r ) dr , (4)
l=n l

where Gee  me  mm  em 
kl (r, r ), Gkl (r, r ), Gkl (r, r ) and Gkl (r, r ) stand for the dyadic Greens
functions and correspond to the electric or magnetic eld response at the point r of an
unit point current source of electric or magnetic nature placed in r  and are solution of
the Helmholtz equation. Thanks to the duality principle, only the dyads Gee 
kl (r, r ) and
me 
Gkl (r, r ) have to be computed and the two others are respectively obtained through:

Gem me
kl (, ) = Gkl (, ) (5)

Gmm ee
kl (, ) = Gkl (, ) (6)

Finally the response of the probe, in this case a single emitting/receiving coil, is
given by:
m  & '
1 
Z = 2
E0k (r  ) Jk (r  ) + jHk (r  ) M0k (r  ) dr  (7)
I k
k=n

where I is the feeding current of the probe.


In order to numerically solve the system (Eq. (3)) and (Eq. (4)), a discretization by
means of a Galerkins version of the method of moments [5] is applied. The volume of
the aw is divided into N parallelepiped voxels in which the electric and magnetic
elds are taken as constant values. In doing so a linear system of equations is obtained
where the system matrix is symmetric (from properties of symmetry of the Green dyads).
This system, however, can be associated to stiff matrices since the electric and magnetic
elds might differ by orders of magnitude depending on the characteristic impedance of
the medium. From a numerical point of view, it is advantageous to equilibrate the linear
system to solve in order to avoid numerical instabilities [8].
C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium 137

2. Numerical results for a single layer

Our approach is validated in a single layer conguration at a frequency f = 1.5 kHz.


According to our notation we have then N = 3 and k = 2, from now the latter will
be implied. Its thickness e, its relative permeability r , its conductivity and the corre-
sponding skin depth are given in Table 1.

e r
1.25 mm 100 1 MS m1 2.3 mm
Table 1. Geometrical and physical parameters of the ferromagnetic layer.

The ferromagnetic layer is affected by cuboid-shaped voids of length L, width w and


depth d given in Table 2. Three kinds of defects are dealt with, the through-wall defect
and two having a depth equal to 40 % of the thickness of the plate called inner defect
when breaking the surface on the same side of the acquisition probe and outer defect
when breaking the surface on the opposite side.

L w d
6 mm 0.2 mm [40 %; [100 %] e
Table 2. Description of the defect affecting the layer.

The variation of impedance Z is simulated by moving a single emitting/receiving


coil along the largest dimension of a defect affecting the ferromagnetic layer. The
acquisition conguration is described in Table 3.

outer diameter inner diameter height number of turns lift-off feeding current
3.2 mm 1.2 mm 0.8 mm 140 0.5 mm 1 mA
Table 3. Description of the acquisition conguration.

2.1. Comparison with data from literature

A rst comparison between our semi-analytical model and data from the literature [7] is
presented for the conguration described in the previous section. A ferromagnetic plate
with an outer defect of 40 % of the plates thickness is considered. The number of
voxels N used to discretize the volume of the defect is 20 20 6 on x, y and z axis
respectively.
The result of this simulation is compared with a nite-elements model result from
literature [7]. In Figure 2 the impedance variations obtained with the two different models
are superposed and a good agreement between the models is observed.

2.2. Comparison with the equivalent ECT conguration in cylindrical geometry

An equivalent conguration in cylindrical geometry has been simulated (see Figure 3) in


order to test the behavior of our model in different ECT cases. The model in cylindrical
geometry is integrated in CIVA and has already been experimentally validated [2]. In
order to be able to simulate a conguration as close as possible to the previous one
138 C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium

6
x 10
0

Imaginary part [ohm]


4

8
iterature
model
9
6 5 4 3 2 1 0
Real part [ohm] x 10
6

(a) (b)

Figure 2. Simulated ECT conguration (a) and the corresponding results (b) for a 40 % depth outer aw.
Signals show the impedance variations corresponding to the passage of the emitting/receiving coil over the
aw. The result is compared with a nite element result from the literature [7]

(a) (b)

Figure 3. Equivalent ECT conguration in cylindrical geometry (a) and the corresponding angular extension
(b). Geometrical and physicals parameters are given in Table 4.

outer diameter inner diameter w d


42.5 mm 40 mm 17.19 0.2 mm [40 %; 100 %] e
(a) Tube parameters (b) Defect parameters affecting the tube
Table 4. Parameters of the equivalent ECT conguration in cylindrical geometry described in Figure 3.

the inner and outer diameters of the tube (given in Table 4a) have been chosen large
enough with respect to those of the acquisition probe. The geometrical parameters of the
corresponding 100 % defect and 40 % inner and outer defects are reported in Table 4b.
To validate the similarity between the planar and the cylindrical congurations both
simulated results are compared in the case of an outer defect (see Figure 4). As expected
the two Z have the same shape and the same order of magnitude.
Two others congurations have been then simulated in planar and cylindrical ge-
ometries: an inner defect of 40 % (Figure 5) and a through-wall crack (Figure 6). Those
results show a good agreement between the two ferromagnetic models even if, for the
100 %-crack, a discrepancy between the two curves exists and might be caused by the
two different geometries.
C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium 139

6
x 10

Imaginary part [ohm]


3

7
cylindrical model
planar model
8
8 6 4 2 0
Real part [ohm] 6
x 10
(a) Comparison between the pla- (b)
nar conguration and the cylindrical
one.

Figure 4. Results for a 40 % depth outer aw. Signals show the impedance variations corresponding to the pas-
sage of the emitting/receiving coil over the aw. Comparison with the equivalent case in cylindrical geometry
of CIVA.

5
x 10

6
Imaginary part [ohm]

10

12

14

16 cylindrical model
planar model
10 5 0 5
Real part [ohm] 5
x 10
(a) (b)

Figure 5. Results for a 40 % depth inner aw. Signals show the impedance variations corresponding to the pas-
sage of the emitting/receiving coil over the aw. Comparison with the equivalent case in cylindrical geometry
of CIVA.

3. Conclusions and perspectives

A model has been developed in order to simulate the response of the ECT probe, consist-
ing in a set of emitting and receiving coils, to the presence of a aw in a planar stratied
and ferromagnetic medium. Simulations results and comparisons with the literature and
140 C. Zorni et al. / ECT of Ferromagnetic Materials: Modelling of Flaws in a Planar Medium

5
x 10
0

Imaginary part [ohm]


4

10

cylindrical model
planar model
12
14 12 10 8 6 4 2 0
Real part [ohm] x 10
5

(a) (b)

Figure 6. Results for a 100 % depth aw. Signals show the impedance variations corresponding to the pas-
sage of the emitting/receiving coil over the aw. The result is compared with a nite element result from the
literature. Comparison with the equivalent case in cylindrical geometry of CIVA.

the equivalent cylindrical model of CIVA carried out in the case of a single layer and a
single emitting/receiving coil show a good agreement between the simulated signals.
Experimental validations are currently in progress and will take into account also
the stratied case and the multiple aws aspect. The perspective for this work is the
extension of the model to simulate the response of a magnetic probe containing GMR
elements.

References

[1] CIVA: State of the art simulation platform for NDE. http://www-civa.cea.fr.
[2] A. Skarlatos, G. Pichenot, D. Lesselier, M. Lambert, and B. Duchne. Electromagnetic modeling of a
damaged ferromagnetic metal tube via a volume integral formulation. IEEE Trans. Mag., 44:623632,
2008.
[3] Sverine Paillard, Grgoire Pichenot, Marc Lambert, and Hubert Voillaume. Eddy current modelling for
inspection of riveted structures in aeronautics. In S. Takahashi and H. Kikuchi, editors, Electromagnetic
NonDestructive Evaluation (X), Studies in Applied Electromagnetics and Mechanics, pages 2532. IOS
Press, Amsterdam, 2007.
[4] Weng Cho Chew. Waves and Fields in Inhomogeneous Media. Wiley-IEEE Press, New-York, 1999.
[5] R.F. Harrington. The method of moments in electromagnetics. J. Electromagn. Waves Appl., 1:181
200(20), 1987.
[6] B. Auld, F. Muennemann, and M. Riaziat. Research Techniques in Nondestructive Testing, volume 7,
chapter Quantitative modelling of aw responses in eddy current testing, pages 3776. 1984.
[7] Haoyu Huang, Toshiyuki Takagi, and Tetsuya Uchimoto. Fast numerical calculation for crack modeling
in eddy current testing of ferromagnetic materials. J. Appl. Phys., 94(9):58665872, 2003.
[8] A. Abubakar and P. M. van den Berg. Iterative forward and inverse algorithms based on domain integral
equations for three-dimensional electric and magnetic objects. J. Comp. Phys., 23:236262, 2004.
Electromagnetic Nondestructive Evaluation (XIV) 141
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-141

Efficient Propagation of Uncertainty in


Simulations Via the Probabilistic
Collocation Method
Jeremy S. KNOPPa,1, John C. ALDRINb,
and Mark P. BLODGETTa
a
Air Force Research Laboratory, USA
b
Computational Tools, USA

Abstract. Eddy current models have matured to such a degree that it is now
possible to simulate realistic nondestructive inspection (NDI) scenarios. Models
have been used in the design and analysis of NDI systems and to a limited extent,
model based inverse methods for Nondestructive Evaluation (NDE). The science
base is also being established to quantify the reliability of systems via Model
Assisted Probability of Detection (MAPOD). In realistic situations, it is more
accurate to treat the input model variables as random variables rather than
deterministic quantities. Typically a Monte Carlo simulation is conducted to
predict the output of a model when the inputs are random variables. This is a
reasonable approach as long as computational time is not too long; however, in
most applications, introducing a flaw into the model results in extensive
computational time ranging from hours to days, prohibiting Monte Carlo
simulations. Even methods such as Latin Hypercube sampling do not reduce the
number of simulations enough for reasonable use. This paper presents the
Probabilistic Collocation Method as a non intrusive alternative to other
uncertainty propagation techniques.
Keywords: eddy current, Polynomial Chaos, Probabilistic Collocation Method

Introduction

Recent model development efforts have demonstrated significant advances in the


simulation of eddy current inspection scenarios [1-3]. This work is a critical
component of a strategy to use modeling and simulation in NDE to reduce the cost and
experimental burden of Probability of Detection (POD) studies, characterize damage,
and optimize probe design and inspection configuration.
The authors have investigated the utility of several numerical methods for
electromagnetic NDE. These methods include analytical methods, the finite difference
method, the finite element method (FEM), meshless FEM, the boundary element
method (BEM), and the volume integral method. In addition to these methods, hybrid
methods are also being developed [2]. While incremental advances in computational
efficiency are expected, the future development of model-assisted probability of

1
Air Force Research Laboratory (AFRL RXLP), Wright Patterson AFB, OH 45433, USA, Email:
Jeremy.Knopp@WPAFB.AF.MIL
142 J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM

detection (MAPOD) protocols and model-based inversion schemes should not depend
on a revolution in computational efficiency. One technical capability necessary for the
realization of MAPOD and deployable model-based inverse methods is the propagation
of uncertainty in input variables. In this work, variable refers to a measurable quantity,
and parameter refers to a quantity that must be estimated such as liftoff.
Stochastic numerical methods have been developed to propagate uncertainty
efficiently without relying solely on advances in computational power. Significant
advances in stochastic computation have been made especially in the last 20 years.
This paper briefly traces the development of stochastic numerical methods pertinent to
NDE modeling and simulation. The probabilistic collocation method (PCM) is
introduced and applied to two eddy current forward models.

1. Probabilistic Collocation Method

Spectral representation of uncertainty involves decomposing a random variable into


deterministic and stochastic components. Following the work of Norbert Wiener on
Homogeneous Chaos [4], Cameron and Martin pointed out that any second-order
functional of Brownian motion can be expressed as a mean-square convergent series in
terms of infinite-dimensional Hermite polynomials in Gaussian variables [5]. Any
random process with finite second-order moments, (which encompasses most physical
phenomena) can be represented by a Polynomial Chaos expansion using Hermite
polynomials and will converge according to the Cameron-Martin theorem. If this
random process is Gaussian, the convergence is exponential. An important property of
Hermite polynomials is that they are orthogonal with respect to Gaussian probability
measure. In fact for Gaussian processes, optimal convergence is achieved with the use
of Hermite polynomials because the weighting function is the same as the probability
density function (PDF) for the Gaussian distribution. To illustrate, the definition of an
orthogonal polynomial with respect to a given weighting function w(x) is given by:

b
( f g )w = f ( x) g ( x) w( x)dx = 0 (1)
a

The inner product of functions f(x) and g(x) is zero if they are orthogonal with w(x).
The only polynomial functions that satisfy this for the case where the weighting
function is equal to the PDF of a standard normal distribution are Hermite polynomials.
Ghanem and Spanos combined this concept with the finite element method to
simulate problems in solid mechanics [6]. For non-Gaussian processes, the
convergence is not optimal, but fortunately there are other orthogonal polynomial sets
for other general types of random variables that converge exponentially [7].
In work mentioned thus far, the formulations are often intrusive, meaning that
alternation of deterministic code is necessary for implementation. A non-intrusive
method was introduced in the mid-1990s [8]. Essentially, a reduced order polynomial
form of the model is derived with minimal simulations. The input values for these
simulations are derived from the roots of orthogonal polynomials which are dependent
only on the input distributions. Hermite polynomials are associated with Gaussian
distributions as described in the preceding paragraph. The weighting function for the
J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM 143

Uniform distribution is simply a constant on an interval bounded by [a,b]. Legendre


polynomials are orthogonal with this weighting function.

2. Case Study 1.

Application of the Probabilistic Collocation Method is model independent, but the


decision of what order and whether to include interaction terms does depend on the
complexity of the mapping of input variables to the output response. The orthogonal
polynomials and their associated roots are dependent on the type of input distribution
and its parameters. For this case study, the team15B workshop problem presented by
Steve Burke is used [9]. All simulations are conducted using VIC-3D. The only
variable that is changed in this example is the frequency, which is 500 Hz here instead
of 7 kHz. In this problem, the probe is scanned along the notch and the real and
imaginary components of the impedance are measured. A peak in the magnitude of the
impedance is observed when the center of the coil translated along the notch at a
distance of -12 mm and 12 mm away from the origin which is the center of the notch.
The objective of this exercise is to predict the probability density function (PDF)
of the peaks of the real and imaginary components of the impedance when the liftoff of
the coil is uniformly distributed between 1.53 mm and 2.53 mm, and the depth of the
notch is normally distributed with a mean of 5 mm and standard deviation of 1 mm. In
the original problem the coil liftoff is 2.03 mm and the depth of the flaw is 5 mm.
The first step is to derive orthogonal polynomials with respect to the input
distributions. From now on the parameter liftoff will be denoted by A, and notch
depth will be denoted by B. For liftoff, Legendre polynomials associated with that
particular distribution and parameter set are shown in equation 2. For the depth, the
Hermite polynomials are shown in equation 3.

L1 ( A) = A 2.03
L2 ( A) = A2 4.06 A + 4.0376
L3 ( A) = A3 6.09 A2 + 12.2127 A 8.0609 (2)
4 3 2
L4 ( A) = A 8.12 A + 24.5111A 32.5917 A + 16.1041
L5 ( A) = A5 10.15 A4 + 40.9312 A3 81.9626 A2 + 81.4899 A 32.1796
H 1 ( ) =
H 4 ( ) = 4 6 2 + 3
H 2 ( ) = 2 1 (3)
H 5 ( ) = 5 10 3 + 15
H 3 ( ) = 3 3

Note that since the mean of the Gaussian distribution is 5, parameter B is translated by
5.

B = 5 + H1 ( ) (4)

Next, a first order model might be used to approximate the real model. In this
case, there are three unknowns X0, X1, and X2 as shown in equation 5. Since we are
interested in both the resistance and reactance, this process is done separately for each.
144 J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM

(a) (b)

Figure 1. Change in the (a) real and (b) imaginary part of the impedance due to the
plate and flaw.

Three simulations are required to solve for the three coefficients. Four pairs of
collocation points for the inputs of the simulations are available from the roots of the
2nd order polynomials. Three pairs are selected to solve for the three unknowns. L 1
and H1 are simply evaluated at the three pairs of collocation points. This is all that is
needed for the a 1st order approximate model. Then a Monte-Carlo simulation can be
conducted on this simple model to calculate the PDF of the resistance and reactance.
To check the accuracy of this approximation, the next higher order collocation points
are necessary. The roots of the 3rd order orthogonal polynomials are used and six
simulations using the full model are conducted and compared with the approximate
model. Ultimately nine simulations are necessary for construction and evaluation of
the 1st order approximate model. Fortunately the same six simulations used for the
error check of the 1st order model can be used to construct the 2nd order approximate
model. The 4th order orthogonal polynomials are then used to check the accuracy of the
2nd order approximate model requiring eight more simulations for a total of fourteen
simulations. The authors have generally found that a 3rd order approximate model with
one interaction term is usually sufficient to predict output PDFs with good accuracy
for forward eddy current simulations, but there may be situations where higher order
approximations with more interaction terms are needed.

Y = X 0 + X 1 L1 ( A) + X 2 H1 ( B) (5)

The form of the 3rd order reduced model as described in equation 6 has eight
coefficients, one of which is an interaction term.

Y = X 0 + X 1 L1 ( A ) + X 2 H 1 ( ) + X 3 L 2 ( A ) + X 4 H 2 ( )
(6)
+ X 5 L 3 ( A ) + X 6 H 3 ( ) + X 7 L1 ( A ) H 1 ( )

To solve for the eight unknowns, eight simulations must be run. The roots of the
4th order Legendre and Hermite polynomials are used to select the input values for A
and B in these simulations. This is done separately for the real and imaginary
components of the impedance. Once again, everything is known except for the X
coefficients. The resulting system of equations is then solved and the coefficients are
J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM 145

listed in Table 1 for both the real and imaginary components of the impedance.
Incidentally, X0 is the expected value, which is one of the primary benefits of using
orthogonal polynomials. As expected, the interaction term X7 is close to zero in both
cases.
To check the error of this third order approximation, the roots of the 5 th order
polynomials are used to select collocation points. Ten more simulations are needed for
comparison with the reduced order model predictions. The results of the simulations
and the results of the reduced model are displayed in Table 2.

Table 1. Coefficients for 3rd order approximation model.


Model term Real Imaginary
X0 0.9693 2.4132
X1 0.2209 0.4386
X2 0.0012 0.0072
X3 0.0284 0.0479
X4 0.0004 0.0011
X5 0.0050 0.0077
X6 0.0001 0.0003
X7 0.0003 0.0018

Table 2. Evaluations of original and approximation model for error calculation.

Y real Y real Y imaginary Y imaginary


1.027 1.0270 2.5444 2.5415
1.0284 1.0280 2.5272 2.5295
1.0306 1.0305 2.5202 2.5214
0.9656 0.9656 2.4219 2.4195
0.9670 0.9666 2.4059 2.4081
0.9690 0.9690 2.3994 2.4007
0.9084 0.9084 2.3066 2.3043
0.9098 0.9092 2.2918 2.2937
0.9117 0.9116 2.2857 2.2869
0.8774 0.8778 2.2073 2.2053

The residual is defined as the difference between the full and reduced model in
equation 7, and the sum of squares of the residuals is defined in equation 8. The
relative sum of squares of the residuals is used to quantify the error in the
approximation. The relative sum of squares of the residuals is calculated by simply
dividing by the expected values as shown in equation 9. These quantities are summed
over the model results for each collocation point.

Y Y (7)

i
2

(8)
ssr = i
n

ssr
rssr = (9)
E[Y ]
146 J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM

The errors for the real and imaginary parts are given by: ssr(resistance) =
2.942810-4, rssr(resistance) = 3.035910-4, ssr(reactance) = 0.0020, and
rssr(reactance) = 8.479010-4.
Now there are two 3rd order polynomials with good accuracy for the real and
imaginary components of the impedance. Monte Carlo simulations can now be run
with ease on this 3rd order approximate model. The PDF for the real and imaginary
component of the impedance is shown in Figure 2(a) and 2(b) respectively.

6 25

5
2

4
15
PDF

P DF
3

1
2

05
1

0 0
0 85 09 0 95 1 1 05 11 27 26 25 24 23 22 21 2
Resistance Reactance
(a) (b)

Figure 2. PDF of the (a) real and (b) imaginary component of the impedance.

3. Case Study 2.

A second case study is presented for the characterization of a gap between multiple
plates, representing the problem of characterizing corrosion at the faying surface of a
multilayer aircraft structure. A diagram of the problem is shown in Figure 3 including
material properties and dimensions. In this hypothetical study, the thickness and
conductivity of the layer are treated as random variables. The conductivity of the top
plate is assumed to be a normal random variable with a mean value of 1.876107 S/m
and standard deviation of 3.384106 S/m as shown in Figure 4(a). The liftoff is defined
as a uniform random variable with the range of 0.55 +/- 0.05 mm as shown in Figure
2(b). VIC-3D was used to generate the simulated results for the study. The gap
between the plates was fixed at 0.3 mm and the frequency for the study was set to 4.0
kHz. Both 1st order and 2nd order PCM models were evaluated for this case requiring
only 3 and 6 model calls respectively. Monte Carlo simulations were also run with
100,000 model calls used in the study.



 1 ^

 air ^

2 ^

Figure 3. Diagram of case study 2 problem.


J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM 147

600 200
160
500 140
150
120
400
100
300 100
80

200 60

40 50
100 MC
20
PCM
0 0 0
15 2 25 05 0 52 0 54 0 56 0 58 06 0 085 0 09 0 095
(S/m) x 10
7 (a) dz (mm) (b) R (c)
g
300 0 704 0 704

0 706 0 706
250
0 708 0 708
200
0 71 0 71

150 0 712 0 712


X

X
0 714 0 714
100
0 716 0 716

50 MC 0 718 0 718
PCM 0 72 0 72
0
07 0 71 0 72 0 73 0 74 0 09 0 091 0 092 0 093 0 09 0 091 0 092 0 093
X R R
(d) (e) (f)

Figure 4. (a) Conductivity and (b) liftoff input distributions with ouput (c) resistance
and (d) reactance output 1D distributions and (e) joint distribution calculated using
PCM. (f) A joint distribution calculated using Monte Carlo simulations is provided for
comparison.

Results for the 2nd order PCM model are presented in Figures 4(c)-(f). The results
in Figures 4(c)-(d) show that the resistance is basically distributed uniformly and the
reactance is basically distributed as a normal random variable. This relationship is
expected based on the different effect of liftoff and conductivity changes in the
impedance plane. A comparison of the joint distribution calculated using both PCM
and Monte Carlo methods are shown in Figures 2(e) and 2(f) respectively. By
observation, good agreement was achieved using PCM with Monte Carlo methods
while only requiring 6 model calls here. Values for the error between the PCM and
Monte Carlo simulations for the 1st order PCM model, ssr and rssr, were 43.58 and
0.3177 respectively. Error values for the 2nd order PCM model, ssr and rssr, were
15.49 and 0.1129 respectively. Thus, the 2nd order model provides a 64.5% reduction
in the error between the PCM and Monte Carlo simulations with respect to the 1 st order
model. In conclusion, the PCM approach is highly efficient at propagating parameter
variation through models with minimal model calls.

4. Summary

A non-intrusive approach to propagating random inputs in eddy current forward


models was presented. The selection of collocation points in this method depends on
the type and parameters of the input statistical distribution and not the model itself.
Prediction of the PDF can be achieved with fewer simulations than traditional
uncertainty propagation methods. Future work will investigate using the method with
148 J.S. Knopp et al. / Efcient Propagation of Uncertainty in Simulations via the PCM

more input variables and also for inverse problems. The robustness of this method for
inverse problems will likely be much more sensitive to choice of order and interaction
terms.

5. Acknowledgements

Funding was provided in part, by the Air Force Office of Scientific Research.
Many thanks to Matt Cherry for reviewing the manuscript.

References

[1] J. S. Knopp, J. C. Aldrin, and K. V. Jata, Computational Methods in Eddy Current Crack Detection at
Fastener Sites in Multi Layer Structures, Nondestructive Testing and Evaluation, 24 (2009), 103 120.
[2] A. Skarlatos, C.G. Pascaud, G. Pichenot, G. Cattiaux and T. Sollier, Modelling of steam generator tubes
inspection in the proximity of support plates area via a coupled finite elements volume integral
method approach, in Electromagnetic Non Destructive Evaluation (XII), Studies in Applied
Electromagnetics and Mechanics, Y. K. Shin, H. B. Lee, and S. J. Song, (Eds.). Amsterdam, IOS Press
(2009), 51 58.
[3] H. A. Sabbagh, J. C. Aldrin, R. K. Murphy, and E. H. Sabbagh, Application of Model Based Inversion
to Eddy Current NDE of Heat Exchanger Tubing, in Electromagnetic Non Destructive Evaluation
(XII), Studies in Applied Electromagnetics and Mechanics, Y. K. Shin, H. B. Lee, and S. J. Song,
(Eds.). Amsterdam, IOS Press (2009), 26 33.
[4] N. Wiener, The homogeneous chaos. American Journal of Mathematics, 60 (1938), 897 936.
[5] R. H. Cameron and W. T. Martin, The orthogonal development of non linear functionals in series of
Fourier Hermite functionals. Annals of Mathematics, 48 (1947), 385 392.
[6] R. G. Ghanem and P. D. Spanos, Stochastic finite elements: a spectral approach. Springer Verlag, New
York, 1991.
[7] D. Xiu and G. E. Karniadakis, The Wiener Askey polynomial chaos for stochastic differential
equations, SIAM J. Sci. Comput., 24 (2002), 619 644.
[8] M.A. Tatang, W.W. Pan, R.G. Prinn, and G.J. McRae, An efficient method for parametric uncertainty
analysis of numerical geophysical model, J. Geophy. Res., 102 (1997), 21925 21932.
[9] www.compumag.org/jsite/images/stories/TEAM/problem15.pdf
Electromagnetic Nondestructive Evaluation (XIV) 149
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-149

Computation of the magnetic eld due to


a defect embedded in a planar stratied
media: Application to AC eld
measurement techniques
Denis PRMEL 1 and Grgoire PICHENOT
CEA, LIST, Centre de Saclay, F-91191 Gif-sur-Yvette cedex, France

Abstract. Eddy current techniques are useful for the detection and the characteriza-
tion of aws in conductive structures. The aw signal results from the measurement
of the local punctual value of the perturbed magnetic eld above the metal surface
in the vicinity of the defect by using some magnetic eld sensor (Giant Magneto
Resistance sensor (GMR) or Giant Magnetic Impedance sensor (GMI). This paper
aims to describe a semi-analytical numerical model integrated into the CIVA plat-
form which is able to compute the aw signal due to a given notch embedded in
a planar stratied media. The inducer may be chosen in a list of available exciting
coils into CIVA. Some numerical experiments and a comparison between simulated
data and experimental data show the validity of the new developments.
Keywords. Eddy current non destructive testing, Greens dyad formalism, Giant
Magneto-Resistance sensor, Giant Magneto-Impedance sensor

1. Introduction

Eddy current techniques are very useful for the detection and the characterization of
aws in conductive structures. Eddy currents are induced in the workpiece under test by
carrying a sinusoidal current in one or a set of exciting coils. According to the operat-
ing frequency, eddy currents induced in a thin "skin" near the surface of the conductive
media interact with a aw in such a way that currents ow around the ends of the defect
and down the faces of the defect. The deviation of eddy currents results in the perturba-
tion of the magnetic eld above the metal surface and in the vicinity of the defect. The
awed signal is analyzed in order to locate and size surface defects. Considering Non
Destructive Techniques (NDT), the variations of the magnetic eld can be detected by
the changes in the impedance of the exciting coil as it is commonly achieved in Eddy
Current (EC) methods. The aw signal is thus due to the global variation of the per-
turbed magnetic eld. An alternative NDT method, Alternating Current Field Measure-
ment (ACFM) consists in analyzing the aw signal provided by some magnetic eld sen-
sor as Giant Magneto-Resistance (GMR) or Giant Magneto Impedance sensors [1,2,3,4].
1 Corresponding Author: Denis Prmel, CEA LIST; E-mail: denis.premel@cea.fr
150 D. Prmel and G. Pichenot / Computation of the Magnetic Field

The aw signal results from the measurement of local, punctual values of the surface
magnetic eld. This signal is obtained by using a tiny inductive sensor or a magnetic
eld sensor [5].
ACFM technique is very useful for sizing any crack with a simple geometry. The
prediction of crack parameters (i.e. the length, the opening and the depth) from the mag-
netic eld is of great interest in practice. The resolution of this inverse problem requires
preliminarily solving efciently the forward problem, i.e. to be available to compute the
components of the magnetic eld resulting from the interaction of the crack with eddy
currents generated by some appropriated inducer. This paper aims to describe a numer-
ical model based on the volume integral approach [6,7,8] which allows one to compute
the aw signal due to a given notch embedded in a planar stratied media. The inducer
may be chosen arbitrarily from a list of exciting coils [9,10] available into CIVA [11] (a
NDT numerical platform dedicated to several NDT techniques). In ACFM techniques,
rectangular coils or current foils are preferably used in order to generate a uniform cur-
rent ow orientated perpendicularly to the length of the aw. This numerical model is
rstly numerically validated by using two kinds of numerical models. These ones differ
from the observation equation which is used for computing the response of the magnetic
sensor. Then, some validation experimental results are presented. These new develop-
ments, translated into new functionalities in the last version of CIVA allow one to con-
duct some parametric studies such as the lift-off of the magnetic sensor, the tilt of the
inducer, and other parameters which could be able to inuence the sizing of the defect by
using one or two components of the magnetic eld. This work precedes also some work
dedicated to the inversion problem [12].

2. An example of ECT conguration

Figure 1 displays an example of ECT conguration consisting of a stratied slab affected


by aws. Eddy currents in the slab are induced by an exciting coil driven by an alternative
current of angular frequency . Into the CIVA software, this coil may be chosen among
cylindrical coils, with a ferrite core or not, or rectangular coils. In this gure, we consider
two rectangular coils in order to approach the behavior of a uniform current in the Y
direction. In this gure, we also assume a magnetic sensor which is able to provide the
three components of the perturbed magnetic eld due to the presence of the defects. It
is important to note that the response of the sensor is obtained taking into account the
sensitivity of the sensor, expressed in Volts per Tesla and that the response results from
the difference between the eld obtained in the presence of the aw and the eld obtained
when there is no aw. The computation module may consider an arbitrary number N of
layers made of materials being linear, isotropic and not magnetic (with the permeability
of the vacuum denoted by 0 ). The conductivity l of each layer is constant, except when
there is a aw. This one is indeed represented by local variations of the conductivity (r)
in the lth layer.
The continuation of this paper is organized as follows. In section 3, the eddy current
modeling problem is solved by using the Volume Integral Method implying a system of
two integral equations for which the kernels are based on the Greens dyad formalism.
In the previous version of the CIVA platform, the computation of the e.m.f induced in
a receiving pick-up coil is available but we aim at the addition of new functionalities in
D. Prmel and G. Pichenot / Computation of the Magnetic Field 151

Figure 1. An example of ECT conguration.

order to have the capability to compute the punctual values of each component of the
perturbed magnetic eld above the awed material. So, two approaches are considered
for obtaining the numerical solutions of the forward problem. These two approaches are
used for the numerical validation of the new functionalities. Numerical results used for
the numerical validation are compared in section 4 while a real experiment is proposed in
section 5. Finally experimental data are compared with simulated data in order to show
the experimental validation. A short conclusion follows in section 6.

3. The Volume Integral Method and the specic dyad for the problem

The computation of the perturbed magnetic eld above the plate requires computing
rstly the electrical internal eld in a nite domain surrounding the aw. Each layer
may include one nite volume containing an object and the total internal eld in each
layer k results from the interaction from the primary eld Epk (r) in the layer due to the
driving probe and the perturbation eld due to a set of elementary aws, each one being
considered in a single layer l. The total electric eld in each layer, denoted by Ek (r), is
given by:

N 

Epk (r) Gkl (r, r ) [l (r )] El (r ) d
(ee)
Ek (r) = j0 (1)
l=1 aw

The contribution is null if there is no aw in the lth layer ((r ) = l ). The superscript
(ee) means that we compute the components of the electrical eld due to an unit point
source which is a solution of:

Gkl (r, r ) kk2 Gkl (r, r ) = kl I (r r ).


(ee) (ee)
(2)

In this last equation, the observation point r is assumed to be in the layer k while the
source point r is in the layer l. kl stands for the Kronecker symbol and I is the unit
dyad. According to the quasi-static regime, the wave number kk2 in the k th layer is given
152 D. Prmel and G. Pichenot / Computation of the Magnetic Field

by kk2 = j0 k . The Greens dyads satisfy as usually the appropriated boundary condi-
tions at the innity in the transverse directions and at the interfaces between two different
layers in the normal direction to the planar surface of the stratied medium. The analyt-
ical expression of the Greens dyad corresponding to a multilayered planar medium is
given in [13].
The total internal eld in each layer appears inside and outside the integral. This
state integral equation must be solved numerically by the Moment Method (MoM) [14].
Knowing the total electrical eld in each layer, the perturbed magnetic eld can be com-
puted above the plate, in the air region 1, by an observation equation which introduces
the Greens dyad G1k (r, r ). The superscript (me) means that the components of the
(me)

magnetic eld is calculated due to a current dipole:

(r, r ) = Gkl (r, r )


(me) (ee)
Gkl (3)

The total perturbed magnetic eld takes into account all the contributions of each
layer k:

N 

G1k (r, r ) [k (r )] Ek (r ) d
(me)
B1 (r) = 0 (4)
k=1 aw

This equation has been implemented into CIVA. In order to validate these new de-
velopments, some other numerical data have been computed by using another numerical
code which has been previously validated. Let us consider that the components of the
magnetic eld can be estimated by the e.m.f induced in a tiny pick-up coil when the sec-
tion of the pick-up coil tends towards zero. In this case, the response of three receiving
pick-up tiny coils is given by the changes in the mutual impedance and can be obtained
via the reciprocity theorem:

N 

I1 I2 Z = [l (r)] El (r) Epl (r) d (5)
l=1 aw

where Epl (r ) stands for the primary eld that would be induced by the receiving coil
in the awed region, in the lth layer, if it was driven by a current I2 , and I1 is the
driving current of the exciting coil. The total electrical eld El (r) in the lth layer comes
from the numerical resolution of equation (1). According to the orientation of the pick-
up receiving coil, one can obtain an approximation of one component of the magnetic
eld by using the Faraday-Lorentz law. The numerical results which can be obtained by
these two approaches can be now compared in order to show the validity of these new
functionalities.

4. Numerical Validation

Figure 2 displays the overview and the front view of an ECT conguration consisting of
a non magnetic slab constituted by four layers. The thickness of each layer is xed at
D. Prmel and G. Pichenot / Computation of the Magnetic Field 153

0.8 mm, the values of the conductivities of each layer expressed in MS/m are given by
1 = 0.1, 2 = 0.5, 3 = 1, 4 = 10. The operating frequency is chosen at 100 kHz and
let us consider two notches. The rst one is breaking the surface of the rst layer, it is 2
mm length, 0.3 mm depth and its opening is 0.1 mm. The second aw is embedded at the
top of the third layer, its length is equal to 4 mm, its depth is 0.5 mm and its opening is
also 0.1 mm. These two aws are differently orientated, the rst one is orientated along
the X axis while the second one is orientated along the Y direction. The eddy current
(EC) probe is constituted by two exciting rectangular coils. For each one, the inner width
is 4 mm , the inner length is 8 mm and the height is equal to 0.1 mm. The number of
turns is 5 and nally the width of the current foil is 1 mm.

Figure 2. An overview (on the left) and a front view (on the right) of an ECT conguration used for a numerical
validation.

The EC probe is scanning along three lines in the X direction and three values of
the Y position have been chosen in order to verify the validity of the results in the two
directions of scanning. Figure 3 and 4 show a comparison between the two numerical
approaches resulting from the implementation of equations (4) and (5). The real and the
imaginary parts of the Bx , Bz components are respectively shown. Three curves are dis-
played corresponding to three values of the Y position of the probe. These numerical re-
sults show a good agreement between the two numerical approaches. In the next section,
an experimental validation is achieved.

5. Experimental validation

Experimental data have been obtained from an experimental set-up which is depicted in
Figure 5. Let us consider a non magnetic slab made of Inconel 600, the conductivity is
assumed to be 1.02 MS/m. The operating frequency is 100 kHz. Eddy current are in-
duced in the slab by a rectangular coil which is etched on a exible lm. The inner width
of the coil is 23 mm, the inner length is 62 mm and the height is equal to 35 m. The
number of turns is 100 and nally the width of the current foil is 7 mm. The distance from
the bottom of the exciting coil to the slab is about 2 mm. A Giant Magneto Resistance
sensor (GMR) measures the Bx component of the magnetic eld. The EC experimental
signal is obtained by subtracting the response of the probe in the awed region and the
154 D. Prmel and G. Pichenot / Computation of the Magnetic Field

Figure 3. Real and Imaginary parts on the Bz component.

Figure 4. Real and Imaginary parts on the Bx component.

signal without the aw. The distance from the GMR to the target is about 0.1 mm but
this value can be affected by an uncertainty due to the encapsulation of the sensitive part
of the sensor in the chip. The thickness of the slab is 1.55 mm and we consider a rst
notch for a preliminary procedure denoted by "calibration" . Indeed, in most industrial
applications, due to uncertainties in the experimental system, the measured EC signal
has to be calibrated with respect to a reference conguration. At this stage, we choose a
rst notch of 10 mm length, 0.93 mm depth and the opening is 0.1 mm. The EC probe is
scanning a line along the length of the defect and nally simulated data and experimental
data are normalized in order to t the maximum value of the magnitude of the EC signal.
The shape of the two signals are compared in Figure 6(a). In this gure, the real and the
imaginary part of the signals are represented, the same signals are also displayed in the
impedance plane diagram. As it is expected, the comparison shows a very small discrep-
ancy between simulated data and experimental data due to, probably, local variations of
the lift-off during the movement of the probe. The complex value of the normalization
constant is xed for all the following results of comparison. Figure 6(b) displays the EC
signals when the probe is scanning perpendicularly to the defect.
In the last experiment, let us consider two identical aws separated by a gap of 3
mm (See Fig. 8(a)). The length of each defect is 10 mm, the opening is 0.1 mm and
the depth is 0.76 mm. Figure 7(a) and 7(b) display two cartographies of the EC signals
D. Prmel and G. Pichenot / Computation of the Magnetic Field 155

Figure 5. A scheme of the experimental set-up and an example of the EC probe including the GMR Sensor.
Only one rectangular coil is activated for obtaining experimental data.

(a) the probe scans a line parallel to the notch (b) the probe scans perpendicularly to the notch

Figure 6. A comparison of the EC signals of the rst aw.

collected when the probe is scanning above the awed region. Figure 8(b) shows a slice
view when the probe is moved along a line parallel to the X axis.

(a) Simulated data (b) Experimental data

Figure 7. Two cartographies corresponding to the magnitude of the EC signals.

6. Conclusion

The Volume Integral Method (VIM) and the Greens dyadic formalism has been applied
in order to compute the response of a magnetic sensor due to an arbitrary number of aws
in a stratied planar structure. The new functionalities which have been developed are
able to compute the punctual value of the perturbed magnetic eld above a awed region.
Two numerical approaches have been compared in order to show the validity of the new
developments and some comparison results between simulated data and some experi-
156 D. Prmel and G. Pichenot / Computation of the Magnetic Field

(a) The experimental set-up. (b) the probe scans perpendicularly to the notch

Figure 8. A comparison if the EC signals due to two close aws when the probe is scanning along a line
parallel to the notch.

mental data conrm the validity of the global modeling approach. These developments
will be available in the new version of the CIVA platform and some new developments
are today already engaged for considering magnetic materials of constant permeability.

References

[1] R. F. Mostafavi and D. Mirshekar-Syahkal, "AC Fields Around Short Cracks in Metals Induced by
Rectangular Coils", IEEE Transactions on Magnetics, 35(3), (1999), 2001-2006.
[2] M. Ravan, S.H.H. Sadeghi and R. Moini, "Neural network approach for determination of fatigue crack
depth prole in a metal, using alternating current eld measurement data", IET Sci. Measurement Tech-
nology, 2(1), (2008), 3238.
[3] R. K. Amineh, M. Ravan, S. H. Sadeghi, and R. Moini, "Removal of Probe Liftoff Effects on Crack
Detection and Sizing in Metals by the AC Field Measurement Technique", ", IEEE Transactions on
Magnetics, 44(8), (2008), 20662073.
[4] R. K. Amineha,1, M. Ravan, S.H. Sadeghi R. Moini, "Using AC eld measurement data at an arbi-
trary liftoff distance to size long surface-breaking cracks in ferrous metals", NDT&E International, 41,
(2008), 169177.
[5] B. Marchand, C. Zorni, J.M. Decitre and O. Casula, "Recent Developments of Eddy currents
probes",ENDE 2009, to be published.
[6] J.R. Bowler and L.D. Sabbagh and H.A. Sabbagh, "A Theoretical and Computational Model of Eddy
Current Probes incorporating volume integral and Conjuguate Gradient Methods", IEEE Transactions
on magnetics, 25(3), (1989), 26502664.
[7] J.R. Bowler and S.A. Jenkins and L.D. Sabbagh and H.A. Sabbagh, "Eddy Current Probe Impedance
due to a volumetric aw", Journal of Applied Physics, 70(3), (1991), 11071114.
[8] J.M. Decitre, D. Prmel and M. Lemistre, "3D Modelling of a Magneto-optic imager by a dyadic Greens
functions approach", Review of quantitative Nondestructive Evaluation, 22, ed. D.O. Thompson and
D.E. Chimenti, IOP, (2003), 695-702.
[9] T. Theodoulidis, G Pichenot , "Integration of tilted coil models in a volume integral method for realistic
simulations of eddy current inspections", Electromagnetic Non-Destructive Evaluation (XI), Studies in
Applied Electromagnetics and Mechanics, A. Tambourino, Y. Melikhov and L. Udpa, Amsterdam, IOS
Press, 2008.
[10] T.P. Theodoulidis and E.E. Kriezis, Impedance evaluation of rectangular coils for eddy current testing
of planar media, NDT E International, 35(6), (2002), 407414.
[11] www-civa.cea.fr.
[12] D. Prmel and A. Baussard, "Eddy current evaluation of three-dimensional aws in at conductive
materials using a Bayesian approach, Inverse problems, 18(6), (2002), 18731889.
[13] W.C. Chew,"Waves and elds in inhomogeneous media", IEEE Press (2nd edition), Piscataway,1995.
[14] R.F. Harrington, "The method of moments in electromagnetics", Journal of Electromagnetic Waves and
Applications, 1(3), (1987), 181200.
Electromagnetic Nondestructive Evaluation (XIV) 157
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-157

Automatic Detection and Identification of


Image Quality Indicators in Radiograms
Piotr BANIUKIEWICZ1 and Ryszard SIKORA
Department of Electrical and Computer Engineering, Faculty of Electrical
Engineering, West Pomeranian University of Technology, al. Piastow 17, 70-310
Szczecin, Poland

Abstract. In this paper the authors focus on detection of wire image quality
indicators, which are widely applied in industrial radiography. Two algorithms for
IQI detection are proposed. The first algorithm is intended for fast and rough
detection of position of IQIs. It is based on evaluation of correlation between the
radiographic image of IQI (together with a plastic case) and the fragments of the
analyzed radiogram. The second algorithm uses the Radon transform applied to the
region of interest (ROI) obtained from the first algorithm to extract the exact
position of wires as well as information on visibility of each wire.

Keywords. object recognition, radiography

Introduction

Welding is a fabrication process that joins materials by causing coalescence. This is


often done by melting two workpieces and joining them to make coalescence possible.
Many different energy sources can be used for melting materials. The most popular are
gas flame, electric arc, laser beam, friction and ultrasound. Welding can be done in
many different environments, including open air, underwater and space. The chemical
and physical processes that occur during melting and joining materials are very
complicated and not fully understood. For example, metal cooling is a highly nonlinear
physical process which proceeds simultaneously with chemical reactions. Those
processes result in raising mechanical tensions in material as well as forming new
chemical compounds in joint area (e.g. gases, slag, etc). Those undesired products are
often the source of potential cracks and discontinuities in the joint. Additionally, the
heat applied to the joint area in the process of welding causes overheating of the
surrounding material and in the result the changes in its mechanical properties.
Therefore, welding without proper precautions can be dangerous because of possible
occurrence of defects and discontinuities in the weld. Due to high complications, the
weld testing process mostly relies on the experience of human being. Defects and
discontinuities in the weld can reduce the strength of the joint and can lead to a
catastrophic failure. This is critical due to the application of a welding technique for
joining stress-exposed parts such as ship hulls, skyscrapers, bridges and many more.
There are many methods of nondestructive testing of the weld. Unfortunately, in most

1
Corresponding Author: Piotr BANIUKIEWICZ, West Pomeranian University of Technology, al Piastow 17,
70 310 Szczecin, Poland, email: baniuk@zut.edu.pl
158 P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms

cases, the weld area is inhomogeneous, which is the result of processes which take
place during metal cooling. This is a potential source of unwanted signals and
disturbances that affect all methods of weld testing.
Through the years the technology that utilizes penetrating rays has been the most
important nondestructive testing (NDT) method. Fast and accurate x-ray converters that
have been developed recently open new possibilities for x-ray applications and make
this technique as flexible as never before. Digital radiography joins modern digital
image processing algorithms with traditional x-ray testing method. Together with the
fast development of digital radiography, there arise new nondestructive testing
technologies based on it. One of the biggest problems of assuring the demanded quality
of a product is the need of analyzing a large number of radiograms. In shipbuilding
industry this number can reach even thousands. All radiograms must be analyzed by a
qualified operator in order to find those welds which do not fulfill very restrictive
safety norms. This process takes a lot of time and effort but it tends to be fully
automated.The advanced computerized systems for radiogram analysis are still under
development [1] and the decision taken by a human being is still considered as the most
reliable and trustworthy. The process of radiogram analysis consists of a few stages
which are strictly regulated by adequate norms. The first stage is always accepting or
rejecting the radiogram according to its quality. Too dark, too bright, blurred or noised
radiograms do not provide trustworthy information on internal structure of the
specimen. In this paper the authors present an algorithm for detecting Image Quality
Indicators (IQIs), which are commonly used for obtaining the quality of radiogram.
This algorithm can be applied in the first stage of radiogram analysis process and it
stands as a part of the automatic radiogram analysis system (ISARView) which is being
developed by the authors. The aim of this procedure is only to find IQI in the
radiogram and pass its position to other procedures implemented it the system for
further analysis (e.g. accepting or rejecting radiogram because of its quality). The main
procedure has been implemented in C++ and optimized for 64-bit environment.

1. Image Quality Indicators

In radiography, one usually has a wide choice in the parameters of the radiographic
technique. Most of them are regulated by norms but there is still margin for changes
that can be done making the radiogram better or worse. For example, one has a choice
of source-to-specimen distance, a choice of film type, a choice of film density, a choice
of X-ray kilovoltage, X-rays or gamma-rays etc. The resulting radiograph therefore can
vary between high quality and poor quality; quality in this case meaning the ability to
detect small critical flaws such as cracks [2]. There are a lot of norms, both national
and international, issued for particular applications as well as for a general purpose,
giving advice or instructions on the choice of technique details, in order to ensure good
quality radiographs. The usual method for measuring image quality in radiography is
the use of Image Quality Indicators (IQIs). The quality of a radiographic image can be
estimated in terms of three factors:
Image unsharpness
Image contrast
Image noise
P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms 159

Figure 1. Exemplary image quality indicators of one wire (a). The effects of increasing geometric
unsharpness in a duplex type IQI. A pair where the individual wire images are merged (c) and one can
no longer see the two separate (b) wires stand for the unsharpness measure.

Using IQI it is possible to obtain all these three parameters and assess the quality of
picture. There are four main types of IQIs. Those are: the wire type, the step/hole type,
the hole-in-plaque type and the duplex wire type. Here, only the wire type and the
duplex wire type are considered. A typical wire IQI contains six or seven straight wires
of the same or similar material to the specimen having various diameters. The wires are
held parallel to one another in a plastic mount with appropriate identification symbols.
The double wire indicator is quite similar to wire type one. It consists of a series of
pairs of wires of high density material (tungsten and platinum) where each pair of wires
of diameter d is spaced at a distance d apart. Typical construction details and
dimensions are specified in norms [3],exemplary indicators are depicted in Figure 1.
The quality of the radiogram is defined based on the visibility of the wires. The duplex
wire IQI measures only unsharpness as it is explained in Figure 1.

2. Problem statement

The most important problem relating to IQI detection is that the low signal amplitude
of the IQIs contributes to their poor visibility. Additionally, a typical radiogram
consists of nonlinear trend that results from nonhomogeneity of the specimen and its
non-uniform thickness. The amplitude of the trend signal is much greater than the
intensities of particular wires. Thus, the process of the automatic determination of the

Figure 2. Exemplary real radiograms (fragment with IQI only) and corresponding averaged intensity
profiles taken along weld line.
160 P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms

quality of the radiogram needs special algorithms for accurate detection of IQIs.
Exemplary real radiograms and intensity profiles are showed in Figure 2. The intensity
profiles are averaged from dozens of lines taken along the weld line. As one can see,
the signal from the wires is very poorly visible due to the low SNR. On the other hand,
the plastic case form IQI gives a clear signal that is visible in intensity profile. This
gives a possibility to detect position of IQI by means of image recognition algorithms.

3. The IQI detection process

The process of detection of IQIs was divided into two steps. In the first stage only the
presence and the position of the indicator are obtained whereas, the second step
provides exact information on the position of particular wires. Thus, it is possible to
evaluate the quality of the radiogram based on the visibility of wires in the wire type
indicator or merging a pair of wires for the double wire indicator. Under fundamental
initial assumptions for both algorithms lies norm [3] that regulates placement of IQI in
the radiogram. In the case of welds, it is expected that IQIs are always located on the
weld line almost perpendicular to it. Thus, the region of the analysis is limited to the
weld line and a small surrounding area.
The plastic case covering the wires in IQI is the best visible part of the whole
indicator in the radiogram as well as in the intensity profile. This fact is used in the
process of detection and localization of IQI. The algorithm works with averaged
intensity profiles taken along the weld line. The position of IQI is obtained from a
fusion of two methods commonly used in the pattern recognition. Those methods are:
1) to measure the dependence between sought pattern p and signal I using Pearson
product-moment correlation coefficient and 2) evaluating phase correlation between
Fourier transform of pattern p and Fourier transform of signal I. The pattern p is
generated artificially as an average calculated from dozens of intensity profiles of IQIs
obtained from various radiograms. The resulting waveform can be assumed as the best
representative of IQI signal holding general features of that signal.The correlation
coefficients were calculated between the pattern p and the window of size p sliding
over the signal I with step of 1:

n Ii p Ii p
ri = (1)
n I ( Ii ) n p ( p )
2 2 2 2
i

where Ii stands for i-th position of the window in the signal I, n is the number of
elements of p and Ii. The position of the maximum of function r is correlated with the
position of IQI in the radiogram. In order to make the detection process more reliable
the second criterion based on phase correlation has been proposed. In image processing,
phase correlation is a method of image registration, and uses a fast frequency-domain
approach to estimate the relative translative offset between two similar images. Here,
this approach was applied to one-dimensional waveforms, namely intensity profile I
and pattern p:

{
rp = F 1 e
(
i I p )
} (2)
P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms 161

Figure 3. Plots of Pearson product moment correlation coefficient r and phase correlation rp evaluated
for intensity I (Figure 1b) and searched pattern p. Function R is fusion of r and rp clearly pointing
position x of IQI (x=1182 for radiogram depicted in Figure 1b).

where F-1 stands for inverse Fourier transform and I, p are phases of Fourier
transforms of I and p respectively.
Because of the noise and the linear trend that are mostly present in radiographic
images there is no clear maximum in either of the functions visible. Thus, the final
criterion R was proposed as a fusion of correlation function r and phase correlation rp.
All those functions are depicted in Figure 3. The position of the maximum of R stands
for the position of IQI in the radiogram.
In practice, the smallest element readily visible in the area under inspection is used
to determine IQI sensitivity. Thus, the information on the presence or the position of
IQI in the radiogram is not sufficient to draw any conclusions on its quality. One must
know where exactly the wires are located and which of them are visible. For this reason
the second algorithm has been developed. It uses information on location of IQI
obtained from the first method to find positions of wires in the indicator. One of the
well known methods commonly used for detecting straight lines in pictures is the use
of Radon transform or related Hough transform. Using the Cartesian coordinate system
to describe line integrals and projections, the IQI is represented by a two-dimensional
function f(x, y) and each line integral by the (, t) parameters. The equation of the line
(being a particular wire in our case) is:

t = x cos + y sin (3)

The integral of line (3) is given as follows:

P (t ) =
( , t ) line
f ( x, y ) ds (4)

what yields :


P (t ) =

f ( x, y ) ( x cos + y sin t ) dxdy (5)

where stands for the projection angle, t defines the position of the integral line on the
projection and P(t) is known as Radon transform of function f(x, y) [4].
The most important feature of Radon transform is amplifying straight lines to be
found in pictures invariantly to their rotation angle. This makes it especially useful in
the IQI detection process. Figure 4 shows Radon transform (sinogram) of the
162 P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms

Figure 4. a) The IQI wires pattern generated artificially according to norms, b) Radon transform of part
of the radiographic image and c) integrated intensity for =6 as function of t. d)Results of detection of
IQI position.

radiogram section that contains IQI. It is clearly visible on the line integral that peaks
from the wires have been amplified in comparison to those original ones depicted in
Figure 2. The detection algorithm of these peaks was realized in the same way as in the
case of the first step described above. The correlation function r (1) is evaluated for all
angles from the sinogram and an artificially generated pattern representing the ideal
IQI signal. The pattern is a sum of six Gaussian functions with increasing amplitudes
(Fig. 4a). But now, unlike the previous case, the wires are under consideration and not
the plastic case. Therefore, the maximum of r is found for two arguments of function
P(t) and it stands for the best matching of the whole artificial IQI pattern to the line
integral for given (Fig. 4c). The angle is the rotation angle of the IQI relating to x
axis whereas t is the radial coordinate of the first wire (beginning of the pattern) in the
indicator. Having and t, the first wire can be expressed as linear function y(x) defined
for analyzed region of interest, which has been found by first algorithm, in the
radiogram:


y ( x ) = tan + ( x x0 t cos ( ) ) + y0 (6)
2

where x0 and y0 stand for the centre of the image. The result of the wires detection is
presented in Figure 4d.

Conclusions

Radiography is the most popular and well known method of nondestructive testing
widely applied in industry. Information on the quality of radiographic picture is
especially important in the case of the method of nondestructive testing. The algorithms
for IQI detection presented in this paper have been applied in the automatic system of
radiogram analysis (ISARView) in the first stage of the whole process of the analysis.
Taking into consideration the fact that even thousands of radiographic pictures can be
taken during manufacturing of e.g. a ship, the methods presented in this paper can
significantly speed up the whole process of nondestructive testing in various branches
of industry.
P. Baniukiewicz and R. Sikora / Automatic Detection and Identication of IQIs in Radiograms 163

Acknowledgments

This work was conducted in a framework of the research project The intelligent
system for radiograms analysis, supported by the Polish Ministry of Science and
Higher Education. Grant no. NNR01003706/2009 (2009-2012).

References

[1] European Commission sponsored project Development of novel digital radiography technology. To
facilitate the traditionally less research intensive inspection industry sector change from manual film
radiography to automated digital, contract No NMP2 C 2005 515746.
[2] R. Halmshaw, T. Kowol, Image quality indicators in industrial radiography, IE NDT LTD technical
materials
[3] European norm EN 462 part 1:1994
[4] A. G. Rann, and A. I. Katsevich, The Radon Transform and Local Tomography, Boca Raton, FL: CRC
Press, 1996.
164 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-164

Numerical Simulation of Acoustoelastic


Effect in Pre-stressed Media
Cuixiang PEI1 and Kazuyuki DEMACHI
Department of Nuclear Engineering and Management, University of Tokyo,
Tokyo 113-8656, Japan

Abstract: In this work, a code of finite element method is developed for numerical
simulation of the acoustoelastic effect in pre stressed media. The possibility of
assessing the stress status, by using the EMAT (electromagnetic acoustic
transducer) receiver for precise measurement of LCR (refracted longitudinal) wave,
is investigated. Besides that, the relationship between the polarization of transient
Rayleigh wave and the state of stress is also predicted by simulation.

Keywords: Ultrasonic NDE, Residual stress, Acoustoelastic effect, EMAT,


Polarization

Introduction

Ultrasonic NDE is one of the most promising methods for quantitative determination of
residual stress in materials. And, most of ultrasonic methods for stress measurement are
based on the effect of acoustoelasticity. The acoustoelastic effect, related to the change
of ultrasonic wave speed and the polarization of Rayleigh wave, is a well established
phenomenon that has been extensively reported in lots of literatures [1-4]. At present,
the residual stress measurement by using acoustoelastic effect is mainly studied with
experiment method. And simultaneously, the numerical simulation method can provide
an another useful tool for the research of stress measurement with ultrasonic wave. In
this work, the possibility of numerical simulation of acoustoelastic effect related to the
stress status was investigated by using FEM method.
Now in the practical application, the stress measurement using ultrasonic wave is
mainly based on measuring the variation of ultrasonic wave velocity by time-of-flight.
As the LCR wave is more sensitive to the in-plane stress and easy to yield, it is widely
used in the subsurface stress measurement [9]. In this work, a kind of non-contact
transducer named EMAT is proposed to precisely measure the acoustoelastic effect on
LCR wave.
There are some literatures have demonstrated that the polarization of Rayleigh
wave related to the state of stress is usually more sensitive than the wave speed. The
relationship between Rayleigh wave polarization and state of stress in homogeneous
material was investigated by Junge et al[3]. However, in their analytical model, only a
steady sinusoidal Rayleigh wave propagates in the material. Actually it is very difficult
to generate a pure steady Rayleigh wave in the media, the Rayleigh wave in sinusoidal
mode is usually interfered by the longitudinal and shear wave. In that case, it is
1 Email: peicx2009@gmail.com
C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media 165

impossible to get an accurate estimation of the Rayleigh wave's polarization in the


steady mode by experiments[5, 6]. However, it is possible to measure the polarization
of the Rayleigh wave in a single pulse when it is separated from the longitudinal or
shear wave. Therefore, it is very meaningful to investigate the polarization of the
transient pulse Rayleigh wave related to the stress status.

1. Analytical model and numerical method

1.1 The acoustoelastic effect in solids

An ultrasonic wave through a stressed body would give rise to further stresses, and the
traditional theory of acoustoelasticity applied in the unstressed medium cannot be used
here. Instead, based on the deformation process, Duquennoy et al. defined a theory of
three states of a body, shown in Figure 1 [3, 4]. A solid body undergoes a series of
deformations from a stress free state to a static deformation or a dynamic deformation.
The position vector defines the position of a point in the natural state of zero stress
and zero strain. X defines the position of a point in the initial state when the body
undergoes a static deformation due to residual stress during the manufacturing
processes or due to the applied stresses. Similarly x defines the position in the final
states when a dynamic deformation such as the ultrasonic wave through the body. The
displacement of a point from one state to another can be described mathematically as:
u i ( ) X ; u f ( ) x ; u ( ) x X uf ui

Figure 1. Coordinates for material point in three states

Based on the three-state theory, the controlling equation of the acoustic wave in a
solid with initial stress can be written as

( ) X uX
2
i u I 2u I
IK JL + C IJKL
K
+f i
0 (1 NN ) (1)
J L t t 2
where i is the Cauchy stress tensor in the initial state, i the initial strain tensor and
0 the mass density in the natural state, is the acoustic damping coefficient, f is
the external force loading on the body. And for isotropic material the elastic constant
can be expressed as
166 C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media

C IJKL IJ KL + ( IK JL + IL JK ) + [( + v1 ) IJ KL
i
+ ( + v2 )( IK JL + IL JK )] NN + 2( + v2 )( IJi KL + KL
i
IJ ) (2)
i i i i
+ 2( + v3 )( JL + JK + IL + IK )
IK IL JK JL

where and are the well known Lame constants and vi (i 1, 2,3) are the
third-order elastic constants (TOE constants).

1.2 Numerical simulation method for Ultrasonic wave

According to FEM, Eq. (1) can be calculated by solving the discrete wave motion Eq.
(3)
[ M ]{U&&} + [C ]{U& } + [ K ]{U } {F } (3)
where [ M ] denotes the mass matrix, [C ] the damping matrix, [ K ] the stiffness
matrix, {U } the nodal displacement and {F } the force vector. Eq. (3) is rearranged
into the following equation by using the explicit integration method in time-domain [7]
{U& }t +t = {U& }t t 2t [ M ]1 [C ]{U& }t 2t [ M ]1 [ K ]{U }t + 2t [ M ]1 {F }t
(4)
{U& }t +t + {U& }t
{U }t +t = {U }t + t
2

1.3 Numerical method for EMAT Receiver

A typical configuration for an EMAT receiver of acoustic waves in a test-piece is


shown in Figure 2. It is usually composed of a static magnet and a set of coils (named
pick-up coil). When the ultrasonic wave occurred in the test-piece under the EMAT
receiver, the vibration by the ultrasonic wave would interact with the static magnetic
flux density of the static magnet B0 to yield transient eddy current J e in the metal.
Je e v B0 , v is the velocity of a particle, e is the conductivity of the metal.
And a micro current would be induced by the electromagnetic induction effect. The
magnetic flux density B0 of a permanent magnet can be calculated by equivalent
magnetic charge approach [7]. The most direct way to calculate the pick-up signals of
the EMAT receiver is to use Biot-Savart's law. According to Faraday's law, the
electromotive signal of the pick-up coil is produced in the pick-up coil by the flux
variation,
N
B ef
Ve
i =1 Si t
dS (5)

f
where B e is the reaction magnetic flux density at a position of the pick-up coil
produced by the transient eddy current, N is the number of turns of the pick-up coil,
and Si is a surface surrounded by the wire of i-th pick-up coil. Substituting
B ef A ef into Eq. (5) and applying Bio-Savat's law, we have,
C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media 167

(6)

where i is boundary of surface Si and r denotes the distance from a current


source point in the conductor to a point in the pick-up coil.

2. Simulation of the Acoustoelastic effect in LCR wave

2.1 Numerical model

In order to demonstrate the feasibility of this simulation method, the LCR waves
acoustoelastic effect related to the in-plane stress was simulated in a three-dimension
model, shown in Figure 2. As the piezoelectric (PZT) ultrasonic transducer is better as
generator; and EMAT is better as detector. A PZT angle transducer was used to
generate the LCR wave, and an EMAT probe was used to receive the time behavior of
the LCR wave signal for different initial stresses x , uniaxially directed along the x
axis. As shown in Table 1[6, 8], two kinds of material have been used here. Where, the
sensitivity constant kc is defined as a relative change in wave speed per unit change in
stress. When the longitudinal wave speed through a wedge is 2720 m/s, based on
Snells law, the incidence angle w should be set at about 26o to generate the LCR
wave in the model.

2.2 Numerical results and analysis

To give a preliminary judgment of the validity of this method, the ultrasonic pulse was
excited by one damped cycle of a 1 MHz sinusoidal force loading on a wedge with an
angle. The simulation result of the ultrasonic wave field in a cross section at a given
time is shown in Figure 3. The time behavior of the EMAT coil voltage is plotted in

z y
x

Figure 2. A pitch catch setup model in simulation

Table 1 Material parameter of the model


Material / kg m 3
 e / S m
1 / Pa / Pa v1 / Pa v2 / Pa v3 / Pa kc / MPa
Al Alloy 2719.0 38.2E6 49.1E9 26.0E9 379.0E9 198.0E9 80.0E9 8.79E 5
D54s
Rail Steel 7800.0 1.1E6 115.8E9 79.9E9 36.0E9 266.0E9 178.5E9 1.214E 5
168 C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media

Figure 3. Ultrasonic wave field excited by PZT

Figure 4. LCR wave received by EMAT

Figure 4. Compared to the Ultrasonic wave field, it can be seen that the first pulse in
Figure 4 is the receiving signal of LCR wave.
In Figure 5 and Figure 6, the close-up of the LCR wave signals for two initial
stresses in Al Alloy and rail steel is provided. The acoustoelastic effect with a small
phase delay induced by the stress can be observed. It can be seen that the variation in
LCR wave speed due to acoustoelastic effect in rail steel is much smaller than that in
Al alloy. The relative change in LCR wave speed for different initial stresses is
simulated, shown in Figure 7. It can be seen that the simulation results show a very
good agreement with the theoretical values.

Figure 5. Acoustoelastic effect in Al alloy


C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media 169

Figure 6. Acoustoelastic effect in rail steel

Figure 7. Relative change of LCR waves speed in Al Alloy and rail steel

3. Simulation of the polarization of transient Rayleigh wave

When Rayleigh wave travels across the surface of isotropic solids, the surface particles
move in ellipses in planes normal to surface and parallel to the direction of the
propagation. The polarization of Rayleigh wave is defined as the ratio between the
maximum in-plane, and the maximum out-of-plane displacement components. As
shown in Eq. (7), the sensitivity constant k p is defined as a relative change in
polarization per unit change in stress
0
k p (7)
0
where, the subscript 0 indicates the parameter at zero stress. In this work, the
polarization of pulse Rayleigh wave excited by the PZT transducer was investigated by
170 C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media

FEM. In order to make a comparison with the results based on the theory of steady
Rayleigh wave, three kinds of materials used in Junges paper were chosen here. As
shown in Table 2, k p' is the sensitivity constant in polarization of the Rayleigh wave
in steady mode[6].
Table 2 Material parameters
Material / kg m 3 / Pa / Pa v1 / Pa v2 / Pa v3 / Pa k 'p / MPa
Al Alloy 2719.0 49.1E9 26.0E9 379.0E9 198.0E9 80.0E9 6.673E 5
D54s
Al (99.3%) 2710.0 50.9E9 26.1E9 27.7E9 91.7E9 89.3E9 3.5809E 5
Rail Steel 7800.0 115.8E9 79.9E9 36.0E9 266.0E9 178.5E9 9.011E 6

The model is the same as in Figure 2. In order to yield a pure surface wave, the
o
angel of incidence w is set at about 70 . The simulation result of Rayleigh wave
field in a cross section at a given time is shown in Figure 8. Figure 9 presents the in-
plane and out-of-plane displacement components at point P on the surface of the
unstressed Al alloy model. As can be seen, the two components have a phase-shift
of / 2 . The polarization of Rayleigh wave is presented in Figure 10.

Fig. 8 Pulse Rayleigh wave field at a given time in Al alloy model

Figure 9. Rayleigh wave components at point P


C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media 171

Figure 10. Rayleigh wave polarization at point P

Figure 11. Polarization of Rayleigh wave for different initial stresses in Al alloy model

Figure 11 presents the polarization of Rayleigh wave for different initial stresses in
Al alloy model. It is observed that the polarization increases with increasing tension
and decrease with increasing compression. Figure 12 presents the relative change of
polarization for different initial uniaxial stresses x . It shows that the relative change
in polarization of pulse Rayleigh wave is much larger than the reference values based
on the theory of steady Rayleigh wave.

4. Conclusion

In this paper, a numerical code is developed for numerical simulation of acoustoelatic


effect in pre-stressed media. The possibility of assessing the stress status, by using the
EMAT for precise measuring the LCR wave, was investigated. The numerical results
showed a very good agreement with the theoretical values. Finally, the relationship
between the polarization of transient Rayleigh wave to the state of stress was also
predicted. The numerical results showed that the relative change in polarization of
transient Rayleigh wave have a much larger value than that of steady Rayleigh wave
due to the initial stress.
172 C. Pei and K. Demachi / Numerical Simulation of Acoustoelastic Effect in Pre-Stressed Media

Figure 12. Comparison of acoustoelastic effect of pulse Rayleigh wave with reference values of steady
Rayleigh wave

Acknowledgement

This work was supported by China Scholarship Council (CSC).

References

[1] Shamachary Sathish, Thomas J. Moarn, Residual stress measurement with focused acoustic wave,
Materials Science and Engineering A 399 (2005), 84 91.
[2] Development of non contact stress measurement system during tensile testing using the
electromagnetic acoustic transducer, NDT&E 39 (2006), 299 303.
[3] Michael Junge, Jianmin Qu, Relationship between Rayleigh wave polarization and state of stress,
Ultrasonics 44 (2006), 233 237.
[4] Dequennoy M. Ouaftouh M., Ultrsonic evaluation of stresses in orthotropic materials using Rayleigh
waves, NDT&E 32 (1999), 189 199.
[5] Shailesh Gokhale, Determination of applied stress in rails using the acoustoelastic effect of ultrasonic
waves, Master's thesis, Texas A&M University, Dec. 2007.
[6] Michael Junge, Measurement of applied stress using polarization of Rayleigh waves. Masters thesis,
School of Civil and Environmental Engineering, Georgia Institute of Technology, August, 2003.
[7] Cuixiang Pei, Zhenmao Chen, Development of simulation method for EMAT signals and application to
TBC inspection, Int. J. Appl. Electromagn. Mech., Vol.33, 2010.
[8] R.T. Smith, R. Stern, and R.W.B Stephens. Third order elastic module of polycrystalline metals from
ultrasonic velocity measurements. Journal of Acoustical Society of American, 40(5):1002 1008, 1966.
[9] Don E. Bray, Wei Tang, Subsurface stress evaluation in steel plates and bars using the LCR ultrasonic
wave, Nuclear Engineering and Design, 207:231 240, 2001.
Electromagnetic Nondestructive Evaluation (XIV) 173
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-173

BEMLAB Universal, Open source,


Boundary Element Method library applied
in Micro-Electro-Mechanical Systems1
Pawe WIELEBA a,2 and Jan SIKORA b,a
a
Department of Measurement and Diagnostic Systems, Electrotechnical Institute,
ul. Pozaryskiego 28, 04-703 Warszawa, Poland
b
Department of Electronics, Lublin University of Technology,
ul. Nadbystrzycka 38a, 20-618 Lublin, Poland

Abstract. BEMLAB [1] is the open source project implementing Boundary Ele-
ment Method (BEM) [2,3] comprehensively. BEMLAB binary packages and the
source code are distributed under GNU LGPL (Lesser General Public License) li-
cense terms. The project provides universal library and the reference application,
which is the easiest way for solving problems using BEM. The article presents ex-
ample BEMLAB application in modelling comb capacitors used in Micro-Electro-
Mechanical Systems (MEMS). Finally capacitance will be calculated using BEM
results. Particularly Asymptotic Boundary Conditions (ABC) will be used to model
external space of the capacitor (the geometry of innite half-lane) as an internal
BEM problem. This approach will allow to calculate total capacitance with dis-
persed capacitance included. Results will be compared with the simplied model.
Keywords. BEMLAB, BEM, Boundary Element Method, MEMS, Micro-Electro-
Mechanical Systems, Comb capacitors, Comb actuators, Asymptotic Boundary
Conditions, Open source, Objectivity, Multi-threading

Introduction

Micro Electro-Mechanical Systems (MEMS) are very small devices which dimensions
vary in the range from 0.1m to 0.1mm. They exchange energy either from electri-
cal to other type for example mechanical, or mechanical to electrical. The most com-
monly known MEMS devices are ink-jet printers heads or accelerometers used in airbags.
MEMS design process involves two modelling types electrostatic and mechanical. This
article incorporates electrostatic one and will concentrate on comb capacitors used in
MEMS devices. The comb capacitors design stage problem requires to calculate capac-
itance C [4]. One method of calculating capacitance is usage of the simplied model
based on the ideal parallel-plate capacitor. However this approach does not incorporate
dispersed capacitances appearing in the corners and thus results are not precise enough.

1 http://bemlab.org/
2 E-mail: p.wieleba@iem.pw.edu.pl
174 P. Wieleba and J. Sikora / BEMLAB

The most precise modelling method which can be used for associating real physical
properties and processes is the eld method. Field method applied in electromechanics
satises the Laplace equation. The most common numerical method of solving Laplace
equation is the Finite Element Method (FEM) [5,6]. Many open source and commer-
cial packages, which can be used to solve problems using FEM exist [7]. However this
method is a domain one and requires discretization of the whole domain. While estimat-
ing condenser capacitance it is a worthless overhead. Therefore it is tempting to use the
boundary method. However, there was no universal package which made it possible until
BEMLAB [8,1] came into existence. Now it is possible and fairly easy.

1. BEMLAB open source and objective project

BEMLAB [1] is the open source project implementing Boundary Element Method
(BEM) [2,3] comprehensively. BEMLAB binary packages and the source code are dis-
tributed under GNU LGPL (Lesser General Public License) license terms. The project
provides universal library and the reference application, which is the easiest way for
solving problems using BEM. There are also auxiliary programs provided to facilitate
engineers tasks. A very important factor for the end user is the input/output data for-
mat, therefore the title library uses format compatible with Matlab M-les, Octave script
les [9] and Scilab sci-les [10]. The project is objective and is developed using Unied
Modeling Language. BEMLAB is implemented in C++. BEMLAB uses multi-threading
(MT) to speed up calculations on multi-processor and multi-core platforms nowadays
CPUs have at least 2 cores. Threads introduced in C++0x specication are used. As an
example GNU Compiler 4.4 or newer provides C++0x API.

2. Boundary Element Method

BEM is the numerical method designed for solving physical problems described by Par-
tial Differential Equations (PDE) but expressed as Boundary Integral Equation (BIE):
  
G
c + d = Gd + f Gd (1)
n n

where: problem domain; domain boundary; eld function potential; G


Green function, a so-called fundamental solution; n boundary element normal vector
directed outwards the problem domain ; n
normal derivative; f domain (source)
function; c coefcient removing or restricting the singularity from the primitive func-
tion of PDE solution.
Field modelling of comb capacitors requires Laplace PDE to be solved in the prob-
lem domain :

2 (x) = 0 (2)

BIE (1) can be used to solve Laplace equation if proper Green function is applied. Green
function (fundamental solution) for Laplace equation in 2D space is:
P. Wieleba and J. Sikora / BEMLAB 175

1 1
G(R) = ln (3)
2 R
where R is the distance between the observation point for which the BIE equation is be-
ing set and the integration point of the element being integrated using the Gauss Quadra-
ture.

3. Comb capacitors in Micro-Electro-Mechanical Systems

Micro-Electro-Mechanical Systems use comb capacitors for detecting mechanical forces


(conversion of mechanical energy into electrical) or for driving mechanical structures
(conversion of electrical energy into mechanical). The structures using comb capaci-
tors for driving mechanical parts are named comb actuators. An example MEMS device
schema is presented in gure 1. It is a micro device of the scanning mirror, which consists
of a mirror and four comb capacitor banks with two nger pairs each.

Figure 1. Scanning micro mirror with vertical comb capacitors.

Comb capacitors may have various shapes, dimensions and types. The article
presents vertical comb capacitors with angular actuation. They are called Angular Verti-
cal Comb (AVC) capacitors. Angular vertical comb capacitor nger in a tilted position
was presented in gure 2 on the left side. Whereas on the right side it was presented in the
rest position from the top perspective. Example dimensions of the AVC capacitor used
Table 1. Angular Vertical Comb condenser ngers dimensions.
Description Symbol Dimension [m]
Finger length l 150
Finger width w 4
Finger height h 40
Gap between ngers d 3
Number of ngers in comb capacitor N 12

for driving micro mirror are gathered in table 1 based on [11]. The capacitor is powered
with the voltage of 20V, where potential of -10V is applied to one electrode and 10V to
the second one.
176 P. Wieleba and J. Sikora / BEMLAB

Figure 2. Angular Vertical Comb capacitor nger in a tilted position from the side perspective (left) and in a
rest position from the top perspective (right).

4. Vertical Comb capacitor eld model using Asymptotic Boundary Conditions

In this section the problem for the AVC capacitor is dened and solved in order to calcu-
late capacitance C. Field model is created for the boundary element method (BEM). All
calculations are done using BEMLAB software.
Firstly the geometry have to be dened. Here 2D model is considered. Gap between
electrodes presented in gure 3 is to be modeled. The model was dened in the way so

Figure 3. Gap between AVC capacitor electrodes.

dispersed capacities are included in calculations. Due to the existence of the open space,
an articial boundary (1t) have to be introduced. The gap is the domain of the problem
P. Wieleba and J. Sikora / BEMLAB 177

marked by (1) . (1) is the boundary around the examined domain (1) . Only nger
halves are modeled because the whole structure of the comb capacitor is repeatable, as
it consists of N nger pairs. Therefore boundary conditions of the second type n = 0
are applied on the left and right side of the presented model. The vertical dashed lines on
the left and right of the gure 3 are placed in the center of particular ngers. The hori-
zontal dashed line at the bottom of the gure also cuts particular ngers into two halves.
Therefore zero boundary conditions of the second type (Neumann Boundary Conditions)

n = 0 are also applied. Electrodes are conductors, therefore Dirichlet Boundary Con-
ditions (boundary conditions of the rst type) are applied on the boundary of electrodes.
On the left nger (yellow) potential = V is applied and on the right nger (right)
potential = +V is applied.
Almost all boundary conditions of the internal problem were dened. The only prob-
lematic boundary is (1t) . The width of the (1t) is a as marked at the gure 3 and is de-
ned by d(t) length away from the electrodes. The (1t) boundary placed in the innity:

d(t) (4)

forms a half-lane beginning on the boundary of electrodes. Therefore special boundary


conditions has to be introduced on (1t) . Setting of zero Dirichlet Boundary Conditions
is a one possibility, but the boundary should be moved far away towards the innity:

d(t)  a (5)

which will increase the domain size and number of boundary elements used. Another one
which will be used in the model, is using of Asymptotic Boundary Conditions (ABC)
[12, chapter 4].

5. Asymptotic Boundary Conditions

Asymptotic Boundary Conditions [12, pp. 52-73] for the half-lane are in the following
form:



+ =0 (6)
y a

where a is the width of the half-lane. According to the BEMLAB software, the follow-
ing notation has been used:


= +0
y a
(7)

= m + n
y

According to dimensions gathered in the table 1, parameters m and n of the


ABCs (7) for the model presented in gure 2 are calculated below:
178 P. Wieleba and J. Sikora / BEMLAB

w w
a= +d+ =d+w =3+4=7
2 2
(8)
m = = = 0, 44879
a 7
n=0

6. Field model calculations using BEMLAB

When appropriate model denition is created, mesh of the geometry has to be generated.
Figure 4 presented the boundary (1) mesh of the model dened in gure 2. It was
0 5.75 11.5 17.2 23
-3.5 6 5

-1.75
8 7

1 2
1.75
Z Y
3.5 X 3 4

Figure 4. Boundary mesh of the AVC capacitor air gap between electrodes for the problem geometry dened
in gure 3 with d(t) = 3.

generated using Gmsh software [13]. Corner geometric nodes has been marked with their
numbers from 1 to 8. The modeled problem is 2D, therefore z coordinate equals 0.
Boundary element method requires boundary elements normal vector n to be di-
rected outwards the analyzed domain . Therefore special attention has to be brought to
the process of generating boundary mesh and after creating one its correctness has to be
checked. BEMLAB software has the functionality of calculating normal vectors for any

3.5

1.5
x

1.5

3.5

0 20 23
y

Figure 5. Boundary elements normal vectors 


n.

dened input mesh. Figure 5 presents normal vectors drawn against the generated mesh.
As can be seen, blue normal vectors n are directed outwards the domain (1) enclosed
P. Wieleba and J. Sikora / BEMLAB 179

by the boundary domain (1) marked with the solid line. The following BEMLAB com-
mand was used to calculate normal vectors of mesh dened in the input le dened by
parameter -i:
% obem_solve -i comb_cap_mesh.m -m 7 -o normal_vecs.m
Results are written to the output le dened by the command parameter -o.
The mesh and boundary conditions described in the previous section are gathered in
one le:
comb_cap_mesh.m
Values of potential and its normal derivative
n are the direct result of BEM. To start
BEM calculations and obtain and n the following command has to be issued:
% obem_solve -i comb_cap_mesh.m -m 1234 -o solution.m
To visualize layout of potential along the domain (1) , values of potential in
internal nodes have to be calculated. This stage is optional and not required to calculate
capacitance C, but is helpful to detect big calculation errors. This stage can be processed
when potential and its normal derivative n are known on the whole boundary . To
proceed calculations the following shell command has to be issued:
% obem_solve -i comb_cap_mesh.m -S solution.m -m 5 \\
-I comb_cap_internal_nodes.m -o internal_potential.m
where -i denes the input le with the input mesh with boundary conditions and
material parameters (as in the previous paragraph), -S denes the input le with the
solution on the boundary (potential and its normal derivative n ), -I denes the
input le with the internal nodes coordinates, -o denes the output le with calculated
internal potential .

Figure 6. Layout of potential inside the AVC capacitors air gap dened in the gure 4.

Figure 6 presents the layout of calculated potential inside the air gap (domain (1) )
calculated using the previous command.
180 P. Wieleba and J. Sikora / BEMLAB

7. Capacitance calculations using BEM results

Capacitance is calculated using the following relation:


 
Q C
C= [F] = (9)
U V

where Q charge gathered on one electrode, U voltage between two electrodes of the
capacitor.
The charge Q can be calculated using the following integral:
   

Q = d = Dn d = En d = d (10)
n

where the surface charge density, Dn electric displacement eld , En electric
 
eld, electric permittivity, and Dn = Dn .
Finally the capacitance can be written in the following form:
 
  
1 
C =  d (11)
U n 

Normal potential derivative n on the boundary is the direct result of BEM, there-
fore it can be directly inserted into equation (11). Figure 7 presents the layout of poten-
tial normal derivative
n on the boundary of the right electrode (starting from node no.
1 to node no. 3 of mesh from gure 4). Potential normal derivative n , presented on the

12

10
potential normal derivative

0
0 10 20 30 40 50 60 70
element number

Figure 7. Layout of potential normal derivative on one electrode.

vertical axis, matches the number of the element presented on the horizontal axis.
Firstly the charge Q was calculated for the half of one nger (as presented in the
model):

Q = 144.863 (12)

The integration was done using BEMLAB software by issuing the following command:
P. Wieleba and J. Sikora / BEMLAB 181

% obem_solve -i int_electrode.m -m 8
integratedValue=[ 144.863 ];
where -i denes the input le with geometry to be integrated and values of integrated
function in the geometric nodes.
Then the capacitance per unit length for half of one nger, using the eld model with
ABC applied, was calculated as follows:
Q 144.863
CI(half) = = = 7.24 [F]
U 20 (13)
= 7.24 106 8.854187817 1012 = 64.104 [aF]

The capacitance between two electrodes of one nger (2 halves) of the comb capac-
itor with the length l = 150 (the z dimension) and the gap width d = 3 is as follows:

C (nger eld) = 2 (l d) CI = 2 (150 3) 64.104 = 18.847 [fF] (14)

The total capacitance C (total) for the whole comb capacitor with N = 12 ngers is
2N times bigger than capacitance between two electrodes of one nger, as one nger has
two sides (vertical). Therefore the total capacitance can be calculated as follows:

C (total) = N C (nger) = 12 18.847 = 226.164 [fF] (15)

8. Vertical comb capacitor simplied model

In the previous section calculations using the eld model were proceeded. In this section
the calculation of capacitance using the simplied model will be presented for the com-
parative purposes. The model for half of the nger as presented in previous section and
on gure 2, consists of one parallel plate capacitor, which can be calculated as follows:
C h(l d) 40(150 3)
C (nger simple) = = = 106 = 17.354 [fF] (16)
S d 3

9. Summary

Universal, open source, objective and multi-threaded BEMLAB software can be suc-
cessively used for modelling MEMS devices. Field modelling using boundary element
method includes dispersed capacitance which occurs in comb capacitors, while simpli-
ed models not. The difference between calculated capacitance values using eld and
simplied model is 9%:

C (nger eld) = 1.09 C (nger simple) (17)

BEMLAB software also allows three dimensional modelling for more complicated struc-
tures or in a tilted position, which cannot be modeled in 2D. BEMLAB input/output text
le format implementation compatible with Matlab M-les, GNU Octave script les and
Scilab .sci les makes cooperation with that software straight forward. Moreover engi-
neering tools are provided to simplify cooperation with mesh formats provided by Gmsh
[13] or Netgen [14].
182 P. Wieleba and J. Sikora / BEMLAB

References

[1] P. Wieleba and J. Sikora: Open Source BEM Library, Advances in Engineering Software 40 (8), 564569,
2009.
[2] L. C. Wrobel and M. H. Aliabadi: The boundary element method, vol. 1 and 2, Wiley, 2002.
[3] J. Sikora: Boundary Element Method for Impedance and Optical Tomography, Ocyna Wydawnicza
Politechniki Warszawskiej, 2007.
[4] G. Of, M. Kaltenbacher and O. Steinbach: Fast multipole boundary element method for electrostatic eld
computations, The International Journal for Computation and Mathematics in Electrical and Electronic
Engineering 28 (2), 304-319, 2009.
[5] O. C. Zienkiewicz and R. L. Taylor: The nite element method, vol. 1, Butterworth-Heinemann, 2000.
[6] R. Sikora: Teoria pola elektromagnetycznego, Wydawnictwa Naukowo-Techniczne, 1997. In Polish.
[7] J. Mackerle: Object-oriented programming in FEM and BEM: a bibliography (1990-2003), Advances
in Engineering Software, 35, 325-336, 2004.
[8] BEMLAB homepage, http://bemlab.org/
[9] Octave homepage: http://www.octave.org/
[10] Scilab homepage: http://www.scilab.org/
[11] W. Piyawattanametha, P. R. Patterson, D. Hah, H. Toshiyoshi and Ming C. Wu: A Surface and Bulk
Micromachined Angular Vertical Combdrive for Scanning Micromirrors, IEEE Optical Fiber Commu-
nications Conference vol. 1, 251-253, 2003.
[12] S. Gratkowski: Asymptotyczne warunki brzegowe dla stacjonarnych zagadnien elektromagnetycznych
w obszarach nieograniczonych algorytmy metody elementw skonczonych, Wydawnictwo Uczelniane
Zachodniopomorskiego Uniwersytetu Technologicznego w Szczecinie, 2009. In Polish.
[13] Gmsh homepage: http://geuz.org/gmsh/
[14] Netgen homepage: http://www.hpfem.jku.at/netgen/
Electromagnetic Nondestructive Evaluation (XIV) 183
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-183

Use of Half-Analytical Method for the


Detection of Defects in Diet Pulses
Hassane MOHELLEBIa,1, Ferroudja BOUALIa, Mouloud FELIACHIb
a
Laboratoire de Gnie Electrique, Universit de Tizi-Ouzou, BP 17 RP, ALGERIA
b
IREENA Laboratory, IUT-CRTT, Nantes University, 44600 Saint-Nazaire, France

Abstract. The present work concerns the development of pulsed eddy currents
non destructive testing techniques applied to defects analysis. The impulse currents
are produced with capacitor discharge bank [1], [2]. Coupled circuits method
employed is based on the mutual inductances calculation [3] and associated to
Kirchhoff laws yields to a real algebraic equation system which is solved at each
time step. The model developed is used to the defects analysis of a cylindrical
device [4].

Keywords. Half analytical method, eddy currents inspection, multi defects


analysis , diet pulses.

Introduction

The eddy currents sensors are used in diverse industrial domains which concern the
evaluation of the physical properties of materials, the load dimensions, defect detection
in a piece etc.
Pulsed eddy current (PEC) sensing is a new emerging technique and has gained
considerable research attention in recent years [5]. Their first applications have been
developed for the measure of material's thickness. The possibility to inject an important
energy in short time with rich spectrum of low frequencies have induced an increase in
use of them in the detection of multi-defects in pieces. This technique is used to inspect
the riveted assembly, the slants and the objects of irregular surfaces, it holds the
voltages of becoming the primary means of detecting corrosion in multilayered
structures and to inspect, characterize the coatings [6],[8],[9],[10].
The present work deals with the study of the differential eddy currents sensor in
pulsed diet by using a half-analytical method coupled with an equivalent electric circuit
[2], [11]. The impulse currents technique is chosen instead of multi-frequency study
which induces some difficulties when the number of frequencies exceeds four [7],[10].
In this case the use of impulse currents method present more advantages. The
application of the method concerns the detection of two defects realized in a
conducting cylinder scanned with differential probe. A cylinder with two layers is
investigated too and a comparison with existing data is realized.

1
Corresponding Autor: Hassane MOHELLEBI, aLaboratoire de Gnie Electrique, Universit de Tizi Ouzou,
BP 17 RP, ALGERIA ; e mail :mohellebi@yahoo.fr
184 H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses

1. Coupled Circuits Method

The half-analytical method used is the coupled circuits one based on a discretization of
the inductor and load sections in (r,z) plane. The elementary sections constructed
permit one to define elementary coils with associated elementary parameters (figure 1).
Electrical parameters of the device as mutual and appropriate inductances exploits an
integral calculus.

Ub1 Ub2

Rb1 Lb1 Rb2 Lb2

Rb1n Lb1n Rb1n Lb2n

Inductor

Rc1 Lc1

Rcm L cm

Short-circuited load

Figure 1. Coupled Circuits Model of device

Ub1, Ub2 are electric voltages of differential probe. Indices n and m are the maximum
numbers of elementary coils of the inductor and short-circuited load.
This process allows expressing the resistance of every elementary turn, the self-
inductance and the mutual inductances between the various turns [12]. The electric
equations of the load and the inductor are given by Eqs (1) and (2) as following:

dI m dI N dI
i + j
0 = Ri I i + L M ij + M ik k (1)
i dt j =1 dt k =1 dt
j i
dI N
dI m dI
k + q (2)
U = Rk I k + L M kq + M ki i
k k dt q =1 dt i =1 dt
qk
H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses 185

Ri, Li and Ii; Rk, Lk and Ik are the resistance, the constant inductance and the current.
Indices i and k are related to load and inductor respectively.
Mij: mutual inductance between both elements i and j of the load.
Mik: mutual inductance between load and inductor elements.
Ij: current in the element j of the load; Ik: current in the element k of the inductor.
Mkq: mutual inductance between both elements k and q of the inductor;
Mki: mutual inductance between the element k of the inductor and the element i of the
load.
Ik: current traversing the element k of the inductor. Iq: current traversing the element q
of the inductor.

1.1. Model parameters computation

Several methods have been proposed for the calculation of the parameters of the
models. We can classify these in two big classes [4], [3]:
- Methods based on the magnetic vector potential
- Methods based on the mutual inductances calculation

1.1.1 Mutual inductance

The mutual inductance betweentwo turns is given by [12]:

2 2 (3)
M = a .b ( k ) K ( k ) E( k )
k k

4. a.b (4)
k2 =
(a + b)2 +h2

: Magnetic permeability; a: radius of the turn 1; b radius of the turn 2 ; h:


distance between the two turns; K(k) and E(k) represent the elliptic integrals of first
and second species respectively;
M2

c a
M1
b
b
h

Figure 2. Distance between two turns

1.1.2 Appropriate inductance


While joining the two turns, we get the expression of the self-inductance given by [12]:
186 H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses

8.a (5)
L = a ln 1.75
Rc

Rc : conductor radius

1.1.3. Resistance
The turn resistance of elementary turn coil is given by Eq. (6):

"
R= (6)
S

: Material resistivity, S: Cross-section; " : Middle length of the turn.

2. Equivalent electrical circuit Model

2.1. Electrical Circuit Model

The relative electric model to the diet pulses is comprised of the equivalent circuit of
the sensor-load coupled to the energizing circuit composed of a capacity and a
resistance. The equivalent electric circuit of the system is given by the following
Figure 3 [2]:

ic(t) Rext

Sensor system
Lb

+ Vc(t)
C
Rb

Figure 3. Equivalent electrical circuit

The electric equation modelling this system is given by Eq. (7):

dic (t )
V C (t) (Rext + Rb )ic (t) Lb =0 (7)
dt

dvc (t )
iC (t ) = C (8)
dt

Rb and Lb are the resistance and the inductance of the system load-inductor; Rext:
external resistance; C: capacity.
H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses 187

2.2. Analytical solution of electrical equation

The resolution of Eq. (7) combined with Eq. (8) yields to the following solution:

1 2 (9)
ic ( t ) = C V0 ( )( e t e t )1 2

1 2

(R ext + R eq ) (R + R eq ) 4 (Rext + Req ) + (R + Req ) 4


2 L eq 2 Leq
With: ext
C
; ext
C
1 = 2 =
2 Leq 2 Leq

1 , 2 are constants values. Initial time conditions v c (t = 0) = V0 , i c (t = 0) = 0

3. Application and Results

The application considered in the present work is the one studied in [5]. It makes use of
a differential probe used for the tubing inspection generator. The defects realized on the
conducting cylinder are axisymmetric with resistivity W= 10-6[.m]. The physical and
geometrical characteristics corresponding to differential probe are: coil width: 1.75*10-
3
[m], distance between coils: 0.5*10-3, high in z direction: 0.75*10-3 [m], internal
radius: 7.5*10-3 [m], magnetic permeability: 4X*10-7 [H/m] and the electrical
conductivity: 5.59*10+7 [.m]-1(Figure 4).
z
J +J
Inductor turns

Cylindrical object r

+J J

Figure 4. Differential sensor with cylindrical tube inspection

3. 1 Analysis of the current response of several defects

In this section we consider a healthy load (flawless) and loads with different types of
defects (internal, in depth and external defect). The electrical circuit parameters used
are: C = 10-7 F, Rext = 0.55 Y, Req= 1.6*10-3 Y, Leq= 1.0054*10-8 H, V0 = 1V.
188 H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses

0.04 0.045

0.035 0.04

0.035
0.03

0.03
0.025
0.025
z [m ]

z [m ]
0.02
0.02
0.015
0.015

0.01
0.01

0.005 0.005

0 0
75 8 8.5 9 9.5 10 10.5 7.5 8 8.5 9 9.5 10 10.5
r[m] 3 r[m] 3
x 10 x 10

0.045
0.045

0.04 0.04

0.035 0.035

0.03 0.03

0.025 0.025
z [m ]
z [m ]

0.02 0.02

0.015 0.015

0.01 0.01

0.005 0.005

0 0
7.5 8 8.5 9 9.5 10 10.5 7.5 8 8.5 9 9.5 10 10.5
r[m] 3 r[m] 3
x 10 x 10

Figure 5. Solving domain for a load presenting different defects

Figures 6 and Figure 7 represent the variation of the current for different defect types:
0.07 0.1
Defect thickness 20%
Healthy load
0.06 Defect thickness 40%
Load with external defect 0.08 Defect thickness 80%
Load with defect in depth
0.05 Load with internal defect
0.06
0.04
C urrent[A ]

Current[A ]

0.03 0.04

0.02
0.02

0.01

0
0

-0.01 -0.02
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5
Time[s] 7 Time[s] 7
x 10 x 10
Figure 6. Variation of the current for different Figure 7. Variation of the current with defect
defect types thickness
H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses 189

While examining the results given in Figure 6 we notice that the current increase
every time the sensor is near to the defect. The internal defect with maximum value of
current Im constitute the more detected one because it is most near the sensor, then
comes the defect in depth and finally the external defect.
In Figure 7 there is represented the current response of the differential probe when
considering different defect thickness (20%, 40%, 80%). In this case we notice that the
parameter which varies more is the crest value of the current that increases
progressively as the defect thickness increases.

3.2. Presence of two defects in the load

In this application we consider a load with a combined defect, internal defect (40 %
width) and an external defect (20 % width) (Figure 8). Figure 9 represents the variation
of the differential current i(t) in the case of the load with an internal defect and in the
case of the load with a combined defect, for C=10-6 F and R=0.2 .
0.045 0.3
Combined defect
0.04 Internal defect
0.25

0.035
0.2
0.03

0.15
C u rre n t[ A ]

0.025
z [m ]

0.02 0.1

0.015
0.05 T0
0.01

0
0.005

0 -0.05
7.5 8 8.5 9 9.5 10 10.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
r[m] 3 Time[s] 6
x 10 x 10

Figure 8. Solving domain with combined defects Figure 9. Differential currents variations

Table 1. Internal and combined defects characteristics

Im[A] T0[s]
1
Internal defect 1.846 10 4.2129 10 7

Combined defect 2.075 10 1 4.4925 10 7

In Table 1, Im represents the maximum value of the current which results from
difference between two currents turns; T0 is the time corresponding to current when it
vanishes.
According to the results presented in Figure 9, we note a difference between the
differential currents i(t) in the case of the internal defect and the same unknown
obtained in the case of the combined defect. The combined defect supplies amplitude
Im with a time of passage by zero T0 more important. We can thus conclude that in spite
of the internal defect is important; it does not hide the presence of the external defect.
190 H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses

3.3. Case of a load presenting two layers

In this application we consider a load with two healthy layers and two layers with
defects.
0.04 0.04

0.035 0.035

0.03 0.03

0.025 0.025
h a u te u r[m ]

h a u t e u r[ m ]
0.02 0.02

0.015 0.015

0.01 0.01

0.005 0.005

0 0
7.5 8 8.5 9 9.5 10 10.5 7.5 8 8.5 9 9.5 10 10.5
rayon[m] 3 rayon[m] 3
x 10 x 10

Figure 10. Load with two healthy layers Figure 11. Load with two layers with defects

The Figure 12 represents the variation of the induced electromotive force (e. m. f) u(t)
in an elementary spire of the inductor for a flawless load, a load with defect, a load
with two layers without defects and a load with two layers with defects.

0.12 1
load w th defect load with defect [1]
0.1 08 healthy load [1]
healthy load
load with defects
load w th two healthy layers
06 healthy load
0.08 load w th two layers with defects
Relative values of e m f

04
0.06
02
0.04
e.m.f [V]

0
0.02
02
0
04

-0.02
06

-0.04 08
0 01 02 03 04 05 06 07 08 09 1
time[s]
-0.06
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time[s] 7
x 10

Figure 12. Induced e.m.f u(t) variation in an Figure 13. Comparison with results
given in [1]
elementary coil of the inductor

This relative value of induced e.m.f (Figure 13) increases with increase of the
layers number and also with the number of defects. These results with relative value
variation with time allow to the same conclusion than the one given in [1].

4. Conclusion

The half-analytic model developed in the current study is applied to the non destructive
eddy current testing device. The study is undertaken in pulsed diet and a response of
current is investigated for several defects nature of conducting tube. The case of the
H. Mohellebi et al. / Use of Half-Analytical Method for the Detection of Defects in Diet Pulses 191

presence of two defects is considered and analysed. A cylinder with two layers is also
studied and the results obtained are in a good agreement according to existing results.
The model developed has shown its capability to analyse multiple defects in a
conducting cylinder or multiple layers one with defect. The advantages of the model
are less time consuming, which is very interesting compared to numerical models
because in this case the air region was note discretized, and its capability to take into
account of the movement of the sensor without changes in domain discretization.

References

[1] N. Boucherou, M. Zergoug, A. Hammouda, G. Kamel, H. Boudjelal, A. Boutaghane, O. Bourdjam,


Conducting materials evaluation using pulsed eddy currents, 18th Frenchs congress of mechanics,
Grenoble (2007), pp. 27 31.
[2] H. Mohellebi, F. Bouali, A. Diche Use of half analytical method for non destructive pulsed eddy current
study, Colloquium of induction, April 13-14, Laghouat (2009), Algeria.
[3] J. Ndinga, Modeling of magneto thermal phenomena : application to induction heating, PhD Thesis, I.N.P
Grenoble, France 1987.
[4] D. Delage, R. Ernst, Courant distribution prediction in axisymmetric inductor dedicated to MF and HF
induction heating. RGE (1984), Vol. 4, no.84, pp.225 230.
[5] J. L. Thomas, Simplified model of eddy current non destructive of generator tubing , Engineer Training
report, CEA, France, 1998.
[6] Young Kil Shin, Dong_Myung Choi, Young Joo Kim, Seung Seok Lee, Signal characteristics of
differential pulsed eddy current sensors in the evaluation of plate thickness, NDT&E International
(2009), pp. 215-221
[7] B. Maouche, R. Alkama, M. Feliachi, Semi analytical calculation of the impedance of a differential
sensor for eddy current non destructive testing, NDT&E International, (2009), pp. 573 580
[8] Y. BinFeng, L. FeiLu, H. Dan, Reseach on edge identification of a defect using pulsed eddy current based
on principal component analysis, NDT&E International, (2007), .pp. 294 299.
[9] T. Chen, G. Y. Tian, A. Sophian, P. W. Que, Feature extraction and selection for defect classification of
pulsed eddy current NDT, NDT&E International, (2008), pp. 467 476.
[10] Deeds W.E., A comparison of multiple frequency and pulsed eddy current techniques, Proc. of the 5th
annual review of progress in quantitative NDE, Lajolla, USA (1978), pp. 117 1 19.
Fields, Leuven Belgium (1994), 18 20.
[11] M. Abdellah, A. Hammel, A. Zaoui, H. Mohellebi, Study of the feeding mode effect on the linear
induction launcher (LIL) performances, in Proc. Wseas Transactions On Systems, (2005) .Issue 10, Vol.
4, pp. 1798 1804.
[12] F. Gardiol, Trait lectricit, Electromagntics, Volume III, Presse Polytechniques et Universitaires
Romandes, Suisse, 1996.
192 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-192

Imperialist Competitive Algorithm Applied


to Eddy Current Nondestructive Evaluation
Ammar HAMELa,1 Hassane MOHELLEBIb and Mouloud FELIACHIc
a
Electrical Engineering Laboratory, Departement of Electrical Engineering,
University of Bjaa, Algeria hamelkane@yahoo.fr
b
Electrical Engineering Laboratory, Departement of Electrical Engineering,
University of Tizi-Ouzou, Algeria b mohellebi@yahoo.fr d hocini.farid@yahoo.fr
c
IREENA Laboratory, IUT of Saint-Nazaire, University of Nantes, France
mouloud.feliachi@univ-nantes.fr

Abstract. The imperialist competitive algorithm (ICA) is a new global search


strategy inspired by the socio political process of Imperialistic competition. ICA
has higher performances such as faster convergence and better global minimum
achievement [1]. Considering its undeniable effectiveness, its limitations and
applicability in various fields, ICA is currently explored. In this paper we adopt
ICA to deal with the inverse problem applied in eddy current non destructive
evaluation. We present test results and a comparison between ICA and the particle
swarm optimisation algorithm (PSO).

Keywords. Non destructive evaluation, eddy currents, inverse problem, imperialist


competitive algorithm.

Introduction

To improve time necessary to the inversion in eddy current non-destructive evaluation,


the specialists tackle the problem by considering two principal aspects, in the event the
development of direct models reliable and applicable to cases of complex geometry and
especially which are fast on the one hand, and the implementation and adaptation of
sure and fast algorithms of inversion on the other hand. In this work we are interested
rather in the second aspect in which the methods which are already used can be shared
in two groups. The first group includes what is called the non-phenomenological
methods and which are based on the techniques of signal processing [1], [2]. If they
are fast, these methods have nevertheless a major disadvantage which is the need for an
enormous data base of signals and various forms with a very good precision. Increasing
interest has been granted in recent years, with the use of stochastic optimization
methods that guarantee the convergence to the global optimum of the function to
optimize. Several algorithms based on these methods are used in eddy current non-
destructive evaluation, namely genetic algorithms, simulated annealing, taboo search
and recently particle swarm optimization [3]- [5]

1
Ammar HAMEL, Electrical Engineering Laboratory, Department of Electrical Engineering,
University of Bjaa, Algeria hamelkane@yahoo.fr.
A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation 193

The main contribution of this paper is an extension to eddy current non-destructive


evaluation of a new inversion algorithm called imperialist competitive algorithm (ICA).
ICA is an iterative evolutionary computation technique and a socio-politically
motivated global search strategy that has recently been introduced for dealing with
different optimization tasks. The proposed method is applied to a simple
axisymmetrical conducting tube with a groove made arbitrarily. The aim is to
reconstruct sizes of the groove by combining the finite element method for the
computation of eddy current with the optimization algorithm in the inverse process.

1. 2D Electromagnetic equation

Starting from an impedance measurement, the aim is to reconstitute the profile of an


axi-symmetrical groove practiced inside an aluminum tube inspected by a differential
probe. The solution is obtained by using a forward model based on the finite element
method and ICA algorithm to solve the inverse problem.
The axisymmetric diffusion equation can be expressed in variational terms by the
energy functional [6],

! 1 1  2
A  j 2
2 
F  " 
r  A   A J s . A  2 .r.dr.dz (1)
 2   r r z  2 
R   

The energy terms corresponding to the magnetic field, eddy current and source
current are represented by the first, second and third terms, respectively, of the
integrand. The impedance of a filamentary circular loop of radius ri can be calculated
from the magnetic vector potential Ai at ri and the value of the impressed current I s
in the loop as follows:

#
E t j.2 .ri . Ai
Zi    (2)
Is Is Is

Hence, the total impedance of the circular coil whose cross section is discretized
into N triangular elements is given by

j.2 .N s N j.2 .J s N
Z
Is
$ ci ci i
r . A % 
I s2
$
rci .% i .Aci (3)
i 1 i 1

Where
N s : turn density [turns / m 2 ] ,
% i : area of i-th element,
rci , Aci : central values of r , A in i-th element.
So the total impedance of a differential probe can be obtained by summing the
impedance of each coil of the differential probe.
194 A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation

2. Imperialist competitive algorithm

Like other evolutionary techniques, ICA starts with an initial population. Population
individuals called countries are divided into two types: colonies and imperialists that all
together form some empires. Imperialistic competition among these empires forms the
basis of ICA. During this competition, weak empires collapse and powerful ones take
possession of their colonies. Imperialistic competition hopefully converges to a state in
which there is only one empire and its colonies are in the same position and have the
same cost as the imperialist. The pseudo code of Imperialist competitive algorithm is as
follows [7]:
& Select some random points on the function and initialize the empires.
& Move the colonies toward their relevant imperialist (Assimilation).
& Randomly change the position of some colonies (Revolution).
& If there is a colony in an empire which has lower cost than the imperialist,
exchange the positions of that colony and the imperialist.
& Unite the similar empires.
& Compute the total cost of all empires.
& Pick the weakest colony (colonies) from the weakest empires and give it
(them) to one of the empires (Imperialistic competition).
& Eliminate the powerless empires.
& If stop conditions satisfied, stop, if not go to second point.
To start the optimization algorithm we generate the initial population of size N pop .
We select N imp of the most powerful countries to form the empires. The remaining
N col of the population will be the colonies each of which belongs to an empire. Then
we have two types of countries: imperialist and colony. To form the initial empires, we
divide the colonies among imperialists based on their power. That is the initial number
of colonies of an empire should be proportionate to its power. Figure 1 shows an
example of an initial population of each empire.

Imperialist 1

Colony 1

Imperialist 2

Colony 2

Imperialist 3

Colony 3

Imperialist n

Colony n

Figure 1. Initial empires.

The movement of colonies toward their relevant imperialist is shown in figure 2 in


which the colony moves toward the imperialist by x units. In this movement,  and x
are random numbers with uniform distribution as illustrated in formula (4) and d is the
distance between colony and the imperialist.
A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation 195

x ~ U
0,  d , ' ~ U
 ,  (4)

Where and  are parameters that modify the area that colonies randomly
search around the imperialist. In our implementation and  are considered as 2 and
0.5 (Radian) respectively.

Imperialist
New position
of colony
x

' d
Colony

Figure 2. Movement of colonies toward imperialist.

In imperialistic competition, all empires try to take possession of colonies of other


empires and control them. Powerless empires will collapse in the imperialistic
competition and their colonies will be divided among other empires. In modelling
collapse mechanism, different criteria can be defined for considering an empire
powerless. In our implementation we assume an empire collapsed and eliminate it
when it loses all of its colonies. Convergence is obtained when all the empires except
the most powerful one will collapse and all the colonies will be under the control of
this unique empire.

2.1. Validation with test functions

ICA was implemented and then tested with some benchmark functions. The results
which we give hereafter are those found with the function F which is written [7]:

F  x  sin
4 x  1.1  sin
2 y (5)

Figure 3 shows a 3D plot of the function F . The global minimum of this function
is in the interval 0 ( x ( 10 , 0 ( y ( 10 , is located in
x, y 
9.039,8.668 and has the
cost of 18.5547 .
196 A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation

Figure 3. 3D plot of the function F.

The initial population is of 30 countries. We choose 4 of the best countries to form


the imperialists which are shown by stars marks in different colors while the colonies
by round points in the same color as the imperialist. Figures 4 (a, b, c, d) show the
countries at iterations 5, 10 and 20 (convergence). We can see that at iteration 5, the
imperialists are located at local minima of the function and one of them has reached to
the global minimum. At iteration 10, only two imperialists are remained and the others
have collapsed. At iteration 20, all the imperialists have collapsed except one and
colonies are in the same position as the imperialist. Due to the mutation at last iteration,
some of the colonies are in different position.
10 10

9 9

8 8

7 7

6 6

5 5
y
y

4 4

3 3

2 2

1 1

0 0
0 2 4 6 8 10 0 2 4 6 8 10
x x
(a) (b)

10 10

9 9

8 8

7 7

6 6

5 5
y
y

4 4

3 3

2 2

1 1

0 0
0 2 4 6 8 10 0 2 4 6 8 10
x
(c) x
(d)
Figure 4. Initial empires (a). Empires at iteration 5 (b). Empires at iteration 10 (c) Empires at iteration 20 (d).
A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation 197

We also applied Particle Swarm Optimization (PSO) algorithm to the function


F [8]. To make a comparison we choose a population of 30 particles. Both global and
local attraction parameters are equal to 1.4. Figure 5 shows average and best cost of
population versus iterations obtained with PSO and ICA.
20 -11
Minimum Cost Minimum Cost
15 Mean Cost Mean Cost
10 -13
F u n c tio n c o s t

F unc tion c os t
5
0 -15
-5
-10 -17
-15
-18.5547 -18.5547
0 40 80 120 160 200 260 0 20 60 100 140 180 220 260
Iterations Iterations
(a) (b)
Figure 5. Mean and minimum cost versus iterations by PSO (a) and ICA (b).

Results show that with ICA the convergence is reached more quickly than with
PSO. Indeed with PSO convergence is obtained after approximately 240 iterations
whereas with ICA it is obtained after 20 iterations. We noted that the execution time of
iteration is practically the same one for both algorithms. Consequently, ICA finds the
global minimum successfully.

3. Application and results

The problem geometry is that studied in [9]. A differential probe is used to scan a
conducting tube (aluminum) having an electrical conductivity  c  1M )s / m* . The
geometrical and electrical data of the eddy current differential probe is as follows:
height of a coil according to z: 0.75e-3 m, inner radius of a coil: 7.75e-3 m, outer radius
of a coil: 8.5e-3 m, vertical distance between the coils: 0.5e-3 m, number of turns of a
coil: 70. The probe is supplied by a current with intensity of 5mA and a frequency of
100kHz . The impedance measured when the medium of the coil is opposite the lower
edge of the groove is Z m  (0.55 j1.45)+ .
The inversion is based on an iterative approach that employs a direct finite element
model to simulate the fundamental physical process, as shown in figure 5. The
inversion algorithm starts with an initial estimate of the groove profile and then
determines the signal by solving a finite element direct problem. The error between the
measured and the calculated signals is minimized iteratively by updating the groove
parameters by keeping the best profile of the previous iteration. When the error is
below a threshold, the profile determined is the desired solution.
198 A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation

Measured signal
Calculated
Z m
Initial groove signal
profile Forward
Z c Yes Desired

Z m Z c , - solution
model

Update groove No
profile

Figure 6. Inversion principle method

Figures 7 (a) and (b) illustrate the evolution of the groove parameters obtained by
PSO and ICA respectively. By PSO, convergence is obtained after 124 iterations. The
value of the height and the depth of the groove are h  3.98mm and p  0.508mm
respectively. On the other hand, by ICA, convergence is practically obtained with 22
iterations and then the value of the height and the depth of the groove are h  3,97 mm
and p  0,509mm respectively. In order to make comparison we used the same number
of particles in PSO as of countries in ICA i.e.30. We noted that the execution time of
iteration is the same one for both algorithms. Consequently, this enables us to
conclude that ICA converges more quickly than PSO algorithm. Nevertheless, with
regard to the precision of the results, PSO proves to be slightly better than ICA.
-3 -3
x 10 x 10
6 7
Groove depth [m] Groove height
5 Groove height [m] 6 Groove depth
Groove parameters [m]

Groove parameters [m]

5
4
4
3
3
2
2
1 1
0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
(a) Iterations (b) Iterations

Figure 7. Groove parameters by PSO (a) and ICA (b).

4. Conclusion

The method proposed in this paper, Imperialist Competitive Algorithm (ICA) uses an
evolutionary algorithm in order to inverse eddy current non-destructive evaluation
signals. The aim is to reconstitute the profile of an axi-symmetrical groove. The
method uses a finite-elements forward model to simulate the physical process and
Imperialist Competitive Algorithm to solve the inverse problem. This evolutionary
optimization strategy has shown great performance in both convergence rate and global
optima achievement. Indeed, from the experimental results, it can be seen that the ICA
method outperforms the PSO one in terms of speed convergence. However, PSO
remains a method which gives results with a better precision. In continuing this work,
the proposed algorithm could be extended to reconstitute complex profiles.
A. Hamel et al. / ICA Applied to Eddy Current Nondestructive Evaluation 199

References

[1] S. Hoole, S. Subramaniam, R. Saldanha, and J. Coulomb, Inverse Problem Methodology and Finite
Elements in the Identification of Cracks, Sources, Materials, and their Geometry in Inaccessible
Locations, IEEE Trans. Magn., vol. 27 (1991), pp. 3433 3443.
[2] Y. Li, L. Udpa, and S.S. Udpa, Three Dimensional Defect Reconstruction From Eddy Current NDE
Signals Using a Genetic Local Search Algorithm, IEEE Trans. Magn., vol 40, No. 2 (2004), pp. 410
417.
[3] M. Rebican, Z. Chen, N. Yusa, L. Janousek, and K. Miya, Shape Reconstruction of Multiple Cracks
From ECT Signals by Means of a Stochastic Method, IEEE Trans. Magn., vol 42, No. 4, (2006), pp
1079 1082.
[4] M. Cacciola, S. Calcagno, F. C. Morabito, and M. Versaci, Swarm Optimization for Imaging of corrosion
by impedance measurements in Eddy Current Test, IEEE Trans. Magn., vol. 43, No. 4, (2007), pp
1853 1856.
[5] A. Hamel, H. Mohellebi, M. Feliachi and F. Hocini, Particle Swarm Optimization for Reconstitution of
Two Dimensional Groove Profiles in Non Destructive Evaluation, ISEF 2009 XIV International
Symposium on Electromagnetic Fields in Mechatronics, No. 4, September 10 12, Arras, France,
(2009), pp. 1079 1082.
[6] O.C.Zienkiewicz and R.L.Taylor, The Finite Element Method, 4th ed., McGraw-Hill, 1989.
[7] E. Atashpaz Gargari, C. Lucas, Imperialist Competitive Algorithm : An Algorithm for Optimization
Inspired by Imperialistic Competition, IEEE Congress on Evolutionary Computation, Singapore ,(
2007), pp 4661 4667.
[8] R. C. Eberhart and J. Kennedy, A New Optimizer Using Particle Swarm Theory, proc. Sixth
International Symposium on Micro Machine and Human Science (Nagoya, Japon), IEEE Service
Center, Piscataway, NJ (1995), pp. 39 43.
[9] J.L Thomas, Modlisation du Contrle Non Destructif par Courants de Foucault des Gnrateurs de
Vapeur, Rapport de Stage 3me anne de cycle dingnieur, CEA, France, 1998.
This page intentionally left blank
Transducers and Techniques
This page intentionally left blank
Electromagnetic Nondestructive Evaluation (XIV) 203
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-203

Pulsed ECT Method for Evaluation of Pipe


Wall-thinning of Nuclear Power Plants
using Magnetic Sensor
Shejuan XIEa, Toshihiro YAMAMOTOb, Toshiyuki TAKAGIa,1 and Tetsuya
UCHIMOTOa
a
Institute of Fluid Science, Tohoku University, Japan
b
Japan Power Engineering and Inspection Corporation, Japan

Abstract. In nuclear power plants, there may happen local wall-thinning on the
inner surface of a pipe due to the flow of coolant flowing inside the pipe. Pulsed
eddy current testing (pulsed ECT) technology is developed in recent years.
Because of its rich frequency components and applicability of large electric current,
a pulsed ECT method may show promising capability of detecting and evaluating
the defect in the deep region of the material. The aim of this study is to discuss the
feasibility of detection and evaluation of local wall-thinning of the bottom surface
in one thick layer and also in the lower layer of two-layer structure pipe using a
pulsed ECT method. Concerning the large thickness of the specimen, a high
sensitive flux gate (FG) sensor has been employed and the corresponding
differential exciting mode has been developed to efficiently apply the FG sensor.
Experimental results show that this FG sensor combined with differential exciting
mode could detect a very small defect that is located in a thick specimen while
which could not be detected by Hall sensor.

Keywords. local wall-thinning, pulsed ECT, differential exciting mode, FG sensor

Introduction
In nuclear power plants, there may happen local wall-thinning on the inner surface of a
pipe due to of the flow of coolant flowing inside the pipe (flow accelerated corrosion
(FAC)). Normally, the thickness of the pipe is very large. In addition, in Japan, there is
a special concern about local wall-thinning under an enforcement plate that covers the
outside of a pipe where a branch pipe is connected to a main pipe. Because the
enforcement plate and the pipe wall form two layers of the metal plate, also the
existence of thick insulators outside the pipe, these make it very difficult to inspect the
inside of the pipe by ultrasonic testing method. Pulsed eddy current testing (pulsed
ECT) technology is one of the methods developed in recent years [1, 2]. Because of its
un-necessity of contact between the probe and the inspected specimen, also its rich
frequency components and applicability of large electric current [3-5], pulsed ECT
method may show promising capability of detecting and evaluating the defect in the
deep region of the material with lift-off. The aim of this study is to discuss the
feasibility of detection and evaluation of local wall-thinning of the bottom surface in

1
Corresponding Author: Toshiyuki TAKAGI. Katahiri 2-1-1, Aoba-ku, Sendai 980-8577, Japan;
E-mail: takagi@ifs.tohoku.ac.jp
204 S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants

one thick layer and also in the lower layer of two-layer structure pipe using a pulsed
ECT method with small lift-off.
When the specimen is rather thick, low frequency is preferred due to the skin effect
phenomenon. It has been pointed out that the sensitivity of pulsed ECT technique can
be greatly enhanced when magnetic sensors are applied instead of pick-up coils for
picking up the field. The enhancement can especially be beneficial when deeply buried
defects are of the prime concern [6].
There is a family of magnetic field sensors, such as, SQUID (Superconducting
Quantum Interference Device), FG (flux gate) sensors, AMR (anisotropic magneto-
resistive), GMR (giant magneto-resistive) and Hall devices [6]. As we know, SQUID
possesses the highest sensitivity but its use is still limited due to practicality and some
cost-related reasons [7]. Hall sensor is not a bad choice when the specimen is not very
thick and also the defect is not very small, but for the small defect detection in a thick
specimen Hall sensor is not preferred due to its relative big noise level (i.e. poor
resolution, according to the reference [6] and also the experience of the author). Thus
FG, AMR and GMR sensors are good candidates for the small defect detection in
rather thick specimen. In this study, FG sensor was employed for the local wall-
thinning detection in large diameter pipes.
Concerning the characteristics of FG sensor, of course, high sensitivity and high
resolution are its obvious merits, but at the same time, small measurement range is its
inherent demerit, normally inside several Gauss. To overcome the trade off of high
sensitivity and small measurement range, in this study, a differential exciting mode has
been developed and the corresponding two identical exciting pancake coils have been
designed.

1. Duty analysis of square wave pulse exciting signal


Figure 1 shows the ideal square wave pulse. The corresponding FS representation in
trigonometric series can be given by equation (1),

d N 1 1
u (t ) = + [ sin(n d ) * cos(nt ) + (1 cos(n d )) *sin( n t )] (1)
T n 1 n n

Where, (1/n)sin(nd)=an, (1/n) (1-cos(nd))=bn, d denotes pulse width (the


duration of exciting pulse) and T denotes pulse period, thus duty can be expressed by
d/T (i.e. = d/T ). The fundamental angular frequency =2/T, n denotes the harmonic
order and N is the number of summation terms (in case of ideal square wave pulse, N is
positive infinite).

Figure 1. Ideal square wave pulse


The amplitude of n-th harmonic frequency could be written by equation (2),
S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants 205

2
An = an 2 + bn 2 = * 1 cos(n d ) (2)
n

So, when duty is , the amplitude of n-th harmonic frequency is,

2 2
An , = * 1 cos(n d ) = * 1 cos(2n ) (3)
n n

When duty is 1-, the amplitude of n-th harmonic frequency is,

2 2
An ,1 = * 1 cos(n (T d )) = * 1 cos(2 n ) (4)
n n

From the above equation (3) and (4), we can see that when the duty of two square
wave pulse signals is complementary to 1, their spectrum distribution should be
completely the same only except the spectrum of DC component. FFT was applied to
the square wave pulse under condition of different duties (0.1, 0.2,, 0.9), then the
energy (here amplitude was considered as energy) percentage of the fundamental
frequency, which means energy of fundamental frequency divided by the energy of all
the harmonic frequencies (except DC component) has been obtained and shown in
Figure 2.

Figure 2. Energy percentage of the fundamental frequency


From Figure 2 we can see that when duty is 50%, the energy percentage of
fundamental frequency is the biggest. Therefore, if the defect is located on the
subsurface of a thick specimen, big pulse period (which means low fundamental
frequency) and 50% duty are preferred because more information from deep position of
the specimen can be expected by low frequency. However, if the defect is located on
the surface of the specimen, short pulse period combined with small duty (e.g. 10%
duty) is more suitable in pulsed ECT technique. In this study, 50% duty of pulse was
applied to detect the local wall-thinning on the bottom side of pipes.

2. Pulsed ECT system


2.1. Specimens
To simulate the local wall-thinning on the bottom side of one layer thick pipe, three
AISI316 austenitic stainless steel flat plates with a slot defect were prepared in our
experiments. The size of the three plates is the same, 300mm in length, 100mm in
206 S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants

width and 20mm in thickness. A rectangular slot was arranged along the center line of
the bottom side of the plate in the shorter direction. The size of the slot defect
(widthdepth) is 2010, 1010 or 105 (mm), respectively. The slot defect is to be
detected from the top surface of the plate. The specimen is shown in Figure 3.

Figure 3. One plate specimen with a slot defect simulating local wall-thinning in one layer thick pipe
To simulate the local wall-thinning on the bottom side of a pipe which is covered
by an enforcement plate, we have prepared two AISI316 austenitic stainless steel flat
plates for our experiments. The size of the two plates is the same, 500mm in length,
300mm in width and 8mm in thickness. One plate is placed on the top of the other. A
rectangular slot has been arranged along the center line of the bottom side of the lower
plate in the shorter direction. The width of the slot is 10mm, and the depth is 1, 3 or
5mm, respectively. The slot is to be detected from the top surface of the upper plate.
The specimen is shown in Figure 4.

Figure 4. Double-plate specimen with a slot defect simulating local wall-thinning in two layer pipes

2.2. Experiment setup


The pulsed ECT experiment system we established consists of a function generator
(WF1945, NF), a power amplifier (BP4610, NF), a scanning stage, an AD board and a
PC etc. In the experiments, a square wave pulse was generated from the function
generator and then amplified by the power amplifier by which the output current could
be controlled instead of the output voltage. Then the amplified current signal of the
square wave pulse was applied to the exciting coils as the exciting signal. A magnetic
sensor is supposed to be the pick-up sensor for picking up the magnetic field. The
pulsed ECT experiment setup is shown in Figure 5.
2.3. Pick-up sensor
As mentioned in the Introduction part, because of demerit of SQUID and Hall device,
in this study a TMF-FG sensor, one kind of FG sensor, was employed to detect local
wall-thinning in large diameter pipes. The manufacturer of this FG sensor is Canon,
sensitivity 1V/G, voltage source DC 5V, offset 2.5V, measurement range about from -
2.5G to 2.5G and resolution is about 1mG according to the manual. The circuit board
of sensor and its output property are shown in Figure 6(a) and Figure 6(b), respectively.
From Figure 6(a) we could see that on the circuit board there are two sensors which can
measure the magnetic field of two directions (measuring direction is parallel to the
length direction of the sensor). In experiments, one sensor or two sensors can be used at
S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants 207

one time according to the experiment request. Figure 6(b) shows that the output has
good linear property to external magnetic field within the measurement range.

Figure 5. Pulsed ECT experiment setup

(a) TMF-FG sensor (b) Output property


Figure 6. TMF-FG sensor and its output property

2.4. Exciting and pick-up mode


Generally speaking, as far as the magnetic sensor is concerned, there is trade off
between the sensitivity and the measurement range, i.e. low sensitivity with large
measurement range while high sensitivity with small measurement range. For FG
sensor, small measurement range is its inherent demerit. To overcome this problem, in
this study, a differential exciting mode has been developed and shown in Figure 7(a).
There are two almost identical exciting coils, coil 1 and coil 2, the parameters of which
are shown in Table 1. Reverse exciting signal has been applied to the two exciting
coils and FG sensor has been located at center of the two coils (see Figure 7(b)).
It has been pointed out that the ratio between the variations of the perturbed field
to the unperturbed field is maximized along the direction which is perpendicular to the
coil circle plane [7]. So in experiments, the magnetic field of vertical direction
(perpendicular to the coil circle plane and surface of specimen) was measured by FG
sensor. The distance from the sensor to the bottom plane of the coil is 6mm (The
smaller, the better. 6mm is the minimum value that we can set because of the big size
of the circuit board of the sensor). To simulate the insulator outside the pipes, firstly
2mm lift-off was applied.
Concerning this mode, there are mainly two aspects of advantage:
(1) FG sensor only measures the signal of the perturbed field that is affected by the
existence of defect because the direct magnetic field of vertical direction (generated by
exciting current in coils) at the position of FG sensor could be counteracted to be zero.
208 S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants

The perturbed field is rather small, thus the small measurement range of FG sensor can
be satisfied.
(2) Eddy current in the specimen decreases very slowly along depth direction and
deeper penetration depth could be expected.

(a) Differential exciting mode (b) Exciting coils and FG sensor


Figure 7. Differential exciting mode and coils
Table 1. Parameter of exciting coils
Items Coil 1 Coil 2
ID 30mm 30mm
OD 60mm 60mm
Height 30mm 30mm
Wire diameter 1mm 1mm
Turn 389 387
Impedance (at 50Hz) 1.847Ohm 1.850Ohm
Distance between two coils 80mm (center to center)

3. Results and discussion


3.1. Results of local wall-thinning detection in one layer pipe
In this study, the local wall-thinning on the bottom side of a plate (shown in Figure 3,
20mm thick) is to be detected using the above pulsed ECT system and FG sensor. The
magnitude of the DC part of the pulse exciting current is 4A for each coil, the period is
0.02s, duty is 50%. The pick-up signal collected by the FG sensor was averaged over
100 cycles of the transient output for duration of 2.0s, to reduce noise. Inspection
condition is shown in Figure 8. In experiments, scanning signals were obtained using
scanning stage and the center of the defect was chosen as the origin. The center of two
exciting coils (also the position of FG sensor locates at) started from the point which is
75mm far from the origin point (coordinate is -75mm) and moved along the scanning
direction to the position of 75mm coordinate in which the interval of every step is 3mm.
The peak value at every scanning point was extracted from the pick-up signal to detect
the slot defect of the plate.
Figure 9(a) shows the results of three specimens with different size of slot defect.
Figure 9(b) shows the relationship between value of peak to peak of scanning signal
and the size of defect. We can see that local wall-thinning in one thick layer could be
detected and different size of defect has different value of peak to peak. This lays a
good foundation for the future quantitative work.
S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants 209

Figure 8. Inspection condition

(a) Scanning signal (b) Relationship between value of peak to peak of scanning signal and size of defect
Figure 9. Pick-up signal and relationship between signal and defect

3.2. Results of local wall-thinning detection in double layer pipes


In this study, the local wall-thinning on the bottom side of the lower layer in double
layer plates (shown in Figure 4, 8mm plus 8mm thick) is to be detected using the same
experiment system and the same parameters of exciting signal as in section 3.1.
Inspection condition is shown in Figure 10. In experiments, S1 is the pick-up signal at
position 1 where sensors are far from defect (ideally S1 is 0) as the reference signal. S2
is the pick-up signal at position 2 where the center of coil 2 locates at the center of the
defect.

Figure 10. Inspection condition


Figure 11(a) shows the results of differential signal (S2-S1) of the three
specimens with different depths of slot defect using FG sensor and Hall sensor,
respectively. The results show that FG sensor is much better than Hall sensor for small
defect detection in a thicker specimen because of the big noise level of Hall device.
Similarly Figure 11(b) shows the relationship between pick value of the differential
signal and the depth of the defect. We can see that local wall-thinning in double layer
pipes could be detected and different depth of defect has different peak value. This also
lays a good foundation for the future quantitative work.
210 S. Xie et al. / Pulsed ECT Method for Evaluation of Pipe Wall-Thinning of Nuclear Power Plants

(a) Pick-up signal using FG sensor and Hall sensor (b) Relationship between peak value and depth of defect
Figure 11. Pick up signal and relationship between signal and defect

4. Conclusion
In the paper, it has been analyzed how to choose a suitable duty of square wave pulse
in pulsed ECT method. A differential exciting mode was developed to efficiently apply
TMF-FG sensor in pulsed ECT experiment system. The experiment results state that
local wall-thinning in one thick layer pipe (20mm) and double layer pipes (8mm+8mm)
with small lift-off (2mm) could be detected using the above detection mode. In addition,
experiment result shows that FG sensor is much better than Hall sensor for small defect
detection in thicker specimen. This study gives a good foundation for the future
quantitative work.

Acknowledgements
This work was conducted as a part of Nuclear and Industrial Safety Agency (NISA)
project on Enhancement of Ageing Management and Maintenance of Nuclear Power
Plants in Japan and supported by the Grant-in-Aid for the Global COE Program,
"World Centre of Education and Research for Trans-Disciplinary Flow Dynamics",
from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of
Japan. The authors would like to thank Mr. Takeshi Sato of Tohoku University, for the
preparation of the specimens and the fabrication of the probe.
References
[1] G.Y. Tian, A. Sophian. Defect classification using a new feature for pulsed eddy current sensors.
NDT&E International 38 (2005) 77-82.
[2] J. Kim, G. Yang, L. Udpa, S. Udpa. Classification of pulsed eddy current GMR data on aircraft structures.
NDT&E International 43 (2010) 141-144.
[3] M. Fan et al. Analytical modeling for transient probe response in pulsed eddy current testing. NDT&E
International 42 (2009) 376 383.
[4] T. Chen et al. Feature extraction and selection for defect classification of pulsed eddy current NDT.
NDT&E International 41 (2008) 467 476.
[5] M. Morozov, G.Y. Tian and D. Edgar. Comparison of PEC and SFEC NDE Techniques. NDT&E
International 24 (2009) 153 164.
[6] G.Y. Tian and A. Sophian. Study of magnetic sensors for pulsed eddy current techniques. Insight 47
(2005) 277-279.
[7] M. Valentino, G. Pepe, A. Ruosi and G. Peluso. Eddy-current nondestructive measurements with
different HTS-SQUID spatial orientations. J. Phys. IV France 8 (1998) 249-252.
Electromagnetic Nondestructive Evaluation (XIV) 211
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-211

Developing ECT System with AMR Sensor


for Combustion Chamber
Dongfeng HE a,1, Mitsuhara SHIWAa, Jianping JIAa, Hisashi YAMAWAKIa, Ichizo
UETAKEa, Junji Takatsubo b, Shinichi MORIYAc, and Koichi OKITAc
a
National Institute for Materials Science, Sengen 1-2-1, Tsukuba, 305-0047, Japan
b
Advanced Industrial Science and Technology1-2-1 Namiki, Tsukuba, 305-8564, Japan
c
Japan Aerospace Exploration Agency, Sengen 2-1-1, Tsukuba, 305-8505, Japan

Abstract. An eddy current testing (ECT) system with a high sensitive AMR sensor
was developed. In our experiments, a specimen of cooper plate with grooves and
slits was used to simulate the cooling grooves of the combustion chamber of liquid
rocket and the defect in it. Three defects, with the width of 0.2 mm, the length of 4
mm and the depth of 0.2 mm, 0.5 mm and 0.8 mm respectively, were made in the
bottom of the grooves. Using a 4.5 mm and 30 turn circular excitation coil, ECT
experiments were done for the AMR sensor with Z, X and Y directions at the
frequency of 2 kHz. The results show that the variance of lift off had big influence
on the ECT result with the Z direction AMR sensor and smaller influence on that
with the X, Y direction AMR sensor. The defects with the depth of 0.8 mm and
0.5mm, 0.2 mm were clearly observed, but the signal of the defect with the depth
of 0.2 mm was very small.

Keywords. ECT, AMR, combustion chamber

Introduction

For the liquid rocket, liquid Oxygen and liquid Hydrogen are used as the propellant.
The wall of the combustion chamber is made of Cu-Cr-Zr copper alloy. High
temperature gas over 3000 K is generated in the combustion chamber and liquid
hydrogen flowing in the cooling groove is used for the cooling. Due to the big thermal
gradient and the excessive thermal strain generated in the cooling groove, two kinds of
damages might be generated in the wall of the combustion chamber. One is from the
inner wall side, which is caused by the oxidation/reduction of the oxide layer; another
is the melt damage from the cooling grooves, which is caused by the growth of the
small crack in the grain boundary of the copper alloy. The aim of this research is to
find an effective method for the tiny defect detection and evaluation of the combustion
chamber.
Eddy current testing (ECT) is an effective method to detect defects in conductive
materials. It has the advantage of low price and easy operation. Inductive coil,
anisotropic magneto resistance (AMR), giant magneto resistance (GMR), flux gate,
superconducting quantum interference device (SQUID) have been used as the sensors
in ECT systems [1-4]. SQUID has the best magnetic field resolution especially at low
frequency. However, SQUID must be cooled by liquid nitrogen or liquid helium, so the
ECT system with SQUID is expensive, complex and is not easy to handle. High
sensitive flux gate is also expensive. The sensitivity of coil become worse at low
212 D. He et al. / Developing ECT System with AMR Sensor for Combustion Chamber

frequency and GMR sensor also has big noise at low frequency. AMR sensor is
cheaper than SQUID and flux gate and has better sensitivity than inductive coil and
GMR sensor at low frequency.
Using the low noise driving circuit developed by us, the magnetic field resolution
of AMR sensor was improved to 12 pT/Hz at the frequency of above 1 kHz [5]. The
driving circuit operated in amplifier mode or feedback mode and the linearity could be
improved for feedback mode, which is important for a moving ECT system in
unshielded environment. ECT experiments were once been done for an aluminum plate
[6]. In this report, we will present our ECT results for the specimen of copper plate
with the X, Y, and Z sensing directions of the AMR sensor.

1. Experimental setup

Figure1 shows the block diagram of the ECT system with AMR sensor. The circular
coil was used to produce the excitation field. Then, eddy-current was induced in the
specimen. If there was some defect in the specimen, the distribution of the eddy-current
was changed and the field produced by the eddy current was detected by the AMR
sensor. Through the lock-in amplifier, the amplitude signal and the phase signal could
be obtained; then it was sent to a computer for data acquisition and data processing. In
our experiment only the amplitude signal was used. The AMR sensor was fixed with
the X-Y stage for the scanning.

Figure 1. The block diagram of the ECT system with AMR sensor

1.1. Specimen

To simulate the real combustion chamber of liquid rocket, the specimen was made of
copper alloy plate. Figure 2 shows it. The thickness of the plate was 4 mm. Grooves
with the width of 1 mm and the depth of 3 mm were made on one side of the copper
plate. Three slit defects were made in the bottom of some grooves with the length of 4
mm, the width of 0.2 mm and the depth of 0.2 mm, 0.5 mm, and 0.8 mm respectively.
The surface of the copper was not flat; bending exists in some part of the copper plate,
which caused the variance of the lift off.
D. He et al. / Developing ECT System with AMR Sensor for Combustion Chamber 213

Figure 2. The specimen of copper plate with grooves and slit defect to simulate the combustion chamber of
liquid rocket.

1.2. AMR sensor and the driving circuit

A commercially available AMR sensor of HMC1001 was used [7]. The magnetic field
sensitivity of the AMR sensor was about 3.2 mV/V/Gauss. To improve the field
resolution, higher bias voltage of 24 V and low noise preamplifier were used. To
improve the linearity of the response, a driving circuit operating in feedback mode was
designed [5]. The AMR sensor has flat response till to 5 MHz and the magnetic field
resolution was about 12 pT/Hz at the frequencies of above 1 kHz.

1.3. The excitation coil and the sensing direction of AMR sensor

A 30 turn, 4.5 mm circular coil was used to produce the excitation field. Figure 3
shows the three sensing direction of the AMR sensor. Figure 3 (a) shows the Z
direction AMR sensor. The coil was attached to the bottom of the AMR and the
position was adjusted to get the biggest output for the excitation field. Figure 3 (b)
shows the X direction sensor. The sensing direction of the AMR sensor was along the
direction of the groove. The coil was attached to one side of the AMR sensor and the
position of the coil was adjusted to get the smallest output. Figure 3 (c) shows the Y
direction sensor. The sensing direction of the AMR sensor was perpendicular to the
grooves direction.

D 
D  E
E  F
F 

Figure 3. The excitation coil and the sensing direction of AMR. (a). Z direction. (b). X direction. (c). Y
direction
214 D. He et al. / Developing ECT System with AMR Sensor for Combustion Chamber

1.4. The experimental condition

The conductivity of the copper alloy is about 5.8106 S/m. In our experiments, the
excitation frequency was 2 kHz. From the formula = 1 f , where, is the
penetration depth, f is the excitation frequency, is the permeability of the material and
is the conductivitythe penetration depth is about 1.5 mm for copper alloy at 2 kHz.
The amplitude of the current flow in the excitation coil was about 20 mA. The
amplitude of the magnetic field at the center of the coil was about 2 Gauss. X
direction was parallel to the direction of the grooves, Y direction was perpendicular to
the direction of the grooves, and Z direction was vertical direction perpendicular to the
surface of the specimen. For the XY scanning, first, fixing the X position and line
scanning along Y direction; then changing the X position. Due to the bending of the
specimen, the variance of the lift off was from 0.2 mm to 1mm.

2. Experimental results

We did the ECT experiments for the AMR sensor with Z direction, X direction and Y
direction. For the Z direction sensor, the big excitation field was also detected by the
AMR senor. For the X and Y direction sensor, the excitation field was not detected by
the AMR sensor.

2.1. Z direction

Figure 4 shows ECT results when using the AMR sensor with Z direction. Figure 4 (a)
shows the Line scanning where the defects exist. Figure 4 (b) shows line scanning
where no defects exist. Figure 4 (c) shows the 2D contour map of the XY scanning
results. The defects with the depth of 0.8 mm and 0.5 mm were observed. Due to the
big background field and the influence of the variance of the lift off, the smallest defect
with the depth of 0.2 mm was not observed.

Figure 4. ECT results using AMR with Z direction. (a). Line scanning result where defects exist. (b). Line
scanning result where no defects exist. (c). Contour map of the XY scanning results.

Figure 5 (a) shows the subtraction results of the line scanning Figure 4 (a) and
Figure 4 (b). Figure 5 (b) shows the 2D contour map of the subtraction results. The
D. He et al. / Developing ECT System with AMR Sensor for Combustion Chamber 215

defects of 0.8 mm and 0.5 mm become clearer; A very small signal appeared for the
defect signal with the depth of 0.2 mm.

Figure 5. (a). The subtraction result of Figure 4 (a) and Figure 4 (b). (b). The contour map of subtraction
results.

2.2. X direction

Figure 6 shows ECT results when using the AMR sensor with X direction. Figure 6 (a)
shows the Line scanning where the defects exist. The defect signal was positive at one
end of the defect, and it is negative at another end of the defect. Figure 6 (b) shows the
2D contour map of the XY scanning results. Using the AMR sensor with the X
direction, the influence of the variance of the lift off was reduced. A small signal
appeared for the defect with the depth of 0.2 mm.

Figure 6. ECT results using AMR with X direction. (a). Line scanning result where defects exist. (b). 2D
Contour map of the XY scanning results.

2.3. Y direction

ECT experiments using AMR sensor with Y direction were also done. Figure 7 (a)
shows line scanning results where defects exist. For each defect, a peak and a valley
signal appeared. The defects with the depth of 0.8 mm and 0.5 mm were clearly
216 D. He et al. / Developing ECT System with AMR Sensor for Combustion Chamber

observed. The signal of the defect with the depth of 0.2 mm was small. The influence
of the variance of lift off was also reduced. Figure 7 (b) shows the contour map of the
results.

Figure 7. ECT results using AMR with Y direction. (a). Line scanning result where defects exist. (b). 2D
Contour map of the XY scanning results.

3. Summary

ECT system with AMR sensor was constructed and ECT experiments for the copper
specimen were done using the AMR sensor with the Z, X and Y directions. Variance of
lift off had big influence when using Z direction AMR sensor. The influence of the
variance of lift off was reduced when using X and Y direction AMR sensor. The
defects with the depth of 0.8 mm and 0.5 mm were observed. The signal of the defect
with the depth of 0.2 mm was very small.

Acknowledgment

We thank the helps from Prof. H. Itozaki and Dr. M. Tachiki.

References

[1] K. Allweins, M. von Kreutzbruck, G. Gierelt, Defect detection in aluminum laser welds using an
anisotropic magnetoresistive sensor array, J. Appl. Phys., 97, 10Q102, (2005)
[2] T. Dogaru, S.T. Smith, Giant magnetoresistance based eddy current sensor, IEEE Trans. Magn., 37,
pp.2790 2793, (2001).
[3] C. Carr, J.C. Macfarlane, The performance of flux gate magnetometers for nondestructive evaluation,
Insight, 41, pp.20 24, (1999).
[4] H. Weinstock, R.A. Welsh, SQUIDs for nondestructive evaluation, Superconductor Industry, 10,
pp.23 28, (1997).
[5] D.F. He, M. Tachiki, H. Itozaki, Highly sensitive anisotropic magnetoresistance magnetometer for
Eddy current nondestructive evaluation, Rev. Sci. Instru , 80, 036102, (2009).
[6] Dongfeng He, Mitsuharu Shiwa, et al., Basic Examination for the Defect Detection of Combustion
Chamber Using ECT, submitted to Journal of JSNDI, (2010).
Electromagnetic Nondestructive Evaluation (XIV) 217
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-217

Flux Leakage Measurements for Defect


Characterization Using NDT Adapted
GMR Sensors
Matthias PELKNER1, Andreas NEUBAUER, Mark BLOME, Verena REIMUND,
Hans-Martin THOMAS and Marc KREUTZBRUCK
Federal Institute for Material Research and Testing
Unter der Eichen 87, D 12205 Berlin, Germany

Abstract. High precision magnetic field sensors are of increasing interest in non
destructive testing (NDT). In particular GMR sensors (giant magneto resistance)
are qualified because of their high sensitivity, high signal to noise ratio and high
spatial resolution. We performed magnetic flux leakage measurements of artificial
cracks with a GMR gradiometer and a 3 axes GMR magnetometer. Cracks of a
depth of 44 m still could be detected with a sufficient high signal to noise ratio.
A semi analytic magnetic dipole model was used for swiftly predicting magnetic
stray fields. The reliable reconstruction based on measurements of artificial
rectangular shaped defects is demonstrated.

Keywords. GMR, NDT, magnetic flux leakage, reconstruction

Introduction

Magnetic flux leakage testing (MFL) is one of the most popular methods for defect
detection and characterization of magnetic components in NDT. The most prominent
method for a MFL application is the magnetic particle inspection (MP). In this method
magnetic particles accumulate at magnetic stray fields generated by surface cracks in a
magnetized component. By means of MP the number of defects and their length can be
determined precisely. However, the estimation of defect depth and the automation of
MP-applications is rather difficult. To examine the depth and width of inhomogeneities
in more detail it is necessary to measure the stray field near cracks. This can be done by
any magnetic field sensor, such as flux gate sensors, hall probes, SQUID, or XMR
sensors (magneto resistance; X = G (giant), A (anisotropic), etc.) [1-6].
In this work we used two types of GMR (giant magneto resistance) sensors optimized
for NDT applications: a gradiometer measuring the normal component of the magnetic
flux leakage and a new 3-axes magnetometer measuring all three directions in space.
GMR sensors are well suited for NDT applications thanks to their high sensitivity, high
signal-to-noise ratio and small sensing area associated with a high spatial resolution. In
contrast to commercial GMR sensors that are usually embedded in a bulky
encapsulation, the sensitive layers of the GMR sensors used in this work are located

1
Corresponding Author: E mail: matthias.pelkner@bam.de. This work is supported by the Federal Ministry
of Education and Research, Germany, project no. 16SV3787.
218 M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT

only a few tens of m from the chip edge, resulting in a small sensor-to-surface
distance. This helps to visualize the defect signature with a very good spatial resolution.
Further, having a fast inverse scheme in mind based on a semi-analytic model for MFL
to estimate the defect parameters like width, depth and length, it is helpful to work with
point-like sensing devices. Therefore the active sensing area should be adapted to the
liftoff and has to be smaller. Otherwise, the field gradients along the active area must
be considered via integration over the whole sensing area, leading to distinct layers
calculation times.

1. GMR sensors and semi-analytical model for MFL.

1.1. Experimental setup

The experimental setup consists of three high precision manipulators (Aerotech) for the
three directions in space, providing a maximum scanning velocity of 0.5 m/s and a
spatial resolution of 1 m. The manipulators are fixed on a 2000 kg bloc of granite
eliminating environmental vibrations.
To record the data of the sensors a fast data acquisition hardware based on a 18 bit
AD-converter with a maximum sampling rate of 600 kHz (National Instruments) was
used. To achieve a high signal-to-noise ratio, the sensor signals were pre-amplified.

1.2. GMR gradiometer and 3D magnetometer

(a) (b)
Figure 1. GMR gradiometer measuring the gradient in x direction of the normal field component Hz. (a)
Sketch of the electric circuit of the GMR chip (Sensitec GmbH). The sensing areas (R1 R4, green and red)
are sensitive to Hz. The distance between R2/R4 and R1/R3 is 250 m. (b) Schematic of an artificial crack with
dimensions length l, depth h and width w. The induced polarization on the defect wall is indicated by
plus signs. The white squares represent the sensing elements of the GMR gradiometer above the defect.

The GMR effect was discovered in 1988 by Peter Grnberg [7] and Albert Fert (Nobel
Prize in Physics, 2007). Independently they found a giant magneto resistance (GMR)
in a multilayer structure consisting of ferromagnetic layers divided by nonmagnetic
M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT 219

layers which depend on the directions of magnetization in each layer. The effect is
based on quantum-mechanical scattering of the conduction electrons at the boundary
surface between the ferromagnetic and nonmagnetic layer.
Generally a GMR gradiometer consists of four not-shielded elements fabricated as
a Wheatstone-bridge as shown in fig. 1(a), resulting in a distinct temperature-
independent output signal. The measured quantity is the difference of the magnetic
field (here the normal component Hz) between the two active areas and it is detected by
measuring the bridge voltage. In our case, the sensing areas have a size of only
7060 m and a transfer function of 14 mV/V(kA/m)-1 in its linear range between
1kA/m.
The 3-axes magnetometer (Sensitec GmbH) consists of three Wheatstone bridges
that allow to measure absolute field values via special flux concentrators. As its
prominent feature the new 3-axes magnetometer detects simultaneously all three spatial
components of the magnetic field in the same parallel plane next to the surface under
investigation. Fig. 2 depicts the layout of the sensor. The x and y components are
measured with special flux concentration layers, which are rotated by 90. Both flux
concentrators have a diameter of 140 m. This results in a spatial resolution that is
worse than for the gradiometer described above. The z-component is measured by the
field distortion in the vicinity of two flux concentrator strips with dimension about
10160 m and separated by a distance of 30 m. The transfer function for the 3-axes
magnetometer is 30 mV/V (kA/m)-1 and it has its linear range for fields between
1kA/m. For further details see [8].

Figure 2. Prototype of a 3 axes GMR magnetometer (Sensitec GmbH) measuring the three directions in
space of the magnetic field in the same plane. On the right a SEM picture of the 3 axes magnetometer is
shown [8].

1.3. Semi analytic model

Inversion of MFL signals requires an adequate forward solver for predicting magnetic
stray field signals. To account for defects we employed the magnetic dipole model
intro-duced by Shcherbinin [9,10]. According to this model MFL signals arise due to
induced magnetic polarizations on the defect walls [11].
This model allows to calculate magnetic stray fields of a rectangular defect with
length l, width w and depth h (see fig. 1(b)). Especially for GMR sensors that can be
positioned very close to the surface of the test piece, it is important to carefully
consider the dimensions of the probe, i.e., the size of their active sensing areas, in the
calculation process. In our calculation we therefore numerically integrated the magnetic
stray field expressions over the active sensor areas of the GMR testing probe (fig. 1(b))
to obtain a fast algorithm.
220 M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT

For the inverse calculation of the defect parameters we employed the trustt-region
reflective algorithm to minimize the difference between the predicted( dsyn F(m) )
and measured ( dobs ) GMR sensor data iteratively using an L2 norm approach. Initially,
the forward solution is estimated for an arbitrarily chosen start model m0 (comprising
the defect parameters l, w and h). Subsequently, the sought-after model parameters are
updated in an iterative fashion by minimizing the objective function
2
dobs j
dsyn with dsyn
j
F(m j )
, (1)

where j denotes the iteration number and F(m j ) the forward solution step as outlined
above. Minimization is continued until convergence is reached yielding an estimate of
the true defect parameters.

2. Results

As test object we used a steel plate with artificial cracks of varying depth (see fig. 3).
The artificial cracks were introduced by electrical discharge machining. The width of
the cracks varies between 100 to 200 m with a length around 5500 m. The depth
ranges between 10 m and 2240 m. The measurements were performed in remanence
(no applied field during measurement), i.e., the test object was magnetized before. The
direction of the applied field was normal to the defect geometry.

Figure 3. Schematic cross section of the steel plate. The field was applied along the x direction.

2.1. Gradiometer detecting the z component of the magnetic field

Using the gradiometer line scans with different liftoffs were recorded. Fig. 4 shows
the gradient of the magnetic stray field as a function of position x. Clear signatures of
the field gradient illustrate the defect positions and the amplitude of H scales with the
depth of the defects [11]. The right panel in fig. 4 illustrates the signals of the defects
with a depth between 44 m and 380 m in more detail. The attenuation of the signal
for an increasing liftoff is in agreement with the theoretical model. Further, it is found
that even the 44 m deep defect can be observed. However, for the 30 m and 10 m
deep cracks no clear signal can be distinguished from the noise. Possible contributions
to the noise level may originate from the roughness of the surface or from spatial
variations of the permeability.
M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT 221

Figure 4. Line scan with a gradiometer for three different distances between sensor and surface of the test
object.

2.2. Reconstruction

Magnetic field sources generally create a blurred magnetic field distribution above
the samples surface. This is due to the divergence of the magnetic field and the
position of the sensing device, which in an exterior investigation is naturally always a
certain distance from the source of the field. What starts as a distinct field distribution
inside the sample turns into a broad field distribution when the field sensor is either too
large or positioned at too far a distance above the samples surface. This blurredness
complicates the data interpretation and begs for additional support through signal
processing to evaluate the mechanical state of the sample. Thus, small sensors
providing high spatial resolution in combination with a large signal-to-noise ratio are
helpful to gain better insight into the type of defect, its size and its location.
In fig. 5 the measurement and the results of the reconstruction of two defects are
shown. For the reconstruction data in the vicinity of the defect was used to calculate the
defect parameters (approx. 120,000 data points).
The false rendering plots in the upper panel of fig. 5(a) show the experimental and
calculated data of the stray field gradient of a 210 m deep crack. In addition 10 line
scans across the defect, extracted from both data sets, are shown aside.
In the lower panel of fig. 5(a) the results of the reconstruction process are shown,
in which the parameters length, depth and width were reconstructed simultaneously. A
relative fast converging process is observed. The length converges after less then 10
iterations and the depth can be reconstructed after less than 5 iterations. In total this
reconstruction requires only a few seconds on a standard PC.
It is worth mentioning that the crack width does not converge to the real value.
This is in agreement with the sensitivity matrix and is due to the fact that in the
analytical model the crack width has a minor influence to the stray field [11].
Remarkably, we find a good reconstruction quality also for the 44 m-deep defect
although dealing with a significant decrease in SNR, as illustrated in fig. 5(b).
222 M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT

Figure 5. False color renderings of two different defects: (a) 210 m and (b) 44 m deep. The top left picture
shows the measurement and the middle one the synthetic generated signal form of the reconstruction
algorithm. Right to them ten lines are exemplary illustrated ( measurement, simulation). In the bottom
diagram the defect parameters and the steps of reconstruction are presented.

2.3. 3 axes Magnetometer

Finally, in this section first measurements of the magnetic flux leakage using a new 3-
axes magnetometer are shown. The same test plate was used as for the gradiometer.
The prototype 3-axes magnetometer measures Hx, Hy and Hz simultaneously at the
same lift off. This provides additional information of the magnetic field which is
required, e.g., for a more voxel-based reconstruction algorithm.
In fig. 6 the results for the Hx and Hz component of the magnetic field are
presented. Panel (a) and (c) show false rendering plots for both magnetic field
components. For each field component three line scans, indicated by the black, red and
green line, are shown in panels (b) and (d). These line scans illustrate the stray field at
the edges (black, green) and the center (red) of the cracks, respectively. As expected
from the analytic model, the stray field signals are most pronounced in the center of a
crack and decline towards the edges. Since a magnetometer was used the defect signals
appear superimposed on magnetic stray fields originating from the magnetic steel plate
itself. The typical signatures of these background fields are a maximum in the center of
the plate for the Hx component and a gradual decline for the Hz component.
The Hy field component (not shown) is characterized by MFL signatures at the
edges of the long sides of the cracks. Since the direction of the magnetization was
along the x direction, only at these end positions distinct signatures of Hy are expected.
M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT 223

Figure 6. Stray field distribution represented in a false color renderings. (a) False rendering plot of field
component Hx. (c) Hz component. (b) and (d): Extraction of 3 line scans across the left and right edge of the
crack (black and green) and the center of the crack (red), where field amplitude is maximum.

In fig. 7 the scanning area was adapted to three defects for the Hx component of the
magnetic field. Compared to the gradiometer we observe a somewhat higher noise level.
In addition to the 380 and 210 m deep defects the 44 m deep defect can be imagined.

Figure 7. Field distribution of the Hx component measured by a 3D magnetometer. Extraction of defects


with a depth of 380, 210 and 44 m.

3. Conclusion

The automation of flux leakage testing is an active research field, where the use of
adapted magnetic field sensors detect stray field distributions of defects above the test
specimen. For this purpose, the sensors require a high spatial resolution and an
excellent field sensitivity. We therefore propose in this work the use of NDT-adapted
GMR sensors that enable us to detect several artificial defects with depth down to 44
224 M. Pelkner et al. / Flux Leakage Measurements for Defect Characterization Using NDT

m. Detecting the magnetic field close to the surface offers the advantage to obtain
more detailed information on the geometrical parameters of the defects. Using a few
m-sized GMR gradiometer, where the sensing MR-layer is positioned close to the
chip edge, we obtained distinct crack signatures with good spatial resolution and
showed that a reconstruction of the length and depth can be reliably carried out. The
width can also be determined, however, less accurate due to the smaller influence of the
width on the stray field. In addition we proposed the use of a 3-axes magnetometer
measuring all three field component simultaneously. These results can be regarded as a
further step towards the use of GMR sensors in NDT providing a more complete
picture of the real magnetic field distribution of the samples outside world. This paves
the way for new visualization techniques, precisely defect localization, defect
characterization, and tomography-like mapping techniques.
Electromagnetic testing based on small magneto resistive layers with high spatial
resolution can be provided with relatively high cost effectiveness. This also allows
GMR-sensors to follow the miniaturization trend providing an adequate testing method
for quality control of small components. GMR-technology thus has the potential to
bridge the micro-gap between the mm-sized conventional induction coils for detecting
macroscopic material defects and the scanning magnetic force microscopy for the
detection of field distribution on the nm-scale.

References

[1] K. Allweins, M. v. Kreutzbruck, G. Gierelt, Defect Detection in Aluminium Laser Welds Using an AMR
Sensor Array, J.Appl.Phys. 97 (2005), 10Q102
[2] H. Pries, Giant Magnetic Resistance Sensoren in der industriellen ZfP, Report of Annual Meeting of the
German Society of Nondestructive Testing 2008, St. Gallen, Switzerland
[3] Y. Kataoka et al., Application of GMR line sensor to detect the magnetic flux distribution for
nondestructive testing, International Journal of Applied Electromagnetics and Mechanics 15
(2001/2002), 47 52
[4] H. J. Krause, M. v. Kreutzbruck, Recent Development in SQUID NDE, Physica C 368 (2002), 70 79
[5] C. Dolabdjian, G. Wach, L. Perez, Improvement in the detection of subsurface fatigue cracks under
airframe fasteners using improved rotating giant magneto resistance magnetometer head, Paper
presented at the 9th European Conference on NDT, 25 29 September 2006, Berlin
[6] F. Gruhl, M. Mck, M. v. Kreutzbruck, J. Dechert, A scanning superconducting quantum interference
device microscope with high spatial resolution for room temperature samples, Rev. Sci. Instrum. 72
(2001), 2090 2096
[7] P. Grnberg, et al., Layered Magnetic Structures: Evidence for Antiferromagnetic Coupling of Fe Layers
across Cr Interlayers, Phys.Rev.Lett. 57 (1986), 2442
[8] C. Glenske, U.Loreit, New 3D Magnetic Field Sensors with GMR Spin Valve Layers, Paper presented
at the 10th Symposium Magnetoresistive Sensors and Magnetic Systems, 31. March/1. April 2009,
Wetzlar (Germany)
[9] N. Zatsepin, V Shcherbinin, Calculation of the magneto static field of surface defects. I. Field topography
of defect models, Defektoskopija 5 (1966), 50 59
[10] V. Shcherbinin, A. Pashagin, Influence of the extension of a defect on the magnitude its magnetic field,
Defektoskopija 8 (1972), 74 83
[11] F. Frster, New findings in the field of non destructive magnetic leakage field, NDT International 19
(1986), 3 14
Electromagnetic Nondestructive Evaluation (XIV) 225
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-225

Evaluation of EMAT signals using


magnetostrictives for imaging
Yoshihiro NISHIMURA a,1 , Akira SASAMOTO a and Takayuki SUZUKI a
a
National Institute of Advanced Industrial Science and Technology, 1-2-1 Tsukuba
Ibaraki-prefecture, 305-8564, Japan

Abstract. Electromagnetic acoustic transducers (EMATs) is a means of contact


free nondestructive testing (NDT). They are useful for high speed scanning of in
spection materials or for scanning high temperature materials. EMATs have inter
nal permanent magnets that induce a strong magnetic eld in the inspected mate
rials. They employ pulsing current to induce eddy currents on the surface of in
spected materials and generates an ultrasound signal by Lorenz force or by magne
tostrictive effect in the inspected material. The ultrasound signals propagate to and
are reected from the edge of the inspected material or aws. The EMAT probe
then acquires a time series of echo signals as the output, similar to those derived by
piezoelectric transducers. However, the time series of data derived by the EMAT
probe is too noisy due to the Barkhausen effect to identify reections from internal
ows, unlike the time series of data derived by the piezoelectric probe. We em
ployed a low pass lter method and Z direction averaging method to decrease the
Bartkhausen noise and compared them.
Keywords. EMAT, image reconstruction, Hilbert transformation, Barkhausen
noise

Introduction

In these years, technology for imaging aws in structural materials using non-destructive
testing methods has become important for the safety of society. The recent progress
of high speed computational ability of PCs has enabled the widespread use of aws
imaging technology.[1][2][3] Conventionally, piezoelectric probes have been employed
for detecting and imaging aws in structural materials. Piezoelectric probes usually are
in contact with samples directly or are indirectly coupled through water and therefore
cannot easily scan samples quickly. Piezoelectric probes are not available for detecting
aws in large samples.
An electromagnetic acoustic transducer (EMAT) is a major nondestructive (NDT)
device that detects aws in a sample using an ultrasound signal induced by electromag-
netic force.[4] EMAT can address the problem of slow-speed scanning because EMATs
can scan without contacting the samples. An EMAT measures a time series of data, that
is waveforms similar to those produced by ordinary piezoelectric transducers. Two kinds
of EMAT are availables, one using Lorenz force and another using magnetostrictive ef-
1 Corresponding Author: Yoshihiro Nishimura, AIST, 1 2 1 Namiki Tsukuba Ibaraki, Japan; E mail:

nishimura.yoshihiro@aist.go.jp.
226 Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging

fects. The latter can be used for easily detecting aws in a sample with ferromagnetic
oxide scale. However, the data generated by EMAT using magnetostrictive effects are
very noisy due to the Barkhausen effect, compared to data produced by ordinary piezo-
electric transducers. It is important to decrease such noise to reconstruct images of aws
in samples because the noise from Barkhausen effects is sometimes too large to ignore.
Of course, some noises come from electoronic amplier circuit also. Such noise can be
decreased by staying at one position and averaging over a time series of data derived
repeatedly. However, if the probe scans a large surface area on a sample in a limited time
for imaging , the probe cannot remain long at one position for averaging. The frequency
of these noise is much higher than that of the signicant signal data. A low-pass lter,
that can remove high frequency noise, is one way to address this problem but introduce
another kind of noise. Some other method for reducing noise when imaging aws should
be explored.

1. EMAT Principles

An EMAT is used in an NDT method similar to the conventional ultrasound testing (UT),
except that its probe output signal has more noise and that it is a contact-free method.
The EMAT probe has a coil (or coils) and a strong internal magnet (Figure.1). Pulsing

Figure 1. EMAT operation principles

current induces eddy current on the surface of the inspected material as much as the
eddy current testing (ECT) probe does. The induced eddy current generates an ultrasound
signal by Lorenz force or by magnetostrictive effect( Figure.1). The ultrasound signal
propagates to and reects from the edges of the sample or the aws (Figure.2). The
probe output is observed as a time series of echo signal intensities on the EMAT screen(
Figure.3).
A typical time series of data derived by an EMAT from reference specimen is de-
picted in Figure.3. The initial peak induced by the rst current pulse and many subse-
quent peaks from multiple reections at the edge of reference sample can be observed.
The initial peak is too large compared to later peaks and so is saturated in the full input
scale range. This measurement was made at 2GHz.
Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging 227

Figure 2. Signal propagation in EMAT operation

Figure 3. EMAT signal observed from reference specimen sample at 2GHz

2. EMAT Signals From Sample With Oxide Scale Surface

A sample with some internal aws was prepared. Its geometry and dimensions are de-
picted in Figure.4. Its oxide scale was generated in an environment similar to those of
actual use. The 3-D scanner depicted in Figure.5 was developed and applied to the
prepared sample. A FA computer controlled the scanning movement of the EMAT probe
and the trigger timing of the pulser/receiver. An AD converter collected the time series
of data from the receiver and the FA computer stored them in its data strorage. A block
diagram of the 3D-scanner is presented in Figure.6. Scanning was done as illustrated in
Figure.7. The specication of the 3D scanner are listed in Table.1.
The measurement was made under the conguration listed in Table 2. The observed
time series of data is very noisy, as illustrated in Figure. 8(a),(b), but averaging mea-
sured data(time-based-averaging) can decrease such noise (Figure.9(a),(b)). These noisy
signals were observed all over the surface and can be decreased by averaging.
228 Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging

Figure 4. Samples geometry and dimensions

Figure 5. 3D scanner

Table 1. 3D Scanner Specications.


Scan speed 1 line/s(max 1000m/s)
Precision(X,Y,Z) 0.5 m
Maximum image resolution 8192 8192
Measurable area 360mm 310mm 80mm

Table 2. Scanning condition.


Scan area 75mm 75mm
X Y resolution 1024x1024
Sampling rate 400MHz
Sampled Number 16000

If the probe scans a large surface area on a sample in a limited time, the probe cannot
stay long at one position for averaging. The time-based-averaging method thus cannot be
Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging 229

Figure 6. Block diagram of 3D scanning system

Figure 7. The movement of the probes scanning over a sample

Figure 8. (a)Raw signal derived from the sample in Figure.4 and (b) Magnied image of reection signal from
bottom edge.
230 Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging

Figure 9. (a)Signal averaged over 400 raw signals derived from the sample(in Figure.4) and (b)Magnied
image of reection signal in (a)

used for scanning a large area. The power spectra of Figures.8(a) and 9(a) are plotted in
Figures.10(a) and (b). Most of the noise can be minimized by averaging.

Figure 10. Power spectra of Figure.8(a) and Figure.9(a)

A low-pass lter that passes frequencies below 5.5MHz was applied to the time se-
ries data in Figure.8(a). The time series data in Figure.8(a) was converted to the one
shown in Figure.11(a),(b). Most of the noise in Figure.8(a),(b) could be removed success-
fully. However, the low-pass-ltered data reveals noises generated by the lter function
at the initial peak in the left half of Figure.11(a).

3. Noise Reduction By Converting 2GHz-sampled Data to 400MHz-sampled Data

We tried another noise reduction method. In this measurement, the data was sampled
in the X and Y directions at a pitch of 100m by 3D-scanner. However, the sampling
interval in the Z direction is equivalent to a sampling pitch of 7.5m because the velocity
of the transverse wave in steel is 3000m/s and the sampling frequency is 400MHz in
Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging 231

Figure 11. (a)Signal ltered from Figure.8 by a low pass lter (in Figure.11) and (b)Magnied image of
reection signal in (a)

Figure.8(a). Compared to the sampling pitch in the X and Y directions, the sampling
pitch in the Z direction is thought to be too small. We therefore tried 1D-position-based
averaging. First, we derived a time series data by sampling at 2GHz. Second, we averaged
every 50 samples to derive a time series data equivalent to the one sampled at 40MHz.
We tried averaging every 10 samples, every 20 samples , every 50 samples and every 100
samples and decided to employ evey-50-samples-averaging. Sampling the time series
data at 40MHz is though to be equivalent to sampling pitch of 75m in the Z direction.
AD converters supporting under 100MHz or 400MHz are used for common uses of UT
equipments. AD converters supporting 2GHz or more are not popular. But a 2GHz AD
converters was found to be particularly useful for real time noise-reduction and imaging
of aws by such noisy EMAT using magnetostrictive effect.

Figure 12. (a)Signal derived by averaging raw data sampled at 2GHz at every 50 samples raw(in Figure.8)
and (b)Magnied image of reection signal in (a)

In this way, the noise depicted in Figure.8 was decreased successfully. The distribu-
tions of reection intensities in a sample have to be derived to reconstruct a 3D-image
of aws in a sample. By applying Hilbert transformation to the time series data in Fig-
232 Y. Nishimura et al. / Evaluation of EMAT Signals Using Magnetostrictives for Imaging

ure.12, a curve illustrating their envelope could be derived in Figure.13(a). Hilbert trans-
formation provides an analytical expression of the signal and its absolute value provides
an envelope curve of the signal. A sectional image of the sample can be reconstructed by
putting the time T-axis in the vertical direction and the X-axis in the horizontal direction
and allocating the the reection from the aws at a position (X,T) on the X-T plane. An
example of the sectional image of sample is depicted in Figure.13(b). The vertical axis
denotes time and the horizontal axis denotes the position in the X direction. The top edge
of Figure.13(b) means the surface. The reection from the aw and the one from bottom
edge can be identied easily and their depth 33mm can be quite precisely measured in
Figure.13(b) because the velocity of the transverse wave in steel id 3000m/s.

Figure 13. (a)Envelope curve derived by Hilbert transformation and (b) Reconstructed sectional image of
reected signal

4. Conclusion

We investigated the noises of EMAT using magnetostrictives to realize high-speed


inspections of large samples. 1D-position-based-averaging method was evaluated by
converting 2GHz-sampled signal to low frequency-sampled signal. 1D-position-based-
averaging was found to produce better signal waves in real time. 1D-position-averaging
requires the higher performance AD than the one for common use. It maybe a problem
for the on-site use.

References

[1] K.Nakahata, S.Hirose, The study of the qualities of the reconstructed image by array probe , JSNDI
preceedings, (2006),39 42
[2] R.G.Pratt:Seismic waveform inversion in the frequency domain I Theory and verication in a physical
scale model, Geographics (1999), 888 901
[3] Toshiki Watanabe:Full wave inversion, JSNDI Jornal vol 53 (2005), 274 279
[4] Nishimura, Y,. Sasamoto, A,. Suzuki, T,. Study of 3D Image reconstruction using EMAT, Int.J. of Ap
plied Electromagnetics and Mechanics, 28(1), p171 176, 2008
Electromagnetic Nondestructive Evaluation (XIV) 233
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-233

8
  - 9      ,,
  
  5 .
 

 ' 6 (7 $
. Q6#' %7 %
 98#D7  "9I6"%#D
'HSDUWPHQWRI(OHFWURPDJQHWLFDQG%LRPHGLFDO(QJLQHHULQJ)DFXOW\RI(OHFWULFDO
(QJLQHHULQJ8QLYHUVLW\RI=LOLQD8QLYHU]LWQD=LOLQD6ORYDN5HSXEOLF

$EVWUDFW   
 
 )   ,
  -          )
  
     . .
  5  .  
3
6  

        - 
  1 )   .
   , 
     5
5  3 8  5 
 5  )

,  C   5  . 


     -
. - 5 ,1
  3  . 7  . 
5  5  -   ,1
 < 3      
 
  ,         
,, 3

.H\ZRUGV 8
  -   7  
 5 7 <       - 3

,QWURGXFWLRQ


     . ,  ,
-  ,  
, 

- 
     3 6)-7  5   .       . .
 
/6 0 )   
- 
1
  
 
2  5   5   4  1 

  5     5 
 1 , 
-   5 3  
 - 5 
 
     1   ,        1 ,
  
55 .
-3
-     / 90      
- 5  6 5  .
 
3 "  1   5  ,  5  5 7 15    5
5
)    
3 +  .    
- 
 ,
  -   7
-     -7 5
<1
,1    ) ,1  ,
-  
     I

   *+ /*     +   0   

,  1  5
3 

5   .    
,  ,. ,1
   5  9
  ,  5   N  
7      ,  , 7  

)
-   ,1      , .
-   1
    ,. , 

 
     . -3
8
  -   .
  /8 90       .  

  6 3 T  5   .  7   
-   5  ,  5


-        )

 . -  N  
       ,
 -3     
   5 8 9 . 1    5  . 
-  

(
9 ,  > 
 ' 7  ,   5
   T  
 7
 
- 5
 
 7 .  - 5 !
7 . 2  (7 A(A :P !
7 '
.N + ,1
 L

>  B5
323N
234 M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation

1  )   
-  

 5   9 ,  
-  
   7 5

)  1-   

    
    / 03
5     ,, 
.  )     7 :  ?  
 
. 1  
-    8 9 ,1
3  
-  

 5 8 9  1 
   5
   
- 5    5
  C  ) 
     5 / 0  $,
 5 /$07 E(F3 - 55  . 1 
   5
7 1 .   5   )N 
 1     

)  8 9
      ,  
3
 ,,       ,  
- 5  N , 7 ,1
,   )

  . 
,     5
      ,

  .
   . ) 5     ,  3

3XOVHGHGG\FXUUHQWHYDOXDWLRQ

8
  /   0 -        ,  ). 5     
   ).  <   
     -   ,
  
   3 T   ,
  ,      5  ).  5  1 1 5
5 4   7   ,
    

 5 4        , 5    
55    , 3   
.    ,
  -      
)5 4  -
,      1
 , 15   ,  5
)     7 E:F3
 N    
,,         )       1C
   .  5 -        
,       
 3  -    7  8 9    <,   1   5
5   .  ,  ,      .  
9   ) 
 <   3   1   ,
  1 5     5  5
 

- 
  
). 5      ) 1 5 4  -   3
   ) -,  5 8 9  7        
     5
  3 T     N)  1   .   N 
.  5   ,  3 &   5  5   1C    1
7 
 5
 -, 8 9     1  7 E?F7 ESF3

)LJXUH 8 ,
 5   -   7 E:F3
M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation 235

&    ,   5  ,
  -    ,    7 
). 5{ 
  ,   ,   5  5    
5 553 

,   ,  ,
 5      
 N   
- 
7   
 5  ). 5 
   . 

 N   )
1  5  5     7 3 (3
8
  -    55   .  .  
,
5 4  - , 
1    {    
   5 4   3  1 55    
  -   ,     .  

 5 4      1) 7
)   
,
5 4  - -   
- ,    ,  5   , 5
5 4   3      ,  5 4    5 - , 5     )


)- 1 .
1
5 -   -    3   
,
5 4  -
 7  ,  5 4       1     .  )

1 55  
5 55      7  4  . 
  .  ,
<   
 7 E:F3

3(&QXPHULFDOLQYHVWLJDWLRQV

6  

  1    5
      #8 + 5 )
. 1     3   . ,  7 )  3 S7 . 

  ,  5   
 
'' ?(P$  ,     -3
  
   . - 5 K (3?H 'O   
 . ,  1
 - 5
 K (3  ,        .  5 3 8  5   5 7 3 3
   ,  
   .     3:3   ,,  5    


   >
(3  .    5
  5    - 5   5    , 
7
:3  .   . 5  5  <       -  
1    , 
)    5    5     , 3


)LJXUH  5  
   5 7  5   3 S3

5  
      5   ,   
  
      / 0   
 -3
236 M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation

1XPHULFDOVLPXODWLRQUHVXOWV

D ,  5  . 


 . 1      
  
.    5
     , 
3 

) -,  5  . 



. 1   > 4 /'V07 
 /+*07 (OS  ). /(OS'"60  8 %

)  .   3  ,  5  55   
 , 

   . ,  5  . 
  . - 
3 "    
, 5  . 
   5 
- 5
   55   
 ,

3 T  5 - , 5  
 7 
- 
  
 5  8 %
. 
)   , N  5 W K ( 7 3 ?7  ,    1
)3 "
, )   1.     . 
, 7 
 8 % 

    .         ,  7


N 5 <,

4 
3    .      . .  7   
.
 /K$|O 03  8 % ). 5  1  
  5

) 4 >

W :
\ W DWH E
7 /(0

)  D  E    3
< .
 5    . 
 , K (   )  3 ?3


)LJXUH 8 % . 
5  <   
3

55   
.
  ,  1   5  8 % 
 
-2  
   
 
  2  5  5  5 -   ,1 3

&RQILJXUDWLRQ
 
5 1
 -, )  3 S )  1 K PAA 1  5     
. -         , 
3   . - 5  

)    K HP 'O / ,,   . -03  


 ,
  .    
5  ,   

-   ,   5 )  
5 55 5 ( 3
M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation 237

)LJXUH',   )    5   
 3
 5

) 
   
 )  3 PJR ,     ,    5 
55   
 , 
    - 5   5 3   5   5 
)  3 :        ,
-   3 S3 "  1    5 
-
 ,  5   5 7    ,     7    /,   63(0 

    1    )    5   ,   ,,<
- S 3 . 
 ,   5   1   1  1 , 
- .
  5    
 , 3



 
 
 

 







     

)LJXUH55   
 , 
7 ,   63 (7 5  (3




 

 



 



     

)LJXUH55   
 , 
7 ,   63 :7 5  (
238 M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation



 





 



 
     

)LJXUH55   
 , 
7 ,   63 ?7 5  (3

'
   
    ,   63: / 3 :7 S07 )  
- 
  5
  5    

- 55     , .  3   5    , 
5 H    55   
 , 
      )    5 
,,<
- :H  
 3 &  1    ,   
  5   5 
.      7 ,   63 ? / 3 :7 S07  55   
 , 
 5
  5  

-  ,  5      ,    5  ,  5 R 3 
  5   
  )  ,   ,      ,

         5   -3 #5  7  
 5    
 ,  
  5 ,   
 1
  
   5  5  ,  
63 ?3
 5

) 
   
 ,    1
) ) )   , 

)    5     1 1    . 5  5 


<        - 3 6  

   , 5  5 
 
,   63(   5  ( )   )  
    
,    1. 3

&RQILJXUDWLRQ
 5  
 
 )  3 M     . -   3   
 
 
)             , 
3 

5  

  ,   

-   ,   5 )  
5 55 ( 3

)LJXUH95  : 5  <       - 3


M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation 239



 
 

 







 
     

)LJXUH55   
 , 
7 ,   63 (7 5  :3

6  

- 

  55   
 , 
  )  3 R3 "  1
  5   
  1     1 )   
 5  ,   5 
 1 1   )   5  : ,    
     
   )   , N ,1 7 3 H3

95  ?
  , 5  <     , N, 
 5  5  ?  
  
5  :7 1 

 <   
  
    5 5   
1-
SH}7 3 (A3


)LJXUH95  ? 5  <       - 3

 

 





 

 


      


)LJXUH55   
 , 
7 ,   63 (7 5  ?3
240 M. Smetana et al. / Pulsed Eddy Current Testing Application to Defect Evaluation

6  

- 

  55   
 , 
 5  5   ) 
3 ((3 "  1 1 .     
  . - 
      )  
5  :7 3 R3 7  
  5  <   
      
5  .
 5   , 
3

&RQFOXVLRQ

  
,     ,
  -   4     5 ,  5
 .    3 6  
.    . 1      .  
5
   5 . ,  5  . 
7 5  ,1  5   N 
 55   
 , 
3 '
  
  ,      ,, 3 T  
 , 5  
    1 
   ,   5
 

5 
-   , 
 >  , 5  . 
/5
   
- 
, 5   
 
07 .  5 
5 J55 ,  /5
    .

5   
 
03  5

) ,  .   5  5


  
  , 
>     5  . 
  , 5    5 
 - -
)   ,    ,   5 4  - 5  . 
3   7 1 
  , 5  . 5  5  <       - 7  )
)    , 
)  ,,, 5     5
   1-
- 5 3    
 
  . 
  1   
  1 )  .

 , 
 5 55    N  1   .  1- ,,  5  5 
,1 3 T    
  ,   - 5   ,  
- 5  )

,  -     5 5  )N3

$FNQRZOHGJPHQW

 )N ) ,,  1-  '


.N +      .
,    -   
  63 8DDA(RSA@3  )N ) 
 ,,  1-  5  '
.N
*   - D *7 ,C 63 (OA?AMOAM3

5HIHUHQFHV

E(F $7 U37 7 *3U37 '7 3  


-  


 5 ,
  -   .
 3 6~
"  
S( /:AAM07 ,  S@@ SM?3
E:F & 1, )))34
 -  3 7     Q- :H7 :A(A7 E#
 F3
E?F 17 37 
N.7 U37 9 7 !37 ,7 $3 >
     . .
  /"07 "''6
(?M? @:M(7 "#' , 7 D
3 ?(7 :AAM3
ESF Q3 &3 &
7 *3U3  > 3XOVHG HOHFWURPDJQHWLF PDWKRGV IRU GHWHFW GHWHFWLRQ DQG FKDUDFWHULVDWLRQ7
6~ 7 .
3 SA7
 .  $ 37 :AA@7 E#
 F3
Electromagnetic Nondestructive Evaluation (XIV) 241
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-241

An ECT Probe with Widely Spaced Coils


for Local Wall Thinning in
Nuclear Power Plants
Toshihiro YAMAMOTOa, Tetsuya UCHIMOTOb and Toshiyuki TAKAGIb,1
a
Japan Power Engineering and Inspection Corporation
b
Institute of Fluid Science, Tohoku University

Abstract. Currently, local wall thinning on the inner surface of a pipe is evaluated
mostly by ultrasonic thickness measurement from the outer surface of the pipe in
nuclear power plants in Japan. However, it has been pointed out that there are two
major issues of this evaluation method. Firstly, this method is not able to evaluate
local wall thinning from the top of a reinforcing plate that covers the outer surface
of a pipe because this two layer structure hinders the propagation of ultrasound.
Secondly, because ultrasonic thickness measurement of a pipe wall is conducted at
certain intervals on the outer surface of the pipe, it may overlook a hole like defect
if the defect size is small compared to this interval. We propose an eddy current
testing (ECT) method using widely spaced excitation and pick up coils to alleviate
these problems. Our experimental results show this method can be applied to in
spection of a double plate that consists of two 8 mm thick austenitic stainless steel
plates, and a defect indication is observed in a much larger area than the defect.

Keywords. Eddy current testing, Local wall thinning, Pancake coil

1. Introduction

In nuclear power plants, local wall thinning on the inner surface of a pipe is evaluated
mostly by ultrasonic thickness measurement from the outer surface of the pipe. How-
ever, some improvements are still required for thorough and efficient inspection of
local wall thinning to maintain aging nuclear power plants. In Japan, it is considered a
problem that ultrasonic thickness measurement is not able to evaluate local wall thin-
ning from the top of a reinforcing plate that covers the outer surface of a main pipe
around a branch pipe perpendicularly connected to the main pipe because this two-layer
structure hinders the propagation of ultrasound (Figure 1). Also, it is of special concern
that ultrasonic thickness measurement may overlook a small hole caused by liquid
droplet impingement (LDI) because ultrasonic thickness measurement of a pipe wall is
conducted at certain intervals on the outer surface of the pipe and the defect size is
sometimes smaller than this interval. Decreasing the interval of course reduces the
incidence of this problem, but piping maintenance in nuclear power plants already
takes an enormous amount of time, and inspection time is not easily allowed to be
extended any longer. The aim of this study is to propose an inspection method for two-

1
Corresponding Author: Professor, Institute of Fluid Science, Tohoku University, 2 1 1 Katahira, Aoba ku,
Sendai, Miyagi, Japan; E mail: takagi@ifs.tohoku.ac.jp
242 T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning

Figure 1. Reinforcing plate

Figure 2. Absolute probe Figure 3. Inspection procedure

layer part of piping and to prevent overlooking a hole-like defect on a pipe.


The authors have recently reported an eddy current testing (ECT) technique for
evaluating local wall thinning [1]. In this technique, an excitation coil and a pick-up
coil are aligned at a distance to detect a defect in a deep region after the model of re-
mote field ECT usually used for inspection of small tubes with bobbin coils aligned at a
distance [2,3]. Prior to our study, remote field ECT using pancake coils to be applied to
flat geometries has been introduced in [4,5], later specialized in inspection of aircrafts.
In this study, we assess the applicability of our ECT technique to detection of local
wall thinning of a thick-walled double plate and a hole-like defect on a thick-walled
single plate.

2. Experiments

2.1. Experimental setup

The ECT probe we used in this study has one excitation coil and one pick-up coil
shown in Figure 2. The excitation and pick-up coils are 10 mm in outer diameter, 5 mm
in inner diameter and 5 mm in height. The number of turns of the excitation and pick-
up coils is 301 and 3,465 respectively. The wire diameter of the excitation and pick-up
coils is 0.2 mm and 0.05 mm respectively. The center-to-center distance between these
coils was set to be 50 mm. The equipment used in our experiments consists of a func-
tion synthesizer (WF1945B, NF Corporation), a bipolar amplifier (HSA4011, NF Cor-
poration) and a lock-in amplifier (LI5640, NF Corporation). In the experiments, the
detection signals were measured while the pair of excitation and pick-up coils was
moved on the surface of a plate as shown in Figure 3. The excitation frequency was set
to be 1 kHz.

2.2. Specimens

Tables 1 and 2 show the information of the specimens used for evaluation of the per-
T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning 243

Table 1. Rectangular slit specimens

Material 316 austenitic stainless steel

Size 500 mm300 mm8 mm

Defect Rectangular slit


(Width 10 mm, Depth 1,3,5 mm)

Table 2. Square hole specimens

Material 316 austenitic stainless steel

Size 500 mm300 mm12 mm

Defect Square hole


(Bottom 10 mm10 mm, Depth 1,3,5 mm)

Figure 4. Double plate

(a) Plate with a rectangular slit (b) Plate with a square hole

Figure 5. Specimens

formance of our ECT technique. The material of these plates is 316 austenitic stainless
steel. A reinforcing plate part of piping was simulated by two 8 mm-thick plates. An
intact plate was placed on the top of the other plate that has a rectangular slit on the
bottom side as shown in Figure 4. The rectangular slit was cut along a half of the center
line on the bottom side in the shorter direction (Figure 5 (a)). A defect caused by LDI
was simulated by a square hole. The square hole was made at the center of a 12 mm-
thick plate (Figure 5 (b)). For each of the rectangular slit and the square hole, three
plates were prepared so as to vary the depth of these simulated defects (1 mm, 3 mm
and 5 mm).

2.3. Evaluation procedure

The specimens described above were inspected from the opposite side of the defect to
be detected. Figure 6 shows how we scanned the surface of the specimens. For a double
plate with a rectangular slit, five equally-spaced lines along the longer direction of the
plate were selected as the scanning lines. Each of the scanning lines has a length of 150
244 T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning

(a) Rectangular slit (b) Square hole


Figure 6. Scanning areas

(a) Original signals (b) Processed signals


Figure 7. Signal processing

(a) Original distribution (b) Processed distribution

Figure 8. Example of signal processing

mm at 1 mm intervals. The left three lines run above the rectangular slit while the other
two do not. For a square hole, a 150 mm60 mm area at the center of the surface of the
plate was scanned. The intervals of scanning were 1 mm in the longer direction and 2
mm in the shorter direction of the plate.

2.4. Signal processing to reduce lift off noise

Because the specimens used in this evaluation have a wide-area surface (500 mm300
mm) and also have a slight warp, the lift-off distance between the probe and the surface
T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning 245

of the specimens varies with the position of the probe, which causes considerable lift-
off noise. To cope with this problem, we processed raw signals obtained from the lock-
in amplifier.
Figure 7 (a) shows the Vx and Vy components of ECT signals obtained from one of
the scanning lines going over a rectangular slit in Figure 6 (a). It can be observed that
the rectangular slit induces a big change in the signals around the center position. Al-
though the end point of the scanning line is far enough away from the slit, the signal
values do not return to zero at the end point. To make the signal values zero at both
ends of the scanning line, the voltage value on the line that connects both ends of a
signal curve in Figure 7 (a) is subtracted from the voltage value on that signal curve for
each point. Figure 7 (b) is obtained in consequence of this processing.
For a signal distribution obtained as described in Figure 6 (b), the above process-
ing is applied in the X direction as well as in the Y direction so as to make the values
on the boundary of the scanning area zero. Figure 8 shows an example of this signal
processing. The removal of gradual changes in the signal distribution highlights a de-
fect indication due to a square hole.

2.5. Results

Figure 9 shows the signal distributions obtained from the specimens with a rectangular
slit. To simulate a reinforcing plate part, an intact plate was placed on the top of a plate
with a rectangular slit as shown in Figure 4. The results indicate that our ECT tech-

(a) 1 mm deep slit (b) 3 mm deep slit (c) 5 mm deep slit


Figure 9. Signal amplitude distributions (rectangular slit)

(a) 1 mm deep hole (b) 3 mm deep hole (c) 5 mm deep hole


Figure 10. Signal amplitude distributions (square hole)
246 T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning

nique can detect a 1 mm-deep defect on the bottom side of two 8 mm-thick plates.
Figure 10 shows the signal distributions obtained from the specimens with a square
hole. A 1 mm-deep square hole on the bottom side of a 12 mm-thick plate is clearly
recognized in the signal distributions. Defect signals are observed in a much larger area
than the defect (The defect area of a square hole is 10 mm10 mm. The area where a
defect indication appears due to a square hole is about 20 mm40 mm). This property
helps prevent overlooking a small defect.
Figure 11 shows the relationship between the maximum signal amplitude and the
defect depth obtained from Figures 9 and 10. There is an almost linear relationship
between them. It implies that the defect depth can be estimated from the maximum
amplitude of a defect indication.
In above experiments, an LDI defect is simulated by a square hole, but a real LDI
defect tends to have a cone shape. Then, a specimen with a cone-shaped hole shown in
Figure 12 was also used for evaluation of the detectability of an LDI defect with our
ECT technique. The specimen is 300 mm in length, 100 mm in width and 7.1 mm in
thickness. The base diameter and the height of the cone-shaped hole are 10.2 mm and
3.6 mm respectively. Because the surface of this specimen is not large enough, the
scanning area was reduced from the area described in Figure 6 (b) by decreasing the
width in the X direction from 60 mm to 40 mm. Figure 13 (a) shows the signal ampli-
tude distribution obtained by this scanning. The appearance of the defect indication is
almost the same as Figure 10, which was obtained with square holes. Figure 13 (b)
provides the profiles on the lines X=0 and Y=0 in the amplitude distribution. Whereas
the defect is located from -5 mm to 5 mm in both the X and Y directions, a big varia-
tion in signal amplitude is still observed at 10 mm in the X direction and at 20 mm in
the Y direction. This indicates that a defect can be found even if the scanning line is 5
mm away from the edge of the defect with the coils aligned parallel to the scanning line.
When the coils are aligned perpendicular to the scanning line, the defect can be found
even if the scanning line is 15 mm away from the edge of the defect (Figure 14).

(a) Rectangular slit (b) Square hole


Figure 11. Relationship between the maximum signal amplitude and the defect depth

Figur 12. Specimen with a cone shaped hole


T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning 247

(a) Amplitude distribution (b) Amplitude profiles in the X and Y directions


Figure 13. Signal amplitude distribution (cone shaped hole)

(a) Parallel to the scanning line (b) Perpendicular to the scanning line

Figure 14. Scanning directions

3. Summary

The aim of this study is to propose an inspection method to evaluate local wall thinning
on a reinforcing plate part of piping and detect a defect located out of the scanning line.
We proposed an ECT technique using widely spaced excitation and pick-up coils to
detect a defect on the bottom side of a thick-walled plate. The experimental results
show this probe can detect a 1 mm-deep rectangular slit on the bottom side of two 8
mm-thick austenitic stainless steel plates and a 1 mm-deep square hole on the bottom
side of a 12 mm-thick austenitic stainless steel plate. There is an almost linear relation-
ship between the maximum signal amplitude and the defect depth. This implies that the
defect depth can be estimated from the maximum signal amplitude due to a defect.
These defect signals are observed in a much larger area than the defect size. When the
coils are aligned perpendicular to the scanning line, this method can detect a defect
located within 15 mm from the scanning line.
The proposed method can be applied to two-layer structure and detects a defect
even if the defect is not on the scanning line. These properties are expected to alleviate
concern over ultrasonic thickness measurement conducted for evaluation of local wall
thinning of piping in nuclear power plants.
248 T. Yamamoto et al. / An ECT Probe with Widely Spaced Coils for Local Wall Thinning

Acknowledgment

This study was conducted as part of Nuclear and Industrial Safety Agency (NISA)
project on Enhancement of Ageing Management and Maintenance of Nuclear Power
Plants in Japan.

References

[1] T. Yamamoto, T. Takagi and T. Uchimoto, Remote field ECT for evaluation of local wall thinning using
pancake coils, The 14th International Workshop on Electromagnetic Nondestructive Evaluation (ENDE
2009), 130 132.
[2] T. R. Schmidt, The remote field eddy current inspection technique, Materials Evaluation, 42 (2) (1984),
225 230.
[3] T. R. Schmidt, History of the remote field eddy current inspection technique, Materials Evaluation, 47
(1) (1989), 14 22.
[4] Y.S. Sun, S. Udpa, W. Lord and D. Cooley, A remote field eddy current NDT probe for the inspection of
metallic plates, Materials Evaluation, 54 (4) (1996), 510 512.
[5] Y.S. Sun, T. Ouang and S. Udpa, Remote field eddy current testing: one of the potential solutions for
detecting deeply embedded discontinuities in thick and multiplayer metallic structures, Materials Evalua-
tion, 59 (5) (2000), 632 637.
Electromagnetic Nondestructive Evaluation (XIV) 249
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-249

PLL BASED EDDY CURRENT


MEASURING SYSTEM FOR
INSPECTION OF OUTER FLAW IN
TITANIUM ALLOY PLATE
Tomasz CHADY1, Jacek KOWALCZYK, Leon NAWOS-WYSOCKI, Grzegorz PSUJ
and Ireneusz SPYCHALSKI
Department of Electrical and Computer Engineering, West Pomeranian University of
Technology, al. Piastow 17, 70-310 Szczecin, Poland

Abstract. In this paper a new nondestructive eddy current system will be


presented. The system was design for detection of shallow outer flaws in thick
titanium alloy specimens (e.g. plates, discs). A Lock in amplifier was used in a
measuring subsystem in order to increase the efficiency of inspections of minor
defects. The verification of a performance of the system was done using 6.9 mm
titanium plate with artificial EDM (Electrical Discharged Machinated) notches.

Keywords. Eddy current testing, measuring system, titanium alloys inspections

Introduction

Titanium because of its mechanical properties (high corrosion resistance, high strength-
to-weight ratio) is an important material in many leading industries e.g. aerospace. It is
commonly used to manufacture new components such as turbine disc. The disc is
machined from a titanium billet which is a cylindrical solid bar of up to 400mm of a
diameter. In-service failures of such components can often lead to catastrophic
consequences with significant economic impact. NDT techniques that are commonly
utilized to examine traditional aerospace materials are not always sufficient to detect
sub-surface defects created during the manufacturing process in titanium alloys [1]. In
the case of components having large diameter a reasonable technique to apply is
Ultrasound UT. However its effectiveness is limited to shallow regions close to the
components surface. This results in a strong need for new advanced NDT technologies
for the inspection of titanium components during the manufacturing process. We
introduce a novel automated quality control system for the inspection of titanium billets
in response to this need in this paper. The system combines two subsystems: eddy
current testing (ECT) for evaluation of the material not deeper than 5mm beneath the
surface and phased array (PA) ultrasonic for deeper inspection. In this paper the ECT

1
Corresponding Author: West Pomeranian University of Technology in Szczecin, Department of
Electrical and Computer Engineering, ul. Sikorskiego 37, 70 313 Szczecin, Poland; E mail:
tchady@zut.edu.pl.
250 T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw

a) b)

Figure 1. Photo of the measuring system: a) front panel, b) top view.

measurement system is introduced. We verify the performance of the system by


measuring the depth of a defect in a flat titanium plate using a dedicated transducer.

1. ECT measuring system

NDT systems in industrial applications have to be robust in that interference coming


from various sources should not affect measurement results. The presence of noise in
EC signals may mislead the flaw identification algorithms. This is important especially
in the case of low testing frequency used for deeper material evaluation (greater
penetration depth) which however cases week EC response signal (low signal-to-noise
ratio). The most widely used noise suppression technique is digital filtering. It gives
the best results if the signal can be separated from noises in a frequency domain. In
many practical cases this condition is not fulfilled, and therefore filters may cause
distortion of the original signals. It also affects the frequency characteristic. Therefore,
a single frequency system with lock-in amplifier is proposed (Figure 1). Such system
allows to detect signals caused by minor defects and having the amplitude lower than
the noise level by extracting from whole spectrum only that signals component that
has the same frequency as the excitation signal (reference signal). In consequence lock-
in technique allows to increase the sensitivity to minor flaws located deeper in the
material.
The proposed system allows to monitor amplitude and phase responses of a
transducer using single testing frequency ranging from 1 kHz up to 100 kHz. The main
part of the system is a lock-in amplifier based pick-up signal block.

1.1. Portable device of the system

The block diagram of the measuring system is shown in Figure 2. A high speed 16-bit
multifunction data acquisition converter (DAQ) with D/A and A/D modules (NI USB-
6251) are used in the system for both generating and acquiring signals. The DAQ
device in connected via USB interface to the PC class computer which accumulates,
computes data and controls the system. A generated in the D/A DAQ module
sinusoidal signal is gained utilizing a high speed power amplifier (TDA 7294) before
driving transducer's excitation coils. TDA 7294 is very low distortion and noise power
amplifier with DMOS power stage. The driving current was controlled through LEM
Components electronic transducer with no galvanic connection to the main circuit and
was cut off if its value exceeds the maximum calculated for a measuring transducer.
T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw 251

NI 6251 LIA BVD 150 H


REAL
LOCK IN SIGNAL
A/C IMAG IN
LOCK IN

DIGITAL

REF IN
PC

TRANSDUCER
OUT
USB
COMPUTER

C/A
TDA 7294 SWICH

POWER AMPLIFIER
Figure 2. Block diagram of measuring system.

Induced voltage in the pickup coils are amplified by a programmable gain


amplifier in lock-in LIA-BVD-150-H module [2]. It is a dual phase lock-in amplifier
with digital phase shifter and detection of real component, imaginary component and
amplitude of input signal. Such arrangement enables significant noise suppression.
Lock-in amplifiers are commonly used to detect and measure very small AC
signals - all the way down to a few nanovolts. Accurate measurements can be made
even when the small signal is obscured by noise sources many thousands of times
larger. Lock-in amplifiers use a technique known as phase-sensitive detection in order
to separate real and imaginary components of a signal at a specific reference frequency
and phase. Noise signals, at frequencies other than the reference frequency, are rejected
and do not affect the measurement. Signals from lock-in module are then supplied to
A/D DAQ module.

Figure 3. View of the LabView environment software application.


252 T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw

The proposed system works with specialized software operating in the National
Instruments' LabView environment. The software interface is shown in Figure 3. The
application is used to drive the DAQ converter in order to control excitation
instruments, configure PLL operating parameters (such as sensitivity and time
constant) and acquire measuring signals (a real and imaginary part of measuring
signal). All parameters of lock-in amplifier, DAQ and excitation signal can be
introduced to the system through a configuration file. The application enables external
and internal triggering of acquisition events. During the measurements, real time results
of acquired signal, its absolute value, real and imaginary part, phase shift and finally
plot of real versus imaginary part are presented to an operator. In the system there were
three processing algorithms implemented utilizing statistical analysis, gradient and
lowpass filtering. They allows to eliminate trends in the signal caused by the lift off
and surface roughness. The algorithms compute correction values for a signal basing on
value measured at present step and past ones and can be applied in-situ or after the
measurements.

1.2. The transducer

The transducer used in the system is presented in Figure 4. In order to make possible
defect detection located 5 mm beneath the sample surface its dimensions (Figure 4b)
and excitation frequency were optimized using FEM (Finite Element Method) and
Comsol software.
The transducer is built using an E-type ferrite core with excitation coils wound on
exterior columns of the ferrite core and a pick-up coil on a central one. The excitation
coils generate contrary directed fluxes in the pick-up coil. Therefore, the flux flowing
through pick-up coil is close to zero in equilibrium state. If a flaw appearing close to
one of the excitation coils distorts the generated flux flow, the signal different from
zero is induced in the pick-up coil. Such configuration of the transducer allows us for
the optimal usage of the dynamic range of the A/D converter module. In order to
achieve the greatest equilibrium the excitation coils of the transducers were driven by
separate power amplifier.

a) 1 b) c) 4
44
3
1 5
3
25

2 2

Figure 4. Photo and view with dimensions of the transducer: a) photo of the transducer, b) 3D view of the
transducer, c) photo of the transducers head; all dimensions are in [mm]; 1 ferrite core, 2 excitation coils,
3 pick up coil, 4 transducer head, 5 sample
T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw 253

Figure 5. Photo and view of the titanium plate: a) photo, b) drawing of the notch, c) cross section drawing of
the notch; x axis is the scanning axis

2. The verifying experiments

The system is designated for the detection of minor outer flaws OF (casing
discontinuity in the surface at the opposite site of the plate to the scanning side) in thick
titanium specimens. In order to verify the performance of the proposed system several
experiments were carried out.

2.1. Object of the experiment

A titanium alloy plate having artificial EDM notches of different depth d was used
(Figure 5.). The depth was ranging from 10 to 100% of the plate thickness (6.9 mm).
All measurements were carried out for outer flaws.

2.2. Results of the experiment

Results presented in the paper were obtained for 40% notch. First, the A-scan
measurements using different exciting frequencies were carried out in order to find the
optimal one. Figure 6 shows the results of real and imaginary component as well as
amplitude and phase of the signal acquired during the experiments versus excitation
frequency.
From the set of achieved spectrograms one can notice that the frequency range
between 8 and 16 kHz gives the greatest chance to detect the flaw. Considering result
obtained for the real component of the signal (Figure 6a) it is clearly visible that
frequency equal to 11.2 kHz is the optimal one.
In the second stage of the experiments C-scan measurements were done. The two
dimensional distribution of the signals amplitude measured in both scanning direction
over selected area of 40% OF was presented in Figure 7a. In order to better visualize
changes of signal caused by the flaw, plot of amplitude and phase shift as well as real
and imaginary component of a signal measured in selected y line were shown in
Figure 7b and Figure 8. The presented results allow easily to detect the flaw.
254 T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw

a)
f [kHz]
b)

f [kHz]
x [mm] x [mm]
c) d)
f [kHz]

f [kHz]

x [mm] x [mm]
Figure 6. Multi frequency results of measurements obtained for 40% OF: a) real part of measured signal
Re(Usig), b) imaginary part of measured signal Im(Usig), c) amplitude of measured signal |Usig|, d) phase of
measured signal arctg(Im(Usig)/Re(Usig)).

a) b)
Usig [V]
y [mm]

x [mm] x [mm]
Figure 7. Single frequency results obtained for 40% OF: a) amplitude of signal |Usig| measured in selected
area in both scanning direction, b) amplitude of signal |Usig| measured in selected y line.

3. Conclusions

The newly developed ECT system for observation of minor defects in titanium alloy
was presented. The system uses lock-in technique which allows to detect signals having
the amplitude lower than the noise level by taking into consideration only that
component of a signal which frequency is equal to the reference one. In consequence it
T. Chady et al. / PLL Based Eddy Current Measuring System for Inspection of Outer Flaw 255

a) b)
Re(U) [V]

Im(U) [V]
x [mm] x [mm]
c) d)

arctg(Im(U)/ Re(U)) [rad]


Im(U) [V]

Re(U) [V] x [mm]


Figure 8. Single frequency results of 1D scan obtained for 40% OF: a) real part of measured signal Re(Usig),
b) imaginary part of measured signal Im(Usig), c) amplitude of measured signal |Usig|, d) phase of measured
signal arctg(Im(Usig)/Re(Usig)).

leads to increase the possible depth penetration. The results of experiments with the
titanium alloy plate having artificial notches confirm the possibility of using this
system in industrial application. In the future, further experiments will be carried out
for titanium discs with artificial flaws.

Acknowledgement

This work was supported in part by Polish Ministry of Science and Higher Education
and by European Commission sponsored project QualiTi which is a collaboration
between the following organizations: I.S.O.TEST Engineering s.r.l, West Pomeranian
University of Technology (ZUT), Tecnitest Ingenieros S.L., TIMET UK Ltd, TWI Ltd
and Vermon SA. The project is co-ordinated and managed by TWI Ltd and is partly
funded by the EC under the Research for the Benefit of Specific Groups Project (ref:
FP7-SME-2007-1-GA-222476.)

References

[1] National Transportation Safety Board Aircraft Accident Report NTSB AAR 98/01
[2] http://www.femto.de/datasheet/FEMTO_Product_Overview.pdf
256 Electromagnetic Nondestructive Evaluation (XIV)
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-256

Giant Magneto-Resistive Sensor based


Magnetic Flux Leakage Technique for
Inspection of Track Ropes
W. Sharatchandra SINGH, B.P.C. RAO1, S. MAHADEVAN, T. JAYAKUMAR
and Baldev RAJ
Nondestructive Evaluation Division, Indira Gandhi Centre for Atomic Research,
Kalpakkam, TN-603102, India

Abstract. Giant magneto resistive (GMR) sensor based magnetic flux leakage
(MFL) technique is proposed for detection of defects in 64 mm diameter track
ropes. Helmholtz coil is used for magnetization and tangential component of
leakage flux from defects is measured using a GMR sensor. This technique is able
to detect local flaws and loss of metallic cross sectional area type defects in the
track rope.

Keywords. GMR sensor, magnetic flux leakage, track rope, finite element model

Introduction

Steel wire ropes are used for material handling in mines and hauling of men in ski-lift
operations [1]. In wire ropes, corrosion and wear are the main causes for damage such
as formation of local flaws (LF) and loss of metallic cross-sectional area (LMA) [2-4].
LFs are external and internal discontinuities such as broken wires, cracks and corrosion
pitting. Reasons for wire breaking can be due to fatigue, inter-strand nicking or
martensitic embrittlement. LMAs are distributed defects such as missing of wires
caused by corrosion, abrasion and wear resulting in loss of cross-sectional area.
Periodic inspection of wire ropes is important to assess the structural integrity and to
take corrective actions. Nondestructive inspection of wire ropes is challenging due to
the heterogeneous structure of wire ropes, multiplicity, uncertainty of broken wires and
hostile working environment. Visual and magnetic flux leakage (MFL) techniques are
widely used for nondestructive inspection of wire ropes [4]. A variety of procedures
that use different types of sensors and magnetizing devices have been employed for
detection of defects in wire ropes by MFL technique [6-10]. Jomdecha et al. [7] used
printed circuit-shaped coils connected in series as field sensors and solenoid as
magnetization unit for inspection of wire ropes and reported detection of 2 mm deep
surface defects in a 38 mm diameter wire rope. Kalwa et al. [9] developed an MFL
system comprising of magnetic concentrators, Hall sensors and sensing coils and
suggested that measurement of tangential component is more versatile than normal
component for detection of multiple defects in wire ropes.
1
Corresponding Author: Dr. B.P.C. Rao, Head, EMSI Section, NDE Division, Indira Gandhi Centre for
Atomic Research, Kalpakkam, TN 603102, India, E mail: bpcrao@igcar.gov.in.
W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes 257

In this paper, we propose MFL technique for detection of defects on the outer
surface of 64 mm diameter track rope used for transportation of coal. The track ropes
are stationary and are rigidly supported at periodic intervals. They are operated for
about 10 hours every day transporting nearly 3000 tons of coal with the help of 256
numbers of stationary buckets each carrying nearly 1.6 tons of coal. The schematic of
cross-section of the track rope is shown in Figure 1. The track rope has 8 layers of
stranded wires of different diameters as detailed in Table 1. The width of the outer
surface of the Z wire is 6.45 mm and the gap width between two outer Z wires is 0.76
mm. The carriage wheels of the bucket are in contact with the track rope and thus
causing damage to the outer surface of the Z-wire. Service induced surface flaws are
the major causes of failure in these ropes. When more than two Z wires of the outer
layer are broken, they will be separated from the adjacent layers.

Table 1 Track rope cross sectional details.

Layer Wire details Wire diameter or


thickness, mm
1 1 centre wire 4.00
2 6 round wires 4.66
3 6 filler wires 1.93
4 12 round wires 4.33
5 18 round wires 4.27
6 13 round wires 4.95
7 32 Z wires 5.50
8 34 Z wires 6.48

Figure 1. Cross section of track rope.

Recently, GMR sensors are being widely used in MFL testing for detection of
defects in carbon steel plates [10], pipelines [11] in view of their high sensitivity for
low magnetic fields, good signal-to-noise ratio (SNR) and high spatial resolution. GMR
sensors are attractive for measurement of feeble magnetic fields from shallow surface
defects and deeply located sub-surface defects. They are also useful for detection of
stress and fatigue damage [12]. This paper discusses MFL technique proposed using
Helmholtz coil and GMR sensors for detection of defects in the track ropes. The results
of the developmental studies carried out on artificial LFs and LMA defect in track rope
are discussed. In order to interpret the measured MFL signal from LMA defect in the
track rope, 3-D finite element modeling is performed and results are discussed.

1. GMR based MFL Technique

1.1. Measurement Set-up

The schematic of the proposed MFL technique is shown in Figure 2. It consists of


magnetization device, variable DC power supply, GMR sensor, track rope with defects,
amplifier, and personal computer for data acquisition. The magnetization device
consists of a Helmholtz coil with two circular coils of 70 turns each separated by 50
mm. The width and height of each coil are 20 mm and 15 mm respectively. The use of
Helmholtz coil ensures uniform axial magnetization of the track rope. GMR sensor is
258 W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes

used for detection of tangential component (along the scan direction) of leakage flux
from defects and is kept at the middle of the Helmholtz coil.
Measurements are made by moving the GMR sensor and Helmholtz coil together
as a single unit over the track rope in steps of 1 mm and at each axial location GMR
sensor is scanned circumferentially. A constant lift off of 0.3 mm is maintained
between the GMR sensor and the track rope to avoid physical damage to the sensor. In
order to enhance the sensitivity of GMR sensor, its output is first amplified using a
low-noise amplifier consisting of a differential amplifier, notch rejection filter at 50 Hz,
100 kHz low-pass filter and a single-ended variable gain amplifier with DC
suppression. The variable gain amplifier is set such that it amplifies the sensor output
10 times. The amplified GMR sensor output is digitized using a 2-channel data
acquisition system (16 bit) and stored in the computer for analysis.
&RPSXWHU
'&SRZHUVXSSO\ %LDVVXSSO\$PSOLILHU



PP

6FDQGLUHFWLRQ

*05
6HQVRU

:LUH5RSH PP
+HOPKROW]FRLO

Figure 2. GMR sensor based MFL testing set up.

1.2 GMR Sensor

GMR sensor works on giant magneto-resistive effect in which there is a large drop in
electrical resistance of the multilayer for an incident magnetic field due to spin
dependent scattering of electrons at the interface between two ferromagnetic layers
(few nm thick) separated by a nonmagnetic layer. The GMR bridge sensors (AA003-
02) manufactured by M/s NVE Associates are used. These sensors measure differential
output voltage and ensure high stability with low noise (sensitivity ~260 VT -1 at 5V
bias voltage). They exhibit linear response in 0.2-1.3 mT range followed by saturation.

1.3 Artificial Defects

Localized flaws are simulated by introducing electro-discharge machined (EDM)


notches on the outer surface of the track rope. Four axial (A, B, C, D) notches and four
circumferential (E, F, G, H) notches of 5.5 mm length, 2 mm width and of different
depths are machined as shown in Figure 3. The depths of the EDM notches measured
using replica technique are given in Table 2. The distance between the notches is
maintained at 80 mm which is slightly more than the length of Helmholtz
magnetization unit. One LMA type defect is simulated by removing material (42.0 mm
length, 9.2 mm width and 3.0 mm depth) from the outer surface of the Z wire and this
is around 46% of the outer Z-wire (6.48 mm diameter).
W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes 259

         PP


PP PP PP PP PP PP PP PP PP

PP
$ % & ' ( ) * +

PP PP
PP

Figure 3. Schematic of track rope having axial and circumferential machined artificial notches.

1.4 Magnetizing Current

The magnetizing current in the Helmholtz coil is optimized such that detectable leakage
field comes from all the notches. Current is optimized by studying the MFL signal peak
amplitudes for shallow axial notches (A and C) and shallow circumferential notches (E
and G). The signal amplitude is found to increase up to 6 A, beyond that it is decreased
due to the penetration of significant amount of magnetic flux into the inner layer of the
rope, as shown in Figure 4. Hence, a magnetizing current of 6 A is considered optimum
and is used in this study.

2.05 mm deep axial notch (A)


2.5 5.86 mm deep axial notch (C)
1.94 mm deep circum. notch (E)
5.90 mm deep circum. notch (G)
GMR Signal Amplitude, V

2.0

1.5

1.0

0.5

0.0
0 2 4 6 8 10
Magnetizing Current, A

Figure 4. Optimization of magnetizing current.

2. Results and Discussion

2.1. LF defects

GMR signals (tangential component of the leakage flux) of axial notches A, B, C and D
are shown in Figure 5. The GMR sensor output for the shallowest axial notch (2.05 mm
deep) is approximately 2.0 times the background noise that comes mainly from 0.76
mm wide stranded structure of the track rope. As can be observed all the four axial
notches are detected by the technique with two distinct peaks correlated to the edges of
axial notches. To access the detection and sizing capability of the technique, peak
amplitude and Full Width at Half Maximum (FWHM) are determined after Gaussian
fitting and interpolation and the results are shown in Table 2. The error in the
260 W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes

determination of FWHM is 0.1. As can be noted the amplitude of the MFL signals is
found to increase with increase in notch depth. The FWHM is found to be nearly the
same for all the axial notches, because of the constant notch length (5.5 mm).
MFL signals of circumferential notches viz. E, F, G, and H are shown in Figure 6.
The GMR response of circumferential notches is seen sharp with a single peak and the
amplitude of 1.94 mm deep notch is 3.2 times the background noise. As can be seen
from Figure 6, all the four notches are detected by this technique. The signal amplitude
increases up to a depth of 5.90 mm and then decreases while the FWHM remains
nearly constant (refer Table 2). It is observed from Table 2 that the amplitude of the
circumferential notches is 2 to 3 times higher than that of axial notches of similar depth.
However, the FWHM of circumferential notches is found to be nearly 2 to 3 times
lower than that of the axial notches. These observations are due to the axial
magnetization of the track rope causing circumferential notches to leak out higher flux
in comparison to the extended axial notches. Using a simple threshold FWHM, it is
possible to identify axial and circumferential defects in the track rope. Alternately, a set
of array sensors covering the circumference can be used.

1 .2
A x ia l N o tc h e s C D
1 .0

0 .8 B
GMR Output, V

0 .6
A
0 .4

0 .2

0 .0

-0 .2

-0 .4
0 50 100 150 200 250 300
S c a n D is ta n c e , m m

Figure 5. MFL signals of axial notches.

2.5
Circumferential Notches H
G
F
2.0 E
GMR Output, V

1.5

1.0

0.5

0.0

-0.5

0 50 100 150 200 250 300


Scan Distance, mm

Figure 6. MFL signals of circumferential notches.


W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes 261

Table 2. The MFL signal amplitude and FWHM of axial and circumferential notches.
Notch Notch Depth, mm Amplitude, V FWHM, mm
A (axial) 2.05 0.58 4.8
B (axial) 4.11 0.91 5.5
C (axial) 5.86 1.22 4.8
D (axial) 7.91 1.25 4.6
E (circumferential) 1.94 1.86 1.8
F (circumferential) 3.88 2.33 1.6
G (circumferential) 5.90 2.56 2.0
H (circumferential) 8.24 2.50 1.7

The performance of the MFL technique is assessed for detection of a fine


circumferential saw cut that simulates a fatigue crack in the track rope. The saw cut
(length 15 mm, width 0.5 mm and depth 2 mm) and corresponding GMR sensor output
are shown in Figure 7. Clearly GMR sensor has detected the saw cut. The signal
amplitude nearly matches with the MFL signals of circumferential notches of Figure 6.
The signal amplitude is found to be slightly less than the signal amplitude of
circumferential notch E of similar depth due to smaller width of the former. Thus, this
technique can be used for early detection of localized cracks in track ropes.
In order to assess the resolution to multiple cracks, 5 saw cuts (C1, C2, C3, C4 and
C5) are simulated with each of length 23 mm, width 1 mm and depth 2 mm, separated
by a distance of 13.5 mm, 10.0 mm, 5.5 and 3.2 mm and the GMR sensor output is
shown in Figure 8. As can be seen, the technique is able to detect all the 5 saw cuts and
resolve the saw cuts separated by 3.2 mm distance.

2.2. LMA defects

Photograph of LMA defect and its GMR sensor response are shown in Figure 9. The
depth distribution (profile) of the LMA defect is measured using optical method. As
can be observed, the GMR sensor has detected the LMA with good SNR.

1.2

1.0

0.8
GMR Output, V

0.6

0.4

0.2

0.0

-0.2
2 4 6 8 10 12
Scan Distance, mm

Figure 7. Photograph of a 2 mm deep saw cut in track rope and the measured GMR sensor output.
262 W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes

2.0
C1 C2 C3 C4 C5
1.5

1.0

GMR Output, V
0.5

0.0

-0.5

-1.0
0 10 20 30 40 50
Scan Distance, mm

Figure 8. Photograph of series of 2mm deep saw cuts machined in track rope and the corresponding
measured GMR sensor output.

The LMA signal is found, as expected, extended as compared to the LF signal.


There is a decrease in sensor output with respect to the baseline background, observed
for the LMA defect. In order to understand this aspect, 3-D finite element modeling has
been performed using COMSOL 3.4 Multiphysics software package. The model
predicted tangential component of leakage field profile is also shown in Figure 9. In the
model, the relative magnetic permeability and magnetization of the track rope are
assumed as 100 and 106 A/m respectively. There is a good agreement between the
measured and the model predicted signals and the decrease in signal amplitude for
LMA defect appears to be due to the pulling of magnetic field lines into the track rope.

Distance along LMA, mm


-5 0 5 10 15 20 25 30 35 40 45 50
2.8 1
GMR response (experimental)
2.6 GMR response (model)
LMA depth distribution
0
2.4
LMA Depth, mm

2.2
GMR Output, V

-1
2.0

1.8 -2

1.6
-3
1.4

1.2 -4
0 10 20 30 40 50 60 70 80
Scan Distance, mm

Figure 9. Photograph of LMA (42.0 mm length, 9.2 mm width and 3.0 mm depth) machined in the track rope,
GMR sensor response scanned along dotted line direction and optically measured depth profile.
W.S. Singh et al. / GMR Sensor Based MFL Technique for Inspection of Track Ropes 263

3. Conclusion

GMR sensor based MFL technique has been proposed for detection of LF and LMA
type of defects on the outer surface of 64 mm diameter track ropes. Performance of this
technique has been validated by detecting a few saw-cuts and EDM notches simulating
cracks in the track rope. For LMA type defects, a decrease in signal amplitude has been
observed and confirmed by 3-D finite element model. This proposed MFL technique
can be used for detection of both LF and LMA types of defects in track ropes. In order
to further enhance the detection sensitivity and to implement the technique in field,
work on the use of split type coils and array GMR sensors is underway.

4. Acknowledgements

Authors thank Mr. S. Vaidyanathan and Mr. P. Krishnaiah, Electromagnetics,


Modeling, Sensors and Imaging Section, NDE Division, Metallurgy and Materials
Group, IGCAR, Kalpakkam for their help during the experiments.

References

[1] S.S. Udpa and M.O. Patrick, Nondestructive Testing Handbook - Electromagnetic Testing, ASNT, 3rd
Edition, Vol. 5, 2004, p. 437.
[2] D. Basak, S. Pal and D. Chandra Patranabis, In situ non destructive assessment of a haulage rope in a
monocable zigback passenger ropeway, Insight, Vol. 50, No. 3, March 2008, pp. 136 137.
[3] Herbert R. Weischedel and R. P. Ramsey, Electromagnetic testing, a reliable method for the inspection
of wire ropes in service, NDT International, June 1989, pp. 155 161.
[4] Herbert R. Weischedel, The Inspection of Wire Ropes in Service: A Critical Review, Materials
Evaluation, Vol. 43, December 1985, pp. 1592 1605.
[5] N. Sumyong, A. Prateepasen and P. Kaewtrakulpong, Influence of Scanning Velocity and Gap Distance
on Magnetic Flux Leakage Measurement, ECTI Transactions on Electrical Engineering, Electronics
and Communications, Vol. 5, No. 1, February 2007.
[6] E. Kalwa and K. Piekarski, Design of Hall effect sensors for magnetic testing of steel ropes, NDT
International, Vol. 22, No. 5, October 1987, pp. 295 301.
[7] C. Jomdecha and A. Prateepasen, Design of modified electromagnetic main flux for steel wire rope
inspection, NDT&E International, Vol. 42, 2009, pp. 77 83.
[8] Wang Hong yao, Hua Gang and Tian Jie, Research on Detection Device for Broken Wires of Coal
Mine Hoist Cable, Journal of China University Mining & Technology, Vol. 17, No. 3, pp. 376 381.
[9] E. Kalwa and K. Piekarski, Design of inductive sensors for magnetic testing of steel ropes, NDT
International, Vol. 20, No. 6, December 1987, pp. 347 353.
[10] W. Sharatchandra Singh, B. P. C. Rao, S. Vaidyanathan, T. Jayakumar and Baldev Raj, Detection of
Leakage Magnetic Flux from near side and far side Defects in Carbon Steel Plates using Giant
magnetoresistive Sensor, Measurement Science and Technology, Vol. 19, 2008 015702 (8pp).
[11] L. Chen, P.W. Que and T. Jin, A Giant Magnetoresistance Sensor for Magnetic Flux Leakage
Nondestructive Testing of a Pipeline, Russian Journal of NDT, Vol. 41, No.7, 2005, pp. 462 465.
[12] T. Chady and G. Psuj, Data fusion from multidirectional remanent flux leakage transducers for NDT
of stress and fatigue loaded steel samples, IEEE Trans. on Magn., Vol. 44, Nov. 2008, pp. 3285 3288.
This page intentionally left blank
Application of Electromagnetic
Nondestructive Techniques
This page intentionally left blank
Electromagnetic Nondestructive Evaluation (XIV) 267
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-267

Pipe Wall Thickness Measurement by


Electro-Magnetic Acoustic Transducer
using Band Exciting Method
Daigo KOSAKA a, Fumio KOJIMA b,1 and Kosuke UMETANI c
a
Graduate School of Engineering, Kobe University, Kobe, Japan
b
Organization of Advanced Science and Technology, Kobe University, Kobe, Japan
c
Graduate School of System Informatics, Kobe University, Kobe, Japan

Abstract. This paper is concerned with a quantitative nondestructive evaluation of


pipe wall thinning using an Electro-Magnetic Acoustic Transducer (EMAT). First,
two methods for sizing wall thickness, pulse-echo method (PEC) and
electromagnetic resonance measurement technique (EMAR) are examined using
calibration specimen of SS400. It was shown that EMAR is superior to PEC for its
sizing accuracy. Secondly, we discuss a feasible measurement strategy based on
EMAR technique. Finally, the feasibility studies of the method are summarized for
evaluating mock Flow-Accelerated Corrosion (FAC) test samples.

Keywords. Nondestructive test, Condition monitoring, Signal processing, Flow-


accelerated corrosion

Introduction

Phenomena of pipe wall thinning are critical issues on ageing management of nuclear
power plants [1-7]. Pipe wall thinning management in nuclear power plants is aimed at
providing a life management process ensuring replacement or repair prior to in-service
failure. The main objective of pipe wall inspection is to identify the location of
maximum thinning, to ascertain the extent and depth of the thinning, and to evaluate
the wear rate. Currently the conventional technique is ultrasonic testing (UT) which
provides the accurate resolution for the pipe wall thickness measurement. However, in
UT, we must remove insulation from the piping area at each in-service inspection.
There also exist so many inaccessible locations for the in-situ inspection by
surveyors in practical plants. Therefore, continuous surveillance during operation is
much more important than the periodical inspection by UT. Recently, we propose a
monitoring technique of pipe wall thinning using Electro-Magnetic Acoustic
Transducer (EMAT) [8]. EMAT is a non-contacting inspection device that generates an
ultrasonic pulse in the sample inspected [9-13]. The advantage of our method is that
measurements can be remotely implemented and the exciting test signals can be
automatically reproduced by electrical circuits. EMAT consists of a magnet and a coil
of wire and relies on electro-magnetic acoustic interaction for elastic wave generation.

1
Corresponding Author: Fumio KOJIMA, Organization of Advanced Science and Technology, Kobe
University, 1-1 Rokkodai-cho, Nada-ku,Kobe 657-8501; Japan, E-mail: kojima@koala kobe-u.ac.jp
268 D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method

Figure 1 illustrates our experimental apparatus. It is composed of EMAT, a burst power


supply, an oscilloscope and a computer. A using sensor is shown in Figure 2. The
sensor is a vertical type of EMAT which produces SH wave. The magnets generate a
static magnetic field. The coil generates a dynamic magnetic field that induces eddy
currents in a surface of a sample specimen. An ultrasonic wave is generated in the
surface by the static magnetic field and the eddy currents. This type of sensor fits
thickness measurement of pipe. There are two thickness measurement techniques for
EMAT. Pulse-echo method (PEC) is a fundamental technique based on time domain
approach. PEC is ease of use and rapid processing. However, unlike the conventional
UT technique, the detecting signal by EMAT is not so large. Therefore, the use of PEC
is not adequate for pipe wall thickness measurement. Electromagnetic resonance
method (EMAR) is another thickness measurement technique based on frequency
domain [14-15]. The method can estimate wall thickness from seeking a sequence of
peak spectrums of detecting signals corresponding to sweeping frequencies of exciting
test signals. Comparing that PEC method use one single frequency for exciting,
measurement process by EMAR requires tedious treatment of signal processing.
Nevertheless EMAR has advantages on the accurate resolution for sizing thickness of
pipe.
This paper is organized as follows: First, two measurement methods were
compared in the point of thickness measurement method. Second, we propose a
measurement method that uses an exciting voltage with a wide band. A receiving
voltage of a wide band exciting voltage has information of with wide band frequency.
This means this method is able to obtain the resonant frequency at one time. Finally,
the method was applied to evaluate mock Flow-Accelerated Corrosion sample
specimens.

Figure 1. Experimental setup. Figure 2. Top view of EMAT.

1. Thickness measurement by EMAT

Our experimental apparatus is composed of an EMAT, a pulser receiver, an


oscilloscope, an auto stage and a computer. Burst waveform of the excitation voltage of
800Vpp is added to EMAT by pulsar. Two methods for sizing wall thickness are
examined using calibration plate sample specimen of SS400.
D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method 269

1.1. Pulse-echo method

The pulsar signal was taken by two pulses with 2MHz. The reflective waveforms were
recorded by a computer coupled to an oscilloscope. Using PEC, thickness calculation
can be represented by

vt
T= (1)
2
where T , v and t denote thickness, sound velocity and time of flight (TOF). Sound
velocity of SS400 is 3.216103m/s. Detecting waveforms are shown in Figure 3. Figure
3(a) is a waveform of a plate sample specimen with 8mm thickness. Noting that the
time of flight (TOF) t was 4.9s and from Eq. (1), the estimated thickness was taken
as 7.9mm of which error became 0.1mm. Figure 3(b) is a waveform of a plate sample
specimen with 6mm thickness. Since TOF was t = 3.6s, the estimated thickness was
5.8mm where the error was 0.2mm. Taking into account that TOF t becomes
proportional to the thickness, it is more difficult to recognize TOF t for a smaller
thickness.

Figure 3. Detecting waveforms using pulse-echo method.


270 D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method

1.2. Resonance method

EMAR utilizes collections of resonance frequencies by sweeping exciting voltages


with wide frequency range. In our experiment, exciting time is 100s, and exciting
frequency range is 0.6MHz to 2.6MHz. The interval of the sweeping frequency was
taken as 1kHz. For measurements by EMAR, the estimate of thickness can be derived
from

v
T= (2)
2f

where f denotes a peak interval with respect to the frequency. The detecting
waveforms of 4.7mm and 4.9mm thickness are shown in Figure 4. The f is very clear
that, if T is smaller, then the t of pulse-echo method becomes smaller interval while
the f becomes larger interval. The f of 4.7mm thickness and the f of 4.9mm
thickness are clearly distinguishable. This implies that EMAR method is more accurate
measurement technique than PEC method. The f of the detecting waveform of 4.7mm
and 4.9mm thickness are 0.343MHz and 0.330MHz. Detecting thickness of the sample
specimens is 4.69mm and 4.87mm.
Figure 5 depicts the total resolution of the method presented here. In Figure 5, we
admit the detecting accuracy was up to 0.1mm for SS400. On the other hand,
measurement becomes time-consuming because EMAR requires sweeping exciting
frequencies with wide range. In our experiment, each measurement spent about 2
minutes for obtaining sizing estimate.

Figure 4. Resonance frequency using EMAR method


D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method 271

Figure 5. Thickness resolution of resonance method.

2. Measurement method using band exciting in EMAR

In this section, we discuss a feasible measurement strategy for implementing EMAR


method. As already suggested in the previous section, EMAR is time consuming in
order to seek resonant state with wide frequency ranges. Figure 6 and 7 demonstrate
one typical example for measurement strategy using an exciting waveform with a large
frequency band. Figure 6, a frequency band of the exciting waveform was chosen from
0.9MHz to 1.8MHz. Then it was found that the detecting waveform in Figure 7 has
some resonance frequencies at one shot. Hence we can reduce the measurement time by
using a set of exciting signals with wide band.

Figure 6. Exciting waveform with wide band.


272 D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method

Figure 7. Spectrums waveforms of band exciting method.

3. Application to pipe wall thickness measurement

In this section, the band exciting resonance method is applied to evaluate a mock FAC
sample specimen. As shown in Figure 8, a 6B-SS400 pipe was tested and the corrosion
shape was fabricated to simulate flow accelerated corrosion (FAC). The extent of FAC
was taken as 200 mm with the maximum thinning depth b = 1.0mm. Figure 9 shows
estimating sizing results of the sample specimen by our measurement method. For the
purpose of the comparative discussions, the conventional UT techniques were also
performed.

Figure 8. Sample specimen of FAC model (a=160mm, b=1.0mm)


D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method 273

Figure 9. Estimated sizing results for mock Flow-Accelerated Corrosion sample specimen.

4. Conclusions

Two measurement methods for pipe wall thickness measurements were examined
using calibration specimen of SS400. It was shown that EMAR is superior to PEC
for its sizing accuracy. It was shown that the measurement strategy using the
exciting voltage with a wide band has achieved the rapid processing for thickness
measurements. The feasibility studies of the method are summarized for evaluating
mock Flow-Accelerated Corrosion (FAC) test samples. Detecting results of the
method agreed well with detecting results of ultrasonic testing.

Acknowledgments

This work has been performed as a part of the National Research Project for
Enhancement of Measures against the Ageing of Nuclear Power Plants sponsored by
the Nuclear and Industrial Safety Agency (NISA). The authors also gratefully
acknowledge the Institute of Nuclear Safety System Inc. (INSS).

References

[1] C. Schefski, J. Pietralik, T. Dyke, M. Lewis, A physical model to predict wear sites engendered by flow-
assisted corrosion, US DOE Rep, pp. 149-154, 1995.
[2] B. Chexal et al., Flow-accelerated corrosion in power plants, EPRI report, TR-106611, 1996.
[3] R. B. Dooley, V. K. Chexal, Flow-accelerated corrosion of pressure vessels in fossil plants, International
Journal of Pressure Vessels and Piping, Volume 77, Issues 2-3, pp. 85-90, 2000.
[4] Flow-accelerated corrosion in nuclear power plants: Application of CHECWORKS at Darlington,
[5] T.R. Allen, P J. King, and L. Nelson: Flow Accelerated Corrosion and Cracking of Carbon Steel Piping
in Primary Water Operating Experience at the Point Lepreau Generating Station, Proceedings of the 12th
International Conference, pp. 773-784, 2005.
[6] K. D. Efird, Flow accelerated corrosion testing basics, Pap Corros, p. 16, 2006.
[7] S. Uchida, M. Naitoh, Y. Uehara, H. Okada, N. Hiranuma, W. Sugino, S. Kishizuka and D. H. Lster:
Evaluation of Wall Thinning Rate with the Coupled Model of Static Electrochemical Analysis and
Dynamic Double Oxide Layer Analysis, Journal of Nuclear Science and Technology Vol. 46, No. 1, pp.
31-40, 2009.
[8] D. Kosaka, F. Kojima, H. Yamaguchi and K. Umetani, Monitoring System for Pipe Wall Thinning
Management using Electromagnetic Acoustic Transducer, Vol.2, No. 1, pp.34-42, 2010.
274 D. Kosaka et al. / Pipe Wall Thickness Measurement by EMAT Using Band Exciting Method

[9] H.M. Fros, Electromagnetic-ultrasonic transducer: Principles, practice and applications, in: Physical
Acoustics, Vol. 14, W.P. Mason and R.N. Thurston eds., Academic Press, New York, pp. 179-275, 1979.
[10] B.W. Maxfield and C M. Fortunko, The design and use of electromagnetic acoustic wave transducers,
in: Material Evaluation, Vol.41, pp. 1399-1408, 1983.
[11] R.B. Thompson, Physical principles of measurements with EMAT transducers, in: Physical Acoustics,
Vol.19, Academic Press, New York, pp. 157-200, 1990.
[12] K. Mirkhani et al. Optimal design of EMAT transmitters, in: NDT&E International, Vol. 37, pp. 181
193, 2004.
[13] D. MacLauchlan, S. Clark, B. Cox, T. Doyle, and B. Grimmett, Recent advancements in the application
of EMATs to NDE, in: Proc. of the 16th WCNDT 2004, Montreal, Canada.
[14] K. Kawashima, Very high frequency EMAT for resonant measurement, in: Proc IEEE Ultrasonic
Symposium, No. 2, pp.1111-1119, 1994.
[15] H. Ogi, Field dependence of coupling efficiency between electromagnetic field and ultrasonic bulk
waves, J. Appl. Phys. 82 (8), 15, pp. 3940-3949, 1997.
Electromagnetic Nondestructive Evaluation (XIV) 275
T. Chady et al. (Eds.)
IOS Press, 2011
2011 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-60750-750-5-275

Applicability of Magnetic Flux Leakage


Method for Wall Thinning Monitoring in
Nuclear Power Plants
Hiroaki KIKUCHI1, Isamu SHIMIZU, Katsuyuki ARA,
Yasuhiro KAMADA and Satoru KOBAYASHI
NDE & SR-C, Faculty of Engineering, Iwate University, 4-3-5 Ueda, Morioka, Iwate,
020-8551, Japan

Abstract. This paper presents an applicability of magnetic flux leakage (MFL)


method for a nondestructive evaluation (NDE) of wall thinning in nuclear power
plants. Since MFL method has already been applied to other industry field, this
method should be an effective tool in nuclear power plants. The ancillary yoke
poles were employed to adapt to the piping shape, and higher sensitivity was
achieved to estimate the depth in a slit on piping. The distribution of magnetic flux
density corresponding with profile of wall thinning depth was also obtained.
Additionally, an applicability of MFL method for probing a slit fabricated on
superposed specimens was investigated.

Keywords. Wall thinning, magnetic flux leakage (MFL), ancillary yoke pole,
reinforcing plate, dual layer

1. Introduction

An ageing management for nuclear power plants is quite important issue due to its
long-term operation in Japan. One of ageing problem is wall thinning on pipe. Wall
thinning occurs at orifices, elbows and under reinforcing plates, etc., through Flow-
Accelerate Corrosion and Liquid Droplet Impingment erosion. Although several
methods like ultrasonic, eddy current are available tools, the magnetic flux leakage
(MFL) [1-4] also must be an effective nondestructive evaluation (NDE) technique for
the wall thinning, because this method has already been applied to industry field such
as gas pipeline. It is difficult to assess wall thinning under a reinforcing plate, where
steels are superposed for reinforcements, using ultrasonic and eddy current, because of
the reflection of signal and the wall thickness. Therefore MFL method is expected to
apply for an evaluation of wall thinning under the reinforcing plate.
The fundamental experiment of MFL using a single-yoke was examined and it
was clarified that sizing of a rectangular slit fabricated on steel plates is possible [5].
However, a development of a probe feasible for piping and a feasibility study of MFL
for application in a reinforcing plate are required for practical uses. Therefore, in this
study, an applicability of MFL method for wall thinning monitoring in nuclear power
plants was investigated, and a possibility to evaluate wall thinning under reinforcing
1
Corresponding Author: Hiroaki Kikuchi, NDE & Science Research Center, Faculty of Engineering,
Iwate University, 4 3 5 Ueda, Morioka 020 8551, Japan; Phone:+81 19 621 6890; Email: hkiku@iwate
u.ac.jp
276 H. Kikuchi et al. / Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring

plates using MFL was examined. The ancillary yoke poles were utilized to solve the
problem in contact between the magnetic-yoke and the pipe surface and to achieve
higher sensitivity for estimation of slit depth on piping. Additionally, steels superposed
were prepared and a slit modeled wall thinning was formed in underlayer, then an
evaluation of the slit was performed using MFL method.

2. Utilization of ancillary yoke pole

2.1 FEM analysis

Firstly, to confirm an effect of ancillary yoke pole, two-dimensional (2D) FEM


analysis was done. Figure 1(a) shows the analysis model for the calculation. Since a
conventional magnetic-yoke has a flat plane on contact part, there is a large air gap
between the yoke and the surface of the pipe. The magnetic flux density distribution
was calculated along with the dot line shown in Figure 1(a) when the pipe has a slit
with 10 mm width and 2 mm depth and when the pipe has also no slit. The calculation
was done for both x- and z-components of magnetic flux density, Bx, By. An applied
current to the excitation coil was 1 A. Figure 2 shows the calculated results. The results
show the same behavior in both cases being independent of existence of slit. This
means detection of a slit on pipe is difficult if there is a large air gap between the yoke
and the pipe surface.
Figure 1(b) shows the FEM analysis model when ferromagnetic material
exists between the yoke and the surface of the pipe as ancillary yoke poles. The
distribution of both x- and z-components, Bx, By, of magnetic flux density was
calculated along with the dot line shown in Figure 1(b). Figure 3 shows those
calculated field distributions. One can see the notable difference in both distribution
profiles. These calculated results indicate the ancillary yoke pole is very useful to
achieve higher sensitivity.

2.2 Experimental procedure

Figure 4(a) shows the dimension of the magnetic-yoke used for experimental
investigating the utilization of ancillary yoke poles. The geometry and the size of the
specimen are shown in Figure 5. The diameter is 60.5 mm and thickness is 5.2 mm, and
carbon steel STPT is used for their material. A slit with 10 mm width is fabricated at

Air gap Conventional yoke Ferromagnetic Conventional yoke


Coil: 1000 T material Coil: 1000 T

Silicon Iron Silicon Iron


50 50
Coil Coil
60 60

13 Bz 13 Bz
Unit: mm Bx Unit: mm Bx
x x
0 0
9 S15C Pipe 9 S15C Pipe

Slit Slit
(a) with air gap (b) using ancillary yokes
Figure 1. 2D FEM analysis model.
H. Kikuchi et al. / Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring 277

0 800
without slit
Leakage flux density Bx (Gauss)

Leakage flux density Bz (Gauss)


-200 600
with slit
-400 400
-600 200

-800 0

-1000 -200

-1200 -400
without slit
-1400 with slit -600

-1600 -800
-16 -12 -8 -4 0 4 8 12 16 -16 -12 -8 -4 0 4 8 12 16
Position x (mm) Position x (mm)
(a) x component, Bx (b) z component, Bz
Figure 2. Leakage magnetic flux density distribution with air gap between yoke and pipe.

0 100
without slit without slit
Leakage flux density Bx (Gauss)

Leakage flux density Bz (Gauss)


-20
with slit with slit
-40 50
-60

-80 0

-100

-120 -50

-140

-160 -100
-16 -12 -8 -4 0 4 8 12 16 -16 -12 -8 -4 0 4 8 12 16
Position x (mm) Position x (mm)
(a) x component, Bx (b) z component, Bz
Figure 3. Leakage magnetic flux density distribution with ancillary yokes between yoke and pipe.

the center part of the pipes, and its depths are 0.5, 1, 4, 4.5 mm, respectively. The
coordinate for measurement system is defined as shown in Figure 4(a). The center part
of slit is the origin of x-direction (horizontal), and the vertical direction to horizontal is
z-direction (the surface of specimen is z = 0).
Figure 4 (b) shows the geometry and the size of ancillary yoke pole adjusted
for the pipe shown in Figure 5. The material for this pole is pure iron. To obtain soft
magnetic properties, the pole was annealed at 700 C for 1 hour in the atmosphere after
its machining. After annealing, the poles were polished to remove oxide layer.

2.3 Results

Figure 6 plots x- and z- component of magnetic flux density, Bx, Bz against position x
(Here, center of the slit is x = 0). On the measurement, the applied current to the
excitation coil was 3 A, and the magnetic-yoke was fixed as the center of the yoke
corresponding with the center of the slit, and a gauss meter (Lakeshore 460) scanned
the pipe surface in x-direction at z = 0.5 mm. The pipes with slit depth of 4 and 4.5 mm
were used here. The results obtained with ancillary yoke poles are shown as compared
with measurement without ancillary yoke poles. It is clarified that the higher sensitivity
was obtained when the ancillary yoke poles were used. As for the pipe with slit depth
of 4 mm, the intensity of Bx has no change and Bz has no peak at the edge of slit in the
case without ancillary yoke poles, while the profile using ancillary yoke poles shows
large increment in Bx profile and has a peak in Bz at the edge of slit (x = 5 mm). For the
278 H. Kikuchi et al. / Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring

Figure 4. Measuring system for MFL and dimensions of magnetic yoke and ancillary yoke pole.

Wall thinning Wall thinning


d

# 2345mm
w 10 mm

60.5mm

t = 5.2mm 60mm
Figure 5. Dimension of pipe with slit.

-20 200
Magnetic flux density Bx (Gauss)

Magnetic flux density Bz (Gauss)

-40
100
-60

-80 0

-100 -100
-120 4mm NAY 4mm NAY
4 5mm NAY -200 4 5mm NAY
-140 4mm AY 4mm AY
4 5mm AY 4 5mm AY
-160 -300
-2 0 2 4 6 8 10 12 14 -2 0 2 4 6 8 10 12 14
Position x (mm) Position x (mm)
(a) x component, Bx (b) z component, Bz
Figure 6. Experimental leakage magnetic flux density distribution with and without ancillary yoke poles
between yoke and pipe (AY: with ancillary yoke, NAY: without ancillary yoke).

pipe with slit depth of 4.5 mm, the changes in Bx and the peak intensity in Bz at x = 5
mm were enhanced using ancillary yoke poles.
Figure 7 plots x- and z- component of magnetic flux density, Bx, Bz against
position x, when the probe includes the magnetic-yoke, ancillary yoke poles and the
gauss meter scanned the pipe surface in x-direction (In this case, gauss meter and
magnetic-yoke scanned together). The depth of slit changes from 0.5 to 4.5 mm. The x-
component Bx of flux has rapid increase at x = 5 mm, and its rate of change increases
with increasing slit depth. On the other hand, the z-component Bz of flux has a peak at x
= 5 mm, and its intensity increases with increasing depth of slit. The position x = 5 mm
is consistent with the edge of slit.

3. Evaluation of wall thinning on pipe

Pipe specimens with modeled wall thinning being similar to the practical cases were
H. Kikuchi et al. / Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring 279

-20 300
Magnetic flux density Bx (Gauss)

Magnetic flux density Bz (Gauss)


d = 0.5 mm
-40 250 d = 1 mm
d = 4 mm
-60
200 d = 4.5 mm
-80
150
-100
100
-120
d = 0.5 mm 50
-140 d = 1 mm
d = 4 mm 0
-160 d = 4.5 mm
-180 -50
-5 0 5 10 15 20 25 -5 0 5 10 15 20 25
Position x (mm) Position x (mm)
(a) x component, Bx (b) z component, Bz
Figure 7. Experimental leakage magnetic flux density distribution with ancillary yoke poles, when slit
depth d changes.

60

14 20

57
63.4 Unit: mm

Figure 8. Dimension of pipe with wall thinning. a is Figure 9. Dimension of ancillary yoke
maximum depth, b width of slit and ' central angle of wall pole.
thinning part.
prepared, and MFL method using ancillary yoke poles was examined to assess wall
thinning of those specimens, here.
The configuration and the size of pipe are shown in Figure 8. Its diameter is
114.3 mm and its thickness is 8.6 mm, respectively. The material of the pipe is carbon
steel STPG. We prepared two specimens, A and B. The value of a, b and ' shown in
Figure 8 are 4.4 mm, 100 mm, 130 deg., respectively for specimen A, and those values
for specimen B are 2.15 mm, 50 mm, 90 deg., respectively. The magnetic-yoke used
here was the same as mentioned in Figure 1 (a), while another ancillary yoke poles
adjusted for the pipe shown in Figure 8 were prepared. Figure 9 shows the geometry
and the size of ancillary yoke pole. The gauss meter was fixed at the center part of the
magnetic-yoke, and the magnetic-yoke with the ancillary yoke poles scanned the pipe
surface in x-direction. The x- component of magnetic flux density, Bx was measured
with application of 3 A current to the excitation coil. The scans were performed along
with the line passing through the deepest wall thinning point in both A and B
specimens and were also performed along with the line passing through out of wall
thinning area; this means scanning excludes wall thinning part.
Figure 10 shows the distribution of x- component of magnetic flux density Bx
against position x. For both specimen A and B, magnetic flux density is constant on the
line passing through out of wall thinning area, while the distributions on the line
passing through deepest wall thinning point reflect the depth profiles of wall thinning
in each specimen (the cases of A: a = 4.4 mm and B: b = 2.15 mm). The magnetic flux
density at deepest wall thinning point in specimen A is larger than that of specimen B.
At deeper wall thinning point, larger magnetic flux density is obtained.
280 H. Kikuchi et al. / Applicability of Magnetic Flux Leakage Method for Wall Thinning Monitoring

40

Magnetic flux density Bx (Gauss)


Specimen A: Wall thinning area
35

30

Specimen B: Wall thinning area


25
A a 4 4 mm
A line out of slit
B a 2 15 mm
B line out of slit
20
-100 -50 0 50 100