You are on page 1of 317

ELECTROMAGNETIC NONDESTRUCTIVE

EVALUATION (XI)

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 31
Previously published in this series:
Vol. 30.
Vol. 29.
Vol. 28.
Vol. 27.
Vol. 26.
Vol. 25.
Vol. 24.
Vol. 23.
Vol. 22.
Vol. 21.
Vol. 20.
Vol. 19.
Vol. 18.
Vol. 17.
Vol. 16.
Vol. 15.
Vol. 14.
Vol. 13.
Vol. 12.

S. Wiak, A. Krawczyk and I. Dolezel (Eds.), Advanced Computer Techniques in


Applied Electromagnetics
A. Krawczyk, R. Kubacki, S. Wiak and C. Lemos Antunes (Eds.), Electromagnetic
Field, Health and Environment Proceedings of EHE07
S. Takahashi and H. Kikuchi (Eds.), Electromagnetic Nondestructive Evaluation (X)
A. Krawczyk, S. Wiak and X.M. Lopez-Fernandez (Eds.), Electromagnetic Fields in
Mechatronics, Electrical and Electronic Engineering
G. Dobmann (Ed.), Electromagnetic Nondestructive Evaluation (VII)
L. Udpa and N. Bowler (Eds.), Electromagnetic Nondestructive Evaluation (IX)
T. Sollier, D. Prmel and D. Lesselier (Eds.), Electromagnetic Nondestructive
Evaluation (VIII)
F. Kojima, T. Takagi, S.S. Udpa and J. Pv (Eds.), Electromagnetic Nondestructive
Evaluation (VI)
A. Krawczyk and S. Wiak (Eds.), Electromagnetic Fields in Electrical Engineering
J. Pv, G. Vrtesy, T. Takagi and S.S. Udpa (Eds.), Electromagnetic Nondestructive
Evaluation (V)
Z. Haznadar and . tih, Electromagnetic Fields, Waves and Numerical Methods
J.S. Yang and G.A. Maugin (Eds.), Mechanics of Electromagnetic Materials and
Structures
P. Di Barba and A. Savini (Eds.), Non-Linear Electromagnetic Systems
S.S. Udpa, T. Takagi, J. Pv and R. Albanese (Eds.), Electromagnetic
Nondestructive Evaluation (IV)
H. Tsuboi and I. Vajda (Eds.), Applied Electromagnetics and Computational
Technology II
D. Lesselier and A. Razek (Eds.), Electromagnetic Nondestructive Evaluation (III)
R. Albanese, G. Rubinacci, T. Takagi and S.S. Udpa (Eds.), Electromagnetic
Nondestructive Evaluation (II)
V. Kose and J. Sievert (Eds.), Non-Linear Electromagnetic Systems
T. Takagi, J.R. Bowler and Y. Yoshida (Eds.), Electromagnetic Nondestructive
Evaluation

Volumes 16 were published by Elsevier Science under the series title Elsevier Studies in
Applied Electromagnetics in Materials.
ISSN 1383-7281

Electromagnetic Nondestructive
Evaluation (XI)

Edited by

Antonello Tamburrino
University of Cassino, Italy

Yevgen Melikhov
Cardiff University, UK

Zhenmao Chen
Xian Jiaotong University, China

and

Lalita Udpa
Michigan State University, USA

Amsterdam Berlin Oxford Tokyo Washington, DC

2008 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 978-1-58603-896-0
Library of Congress Control Number: 2008932761
Publisher
IOS Press
Nieuwe Hemweg 6B
1013 BG Amsterdam
Netherlands
fax: +31 20 687 0019
e-mail: order@iospress.nl
Distributor in the UK and Ireland
Gazelle Books Services Ltd.
White Cross Mills
Hightown
Lancaster LA1 4XS
United Kingdom
fax: +44 1524 63232
e-mail: sales@gazellebooks.co.uk

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 (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.

Preface
The 12th International Workshop on Electromagnetic Nondestructive Evaluation
(ENDE07) was held from 19th21st June 2007. The Workshop was hosted by the
Wolfson Centre for Magnetics at Cardiff University, Cardiff, Wales, UK with sponsorship from Cedrat SA, Serco Assurance, Rolls-Royce plc, Welsh Assembly Government,
Computer Simulation Technology, The Japan Society of Applied Electromagnetics and
Mechanics, Engineering and Physical Sciences Research Council. The organizers
gratefully acknowledge their support.
The aim of this annual workshop is to bring together engineers and scientists from
universities, research institutions and industry to discuss and exchange the latest ideas
and findings in basic research and development as well as industrial applications of
Electromagnetic Nondestructive Evaluation.
After the introductory welcoming remarks from Dr. David Grant (Vice Chancellor
of the Cardiff University), Prof. Hywel Thomas (Head of the School of Engineering,
Cardiff University) and Prof. David Jiles (Chairman of the Workshop), the technical
program of the Workshop commenced with a plenary talk NDE Research Makes a
Difference by Prof. Chris Scruby, Director, UK Research Centre in NDE, Imperial
College London, U.K. Four distinguished invited speakers discussed the challenges and
achievements in various fields of ENDE. Prof. J. Bowler (Iowa State University, USA),
gave a talk titled Integral methods for calculating the interaction of eddy currents with
cracks. Prof. K. Miya (Keio University, Japan), presented the second invited talk on
The Start of a New Field of Electromagnetic and Mechanical Maintenance Engineering. Prof. G. Dobmann (Fraunhofer-Institute for Non-destructive Testing, Germany),
was invited to present Industrial Applications of 3MA Micromagnetic Multiparameter Microstructure and Stress Analysis. Finally Prof. N. Takahashi (Okayama University, Japan) presented an invited talk on 3D Nonlinear Eddy Current Analysis of Electromagnetic Inspection of Defects in Steel. A total of 75 technical papers were divided
into 30 oral and 45 poster presentations. The oral presentations were organized into 7
sessions covering a variety of topics on both theoretical and experimental aspects of
NDE in eddy currents, magnetic measurements, magnetic flux leakage, Barkhausen
methods, new methods and inverse problems for crack detection. These sessions were
chaired by experts in the field including Profs. S. Udpa, L. Udpa, D. Jiles, C. Scruby,
P. Nagy, T. Moses, G. Dobmann, K. Miya, S. Takahashi, A. Tamburrino and others.
During closing remarks it was announced that the next ENDE Workshop (ENDE2008)
will be held June 1012, 2008 in Seoul, Korea. The ENDE Workshop 2009 will be
held in Dayton, Ohio, U.S.A. The conference concluded with remarks from the chairman Prof. David Jiles.
A total of 73 participants from 16 countries were registered for the Workshop. The
short versions of the papers were published in the Workshop digest and 39 reviewed
full papers were accepted for publication in this proceeding. The organizers would like
to thank all the participants for their contribution and all the referees for their role in
reviewing the full papers. Lastly, the editors gratefully acknowledge the help and hard
work of Ms. Linda Clifford in putting this volume together.

This page intentionally left blank

vii

List of Referees
J. Aldrin
Z. Badics
J. Bowler
N. Bowler
T. Chady
M. Chan
Z. Chen
W. Cheng
G. Dobmann
Y. Gotoh
R. Grimberg
X. Hao
H. Huang
G. Hwang
L. Janousek
H. Kikuchi
S. Kobayashi
F. Kojima
J. Lee
D. Lesselier
L. Li
Y. Melikhov
V. Melapudi
O. Mihalache
T. Moses
G. Ni
J. Pv
P. Ramuhalli
G. Rubinacci
O. Stupakov
J. Taggart
A. Tamburrino
T. Takagi
N. Takahashi
S. Takahashi
T. Theodoulidis
G.Y. Tian
I. Tomas
Y. Tsuchida
L. Udpa
S. Udpa
M. Vaidhianathasamy
S. Ventre

Computational Tools, USA


Rhythmia Medical, Inc, USA
Iowa State University, USA
Iowa State University, USA
Technical University of Szczecin, Poland
Michigan State University, USA
Xian Jiaotong University, China
Tsurumi R&D Center, JAPEIC, Japan
Fraunhofer-IZFP University, Germany
Oita University, Japan
National Institute of R&D for Technical Physics, Romania
University of Birmingham, United Kingdom
IIU Corporation, Japan
Shanghai Jiaotong University, China
University of Zilina, Slovakia
Iwate University, Japan
Iwate University, Japan
Kobe University, Japan
Chosun University, Korea
Suplec, France
Tsinghua University, China
Cardiff University, United Kingdom
Michigan State University, USA
JAEA, Japan
Cardiff University, United Kingdom
Zhejing University, China
Budapest University of Technology and Economics, Hungary
Michigan State University, USA
Universit degli Studi di Napoli Federico II, Italy
Tohoku University, Japan
Serco Assurance, United Kingdom
University of Cassino, Italy
Tohoku University, Japan
Okayama University, Japan
Iwate University, Japan
University of Western Macedonia, Greece
University of Newcastle upon Tyne, UK
Institute of Physics, Czech Republic
Oita University, Japan
Michigan State University, USA
Michigan State University, USA
University of Newcastle Upon Tyne, United Kingdom
University of Cassino, Italy

viii

M. Versaci
F. Villone
J. Wilson
L. Xin
N. Yusa
Z. Zeng

Universit Mediterranea di Reggio Calabria, Italy


University of Cassino, Italy
University of Newcastle Upon Tyne, United Kingdom
Michigan State University, USA
IIU Corporation, Japan
Michigan State University, USA

ix

Organizing Committees
International Committee
Chairman

S. Udpa, Michigan State University, USA

Members

J.R. Bowler, Iowa State University, U.S.A.


N. Bowler, Iowa State University, U.S.A.
Z. Chen, Xian Jiaotong University, China
G. Dobmann, Fraunhofer Institute for NDT, Germany
H.K. Jung, Seoul National University, South Korea
F. Kojima, Kobe University, Japan
D. Lesselier, DRE-LSS CNRS-SUPELEC-UPS, France
K. Miya, Keio University, Japan
G.Z. Ni, Zhejing University, China
J. Pavo, Budapest University, Hungary
G. Pichenot, CEA SACLAY, France
A. Razek, LGEP CNRS-SUPELEC-UPS-UPMC, France
G. Rubinacci, Universita di Napoli Federico II, Italy
S.J. Song, Sung Kwan University, Korea
T. Takagi, Tohoku University, Japan
S. Takahashi, Iwate University, Japan
A. Tamburrino, Universita degli Studi di Cassino, Italy
L. Udpa, Michigan State University, U.S.A.

Organizing Committee

Tony Dunhill, Rolls Royce Plc


Keith Jenkins, Cogent Power
David Jiles, Cardiff University (Chairman)
Chester Lo, Iowa State University
Tony Moses, Cardiff University
Ian Nicholson, TWI Ltd
Allan Rogerson, Serco Assurance Plc
Chris Scruby, Imperial College London
Gui Yun Tiann, University of Huddersfield

Local Committee

Phil Anderson
Jeremy Hall
Eugene Melikhov
John Snyder
Paul Williams
Stan Zurek

This page intentionally left blank

xi

List of Participants
Mr. Kavoos Abbasi
Tohoku University, Japan
abbasi@karma.qse.tohoku.ac.jp

Dr. Dagmar Faktorova


University of Zilina, Slovak Republic
faktor@fel.uniza.sk

Prof. Purnachandra Rao Bhagi


Indira Ghandi Centre for Atomic
Research, India
bpcrao@igcar.gov.in

Mr. Fabrice Foucher


CEDRAT, France
karoline.ballini@cedrat.com

Prof. John Bowler


Iowa State University, USA
jbowler@iastate.edu

Dr. Raimond Grimberg


National Inst. of R & D for Technical
Physics, Romania
grimberg@phys-iasi.ro

Dr. Nicola Bowler


Iowa State University, USA
nbowler@iastate.edu

Dr. Xinjiang Hao


Birmingham University, UK
x.hao@bham.ac.uk

Dr. John Burd


JB Consulting Ltd, UK
jbc.ltd@ntlworld.com

Dr. Jeremy Hall


Wolfson Centre for Magnetics, UK
halljp@cf.ac.uk

Hee Jun Chen


Sungkyunkwan University, Korea
rokmc763@ckku.edu

Mr. Jiseong Hwang


Chosun University, Republic of Korea
atithi1004@nate.com

Dr. Zhenmao Chen


Xian Jiaotong University, China
chenzm@mail.xjtu.edu.cn

Dr. Richard Ireland


QinetiQ, UK
rcireland@qinetiq.com

Prof. Gerd Dobmann


Fraunhofer-Institut IZFP, Germany
gerd.dobmann@izfp.fraunhofer.de

Dr. Ladislav Janousek


University of Zilina, Slovak Republic
janousek@fel.uniza.sk

Mr. David Edgar


University of Nottingham & Qinetiq, UK
dedgar@qinetic.com

Dr. Mohan Jayawardene


CST, UK/Germany
mohan.jayawardene@cst.com

Dr. Christiaan Eggink


Shell Global Solutions Int., Netherlands
christiaan.eggink@shell.com

Dr. Steve Jenkins


Current Enterprises, UK
stevej@current.co.uk

xii

Prof. David Jiles


Wolfson Centre for Magnetics, UK
jilesd@cf.ac.uk

Dr. Dmitriy Makhnovskiy


University of Plymouth, UK
dmakhnovskiy@plymouth.ac.uk

Dr. Tonphong Kaewkongka


Chulalongkorn University, Thailand
tonphong.k@chula.ac.th

Dr. Eugene Melikhov


Wolfson Centre for Magnetics, UK
melikhov@cf.ac.uk

Mr. Yuchiro Kai


Oita University, Japan
ykai@cc.oita-u.ac.jp

Prof. Kenzo Miya


IIU, Japan
miya@iiu.co.jp

Dr. Hiroaki Kikuchi


NDE & SRC Ueda, Japan
hkiku@iwate-u.ac.jp

Prof. Tony Moses


Wolfson Centre for Magnetics, UK
mosesaj@cf.ac.uk+D70

Mr. Jeremy Knopp


Air Force Research Laboratory, USA
jeremy.Knopp@wpafb.af.mil

Dr. Shoichiro Nagata


University of Miyazaki, Japan
nagata@ee.miyazaki-u.ac.jp

Dr. Satoru Kobayashi


NDE & Science Research Centre, Japan
koba@iwate-u.ac.jp

Prof. Peter BN Nagy


University of Cincinnati, USA
peter.nagy@uc.edu

Prof. Fumio Kojima


Kobe University, Japan
kojima@koala.kobe-u.ac.jp
Dr. Yann Le Bihan
LGEP- SUPELEC, France
yann.le-bihan@lgep.supelec.fr
Prof. Jinyi Lee
Chosun University, Republic of Korea
jinyilee@chosun.ac.kr

Severine Paillard
CEA, France
severine.paillard@cea.fr
Dr. Manuele Papais
University of Udine, Italy
papaismanuele@libero.it
Dr. Grgoire Pichenot
CEA, France
gregoire.pichenot@cea.fr

Dr. Yohan Le Diraison


ENS-CACHAN, France
yohan.le-diraison@satie.ens-cachan.fr

Mr. Grzegorz Psuj


Szczecin University of Technology,
Poland
gpsuj@ps.pl

Dr. Dominique Lesselier


SUPELEC, France
lesselier@lss.supelec.fr

Mr. Brian Radtke


Iowa State University, USA
bradtkc@iaste.edu

Prof. Li Luming
Tsinghua University, China
lilm@tsinghua.edu

Miss Alicia Romero Ramirez


Swansea University, UK
a.romero-ramirez.436912@swan.ac.uk

xiii

Dr. Cyril Ravat


ENS-CACHAN, France
ravat@satie.ens-cachan.fr

Prof. Norio Takahashi


Okayma University, Japan
norio@elec.okayama-u.ac.jp

Dr. Alan Rogerson


Serco Assurance, UK

Prof. Seiki Takahashi


Iwate University, Japan
seiki.t@iwate-u.ac.jp

Dr. Adriana Savin


National Inst. of R & D for Technical
Physics, Romania
asavin@phys-iasi.ro

Prof. Antonello Tamburrino


Daeimi University of Cassino, Italy
tamburrino@unicas.it

Prof. Chris Scruby


Imperial College, UK
c.scruby@imperial.ac.uk

Dr. Alan Tassin


ENS-CACHAN, France
tassin@satie.ens-cachan.fr

Dr. Gongtian Shen


CSEI, China
shen-gongtian@csei.org.cn

Dr. Theodoros Theodoulidis


University of West Macedonia, Greece
theodoul@uowm.gr

Prof. Young-Kil Shin


Kunsan National University, South Korea
ykshin@kunsan.ac.kr

Prof. Gui Yun Tian


Newcastle University, UK
g.y.tian@ncl.ac.uk

Dr. Anastassios Skarlatos


CEA, France
gregoire.pichenot.cea.fr

Mr. Yuji Tsuchida


Oita University, Japan
tsuchida@cc.oita-u.ac.jp

Dr. Sung-Jin Song


Sungkyunkwan University, Korea
sjsong@skku.edu
Mr. Giuseppe Sposito
Imperial College, UK
g.sposito@imperial.ac.uk
Dr. Vladamir Syasko
Constanta, Russia
office@constanta.ru
Dr. John Taggart
Serco Assurance, UK
john.taggart@sercoassurance.com
Prof. Toshiyuki Takagi
Tohoku University, Japan
takagi@ifs.tohoku.ac.jp

Dr. Lalita Udpa


Michigan State University, USA
udpal@egr.msu.edu
Prof. Satish Udpa
Michigan State University, USA
udpa@cgr.msu.edu
Dr. Tetsuya Uchimoto
Tohoku University, Japan
uchimoto@ifs.tohoku.ac.jp
Dr. Moorthy Vaidhianathasamy
Newcastle University, UK
v.moorthy@ncl.ac.uk
Dr. Haitao Wang
Nanjing University of Aeronautics &
Astronautics, China
whtlqf@hotmail.com

xiv

Dr. Casper Wassink


Applus RTD, The Netherlands
casper.wassink@applusrtd.com

Mr. John Wilson


Newcastle University, UK
john.wilson2@ncl.ac.uk

Mr. Chris Ward


University of Nottingham/RWE NPower,
UK
chris.ward@rwenpower.com

Mr. Chen Xing


Tsinghua University, China
chenx98@mails.tsinghua.edu.cn

Dr. Paul Williams


Wolfson Centre for Magnetics, UK
williamspi@cf.ac.uk

Dr. En Tao Yao


Nanjing University of Aeronautics &
Astronautics, China
yaoentao@yahoo.com.cn

xv

Contents
Preface

List of Referees

vii

Organizing Committees

ix

List of Participants

xi

Invited Speakers
A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering
Kenzo Miya
NDE Research Makes a Difference
C.B. Scruby

3
10

Industrial Applications of 3MA Micromagnetic Multiparameter Microstructure


and Stress Analysis
Gerd Dobmann, Iris Altpeter, Bernd Wolter and Rolf Kern

18

3D Nonlinear Finite Element Analysis of Electromagnetic Inspection of Defects


in Steel
N. Takahashi and Y. Gotoh

26

Magnetic Materials
Evaluation of Irradiation Embrittlement in Fe-Cu-Ni-Mn Model Alloys by
Measurements of Magnetic Minor Hysteresis Loops
Satoru Kobayashi, Hiroaki Kikuchi, Seiki Takahashi, Katsuyuki Ara and
Yasuhiro Kamada
Analysis of Barkhausen Noise Characteristics and Mechanical Properties on
Cold Rolled Low Carbon Steel
Hiroaki Kikuchi, Tomoki Koshika, Tong Liu, Yasuhiro Kamada,
Katsuyuki Ara, Satoru Kobayashi and Seiki Takahashi
A Bridge Between NDE and Charpy Impact Testing
Seiki Takahashi and Satoru Kobayashi
ND-Materials Characterization of Neutron-Induced Embrittlement in German
Nuclear Reactor Pressure Vessel Material by Micromagnetic NDT Techniques
Gerd Dobmann, Iris Altpeter, Melanie Kopp, Magdalena Rabung and
Gerhard Hbschen
Evaluation of Chill Contents in Flake Graphite Cast Irons Using AC
Magnetization Method
Tetsuya Uchimoto, Jun Matsukawa, Toshihiko Abe, Toshiyuki Takagi,
Takeshi Sato, Hiroyuki Ike, Takahito Takagawa and Noritaka Horikawa

37

42

46

54

62

xvi

Characterisation of Microstructures in Heat Treated Maraging Steel Using Eddy


Current and Barkhausen Emission Techniques
K.V. Rajkumar, B.P.C. Rao, B. Sasi, S. Vaidyanathan, T. Jayakumar and
Baldev Raj

70

Novel Acoustic Barkhausen Noise Transducer and Its Comparison with


Electromagnetic Acoustic Transducer
John Wilson, Gui Yun Tian, Rachel S. Edwards and Steve Dixon

78

Modelling and Measurement of Decarburisation of Steels Using a


Multi-Frequency Electromagnetic Sensor
X.J. Hao, W. Yin, M. Strangwood, A.J. Peyton, P.F. Morris and C.L. Davis

86

Assessment of Grinding Damage on Gear Teeth Using Magnetic Barkhausen


Noise Measurements
Moorthy Vaidhianathasamy, Brian Andrew Shaw, Will Bennett and
Peter Hopkins
Evaluation of Contact Fatigue Damage on Gears Using the Magnetic Barkhausen
Noise Technique
Moorthy Vaidhianathasamy, Brian Andrew Shaw, Will Bennett and
Peter Hopkins

90

98

Inverse Problems
3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections
Alessandro Pirani, Marco Ricci, Antonello Tamburrino and
Salvatore Ventre

109

Multi-Frequency Eddy Current Imaging for the Detection of Buried Cracks in


Aeronautical Structures
Yohan Le Diraison and Pierre-Yves Joubert

117

A Comparative Study of Source Separation Techniques for the Detection


of Buried Defects in the EC NDE of Aeronautical Multi-Layered Lap-Joints
Alan Tassin, Yohan Le Diraison and Pierre-Yves Joubert

125

Fundamental Feature Extraction Methods for the Analysis of Eddy Current Data
Jeremy S. Knopp and John C. Aldrin

133

Inversion of Potential Drop Data for the Reconstruction of Crack Depth Profiles
Giuseppe Sposito, Peter Cawley and Peter B. Nagy

141

Automatic Classification of Defects with the Review of an Appropriate Feature


Extraction
Alicia Romero Ramirez, Neil Pearson and J.S.D. Mason

148

Microwave Nondestructive Detection of Longitudinal Cracks in Pipe with


U-Bend and Prediction of Its Location by Signal Processing
Kavoos Abbasi, Satoshi Ito and Hidetoshi Hashizume

154

Some Experiences with Microwave Investigation of Material Defects


Dagmar Faktorov

162

xvii

Modeling
Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue
Cracks
Zhenmao Chen, Noritaka Yusa, Kenzo Miya and Hideaki Tokuma
Toward the Reconstruction of Stress Corrosion Cracks Using Benchmark Eddy
Currents Signals
Maxim Morozov, Guglielmo Rubinacci, Antonello Tamburrino,
Salvatore Ventre and Fabio Villone
Evaluation of Subsurface Cracks in Riveted Aluminium Joints Using Industrial
Eddy Current Instrumentation
Maxim Morozov, Guglielmo Rubinacci, Antonello Tamburrino and
Salvatore Ventre
Design of a System for the Long Defects Detection with Advanced Methods for
Eddy-Currents Analysis
E. Cardelli, A. Faba, A. Formisano, R. Martone, F.C. Morabito, M. Papais,
R. Specogna, A. Tamburrino, F. Trevisan, S. Ventre and M. Versaci

171

179

187

195

Theory of Four-Point Alternating Current Potential Drop Measurements


on a Layered Conductive Half-Space
Nicola Bowler and John R. Bowler

203

Integration of Tilted Coil Models in a Volume Integral Method for Realistic


Simulations of Eddy Current Inspections
Theodoros Theodoulidis and Gregoire Pichenot

211

Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics


S. Paillard, G. Pichenot, Y. Choua, Y. Le Bihan, M. Lambert,
H. Voillaume and N. Dominguez
Numerical Modeling of Eddy Current Nondestructive Evaluation of
Ferromagnetic Tubes via an Integral Equation Approach
Anastassios Skarlatos, Grgoire Pichenot, Dominique Lesselier,
Marc Lambert and Bernard Duchne
Design of Reflection Type Pulsed Eddy Current Nondestructive Testing
Young-Kil Shin, Dong-Myung Choi and Hee-Sung Jung

217

225

231

Applications
Noninvasive Characterization of Bjork-Shiley Convexo-Concave Prosthetic
Heart Valves Using an Electromagnetic Method
Raimond Grimberg, Shiu C. Chan, Adriana Savin, Lalita Udpa and
Satish S. Udpa
Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer.
Application to Pressure Tubes Examination
Adriana Savin, Lalita Udpa, Rozina Steigmann, Alina Bruma,
Raimond Grimberg and Satish S. Udpa

241

249

xviii

Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants Using


Eddy Current Data
Raimond Grimberg, Adriana Savin, Rozina Steigmann, Aurel Andreescu,
Nicoleta Iftimie and Marius Mihai Cazacu
Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars
Maxim Morozov, Guglielmo Rubinacci, Antonello Tamburrino
and Salvatore Ventre

257

263

Advanced Probe with Array of Pick-Up Coils for Improved Crack Evaluation
in Eddy-Current Non-Destructive Testing
Ladislav Janousek, Klara Capova, Noritaka Yusa and Kenzo Miya

271

Observation of Stress Loaded Ferromagnetic Samples Using Remanent Flux


Leakage Method
Tomasz Chady, Grzegorz Psuj and Ryszard Sikora

276

Evaluation of Complex Multifrequency Eddy Current Transducer Designed


for Precise Flaw Depth Measurements
Tomasz Chady, Piotr Baniukiewicz, Ryszard Sikora and Grzegorz Psuj

283

Comparative Study of Coil Arrangements for the EC Testing of Small Surface


Breaking Defects
Cyril Ravat, Yann Le Bihan, Pierre-Yves Joubert and Claude Marchand

288

Author Index

295

Invited Speakers

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-3

A Start Of New Field Of Electromagnetic


And Mechanical Maintenance Engineering
Kenzo MIYA1
IIU, Corp., 2-7-17-7F, Ikenohata, Taitoku, Tokyo, 110-0008, Japan

Abstract. Maintenance to ensure the integrity of structures is one of the most


important issues in modern industry, but more so in the nuclear power industry.
This is primarily because complete prevention of all degradation of industrial
materials is not possible, at least not in a realistic way. Consequently it is
extremely important to detect degradation before machines start to loose their
ability to function which may lead to harmful failures. Usually there is some kind
of significant interval between the start of material degradation and failure; in
other words, degradation usually progresses slowly with time. The problem is
whether we can detect precursors of failure in the interval or not. If it is possible,
we can have economic as well as safety related benefits because we are able to
stop the machine before a functional failure and prevent an accident initiated by
the failure of the machine. The benefits are not restricted to the nuclear power
industry because it can be applied to any other heavy industry. There are many
reasons why many conferences, such as ENDE, have been established to discuss
and promote the progress of nondestructive inspection techniques. Whereas
nondestructive inspection plays an important role in maintaining structural
integrity and the performance of nondestructive inspection techniques should be
enhanced in that sense, we need to regard it as one of several components
composing maintenance engineering. Moreover, although nondestructive
inspection is a proactive measure of maintenance, it is not a predictive tool. In fact,
it is effective for detecting existing defects and does not say anything about the
temporal evolution of defects. On the other hand, a condition monitoring system
can offer significantly more useful information as mentioned above. In this paper I
would like to introduce the concept of maintenology as a new science and
technology in contrast to conventional maintenance engineering and to present
several important results on electromagnetic maintenance for nuclear power plants
as condition monitoring techniques (CMT). In particular, the introduction of
electromagnetic maintenance is expected to play a very promising role in
abnormality predictions of many dynamic machines that are required to be
inspected regularly by law. Application of the technique would change
conventional wisdom in thinking that machines should be taken apart and
inspected regularly based on regulations. In many cases, this TBM (time based
maintenance) is not too conservative in achieving an optimal maintenance
approach.

Introduction
The nuclear energy renaissance is occurring not only in advanced countries but also in
developing ones like China, India and Brazil, due to demands for renewable energy
sources and to protect the earth against abnormal climate caused by the green house
effect. Nuclear power plants are one of the greenest sources of energy, and operated

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

safely, can provide energy reliably to meet base loads. Thus, there is considerable
interest in the safe and economic operation of nuclear power plants (NPPs).
Maintenance engineering is extremely important to achieve these two goals. Two
approaches are usually pursued: time-based maintenance (TBM) and condition-based
maintenance (CBM). Optimal maintenance may be achieved by planning the best mix
of the two approaches. The development of CBM techniques is greatly needed to
enhance the safety level of NPPs.
In this paper we explain the possibility of using electromagnetic methods and the
application of the technology to maintenance operations at NPP. The essential part of
electromagnetic maintenance technology is the utilization of u x B electromagnetic
motive force (u: vectored velocity, B: magnetic field). In principle, diagnostics based
on the technique can be applied to rotating machines at present although application
could be extended to static components in future. Theoretical issues underlying the
approach and experimental work associated with electromagnetic maintenance will be
introduced first. Numerical simulations and experiments will then be presented.

1. Construction of maintenance engineering as a new field


Maintenance issues comprise two different aspects, one is related to human behaviour
and another relates to technical matters. Complicated human behaviour relating to
maintenance of machines is considered to consist of three principles, as shown in
Figure 1: the selection principle, connection principle and projection principle. Their
relationship can be understood well if we imagine a process of making a sentence
where proper words are selected first (selection principle), words are properly arranged
to meet grammatical requirements (connection principle), and the intended meaning is
realized by the sentence. The grammatically correct arrangement of words is called a
projection principle. Concept of such a view is translated into maintenance activities
like words corresponding to various kinds of technology supported by natural science,
sociology, codes and standards, the connecting principle corresponding to maintenance
scheme, and projection principle corresponding to the selection of necessary techniques
from existing technological systems. These basic principles should be favourably
applied to planning the maintenance scheme.
In Fig. 2, a flow chart of maintenance activities is shown schematically. They start
from the selection of the component to be maintained followed by the selection of the
maintenance method, i.e. TBM or CBM, and then finally we proceed to the inspection
process. This, in general, is the most universal process.
In Fig. 3, concepts of maintenology are newly defined and explained to introduce a
new approach to conventional maintenance engineering. The new approach consists of
three principles depicted in Fig. 1; the basic form of maintenance action is shown in
Fig.2. This recognition may contribute to the construction of revised maintenance
engineering by applying theoretical aspects.

2. Electromagnetic maintenance engineering


Principles of electromagnetic maintenance may be called the EM method. It is
associated with the following processes:
1) Creation of a static magnetic field in a region of rotating components

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

2) Measurement of a dynamic magnetic field created by eddy currents


3) Diagnosis of conditions of rotating parts inside, which can not be observed from
the outside
4) Understanding the precursor of component abnormalities
Measurements were carried out for real pumps, real fans, bearings, generating
motors, etc. It was surprising to observe signals through thick casing of components
overcoming the skin effect due to very low frequency.
Simulation software was developed to predict eddy current distribution in
components allowing one to judge whether conditions are normal or abnormal. This is
a classical inverse problem but a very difficult one to solve due to ill-posedness.
Selection Principle

Projection principle

Connection principle

System
s

Component

Selection of applicable engineering theories

Equipments

Reflection
Reflection of
maintenance
technology on
planning

Select appropriate technical knowledge


from the engineering science system
Codes&standards
Sociology, economy
Engineering
science

Maintenanology

humanities

maintenance
scheme= Flow
of
maintenance
activities

A-B-C-D-E-F

Three elements of maintenance

1RGTCVKQP

Corrective
actions

Characteristics
of Aging
(predictable
unpredictable)

Analysis Evaluation
Check
Evaluation
Investigations of
root causes

Natural science

Component
attribution Design
Operation

+PUGTXKEG

Actions/Measures

Understand the
current state
Inspections
Tests
Monitoring

Maintenance
engineering

Nuclear
facilities

Classifications of
maintenance methods
Options

Logic of choice for


maintenance methods

Inspection plans

Maintenance methods :
Decomposition testing :

Nondestructive testing :

Function tests :

Condition monitoring :

Corrective
Maintenance

Safety
Safety Operation
Operation

Figure 1. Principles underlying maintenance


activities
Maintenance
Maintenance
Engineering
1.
2.
3.
4.
5.

Maintenance
Engineering

Maintenance Science
Maintenance Engineering &
Technology
Maintenance Sociology
M ethodology of optimization
Codes and Standards
development

Figure 2. Flow Chart of Maintenance

Theoretical
Approach
1.
2.
3.
4.
5.

Theoretical
Approach

Internal form of
maintenance activities
Basic form of maintenance
action
Academic development
Application of P-D-C-A
Theory of code and standard

Figure 3. Structure of maintenology

Figure 4. Eddy current distribution in blades (The


largest eddy current will be generated in the area of
the blade near the exciting coil. Due to this,
magnetic flux changes in the direction coil are large.
The excitation flux is constant and is not measured)

Casing

Impeller

Blades

Figure 5. Impeller model and meshes

Figure 6. Eddy current in ball bearings

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

5.00E-07

Left scale for 5mm

4.00E-07

U (v)

3.00E-07

U: 5mm
U: 10mm
U: 15mm

5.00E-08

5.00E-07

4.00E-08

4.00E-07

5.00E-08

Left scale for 5mm

U: 5mm
U: 10mm
U: 15mm

4.00E-08

3.00E-08

3.00E-07

2.00E-07

2.00E-08

2.00E-07

2.00E-08

1.00E-07

1.00E-08

1.00E-07

1.00E-08

0.00E+00

0.00E+00

0.00E+00

0.00E+00

-1.00E-07

-1.00E-08

V)-1.00E-07

-1.00E-08

-2.00E-07

-2.00E-08

-2.00E-07

-2.00E-08

-3.00E-07

-3.00E-08

-3.00E-07

-3.00E-08

-4.00E-07

-4.00E-08

-4.00E-07

-5.00E-07

-5.00E-08
0.02

-5.00E-07

V)

0.002 0.004 0.006 0.008

0.01

0.012 0.014 0.016 0.018

U (v)

3.00E-08

-4.00E-08

0.002 0.004 0.006 0.008

Time(s)

0.01

0.012 0.014 0.016 0.018

-5.00E-08
0.02

Time(s)

With a holder

Without a holder

Figure 7. Signals from sensor. (1. Decrease of signal amplitude by 1/10 when distance becomes longer, from
5 to 10 mm. 2. Effect of a holder on signal is not significant, about 10% decrease)
3.00E-08
2.00E-08

U (v)

1.00E-08
0.00E+00
-1.00E-08
-2.00E-08

U: crack free
U: V carck
U: H crack

-3.00E-08
-4.00E-08
0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0.016

0.018

0.02

Time(s)

Figure 8. Ball bearings with a defect

Figure 9. Signals from defect

In Figure 4, four blades rotate around an axis which is not shown in the static
magnetic field created by a permanent magnet (not shown here). Eddy current
distribution induced in blades is shown in the figure together with a pick-up coil. If a
crack is present or the blade is greatly deformed, we measure some changes in signals
when abnormalities are present.
In Figure 5 the finite element mesh for the impeller of a pump is shown. The finite
element method is employed to evaluate eddy currents in the impeller.
In Figure 6 ball bearings witness an orbital motion. The rotational eddy current is
not significant, but the orbital motion produces a considerable amount of eddy currents
in the balls. Thus, defects in a ball bearing may be easily detected by the method. If
worn particles are present between bearings and steel plates, the speed of the orbital
motion will change and this change can be detected using this method. Simulation
results are shown in Fig. 7 for two cases: one is without a holder and the other is with a
holder. Eddy currents in these cases are not different from each other indicating small
effects of the holder. In the figure, U is a measure of distance between a bar magnet
and surface of a casing. The measured signals largely depend on the distance, which is
easily estimated by the decrease in the field due to the distance. The number of signal
oscillations corresponds to the number of blades.
In Fig. 8, the possibility of the detection of two notches on the surface of balls was
tested by numerical simulation. The notches are in horizontal and vertical direction as
shown in the figure. Results are shown in Fig.9. In the case of the vertical notch,
differences from the case without a notch are evident but are difficult to find. This
result can be explained from the magnitude of eddy currents due to positional
differences of the two notches. Therefore, it becomes essentially important to optimize
the relative position of a magnet and rotating blade.

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

3. Experimental results
The numerical results above were introduced on the basis of the explained concept to
apply electromagnetic phenomena to defect inspection of rotating machines. The
numerical results showed that there is a valid basis for the concept. The next step is to
show experimental verification of the concept. Up to now, our research group has
conducted many experiments with a small fans and pumps. We have also measured
electromagnetic signals generated by rotating parts of real machines. When we can set
a sensor and magnet close to the rotating parts, we can successfully measure a signal.
However, when access is limited, it is difficult to obtain a useful signal.
In Fig.10, measurement results are shown for the case of a fan. The signals were
obtained as a function of distance between the sensor and the rotating part. When the
distance is 5 mm, the signal is the largest and in the case of 20mm distance, the signal
is the smallest. These results are from the set-up without fan casing. But, the 20mm
distance result indicates that the measurement with a set-up that includes fan casing is
possible. Results with the casing are not shown here, but a sufficient number of signals
were obtained. Cyclic signals are obtained for 5 blades. There is the possibility that the
smaller signal on the inside is due to a weight.
The major purpose of this study is to verify the possible application of
electromagnetic phenomena to proactive detection of various types of defects.
Extensive studies must be conducted to find useful relations between signal changes
and the condition of defects. Useful information must be provided to make engineering
decisions on whether operation of the machine should be stopped for repair or
continued with careful monitoring. With this in mind, we carried out an experiment
with a notched blade as shown in Fig. 11. Two curves are shown corresponding to two
types of bar and U-magnets. It is possible to see the movement of the 5 blades from
both results. One of the blades was machined with the notch and signals reflecting an
existence of the notch are observed in both results. Through a two-way excitation
method, a crack of 1/3 the width was easily detected. Significant differences in signals
were seen for cracks 1/10 the width.
In Fig.12, three pictures of a casing of the real pump, location of a sensor and a
pick-up coil, and an impeller are shown. The casing is cast iron and shows magnetic
property. The rod type permanent magnet is attracted to the casing.
0.08

0.06
0.04
0.02
5mm
10mm
15mm
20mm

-0.02
-0.04
-0.06
-0.08
-2.00E-02 -1.50E-02 -1.00E-02 -5.00E-03 0.00E+00

Figure 10. Fan experiment

5.00E-03

1.00E-02

1.50E-02

2.00E-02

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

1.5
U

0.8

0.6
0.4

0.5
0.2
0

-0.2

-0.5

-0.4
-0.6

-1

-0.8

-1.5
-2.00E-02 -1.50E-02 -1.00E-02 -5.00E-03 0.00E+00 5.00E-03 1.00E-02 1.50E-02 2.00E-02

-1
-2.00E-02 -1.50E-02 -1.00E-02 -5.00E-03 0.00E+00 5.00E-03 1.00E-02 1.50E-02 2.00E-02

Figure 11. Detection of a defective blade

In Fig. 13, measurement results are shown. A careful examination shows that the
number of impellers is 5 and the rotation cycle is 0.02 seconds since a motor operates
at 50 Hz in the Kanto area in Japan. In addition to this, we can recognize a small
difference in the amplitude of the measured voltage corresponding to the difference in
size of the 5 blades. Since the signal amplitudes are different this suggests that the
method can be applied to test for the dimension of axis and impellers, as well as their
condition of motion. The wave form is not sinusoidal and shows a small tooth. The
reason for the tooth is not clear until we see the inside of the pump after opening the
casing. This small observation indicates that the detailed structure inside a rotating
machine will be made clear after we accumulate knowledge on dimensional
information and defects signals in the future.

Figure 12. Casing and impellers (real pump)

K. Miya / A Start of New Field of Electromagnetic and Mechanical Maintenance Engineering

Blade intervals are 0.004s


2
1.5
1

0.5
0
-0.010

-0.005

0.000

0.005

0.010

0.015

0.020

0.025

0.030

0.035

0.040

-0.5
-1
-1.5
-2

Rotation cycles are 0.02s


s

Figure 13. Signal with rod type permanent magnet sensor

3. Conclusion
The author would like to note that there are many challenging problems in the
diagnostics of rotating machines and that there is a strong need for developing methods
of detecting abnormality precursors as a prognosticator of functional failure. There are
well-known methods of condition monitoring. These include vibration monitoring
methods, oil analysis, thermography, etc. These methods will be useful if they are used
properly; however, there are several important problems to be solved in identifying the
root causes of malfunctions occurring in a machine. Electromagnetic methods have
shown their superiority over conventional methods in being able to locate defects and
in identifying conditions at specific locations in rotating parts. Extensive studies are
required to establish the relationship between the measured signal and the nature of
defects. Efforts are urgently needed to translate R&D efforts to industrial applications.

10

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-10

NDE Research Makes a Difference


C B SCRUBY
UK Research Centre in NDE, Imperial College London, SW7 2AZ UK

Abstract. This paper discusses the development of research in NDE and its
impact on industry. Examples will be given of past research projects that have
been translated into solutions to industrial inspection problems, and present day
challenges to industry and the NDE research community. Recurring themes
include the need for quantification, and physical models to give scientific
understanding and hence improved confidence for product quality and safetycritical application. Timely technology transfer is a continuing challenge and
lessons from past experience will be discussed. Finally, the author will discuss
future research strategy, including opportunities for interdisciplinary collaboration
in order to integrate NDE more effectively into the engineering life cycle.
Keywords. Non-destructive evaluation, NDE, electromagnetic.

Introduction
There has been steady growth in non-destructive evaluation (NDE) research since the
1950s, reflecting the increasing demand for greater safety and environmental protection
by the public. NDE research has always been particularly challenging. Firstly, the
NDE field makes use of multiple technologies & disciplines, including magnetic,
electrical, ultrasonic, radiographic, optical, and thermographic methods. Secondly,
NDE is used for a wide range of materials (from metals to plastics, ceramics and
composites) & applications (manufacturing process control, in-service inspection,
defect sizing, corrosion measurement, and degradation monitoring). NDE is used in
most sectors of industry: aerospace, power, oil & gas, transportation, infrastructure,
built environment, manufacturing, nuclear, process, defence, electronics, packaging,
etc., which have vastly different products and assets to be inspected everything from
electronic devices and foodstuffs through to entire rail networks and huge refineries.
To enhance the challenge there is a complex supply chain to take new technologies
from university research through to routine use, involving research and technology
organizations (RTOs), service companies, equipment suppliers, certification and
standards authorities, trainers and consultants (Figure 1).
However, on the positive side, there is much commonality and overlap at the
research stage in terms of techniques and generic applications. Many universities and
RTOs have begun to recognize the benefits of an interdisciplinary approach to NDE
research. Until the late 1980s much research was funded by government research
laboratories and state-owned industries, especially in the nuclear and defence sectors.
There were then changes due to privatisation that made a major impact on the funding
of NDE research. Although the value of an NDE solution to an individual company or
single industry sector may be too small to justify large investment, there are significant
opportunities for collaborative funding to increase benefit, share risk and reduce cost.

C.B. Scruby / NDE Research Makes a Difference

Technique
development

Service
suppliers

Product
development

Equipment
suppliers

11

Asset owners
and operators

Research

Product
Manufacturers

Figure 1. The technology supply chain for NDE

Furthermore, generic research may attract government funding through its research
councils, while pre-competitive, inter-sectoral, research collaborations lever public
funding via for instance EUs Framework Programmes.
There are several key common themes affecting the effective application of NDE,
and which are the drivers for research in NDE. The major ones are: coverage, overall
inspection speed, quality of information delivered (sensitivity, accuracy, resolution,
etc), accessibility of region to be inspected, reliability of the NDE measurement,
applicability to a wide range of damage processes, materials and applications, reducing
the need for skilled operator intervention (which implies autonomous inspection or
monitoring systems), and of course the reduction of costs of the whole NDE process.
The UK Research Centre in NDE (RCNDE) has a wide portfolio of NDE research
within most of these theme areas.

1. The Research & Innovation Process for NDE


The innovation process for NDE is no different from any other area of technology.
Strong relationships in the supply chain (Figure 1) aid idea generation. As Figure 2
shows, there is usually some pioneering or opportunistic research before a more
substantial research programme is undertaken. This is often followed by trials to
evaluate the likely industrial benefits of the research. Depending upon the outcome,
this may be followed by further, more applied research and development. If successful
the new technology needs to be transferred or translated from research into product, a
crucial and vulnerable step that will be discussed again. Commercialization, involving
development of a market and sales, follows. However, development of the full
potential of a product (or service), market penetration and acceptance, tends to be
hindered if the technology is not embodied into standards and routine inspection
procedures. Note that both scientific push and industrial pull are needed to drive
this process; they need to overlap to ensure successful innovation.
For a variety of reasons product development and commercialization sometimes
accelerates beyond what can be supported by the market or the scientific foundation.
As a result the technology is oversold which may lead to disillusionment. The only
way forward is a realistic evaluation of both the underlying science and the proposed
market, followed (if the conclusion is positive) by further research, development,
optimization of the product to match industrial requirements, and hence through
standardization into market acceptance.

12

C.B. Scruby / NDE Research Makes a Difference

230

Acceptance

Standardisation
Commercialisation
Product
development
100
Further research
&
development

Trials &
evaluation
Research
Pioneers
1

Technology development

121

-30

Scientific push
Industrial pull

Figure 2. Research and innovation process for NDE

Within the defence and aerospace sectors of industry the innovation process is
often described in terms of technology readiness levels (TRLs), as shown in Figure 3.
Here the inventive phase is described as technology assessment. This may take place
in an industrial R&D laboratory or an university. Production implementation is the
remit of the end-user company. The challenge is the intermediate steps from 4 to 7;
there is a need not only for organizations to undertake work at TRLs 5 and 6, but also
to interact effectively with the organizations who carry out TRLs 1-4 on the one hand
and TRLs 7-9 on the other. Experience has shown that the most risky part of the
process is this translation from research to production. This may be because research
and production are carried out by organizations with different objectives and cultures.
The NDE industry has a relatively complicated supply chain for new technology
(Figure 1) because the technology is used in both production and operation, and
operators often sub-contract to service companies. There need to be routes to market
for NDE services as well as products. In either case, improved ways are needed to
facilitate the movement of new technology from left to right along the chain.

2. Case Studies of NDE Research and Innovation


There are many potential examples within the field of electromagnetic NDE, but only a
small number can be examined here with a view to understanding the process whereby
a research idea is converted into an industrial technology. The examples are selected
from the knowledge and experience of the author, and selection or omission is not a
value judgment of the product in question.
Alternating Current Field Measurement (ACFM) [1] is a current perturbation
technique related to, but different from eddy current testing methods. Its development
followed closely the pattern described by Figure 2. Thus there was scientific push in
the form of university research at University College London (UCL) including

13

C.B. Scruby / NDE Research Makes a Difference

Proven
Method

9. Track record established through successful applications


8. Production system qualified through test and demonstration

7. System prototype demonstration in an operational environment

6. Inspection optimised for capability and speed

Pre-Production

5. Inspection validated in relevant environment (prototype equipment)

4. Inspection validated in laboratory environment

Technology
Assessment
and Proving

Development

Production
Implementation

3. Experimental proof of concept completed


2. Technology concept or application vetted or demonstrated
1. Basic principles observed and reported

Idea

Figure 3. Innovation process described in terms of Technology Readiness Levels

substantial modelling from 1987-9. This was followed in 1991 by application trials on
model offshore nodes supported in part by the oil & gas industry, demonstrating
growing industrial pull. In 1993 ACFM was licensed to TSC (a small university spinout company), with (1993) a royalty agreement to cover the UCL developments needed
to commercialize the technology. 1994-8 saw the development of an applications
market, mainly in oil & gas sector. This was supported by applications R&D in both
university and company (1996-2000) with strong industrial pull. Standards for ACFM
were published in 2003, and from 2004 onwards the focus was growing the application
market and moving into the rail and nuclear sectors as the technology gained
acceptance. The development was over a shorter period than some other examples,
initial research to standard taking some 16 years. This was perhaps partly because the
risk was reduced by the existence of an established market for eddy current technology.
There was a good combination of scientific push and industrial pull for most of the
development phase. But also, significantly, a strong supply chain based on close
organizational and individual relationships was established early on.
A second example is Pulsed Eddy Currents (PEC). It is based upon pioneering
research at Argonne National Laboratory (USA) in 1950 with nuclear applications in
mind. It was then picked up by the aerospace industry for detecting cracks in
aluminium. In 1987 PEC was adapted by ARCO in their TEMP instrument (a pulsed
eddy current device) to detect steel corrosion under insulation (CUI) for the oil and gas
industry. In 1990 TEMP was licensed to RTD in Holland, being commercialized as
INCOTEST. In 1994 Shell investigated PEC for detecting CUI, but found some
shortcomings. However, at the same time, Shell carried out a laboratory investigation
of PEC for other industrial applications. This led in 1996 to the launch of a
commercial PEC system for a range of oil and gas applications, such as corrosion under
fireproofing. There are no standards for PEC at the time of writing. The research from
the 1980s onwards mainly took place in companies rather than universities, which
ensured a strong industrial pull through most of development of PEC and a good
potential market, once the target applications had been identified [2].
Basic research into eddy currents was started at RAE Farnborough (MoD) in
collaboration with Surrey University in the 1970s. In the 1980s this moved on to eddy

14

C.B. Scruby / NDE Research Makes a Difference

current research using Hall sensors, while from 1985-9 industrial pull joined scientific
push and a prototype EddyScan system for crack detection in aircraft was developed.
The scientific focus then moved to transient eddy currents in the 1990s [3], as further
research was undertaken, first on the technique and then (1995-2000) on applications as
industrial pull increased. Activity switched to a product and the TRECSCAN system
was developed from 2000-3, addressing aircraft inspection needs in the defence and
aerospace sectors. Like ACFM and PEC, the development of transient eddy currents
was heavily dependent upon background research into electromagnetic induction and
eddy currents in particular. Here the early work benefited from collaboration between
a government laboratory and a university. There was a requirement to deliver useful
research outputs to the MoD customer in the early days of this development, but
industrial pull appears to have strengthened progressively. This is consistent with the
growing market for methods to prolong the life ageing aircraft.
The first three examples build on related research into eddy currents. Apart from
isolated pieces of pioneering research in various countries, the next example is of a
technology developed from first principles, and it is therefore not surprising that it is
still some way from acceptance, even though about 20 years have passed. In 1982-9
some fundamental scientific studies of magnetic methods for materials and stress
characterisation were undertaken at Oxford University and Harwell Laboratory [4].
This led on to a programme of research into NDE stress measurement, 1989-93, with
progressively stronger industrial pull and support from the oil and gas industry; this led
to trial applications of what became known as MAPS in 1993-5. The early applications
demonstrated the need for further research into the technique as well as its application
(1995-00). It was found necessary to implement further refinements to the
underpinning physical model as well as to the MAPS instrument (2000-7), while at the
same time beginning to commercialise the technology for niche applications in the oil
and gas and rail sectors. The early stages of the development showed the importance of
collaboration between university, RTO and end-user industry. Industrial pull waned in
the middle period of the development and this, coupled with privatisation and
successive company reorganisations, may have delayed the innovation process.
The final example is of an important ultrasonic innovation, time-of-flight
diffraction (TOFD), rather than another electromagnetic technique. Maurice Silk
started research into TOFD [5] at the NDT Centre, Harwell Laboratory (UKAEA) in
1974 using ideas from previous research into neutron time-of-flight spectrometry.
During the ensuing years (1974-82) the work was mainly laboratory research (scientific
push). Other scientists became involved and the theory of ultrasonic diffraction was
developed in 1981-83. The first major industrial application was to UKAEAs
Winfrith Reactor 1982-4. Soon afterwards (1983-4) TOFD used by UKAEA for their
nuclear defect detection trials. Almost simultaneously (1984-90) there was further
industrial pull when TOFD was developed for offshore and undersea use as an early
activity of Harwells HOIS project. The first commercialisation came 1982-4 when
Zipscan was developed and licensed. Soon afterwards in 1984, TOFD started to be
used for major industrial inspections, especially nuclear pressure vessels and offshore
structures. The first TOFD standard was published in 1993 and TOFD was soon
accepted as a mainstream NDE method across industry. TOFD took about 20 years to
move from initial research idea through to an industry standard. It is salutary to
wonder how long exploitation would have taken had it not been for strong industrial
pull in the mid-1980s from two separate energy sectors. TOFD had little input from the
university research base in its early days, reflecting the strength of public sector

C.B. Scruby / NDE Research Makes a Difference

15

laboratories then, a different situation from the 21st century. Throughout the central
period of development there was strong collaboration between researchers, developers
and end-users, helped by national imperatives in the nuclear case and a well-established
and effective joint industry collaborative project (HOIS) in the other. Another
important element was the utilisation of ideas and technology from other fields.
A common theme in all of these and similar developments has been the need to
develop physical models of the technique under investigation. In many cases using
models to perform predictive or interpretive calculations was a major exercise, with
more limited computing facilities prior to the 1990s. A model was vital to understand
the amplitude variations in TOFD data, as it has been for example to interpret the
electrical signals from ACFM, TRECSCAN and PEC techniques. In the case of MAPS,
modelling the effects of stress on magnetic parameters from first principles involves
very difficult physics as well as computational power. However difficult or timeconsuming, robust physical models are, in the authors view, vital for any new
technology. They are needed, not only to interpret the data and design the best way to
use it, but also to give credibility within the scientific and engineering communities.

3. Lessons to be learnt
From the authors experience and knowledge, there are a number of barriers to or
brakes that slow down effective NDE research, i.e. research whose results make a
positive difference to industrial practice. Two important hurdles are, on the one hand
lack of understanding by researchers of industrial needs and culture, and on the other
hand lack of understanding by industry of scientific issues and research culture.
Research itself, or the development process that follows can be hindered by the wrong
level of funding, more commonly too little but occasionally too much at a time when
the team is unable to deliver what is anticipated. This leads on to the importance of
correctly managing expectations. Often researchers are over-optimistic about what
they will achieve and their speed of progress in order to secure funds. When they fail
to meet their customers expectations the development may be dropped for the wrong
reasons. Intellectual property (IP) can cause problems. Sponsors of research may insist
on IP rights that are too restrictive and stifle creativity, while loose IP arrangements can
dissuade companies from investing in technology.
As already stated, a serious issue concerns gaps in the supply chain, i.e. between
researchers, developers and end-users. In terms of TRLs, the problems usually arise
when progressing from level 4 to level 7. Especially during the past 10 years, members
of the NDE supply chain have undergone reorganisations, changes of status, mission
and objectives. Staff have been lost, working relationships destroyed and the supply
chain disrupted. Finally, human factors always play a part, the biggest hurdles being
lack of trust and poor communication for whatever reason. The Engineering Doctorate
in NDE is a very encouraging recent development, already proving its worth in terms
of technology transfer from research to application. Accountable to both university and
company, the student (research engineer) begins to bridge that TRL gap.
What lessons can be learnt of the ingredients for successful NDE research? The
following list is not exhaustive, nor is every ingredient always necessary:
1. Adequate resources - people, facilities, environment, timely funding
2. Scientific excellence - creativity and lateral thinking balanced against
penetration and focus; correct fundamentals essential for robust application

16

C.B. Scruby / NDE Research Makes a Difference

3.
4.
5.
6.
7.
8.
9.
10.
11.
12.

Technology push balanced against industry pull throughout


Interdisciplinary collaboration, networking and international links
Import of ideas, results & technology from other fields
Industrial pull founded on sound business needs or regulatory imperatives
Supportive organisational structures with long-term vision
Viable exploitation route along joined up technology supply chain
Committed individuals to champion and drive innovation process
Trust, good communication and shared objectives within the whole team
Proprietary secrecy balanced against publication and scientific credibility
Good timing, grasp of opportunity, achievable timescales

The author is tempted to add: serendipity, since chance events often seem to
unlock research and stimulate vital creative leaps; patience, since it is difficult to
predict what research will produce and when.

4. Concluding Remarks
Before discussing strategy for successful research in NDE, it is important to capture the
present trends in the field of NDE. Encouragingly, most recent market surveys have
predicted a steady growth in use of NDE by industry. This is driven ultimately by the
needs of society for greater assurance of the safety of engineering structures. Linked
with this is a heightened awareness of the need for environmental protection. New
materials & new designs are being used that require new inspection technology. There
is also a move towards greater automation, and the willingness to invest in high
technology solutions to reduce inspection costs, improve quality, speed up inspection
and sentencing. In some cases the move is towards reducing the need for inspections
that interrupt service by new strategies such as structural health monitoring (SHM).

Mathematics
Materials
characterisation

Physics

Manufacturing
& product
quality
Techniques

Technology &
science transfer
Mechanical
engineering
Materials
science

NDE Research
Applications

Supply chain

Electronic
engineering

In-service
structural integrity
& assessment

Condition & health


monitoring

Other disciplines
Figure 4. NDE research and linked engineering disciplines

C.B. Scruby / NDE Research Makes a Difference

17

As a long-term strategy there is a desire to integrate NDE more strongly into the
engineering life cycle, so that the benefits of NDE are weighed against reductions in
the whole life cost of an asset. To do so requires the development of linkages with
material science, engineering design, structural integrity and assessment. NDE sits at
the centre of a complex network of engineering disciplines as Figure 4 shows; these
needs to be understood and exploited. Recent years have seen large changes not only
in the large users of NDE technology, but also in the supply chain. Equipment supply
and inspection service companies have changed hands, merged and in some cases been
bought by large companies. Such changes are likely to continue in the future.
Without any doubt the NDE industry will continue to encounter technical
challenges that can only be solved by consistent, high quality research. What are likely
to be the main elements of a future NDE research strategy? The following list indicates
some of the likely priorities:
1. Modelling, always essential for scientific understanding and to give
confidence in product quality and for safety-critical applications
2. Improved quantification of results, characterisation and discrimination
3. More advanced data analysis, imaging and visualisation
4. Greater understanding of and improvement to reliability
5. Raising inspection speeds and reducing human factors through automation and
autonomous systems
6. Incorporation of technological advances from other fields
7. Addressing new materials, designs, difficult applications & environments
8. Earlier detection of materials & structural degradation & failure
9. Technologies and strategies to facilitate reliable SHM
This is a very full and challenging list. It can only be achieved through greater
collaboration & integration with other disciplines & fields. It also requires the long
term education and training of high calibre engineers for all stages of the research and
innovation process.
The UK Research Centre in NDE is a university-industry partnership that was
established nearly 5 years ago to harness the UKs research base to meet these longterm challenges, building on and learning from past experience, and hopefully avoiding
some of the pitfalls. Its vision is to make a lasting difference to NDE technology
through well-planned world-class research, and to invigorate the NDE profession
through the provision of highly trained engineers.
To conclude, NDE research does make a difference. It is impossible to respond to
the challenges of tighter regulation, new materials and applications, new products and
plant, more stringent operational conditions, longer life, higher accuracy, reliability,
long terms global trends, changes in industry, rising public expectation of safe and
environmentally secure operation, and the commercial drivers of faster, better, cheaper
products without research.

References
[1]
[2]
[3]
[4]
[5]

W.D.Dover, R.Collins and D.H.Michael, Phil. Trans. R. Soc. Lond. A320, (1986) 271-283.
P.Crouzen, Proceedings of 9th European Conference on NDT, Berlin, (2006) in press
S.K.Burke, G.R.Hugo and D.J.Harrison, Review of Progress in Quantitative Non-Destructive
Evaluation, 17A, (1998) 307-314
D.J.Buttle, C.B.Scruby, J.P.Jakubovics and G.A.D.Briggs, Proc Roy Soc Lond. A414, (1987) 469 497.
M.G.Silk and B.H.Lidington, Harwell Report, AERE-R7774 (1974).

18

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-18

Industrial Applications of 3MA


Micromagnetic Multiparameter
Microstructure and Stress Analysis
Gerd DOBMANN1, Iris ALTPETER1, Bernd WOLTER1, Rolf KERN1
1
Fraunhofer IZFP, Germany

Abstract. Micromagnetic NDT techniques like the measurement of the magnetic


Barkhausen noise, the incremental permeability and the harmonic analysis of the
tangential magnetic field allow deriving inspection procedures to online
monitoring and control machinery parts and components in production processes
in order to characterize mechanical properties like hardness, hardening depth, yield
and tensile strength. These types of inspection procedures continuously were
further developed in the last two decades so that today the second generation of
system hard and software is in industrial use. The application is in steel industry
where steel sheets in hot-dip-galvanizing lines were annealed after cold rolling but
also in heavy plate rolling mills where after thermo-mechanical rolling special
textures and texture gradients can occur. An increasing number of applications are
also to find in the machinery building industry and here especially in case of
machinery parts of the car supplying industry. Besides mechanical hardness
determination the measurement of residual stresses and the detection of inhomogeneities in the surface of machined parts is an inspection task. In different
case studies the advantage to implement a micromagnetic NDE technique into the
industrial processes is discussed.
Keywords. Micromagnetic NDE, hardness, hardening depth, residual stresses,
yield strength, steel industry, steel sheets, heavy plates, machinery building,
automotive

Introduction
The reason to develop 3MA (Micromagnetic-, Multiparameter-, Microstructure-, and
stress-Analysis), starting in the late seventies in the German nuclear safety program,
was to find microstructure sensitive NDT techniques to characterize the quality of heat
treatments, for instance the stress relieve of a weld. George Matzkanin [1] just had
published a NTIC report in the USA to the magnetic Barkhausen noise. The technique
was sensitive to microstructure changes as well as to load-induced and residual stresses.
Therefore a second direction of research started in programs of the European steel
industry and the objective was to determine residual stresses in big forgings. Beside the
magnetic Barkhausen effect also a magneto-acoustic-one became popular [2]. The
technique has based on acoustic emission measurements during a hysteresis cycle and
was because of the high amplification also sensitive to electric interference noise.
Therefore the acoustic Barkhausen noise technique has never found a real industrial

G. Dobmann et al. / Industrial Applications of 3MA

19

application. Later further micromagnetic techniques were developed: the incremental


permeability measurement, the harmonic analysis of the magnetic tangential field and
the measurement of the so-called dynamic or incremental magnetostriction by use of an
EMAT [3, 4].
The methodology of the Micromagnetic-, Multiparameter-, Microstructure-, and
stress- Analysis (3MA) in detail is described in [4]. On the basis of a multiple
regression model, describing target quantities like hardness or yield strength as a
function of measured micromagnetic quantities, the unknown model parameters are
determined in a calibration step. By using a least squares approach the unknown
parameters are the solution of a system of linear equations. 3MA only can be applied at
ferromagnetic materials. Here the techniques are especially sensitive to mechanical
property determination as the relevant microstructure is governing the material
behavior under mechanical loads (strength and toughness) in a similar way as the
magnetic behavior under magnetic loads, i.e. during the magnetization in a hysteresis
loop. Because of the complexity of microstructures and the superimposed stress
sensitivity there was a need to develop the multiple parameter approach.
Whereas the first generation of 3MA equipments was basing on the magnetic
Barkhausen noise and magnetic tangential field analysis only, 3MA equipment exist
now in the forth generation also integrating incremental permeability and eddy current
impedance measurements (see Figure 1). More than 100 installations are in use in
different industrial areas. This mainly covers the steel and machinery building
industries.

Figure 1. TCP-IP-based 3MA equipment and software in combination with a laptop

1. Applications in the steel industry


1.1 Steel Strip Inspection
A lot of experiences with 3MA in the last 2 decades were to the continuous mechanical
property determination at steel strips, designed to produce car bodies [5, 6], running
with a speed of 300m/minute for instance in a continuous galvanizing and annealing
line. Yield strength (Rp0.2), tensile strength (Rm), planar and vertical anisotropy
parameters (rm, 'r) are in the focus of quality assurance measures [4], all of them are
defined by destructive test and cannot be measured continuously. Therefore 3MA
correlations were calibrated.
Figure 2 shows a yield strength profile along a coil of 2.5 km length [5]. At the
beginning and the end an unacceptable increasing of strength is detected higher than

20

G. Dobmann et al. / Industrial Applications of 3MA

the upper acceptance level (blue line). The strength values are calculated by the 3MA
approach from measured micromagnetic data. The red dots indicate the selection of
specimens taken to destructive verification tests after performing NDT. The residual
standard errors found by validation are in the range 4-7 % concerning the yield strength.
Figure 3 shows a 3MA installation in the line of a strip producer; a robot is used to
handle the transducer.

Figure 2. 3MA predicted Yield strength [5]

Figure 3. 3MA probe with robot at a strip line

1.2 Heavy plate Inspection

1600

1600

Elastic limit R p0.2 [MPa] (3MA)

Tensile strength R m [MPa] (3MA)

Ongoing research is to heavy plate inspection. The steel producer asks for the
measurement of geometrical and mechanical properties, which have to be uniform
along the product length and width, especially in the case of high-value grades used in
off-shore application. Destructive tensile and toughness tests are performed by highly
qualified and certified personnel according to codes and delivery conditions. The tests
cannot be integrated into online closed loop control with direct feedback. To reliably
test the mechanical hardness the surface must be carefully prepared by removing scale
and decarburized surface layers and residual stresses are to relieve. The extraction of
the test pieces and testing is very time and cost extensive. Costs in the range of several
thousands Euro per year arise in a middle-sized heavy plate plant only by destruction of
the test pieces.
1400
1200
1000
800
600
400
200
0

1400
1200
1000
800
600
400
200
0

200

400

600

800 1000 1200 1400 1600

Tensile strength R m [MPa] (destructive)

Figure 4. Tensile strength predicted by 3MA [6]

200

400

600

800 1000 1200 1400 1600

Elastic limit or yield strength


Rp0.2 or R EH [MPa] (destructive)

Figure 5. Yield strength predicted by 3MA [6]

G. Dobmann et al. / Industrial Applications of 3MA

21

In case of a mother plate of several meters length the edges are usually subjected to
other cooling conditions than the rest. Indeed, especially the plate ends are known to
cool faster, generating an undesired increase in tensile strength Rm and yield strength
Rp0.2. State-of-the-art is to cut-off the plate edges with non-conform properties based
on empirical values concerning the cut-off length. As the destructive tensile test
follows directly after the cut-off process of the edges, only the result of these tests can
reveal the selection of a not appropriate cut-off length. This results in high costs due to
reworking, pseudo-scrap and delayed shipment release; the European steel producers
estimate their annual costs in the range of 11 million Euros. Knowing exactly the
contour of the zone with unacceptable material properties would allow an open loop
control of the cut-off process. Therefore heavy plate producers will replace the
destructive quality inspection of test pieces by a NDT technology [6] applying 3MA
(see Figure 4 and 5). By a manufacturer-specific calibration residual standard errors of
10 MPa (Rm), 20 MPa (Rp0.2), and 4HB in the Brinell hardness can be obtained. It
should mention here that in the 3MA calibration also other measuring quantities can be
integrated so far they provide other independent information, for instance elastic
properties. By using ultrasonic waves propagating in thickness direction, i.e. a
compressive wave excited by a piezoelectric transducer (index L) and two linearly
polarized shear waves (polarized in, index SHR, and transverse, index SHT, to the
rolling direction) excited by a EMAT, normalized time-of-flight quantities can be
derived describing crystallographic texture effects. Taking into account these quantities
(tSHR/tL, tSHT/tL, (tSHR-tSHT)/tL) together with the micromagnetic parameters then a
regression result is obtained again reducing the residual standard error.

2. Application in Automotive and Machinery Building Industry


2.1 Car Engine casting
To reduce the weight of the power supply unit the car combustion engines cylinder
crankcases can be made of cast iron with vermicular graphite (GJV), because this
material in a Diesel engine allows a higher loading pressure even by reduced wall
thickness. However, the service live of machining tools is during processing an engine
block made from GJV substantially smaller compared with a block from cast iron with
lamellar (flake) graphite (GJL).
work on - relevant range

lamellar graphite

GJL

transition zone

vermicular graphite

GJV

Figure 6. Microstructure gradient obtained in a cylinder region of a cast engine [7]

This disadvantage can be eliminated by an innovative casting technology that


produces a continuous microstructure gradient in the cast iron from lamellar graphite at

22

G. Dobmann et al. / Industrial Applications of 3MA

the inner surface of the cylinders to vermicular graphite in radial direction. By


implementing some chemical additives into the core of the mould which can diffuse in
the cast iron during the solidification process in the mould the gradient with a
continuous transition from lamellar graphite and finally vermicular graphite is obtained.
However, the technology can only be used by the casters so far the gradient quality can
be characterized and monitored by NDT. Figure 6 documents in a micrograph such a
gradient beginning at the left side with cast iron (inner cylinder surface) and lamellar
graphite followed by a transition region and vermicular graphite on the right side.
3MA techniques always cover a certain analysing depth depending on the
magnetising frequency and geometrical parameters of the magnetisation yoke, etc. So
far the gradient has different graphite compositions within the analysing depth, 3MA
quantities should be influenced. Based on measurements at an especially designed
calibration test specimen set 3MA quantities were selected to image the gradient with
optimal contrast. As reference quantity to calibrate 3MA the local thickness of the
GJV-layer was evaluated by using micrographs and optimized pattern recognition
algorithms in the microscope. A special designed transducer head was developed to
scan the cylinder surface by line scans in hoop direction and rotating the head, then
shifting the head in axial direction to perform the next line scan. Figure 7 and Figure 8
show as example the coercivity images derived from the tangential field strength
evaluation[3] (HC0 in A/cm) and line scans covering an angle range of 190.

Figure 7. Coercivity image of a reference block


made from GJV

Figure 8. Coercivity image of block with GJV/GJL


gradient

Combining different 3MA quantities in a multiple regression the thickness of the


GJL layer was predicted. A regression coefficient of R2= 0.93 and a residual standard
error of V = 0.06 mm was obtained [7].
2.2 Wheel Bearing Inspection
The fixation of the inner ring of wheel bearings is performed by a wobble riveting
process. As a consequence a residual stress is built up in the ring which may not exceed
a limit value of about 300 MPa to get a perfect quality.
The usual technique to inspect the residual stress state is x-ray diffraction which is
destructive in nature because it requires a preparation of the test location. Furthermore
it can only be performed statistically. The 3MA technique allows a fast non-destructive
estimation of the residual stress level (Figures 9, 10). After a calibration step by using
x-ray reference values a 100% quality inspection of these parts is possible. The
calibration procedure requires a coincidence of the 3MA and x-ray calibration positions

23

G. Dobmann et al. / Industrial Applications of 3MA

because residual stress varies along the circumference. That means the 3MA data have
to be recorded in a first step before the x-ray test location is prepared by etching.
According to Figure 10 the residual standard error in the calibration is in the 20 MPa
range. Besides the residual stress additionally the surface hardness can be measured.
250

Analysing Depth:
100 m

Residual Stress [MPa] (X-ray)

200

150

100

50

Adj.R = 0.941
1 = 19.0 MPa

-50
-50

50

100

150

200

250

Residual Stress [MPa] (3MA)

Figure 9. 3MA-Probe at test location

Figure 10. Residual stress calibration

2.3 Evaluation of Microstructure and Stress Gradients


Machined parts in most of the cases have more or less steep gradients in their
properties near the surfaces. To improve the lifetime of mechanical highly stressed
machinery components the bearing areas are surface-hardened from the m- up to the
millimetre range depending on the requirements and on the hardening technology.
80
70

3MA / Nht 700 [m]

Adj R = 0,8211

60
50
40

Adj R = 0,9151

30
20

3MA Value

10

Nht700 Value

0
0

10

20

30

40

50

60

70

Optical Result [m]

Fig.11. Comparison of nitrating hardening depth measured by 3MA and Nht 700 (Vickers) versus optical
result

Additional surface finishing by grinding can superimpose surface near defects of


microstructure and residual stresses which can result in a part breakdown. To inspect
the production quality in many cases not only the properties immediately at the surface
but also information of the properties below the surface are desired. 3MA is an
effective tool to investigate the properties near the surface as well as the range below
the surface up to several millimetres in depth.
One example of a 3MA application in industry is the determination of nitrating
hardening depth NHD of piston rings on the flank side and on the tread surface.
Typical values of nitrating hardening depth are between 60 and about 100 m. It is
found by the user that the reproducibility of the non-destructive values of hardness and
hardening depth in piston rings is better than the conventional testing by a

24

G. Dobmann et al. / Industrial Applications of 3MA

metallographic Vickers hardness test (Nht700), as can be seen in Figure 11 [8]. The
reason of that behaviour seems to be the difference in the lateral resolution of the
conventional and the non-destructive testing method. Due to a diameter of the 3MA
receiver coil of about 2 mm the 3MA values are covering a much larger inspection area.
Fast data evaluation by 3MA allows a complete production feedback control.
The occurrence of grinding defects, e. g. in gear wheels, is a main problem since
many years which is caused by too much heat input during the grinding process.
Modern grinding tools allow much higher grinding speed compared to former machines
but on the other side this can result in more defects. To get information on the quality
of grinded microstructure states the common method in industry is the nital etching
technique. Grinding defects are indicated by the discoloration of the surface. This
technique is effective as long as the surface information is sufficient to estimate the
quality. But it fails if in a preceding production step defects are produced below the
surface which are covered in the next production step by a perfect finishing. Several
examples of defective gear wheels investigated by hole drilling method and x-ray
diffraction have shown that in a depth of 100 m high tensile stresses up to several 100
MPa can be present whereas at the surface a perfect compressive state of several
hundred MPA has been found. These hidden defects cannot be detected by nital etching.
As a consequence after some time small cracks are covering the surface due to stressrelieve even without a mechanical load.

Figure 12. Hardness calibration at various depths;


hardness values determined by 3MA versus target
values

Figure 13. Residual Stress calibration at various


depths; RS values determined by 3MA versus X-ray
reference values

Since several years IZFP has gained experience in the non-destructive detection
and quantitative evaluation of such grinding defect gradients by 3MA in cooperation
with industrial partners and in different research and development projects [9, 10].
After a calibration step 3MA can be used to evaluate different target values
simultaneously, especially the hardness and the residual stress at the surface and in
several depths below the surface (Figures 12 and 13). To get unambiguous results
calibration must be done carefully. Calibration is mainly determined by well defined
calibration specimens and only valid to the target ranges available by calibration. In
most cases calibration is restricted e. g. to the material, to the actual machining
parameters and even to the 3MA probe in use. If any variation occurs, its influence on
the validation of the existing calibration has to be checked and if necessary a
recalibration or extension of the existing calibration has to be performed to include any
disturbances. These limitations and the calibration effort may be seen as a disadvantage
of 3MA. But if an optimal calibration is developed the fast non-destructive

G. Dobmann et al. / Industrial Applications of 3MA

25

determination of various quality parameters which is desired concerning expensive


security related parts justifies this effort.

3. Conclusion
3MA is a matured technology and a wide field of applications is given. However,
besides the success story we also can find critical remarks from industrial users. These
are mainly to the calibration efforts and problems of recalibration if a sensor has to be
changed because of damage by wear. Therefore actual emphasis of R&D is to
generalize calibration procedures.

4. Acknowledgements
The authors very much appreciate to acknowledge the contribution to the result by
companies as ThyssenKrupp Stahl AG, Duisburg, ArcelorMittal Research, Metz,
Dillinger Htte GTS AG, Dillingen, Halberg Guss GmbH, Saarbrcken, and Schaeffler
KG, Schweinfurt.

References
[1]

G.A. Matzkanin, et al., The Barkhausen Effekt and its Application to Nondestructive Evaluation,
NTIAC report 79-2 (1979) (Nondestructive Testing Information Analysis Center, San Antonio, Texas)
1-49.
[2] W.A. Theiner, E. Waschkies, Method for the non-destructive determination of material states by use of
the Barkhausen-effect (in German), Patent DE 2837733C2 (1984).
[3] G. Dobmann et al, Barkhausen Noise Measurements and related Measurements in Ferromagnetic
Materials; in Volume 1: Topics on Non-destructive Evaluation series (B.B. Djordjevic, H. Dos Reis,
editors), Sensing for Materials Characterization, Processing, and Manufacturing (G. Birnbaum, B. Auld ,
Volume 1 technical editors), The American Society for Non-destructive Testing (1998) ISBN 1-57117067-7.
[4] I. Altpeter, et al., Electromagnetic and Micro-Magnetic Non-Destructive Characterization (NDC) for
Material Mechanical Property Determination and Prediction in Steel Industry and in Lifetime Extension
Strategies of NPP Steel Components, Inverse Problems 18 (2002) 1907-1921.
[5] M. Borsutzki, Process-integrated determination of the yield strength and the deep drawability properties
rm and 'r on cold-rolled and hot-dip-galvanized steel sheets (in German); Ph.D. thesis, Saar University,
Saarbrcken, Germany, 1997.
[6] B.Wolter, G. Dobmann, Micromagnetic Testing for Rolled Steel, European Conference on Nondestructive Testing (9) (2006) Th. 3.7.1, 25.-29. 09. 2006, Berlin.
[7] M. Abuhamad, I. Altpeter, G. Dobmann, M. Kopp, Non-destructive characterization of cast iron
gradient combustion engine cylinder crankcase by electromagnetic techniques (in German), DGZfPAnnual Assembly (2007), Frth (to be published).
[8] IZFP Annual Report, (2004), Saarbrcken, Germany.
[9] W. A. Theiner et al., Process Integrated Nondestructive Testing For Evaluation Of Hardness; in the
proceedings of the 14th World Conference on Nondestructive Testing (14th WCNDT), (1996) 573, New
Delhi, India.
[10] B. Wolter et al., Detection and Quantification of Grinding Damage by Using EC and 3MA Techniques,
in the proceedings of the 4th International Conference on Barkhausen Noise and Micromagnetic Testing,
03-04 July 2003, Brecia, Italy, 159-170.

26

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-26

3D Nonlinear Finite Element Analysis of


Electromagnetic Inspection of Defects in Steel
N.Takahashi 1 and Y.Gotoh 2
Abstract. The electromagnetic inspection technology is commonly used to evaluate defects in
steel in various power generation plants because of its ability to detect quickly. In this paper, the
electromagnetic inspection methods using an ac magnetic field, or a biased ac field are proposed.
The behavior of the flux in steel is examined using 3-D FEM that takes into account the initial
magnetization curve, hysteresis (minor loop) and eddy currents in order to improve the
inspection method.
Keywords. Electromagnetic inspection, 3-D nonlinear FEM, hysteresis, magnetization curve,
opposite side defect in steel

1. Introduction
The electromagnetic inspection method offers the possibility of quick detection of
various kinds of defects, such as the outer side defects and plural cracks [1]. Since the
permeability of the steel is usually not uniform, the electromagnetic inspection of steel
has a magnetic noise. Therefore, a strong magnetic field is used for the inspection of
defects in a ferromagnetic material, such as a steel wall of the oil tank or the steel tube
etc. [2].
The non-uniformity of permeability in steel is reduced when a magnetic field is
increased. Therefore, the ac or dc strong magnetic field is used for the inspection
method using dc field and eddy current testing or the magnetic flux leakage testing
(MFL) etc. [3,4].
In order to improve these inspection methods, the detailed analysis of 3-D flux and
eddy current distribution should be performed. However, a detailed examination of the
behavior of flux distribution etc. under the electromagnetic inspection using a strong
magnetic field is difficult, because the magnetic property of steel is nonlinear and eddy
currents are induced. Although there has been a lot of research in defect detection
analysis [5-7], little work has been done in the 3-D analysis when the material is
magnetic.
In this paper, problems of 3-D finite element analysis in electromagnetic inspection,
such as the inclusion of minor loops for the detection under ac and dc excitation, are
discussed. The electromagnetic inspection method using a dc magnetic field and a
minute alternating magnetic field has been proposed [8]. The behavior of flux in steel is
examined using a 3-D finite element method that takes into account hysteresis (minor
loop) and eddy currents. It is shown that the detection of a defect is possible by the
differential permeability of the minor loop as its position in the BH plane is affected by
1
Norio Takahashi is with Department of Electrical and Electronic Engineering, Graduate School of Natural
Science & Technology, Okayama University, 3-1-1 Tsushima, Okayma 700-8530, Japan. (telephone: +8186-251-8115, fax: +81-86-251-8258, e-mail: norio@elec.okayama-u.ac. jp).
2
Yuji Gotoh is with Department of Mechanical and Energy Systems Engineering, Faculty of Engineering,
Oita University, 700 Dannoharu, Oita, 870-1192, Japan. (telephone: +81-97-554-7795, fax: +81-97-554-7790,
e-mail: goto-yuuji@cc.oita-u.ac.jp).

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

27

the existence of the defect. The flux and eddy current distribution in steel, having plural
cracks, are analyzed and it is shown that the detection of plural cracks, when the
distance between them is very short, is possible using the differential type two search
coils which is set parallel to the steel plate [9]. Steel tubes are used in the heat
exchanger in petrochemical plants. Since these steel tubes are grouped in a bundle, it is
necessary to inspect the existence of defects from the inside of each tube. Therefore,
the possibility of detection of outer side defects in steel tubes using an inner coil in the
alternating flux leakage test is examined by analyzing the detailed behavior of flux and
eddy current distribution [10].
2. Evaluation of Detection Method of Opposite Side Defect using DC Field and
Minute AC Field Taking Account of Minor Loop
2.1 Model and Method of Analysis
Figure 1 shows the model of 1/2 the domain needed to analyze the detection of the
opposite side defect in a steel plate. This model is composed of the yokes for dc and a
minute ac magnetic field and a search coil. The dc exciting current is 3A and the
minute ac exciting current is 0.5A (1kHz).
The flux density B in the steel is produced by the dc magnetic field and the minute ac
magnetic field. The magnetic field is analyzed using 3-D edge-based hexahedral
nonlinear FEM and the step-by-step method taking account of hysteresis (minor loop)
and eddy currents in the steel plate. The basic equation of eddy current analysis using
the A-Imethod is given by:

wA
Jo  V
 gradI

wt
wA

div  V
 gradI 0

wt

rot (QrotA)

(1)
(2)

where A is the magnetic vector potential, I is the scalar potential, v is the reluctivity, Jo
is the current density and V is the conductivity. The minor loop is modeled using these
upper hysteresis curves (SS400) [1]. It is assumed that the obtained B and H are at the
point b (Hmin, Bmin) on the upper loop as shown in Figure 2. If the calculated flux
density Bc at the Newton-Raphson (N-R) iteration is larger than Bmin, then Bc should be
located at the point d (Hd, Bc) on the lower minor loop. Hd on the lower minor loop is
given by the following equation, if the upper loop is symmetrical with respect to the
middle point e:
Hd

H g  2 H e  H g .

(3)

2.2 Behavior of Minor Loop and Inspection of Opposite Side Defect


Figure 3 shows the calculated result of a minor loop at a surface point in the steel
with and without an opposite side defect. The figure indicates that the dc flux density
near the defect in steel is increased when there exists an opposite side defect and a
minor loop is generated at the high flux density on an initial magnetization curve. The
figure shows that the differential permeability of the minor loop becomes small, when
there exists an opposite side defect in the steel.
Figure 4 shows the distribution of the differential relative permeability Pd on a minor
loop with and without an opposite side defect. Figure 4 (a) shows that the differential
relative permeability is increased near the surface of the steel plate when there are no

28

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

22

52.5

.5

.5
16 20

19

dc-yoke
dc-exciting coil (3A)

initial magnetization curve


upper minor curve

29

33

steel plate (SS400)

3.35

1.35
7.65

lift-off=0.1
3

f (Hf, Bf)

150
ac-exciting coil
(1kHz, 0.5A)

g (Hg, Bg)

z
y

a (Hmax, Bmax)
2.7
9 1.4
4 ac-yoke
search coil
10.18
0.56

4.28

4.7

10
5

c (Hc, Bc)

2.95

e (He, Be)
d (Hd, B)

Dd=1

opposite side defect

b (Hmin, Bmin)
Dw

lower minor curve

Dw=0.5

Figure1. Inspection model of steel plate with outer side


defect (1/2 domain).
2

Figure 2. Explanation of minor loop.


ac-yoke

search coil (40turns)


differential Pd

1.6

1200

B (T)

with an opposite side defect

1000

1.2

800

without an opposite side defect

600

0.8

400

initial B-H curve

200

0.4
0

2000

4000
6000
H (A/m)

8000

steel plate (SS400)

(a) without an opposite side defect

differential Pd
1200

Figure 3. Effect of opposite side defect on


minor loop (calculated).

1000
800
600
400
200
0

opposite side defect

(b) with an opposite side defect


Figure 4. Distribution of the differential relative
permeability Pd on a minor loop in the steel plate
with and without an opposite side defect (dc=3A,
ac=1kHz, 0.5A).

defects. However, the permeability between the opposite side defect and the surface of
the steel is reduced as shown in Figure 4 (b). Since the dc flux is distributed near the
surface of the steel plate, the minute ac flux density dose not penetrate as far as the
opposite side defect.
During the inspection, the dc and ac yokes are moved along the x-direction by 1mm
pitch, keeping the lift-off at 0.1mm. Figure 5 shows the measured inspection result of
the change of ac flux density 'B detected by the search coil due to the outer side defect.
'B is defined by

'B=|B| (at defect position)-|B| (at no defect position, x=10mm)

(4)

29

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

'B in the search coil is calculated by 3-D non-linear FEM taking account of
hysteresis (minor loop) and eddy current. After 35 steps (about 2 periods), almost
steady state result can be obtained. The figure shows that the measured and calculated
peak values of 'B are almost in good agreement. Moreover, 'B is decreased near the
position of the opposite side defect. This is due to the fact that the dc flux density near
the position of the opposite side defect in steel plate is increased and as a result, the
permeability of the steel is decreased. The figure suggests that it is possible to detect
the opposite side defect in a steel plate by using the proposed inspection method.
measured

'B (x10-4T)

calculated
-20

-40

Dd=1

-60
-10 -7.5 -5 -2.5

2.5

5 7.5 10
position x (mm)
opposite side defect

Dw=0.5

Figure 5. Change of flux density 'B when there exists an opposite side defect (dc=3A, ac=1kHz, 0.5A).

3. Alternating Magnetic Flux Leakage Testing for Detection of Plural Cracks


3.1 Inspection Method and Modeling
Figure 6 (a) shows a model of the alternating magnetic flux leakage test that detects
plural cracks. The amplitude of the current is 1A(rms). The gap between the leg of the
yoke and the surface of the steel is 0.2mm. Figure 6 (b) shows the proposed differential
type search coils for detecting the parallel component Bx of leakage flux from plural
cracks. The leakage flux Bx which is uniformly distributed over the steel surface is
measured using the search coil E. The local leakage flux Bx from the cracks is
measured using the search coil D. The difference between Bx obtained by the search
coil D and that by the search coil E is used for detecting the number and the size of
plural cracks. The sizes of these coils were optimized using the evolution strategy. The
search coils (thickness < 0.1mm) can be located under the legs of the yoke. The
frequency is chosen as 1kHz, resulting in a skin depth of 0.15mm (the relative
permeability is assumed to be 2500).

yoke
Yoke

Plate

Distance L:0.5mm
y

29

w=S

16.5

Sw
-

52

29

Steel
plate
steel plate
(SS400)

22
.5

19

52.552.5

19

56.5

exciting
Exciting
coilcoil

10
5

Bx search coil E

SlSl-E=59
=59



0.1

0.06

150

search coils
Search
coil
y

Crack (Width Cw:: 0.01mm)

Depth Cd:1mm
:1mm

Bx search coil D

Position of search coil D (D)

(a) birds eye view (1/2 region)


(b) search coils for detecting plural cracks
Figure 6. Model for alternating flux leakage testing of plural cracks.

30

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

Cd=1

Bx [x10-4 T]

100
80

Cd=0.5

60

Cd=0.1

40
20
0
-20
-1.5

No-crack
-1

-0.5

L=0.5

Figure 7. Detection of Bx using differential search coil


(Cw=0.01mm, Cd=1mm, L=0.5mm, 1kHz, 1A,
calculated, with hysteresis).

0.5

1
1.5
Position [mm]

Crack position
(Cw=0.01)

Figure 8. Effect of depth of cracks on Bx


(1kHz, 1A, calculated, with hysteresis)

3-D FEM using the 1st order hexahedral edge element is applied. The flux and eddy
current are analyzed by the step-by-step method taking account of the non-linearity of
the steel plate. In order to obtain the steady state result, the calculation is carried out for
2.5 periods (=160 steps). The time interval t of the step-by-step method is chosen as
1.5625x10-5 sec.
3.2 Detection of Plural Cracks using Horizontal Coils
The detection characteristics are analyzed by a nonlinear calculation using the
hysteresis curves. Figure 7 shows the distribution of Bx of one, two and four cracks
detected using the differential type coil. The figure illustrates that it is possible to
evaluate plural cracks by using the amplitude of the parallel component Bx of leakage
flux detected by the differential search coil. Figure 8 shows the effect of the depth of
the cracks when there are two cracks. The figure shows that the amplitude of Bx is
increased as the crack depth grows.
4. Electromagnetic Inspection Method of Outer Side Defects on Steel Tubes using
an Inner Coil
4.1 Inspection Model and Conditions
Figure 9 shows the inspection model of the outer side defect of a steel tube
(SUS430). The proposed inner inspection probe is also shown in the figure. The
inspection probe is composed of a yoke, an exciting coil and a search coil for detecting
the perpendicular component |Bx| of leakage flux due to the outer side defect. The outer
side defect is a circumferential one. The frequency is chosen as 60Hz (commercial
frequency) resulting in a skin depth equal to 2.31mm (maximum relative permeability
(=434) in the steel tube is used for the calculation of the skin depth). The exciting
current is 5A(rms). The conditions of analysis and experiment are as shown in Table 1.
4.2 Outer Side Defect in a Steel Tube
Figure 10 shows the distribution of calculated and measured |Bx| in a search coil. |Bx|
is obtained by moving the inspection probe in the z-direction inside the steel tube at the

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

Table 1. Condition of analysis and experiment


202 turns, 60Hz, 5A (rms)
50 turns, Width (Sw)=3mm, Height (Sh)=1.65mm,
Length (Sl)=2.56mm, Lift-off (Lo)=0.2mm
SUS430, V=1.82x106 S/m, Diameter=25I, Thickness (t)=1.5mm, Pr-max=434
SS400, V=7.51x106 S/m, Maximum relative permeability Pr=3000
Width (Dw)=2mm, Depth (Dd) =0.5mm, Circumference defect
88095, 83232
N-R method: 1.0x10-3 T, ICCG method: 1.0x10-5

Exciting coil
Search coil
Steel tube
Yoke
Outer side defect
Nodes and elements
Convergence criterion

steel tube
(SUS430, 25)
exciting coil
(202turns)

70
60

-4

search coil (Bx)

Bx [x10

Sw=3

80
59.6

50turns

me asured
40
30
20

4.5
Sl=2

20.4

calculated

50
T]

10.5

31

.56
S

5
1.6
h=

yoke
(SS400)

10
0
-10

-5

z
t=1.5

x
D w=2

Figure 9. Inspection model for outer side


defect of steel tube (1/4 domain).

10
position z [ mm]

position of outer side defect

Figure 10. Inspection waveform of leakage flux (60Hz,5A).

lift-off of 0.2mm. The calculated result is in agreement with measurement. The figure
illustrates that the outer side defect in the steel tube can be detected using the proposed
inspection method.
4.3 Application to the Steel Tube with Aluminum Cooling Fin
In a petrochemical plant, the steel tube is sometimes used with aluminum cooling
fins. Therefore, the possibility of applying the proposed inspection method for
detecting the outer side defects on a steel tube with cooling fin is examined. Figure 11
shows the inspection model of the steel tube (SUS430) with aluminum cooling fin. The
outer side defect width (Dw), defect depth (Dd) and defect length (circumference
direction, Dl) are 0.5mm, 0.5mm and 10mm, respectively. The nonlinear analysis was
carried out. The non-uniformity of permeability and conductivity in steel is neglected.
The conductivity V of the aluminum fin is 3.5x107 S/m.
Figure 12 shows the eddy current distributions in the steel tube with and without fin.
The figure shows that the eddy current density is maximum at the inner boundary of the
fin with high conductivity. The maximum flux density |B|max in the steel tube with fin is
1.68T and |B|max without the fin is 1.51 T. The flux density in the steel tube is increased
when the fin is put on the tube because of the opposing flux due to the eddy currents.
Figure 13 shows the average flux density |Bx| in a search coil calculated along the zposition inside the steel tube with and without the fin. The results under the dc
excitation are also shown. The figure shows that the leakage flux from the steel tube
with the fin is larger than that without the fin under ac excitation. The leakage flux is
increased due to the larger than that without the fin under an ac excitation. The leakage

32

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection


steel tube (SUS430, 25)

1 .5

2.5

2 2

search coil (Bx)

tube
fin

yoke
fin of aluminum

exciting coil

exciting coil
(202turns)
yoke (SS400)

outer side
defect
Dw=0.5,
Dd=0.5,
Dl=10

z
y

27.5

Figure11. Inspection model of the steel tube with the aluminum fin.
eddy current
[x105 A/m2]
14
12
10
8
6
outer side
defect

4
2

steel tube
(SUS430)

1.5 2.5

fin of aluminum

(a) without fin


 (b) with fin
Figure 12. Distribution of eddy current density in steel tube with the
aluminum fin (60Hz,5A, Dw=0.5mm, Dd=0.5mm).
with fin (60Hz)
30

without fin (60Hz)

15

with fin

20
10

with fin and without fin (DC)


: with fin
: without fin

0
-8

-6

-4

-2

10

without fin

4
6
8
position z [mm]

position of the fin of aluminum


position of outer side crack (Dw=0.5, Dd=0.5, Dl=10)
2

Bx [x10-4 T]

Bx [x10-4 T]

20

0
0

20

40

60 80 100 120 140 160 180 200


exciting frequency [Hz]

Figure 13. Effect of aluminum fin and alternating


magnetization (5A, Cw=0.5mm, Cd=0.5mm, calculated).

Figure 14. Effect of exciting frequency


(5A, calculated).

N. Takahashi and Y. Gotoh / 3D Nonlinear Finite Element Analysis of Electromagnetic Inspection

33

increased due to the eddy current in the steel tube and the fin. The ac excitation at 60Hz
is appropriate from the viewpoint of the amplitude of the output signal as compared
with the dc excitation.
Figure 14 shows the maximum flux density in a search coil calculated by changing
the exciting frequency in a model with and without fin when the exciting current is 5A
(rms). As seen in the figure, the exciting frequency that is suitable for the inspection is
around 20Hz 60Hz.
5. Conclusions
The results obtained are summarized as follows:
(1) The principle of the inspection method of opposite side defects with ac and dc
excitations is clarified by analyzing the magnetic property of the material. When the
large dc magnetic field is impressed on a steel plate with the opposite side defect, the
dc flux in steel bypasses the defect because of the magnetic saturation. As a result,
the flux density between the opposite side defect and the surface of the steel is
increased. The differential permeability of the minor loop due to the minute ac
magnetic field is therefore decreased. This phenomenon can be used for the
inspection of opposite side defects.
(2) It is possible to evaluate plural cracks by using the amplitude of the parallel
component Bx of the leakage flux detected by the differential search coil.
(3) The alternating flux leakage test method using an inner coil at commercial
frequency is able to detect the outer side defect in steel tubes with or without
aluminum cooling fin. The ac excitation is more suitable than the dc excitation.
References
[1] Y.Sun, An introduction to electromagnetic nondestructive testing, Applied Electromagnetics and
Mechanics, vol.13, pp.145-152 (1998).
[2]N.KasaiK.Sekine, and H.Maruyama, Non-destructive evaluation method for far-side corrosion type
flaws in oil storage tank bottom floors using the magnetic flux leakage technique", J. Jpn. Petrol. Inst.,
vol.46, no.2, pp.126-132 (2003).
[3] H.Fujiwara, T.Sakamoto, T.Nishimine, and K.Kokubo, Development of ac magnetic leakage flux testing
system, Proc. Int. Symp. Applied Electromagnetics and Mechanics, pp.527-528 (2001).
[4] S.Nishino, Development of magnetizing eddy current pipeline inspection tools, The First US-Japan
Symposium on Advances in NDT, pp.254-258 (1996).
[5] G.Chen, T.Sugibayashi, M.Shiwa, and H.Yoneyama, Investigation of subsurface flaw detectability of
magnetic flux leakage testing, Proc. Int. Symp. Applied Electromagnetics and Mechanics, pp.525-526
(2001).
[6] K.Sekine, Y.Zhang, A.lizuka, and K.Nonaka, A Theoretical analysis of magnetic force acting on
magnetic particles in the Immediate vicinity of surface flaws, The First US-Japan Symposium on
Advances in NDT, pp.396-401 (1996).
[7] M.Katoh, K.Nishio, T.Yamaguchi, and S.Mukae, FEM study on magnetic test of square bar by direct
contact method, Proc. FENDT 94 and ROCSNT 9th Annu. Conf., pp.79-85 (1994).
[8] Y.Gotoh and N.Takahashi, Evaluation of detecting method with AC and DC excitations of opposite side
defect in steel using 3D non-linear FEM taking account of minor loop , IEEE Trans. Magn., vol.44 (2008).
[9] Y.Gotoh and N.Takahashi, Detection of plural cracks in steel using horizontal coils 3D FEM analysis
considering hysteresis and non-uniformity of steel , IEEJ Trans. FM, vol.125, no.10, pp.835-840 (2005).
[10] Y.Gotoh and N.Takahashi, 3D FEM analysis of electromagnetic inspection of outer side defects on
steel tube using inner coil, IEEE Trans. Magn., vol.43, no.4, pp.1733-1736 (2007).

This page intentionally left blank

Magnetic Materials

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-37

37

Evaluation of Irradiation Embrittlement in


Fe-Cu-Ni-Mn Model Alloys by
Measurements of Magnetic Minor
Hysteresis Loops
Satoru KOBAYASHI a,1 , Hiroaki KIKUCHI a , Seiki TAKAHASHI a , Katsuyuki ARA a
and Yasuhiro KAMADA a
a
NDE and Science Research Center, Faculty of Engineering, Iwate University, 4-3-5
Ueda, Morioka 020-8551, Japan
Abstract. Magnetic minor hysteresis loops of neutron-irradiated Fe-Cu-Ni-Mn
model alloys varying Cu and Ni contents have been measured. For almost alloys
minor-loop coefcients which are in proportion to internal stress decrease after neutron irradiation to a maximum uence of 0.44 1019 n cm2 . The decrease of
the coefcients is strongly enhanced for alloys with high Cu and high Ni contents.
Both magnetic and mechanical properties are correlated with each other and the
coefcients are roughly in inverse proportion to yield strength.
Keywords. Neutron irradiation, Magnetic minor hysteresis loops, precipitates

Introduction
Neutron irradiation induces microstructural changes associated with an increase in a
number density of nanoscale defects including vacancies, dislocation loops and precipitates, and makes materials brittle and more susceptible to fracture [1]. Currently, the
irradiation embrittlement of pressure vessels in nuclear reactors is evaluated by ductilebrittle transition temperature (DBTT) obtained in Charpy impact tests. However, the
stock of Charpy specimens preinstalled in the reactors is diminishing and nondestructive
evaluation of the irradiation embrittlement has become an urgent matter of study.
Recently, we measured magnetic minor hysteresis loops of some neutron irradiated
Fe-Cu-Ni-Mn model alloys with high Ni content to investigate the inuence of neutron
irradiation on magnetic properties [2]. It was found that minor-loop coefcients which
are obtained from power-law relations between minor-loop parameters and are sensitive
to internal stress [3], decrease with uence. This shows the reduction of internal stress
during neutron irradiation and was explained as being due to a compensation of internal
stress of dislocations by Cu precipitates grown around the dislocations. Nevertheless, dependence of elemental content on the coefcients as well as their relation with mechan1 Corresponding Author: Satoru Kobayashi, NDE and Science Research Center, Iwate University, 4-3-5
Ueda, Morioka 020-8551, Japan, E-mail: koba@iwate-u.ac.jp .

38

S. Kobayashi et al. / Evaluation of Irradiation Embrittlement in Fe-Cu-Ni-Mn Model Alloys

Table 1. Chemical compositions of measuring samples. (wt %)


Sample

Cu

Ni

Mn

OV4

0.05

0.8

1.6

OV5

0.05

1.6

1.6

OV6
OV7

0.1
0.1

0.8
1.6

1.6
1.6

OV8

0.05

1.6

1.6

OV17

0.2

1.6

1.6

Si

0.025

0.5

Table 2. Neutron irradiation conditions in this study.


Fluence t
(1019 n cm2 )

Effective uence te
(1019 n cm2 )

Flux
(1012 n cm2 s1 )

Flux regime

T29

0.89

high

0.44

0.44

T30

0.07

low

0.02

0.07

T31
T32

0.72
0.26

high
intermediate

0.06
0.02

0.07
0.04

ical properties were not investigated in detail. In this paper, we study the inuence of
Cu and Ni contents on minor hysteresis loops in neutron-irradiated Fe-Cu-Ni-Mn model
alloys.

1. Experimental
Neutron-irradiated tensile test samples with dimensions of 24 5 0.5 mm 3 were
prepared by Odette group of University of California Santa Barbara (UCSB). Fe-Cu-NiMn model (OV) alloys with variable combinations of Cu and Ni contents listed in Table 1
were neutron irradiated at 290 C. We examined four irradiation conditions with uence
t up to 0.44 10 19 n cm2 and with neutron ux in the range 0.07-0.89 10 12 n
cm2 s1 as listed in Table 2.
A set of magnetic minor hysteresis loops with various eld amplitudes H a was measured at room temperature using a closed magnetic circuit with a sample sandwiched
between two magnetic yokes. A cyclic magnetic eld with a frequency of 1 Hz and H a
up to 6 kA/m was applied along the long axis of the sample. Our analysis showed that
there exist several power-law relations between minor-loop parameters in a limited range
of Ha in which irreversible movement of Bloch wall mainly contributes to magnetization [3]. From the relations, minor-loop coefcients in proportion to internal stress were
determined. In this study, we paid attention to the minor-loop coefcient W F0 , obtained
from the power-law relation, given by
 nF
Ma

0
WF = WF
,
(1)
Ms
where Ma and WF are magnetization and hysteresis loss of a minor hysteresis loop
and Ms are saturation magnetization [3]. From least-squares ts of W F Ma curves
to equation (1), the exponent of n F = 1.70 0.03, which is independent of chemical
compositions and neutron uence, as well as W F0 were determined. Here, minor loops

S. Kobayashi et al. / Evaluation of Irradiation Embrittlement in Fe-Cu-Ni-Mn Model Alloys

39

with 0 Ma = 0.31.3 T were used for the ts. The obtained coefcient was typically
averaged over 2 - 3 samples for each alloy-irradiation condition and an experimental
error of WF0 mainly results from the sample dependence.
To compare W F0 obtained with various neutron uxes, an effective neutron uence
t
e , given by te = t (r /)1/2 , was introduced [2,4]. Here, r is a reference ux
and r = 0.89 10 12 n cm2 s1 was assumed. This equation was originally proposed to
normalize a ux-dependent change of yield strength [4] and is also useful for magnetic
properties; on the te scale, all minor-loop coefcients obtained for various neutron
uxes fall in the same smooth curve, although lower ux tends to shift the coefcients to
lower t [2].

2. Results and Discussion


Figure 1 shows WF0 as a function of te . For all OV samples, the behaviour of W F0 seems
to be classied into two groups. One is a behaviour seen for OV6 and OV7 that W F0
increases at low uence, shows a maximum around te = 0.05 10 19 n cm2 , and then
gradually decreases with uence. The other is a behaviour that W F0 shows a monotonic
decrease with uence (OV5, OV8, OV17) or is almost constant against uence (OV4).
For OV5, OV7, and OV17, both Ni and Mn contents are 1.6 wt% and only the Cu content
is different from each other. The decrease of W F0 after irradiation to te = 0.44 10 19 n
cm2 is proportional to Cu content and is most pronounced for OV17 with 0.2 wt% Cu.

Figure 1. WF0 as a function of effective neutron uence.

On the other hand, a behaviour of W F0 is also inuenced by Ni content. For OV4


and OV6 with 0.8 wt% Ni, W F0 is weakly dependent on uence and the decrease after

40

S. Kobayashi et al. / Evaluation of Irradiation Embrittlement in Fe-Cu-Ni-Mn Model Alloys

irradiation is largely reduced compared with that for OV5 and OV7 with 1.6 wt% Ni,
respectively. There observations clearly show that the higher level of Cu and Ni inclusion
enhances the decrease of internal stress during neutron irradiation. Note that Si inclusion
also seems to enhance a decrease in W F0 after irradiation when data for OV5 and OV8 are
compared. The effects of Si inclusion on minor-loop properties are subjects for further
study.
We now examine a relation of W F0 to yield strength y obtained by tensile tests
performed by UCSB group. Figure 2 shows relations between W F0 and a change in yield
strength y for all uences. Since y generally increases with uence, higher y
corresponds to that for higher te . For all OV samples, WF0 is almost constant or linearly
decreases against y . Moreover, the higher Cu and Ni contents the larger a change both
in WF0 and y . This correlation implies that a decrease in W F0 originates from the same
irradiation mechanism as that for the increase of yield strength.

Figure 2. Relation between WF0 and y at all measuring uences.

For all materials, neutron irradiation induces changes in mechanical properties.


These changes are primarily due to the formation of irradiation defects such as Cu rich
and Cu-Mn-Ni precipitates [5]. Since these defects act as obstacles to the movement of
Bloch wall, minor-loop coefcients will increase with uence and Cu and Ni contents.
However, WF0 rather decreases with uence, whereas for some OV samples a peak appears at low uence. This implies the presence of two irradiation mechanisms that affect
magnetic properties; the rst mechanism is due to the formation of irradiation defects
in the matrix and contributes to an increase in minor-loop coefcients and the other is
a mechanism that contributes to their decrease. One of possible mechanism to explain
such decrease is the preferential formation of Cu rich and/or Cu-Mn-Ni precipitates on
dislocations [6,7]; precipitates are preferentially formed on dislocations to minimize the
elastic energy and reduces the internal stress around the dislocations. With increasing
Cu and Ni contents, the volume fraction of precipitates also increases [5]. This would
enhance the precipitation around the dislocations, resulting in a large decrease of W F0

S. Kobayashi et al. / Evaluation of Irradiation Embrittlement in Fe-Cu-Ni-Mn Model Alloys

41

for OV17 with high Cu and high Ni contents. This enhanced precipitation, on the other
hand, makes the material mechanically harder [1], which is reason why W F0 is inversely
proportional to y .
In conclusion, we have shown that the minor-loop coefcient is roughly in inverse
proportion to yield strength, which is an important mechanical property related with
irradiation embrittlement. This result clearly shows that magnetic method using minor
hysteresis loops is useful for nondestructive evaluation of irradiation damage in nuclear
reactor pressure vessels.

Acknowledgements
This research project has been conducted under the research contract with the Japan
Nuclear Safety Organization (JNES). We thank Prof. G. R. Odette of UCSB for allowing
us to measure neutron irradiated samples.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]

G. R. Odette, G. E. Lucas, Radiat. Eff. Def. Solids 144 (1998) 189.


S. Kobayashi, H. Kikuchi, S. Takahashi, K. Ara, Y. Kamada, Studies in Applied Electromagnetic and
Mechanics 28 (2007) 217.
S. Takahashi, S. Kobayashi, H. Kikuchi, Y. Kamada, J. Appl. Phys. 100 (2006) 113908.
G. R. Odette, T. Yamamoto, D. Klingensmith, Philos. Mag. 85 (2005) 779.
S. C. Glade, B. D. Wirth, G. R. Odette, P. Asoka-Kumar, J. Nucl. Mater. 351 (2006) 197.
S. Takahashi, H. Kikuchi, K. Ara, N. Ebine, Y. Kamada, S. Kobayashi, M. Suzuki, J. Appl. Phys. 100
(2006) 023902.
M. K. Miller, K. F. Russell, M. A. Sokolov, R. K. Nanstad, J. Nucl. Mater. 361 (2007) 248.

42

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-42

Analysis of Barkhausen Noise


Characteristics and Mechanical Properties
on Cold Rolled Low Carbon Steel
Hiroaki KIKUCHI a,1, Tomoki KOSHIKA a , Tong LIU a , Yasuhiro KAMADA a,
Katsuyuki ARA a, Satoru KOBAYASHI a and Seiki TAKAHASHI a
a
NDE & SRC, Faculty of Engineering, Iwate University, 4-3-5 Ueda, Morioka, Iwate,
020-8551, Japan

Abstract. Low carbon steel specimens cold rolled at ratios of 0 - 40 % have been
examined comprehensively by magnetic Barkhausen noise (MBN) method, and
their microstructure were studied by transmission electron microscope. In order to
correlate MBN parameters with those mechanical properties, Vickers hardness and
ductile-brittleness transition temperature (DBTT) were also evaluated. MBN
energy and rms voltage rise rapidly with cold rolling below 10 %, and saturate at
higher rolling ratio. This phenomenon is attributed to the combined effects of cell
texture and dislocation density. It is also found that good correlation between
MBN parameters and Vickers hardness, DBTT.
Keywords. Barkhausen noise, Microstructure, Vickers hardness, DBTT

1. Introduction
Magnetic Barkhausen noise (MBN) has been widely used to study the ferromagnetic
materials nondestructively and there have been investigations of MBN changes under
various conditions including fatigue, irradiation, stress and so on [1-3]. It has been
proved to be sensitive to grain size, composition, hardness and so on [4, 5]. Therefore,
this method is becoming a potential tool for many nondestructive evaluation (NDE)
applications. However, the effect of microstructure changes induced from cold rolling
on MBN has been rarely investigated. Though Athertons group has studied MBN
properties in cold rolled nuclear vessel steel [6], their specimens possessed an easy axis
in the unrolled state as a result of crystallographic texture, which made the exclusive
study of rolling texture by MBN method very difficult. For practical uses, it is
important to investigate that effect and a relation between MBN parameters and
mechanical properties. In this work, low carbon steel plates with negligible
crystallographic texture before deformation by cold rolling were investigated to
understand the origin of MBN dependence on rolling texture, and to correlate those
MBN parameters with those mechanical properties.

1
Corresponding Author: Hiroaki Kikuchi, NDE & Science Research Center, Faculty of Engineering,
Iwate University, 4-3-5 Ueda, Morioka 020-8551, Japan; Phone:+81-19-621-6350; Email: hkiku@iwateu.ac.jp

H. Kikuchi et al. / Analysis of Barkhausen Noise Characteristics and Mechanical Properties

2. Experimental Procedure

43

Excitation coil

Low carbon steel (S15C) plates were


prepared and then annealed at 1173 K for
one hour, followed by air-cooling. Then,
50 mm
40 mm
13 mm
Fi-Si
those steels were deformed by cold rolling
60 mm
with different reduction ratios of 0, 5, 10,
Fig. 1 Schematic view and dimensions of yoke
20 and 40 %. The steel contains 0.15
probe.
0.20wt.%C, 0.150.35wt.%Si, 0.30
0.60wt.%Mn and Fe in balance. For
MBN measurement, five kinds of
specimens were ground into the same
dimension 40 60 10 mm3.
In order to evaluate ductile-brittleness
transition temperature (DBTT) Charpy
impact test pieces, 10 10 55 mm3,
were machined from each plate along the
rolling direction. Disk specimens were
Fig. 2 MBN measurement setup.
also prepared to evaluate microstructures
by Philips-tecnai 30 transmission electron microscope (TEM).
The plate specimens were magnetized parallel to the rolling direction by a triangle
wave of 1 Hz using a yoke probe. The yoke probe was composed of a U-type Fe-Si
yoke and an excitation coil as shown in Fig. 1. MBN signals were detected by a MBN
sensor, i.e., an air-core coil with 305 turns and 10 mmI, attached on the surface of
specimens. The measurement system is illustrated in Fig. 2. The original MBN signals
were amplified (60 dB), filtered (10-100 kHz), and finally sampled at 380 kHz. The
MBN results of each sample were measured 10 times and averaged.
The MBN parameters dependencies on the cold rolling ratio were studied in terms
of MBN energy and rms voltage. MBN energy, EMBN, and rms voltage, Vrms are defined
as follows.

EMBN

0 VMBN dt

(1)

1 T 2
(2)
VMBN dt
T 0
where VMBN is original MBN signal and T is half period of excitation field.
The absorption energy of the Charpy impact test pieces was obtained by using the
Charpy test machine at temperatures between 200 and 360 K. The ductile-brittleness
transition temperature (DBTT) was estimated from the absorption energy temperature
dependence of the Charpy impact test pieces [7]. The Vickers hardness of the
specimens was also measured using a Vickers hardness meter with a load of 500 g.
Vrms

3. Results and Discussion


Fig. 3 shows typical time dependent of MBN signals and the excitation current applied
to the yoke probe for 0% and 40 % reduction ratio specimens. From these profiles,
MBN energy and rms voltage was calculated using equations (1) and (2). MBN
parameters were plotted as a function of reduction ratio in Fig. 4. They rise rapidly in

H. Kikuchi et al. / Analysis of Barkhausen Noise Characteristics and Mechanical Properties

MBN energy EMBN (mV2)

480

32
MBN energy
rms voltage

400

28

320

24

240

20

160

16

80

rms voltage Vrms (mV)

44

12
0

10

20

30

40

50

Reduction ratio (%)

(a) 0 % reduction ratio

Fig. 4 The relation between MBN energy, rms


voltage and reduction rate.

the beginning with reduction ratio below


10%, and then slightly increase above 10%
cold rolling. In Arthertons work [6], cold
rolling brought about the increase of MBN
parameters. The authors attributed these
phenomena to crystalline anisotropy before
rolling. In our work, on the other hand, Xray analysis revealed that the crystalline
anisotropy is not observable before rolling
(b) 40 % reduction ratio
and the surface residual stress effect is also
Fig. 3 typical time dependent of MBN signal and
negligible after a surface grinding process.
excitation current applied to yoke probe.
Fig. 5 shows the micrograph of TEM
observation. One can see that dislocations are distributed homogeneously for 0% cold
rolling and a large amount of dislocations are generated with the increase of reduction
ratio. A cell substructure also comes into being gradually during this rolling process
and aligns with the direction parallel to the rolling direction. The formation of rolling
texture <110> was observed by X-ray analysis. The dislocation density Uestimated by
the Ham method [8] is 5 109 cm-2, 2 1010 cm-2, and 6 10 1010 cm-2 for 0%, 5%,
and 10% reduction ratio, respectively; for higher reduction, is not shown because of the
formation of cell structures, which makes it difficult to estimate a reliable value of U. It
is proposed that the increase of MBN parameters is attributed to the combined effects
of cell structure and dislocation density. TEM observation reveals that dislocations tend
to aggregate at the boundary of cell structure during the cold rolling, and the
dislocation density inside the cell texture is relatively low so that the domain wall can
come across. Consequently, increases in dislocation density with cold rolling lead to
sensitive increases in the domain wall gradient at pinning site, resulting in a larger
MBN response up to 10 % reduction ratio. On the other hand, above 10 % increases in
dislocation density becomes not significantly, whereas a higher rolling ratio induces a
large number of smaller cell structures, namely the boundary area increases, which
results in low MBN signal. Therefore, the total MBN response tends to saturate at high
reduction ratio. This MBN analysis method shows a potential application in evaluating
microstructure changes nondestructively in cold rolled steel.
The relations between MBN parameters and mechanical properties, DBTT and
Vickers hardness, are shown in Fig. 6. DBTT and Vickers hardness increases as
reduction ratio increases due to work hardening. MBN parameters rise rapidly with
increasing mechanical parameters below 10 % reduction ratio, and then they increase

45

H. Kikuchi et al. / Analysis of Barkhausen Noise Characteristics and Mechanical Properties

(a) 0 %
(b) 5 %
(c) 10 %
Fig. 5 TEM micrographs showing dislocations in S15C steel.

24

300

22

260

20

220

18

180

16

140
130

140

150 160

170

180 190 200

Vickers hardness (Hv)

MBN energy EMBN (mV2)

340

26

380
340

24

300

22

260

20

220

18

180

16

140
245

14
210

MBN Energy
rms voltage

250

255 260

265

270 275

280

rms voltage Vrms (mV)

26
MBN Energy
rms voltage

rms voltage Vrms (mV)

MBN energy EMBN (mV2)

380

(d) 40 %

14
285

DBTT (K)

(a) MBN parameters vs. Vickers hardness


(b) MBN parameters vs. DBTT
Fig. 6 The relations between MBN parameters and mechanical properties, DBTT and Vickers hardness.

gently over 10 % reduction ratio. Though the relation between MBN parameters and
mechanical parameters is not simple proportional, MBN parameters shows the
increases monotonically as function of mechanical parameters. Thus, these results
represent a possibility of NDE for the mechanical properties using the Barkhausen
noise technique.

4. Conclusion
The MBN and mechanical properties of cold rolled low carbon steel were studied.
MBN energy and rms voltage rise sharply in an initial stage of cold rolling and gently
over 10% cold rolling ratio. The Vickers hardness and DBTT exhibits the same
tendency as MBN parameters. Dislocations increase with the increase of the rolling
reduction, which cause the increase of Vickers hardness and DBTT. The formation of
cell texture and the changes in dislocation density dominated the MBN properties.
Good correlations between MBN parameters and mechanical parameters were derived,
which show the Barkhausen technique is a good candidate of NDE for the mechanical
properties of cold rolled steel.

References
[1] C. C. H. Lo, J. Paulsen and D. C. Jiles, IEEE Trans. Magn., 40 (2004) 2173-2175.
[2] S. Palit Sagar, N. Parida, et al., Int. J. Fatigue, 27 (2005) 317-322.
[3] V. Moorthy, B. A. Shaw, S. Day, Acta Mater., 52 (2004) 1927-1936.
[4] J. Anglada-Riveraa, L. R. Padoveseb, J. Cap-Sncheza, J. Magn. Magn. Mater., 231 (2001) 299-306.
[5] O. Saquet, J. Chicois, A. Vincent, Mater. Sci. Engineering, A269 (1999) 7382.
[6] C. G. Stefanita, L. Clapham, J. K. YI, D. L. Atherton, J. Mater. Sci., 36 (2001) 2795-2799.
[7] Y. Kamada, T. Nakano, S.Takahashi, et al., J. Magn. Jpn., 28 (2004) 409412.
[8] R. Ham, Philos. Mag., 6 (1961) 11831184.

46

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-46

A Bridge Between NDE and Charpy


Impact Testing
Seiki TAKAHASHI 1 and Satoru KOBAYASHI
NDE and Science Research Center, Faculty of Engineering, Iwate University, Morioka
020-8551, Japan
Abstract. Charpy impact test is explained from the viewpoint of the transition of hammers
energy to the kinetic energy of dislocations. Ductile-brittle transition temperatureT is
represented as a simple function of the nucleation energy of dislocations Hn, the dislocation
density U, the maximum density of mobile dislocations Uo and the interaction energy of
obstacles with dislocations Uob; kT = UobUo +(Uo-U)Hn when U< Uo, while kT = UobUm when
U> Uo, where k is Boltzmann constant and Um is the density of mobile dislocations. The
relationship is compared with the experimental results for cold rolled S15C steels and
neutron irradiated low and high copper A533B steels of nuclear reactor pressure vessel
materials. The experimental results are qualitatively explained by the present model.
Keywords. Charpy impact test, dislocations, ductile-brittle transition, plastic deformation,
neutron irradiation

1. Introduction
Charpy impact test is an useful and reliable method to get information on ductility
of materials and has been traditionally used for a long time in the engineering
field.[1,2] The surveillance test of nuclear reactor pressure vessels (NRPVs) has been
carried out by Charpy impact method and the obtained ductile-brittle transition
temperature (DBTT) decides the lifetime of NRPV.
The mechanism of age degradation in NRPV has been investigated from the
viewpoint of microstructure and it is widely accepted that copper precipitates with size
of 2-3 nm, which nucleate and grow by the neutron irradiation, make material
brittle.[3,4] On the other hand, the nucleation mechanism has been recently
investigated by magnetic method in NRPV material and it was suggested that copper
precipitates gathering around dislocations through the elastic interaction also contribute
to the degradation.[5] Currently, the degradation in NRPV is evaluated by DBTT
obtained by Charpy impact test that is a macroscopic property. Nevertheless, there exist
few ideas to connect the macroscopic property with the microscopic one and its
physical meaning is vague at present. We need a bridge between the microstructure of
irradiation damages and the traditional mechanical properties for the practical use.
The relationship between DBTT and magnetic properties has been investigated
experimentally and it was found that coercive field increases in proportion to DBTT.[6]
Charpy impact method is a destructive test, whereas the magnetic method has a
characteristic of non-destructive evaluation. We need the physical model of DBTT to
explain the relationship of DBTT with magnetic properties. The purpose of the present
1
Corresponding Author: NDE&Science Research Center, Faculty of Engineering, Iwate University, 4-3-5
Ueda, Morioka 020-8551, Japan; Phone:+81-19-621-6431; Email: seiki.t@iwate-u.ac.jp

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

47

study is to give physical explanation for the DBTT and absorption energy from the
microscopic viewpoint.
In this study, we introduce a dislocation model representing the absorption energy
versus temperature and express DBTT by physical properties. The experimental results
are analyzed and DBTT and upper shelf energy are explained on the basis of the
present dislocation model in the neutron irradiated low carbon steels.

2. Charpy impact test


Ductile and brittle states are examined by Charpy impact test in which a hammer
accelerated by gravity potential collides with a test piece and the hammer energy lost
by the collision is measured. The lost energy is very small in the brittle state, whereas
in ductile state the hammer energy is absorbed into the test piece. The metallic
materials are brittle at low temperatures but change to ductile above the critical
temperature. This critical temperature is called ductile-brittle transition temperature.
The value of DBTT depends on the kinds of materials and increases with the
progression of degradation. The absorption energy in the ductile state also depends on
the degree of degradation and the ductility of materials is therefore evaluated by both
DBTT and the upper shelf energy. [1,2]
Samples are deformed plastically in the ductile state when a hammer collides. The
hammers energy is consumed for plastic deformation and is then transformed into
thermal energy. Plastic deformation is cause of the nucleation and motion of
dislocations from the microscopic viewpoint. On the other hand, in the brittle state
dislocations cannot move and the samples are broken down without the dislocation
motion.

3. Absorption energy and dislocations


In the ductile state, most of a potential energy of a hammer are transferred to a
kinetic energy of dislocations during the collision that is the main part of the absorption
energy in Charpy impact test. The mechanism of the energy transition depends on the
dislocation density. If there exist many dislocations enough to receive the hammers
energy, the interaction of dislocations with their obstacles decides the energy transition.
If dislocations do not exist enough, the hammers energy would be consumed for
nucleation of dislocations and their kinetic energy.
If the energy transformation is adiabatic, though transformation in actual Charpy
impact test is not adiabatic, the absorption energy is represented as a step function of
temperature as shown in Figure 1. The energy is discontinuously transformed at the
transition temperature T. The absorption energy of the step function can be
approximately expressed by the equation
Eabs = Uo exp[- (U/kT)n] ,

(1)

where n = f. Here, Uo is an upper shelf energy, k is Boltzmann constant and U is the


transition energy. The transition temperature T is given by

48

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

Absorption energy Eabs/ U0

1.2
adiabatic (n=)
1
n = 50
n = 30
n = 15

0.8
0.6
0.4

static (n = 1)

0.2
0
0

0.5

1.5

T/T
 
Figure 1. Energy transition processes for n = 1 (static), 15, 30, 50, f (adiabatic) cases.

U = kT

(2)

and is not the same value as the definition of DBTT.


Another extreme case of energy transformation is the quasi-static transition where
the energy is transformed at the constant temperature near T, during the infinite time. In
this case, the absorption energy is written as
Eabs = Uo exp[-(U/kT)],

(3)

which is represented in Figure 1.


The actual transition in Charpy impact test is represented as a model of the
intermediate of two examples, taking an adequate n in Equation (1), and an absorption
energy for n = 15, 30, 50 is plotted in Figure 1 as an example.
The energy transition in Charpy impact test is explained under supposition that any
phase transition does not occur in the temperature range of the measurement. The
hammers energy is transformed into the kinetic energy of dislocations during the
collision. The energy transition is performed perfectly near Twhen the dislocation
density is high enough to receive the potential energy and their mean free path is long
enough for dislocations to accumulate their kinetic energy. However, there exist a lot of
obstacles against the dislocation motion in real materials. The obstacles interact with
dislocations and restrict the dislocation motion. The interaction of obstacles with
dislocations is represented as an interaction energy Uob that depends on the strength of
the interaction, the density and distribution of obstacles. Dislocations themselves act as
the obstacles, make the mean free path short and the number of mobile dislocations
small near and above T.
When the dislocation density is not enough to receive the hammers energy, the
dislocations would be nucleated by the energy supplied from the hammer during the
collision. The absorption energy is written as, neglecting the obstacle effect,
Eabs = Uo exp[- ((Uo - U)Hn/kT)n] ,

(4)

where U is the dislocation density before the collision, Hn is the nucleation energy of
dislocations per unit length and Uo is the maximum dislocation density to receive the

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

49

potential energy of the hammer.


Generally, materials include obstacles and both obstacles and the dislocation
density contribute to the absorption energy. When U is lower than Uo, dislocations with
density of Uo-U should be nucleated and the absorption energy would be represented by
Eabs = Uo exp[-((Uob Uo +(Uo - U)Hn)/ kT)n]

(5)

On the other hand, when U is higher than Uo, the absorption energy is represented as
Eabs = Uo exp[-(Uob Um/ kT)n].

(6)

Here, Um is the density of mobile dislocations that is lower than Uo. Uo is the maximum
of Um. The dislocation density is not included in Equation (6) explicitly but Uob is
related to the density and distribution of dislocations, because dislocations play an
important role as obstacles even in the initial stage of plastic deformation.
The upper shelf energy Uo depends on the density of mobile dislocations and their
mean free path in the collision. Uo is the kinetic energy of dislocations; i.e.

Uo
F't

p2
U m ,
2m

'( pU m )
p

(7)

U m 'p  p'U m
'x
't

(8)

(9)

where p is the momentum of a dislocation per unit length and depends on the mean free
path x. m is the effective mass of a dislocation and F is the force acting on dislocations
during collision time 't. Dislocations are accelerated during 't. When x is long enough,
the momentum would become large. If the density of mobile dislocations Um is small
and x is short, 't would become short and the material is brittle. Uo depends on Um. The
strength of interaction energy and the test temperature restrict the density Um;
dislocations get over obstacles through the thermal activation process and the external
force F that is decided by the external conditions of a hammer.

4. Comparison with experimental results


The relationship between DBTT and dislocations has been investigated in S15 C
steel, where the dislocations are induced by cold rolling with 0, 5, 10, 20 and 40%
strain. The chemical composition of S15C steel is listed in Table 1. The absorption
energy was measured by use of Charpy impact test pieces with the standard size of
Table 1.

Chemical compositions of S15C steel.

S15C

Si

Mn

Fe

wt.%

0.16

0.20

0.44

balance

50

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

101055 mm. The impact tests were carried out with a pendulum of 27.6 kG and lift
angle of 138.5 in the temperature range of 200-363 K. Five V-notched Charpy samples
were tested at each temperature and both the largest and smallest values of absorption
energy were eliminated when averaging the data. Figure 2 shows the temperature
dependence of absorption energy. The lines are obtained by fitting the experimental
results to Equation (1). The value of n changes from 15 to 30 depending on plastic
deformation; it increases with rolling reduction initially, takes the maximum for 20%
rolling reduction and decreases above 40% reduction. The dislocation density is not
enough before cold rolling, whereas obstacles are present. The hammers energy would
be consumed by the two processes; the nucleation of dislocations and climbing over
obstacles. The dislocation density becomes enough by 5% rolling reduction and the
process becomes only climbing of obstacles. The value of n increases from 15 to 30.
The dislocations play the roll of obstacles above 20% rolling reduction. The pinning
processes of dislocations increase and the value of n decreases from 30 to 15.
 The value of DBTT was obtained experimentally as the temperature at which the
upper shelf energy becomes a half. Figure 3(a) shows the relation between DBTT and
rolling reduction, which is compared with the calculated value of T and the upper shelf
250

0%
5%
10%
20%
40%

Absorption energy (J)

200

150

100

50

0
180 200 220 240 260 280 300 320 340 360 380

    

Temperature(K)

Figure 2. Temperature dependence of absorption energy in cold rolled S15C steel. The carved
lines are calculated results n = 15, 31, 61, 25 and 11 correspond to H = 0, 5%, 10%, 20%, and
40% in strain, respectively.

(a)

180

270

160

260

140

250
120
240
100
230

600

10

20

30

40

Rolling reduction (%)

50

U0

DBTT, T (K)

280

200

Coercive field Hc (A/m)

290

(b)
500

400

300

200
240

250

260

270

280

290

DBTT (K)

Figure 3. (a) DBTT and theoretical values of T and Uo as a function of rolling reduction in
S15C steel. The solid and open circles, and triangles denote DBTT, T and Uo, respectively. (b)
The relation between DBTT and coercive field of S15C steel.

51

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

Table 2.

Chemical compositions of A533B steel.

A533B

Si

high Cu

0.19

0.19

1.47

0.16

0.64

0.51

0.14

balance

low Cu

0.19

0.18

1.46

0.05

0.64

0.50

0.14

balance

Table 3.

Mn

Cu

Ni

Mo

Cr

Fe

n, T and U0 of A533B steels before and after neutron irradiation.

T .

A533B

Uo (J)
223 75  

high Cu (before)

72

232 5

    (after)

42

312 30

low Cu (before)

11 4

215 6

 174 27

    (after)

12 4

238 4

 153 11

181 50

energy Uo. The relation between DBTT and coercive field obtained in our previous
work[6] is also given in Figure 3(b).
  DBTT has been measured before and after the neutron irradiation in NRPV
A533B steels with low and high copper contents. Their chemical contents are shown in
Table 2. The neutron radiation was performed at 563 K in helium atmosphere in a 50
MW nuclear reactor of Japan Materials Testing Reactor(JMTR). The radiation effect
that yields the brittleness has been examined after the neutron fluence to 5 1019 cm-2.
The number of test pieces is limited to two for each temperature. Figure 4(a) shows the
temperature dependence of absorption energy in high copper A533B steel. The lines are
obtained by fitting the experimental results to Equation (6). The values of n , Tand Uo
change by the neutron radiation, from 7 2 to 4 2, 232 5 K to 312 30 K and
22375 J to 181 50 J, respectively as listed in Table 3. Figure 4(b) shows the
temperature dependence of absorption energy in low copper A533B steel. The values of

250

250

(a)

high Cu

unirradiated

Absorption energy (J)

Absorption energy (J)

300

200
150
100

irradiated

50
0
150

200

250

300

350

Temperature (K)

400

450

(b)

low Cu

200

unirradiated

150

irradiated
100

50

0
150

200

250

300

350

400

450

Temperature (K)

Figure 4. Temperature dependence of absorption energy in A533B steel with (a) high copper
and (b) low copper contents, before and after neutron irradiation to a fluence of 5 1019 cm-2.
The solid lines through the data shows the least squares fits.

52

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

n ,Tand Uo change by the neutron radiation from11 4 to 12 4, 215 6 K to 238 4


K and 174 27 J to 153 11 J, respectively as listed in Table 3. The large deviation of
n , Tand Uo is attributed to the lack of data points and the shortage of test pieces.
Neutron radiation would not exert any remarkable influence on the distribution of
dislocations as well as the dislocation density but on the interaction energy Uob. The
change of properties due to neutron radiation is smaller in low copper A533B steel than
in higher copper one.

5. Discussion
In cold rolled S15C steel, the transition temperatures DBTT and T show similar
change against rolling reduction though the value of DBTT is slightly larger than that
of T as shown in Figure 3. The obstacles to the dislocation motion are dislocations
themselves and their interaction would be proportional to the applied stress. The value
of T shows a gentle increase from 5% to 40% in rolling reduction, whereas it increases
rapidly from 0 to 5% in rolling reduction. The upper shelf energy U0 decreases
monotonically from 0 to 40% reduction. The monotonic decrease of U0 indicates that
the mean free path would decrease and the density of mobile dislocations would not
change remarkably.
The dislocation density of A533B steels is higher than 1010 cm-2 before neutron
radiation and dislocations make cell structure. The dislocations that do not contribute to
the cell structure can move the same distance as the size of a cell, because there exist
little dislocations inside of the cell. Therefore, the mean free path would be large,
whereas the value of Um is small. Recently, it was suggested by the magnetic method
that the copper precipitates and dislocation loops created by neutron radiation would
gather around dislocations and disturb the dislocation movement.[5] Since copper
precipitates have stress field and edge dislocations include both compressive and
repulsive stress field, copper precipitates gathering around the dislocations would
compensate the stress field of dislocations in order to reduce the elastic energy. The
copper precipitates strongly disturb the dislocation movement and makes high copper
A533B steel brittle. On the other hand, copper precipitates inside of cells also disturb
the dislocation movement and make the mean free path x decrease. Since the value of
Um would not change by the neutron radiation, the value of Uo decreases as is listed in
Table 3. Copper precipitates make the value of Uob increase and their inhomogeneous
distribution makes the distribution of Uob wide. This results in a decrease of n due to
neutron radiation in the high copper A533B steel.
The amount of copper precipitates nucleated by neutron radiation is small in A533B
steel with low copper content in comparison with that for high copper content. The
change of T (DBTT) is 18 5 K in low copper A553B steel that is much smaller than
80 30 K in high copper A533B steel. The difference of the change in T is attributed to
the number of obstacles, which decide the value of Uob. The value of n does not change
remarkably by the radiation in low copper A533B steel. This result indicates that the
obstacles distribute homogeneously. The change of Uo due to neutron radiation in low
copper A533B steel is also smaller than that of the high copper A533B one, being
consistent with the results of T. The main obstacles to dislocation motion in A533B
steels are copper precipitates and their amount depends on the copper contents.
DBTT gives us the direct information about ductility and brittleness from a
macroscopic viewpoint and can be qualitatively explained by the dislocation theory. On

S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing

53

the other hand, dislocations also interact with magnetic domain walls and influence
their movement. According to earlier theory for micromagnetism, arrangement of
magnetization is determined so as to minimize magnetic Gibbs free energy consisting
of exchange energy, magnetocrystalline anisotropy energy, magnetostatic energy and
magnetoelastic energy.[7] In ferromagnetic materials including dislocations, the Gibbs
free energy is lowered when domain walls are located at dislocations and dislocations
act as obstacles to the domain wall motion, yielding changes in magnetic properties.
Therefore, both magnetic and mechanical properties have an intimate connection with
each other through dislocations as is seen in the simple relation between DBTT and
coercive field in Figure 3(b). Such connection is also true for NRPV steel irradiated by
neutron where various kinds of lattice defects such as precipitates which disturb
dislocation motion are formed; these irradiation defects interact with domain walls as
in the case of dislocations.[7-9] However, the crucial difference between these
properties is the fact that magnetic properties can be obtained by nondestructive
measurements whereas Charpy impact test is destructive one. The practical application
of magnetic methods to NDE is therefore expected for the pressure vessel of nuclear
reactors exposed to neutron radiation.

Acknowledgements
The authors express thanks to Dr. Y. Kamada, for the Charpy impact test
measurement, Dr. H. Kikuchi, K. Ara and N. Ebine for the operation of the nuclear
reactor. This research was supported by a Grant-in-Aid for Scientific Research (S),
Grant No. 14102034, from the Ministry of Education, Culture, Sports, Science and
Technology of Japan.

References
[1] E. R. Parker, in Brittle Behavior of Engineering Structure, (John Wiley & Sons, 1957).
[2] D. Francois, in From Charpy to Present Impact Testing, (ESIS Publication, 2002)
[3] J. Koutsk and J. Kock, in Radiation Damage of Structural Materials, (Elsevier Science
Publishers, 1994).
[4] G. R. Odette and G. E. Lucas, JOM 53 (2001) 18.
[5] S. Takahashi, H. Kikuchi, K. Ara, N. Ebine, Y. Kamada, S. Kobayashi and M. Suzuki, J. Appl.
Phys. 100 (2006) 023902.
[6] S. Takahashi, S. Kobayashi, Y. Kamada, H. Kikuchi,, J. Appl. Phys. 100 (2006) 113908.
[7] H. Kronmller and M. Fhnle, in Micromagnetism and the Microstructure of Ferromagnetic
Solids (Cambridge, 2003).
[8] L. J. Dijkstra and C. Wert, Phys. Rev. 79 (1950) 979.
[9] J. B. Goodenough, Phys Rev. 95 (1954) 917.

54

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-54

ND-materials characterization of neutroninduced embrittlement in German nuclear


reactor pressure vessel material by
micromagnetic NDT techniques
Gerd DOBMANN1, Iris ALTPETER1, Melanie KOPP1, Magdalena RABUNG1, and
Gerhard HBSCHEN1
1
Fraunhofer IZFP, Germany

Abstract. Depending on the neutron fluence and the special design of the pressure
vessel of nuclear power plants (NPP) the microstructure of the steels change by
neutron induced embrittlement. Embrittlement is on the basis of vacancies and Curich precipitates which in the size range of 1-3 nm contribute with coherent
residual stresses of the 3rd kind to an increase in hardness and strength (yield
strength and tensile strength) as well as with a reduction of the upper shelf value of
Charpy energy and a shift in the brittle-to-ductile transition temperature to higher
temperatures. Micromagnetic investigations sponsored by the German minister of
economics were performed at full Charpy specimen and material of the last
generation of German NPP in order to characterize the material degradation. The
contribution reports to the results obtained by the application of the
Micromagnetic-, Multiparameter-, Microstructure-, and stress-Analysis (3MA) and
the magnetostrictive excitation of ultrasound using an EMAT. Both technologies
document potential to be further developed to an in-service inspection technique.

Keywords. Neutron irradiation, embrittlement, pressure vessel material,


micromagnetic NDE techniques, multiple regression, calibration

Introduction
In the year 2003 IZFP has participated in the EURATOM project GRETE where the
characterization of neutron degradation of pressure vessel material was one project task.
Materials came from surveillance and irradiation programs. All specimens were
investigated in the hot cells of the research reactor in Petten, the Netherlands. The
specimens were half Charpy specimens obtained after the performance of the Charpy
impact energy test, i.e. the specimen have had plastic deformation and residual stresses

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

55

as disturbing influences. Concerning micromagnetic NDT techniques proposed by


IZFP the effects strongly influence the results. However, performing the so-called
3MA-approach (Micromagnetic, Multiparameter, Microstructure and stress Analysis)
[1] after calibration correlation coefficients in the 0.98 range were obtained with a
residual standard deviation of 16C when, for instance, the shift compared with the
non-irradiated material state - of the ductile to brittle transition temperature 'T09 was
predicted. When ISO-v-notched specimen, for instance according to the ASTM
standard A370 are tested in the Charpy impact test in order to evaluate the Charpy
impact energy as function of temperature, each half of the broken specimen has shear
lips in the fracture plane, i.e. the broken specimen shows a broadening compared with
the initial state. The broadening is defined as the difference value between the
specimen width with shear lips after break and the initial width value. T09 is derived
from the fitted and averaged Charpy impact energy versus temperature curve as the
special temperature where the broadening meets the value 0.9mm. However, a
verification test of the approach by independently selected specimen resulted in much
larger deviations because of the above mentioned disturbing effects [2, 3].

1. Material Selection

Figure 1. Neutron fluence in a German NPP of the last generation as function of lifetime

As directed by the German minister of economics responsible for nuclear safety a


project was performed where the objective in a feasibility study was to demonstrate the
potential of micromagnetic NDT techniques to characterize the material degradation by
neutron irradiation when the neutron fluence is much smaller compared to the former
discussed European project.
Background is that in Germany (see Figure 1), according to the German codes, the
design lifetime is 32 years and the fluence is restricted to values of 51018 n/cm2
(Energy >1MeV). This design end-of-life value (green curve in Figure 1.) is one order

56

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

in size smaller than in other countries and is obtained by a much larger gap between
PV-wall and core internals in the German design compared with others. When the real
behavior of the material is discussed resulting from fluence measurements then the
fluence curve follows (in an extrapolation) the red curve. After 48 years of full power
service only a value of 3.351018 n/cm2 is obtained documenting also the potential of a
possible lifetime extension.
The materials selected to perform the non-destructive tests were from the two steel
types 22NiMoCr37 and 20MnMoNi55. The important contents of the elements Cu, P
and Ni are indicated in Table 1 where also the available fluence values are documented.
As can be seen, the specimens named P16 with the highest Cu, P, and Ni-content have
the highest fluence values, i. e. they have the highest degree in degradation.
Table 1 Steel grades selected for materials characterization
Indication

P140

Weld
Material
/Base
Material

Cu [%]

WM

0,07

P [%]

Ni [%]

Fluence
[n/cm]

0,009

0,9

3,72E+18
7,55E+18
1,04E+19
3,71E+19

P141

BM

0,06

0,008

0,8

3,78E+18
7,66E+18
1,05E+19

P16

WM

0,08

0,012

1,7

4,15E+18
8,04E+18
1,16E+19
5,22E+19

According to Figure 2 the materials P 16 and P 141 show normal behavior, i.e. an
increase of the of the shift (dt41) of the brittle-to-ductile transition temperature with
fluence, whereas the P 140 material documents a recovery annealing, first increasing
then decreasing. The arrows in Figure 2 indicate the tendency. In this case the
transition temperatureT 41 in contrast to T09 is discussed which is derived from the
fitted and averaged Charpy impact energy curve as function of temperature exactly
where the curve meets the value of 41J.
P16 weld metal
P 16 WM

P 141 BM

P140 weld metal

P141- base metal


P 140 WM

Neutron fluence times 10E18


Figure 2. Shift in the brittle-to-ductile transition temperature (dt41) as function of the fluence (abscissa) of
the three materials

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

57

2. Measurements with 3MA Techniques and Correlation with Toughness


Parameters
The 3MA measurements were performed in the hot cell of AREVA whereas only the
magnetic yoke transducer was inside the hot cell and the electronic equipment outside.
The Charpy specimens were handled by a manipulator. A special adaption piece made
by an Al-alloy was applied to guarantee reproducibility in positioning of the specimen
at the yoke pole shoes.
The 3MA-approach is described in detail elsewhere [1].
However, a short explanation is given here (Figure 3). With 3MA different
micromagnetic quantities are measured. These are derived by analysis of the magnetic
Barkhausen noise M(Ht), the incremental permeability (Ht) as function of a tangential
magnetic field Ht which varies as function of time (t) according to a sinusoidal time
function. In Figure 3, in a sketch, the different measuring quantities are illustrated: In
the middle a magnetic hysteresis (induction B(H) versus the magnetic field H), in the
upper left part the Barkhausen noise M which is an electrical voltage received by
integrating the magnetic flux excited by Bloch wall jumps and rotational processes of
the magnetization vectors of the domains. In the upper right part the incremental
permeability which is the inclination of the small inner loop (see hysteresis) 'B/'H
and proportional to the impedance of an eddy current coil superimposing to the
hysteresis loop the small incremental magnetic field 'H of a frequency, at minimum, a
factor 10 higher than the hysteresis frequency. H, respectively Ht are measured in A/cm,
the time t in s and the induction B in T. In the lower left part of Figure 3 the harmonic
analysis of the tangential field strength is shown and how, for instance, a distortion
factor K can be derived by measuring the odd harmonics up to order of seven. The
lower right part of Figure 3 documents an additionally performed eddy current
impedance measurement with three different frequencies. The 3MA-approach
correlates a target value like, for instance, the dt41 transition temperature shift in a
multiple regression model with the micromagnetic measuring quantities and estimates
the unknown parameters of the model by applying a least squares algorithm.

Figure 3. The 3MA approach

58

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

In the least squares algorithm only one part of the set of specimens is used for
calibration of the model, the other independently selected part is applied to check the
quality of the model (verification test).

Figure 4. 3MA result at the P16 specimens

In Figure 4 as one result the shift in the brittle-to-ductile transition temperature


(dT41) is compared with the predicted values by 3MA. The red colored dots indicate
the calibration specimens, the green colored the verification specimens. The regression
coefficient from the calibration is 0.93 with a residual standard error of 12.8C.
However, because of the inhomogeneous microstructure in the weld material P16
(multiple-path welding under submerged arc) the residual standard error concerning the
verification specimens 24.8C is nearly a factor 2 larger. In the base material the
obtained standard errors are much smaller. This is documented in Figure 5 in case of
the material P 141. The regression coefficient after the calibration is 0.986; the residual
standard error in the verification is only 1.2C.

Determination of the
shift of the ductile to
brittle transition
temperature dT41 by
the 3MA-technique
Base material P141

r2 = 0.986
Standarderror = 0.35 (Calibration)
Standarderror = 1.2 (Verification)

Figure 5. 3MA result at the P141 material

The future plan is to develop an ISI inspection technology to examine the pressure
vessel from the inner side by a 3MA approach. However, in that case the inspection has
to be performed through the austenitic stainless steel cladding which itself has certain
G-ferrite content, a ferromagnetic phase. Therefore the influence of the cladding on a
micromagnetic technique like Barkhausen noise was investigated. Figure 6 documents
the result. Shown are Barkhausen noise profile curves [1] which are obtained when the

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

59

received electrical voltage (ordinate) as time signal in a (x,y)-presentation on the


oscilloscope is visualized versus the time signal of the magnetic field (abscissa).
Even through the 10 mm thick austenitic plate the Barkhausen noise from the
ferritic plate can be recorded. The thickness of the austenitic plate influences the
inspection mainly by a larger lift-off effect. However, there is also a higher eddy
current damping because of the electrical conductivity of the cladding.

Figure 6. Barkhausen-noise M as function of a tangential magnetic field Ht at a ferritic plate (left), through 8
mm austenitic (middle) and 10 mm austenitic (right) plate with G-ferrite simulating the cladding

3. Magnetostrictive Electromagnetic Excitation of Ultrasound (EMUS)


G. Ahlers had previously proposed to excite a standing, low frequency (10 kHz, O=320
mm) ultrasonic wave by a magnetostrictively working electromagnetic acoustic
transducer (EMAT) [4] in the interface between the austenitic cladding and the ferritic
material of the PV wall propagating in thickness direction. In Figure 7 the principle of
such an inspection is shown which up to now is not yet realized for ISI-technology. A
dc- or low frequency ac-electromagnet magnetizes locally the pressure vessel wall from
the inner side through the cladding. A high-frequency (HF)-current tone burst is
exciting a HF-eddy-current coil; the magnetic field is controlled by a Hall probe. The
principle was investigated using cladded test pieces. Figure 8 gives a view on the
experimental set-up using a Bruker water-cooled laboratory electro-resistance dcmagnet with a maximum magnetic induction field of 1.5 T.
transmitter coil

receiver coil

poleshoes of electromagnet

Figure 7. Principle of the magnetostrictive excitation


of a standing wave in thickness direction

Figure 8. EMAT transmitter-receiver prototype


transducer at a cladded test piece

60

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

The HF-coil was designed as transmitter receiver coil with adopted coil windings;
the burst frequency was 50 kHz. The cladded test piece was set-up at the pole shoes of
the laboratory magnet. It was shown that G-ferrite changes in the cladding doesnt
change the results very much. This is documented in the Figures 9 and 10. In the case
of Figure 9 the standing wave was excited in a ferritic plate with wall thickness 30 mm
but with a lift-off of the transducer of 8 mm whereas in Figure 10 the test piece was a
30 mm thick cladded material with a cladding thickness of 8mm. Insonification was
performed from the cladded surface. The electrical signals were obtained by an
excitation with a burst length of 10 cycles of a 50 kHz tone-burst at a magnetic field
strength of 260 A/cm. It is obvious; the cladding is mainly influencing the signal with a
lift-off effect.
There were also investigations concerning the influence of the inhomogeneous
microstructure of a ferritic butt weld beneath the cladding on the magnetostrictive
excitation. This situation occurs at pressure vessels along an inspection path in the
direction of the circumferential welds. By scanning with the EMAT along such a weld
and measuring the resonance amplitude as function of the different positions only a
standard deviation of < 4%, compared with the average value, has been observed. This
reflects the change of magnetostriction with the microstructure of a multilayer
submerged arc weld. This value has to be compared with the measuring effect obtained
at neutron irradiated material. Because the dynamic magnetostriction is sensitive to
lattice defects it was assumed that the resonance amplitude of the standing wave also
reflects the neutron embrittlement and first experiments were also performed with a
special designed magnetostrictive transducer at Charpy specimen in the hot cell of
AREVA in order to principally document the potential.

Figure 9. Magnetostrictive excitation of the


standing wave in a ferritic plate, lift off of the
transducer 8 mm in air (abscissa is the time scale,
the ordinate is the electrical voltage induced in the
EMAT by the ultrasonic wave)

Figure 10. Magnetostrictive excitation of the


standing wave excited beneath the cladded surface,
cladding thickness 8 mm (abscissa and ordinate like
in Figure 9)

The EMAT especially optimized to test the small geometry of Charpy specimens
operates at 1.2 MHz. The echo sequence of the excited standing wave was recorded by
varying the superimposed magnetic field. The time signal was time gated (Figure 11)
and in the gate the peak amplitude was registered. The so obtained measuring quantity
was named E60, indicating that the measurement was performed at the 60% amplitude
level of the maximum magnetic field strength. This is an operating point of the
magnetic field at which the magnetostrictive excitation has not yet obtained its
maximum. It is obvious, at the higher operating frequency of 1.2 MHz the efficiency of
transduction and receiving is better than in the lower frequency range (10 kHz) used at

G. Dobmann et al. / ND-Materials Characterization of Neutron-Induced Embrittlement

61

the plates. In Figure 12 by testing the P16 material the measuring quantity E60 shows a
linear decreasing with the dT41 values. This behavior is expected with the increase of
lattice defects. The scatter in the data documents the natural scatter in the
microstructure of similar irradiated material but different specimens and can also be
observed in the Charpy test data. Compared with the amplitude dynamic as function of
the brittle-to-ductile transition temperature this scatter is smaller than 17%. The
influence of an inhomogeneous weld microstructure which was measured with < 4%
amplitude variation is much smaller.

Time gate
Figure 11. Echo sequence recorded at a Charpy
specimen (abscissa time scale, ordinate the
electrical voltage which is induced in the EMAT by
the ultrasonic wave)

Figure 12. Measuring quantity E60 as function of


the shift of the brittle-to-ductile transition
temperature dT41 at P16 material

Conclusion
Micromagnetic NDT techniques show a high potential when neutron degradation is
characterized. The combination and data fusion of micromagnetic measuring quantities
in a 3MA-approach is suitable to early detect material degradation. The evaluation of a
magnetostrictively transmitted and received ultrasonic wave propagating in the
pressure vessel wall thickness direction as a standing wave has special potential to
enhance ISI. The two techniques are under development.

References
[1]
[2]
[3]
[4]

I. Altpeter, et al., Electromagnetic and Micro-Magnetic Non-Destructive Characterization (NDC) for


Material Mechanical Property Determination and Prediction in Steel Industry and in Lifetime Extension
Strategies of NPP Steel Components, Inverse Problems 18 (2002) 1907-1921.
G. Dobmann, et al., Electromagnetic Characterization of Materials Degradation due to Neutron
Irradiation and Fatigue, Applied Electromagnetics and Mechanics (2003), 30-31.
G. Dobmann, et al., Aging Material Evaluation and Studies by Non-Destructive Techniques (AMESNDT) - a European Network Project, Nuclear Engineering and Design 26 (2001) 373-374.
I. Altpeter, et al., Review of Progress in Quantitative Nondestructive Evaluation. 22A, Melville, New
York, American Institute of Physics (AIP) (2003) 15-21.

62

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-62

Evaluation of Chill Contents in Flake


Graphite Cast Irons Using AC
Magnetization Method
Tetsuya UCHIMOTO a,1, Jun MATSUKAWAb, Toshihiko ABEa,
Toshiyuki TAKAGIa Takeshi SATOa, Hiroyuki IKEb, Takahito
TAKAGAWAc and Noritaka HORIKAWAd
a
Institute of Fluid Science, Tohoku University, Japan
b
Graduate School of Engineering, Tohoku University, Japan
c
Iwate Industrial Research Institute, Japan
d
Graduate School of Engineering, Hokkaido University, Japan
Abstract. In this study feasibility of evaluation for chill contents in flake graphite
cast iron is investigated based on AC magnetization method, which is one of the
electromagnetic nondestructive evaluation methods. Magnetic properties of flake
graphite cast iron samples with different chill contents were measured by a B-H
loop analyzer in order to discuss their relation to chill contents. It was found that
some magnetic parameters of the flake graphite cast iron depend on the contents of
matrices and graphite. Especially, the hardness, which reflects chill contents, has
good correlation with the area of hysteresis loop at relatively high frequencies.
Focusing on the finding, AC magnetization method was applied to evaluation for
the chill contents. Through the experiment, it was found that there is a correlation
between hardness of flake graphite cast irons with chill and signals. In
consequence, AC magnetization method has a capability of evaluating chill
contents in flake graphite cast irons.
Keywords. cast iron, AC magnetization method, nondestructive evaluation, chill

1. Introduction
In foundry industry, thin-walled mechanical components of cast irons provide weight
saving of automobiles, resulting in realization of energy saving vehicle. However, thinwalled components of cast iron leads to the increase in cooling rate during
solidification of the cast irons, so that chill, namely proeutectic cementite
microstructure, is crystallized. Chill deteriorates the mechanical properties because cast
irons including chill are hard and brittle [1]. In contrast, it was reported that chill
improves wear resistance, and it can be applied to sliding parts of automobile
components such as cylinder bores and piston rings to be endured in severe wear
conditions [2]. Therefore, it is highly required in foundry industry to evaluate and
1
Corresponding Author: Institute of Fluid Science, Tohoku University, 2-1-1 Katahira, Aoba, Sendai,
Miyagi 980-8577 Japan; E-mail: uchimoto@ifs.tohoku.ac.jp

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

63

control chill contents.


At present, visual test of fracture surface, microscopic observation and indentation
test are applied to evaluate chill contents. However, they are all destructive testing and
total inspection is impossible by these methods. For the reason, there is a requirement
for establishment of the method to evaluate chill contents nondestructively and
quantitatively.
Cast iron is composed of graphite and matrices consisting of ferrite, pearlite, chill
and so on. Therefore, it is indispensable to extract the information concerned with chill
contents from other information including that of graphite and other matrices contents
when nondestructive evaluation is applied to estimation of chill contents. Focusing on
the difference of electromagnetic properties of each matrix and graphite, some attempts
have been done to estimate hardness, other mechanical properties, ratio of ferrite to
pearlite in ductile cast iron and graphite size in flake graphite cast iron by the method
based on electromagnetic phenomena [3-6]. Feasibility of evaluation for chill contents
in ductile cast iron was reported [7].
This work proposes an evaluation method of chill contents by means of the
electromagnetic nondestructive method. In this paper, firstly, the contents of matrices
and graphite in the flake graphite cast irons with different contents of chill are quantified from the analysis of microstructure. Secondly, the relations between
electromagnetic characteristics and the contents of each matrix and graphite are
discussed. Finally, focusing on the relation, AC magnetization method is applied to
evaluation for chill contents.

2. Cast Metals
The flake graphite cast irons which have chemical composition listed in Table 1 were
prepared. Materials without chill were divided into three groups: FC150, FC200 and
FC250 (JIS standard). In addition, each group has three types of materials depending
on three kinds of heat treatments: as cast, furnace cooling and air cooling. In the case of
furnace cooling and air cooling, the cast material was reheated at 850 oC in the furnace
for one hour. Then, the material was cooled in the furnace or in air, respectively.
Therefore, there were 9 materials in the group of the flake graphite cast irons without
chill. There were other sets of flake graphite cast iron materials which include chill: I-1
and I-2. They included different contents of chill. The molten metal of cast iron in the
group of I-1 and I-2 was same as FC250 and FC150, respectively. In order to obtain
chill, following casting process were carried out. Source material, consisting of pig iron,
ferrosilicon, electrolytic manganese and electrolytic iron, was melted at 1500 oC. The
chemical composition of materials was controlled by retention time of molten metals at
1500 oC. As retention time of the molten metal at 1500 oC increase, CE value, which is
defined by
1
CE mass %C  ( mass % Si  mass % P )
(1)
3
was decreased, and chill was obtained more easily. Chill contents were increased from
I-1-1 material to I-1-3 one. It is same on the group of I-2 materials. There was no
inoculant into molten metals.
Microstructures of the flake graphite cast iron materials were observed by optical
microscope to evaluate the size, the shape and the contents of graphite and matrices in

64

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

them. Pearlite was etched by picric acid ethanol. Some representative photographs are
shown in Figure 1. The photos of all materials were analyzed and the areas of each
matrix and graphite were quantified. Graphite area was calculated from the photos of
the microstructure before etching, and that of matrices was quantified from the picture
of microstructure after etching. The contents of graphite and matrices in the materials
are summarized in Table 2.

Table 1 Chemical composition of materials.


C%

Si%

Mn%

P%

S%

CE%

FC150

3.77

2.78

0.78

0.025

0.015

4.71

FC200

3.36

2.15

0.69

0.018

0.01

4.08

FC250

3.13

1.66

0.72

0.017

0.002

3.69

I-1-1

3.40

1.81

0.66

0.016

0.008

4.01

I-1-2

3.23

1.84

0.66

0.016

0.009

3.85

I-1-3

3.01

1.84

0.66

0.016

0.009

3.63

I-2-1

3.99

2.59

0.77

0.025

0.011

4.86

I-2-2

3.78

2.62

0.77

0.025

0.010

4.66

I-2-3

3.48

2.63

0.76

0.024

0010

4.36

Table 2 Contents of graphite and matrices in each material.


Graphite%

Ferrite%

Pearlite%

Chill%

FC150 as cast

11.8

11.3

72.4

0.00

FC150 furnace

13.6

69.3

17.1

0.00

FC150 air

11.1

8.85

71.0

0.00

FC200 as cast

8.25

6.60

85.2

0.00

FC200 furnace

6.45

78.5

15.1

0.00

FC200 air

6.45

4.45

89.1

0.00

FC250 as cast

4.33

3.50

92.2

0.00

FC250 furnace

4.65

40.5

54.9

0.00

FC250 air

4.50

5.20

90.3

0.00

I-1-1

0.45

0.00

68.2

22.9

I-1-2

0.17

0.00

71.5

28.3

I-1-3

0.25

0.00

76.8

31.4

I-2-1

16.4

0.00

75.5

8.1

I-2-2

6.54

0.00

78.5

15.0

I-2-3

1.78

0.00

77.0

21.2

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

(1)G FC150 as cast sample (before etching)

65

(2)G FC150 as cast sample (after etching)

(3)G I-1-1 sample (before etching)


(4)G I-1-1 sample (after etching)
Figure. 1 Microstructure in flake graphite cast iron.

3. Magnetic Properties
Magnetization curve was measured by the B-H analyzer. The B-H analyzer is
composed of pickup coils and an exciting coil. Cylindrical sample is inserted in pickup
coil. AC current flowing in an exciting coil generates uniform magnetic field around a
sample. Magnetization process in the sample is obtained by a signal of pickup coil.
Some parameters such as loop area, remanence and coercivity are calculated from the
hysteresis curves obtained by the B-H loop analyzer. The cylindrical samples were
processed from the materials. Length and diameter of samples were 30 mm and 3 mm,
respectively. Hysteresis curve was measured at the frequencies of 10 Hz and 100 kHz.
In the case of 10 Hz, we consider that the measurements are quasi-static, and
demagnetizing field correction was made to measured loops. In the case of 100 kHz, no
correction was made, which means that the loops should be compared relatively. Figure
2 shows hysteresis curves of some samples at the frequency of 100 Hz. Shape of
Magnetization loop depend on the contents of matrices and graphite. Especially,
magnetic flux density of the sample without chill at relatively high magnetic field
increases with decreasing graphite contents. However, magnetic flux density of I-1-1
sample is relatively small in spite of its low graphite contents. It was found that the
sample with chill possesses specific difference from other sample without chill in

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

/ CIPGVKEHNWZFGPUKV[
)







(% CUAECUV
(% CUAECUV
(% CUAECUV
+
+





   







 

/ CIPGVKE(KGNF
1 G

Figure 2. B-H characteristics at the frequency of 10Hz.



.QQRCTGC
,O 





(% CUECUV
(% HWTPCEG
(% CKT
(% CUECUV
(% HWTPCEG
(% CKT
(% CUECUV
(% HWTPCEG






(% CKT
+
+
+
+
+
+

        

* CTFPGUU* 8

Figure 3. Relation between loop area and hardness at the frequency of 10Hz.



/ CIPGVKEHNWZFGPUKV[
)

66




(% CUAECUV
(% CUAECUV
(% CUAECUV
+
+













/ CIPGVKEHKGNF
1 G



Figure 4. B-H characteristics at the frequency of 100 kHz.

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

67


(% CUAECUV
(% CKT
(% HWTPCEG
+
(% CKT
+
(% CUAECUV
+
(% HWTPCEG
+
(% CKT
+
(% CUAECUV
+
(% HWTPCEG




.QQRCTGC,O 











* CTFPGUU* 8 





Figure 5 Relation between loop area and hardness at the frequency of 100kHz.

magnetic characteristics. Hardness of a cast iron mainly depends on both chill and
graphite contents. Focusing on the fact, the loop area calculated from hysteresis curve
was evaluated as function of hardness of the sample. There is little correlation between
the hardness and loop area of each specimen. Figure 4 presents hysteresis curves of the
same sample shown in Figure 2 at the frequency of 100 kHz. Shape of magnetization
loop is also related to the contents of matrices and graphite. Loop area of hysteresis
curve of each sample was also evaluated as function of hardness in Figure 5. In contrast
with the relation at 10 Hz, they decreased linearly as hardness increases. The set of
samples plotted in Figure 5 includes flake graphite cast iron samples without chill
structure, FC150, FC200 and FC250 as well as ones with chill structure. In addition,
each series of FC150, FC200 and FC250, which has different graphite shapes, includes
samples with different matrices owing to heat treatments. Therefore, hardness of the
samples reflects chill contents, graphite shapes and matrices, which implies that loop
area at relatively high frequencies depends on chill contents, graphite shapes and
matrices. Since there is no correlation between loop area and hardness at low frequency,
linear correlation between them at relatively high frequencies is due to effects of eddy
currents which depends on permeability and conductivity. Permeability and
conductivity of all samples were measured the by BH analyzer and four-terminal
method, respectively. Both of them do not have any correlation with hardness. Its
mechanism should be complicated since eddy currents flows in heterogeneous media
consisting of different types of matrices and graphite structures, which will be
discussed in future.

4. AC Magnetization Method
Based on the finding acquired from the evaluation of magnetic properties by means of
B-H analyzer, AC magnetization method was applied to evaluation for chill contents,
and its feasibility was discussed. Schematic drawing of experimental setup is shown in
Figure 6. The probe consists of two coaxial pancake coils with a ferrite core; upper coil
is an exciting coil, and lower one is a pickup coil. A ferrite core is put to obtain

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

Function generator

FFT analyzer
PC

exciting
ferrite
exciter
detector
sample
pickup
Figure 6 Experimental setup for AC magnetization method.

2 KEMWR
8



(% CUECUV





    
'ZEKVG
8





Figure 7. Hysteresis equivalent curve at the frequency of 3kHz.

13.0

FC150 as_cast
FC150 furnace
FC150 air
FC200 as_cast
FC200 furnace
FC200 air
FC250 as_cast
FC250 furnace

12.5
2

Loop area (V )

68

FC250 air
I-1-1
I-1-2
I-1-3
I-2-1
I-2-2
I-2-3

12.0

11.5

11.0

100

200

300

400

500

Hardness HV

Figure 8. Relation between loop area of lissajous and hardness.

T. Uchimoto et al. / Evaluation of Chill Contents in Flake Graphite Cast Irons

69

stronger magnetic field. AC current flowing in exciting coil induces AC magnetic field
into a sample. Magnetization process of the sample is detected by the voltage of a
pickup coil. Changes of hysteresis curve were easily acquired by plotting lissajous
waveform composed of signals of an exciter and a detector (hysteresis curve
equivalent), and amplitude of 3rd harmonic wave acquired by a signal of pickup coil.
The samples were large enough not to neglect edge effect of them. Outer and inner
diameter of exciting coil and pickup one is 8.4 mm and 5.4 mm, height is 4.25 mm and
turn number is 150. Applied voltage was 5 V, and measurement was carried out at 3
kHz.
Figure 7 shows hysteresis curve equivalent measured from the sample of FC150 as
cast. We investigated relationship between the amplitude of 3rd harmonics and
hardness. Correlation between them was not confirmed. Figure 8 shows relation
between hardness and loop area calculated from the hysteresis curve equivalent
obtained from each sample. Tendency of decrease in loop area accompanied with
decrease in hardness indicates the feasibility of evaluation of hardness. It is expected
that correlation between hardness and loop area improves due to accurate evaluation of
loop area accomplished by optimization of the measurement condition.

5. Summary
In this paper, dependence of chill contents on magnetic properties was discussed
through investigation of magnetic properties of flake graphite cast iron with different
chill contents. Based on the insight obtained from the investigation, AC magnetization
method was applied to evaluation for chill contents. As the results, it was found that the
loop area of hysteresis curve equivalent has correlation with hardness which reflects
chill contents, which implies the feasibility of nondestructive evaluation for chill
contents by AC magnetization method.
Acknowledgements
This work is partially supported by New Energy and Industrial Technology
Development Organization, Japan, Industrial Technology Research Grant Program,
Characterization of microstructure of advanced cast iron for energy-saving automobile
based on multi-scale electro magnetic approach, 04A48512. The author appreciates
many supports by Tsutomu Watanabe and technical staffs in the Institute of the Fluid
Science in processing the samples prepared in this study.
References
[1]
[2]
[3]
[4]

Handbook of foundry engineering (MARUZEN). J.JFS, 2002, 227


T. Nagai, M. Uemura, J. Fujioka, H. Hattori. J.JFS. 2004, 76(6), 440-446
T. Abe, T. Uchimoto, T. Takagi, S. Tada. J.JFS. 2003, 75(10), 675-681
T. Uchimoto, T. Takagi, S. Konoplyuk, T. Abe, H. Huang and M. Kurosawa. Journal of Magnetism
and Magnetic Materials, 2003, 1(258-259), 493-496
[5] S. Konoplyuk, T. Abe, T. Uchimoto, T. Takagi, M. Kurosawa, NDT&E International, 2005, 38(2005),
623-626
[6] G. Vrtesy, T. Uchimoto, T. Takagi, I. Tom, O. Stu-pakov, I. Mszros, J. Pv, Physica B,
372(2006), 156-159
[7] M. Kurosawa, T. Uchimoto, T. Abe, T. Takagi, T. Sato, H. Kage, T. Noguchi, J.JFS. 2005, 77(12), 826832

70

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-70

Characterisation of Microstructures in
Heat Treated Maraging Steel using Eddy
Current and Barkhausen Emission
Techniques
K.V. Rajkumar, B.P.C. Rao, B. Sasi, S. Vaidyanathan,
T. Jayakumar and Baldev Raj
Nondestructive Evaluation Division, Indira Gandhi Centre for Atomic
Research, Kalpakkam 603102, India; Phone: 91 44 27480232;
Fax: 91 44 27480356; email: bpcrao@igcar.gov.in
Abstract. The effects of ageing induced microstructural changes in M250 Maraging
steel widely used in aerospace industries are characterized using electromagnetic nondestructive evaluation (NDE) methods. Eddy current and magnetic Barkhausen
emission parameters have been studied and the results are compared with hardness,
XRD and transmission electron microscopy (TEM).
Keywords. Maraging steel, NDE, Eddy current, Barkhausen emission, Microstructure

Introduction
Maraging steel is one of the most preferred structural materials for critical
application in aerospace industries due to its excellent mechanical properties [1]. The
aging behaviour of the maraging steels has been extensively studied [2-6]. M250
maraging steel components are subjected to solution annealing (SA) at 1093 K for 1 h
followed by aging at 755 K for 3-10 h. This heat treatment produces the best
combination of mechanical properties i.e. ultra high strength coupled with good
fracture toughness due to the precipitation of intermetallic phases in low carbon soft
martensitic matrix [5-6]. The early aging period comprises of recovery of martensitic
structure and hardening due to precipitation of hexagonal Ni3Ti intermetallic
precipitates. The intermediate aging duration is characterized by reversion of austenite
accompanied by precipitation of hexagonal Fe2Mo intermetallic phase. As these two
processes occurring during the intermediate aging drastically affect the hardening in
opposite manner, overall hardening levels off after reaching a maximum. Decrease in
hardening, observed during longer aging durations is attributed essentially to the
formation of reverted austenite rather than the precipitate coarsening. The amount of
reverted austenite is reported to increase with increase in aging temperature (upto ~ 900
K) and time [2]. Hence, this regime is of great technological importance. Quantitative
characterization of microstructures using particularly, electromagnetic (NDE) methods
is of much practical interest. These techniques exploit measurements of changes in

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

71

electrical and magnetic properties of materials. Among others eddy current (EC) and
magnetic Barkhausen emission (MBE) techniques are preferred, essentially because
they are non-contact in nature, sensitive, versatile and field employable.
The present study attempts to investigate the effects of aging induced
microstructural changes on the eddy current and Barkhausen emission measurements
for exploring the possibility of using these methods in shop-floor for verifying the
adequacy of heat treatment. Observations of the electromagnetic methods are correlated
with hardness, X-ray diffraction (XRD), selected area diffraction (SAD) and
transmission electron microscopy (TEM) data.

Figure1. Schematic of experimental setup used for eddy current testing and detailed
dimensional drawing of T/R probe.

1. Experimental
The chemical composition (wt %) of the maraging (M250) steel used in this study is as
follows: 17.89 Ni, 8.16 Co, 4.88 Mo, 0.43 Ti, 0.05 Mn, 0.05 Cr, 0.05 Si, 0.05 Cu, 0.096
Al, 0.003 C, balance Fe. A plate of M250 maraging steel was solution annealed at 1093
K for 1 h followed by air cooling. Specimens of dimensions 30x25x7 mm3 cut from the
solution annealed plates, were encapsulated in quartz tubes under vacuum and aged at
755 K for different durations of 0.25, 1, 3, 10, 30, 40, 70 and 100 h followed by water
quenching.
The eddy current measurements were carried out at 100 kHz using a transmitreceive coil (T/R) type eddy current probe (Figure 1). The EC measurements were
carried out after balancing the probe in air and phase angle of induced voltage was
adjusted such that the signal of reference stainless steel (SS) 304 specimen was along
the positive side of the X-axis [7, 8].
For MBE measurements, the samples were subjected to a continuously varying
cyclic magnetic field in an electromagnetic yoke with a period of 10 s. The current
from the sweep controller circuit was fed to a bipolar high current generator to generate
a symmetrical bipolar triangular field. The applied magnetic field H A was measured at
the centre of the yoke using a Hall probe (Walker Scientific) connected to a Gauss
meter (MG-50 Walker Scientific). The maximum field was set to 1500 Oe for complete
magnetic saturation of the specimen. This corresponds to magnetization field strength
(H) of 1,20,000 A m-1. Calibration of HA was made with respect to the current applied

72

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

to the yoke. The tangential magnetic field HT was measured near the sample surface.
Magnetic Barkhausen emission measurements were performed using an encircling pick
up coil (5000 turns). The MBE signal was amplified using a low noise pre-amplifier
and a post amplifier (80 dB). The magnetic flux density was measured using a 20 turn
coil closely wound on the sample connected to a flux meter (Walker Scientific MF5DP). The output voltage signals for all the magnetic parameters were suitably
conditioned for digitization using PC based data acquisition.
XRD measurements were carried out for estimation of volume fraction of reverted
austenite using MAC Science MXP18 X-ray diffractometer with Cr K radiation in the
complete angular range of 60-130. Vickers hardness measurements were carried out
on these specimens at 10 kg load. Averages of five hardness measurements have been
made for each specimen. The maximum scatter in the hardness measurements was
found to be 5 HV10.

2. Results and Discussion


Variation in hardness, volume fraction of reverted austenite and resistivity reported in
[3] with aging time are shown in Figure 2. The resistivity curve shows continuous
decrease up to 40 h and then leveling off at long aging duration (beyond 40 h). The
initial decrease is attributed to the recovery (that includes the removal of quenched in
point defect and annihilation of dislocations) in martensite matrix and due to
accumulation of solute atoms on dislocations in martensitic matrix. Further decrease in
the resistivity and increase in permeability at intermediate durations is attributed to the
intermetallics precipitation which is also accompanied by increase in the hardness. The
leveling off, observed at longer aging periods (beyond 40 h) is attributed to net
manifestation of two opposing mechanisms occurring simultaneously i.e. precipitation
and austenitic reversion (decreases hardness).

30

10

100

1000

Vol.% of austenite
Hardness
Resistivity

SA

20

Hardness, HV10

Vol. % of austenite

25

15
10
5

650

0.65

600

0.60

550
500
450

Resitivity (ohm-M)

0.25

35

0.55
0.50
0.45

400

SA

0.40

0
0.25

1
10
Aging time, h

100

350

Figure 2. Variation of hardness, electrical resistivity and volume percent of austenite


with aging time (resistivity data is taken from [3]).

Imaginary component of induced voltage, volts

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

73

2.5

2.0
Ferrite

SA
0.25h
1h
3hrs
10hrs
30hrs
40hrs
70hrs
100hrs
Reference

1.5
1.0

Carbon steel

0.5

0.0

SS 304

Air

Al 3003

-0.5
-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

2.5

Real component of induced voltage, Volts

Figure 3. Variation of EC induced voltage in receiver coil with aging time.


The complex plane diagram of induced voltage obtained for heat treated maraging
steel specimens is shown in Figure 3. The induced voltage is segregated into two
distinct clusters marked in square and oval and they are found to be in the second and
first quadrant, respectively. The location of these two clusters confirms that the
specimens are ferro-magnetic. Cluster in the second quadrant (marked as square)
corresponds to ferromagnetic induced voltage of specimens aged up to 10 h and the
cluster in the first quadrant (marked as oval) correspond to samples beyond 10 h of
aging. Formation of two clusters is essentially due to two distinct microstructural
evolution taking place during aging and the associated changes in the resistivity and the
permeability.
The magnitude and phase angle of the induced voltage are determined and are
shown in Figure 4a for various specimens. It can be seen that both magnitude and phase
angle are influenced by microstructure changes and their trends are nearly identical.
The EC induced voltage is found to increase initially from solution annealed (SA)
condition to 0.25 h of aging and then continued to decrease gradually upon aging
further up to 10 h. Aging between 10h and 30 h showed a drastic drop in magnitude
and phase angle of induced voltage and which continued to drop further gradually in
the aging regime of 30-70 h. Beyond 70 h, a drastic drop in induced voltage was
noticed.
The changes observed in EC parameters can be explained as follows: The initial
aging regime (SA-10h) is primarily characterized by dislocation annihilation and
precipitation of Ni3Ti intermetallics typically shown in Figure 5a by TEM studies.
Annihilation of dislocation is attributed to increase magnetic permeability and
reduction in resistivity. Increase in magnetic permeability (associated with dislocation
annihilation), as reported by Sablik et al. [9], tries to increase the induced voltage while
decrease in resistivity, associated with the precipitation of intermetallics and dislocation

74

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

annihilation, tries to decrease the induced voltage. The changes in induced voltage
observed for initial aging regime from SA-10 h hence, can be understood in terms of
net manifestation of two opposing mechanisms i.e. dislocation annihilation and
intermetallic precipitation.
The larger decrease in EC parameters for specimens aged to 30h is attributed to
initiation of non-magnetic reverted austenite in addition to the continuous precipitation
of intermetallics (Ni3Ti and Fe2Mo). Non-magnetic austenite phase lowers the overall
permeability and simultaneous precipitation of intermetallics decreases the resistivity,
both contributing to drastic decrease in EC parameters.
The presence of reverted austenite has been reported detrimental to toughness due
to the deformation concentration at softer austenite, which reaches its critical strain of
fracture at early stage. Hence, the microstructures containing austenite are avoided and
this can be ascertained by using either of the EC parameters with appropriate
thresholds.
The regime of technological importance (3h-10h) can also be identified by
specifying the corresponding EC parameters as they exhibit monotonous decrease in
this regime in Figure 4a. Between 30h and 70h, the EC induced voltage again changed
very gradually showing that the microstructural feature changes occurring do not
influence the EC induced voltage substantially. This is attributed to subtle increase in
the volume fraction of reverted austenite formed along with the precipitation, as evident
from austenite and hardness measurements in Figure 2. For the aging regime 70-100 h,
since the resistivity values Figure 2 remain almost constant, the observed drastic
decrease in the EC parameters is attributed to larger reduction in overall permeability
due to larger volume fraction of reverted austenite as evident from the XRD results in
Figure 2.
This study reveals that both magnitude and phase angle of induced voltage can be
used as NDE parameters for characterization of the microstructures of M250 steel.
Further, time for each measurement is less than 1 second. As compared to phase angle,
measurement of magnitude of induced voltage is accurate and easy. It is also easy to
implement the EC method in production line using portable eddy current instruments
and probes. Thus, this method holds a great promise for quick and reliable
characterization of microstructures in components in shop-floor for ensuring heat
treatment adequacy.
2.4

10

100
EC phase
EC magnitude

SA
SA

100

2.5

96

2.0

92

88

1.8
84

EC phase angle ()

EC magnitude (V)

2.2

2.0

0.25

10

100

10

100

SA

MBE rms voltage, V

0.25

1.5

1.0

0.5

1.6
0.25

10

Aging Time, h

100

80

0.0

0.25

Aging time, h

(a)
(b)
Figure 4. (a) Variation of induced voltage magnitude and phase angle with aging time
and (b) variation in rms voltage of MBE signal with aging time.

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

75

(A)

(B)
(C)

0.5 m
(a)

0.5 m
(b)

Figure 5. (a) Bright field TEM image of specimen solution annealed at 1093 K for 1 h
followed by air cooling (showing high dislocation density) (b) Bright field image of the
specimen thermally aged at 755 K for 100 h followed by water quenching, showing
long and patchy austenite (marked as A), patchy Ni3(Ti, Mo) (marked as B) and
globular Fe2Mo (marked as C).
The MBE rms peak voltage was found to vary with aging time as typically shown
in Figure 4b and the MBE trends were almost identical to eddy current measurements.
The MBE rms peak voltage remained almost constant from solution annealed condition
to 10 h of aging and dropped drastically on further aging. The solution annealed
microstructure is characterized by the martensitic lath/grain boundaries. In this
condition, the magnetic domains have to cross several martensitic lath/grain boundaries
before they give a detectable signal at the sensor coil. The constant MBE rms peak
voltage obtained from SA condition to 10 h of aging can be attributed to the two
opposing mechanisms taking place simultaneously i.e. dislocation annihilation and
intermetallic precipitation. Martensitic recovery, due to the annihilation of dislocations
occurring during initial aging is expected to increase the MBE by reducing the number
of pinning sites for domain wall motion. However, this regime is also characterized by
the continuous precipitation of intermetallics. Aging in this regime results in increase in
the precipitates which act as strong pinning sites to domain wall motion, thereby
reduction in MBE rms peak voltage is expected.
The drastic decrease in MBE rms peak voltage beyond 10 h of aging is attributed
to the initiation of the reversion of non-magnetic austenite phase. It is evident from the
TEM studies that austenite is formed at martensitic lath /grain boundaries. The
austenite formed at these lath and grain boundaries make these region nonmagnetic,
which cannot be easily surmounted by the growing/moving magnetic domains.
Moreover, the non-magnetic austenite phase is expected to be surrounded by stable
closure domains. Under these conditions, magnetization reversal can only advance in

76

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

limited volume fraction of the specimen at any given instant and the actual fraction
depends on the amount of reverted austenite. Aging for longer durations results in
increase in the volume fraction of the austenite, which further decreases the MBE rms
peak voltage. The evidence for increase in austenite volume percent is obtained from
XRD analysis (Figure 2) and TEM studies (Figure 5b).
Hardness and magnetic parameters exhibited different behaviour upon aging.
Hardness increased with the precipitation of intermetallic phases and decreased with
the reversion of austenite. The MBE rms peak voltage was found to be highly sensitive
to austenite reversion and was however insensitive to intermetallics precipitation during
aging. Hardness was found to be influenced more by the intermetallic precipitates as
compared to austenitic phase reversion. The reversion of austenite at 30 h of aging
could not be identified by the hardness due to simultaneous precipitation of
intermetallics, which tend to increase the hardness. The regime of technological
importance 3-10 h could not be identified by hardness or MBE parameter alone.
However, the study clearly revealed that the combination of these two parameters, i.e.
by specifying a minimum hardness (564 VHN) and a minimum MBE rms peak voltage
(1.8V) can be used for unambiguous characterization of the microstructure of
technological importance in M250 maraging steel [10].

3. Conclusion
The present study investigated the influence of microstructural features evolved upon
aging of M250 maraging steel at 755 K for different durations on electromagnetic NDE
(eddy current and MBE) parameters. In eddy current method both magnitude and phase
angle of induced voltage of receiver coil were found to be sensitive to the
microstructure changes through electrical conductivity and magnetic permeability. The
EC parameters could distinctly identify the over-aging due to austenite reversion, a
non-magnetic phase in magnetic matrix. The dislocation annihilation and intermetallics
precipitation were also found to influence the EC parameters. For the first time, it has
been observed that using EC parameters it is possible to study the recovery (i.e.
removal of quenched-in point defect and annihilation of dislocations) during initial
aging which increases the magnetic permeability and decreases resistivity. The
monotonous decrease in EC parameters can be effectively used to identify the aging
regime of technological importance (3-10h) at 755 K.
The MBE rms peak voltage was also found to be highly sensitive to austenite
reversion and was however insensitive to intermetallics precipitation during aging.
Hardness was found to be influenced more by the intermetallic precipitates as
compared to the austenitic phase reversion. The reversion of austenite at 30 h of aging
could not be identified by the hardness due to simultaneous precipitation of
intermetallics, which tend to increase the hardness. The regime of technological
importance 3-10 h could not be identified by hardness or MBE parameter alone.
However, the study clearly revealed that the combination of these two parameters can
be used for unambiguous characterization of the microstructure of technological
importance in M250 maraging steel. The study also established that electromagnetic
non destructive methods hold good promise for shop floor assessment of heat treatment
adequacy.

K.V. Rajkumar et al. / Characterisation of Microstructures in Heat Treated Maraging Steel

77

Acknowledgements
Authors thank Dr. P. Shankar, Dr. Anish Kumar and Mr. S. Mahadevan of Indira
Gandhi Centre for Atomic Research (IGCAR), Kalpakkam, India for many useful
discussions.
References
[1]
[2]
[3]
[4]
[5]

G.P. Miller, W.I. Mitchell, J. Iron Steel Inst. 203 (1965) 899-904.
D.T. Peters, C.R. Cupp, Trans. Met. Soc. AIME 236 (1966) 1420-1429.
V.K. Vasudevan, S.J. Kim, C.M. Wayman, Metall. Trans. A 21, (1990) 2655-2668.
W. Sha, A. Cerezo, G.D.W. Smith, Metall. Trans. A 24 (1993) 1221-1232.
R.F. Decker, S. Floreen, IN: R.K. Wilson (Ed.), Maraging Steels: Recent Developments and
Applications, TMS-AIME, Warrendale, PA, (1988) 138.
[6] Z. Guo, W. Sha, D. Vaumousse, Acta Mater. 51 (2003) 101-116.
[7] B.P.C. Rao, Introduction to eddy current testing, Narosa Publishing, New Delhi, April, 2007
[8] K.V.Rajkumar, B.P.C. Rao, B.Sasi, Anish Kumar, T.Jayakumar, Baldev Raj and K.K. Ray, Materials
Science and Engg. A 464 (2007) 233-240.
[9] M.J. Sablik, J. of Appl. Phys. 89 (10) (2001) 5610-5613.
[10] K.V. Rajkumar, S. Vaidyanathan, Anish Kumar, T. Jayakumar, Baldev Raj and K.K. Ray, J. of
magnetism and magnetic materials 312 (2007) 359-365.

78

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-78

Novel Acoustic Barkhausen Noise


Transducer and its Comparison with
Electromagnetic Acoustic Transducer
John WILSON1, Gui Yun TIAN1, Rachel S. EDWARDS2 and Steve DIXON2
School of Electrical, Electronic and Computer Engineering, Newcastle University,
NE1 7RU, 2Department of Physics, University of Warwick, Coventry, CV4 7AL

Abstract

The analysis of magnetic Barkhausen noise (MBN) has been used to provide
information about the stress state and microstructural properties of ferromagnetic
materials. Recent work has shown that a technique using acoustic Barkhausen
noise (ABN) detection can provide the similar capabilities as traditional MBN
along with additional information for defect characterisation and thickness
measurement in a single system. Because the detection of ABN using a
piezoelectric sensor can be carried out at any point on the material surface, as well
as analysing ABN for microstructural characterisation, the interaction of the
surface propagating waves with defects can also be analysed and used for defect
characterisation, along with frequency analysis for thickness measurement. As
with electromagnetic acoustic transducer (EMAT) systems, the ABN technique
applies totally different physical principles to traditional ultrasonic methods and
couplant is not needed for excitation. The work is carried out through experimental
investigations of calibrated steel samples with machined defects using the ABN
system, in comparison to readings taken using an EMAT system. Test results show
that the ABN technique has potential applications in providing a comprehensive
system for material and stress characterisation along with the additional
capabilities of defect characterisation and material thickness measurement.

Introduction
The inspection of ferromagnetic structures such as oil and gas pipeline [1], rail track [2]
or ferromagnetic components in manufacturing is a common requirement in industry,
and there is wide need for accurate defect detection and characterisation apparatus for
the prediction of failure in ferromagnetic engineering structures and components. This
is to some degree addressed by current inspection techniques, but each method has
drawbacks, for example magnetic flux leakage (MFL) has good detection capabilities,
but there is difficulty in extracting accurate characterisation data [1], although the
recently introduced pulsed MFL (PMFL) system has improved the potential capabilities
of MFL [3]. A reasonable standard of detection and characterisation can be achieved
with ultrasonic testing [2, 4-12], but with some limitations for standard contact
ultrasonic measurements. This has led to the development of alternative non-contact
excitation techniques.

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

79

Acoustic sources for ultrasonic NDE fall into two categories: contact techniques
requiring direct acoustic coupling between the source and the material under
inspection, and non-contact techniques. The major drawback of contact transducers is
the need to provide acoustic coupling between the element and the material under
inspection, so real world inspections usually require extensive surface preparation and
removal of coatings adding to time and expense. There are clear advantages in
techniques where physical contact between transducer and the area under inspection is
not required. Techniques used to address the problem include guided wave technology
[4], where the sensors are located remotely to the area under inspection, water coupled
systems [5, 6], where the material and transducer are immersed in a water tank, pulsed
laser generation of ultrasonic sources [7, 8] and air coupled systems [9,10]. The major
drawback of an air-coupled system is the miss-match in acoustic impedance between
air and solid materials [9], meaning that overall path losses for an air coupled system
can be 100dB + higher than water coupling, with the greatest losses occurring with
ferrous metals [10].
In contrast to other non-contact techniques, electromagnetic acoustic generation
opens up opportunities for ultrasonic generation without direct access to the material
surface [11]. As coupling between transducer and material is provided
electromagnetically, inspection through coatings and corrosion layers is possible and
coupling can be established to materials with irregular surface geometry.
The most commonly used electromagnetic ultrasonic source is the electromagnetic
acoustic transducer (EMAT) [2, 11, 12]. EMAT transducers consist of an excitation
coil driven by a current pulse in the presence of a static magnetic field. In non-magnetic
conductive materials such as aluminium, application of a current pulse to the coil in
presence of the permanent field causes Lorentz forces in the material, which in turn
generate acoustic waves. In magnetic and conductive materials such as steel,
magnetostrictive effects occur in addition to Lorentz forces. Although EMATs provide
a solution to the coupling problem in electrically conductive and magnetic materials,
the ultrasonic generation efficiency of an EMAT is much lower than that of a
piezoelectric transducer.
In this paper, a new electromagnetically generated acoustic source for
ferromagnetic materials is proposed utilising acoustic Barkhausen noise (ABN). As
well as providing a non-contact excitation method for ultrasonic investigation, ABN
uniquely carries information about the stress and microstructure of the material within
the signal itself. The technique has the potential to provide an affordable, low power,
non-contact solution for electromagnetically induced acoustic defect assessment with
the potential to also supply stress and material characterisation data.

1. Experimental Investigation of ABN for Acoustic Source Generation


When a time varying magnetic field is applied to a ferromagnetic material, the induced
magnetism is not continuous; rather, it is made up of jumps in magnetisation
corresponding to domain wall movement, this can be detected by a pick-up coil or
magnetic field sensor and is known as magnetic Barkhausen noise (MBN) [13]. This
domain wall motion also causes a release of elastic energy which manifests itself as an
acoustic pulse and can be measured using a piezoelectric sensor; this is known as
magneto-acoustic emission (MAE) [14] or ABN [15]. As magnetic domain structure is

80

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

intrinsically linked material microstructure, MBN and ABN are sensitive to material
stresses and changes in microstructure [13-15].
This work extends ABN beyond microstructure and stress characterisation to
provide defect assessment capabilities using broadband excitation. In contrast to
EMATs and most other excitation techniques, the frequency of the measured signal is
dependant, not on excitation frequency, but on the frequency of the elastic energy
released by domain wall movement. Low frequency excitation can be used, thus
dramatically reducing the comparative power consumption of the system.

(a)

(b)

Figure 1: a) Probe design, b) One repetition of positive half cycle of pulsed ABN and excitation signal, with
calculated signal profile

The experimental probe is shown in figure 1a. A ferrite core is mounted on the
sample under inspection. A piezoelectric sensor is mounted on the material surface with
petroleum jelly used to provide acoustic coupling between the sensor and the material.
Pulsed excitation is applied to the ferrite core and data acquired from the sensor and the
excitation current simultaneously at a sample frequency of 2MHz. In this work, a
piezoelectric Physical Acoustics R15I-AST integral preamplifier acoustic emission
receiver is used; the sensor is resonant at 150kHz, with a useable frequency range of
around 50kHz 200kHz and a total gain of 72dB is applied to the signal. Figure 1b
shows one repetition of the positive half cycle of the pulsed excitation waveform and
the associated ABN signal. Several signal processing techniques are used in the work;
the most basic of these is calculation of the signal profile or envelope by rectification
followed by a moving average calculation as shown in figure 1b.
1.1. Characterisation of ABN signal
In an ultrasonic defect detection system, the minimum detectable defect depth is
proportional to the wavelength of the signal being measured, so knowledge of the
frequency range of the acoustic source is vital for the developed system, but the
frequency spectrum of ABN has not been widely studied. Figures 2a and 2b show FFT
envelopes for two different materials; a 60mm x 60mm x 1000mm steel block and a
1mm thick mild steel sheet. Sine wave excitation was used in the test, with the ABN
signal from six different frequencies from 2Hz to 1kHz recorded for comparison. It can
be seen from the plots that the basic distribution of frequency components remains the

81

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

same regardless of excitation frequency, although there is a change in amplitude over


the excitation frequencies used. This is as expected, as the frequency spectrum of ABN
is dependant on the frequency of the elastic wave released by the motion of the domain
wall, not the excitation frequency.
FFT
ENVELOPES: SINE EXCITATION, STEEL PLATE
-2
10

5Hz Excitation
10Hz Excitation
50Hz Excitation
100Hz Excitation
1kHz Excitation

-4

10

-5

10

2Hz Excitation
5Hz Excitation
10Hz Excitation
50Hz Excitation
100Hz Excitation
1kHz Excitation

-3

10
Amplitude(Log)

Amplitude(Log)

FFT
ENVELOPES: SINE EXCITATION, STEEL BLOCK
-3
10
2Hz Excitation

-4

10

-5

10

-6

-6

10

100

200
300
Frequency - kHz

400

10

500

(a)

100

200
300
Frequency - kHz

400

500

(b)

(c)
Figure 2: ABN frequency spectrum for; a) thin steel plate, b) large steel block, c) ABN amplitude for
frequency variation

Comparison of figures 2a and 2b show that, although material properties will


certainly play a part in the ABN frequency response, for these unstressed samples,
material dimensions have a much greater influence on the frequency spectrum of the
induced signal than excitation frequency. With the thicker sample, the lower frequency
portion of the signal is enhanced. This is due to the reinforcement of frequency
components with wavelengths corresponding to the material dimensions. So at an
estimated signal velocity of 3000 m/s, with the 60mm sample the frequency component
corresponding to a wavelength of 60mm is 3000 / 0.06 m = 50 kHz and the frequency
component corresponding to half the wavelength is 100 kHz. It can be seen from the
plot that these areas of the plot are indeed enhanced for the thicker sample. For the
thinner sample, the frequency component corresponding to a wavelength of 1mm is
3000 / 0.001 m = 3 MHz, well out of the measured range, so the plot for this sample is
much smoother, without the reinforcement of wavelengths corresponding to the
thickness of the sample.
ABN RMS amplitudes for the frequencies used in the tests are shown in figure 2c.
It can be seen from the plot that the ABN amplitude increases as the excitation

82

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

frequency increases. This is due to increasing overlapping of clustered random event


outputs as the excitation frequency increases [16]. This increase in ABN amplitude will
not continue indefinitely as frequency increases, rather the rate will decrease as the
skin-depth effect becomes apparent [16].

2. Application of ABN to defect detection


An experiment was set up to assess the capabilities of the system to provide depth
information for perpendicular defects in steel. Two samples, shown in figure 3a and 3b,
were used, measuring 60mm x 60mm x 600mm (sample 1) and 60mm x 60mm x
1000mm (sample 2). The blocks, contain five 1mm wide defects, with depths of
0.5mm, 2mm, 2.5mm, 3.5mm and 10mm. The test set-up is shown in figure 3c. In these
tests a single sensor is used at different distances (d) with respect to slot position, with
excitation apparatus kept at a constant distance to the slot. d is given as the distance
between the sensor rim closest to the slot and the slot under inspection, at no time in the
tests is the sensor actually over the slot.
The peak signal strengths for sensor positions from d = 10 to d = -10 are shown in
figure 3e. It can be seen from figure 3e that the peak amplitude for all but the 2mm slot
shows a maximum at d = 0 or d = 1. The EMAT normalised signal amplitude on
approaching a 3mm deep slot in a steel sample is shown in figure 3f [11]. A signal
enhancement is also observable in EMAT systems close to material voids where
reflections and mode conversions from the slot interfere with the direct signal, as
shown in figure 3f [11]. The proportion of transmitted surface wave energy will depend
on the crack depth and the wavelengths within the broadband signal, through the
transmission coefficients [12].
FFTs were calculated from the AE sensor signals, for four slot depths, and the
mean frequencies calculated, as shown in figure 3d. It can be seen from the plot that for
the 2.5mm, 3.5mm and 10mm slots the mean frequency decreases in a fairly linearly as
the slot depth increases. This indicates that as slot depth increases, more of the lower
frequency signal component is reflected back to the sensor. As defect depth increases,
the wavelengths which can be reflected by the defect increase, thus lowering the mean
frequency of the reflected wave. As the capabilities of an ultrasonic system with
respect to measurable slot depth are dependant on the frequency range of the system, it
may be that characterisation of the 2mm slot is beyond the capabilities of the ABN
system using this particular sensor. With ultrasonic measurements using a wideband
Rayleigh wave pulse, the calculated cut-off frequency of the measured signal has been
shown to be proportional to slot depth [12].

(a)

(b)

83

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

Frequency - kHz

ABN MEAN FREQUENCY FOR FOUR SLOTS


121.2

121

120.8

120.6

120.4
2

6
8
Slot depth - mm

1.3
1.2
1.1
1
0.9
-10

-5
0
5
10
Sensor position with respect to slot - mm

(e)

(d)
Normalised signal amplitude

Amplitude - V

(c)
ABN PEAK AMPLITUDE OVER DEFECT AREA
2mm SLOT
2.5mm SLOT
1.6
3.5mm SLOT
1.5
10mm SLOT
1.4

10

1.6
1.4
1.2
1.0
0.8
0.6
0.4
-80

-60

-40

-20

20

40

Scale (mm)

(f)

Figure 3: a) Test sample 1, b) Test sample 2, c) Defect location test set up, d) Mean ABN frequency for four
slot depths, e) ABN peak amplitude over slot area, f) EMAT amplitude approaching a 3mm deep slot in a
steel sample

3. Discussion and Conclusions


Several parameters, summarised in figure 4, influence the form of the ABN signal, with
the greatest influences coming from material properties, sample geometry and
excitation waveform. The peak position of the ABN signal with respect to the
excitation cycle is influenced by the excitation signal itself and the domain structure of
the material, which affects the domain wall activity at different stages in the hysteresis
cycle. By far the greatest influence on ABN frequency spectrum comes from the
geometry of the sample, through internal reflections of the acoustic signal, but the
domain structure will also have some influence through the types of domain wall which
are active in the sample. The overall ABN amplitude is influenced by all three
parameters, through the amount of domain activity in the sample, reinforcement of the
signal through sample geometry and magnetic field intensity in the sample through the
amplitude of the excitation signal. The decay of the ABN signal is greatly influenced
by sample geometry through reflections received by the sensor after the main domain
wall activity has ceased. ABN peak amplitude is of course affected by the overall
amplitude of the ABN signal, but also by the excitation waveform. The higher the rate
of change of the signal, the more ABN activity will be contained within the same time
period, leading to overlapping of acoustic events.

84

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

Figure 4: Contribution of various factors to ABN signal characteristics

ABN can work as an acoustic source which also carries information on material
properties, which is a distinct advantage over current acoustic NDT. Although
preliminary test results show that ABN has some promise as an acoustic source, there
are several problems with technique; the ABN signal intensity is relatively weak; the
signal frequency spectrum is dependent on the frequency of elastic energy from domain
wall, not excitation frequency this could be a problem due lack of controllability, but
also means that low frequency excitation tuned to the optimal frequency for maximum
power transfer can be used to excite high frequency ABN source; unlike the ultrasonic
measurements using EMATs, the mode of propagation of the signal is not fully
understood; the ABN signal is made up of many different acoustic events this could
be problematic for processing. A piezoelectric sensor is used for present tests, but for
true non-contact operation, a non-contact ABN receiver should be developed.
Future tests will be made with real world defects such as cracks, voids and
dislocations and the limitations of the system ascertained. Although the technique will
have limitations in terms of the interaction between the wavelength of the excited
signal and defect depth, the change in the ABN signal caused by the interaction
between domain walls and defects will be incorporated into future measurements.

References
[1]

[2]

[3]
[4]

[5]

A.A. Carvalho, J.M.A. Rebello, L.V.S. Sagrilo, C.S. Camerini and I.V.J. Miranda, MFL signals and
artificial neural networks applied to detection and classification of pipe weld defects, NDT & E
International, Volume 39, Issue 8, December 2006, Pages 661-667.
Y. Fan, S. Dixon, R.S. Edwards and X. Jian, Ultrasonic surface wave propagation and interaction with
surface defects on rail track head, NDT & E International, Volume 40, Issue 6, September 2007, Pages
471-477.
J.W. Wilson and G.Y. Tian, Pulsed electromagnetic methods for defect detection and characterisation,
NDT & E International, Volume 40, Issue 4, June 2007, Pages 275-283.
A. Demma, P. Cawley, M. Lowe, A. G. Roosenbrand and B. Pavlakovic, The reflection of guided
waves from notches in pipes: a guide for interpreting corrosion measurements, NDT & E
International, Volume 37, Issue 3, April 2004, Pages 167-180.
X. Jian, J.P. Weight and K.T.V. Grattan, Miniature wideband ultrasonic transducers to measure
compression and shear waves in solid, Sensors and Actuators A: Physical, Volume 127, Issue 1, 28
February 2006, Pages 13-23.

J. Wilson et al. / Novel Acoustic Barkhausen Noise Transducer

[6]
[7]
[8]

[9]
[10]
[11]
[12]
[13]

[14]
[15]
[16]

85

R.E. Green, Non-contact ultrasonic techniques, Ultrasonics, Volume 42, Issues 1-9, April 2004, Pages
9-16.
B.B. Djordjevic, D. Cerniglia, and C. Cosenza, Guided wave non-contact ultrasonic for NDE, WCNDT
2004.
Y. Hong, S.D. Sharples, M. Clark and M.G. Somekh, Rapid and accurate analysis of surface and
pseudo-surface waves using adaptive laser ultrasound techniques, Ultrasonics, Volume 42, Issues 19, April 2004, Pages 515-518.
E. Blomme, D. Bulcaen and F. Declercq, Air-coupled ultrasonic NDE: experiments in the frequency
range 750 kHz2 MHz, NDT & E International, Volume 35, Issue 7, October 2002, Pages 417-426.
J. Buckley, Air-coupled Ultrasound - A Millennial Review, WCNDT 2000.
R.S. Edwards, A. Sophian, S. Dixon, G.Y. Tian and X. Jian, Dual EMAT and PEC non-contact probe:
applications to defect testing, NDT & E International, Volume 39, Issue 1, January 2006, Pages 45-52.
R.S. Edwards, S. Dixon and X. Jian, Depth gauging of defects using low frequency wideband Rayleigh
waves, Ultrasonics, Volume 44, Issue 1, January 2006, Pages 93-98.
V. Moorthy, B.A. Shaw and P. Hopkins, Surface and subsurface stress evaluation in case-carburised
steel using high and low frequency magnetic barkhausen emission measurements, Journal of
Magnetism and Magnetic Materials, Vol. 299(2), Apr. 2006, pp. 362-375.
D. O'Sullivan, M. Cotterell, D.A. Tanner and I. Mszros, Characterisation of ferritic stainless steel by
Barkhausen techniques, NDT & E International, Volume 37, Issue 6, September 2004, Pages 489-496.
G.Y. Tian, J. Wilson and J. Keprt, Magnetic-acoustic Emission for Stress and Material Characterisation,
ENDE 2006.
H.C. Kim and C.G. Kim, Effect of magnetising frequency and stress on magneto-acoustic emission
from 3% Si-Fe crystals, Journal of Physics D: Applied Physics, Volume 22, Issue 1, pp. 192-198 (1989).

86

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-86

Modelling and Measurement of


Decarburisation of Steels Using a Multifrequency Electromagnetic Sensor
X.J. HAOa,1 , W. YINb, M. STRANGWOODa, A.J. PEYTONb, P.F. MORRISc and C.L.
DAVISa
a
Department of Metallurgy and Materials, University of Birmingham, Birmingham,
B15 2TT, UK
b
School of Engineering, University of Manchester, Manchester, M60 1QD, UK
c
Corus plc. Swinden Technology Centre, Moorgate, Rotherham, S60 3AR, UK

Abstract. Decarburisation of high carbon steel has been simulated, using


composite samples comprised of a 316 stainless steel (paramagnetic) core and a
surrounding tube of ferritic steel (ferromagnetic) with thicknesses between 100
and 600 m, for determining the potential for on-line measurement during steel
processing. Decarburization samples have also been generated, for off-line
measurements, by heat treatment of an Fe-0.8 wt% C steel for various times in air
at 1000 1200C. A multi-frequency (10 106 Hz) electromagnetic sensor was
used to determine variations in inductance (due to differences in permeability) as a
function of decarburisation depth. The relationship between sensor output and
decarburised layer type/thickness has been modelled using finite element software.

1. Introduction
The heat treatment and hot processing of steel usually requires the material to be heated
into the austenite phase field, in the temperature range of 800~1200oC. At these
temperatures, carbon at the surface can be removed by reaction with oxygen in the
surrounding atmosphere and this process is known as decarburisation. Loss of carbon
from the surface is more rapid than replenishment by solid-state diffusion, causing a
gradient in carbon level from bulk to surface. This effect is greater in higher carbon
steels. Loss of carbon at the surface can have a significantly detrimental effect on
mechanical properties of products, since hardness, fatigue, strength, and wear
properties are strongly dependent on carbon content. Commercially the depth and
extent of decarburisation is controlled, which needs accurate measurement of this
phenomenon. Currently, measurement of decarburisation is by destructive methods,
such as metallographic observation or hardness tests on a cross section of samples after
processing. These methods are time consuming and cannot be applied during the
production process. This study aims to develop a multi-frequency electromagnetic
(EM) sensor to monitor the decarburizing process on-line, and to measure
decarburisation depth off-line. Below the Curie temperature ( 770C for carbon steel)
austenite is paramagnetic and ferrite is ferromagnetic. EM sensors work on the basis of
1

Corresponding author: X.J. Hao, E-mail: Xinjiang.Hao@gmail.com

X.J. Hao et al. / Modelling and Measurement of Decarburisation of Steels

87

detecting the difference in relative permeability, R, and conductivity, V, between


microstructural phases or due to changes with temperature. An air cored multifrequency EM sensor can measure the ferrite fraction in steel over the entire ferrite
fraction range: from 0-40% ferrite fraction through low frequency inductance; and from
40-100% ferrite fraction through the zero crossing frequency, by the differences in
effective permeability of the dual phase microstructure [1,2]. Decarburising at high
temperature induces a carbon gradient from surface to interior. During cooling from
austenite, the surface layer, with low carbon content, transforms first to ferrite whilst
the interior transforms to pearlite later. An EM sensor should therefore be able to
measure the decarburisation depth by detecting the ferrite fraction gradient. For on-line
measurement, if the EM sensor is placed above the steel when it is between the Curie
and eutectoid temperatures, any ferromagnetic ferrite surface layer may be detected
compared to the paramagnetic austenite core. For off-line measurements differences in
permeability between ferrite and pearlite, which are much less than between ferrite and
austenite, would need to be detected.

2. Experimental
On-line measurement has been simulated using composite samples comprised of a 316
stainless steel (austenite, paramagnetic) core and a surrounding tube of ferritic steel
(0.17 wt% C, ferromagnetic). 316 stainless steel bars, 8mm diameter, were inserted into
the ferritic steel tubes (8mm inner diameter, 1mm wall thickness) and then cold drawn
to 6.7mm (composite bar outer diameter) to increase contact between core and outer
layer. The composite bars (300mm length) were straightened and center-less ground to
get ferritic layers with thicknesses of about 100, 200, 300 and 600 m. For off-line
measurements, decarburisation samples were generated, by heat treatment of Fe-0.8
wt% C steel bars (10mm diameter, 150mm length) in an air furnace at 1000C for
various times (10min to 5 hours) then cooled in air. Any loose surface oxidation layer
was removed by gentle tapping. Transverse microstructures from each sample were
examined by optical microscopy after sectioning, polishing and etching using 2% nital.
A multi-frequency electromagnetic (EM) sensor was used to determine variations in
inductance (due to differences in permeability) as a function of decarburisation depth.
Samples were measured by inserting them into the air-cored cylindrical sensor, which
had a length of 10mm and diameter of 20mm and was driven using an impedance
analyser at frequencies from 10 to 106 Hz. The relationship between sensor output and
decarburised layer type / thickness was modelled by finite element method (FEM)
using COMSOL Multiphysics [3] software.
Sensor
3. Results and discussion

Austenite
Hot on-line testing was simulated using the
composite
bar
samples.
Microstructure
Ferrite:
observation of the composite bars has confirmed
that the ferrite layer makes good contact with the
austenite. The setup for sensor measurement is Figure 1. Setup of sensor/sample
shown in Fig. 1, which was used to model the for experiments and modelling

88

X.J. Hao et al. / Modelling and Measurement of Decarburisation of Steels

sensor output with decarburisation depth.


-4

10

Inductance (H)

8
6
4
2

Frequency: 100Hz

1.0

Ferrite layer
thickness (Pm)
0
100
200
300
600

10

Normalized Inductance

12

0.8
0.6

Pr

0.4

50
200
1000

0.2

measured

0
-2
1
10

0.0
2

10

10

10

Frequency (Hz)

10

100

10

200

300

400

500

600

Decarb thickness (Pm)

Figure 2. (a) Measured inductance vs. frequency for composite bars and (b) comparison
between measured and modelled results.
5.0

10

-4

4.8
4.6

Inductance (H)

4.4
4.2
4.0
3.8
3.6
3.4

200Pm

At 1000 C for:
10min
Increasing
1hr
decarburisation
2hrs
5hrs

3.2
3.0
1
10

Frequency

10

Figure 3. (a) Microstructure of Fe-0.8wt%C steel after decarburising at 1000C for 2


hours showing surface ferrite (white) and pearlite (black), and (b) measured inductance
change with frequency for samples after various decarburising times at 1000C.
The EM sensor output, in terms of inductance versus frequency is shown in Fig.
2(a). The magnetic field produced by an EM sensor acts on a ferromagnetic target in
two ways. First it tends to magnetise the metal, which increases the coils inductance.
Second, the alternating current magnetic field also induces eddy currents in the metal,
which tend to oppose the driving current and reduce the coils inductance. At low
frequencies, magnetisation dominates the inductance. As the frequency is increased, the
effects of eddy current become more dominant and the inductance decreases,
eventually approaching a constant value at high frequencies. As shown in Fig. 2(a), the
inductance value increases with increasing decarburisation thickness (ferrite layer
thickness) at low frequencies. This indicates that the low frequency inductance value is
a good parameter for measuring decarburisation thickness. Therefore, the measured and
modelled inductance values (normalized for the sample before centre-less grinding,
600m decarburisation depth, for comparison) at low frequency (100 Hz) versus
decarburisation thickness are shown in Fig. 2(b). As the relative permeability of ferrite
could not be experimentally determined (and varies considerably with carbon content),
three values over a large range (50, 200 and 1000) have been chosen for FEM
modelling. Both the measured and modelled results show that inductance changes with
decarburisation thickness in a non-linear manner. With increasing decarburisation
thickness from zero, the inductance increases quickly, and then tends towards
saturation. This phenomenon is due to the demagnetising field effect. When a finite
size ferromagnetic sample is magnetised in an applied magnetic field, a demagnetising
field appears in the sample, which is opposite to the applied field. For the sample with

X.J. Hao et al. / Modelling and Measurement of Decarburisation of Steels

89

the largest relative permeability (1000), the demagnetising field effect is the most
significant. It should be noted that the measured and modelled results are not in perfect
agreement. This may be due to non-uniformity of ferrite layer thickness along the rod
after grinding, and / or the plastic deformation introduced in the ferrite layer, which
affects its permeability [4].
The simulation of on-line testing has shown that the EM sensor has the ability to
detect decarburisation depth based on the fact that the relative permeability of ferrite is
hundreds of times larger than that of austenite. For off-line testing, the sensor needs to
be able to distinguish between the surface ferrite and core pearlite. The relative
permeability ratio of ferrite/pearlite is much smaller than ferrite/austenite because
pearlite is also a ferromagnetic phase. Thompson and Tanner [5] have shown that the
initial relative permeability of pearlite (0.87 wt% C) is 56, whereas ferrite with a small
amount of pearlite (0.17 wt% C) is 280. Modelling work determined that the EM
sensor should have the ability to detect the difference between ferrite and pearlite even
with such a small relative permeability ratio.
Fig. 3(a) shows the microstructure from a Fe-0.8 wt% C sample after decarburising
treatment of 1000C for 2 hours, which consists of a surface ferrite zone, with a ferrite
and pearlite mixed zone followed by fully pearlite on moving into the sample core. The
sensor measured results in Fig. 3(b) show that inductance increases with increasing
decarburising time (and hence decarburised depth) over the whole measured frequency
range used (10 to 100 Hz). At a frequency of 10 Hz, the sensor could not readily
distinguish the two samples with the least amount of decarburisation: decarburisation
times of 10min (containing no separate ferrite layer but only a thin mixed zone), and 1
hour (containing a thin ferrite layer and a thicker mixed zone). By increasing the
frequency these samples become distinguishable. Therefore, to detect thin decarburised
layers, such as seen in industrial processing, the appropriate frequency needs to be
chosen. This is understandable from the point of view that at very low frequency, if the
skin depth is much larger than the thin decarburisation layer thickness, the contribution
to the inductance from the decarburisation layer is too small compared to that from the
core pearlite. On the other hand, if the skin depth is smaller than the decarburisation
layer thickness, the variation of decarburisation layer thickness cannot be detected.

4. Conclusions
Experiments and modelling show that decarburisation in steel rod samples can be
measured using an EM sensor on-line during hot processing (by differences in
permeability between surface ferrite and bulk austenite) or by off-line cold testing (by
differences in permeability between surface ferrite and bulk pearlite). Appropriate
testing frequency needs to be selected dependent on the decarburisation depth and
whether on-line or off-line testing is being carried out to optimise signal output.

References
[1] Papaelias MP, Strangwood M, Peyton AJ, Davis CL, Metall Mater Trans A 2004 (35A) 965-72
[2] Haldane RJ, Yin W, Strangwood M, Peyton AJ, Davis CL, Scripta Mater 54 (2006) 1761-65
[3] http://www.comsol.com/
[4] Thompson SM, Tanner BK, JMMM 132(1994) 71-78
[5] Thompson SM, Tanner BK, JMMM 123(1993) 283-298

90

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-90

Assessment of Grinding Damage on Gear


Teeth using Magnetic Barkhausen Noise
Measurements
Moorthy VAIDHIANATHASAMY, Brian Andrew SHAW, Will
BENNETT*, Peter HOPKINS*
Design Unit, Newcastle University, Newcastle upon Tyne, UK
*Ministry of Defence (Navy), Bristol, UK

Abstract. The marine Gears made with Case-carburised En36 steel were subjected
to grinding damage during manufacturing. The MBN measurements have been
made on different Gear teeth using three different methods, namely, High
Frequency, Medium Frequency and Low frequency MBN measurements with the
optimised measurement device and parameters so that the MBN signal from
different depth ranges can be detected and analysed. The MBN measurements on
these Gear teeth have shown that the grinding damage near the surface (< 10m
depth) can be detected using High frequency MBN, any sub-surface damage
(within ~40m depth) can be detected using Medium frequency MBN and the
damage in the deeper layers (> 40m depth) can be detected using Low frequency
MBN measurements. The thermal damage caused by Grinding Burn is also clearly
revealed by the shifting of the Low frequency MBN peak to lower magnetic field.
Keywords. Magnetic Barkhausen Noise, Gears, Grinding damage.

1. Introduction
The grinding damage induced during the final stage of manufacturing is a major
concern affecting the quality and hence the fatigue life of Gears. It is known that the
grinding damage is associated with alterations in Residual Stress (RS) distribution and
the microstuctural state caused by thermal effects such as Burning or Rehardening.
Conventionally, the grinding damage is assessed using Nital Etching method which
reveals only microstructural changes on the surface. Often, the depth and the extent of
the grinding damage may vary at different locations of the Gear teeth. In Casehardened steel Gears, it has often been found that the surface of the material may not
reveal severe damage due to immediate cooling of the surface. But, the thermal damage
and deformation may have altered the microstructure and RS in the sub-surface
severely [1].
Since, the Magnetic Barkhausen Noise (MBN) signal is generated by the magnetization
process which is strongly influenced by the microstructural and stress states of the
ferromagnetic material, the MBN technique is considered as a potential NDT method to
evaluate the grinding damage in ferromagnetic steels [2]. The MBN measurements can
be made with different set of parameters, varying the maximum level and the excitation

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

91

frequency of external magnetic field strength, the sensitivity of MBN pick-up coil, the
analysing frequency range etc. All these factors decide the range of magnetisation and
the depth from which the MBN signal is detected. In this study, three different MBN
measurement methods have been compared for evaluating the depth range and extent
of Grinding damage in carburised steel Gears. An attempt has been made to relate the
three different MBN measurements with RS alteration below the surface at different
depth ranges and also qualitatively assess the extent of thermal damage.

2. Experimental procedure
The helical gears made from case-carburised En36 steel were subjected to different
levels of damage during final grinding process. Small portions (25 mm long) of the
teeth were cut from different locations of these Gears so that the Residual Stress (RS)
Depth profiles and the MBN measurements can be made using the existing X-ray
diffraction system and the MBN devices at the Newcastle University. The three
different MBN measurement methods, namely, High, Medium and Low frequency
MBN measurements were made using three different set of measurement devices
(electromagnetic yoke and MBN pick-up coil) and parameters (excitation frequency fex,
maximum applied magnetic field Hmax and analysing frequency range, and signal
amplification). The High frequency MBN measurements were made with MBN system
and Gear probe supplied by Stresstech, Finland. The Medium and Low frequency
MBN measurements were made with the MBN system and devices developed at
Design Unit, Newcastle University, UK. The details of the High frequency MBN
measurements (fex=125 Hz, Hmax= 70 Gauss) and Low frequency MBN measurements
(fex=0.2 Hz, Hmax=300 Gauss) are given elsewhere [2-4]. The Medium frequency
MBN measurements were made at 20 Hz magnetic excitation with maximum applied
magnetic field strength of 140 Gauss. The MBN signal was filtered using a 2 kHz
high pass frequency filter and amplified to a Gain of 40 dB. The MBN signal profile is
used for analysis.
The skin depth, from which the MBN signals are detected, strongly depends on several
measurement parameters such as the fex, Hmax, frequency response of the MBN pick-up
coil, analysing frequency range of the MBN signals etc. in addition the effect of
permeability and conductivity of the test materials. Due to the complex and synergistic
influence of these factors, it is not possible to precisely determine the skin depth
theoretically. However, it is well known that the skin depth of the MBN detection
decreases with the increase in frequency of external magnetic excitation and the
analyzing frequency range of MBN signals due to electromagnetic attenuation of the
MBN signals within the test material. Previous studies [2-5] using the High and Low
frequency MBN measurements made with differents set of parameters revealed that the
high frequency MBN measurements did not detect changes in material properties
beyond 10 m depth whereas the low frequency MBN measurements detect changes in
material properties to a depth of 600 m. Based on the previous experiences, the High,
Medium and Low frequency MBN measurements are expected to reveal the changes in
material properties in the near-surface (<10 m), subsurface (<40 m) and deeper
subsurface (upto ~ 400 m).

92

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

3. Results and Discussions


It is expected that the comparison of High, Medium and Low frequency MBN
measurements would indicate the possible distribution of residual stresses (RS) in the
depth direction due to different skin-depth of the MBN signals generated in these
measurements. In this paper, the comparison is shown with two examples.
As an example 1, the MBN profiles measured using the three different types of
measurements on a set of three different teeth from Gear S1 and their RS Depth
profiles are shown in Figure 1. The High frequency MBN profile shows no significant
change in MBN peak among these three teeth indicating no significant variations in RS
near the surface which is also supported by the RS-Depth profile. However, both the
Medium and Low frequency MBN profiles show significant variations between them.
The highest MBN peak for the Tooth-M indicates that it would have higher tensile RS
in the subsurface region compared to other teeth. This is also evident from the RSDepth profiles shown in Figure 1(d).

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

93

94

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

300

Gear S1

(d)

Residual Stress, MPa

200
100
0
0

10

20

30

40

50

-100
Tooth N
-200

Tooth M
Tooth W

-300
Depth, Pm
Figure 1. The MBN profiles measured with (a) High, (b) Medium and (c) Low
frequency measurements and (d) RS-Depth profiles for three teeth from Gear S1.
As an example 2, the comparison of High, Medium and Low frequency MBN profiles
and the RS-Depth profiles for two teeth from Gears T1 and T3 are shown in Figure 2.
In both the high and medium frequency MBN measurements, the T3-L tooth shows
higher MBN level than T1-1 tooth. But, the low frequency MBN profiles show the
opposite that the T1-1 tooth has higher MBN peak than T3-L tooth. The variations in
high and medium frequency MBN profiles suggests that the T3-L tooth may have less
compressive or more tensile RS than T1-1 tooth in the near-surface region (within
30m depth). It can also be observed that the variation in high frequency MBN peak
height shows a difference of ~ 40% and the variation in medium frequency MBN peak
shows a difference of ~ 65% between for T1-1 and T3-L teeth. The larger difference
shown by medium frequency MBN measurement indicates the presence of extended
difference in RS profile even in subsurface layers (up to ~ 30 m depth). It can be
clearly seen from the RS-depth profiles that the T3-L tooth has more tensile RS than
T1-1 tooth even at 20m depth. But, at 40m depth, the T1-1 tooth has tensile RS
value while T3-L tooth has compressive RS value. The presence of more tensile RS in
the deeper region is clearly reflected by the higher MBN peak value for T1-1 tooth than
T3-L tooth in the low frequency MBN measurements.
The distinct variations in the peak position of the Low frequency MBN profiles clearly
show the extent of thermal damage resulting in microstructural softening among these
Gear teeth. The higher applied magnetic field strength in the Low frequency MBN
measurement maximizes the magnetisation range and hence reveals the extent of
thermal damage by distinct shifting of the position of MBN profiles.

Average MBN level, % of 5V

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

Carburised Gears T1 and T3


fex = 125 Hz
'fa = 70-200 kHz

30

(a)

20

25

95

Tooth-T3-L
Tooth-T1-1

15
10
5
0
-100

-50

50

% of voltage applied to electromagnet

100

96

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

200

Carburised Gears T1 and T3

(d)

Residual Stress, MPa

100
0
-100 0
-200
-300
-400

10

20

30

40

50

Tooth-T3-L
Tooth-T1-1

-500
-600
-700
Depth, Pm

Figure 2. The MBN profiles measured with (a) High, (b) Medium and (c) Low
frequency measurements and (d) RS-Depth profiles for teeth from Gears T1 and T3.

M. Vaidhianathasamy et al. / Assessment of Grinding Damage on Gear Teeth

97

4. Conclusions
This study clearly shows that the difference in the RS distribution in the near-surface,
subsurface and deeper regions can be identified by comparing the MBN profiles
obtained from these three different MBN measurements. It is more appropriate to relate
the MBN peak position to the extent of thermal damage and the peak height to the
average RS variation over a certain depth depending on the type of MBN measurement
and the parameters used. However, since the High and Medium frequency MBN
measurements have limited effective magnetic field penetration resulting in smaller
magnetisation range and they could not isolate thermal damage from the RS variations.
References:
[1]
[2]
[3]
[4]
[5]

B.A.Shaw, J.T.Evans and D.E.Yates, The influence of Grinding on the formation of Residual Stress
Distribution in Ground Surfaces, Conference on Drives and Controls., Kamtech Publishing, UK, 1996,
21-26.
V.Moorthy, B.A.Shaw, P.Mountford and P.Hopkins, Magnetic Barkhausen emission technique for
evaluation of residual stress alteration by grinding in case-carburised En36 steel, Acta Materialia, 53
(2005) 4997-5006.
V. Moorthy, B.A. Shaw and P. Hopkins, Surface and subsurface stress evaluation in case-carburised
steel using high and low frequency magnetic Barkhausen emission measurements, J. Magnetism and
Magnetic Materials, 299, 2006, 362-375.
V. Moorthy, B.A. Shaw and P. Hopkins, Magnetic Barkhausen emission technique for detecting the
overstressing during bending fatigue in case-carburised En36 steel, NDT&E international, 38, 2005,
159-166.
V. Moorthy, B.A. Shaw, and K. Brimble, Evaluation of case depth in case-carburised gear steels using
magnetic Barkhausen emission technique, Materials Evaluation, 62 (5) May 2004, 523-527.

98

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-98

Evaluation of Contact Fatigue Damage on


Gears using the Magnetic Barkhausen
Noise Technique
Moorthy VAIDHIANATHASAMY, Brian Andrew SHAW, Will
BENNETT*, Peter HOPKINS*
Design Unit, Newcastle University, Newcastle upon Tyne, UK
*Ministry of Defence (Navy), Bristol, UK
Abstract. Gears made from Case-carburised SAE8620H steel were subjected to
contact fatigue testing at different contact stress levels in the Gear Test Rig facility
at Newcastle University. The High frequency Magnetic Barkhausen Noise (MBN)
measurements were made at different intervals of fatigue cycles. The MBN signal
level gradually increases with progressive fatigue cycles and the rate of change in
MBN level systematically increases with the increase in contact stress level.
However, at higher contact stress levels, the MBN level quickly reaches a
maximum value and then starts to decrease with further fatigue cycles. This
maximum MBN level may be considered as an indication of the critical stage of
fatigue life beyond which a gear may be expected to fail soon. With increasing
numbers of fatigue cycles, the large MBN activity occurs at lower magnetic field
strength which clearly indicates the formation of soft magnetisation regions near
the surface. This may be attributed to the deformation induced transformation of
paramagnetic retained austenite into ferromagnetic martensite.
Key words: Barkhausen Noise, Contact fatigue, Gears.

1. Introduction
One of the most common modes of Gear failure is the contact fatigue damage which is
commonly manifested as the initiation and progression of micro-pitting on the flanks of
gear teeth. The progress of micro-pitting damage alters the Micro-geometry of Gear
profile which in-turn alters the contact pattern of the Gear. This would introduce nonuniform high stress contact points and makes the Gear rotation more noisy, less
efficient and prone to fatigue failure of the Gear teeth. The contact fatigue failure of the
Gear normally occurs by the sudden breakage of the tooth by the crack propagation
initiated from the Gear flank. Hence, the monitoring of contact fatigue using NDE
method would be very useful in assessing the Gear condition and predicting the
remaining life of the Gear.

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

99

The Magnetic Barkhuasen Noise (MBN) is the voltage pulses induced in the pick-up
coil placed on the surface of the ferromagnetic material which is subjected to discrete
changes in magnetization by the external varying cyclic magnetic field. It has been
widely reported in the literature [1-3] that the MBN signal is very sensitive to changes
in microstructural and stress states of the ferromagnetic steels. Since many medium and
large size gears are manufactured using ferromagnetic steels, the MBN technique is
considered as a potential NDT method for assessing the contact fatigue damage in
Gears. This study has been carried out to understand and evaluate the variation in the
MBN signal in response to progressive contact fatigue damage in case-carburised steel
gears.

2. Experimental
The contact fatigue tests have been carried out at Design Unit, Newcastle University,
using 8 mm module, 20 teeth Varigears (Spur Gears with symmetrically varying toothwidth around the circumference Figure 1) made with Case-carburised and Tempered
SAE8620H steel [1], at 3000 rpm speed using OEP-80 Oil at 50qC as lubricant. Typical
contact stresses for different tooth-width and torque levels are shown in Table 1.

Figure 1. Photograph of a VariGear

100

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

Table 1. Typical contact stresses for different tooth width and torque levels

Tooth
Width,
mm

Contact stress
at 1300 Nm

Contact stress
at 1900 Nm

MPa

MPa

28

1301

1517

25

1352

1581

23

1404

1645

21

1457

1711

19

1506

1773

18

1559

1848

The high frequency MBN measurements were made with commercially available MBN
system and the Gear sensor containing ferrite-cored electromagnet and ferrite cored
MBN pick-up coil supplied by Stresstech, Finland. The measurements were made at a
magnetic excitation frequency of 125 Hz. Normally the MBN signals from a ferrite
cored pick-up coil are generated in a wide frequency bandwidth ranging from
excitation frequency to 1 MHz. Generally, avoiding the interference from higher
harmonic contents of the excitation frequency and neglecting very small amplitude
signals at high frequency end, the MBN signal is acquired in a appropriate frequency
bandwidth. In this study, the MBN signal were acquired and analysed within the
frequency range of 70-200 kHz considering the dominance of contact fatigue damage
close to the surface. The total RMS voltage of the MBN signal multiplied by 200
represented as MBN Display M value and the average MBN signal profiles are used
for analysis.

3. Results and Discussions


The variations in MBN levels measured on six different teeth having different width
and hence different contact stresses are shown in Figure 2 with progressive number of
cycles at 1900 Nm and 1300 Nm Torque levels respectively. It can be observed that the
MBN level increases rapidly in the initial stage and then slowly with progressive
number of cycles. At higher stresses >1773 MPa, the MBN level reaches a maximum
and then decreases in the later stages of fatigue life. With increase in number of fatigue
cycles, the cyclic contact plastic deformation results in progressive micro-pitting of the
gear surface as shown in Figure 3(a-c). The micro-pitting initiates micro-cracks on the
surface against the sliding direction of the gear which grows into macro-crack below
the surface along the rolling direction of the gear as shown in Figure 3(d). This results
in macro-pit or failure of the gear teeth. This mechanism is influenced by other

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

101

complex gear running conditions such as changing contact stress distribution,


lubrication condition, temperature, maximum bending stress, material quality and
micro-geometry of the gear. This unpredictable failure event is evident from Figure 4
which shows only micro-pits on one tooth and a macro-pit above the pitch line on the
adjacent tooth. This suggests that the catastrophic failure of the gear can occur without
any prior warning.

MBN Display 'M' Value, A.U.

180
160
140

Tooth Width, mm
Contact Stress, MPa

120

T1-28mm-1517
T2-25mm-1581

Gear- 4S
Torque = 1900Nm
MBN GAIN = 60

100
80

T3-23mm-1645
T4-21mm-1711
T5-19mm-1773

(a)

T6-18mm-1848

60
0

80

20
30
Number of million cycles

70
65

50

40

Gear- 5S
Torque = 1300Nm
MBN GAIN = 30
(b)

75
MBN Display 'M' Value, A.U

10

60

Tooth Width, mm
Contact Stress, MPa

55

T11-28mm-1301

50

T12-25mm-1352

45

T13-23mm-1404

40

T14-21mm-1457
T15-19mm-1506

35

T16-18mm-1559

30
0

10

20

30
40
Number of million cycles

50

60

70

Figure 2. The variations in MBN level with progressive number of cycles at (a) 1900
Nm and (b) 1300 Nm on six different teeth having different widths and contact stresses.
Note the MBN signal amplification gain for (a) is 60 while it is 30 for (b).

102

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

Figure 3. The photographs of the flank of 18 mm width tooth of the gear- 4S showing
the progressive micro-pitting (Grey strain area) after (a) 8m, (b) 20m and (c) 34m
cycles at 1900 Nm and (d) micro-graph showing typical crack growth below the gear
flank surface.

Figure 4. The photograph of gear showing only micro-pitting on one tooth and a large
macro-pit failure on the adjacent tooth.

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

103

Normally, a gear is considered as failed either when there is loss of 20m thickness of
gear flank profile caused by micro-pitting or when there is greater than 4% area of
macro-pits on the gear. However progressive contact fatigue damage has not been
directly related to changes in metallurgical properties such as hardness or
microstructure, even though, they may be affected and causes changes in MBN level.
Hence, this study attempts to monitor the changes in MBN level with progressive
contact fatigue. The variations in the MBN level with progressive number of cycles at
lower contact stresses (in the range of 1301 to 1559 MPa) clearly suggest that the initial
increase in MBN level is also very gradual and systematic in response to stress level
with progressive number of cycles. This suggests that the contact fatigue damage can
be detected in this type of gear steels at very early stage using the MBN technique well
before the development of any visible surface damage. Since the nature of gear failure
is highly unpredictable and catastrophic in many cases, the transition of MBN level
from a maximum may be considered as an indication of the critical stage beyond which
the failure can occur at any time depending on other unknown parameters.
The high frequency MBN profiles measured at the dedendum of 28 mm and 18 mm
width teeth at different number of cycles are shown in Figure 5. It can be observed that
the high frequency MBN level increases with number of cycles in 28 mm width tooth
whereas in 18 mm width tooth, the MBN level reaches a maximum after 8 m cycles
and then decreases with progressive number of cycles. Also, in both cases, initially the
MBN peak position shifts to lower applied voltage (magnetic field) and then again shift
back to slight higher field.
The increase in MBN with shifting of peak to lower field could be due to the combined
effects of deformation induced transformation of retained austenite into martensite and
fatigue softening. The carburised gear surface contains about 25% volume fraction of
retained austenite phases which is a paramagnetic meta-stable phase. During running,
the gear surface is subjected to micro-plastic deformation which transforms the metastable retained austenite into martensite. Since, the transformed martensite is a
ferromagnetic phase, it results in nucleation and movement of additional magnetic
domain walls which effectively contribute to increase in MBN level. The initially hard
tempered martensite poses high density of tangled dislocations. The cyclic deformation
could cause this tangled dislocation to rearrange into dislocation cells. The domain
walls can move with relatively less restraint within the dislocation cells compared to
tangled dislocation structure. This could contribute to increased MBN level with peak
shifting to lower field.
The reduction in MBN from maximum level and the shifting of peak position back to
slightly higher field could be due to the effects of cyclic plastic deformation induced
formation of more compressive residual stresses and cyclic hardening of the
microstructure. After certain number of cycles at a given stress level, the
transformation of retained austenite to martensite would be completed. Normally, the
number of fatigue cycles required for complete transformation of meta-stable retained
austenite increases with decrease in contact stress level. This also depends on other
factors like temperature, lubrication condition etc. Once the phase transformation is
complete, the continued cyclic plastic deformation of near-surface could generate
additional dislocations which could accumulate around the dislocation cell walls and
that results in the reduction of size of dislocation cell interior. This could cause
decrease in the displacement length of magnetic domain walls and stronger pinning of

104

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

domain walls at the dislocation cell walls. This could contribute to the decrease in the
MBN level from maximum and shifting of the MBN peak back to higher field.
The effect of contact stress level on the rate of changes in the material behaviour
introduced by these deformation processes is clearly reflected in the MBN profiles
(Figure 5). At lower stresses, the effect of transformation of retained austenite into
martensite would dominate and cause gradual increase in the MBN level. However, at
higher contact stresses, the phase transformation would complete after smaller number
of cycles. With progressive number of cycles, the cyclic-plastic deformation and the
residual stress alteration would have dominant effect. Hence, the MBN level would
increase to a maximum after certain number of cycles and then would decrease with
progressive number of cycles as observed in Figure 5.

Figure 5. The high frequency MBN profiles measured at the dedendum of (a) 28 mm
and (b) 18 mm width teeth at different number of cycles.

M. Vaidhianathasamy et al. / Evaluation of Contact Fatigue Damage on Gears

105

Even though this explanation seems to be appropriate, very large (> 100%) increase in
MBN level compared to smaller volume fraction (~25%) of retained austenite indicates
that some predominant form of magnetic softening taking place in this steel by
complex dislocation substructure evolution with progressive contact fatigue
deformation. It requires other metallurgical investigations to answer this anomalous
and surprise MBN behaviour.

4. Conclusions
This study shows that the MBN technique can be used to detect the changes in the Gear
caused by contact fatigue damage even at very early stages of fatigue life. The effect of
variation in the contact stress level is also clearly reflected by gradual and systematic
changes in the MBN level and its rate of change with progressive number of fatigue
cycles. The overall response of the MBN suggests that the MBN level increases to
reach a maximum level and then slowly decreases with progressive number of fatigue
cycles. This is attributed to the initial deformation induced transformation of retained
austenite to martensite and fatigue softening followed by cyclic plastic deformation
induced hardening.
References
[1]
[2]
[3]

T.R. Hyde, J.T. Evans, B.A. Shaw, Effect of stress and heat treatment on magnetic Barkhausen
emission in case-carburised steels, Mater. Eval., 58(8)(2000) 985-990.
L.P.Karjalainen, M.Moilanen, Fatigue softening and hardening in mild steel detected from Barkhausen
Noise, IEEE Trans. Magn., MAG-16 (3) (1980) 514-517.
V.Moorthy, B.K.Choudhary, S.Vaidyanathan, T.Jayakumar, K.Bhanu Sankara Rao, Baldev Raj, An
assessment of low cycle fatigue damage using magnetic Barkhausen emission in 9Cr-1Mo ferritic steel.,
Int J. Fatigue, 21 (1999) 263-269.

This page intentionally left blank

Inverse Problems

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-109

109

3D Reconstruction of flaws in metallic


materials by eddy currents inspections
Alessandro PIRANIa, Marco RICCIa, Antonello TAMBURRINOb, Salvatore
VENTREb,1
a
Polo Scientifico Didattico di Terni, Universit degli Studi di Perugia, Via Pentima
Bassa 21, Terni 05100, Italy
b
Ass. EURATOM/ENEA/CREATE,DAEIMI, Universit degli Studi di Cassino, Via G.
Di Biasio 43, Cassino 03043, Italy

Abstract. This work is focused on the image reconstruction of defects in metallic


materials using time-harmonic eddy currents testing measurements. The proposed
approach combines an efficient numerical model of the probe-defect interaction
and a least square error regularized procedure for solving the inverse problem. The
method has been successfully tested to detect multiple volumetric defects on
multiple layers of a planar conductive material. The regularization term is the total
variation, selected to encode the a priori information that defects are supposed to
be blocky. In this manner, the method is potentially capable of treating on the
same foot several defects type such as EDM, fatigue crack, stress corrosion crack
etc.
Keywords. Eddy currents, nondestructive evaluation, total variation regularization,
numerical simulation.

1. Introduction
Eddy currents non-destructive techniques are widely used in the field of
electromagnetic inspection of metallic structures to determine their composition and to
detect possible flaws. In this framework, a main issue is the ability to assess the 3D
conductivity profile of the sample under test. The major difficulty encountered to
achieve this aim is the non-linearity and ill-posedness of eddy currents inverse
scattering models. In the present work, we tackle this issue by combining an efficient
numerical model of the probe-defect interaction and a total variation procedure for
solving the inverse problem and reconstructing the 3D conductivity profile of flawed
samples. Our approach exploits a multi-position and a multi-frequency measurement
system and it has been tested on synthetic noisy data as preliminary proof in the
perspective of an experimental application to real defects.
Here we assume that the defects are volumetric with a finite resistivity. The
modeling of volumetric defects, as opposed to the modeling of zero-thickness defects,
is more intriguing and is receiving increasing attention [1][7]. Here we present an
efficient integral numerical model where the unknown is 'J, the eddy current variation
due to the presence of defects. The methods is in the same line of the approaches
1
Corresponding Author: Salvatore Ventre, Universit degli Studi di Cassino, Via G. Di Biasio 43, Cassino
03043, Italy; E-mail: ventre@unicas.it

110

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

presented in [1] and [8] where perfectly insulating volumetric and zero-thickness
defects are treated. The numerical calculation of 'J combines several advantages,
among them: (i) requires a local discretization in a neighborhood of the defect only, (ii)
the integral formulation involves the static free-space Green function.
The inverse problem in cast in term of minimization of the discrepancy, that is the
least square error between the experimental and numerical computed data. Due to the
intrinsic ill-posedness of the problem, we include a proper penalty term encoding a
priori information about the defects. Our choice is fall on the Total Variation (TV)
regularization term that penalizes functions with great spatial variability with respect to
blocky functions, while preserving the edges [9-11]. The TV has been chosen
because it accounts for the a priori information that the anomalies are blocky.
The paper is structured as follows: the finite element method used to solve the
forward electromagnetic problem is presented in Section 2, the inversion algorithm is
presented in Section 3 and numerical results are presented in Section 4.

2. Forward Problem
The numerical method here proposed has been developed in order to treat efficiently
the typical situation encountered in eddy current testing where anomalies are present in
an a priori known sub-region VT of a conducting domain VC. The anomalies perturb
locally the induced eddy current J and, therefore, it is very efficient [1, 8] to assume as
unknown 'J=J-J0 where J0 is the eddy current when the material is flawless.
Specifically, by combining the electric field expressed in terms of the vector and scalar
potentials with the constitutive relationship we obtain:

KJ
K0J 0

 jZA>J @  M  jZA ext , in VC

(1)

 jZA 0 >J 0 @  M 0  jZA ext , in VC

where J, J 0 L2div VC , J
A : v r o

P0
4S

J0

0 in VC J n

v r '

r  r' dV ,

0, J 0 n

0 on wVC ,

(2)

VC

K is the electrical resistivity of the conductor (in the presence of the defect), Aext is the
vector potential produced by the inducing coil, M is the scalar potential, the subscript 0
refers to the configuration in the absence of defects (defects are modeled as a
perturbation 'K of the electrical resistivity, i.e. K=K0+'K). From (1) it follows the
equation for 'J:
K'J  jZA>'J @ 'KJ 0  'M , in VC

(3)

where 'M=M-M0.
The related numerical model is based on [12]. Specifically, the unknown is represented
as 'J kAE 'I k u Tk where Tk is an edge-element shape function, and AE is a

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

111

proper subset of the edges of the graph made by nodes and edges of the finite element
mesh. Choosing properly this subset it is possible to impose the gauge and boundary
condition. Moreover, since 'J is localized in a neighborhood of the defect, it is
possible to use a local discretization of the conductive material thus achieving a great
reduction in the computational cost [1, 8].
The numerical model is, finally, obtained by imposing (3) in weak form:

VCLOC

u N k (K'J  jZ A>'J @)dV

VCLOC

u N k 'KJ 0dV

N k .

(4)

In (4) VCLOC is a local region in a neighborhood of the tentative region VT

VCLOC VC where 'J is not vanishing, and we exploited that the term involving
'M gives no contribution thanks to the solenoidality of the test functions and their
vanishing normal component on the boundary. The corresponding linear system is:
T

Z 0  'R 'I

'RI 0

(5)

where, Z0=R0+jZL, 'I is the column vector of the coefficients representing the
expansion of 'J and
L ij

'R ij

P0
4S

u N i ( x) u N j ( x' )

x  x'

VCLOC VCLOC

uN

uN

(7)

K 0 u N j dV

(8)

J 0 dV

(9)

VCLOC

I 0,k

uN

(6)

'K u N j dV

VCLOC

R 0,ij

dV dV '

VCLOC

Once the induced current density perturbation 'J has been computed, it is possible to
compute the related voltage induced on the coil. Finally, it is worth noting that the
unperturbed current density J0 can be computed analytically for canonical geometries
or numerically for more complicated configurations.

3. Inverse Problem
We assume that the measurement consists of the value of time-harmonic measurements
of the (complex) voltages induced on the exciting coil due to the presence of the
anomalies. Specifically, the data is the column array V * consisting of measurements
collected at different locations and frequencies (we simultaneously process all of them
to retrieve the overall 3D conductivity profile).

112

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

The proposed inversion method cast the problem in terms of minimization. Specifically,
the solution of the inverse problem is the minimizer of
E x

V *  V x  DTV x

(10)

where x is the column vector containing the parameters describing the conductivity
V=K-1 (here assumed uniform in each element of the finite element mesh) V * and
V x are the measured and numerically computed voltages variation due to the
presence of anomalies, D is the regularization parameter and TV is the total variation
regularization term [9-11] defined as
TV x

VT

V dV

(11)

where V is the conductivity represented by x.


The total variation penalty term, introduced as regularization term, penalizes function
with great spatial variability but preserving the edges [9-11] and, therefore, it is
indicated when the unknown is blocky as in the present problem.
The implemented inversion algorithm update iteratively the current solution x(n) at
step n with the following rule:
x n 1

x n  'x n

(12)

where 'x n minimizes


E n 1 'x

V *  V x n  S n 'x  DTV x n  'x

(13)

S n being the sensitivity map arising from the linearization of V x in a neighborhood

of x(n). The elements of the sensitivity matrix S n represent the derivatives of the
probe voltage with respect to the voxels conductivity. In the present work we compute
the sensitivity by using the method described in [13], that results in a first-order Born
approximation of the electric field inside voxels [14].
A closed expression for 'x n that minimize E n  1 is not available, thus for each
iteration n of the overall inversion algorithm, a further iterative minimization procedure,
labeled henceforth by the index Q must be carried out. In the present work we employ
the Lagged Diffusivity Fixed Point Iteration Method introduced by Vogel and Oman
[10, 11].
The choice of a suitable value for the regularization parameterD is of utmost
importance. We applied the L-curve method for choosing the regularization parameter
(see figure 1(left)). Specifically, defining Deq as the value that balances the LSE and the
TV regularization term, we found that the point of highest curvature, giving the value
of the regularization parameter, is close to 10-3 uDeq.

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

113

Figure 1. Left: L-curve analogous for the TV regularization. Right: Typical graph of the TV error
vs. the TV parameters. Both graphs refer to defect type I.

We have also analyzed the behaviour of E n  1 when D and Q are varied, setting 'x to
the value that minimizes the overall error in the spanned range of parameters D and Q.
Figure 1 (right) show the typical error trend where the solid curve identifies the value
D* that minimizes E n  1 . We found that D* is resulted to be very close to the value
provided of the regularization parameter provided by the L-curve method. Figure 1
refers to defect type I (see figure 3). Similar results have been attained for a wide
variety of numerical cases with different defect types and noise levels.
Since the data are provided by a multi-frequency measurement system, we have
properly weighted each frequency in order to gain useful information from all of them.
To accomplish this aim, we operated a frequency dependent normalization of V and S.
Usually the weights are related to the energy of the signal at the corresponding
frequency, on the contrary we carry out a normalization based on a SVD analysis of the
sensitivity matrix at single frequencies. This normalization is modified to take into
account the presence of additive synthetic noise: in this case the weights are determined
both from the SVD analysis and from the data noise variance.
Finally, to avoid instabilities in the reconstruction, as well as to enhance the
convergence speed, we introduce a priori information about the conductivity range of
values. Specifically, when in a given element the conductivity (represented by x(n)) is
greater than the value of the host material, we set it to the value of the host material. A
similar processing is carried out when the conductivity assumes negative values.

4. Numerical Results
In this section we consider an Aluminium plate (thickness 3mm, V=37.7u106S/m) and
we focus on the conductivity retrieval in a small region (VT) of 12mmu12mmu3mm,
modelled with a 3D regular grid of cubic voxels having the edge of 1mm. Defects are
approximated by a well defined number of voxels with fixed conductivity. The probe
consists of a coil with the following characteristics: inner radius 0.5mm, outer radius
2mm, height 3mm, 200 turns, lift-off 0.5mm. In the procedure the probe is placed in 25
different positions, whose centres build up a regular grid of points oriented at 45
degrees respect to the x axis and equally spaced at a distance of about 2mm that

114

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

corresponds to the coil outer radius, guaranteeing data independence and optimal
covering of the voxel region. The multi-frequency excitation is contextually exploited
in order to inspect the specimen with varying resolution and to examine its inner
structure at different depths. Since any frequency is associated to a specific skin depth
penetration G, we adopt three different values (670, 1500, 6000 Hz) whose skin-depths
correspond to the layer depths.
In eddy-currents applications the coil voltage variation due to a defect may be a
very small part (0.01%- 0.0001%) of the absolute coil voltage. Since the eddy currents
intensity is proportional to the excitation frequency, as the inspection depth increases,
i.e. the operation frequency decreases, the relative voltage variation decreases and the
SNR tends to be determined mainly by the unavoidable limitations of the measuring
apparatuses, usually determined from its operating range. In order to be consistent with
these considerations, we have introduced a synthetic noise, in the numerically
computed voltages vector, represented by an uncertainness circle in the voltage
complex plane which radius is proportional to the coil voltage in air, Vair. Henceforth
*
*
V * VNF
 n , where VNF
(n) is the noise free data array (noise array) and noise level
stands for the ratio between the uncertainness circle radius and the magnitude of Vair.
To gain insight about the actual noise magnitude compared to the defect signal
contribution, Table 1 reports some indicative values concerning defect types I and II for
the noise level adopted (noise level = 0.0005 % ). Specifically, for a given frequency,
*
*
signal level is the ratio between the L1-norm of the voltage variation GV * VNF
 VBG
due to the defect and the coil voltage in air, whereas relative noise is the ratio between
*
the L2-norm of the noise n and GV * . Here VBG
is the so-called background voltage,
that is the coil voltage when defects are not present.
Figure 2(left) shows the trends of the total normalized error LSE versus the overall
iteration index n, having rescaled to one the value at n=1. As expected, the LSE tends
to decrease as the iterations go on. Usually, in order to stop the overall inverse
procedure, a proper LSE error threshold, related to the value of the actual precision of
the experimental measurement, is prescribed. For noisy data, the error asymptotic value
mainly depends on the chosen noise level, and then the threshold has been selected
once a time. In figure 2(right) we report a plot of D* vs. Deq for different defect types
and noise levels: we note that D* always belongs to a neighbourhood of 10-3uDeq.
Moreover, as the noise level increases, the parameter Deq tends to assume a constant
value for all tests.
Figure 3 shows some reconstructed conductivity maps with and without additive
noise. The contribution of a deep defect to the total error is weaker than the one of a
superficial defect and, at a fixed depth of inspection, high-conductivity defect is harder
to reconstruct than a low-conductivity one. Nevertheless the results show that the
conductivity profiles are satisfactory assessed in few iterations.
Table 1. Signal level and Relative Noise

Signal level
Freq.[Hz]
670
1500
6000

Defect Type I
0.00573 %
0.0192 %
0.0995 %

Defect Type II
0.00279 %
0.00830 %
0.0154 %

Relative noise
Defect Type I
12.8 %
3.2 %
0.8 %

Defect Type II
25.4 %
7.1 %
4.5 %

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

115

Figure 2. Left: typical trends of the error LSE vs. the iterations, for no-noise and noisy data: the square
signs and the circle ones refer to the defect type I and type II respectively (see also Figure 3). The noisy data
are obtained with a noise level equal to 5u10-6. Right: plot of D vs. Deq for different defect types and noise
levels. The marks refer to no-noise tests (squares), noise levels 10-6 (triangles), 5u10-6 (stars) and 10-5
(circles).

Figure 3. Examples of reconstructed conductivity map for two defect types: 1st column reports the
actual maps; 2nd and 3rd columns report the maps obtained by processing no-noisy data for a threshold fixed
at 10-4 and 10-6 respectively; in the 4th column we show the maps obtained at the stop iteration (6th and 4th)
by processing noisy data with a noise level of 0.0005 %. The stopping iterations refers to a threshold fixed at
5u10-4 and 5u10-3 respectively.

116

A. Pirani et al. / 3D Reconstruction of Flaws in Metallic Materials by Eddy Currents Inspections

5. Conclusions
In this work we have addressed the imaging of 3D volumetric defects by eddy
current testing. Specifically, we have proposed an efficient integral numerical
formulation where the unknown is 'J, the eddy current variation due to the presence of
defects. The method combine several advantages: (i) it requires a local discretization in
a neighborhood of the defect only and (ii) the integral formulation involves the static
free-space Green function. The inverse problem, on the other hand, has been cast in
term of minimization of the LSE plus a Total Variation regularization term. The Total
Variation term accounts for the a priori information that the defect are piecewise
constant.

Acknowledgements
This work was supported in part by the Italian Ministry of University (MIUR) under a
Program for the Development of Research of National Interest (PRIN grant #
2004095237) and in part by the CREATE consortium, Italy.

References
[1]

[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]

[13]

[14]
[15]

M. Morozov, G.Rubinacci, A.Tamburrino, S.Ventre, Numerical Models of Volumetric Insulating


Cracks in Eddy-Current Testing With Experimental Validation, IEEE Trans. Mag., Vol. 42, No. 5,
May 2006, pp. 1568-1576.
J. R. Bowler, S. A. Jenkins, L. D. Sabbagh, and H. A. Sabbagh, Eddy current probe impedance due to
a volumetric flaw, J. Appl. Phys., vol. 70, no. 3, pp. 11071114, 1991.
Z. Badics, J. Pavo, H. Komatsu, Y. Matsumoto, and K. Aoki, Fast flaw reconstruction from 3d eddy
current data, IEEE Trans. Magn., vol. 34, no. 5, pp. 28232828, Sep. 1998.
Z. Badics, J. Pavo, Y. Matsumoto, and H. Komatsu, Forward solution speed-up for 3D eddy current
inversion, IEEE Trans. Magn., vol. 36, no. 4, pp. 11241127, Jul. 2000.
D. Reis, M. Lambert, and D. Lesselier, Eddy-current evaluation of three-dimensional defects in a
metal plate, Inverse Problems, vol. 18, pp. 18571871, 2002.
D. Prmel and A. Baussard, Eddy-current evaluation of three-dimensional flaws in flat conductive
materials using a Bayesian approach, Inverse Problems, vol. 18, pp. 18731889, 2002.
Y. Li, L. Udpa, and S. Udpa, Three-dimensional defect reconstruction from eddy-current NDE signals
using a genetic local search algorithm, IEEE Trans. Magn., vol. 40, no. 2, pp. 410417, Mar. 2004.
R. Albanese, G. Rubinacci, F. Villone, An integral computational model for crack simulation and
detection via eddy currents, J. of Comp. Phys., Vol. 152, pp. 736-755 (1999).
L. I. Rudin, S. Osher , and E. Fatemi, Nonlinear total variation noise removal algorithms, Proc. of the
11th Annual International Conference the center for Nonlinear Studies, 60, (1992) 259-268.
C. R. Vogel and M. E. Oman, Iterative methods for total variation denoising, SIAM J. Sci. Comput. 17
(1), (1996) 227238.
C. R. Vogel, Computational Methods for Inverse Problems, SIAM: Frontiers in Applied Mathematics
23, (2002)
R. Albanese and G. Rubinacci, Finite element methods for the solution of 3D eddy current problems
in Advances in Imaging and Electron Physics, Peter W. Hawkes (ed.), vol. 102, (Academic Press), 1998,
pp. 1-86.
D. N. Dyck, D. A. Lowther, and E. M. Freeman, A Method of Computing the Sensitivity of
Electromagnetic Quantities to Changes in Materials and Sources, IEEE Trans. On Magnetics, 30, 5
(1994).
S. M. Nair and J. H. Rose, Reconstruction of three-dimensional conductivity variations from eddy
current (electromagnetic induction) data, Inverse Problems, 6, (1990) 1007-1030.
R. Fletcher and C. M. Reeves, Function minimization by conjugate gradients, The Computer Journal, 7
(2), (1964) 149-154.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-117

117

Multi-frequency Eddy Current Imaging


for the Detection of Buried Cracks in
Aeronautical Structures
Yohan LE DIRAISON and Pierre-Yves JOUBERT
SATIE, ENS Cachan, CNRS, UniverSud
61 av. du President Wilson, 94235 CACHAN Cedex, FRANCE
Abstract. The authors present a signal processing method dedicated to the detection of defects buried next to rivets in aeronautical lap joints. The method is based
on a multi-frequency principal component analysis and is applied to the images
provided by an original eddy current imager. The optimization of the method is
carried out thanks to an experimental approach, and validated with the detection
of buried defects, ranging from 2mm to 8mm long and 2mm to 8mm deep. An
extension of the method to a classication scheme is also considered.
Keywords. Eddy currents, magneto-optical imaging, multi-frequency, buried
defect detection, aeronautical riveted lap joints

Introduction
Magneto-optical and eddy current imagers (ECI) are very promising for the efcient
NDE of ageing aircrafts since they simultaneously reduce inspection time and enhance
characterization possibilities [1]. In this paper, the authors implement an original ECI [2]
and propose a buried defect detection method, based on the principal component analysis.
The efciency of the method is characterized and optimized through an experimental
multi-frequency approach, and its extension to a classication scheme is considered.

1. The Experimental Set-Up


The ECI used in this study was developed for the characterization of defects in aeronautical riveted lap joints and was presented in [2]. It is constituted of an EC inductor
designed to create a uniformly oriented EC ow in the inspected structure and a linear
magneto-optical (MO) set-up used to image the normal distribution of the magnetic eld
at the surface of the structure, thanks to a linear MO sensor lm used with a light modulation approach [3]. Finally, this device provides true in-phase and in-quadrature EC
images of a large inspection area (up to 76mm diameter), with a resolution of 100m
100m.
In this study the ECI was implemented for the inspection of a laboratory made riveted lap
joint mock-up, constituted of ve interchangeable aluminum plates featuring a thickness

118

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

of 2mm each and an electrical conductivity of 35M S.m1 . The defects are placed next
to the rivet holes at various depths. They are 0.5mm wide, 2mm high and their length
ranges from 2mm to 8mm (see Figure 1 for an example). EC images obtained in the
conguration depicted in Figure 1 are given in Figure 2 as an illustration.

Figure 1. Lap joint mock-up with a defect in the third Figure 2. EC images (600Hz) of Figure 1 conguraplate.
tion with a 6mm long defect.

2. Signal processing method


As seen in Figure 2, the EC signatures due to the buried defects are rather difcult to
analyze from the raw EC images, rstly because they are masked by the presence of the
rivets, and secondly because of the attenuation of the EC density with the penetration
depth in the material. Considering the rivets and the defects are two different sources
which are mixed to constitute the EC signal, the objective of the signal processing method
is to nd a more adequate representation space in which these sources can be separately
visualized. To that purpose, a method based on a principal component analysis (PCA) is
a good candidate [4,5,6].
2.1. Basic Principle
Let us assume that the transfer function of the device can be expressed by the following
linear expression:

s1



s2
(1)
M = T S = t1 t2 tr .
..
sr
where M is the measurement matrix, T is the unknown transfer matrix, and S is the
matrix of the sources. M is constituted of r lines of p elements. r is the number of
measurements used, gathering the in-phase and in-quadrature measurements obtained at
r
2 different frequencies. Each line is the lexicographical concatenation of the lines of the
measurement images and contains p elements. The dimensions of T is r r and the
dimensions of S is r p. Assuming the sources to be independent and centered, the
source covariance matrix SS t reads:

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

s21
0

SS t = .
..
0

0
s22
..
.
0

0
0

. . ..
. .
s2r

119

(2)

where s2i is the mean square (energy) of the source si .


Furthermore, assuming the row vectors of T to be orthogonal, T reads:

t1  0 0
0 t2  0

T = QT DT = QT .
.. . . ..
..
. .
.
0
0 tr 

with QT QtT = Ir

(3)

where QT is a rotation matrix.


Considering the assumptions expressed in Eqs. 2 and 3, the covariance matrix M M t
reads:
 2 2

t1  s1
0

0
t2 2 s22
0

t
t
t t
M M = T SS T = QT
(4)
QT
..
..
.
.
.
.

.
.
.
.
0
0
tr 2 s2
r

Besides, the singular value decomposition (SVD) of the matrix M reads:

1 0
0 2

M M t = V DV t = V . . .
.. .. . .

0
0
..
.

t
V

(5)

0 0 r
where V is the matrix of the singular vectors vi and D is the diagonal matrix of associated
singular values i , arranged by decreasing order.
Therefore, by a formal identication between Eqs. 5 and 4, one can note that matrix
QT and matrix V generate the same representation space. Therefore, sources S are then
estimated by V t M , so that:


s1
t1  0 0
s1
0 t2  0 s2 s2


V t M = (V t QT )(DT S) = .
.. . . .. .. = .. = S
..
. . . .
.
sr
0
0 tr 
sr

(6)

Moreover, the singular values i are relative to the mean square value (i.e. the energy) of
the estimated sources si .
i = s2 = ti 2 s2i
i

(7)

120

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

2.2. Implementation
In this study, the rivets and the defects are considered to be the only two sources of interest. Since the SVD sorts the singular values by decreasing order, the estimated sources
are expected to be constituted of s1 (rivets) featured by a high 1 (high energy of the
rivet signatures), and s2 (defects), featured by a smaller 2 (low energy of the defect
signatures). As an example, the source separation was carried out on the EC images of
Figure 2, and are shown in Figure 3. This example validates the signal processing approach, since it is clearly seen that estimated source s1 is relative to the presence of the
rivet, and estimated source s2 is relative to the presence of the defect, separated from the
rivet.
In order to characterize the efciency of the source separation for each rivet, a separation
ratio is dened as follows:
SRrivet =

s22
2
=
s21
1

(8)

This ratio is expected to be minimum for a sound rivet (equal to zero in ideal case), and
maximum for large defects.
In the example shown in Figure 3, the computation of this ratio for the sound rivet gives
SRsound = 0.03 and SRf lawed = 0.08 for the awed rivet.

Figure 3. Results of the signal processing method applied to images of Figure 2.

3. Frequency Optimization
The implementation of the signal processing approach requires the acceptance of the assumptions expressed above, which are not a priori obvious to assume. However, if the
assumptions are not entirely veried, it is known that the PCA allows a source rejection
- rather than an actual separation - to be achieved [5]. The rejection (of the rivet) is sufcient here as it is a posteriori veried by the obtained results. Moreover, the choice of the
excitation frequency is particularly important so that the linearity and the orthogonality
of the model can be approached (if not entirely fullled), and the defect detection can be
optimized. To this end, the ECI was implemented for the inspection of defects buried in
the 2nd , 3rd , and 4th plate, using excitation frequencies ranging from 200Hz to 4kHz.
Then, the separation ratio dened in Eq. 8 was computed for each rivet. Separation re-

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

121

sults obtained for that congurations are presented in Figures 4 to 6, and feature the optimum excitation frequencies gathered in Table 1.
The existence of these optimum frequencies can be explained as follows: a low frequency
implies a deep EC penetration and a non-linear mixing of the rivet and defect sources;
conversely, a high frequency implies a reduced EC penetration depth, and therefore a
poor buried defect detection. However, an intermediate frequency features a limited EC
penetration, so that the presence of the defect hardly modies the EC signature of the
rivet. For this particular frequency a linear model can be approached, and the separation is more efcient. Furthermore, one can note that for each plate, the optimum excitation frequency roughly corresponds to a phase-shift = 90 between the surface
EC density and the EC density at the depth d of the defect. Indeed, considering that the
excitation is a plane wave, can be expressed as [7]:

= d f
(9)
where f the frequency, is permittivity and the conductivity of the material. This
feature is consistent with the assumption of orthogonality of the transfer matrix of the
ECI, required for an efcient source separation.

Figure 4. Representation of the separation ratio Figure 5. Representation of the separation ratio
against the frequency, for sound and awed rivets in against the frequency, for sound and awed rivets in
the 2nd plate.
the 3rd plate.

Figure 6. Representation of the separation ratio


against the frequency, for sound and awed rivets in
the 4th plate.

122

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

Table 1. Optimum separation frequencies in each plate, and corresponding penetration depth relative to a
90phase-shift of the EC density.
Plate

Optimum frequency (Hz)

Depth for = 90 (mm)

2nd (2mm 4mm)

f2 = 1200

3.9

3rd (4mm 6mm)


4th (6mm 8mm)

f3 = 600
f4 = 300

5.5
7.7

4. Multi-Frequency Detection
The determination of the three optimum frequencies presented in Section 3 allows an
efcient multi-frequency detection method to be implemented for unknown congurations. Indeed, let us consider the multi-frequency measurement matrix M constituted of
 q ) at the three frequencies so
in-phase EC signals (m
 p ) and in-quadrature EC signals(m
that:

m
 p (f2 )

m
 q (f2 )

m
(f
)

p 3
M =
(10)

m
(f
)

q 3
m
 p (f4 )
m
 q (f4 )
The PCA of M leads to estimated source matrix S in which s1 is relative to the rivets, s2 is relative to the defects (whatever its length or depth), and the other sources are
undened. Indeed, for each considerated rivet conguration, the implementation of this
multi-frequency detection algorithm leads to the same orientation of singular vector v2 .
Therefore, the best defect detection will always be obtained on estimated source s2 .
As an example, EC signals obtained at frequencies f2 , f3 and f4 for a four rivet conguration, summarized in Table 2, are presented in Figure 7 to 9, and the results of the multifrequency PCA are given in Figures 10. One can note the good defect results obtained on
source s2 . Furthermore, the computation of the scalar products between singular vectors
v2 single obtained with the SVD for each rivet and v2 total obtained with the SVD for
the four rivet conguration (presented in Table 2), shows that these vectors are collinear.
This feature validates the multi-frequency detection approach.
Table 2. Four rivet conguration and corresponding scalar product between singular vectors v2 .
Rivet 1

Rivet 2

Rivet 3

Rivet 4

Defect length
Defect plate

4mm
2

none
none

8mm
3

6mm
4

v2 single v2 total

0.990

0.994

0.996

0.984

5. Towards Classication of Buried Defect


For a further characterization of the defects according to their length and depth, we consider the mono-frequency detection carried out in each plate using the optimized fre-

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

123

Figure 7. EC signals obtained at 1200Hz for the in- Figure 8. EC signals obtained at 600Hz for the inspection of the four rivet conguration of Table 2.
spection of the four rivet conguration Table 2.

Figure 9. EC signals obtained at 300Hz for the in- Figure 10. Results of the PCA for the inspection of
spection of the four rivet conguration Table 2.
the four rivet conguration Table 2.

quencies f2 , f3 and f4 . For each frequency, a normalized separation ratio was dened
as:
N SR =

SRf lawed
SRsound

(11)

The N SR was computed and plotted against the defect length, as presented in Figure 11 for frequency f2 , Figure 12 for f3 , and Figure 11 for f4 . The normalized separation ratio is greater than 1 for every considered defect, i.e. all the defects can be detected
when using f2 in plate 2, f3 in plate 3 and f4 in plate 4 (except for the defects shorter
than 3mm in the 4th plate, at f4 = 300Hz). Moreover, the ratio is monotonously increasing with the length of the defect, which enables the classication of the defects according to their length in a considered plate. However, ambiguities may appear between
large and deeply buried defects and small surface defects. To overcome this drawback,
the detection procedure should be successively carried out at frequencies f2 , f3 and f4
and the obtained results should be correlated to build a defect classication according to
length and depth.

6. Conclusion
The proposed multi-frequency signal processing approach shows great efciency for the
detection of buried defects (down to 3mm long and 6-8mm deep), placed next to the rivets in multi-layered riveted structures, and is very promising for classication purposes.
Further works will focus on the implementation of the automatic classication algorithm
using for example a maximum likelihood approach. Also, the method will be extended
to a 2D processing for the enhanced characterization of buried defects of any orientation.

124

Y. Le Diraison and P.-Y. Joubert / Multi-Frequency Eddy Current Imaging

Figure 11. Evolution of the normalized separation ra- Figure 12. Evolution of the normalized separation ratio against the length of the defects at f2 = 1200Hz. tio against the length of the defects at f3 = 600Hz.

Figure 13. Evolution of the normalized separation ratio against the length of the defects at f4 = 300Hz.

References
[1] P. Joubert, J. Pinassaud, G. Rubinacci, A. Tamburrino, S. Ventre, Numerical modeling of a continuous
level Eddy Current Imager, in E NDE, Electromagnetic Non-destructive Evaluation (X), S. Takahashi
and H. Kikuchi (Eds.), pp. 33-40, IOS Press, 2007.
[2] P.-Y. Joubert, J. Pinassaud, Linear magneto-optic imager for non-destructive evaluation, Sensors and
Actuators A: Physical, Volume 129, Issues 1-2, 24 May 2006, Pages 126-130.
[3] R. Grechishkin, S. Chigirinski, M. gusev, O. Cugat, N. Dempsey, Magnetic Imaging Films, Sringer, 2007
[4] P.-Y. Joubert, Y. Le Diraison and J. Pinassaud, Eddy Current Imager for the Detection of Buried Flaws
in Large Metallic Structures, In proceedings of 9th Conference on NDT, Berlin, September 25-26, 2006.
[5] G. Saporta, Thories et mthodes de la statistique, Institut Franais du Ptrole, Technip. br. in 8. 386 p,
1978. [in French]
[6] K.V. Mardia, J.T. Kent and J.M. Bibby, Multivariate Analysis, Academic Press, 1979.
[7] H.L. Libby, Introduction to Non-destructive Test Methods, Robert Kriegger Publiesher Compagny, NY
1979.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-125

125

A Comparative Study of Source Separation


Techniques for the Detection of Buried
Defects in the EC NDE of Aeronautical
Multi-Layered Lap-Joints
Alan TASSIN, Yohan LE DIRAISON and Pierre-Yves JOUBERT
SATIE, ENS Cachan, CNRS, UniverSud,
61 Avenue du Prsident Wilson, 94235 CACHAN Cedex, France

Abstract. The authors present a multi-coil EC sensor dedicated to the rapid


inspection of aeronautical riveted lap joints, and compare the efficiency of two
signal processing methods based on principal component analysis and independent
component analysis, to enhance the detection of buried defects appearing next to
the rivets.
Keywords. Eddy current multi-sensor, buried defects, aeronautical riveted lapjoints, source separation techniques, principal component analysis, independent
component analysis.

Introduction
The detection of buried defects growing next to the rivets in aeronautical riveted lapjoints is a major preoccupation for the NDE of ageing aircrafts. NDE techniques should
be fast, since aircrafts feature a large amount of rivets, and reliable, since buried defects
are particularly difficult to detect. In this context, the authors propose a dedicated eddy
current (EC) sensor, associated to source separation techniques, for an enhanced
detection of buried defects.

1. Experimental Set-Up
In this study, an EC sensor was designed for the rapid inspection of riveted lapjoints. The sensor is constituted of an EC inductor featuring a large coil winded within
a magnetic cup-core, and 2 pairs of pickup coils, connected in series and in phase
opposition, so that a differential measurement can be carried out along the x-axis and
the y-axis (Figure 1). Each pair of coils is constituted of two flat 8-layer PCB coils and
is placed under the exciting coil so that no EC signal is provided in absence of defect
when the sensor is placed above the inspected rivet (Figure 1). This specific
configuration enables the inspection of each rivet to be achieved in a single EC
acquisition.

126

A. Tassin et al. / A Comparative Study of Source Separation Techniques

In this study, the sensor was implemented for the inspection of a laboratory made
riveted lap-joint mockup, constituted of five aluminum based alloy plates featuring a
35MS.m-1 electrical conductivity, a 2mm thickness, and various defects placed next to
the rivets in the 2nd or 3rd plates (Figure 1). The defects are rectangular notches oriented
along the main direction of the lap-joint (x-axis). They feature a 0.5mm width, a 2mm
height, and their length ranges from 1mm to 10mm.
The EC data acquisitions were carried out thanks to a PC controlled impedance
analyzer HP4192A, and the sensor was moved by a PC controlled 3-axis robot, so that
the influence of the sensor mispositioning (lift-off and tilt) can be considered as
negligible.
y

Exciting coil
x

Pickup coils

Magnetic core
2mm

12mm
Plate

1
2
3
4
5

6mm

Defect length

40mm

Figure 1. Structure of the riveted lap-joint mockup and of the dedicated EC sensor.

PCA results

Raw EC signals

ICA results

-10

0.01

S12

In-phase

0.005
0

4
2

20

40

60

80

0.5

20

40

60

80

20

(mm)

(mm)
-9

S22

0.005

40

60

80

60

80

(mm)
-9

x 10

0.01

In-quadrture

x 10

S12

-0.005
-0.01

-9

x 10

x 10

S22

0
0.5

-0.005
-0.01

20

40

(mm)

60

80

0.5

20

40

(mm)

60

80

20

40

(mm)

Figure 2. Raw EC signals (1200Hz) obtained along the x-axis, PCA and ICA result components along the xaxis. Two averaged sound rivets centered at x = 20mm and x = 60mm (dotted lines). Flawed rivet centered at
x = 60mm (solid line). The defect is 5mm long and placed in the second plate.

A. Tassin et al. / A Comparative Study of Source Separation Techniques

127

The sensor was implemented in two different modes: the scanning mode (S-mode)
in which the EC signals are provided while the sensor is moving along the x-axis, and
the punctual mode (P-mode) in which the sensor is only operating above the inspected
rivet, for a rapid lap-joint inspection (Figure 1). Since only defects oriented along the
x-axis are considered in this study, only the pair of pickup coils placed along the x-axis
(Figure 1) is used.

2. Source Separation Techniques for the Enhanced Detection of Buried Defects


As shown in Figure 2, the EC signatures of the defects are rather difficult to analyze
from the raw signals, firstly because of the attenuation of the EC density with the depth
of the defect, and secondly because their EC signatures are masked by the rivets.
In order to enhance the detection of the defects, the authors consider two source
separation techniques: the principal component analysis (PCA) and the independent
component analysis (ICA). In both techniques, the EC signals are assumed to be
constituted of linear mixings of different sources, and the sensor modeling is therefore
expressed by:
G
G
m1 t11 t12 s1
M TS G
(1)
G
m2 t 21 t 22 s2
where the rows of M are the EC signals (in-phase and in-quadrature signals shown in
Figure 2), T is the unknown transfer function of the EC sensor, and the rows of S are
the two sources of interest (rivet and defect sources). Both techniques perform the
estimation of the sources S from the measurements M, but under different assumptions.
2.1. PCA Basic Principle
In the PCA method, the sources are assumed to be centred and uncorrelated, so that:

V 12 0
.

2
0 V 2

SS T

(2)

Furthermore, the column vectors of T are assumed to be orthogonal, which means that
T can be decomposed as the product of a rotation matrix R and a dilatation matrix D:
T

RD

cosT
sin T

 sin T d1

cosT 0

0
.
d 2

(3)

Under these assumptions, the variance-covariance matrix MMT is expressed by:

MM T

TSS T T T

d 2V 2
R 1 1
0

d 22V 22

RT .

Besides, the singular value decomposition of MMT leads to:

(4)

128

A. Tassin et al. / A Comparative Study of Source Separation Techniques

MM T

O
V 1
0

0 T
V
O2

(5)

where V is the matrix of the singular vectors of M associated to the singular values O1
and O2. According to Eqs. (4 and 5), the rotation matrix R can be identified to matrix V.
Finally, the PCA consists in the computation of VTM which leads to the estimation of
the sources S, since:
T

V M

V (T S )

(V R)( D S )

G
d1s1
d sG
2 2

S .

(6)

Since the vectors of R and T appear in an unknown order, the sources of S are rather
separated than estimated. Besides, the vectors of V are chosen to be arranged by
decreasing order according to their singular values [1]. Therefore, in this study, the
implementation of the PCA will carry out a source separation in which the first row 1
of is expected to be relative to the rivet (the rivet features a high contribution to the
EC signals) and the second row 2 is expected to be relative to the defect (the defects
feature a low contribution to the EC signals).
2.2. PCA Implementation
The efficiency of the source separation techniques is related to the acceptance of the
assumptions expressed in section 2.1. which are not obvious. However, if the
assumptions are not entirely verified, the PCA is known to carry out a good rejection of
the higher source (rivet) while estimating the source of lower energy (defect) [1,2],
which is quite satisfactory in the case of defect detection.
Furthermore, the excitation frequency can be optimized so that the assumptions
can be approached, if not entirely fulfilled [3]. In this study, frequency f2 = 1200Hz is
chosen for the detection of 2nd plate defects, and f3 = 600Hz for the detection of 3rd
plate defects, according to the frequency optimization presented in [3].
As an example, the PCA was implemented on the EC signals obtained at frequency
f2 for the inspection of a sound rivet and of a flawed rivet in plate 2. The raw EC
signals and PCA results are shown in Figure 2, in which one can note that the second
PCA component (2) allows the presence of the defect to be highlighted, over the
whole rivet signature (S-mode) or only over the central peak of the signature (P-mode).
In order to characterize the efficiency of the method, the authors define a defect
detection ratio U , expressed by:

S 2d

S 2s

flawed rivet

(7)
sound rivet

where the numerator is the norm of 2 computed over the signature of a flawed rivet (in
S-mode or P-mode), and the denominator is the norm of 2 computed over the signature
of a reference sound rivet (in S-mode or P-mode).

A. Tassin et al. / A Comparative Study of Source Separation Techniques

129

Amplitude of the detection ratio U for T = 90

As an illustration, the detection ratio was computed for the detection of 10 defects
buried in the 2nd plate, ranging from a 1mm length to a 9mm length, and inspected in Smode at f2 = 1200Hz. The obtained results are plotted in Figure 3, as a function of the
rotation angle T, for which T = 0 is relative to the direction of the first vector of V (rivet
source) and T = S/2 is relative to the direction of the 2nd vector of V (defect source).
One can note that the PCA provides a maximum detection ratio U = Umax > 1 for
T = S/2 (i.e., as expected, the defects appear on the second PCA component 2), and
that U | 1 for T = 0 (i.e. only the rivet source appears on the first PCA component 1).

2.5

9mm
7mm
5mm
4mm
3mm
1mm

2
1.5
1

0.5

0
-0.5
-1
-1.5
-2
-2.5
-3

-2

-1

Amplitude of the detection ratio U for T = 0

Figure 3. Defect detection ratio U (defined in Eq. 7) as a function of the projection angle T, in polar
coordinates (U : radius, T : angle). Acquisitions made at f2 = 1200Hz, defects placed in the 2nd plate and
lengths ranging from 1mm to 9mm.

2.3. ICA principle and implementation


The ICA method is based on an extension of the Central Limit Theorem [4], which
states that the distribution of the linear mixing of two independent sources is more
gaussian than the distribution of each source. The gaussianity of a signal s can be
estimated by the kurtosis criterion [4], expressed in Eq. (8), which tends to zero for a
gaussian signal:

kurt ( s)

E[(s  s ) 4 ]  3( E[ s  s 2 ]) 2 .

(8)

Therefore, the estimation of non-gaussian independent sources is achieved by finding


the linear combinations of the considered signals which maximize the absolute value of
the kurtosis [4]. However, the comparison of the kurtosis values of different signals is
only consistent for signals which are centred and of unity variance.
Therefore, the implementation of the ICA requires a previous whitening step [4] by
the means of a PCA of the variance-covariance matrix MMT of the measurements, and
expressed by:

130

A. Tassin et al. / A Comparative Study of Source Separation Techniques

~
M

O
V 1
0

O2

1 / 2

VTM

(9)

~
in which M are the whitened measurements.
The implementation of the ICA then consists in the research of the optimum rotation
angle T = Topt, so that:
kurtmax

cosT opt ~
kurt (
M )
sin T opt

(10)

Finally, the ICA estimation of the sources then reads:


S

cosT opt
sin T
opt

 sin T opt ~
M
cosT opt

(11)

In this method, the estimated sources are not likely to appear in a predefined order.
Therefore, the identification of the sources requires a calibration step obtained on
known flawed situations. In this study, the source relative to the rivet is chosen to be
placed in 1 and the source relative to the defect (which maximizes the detection ratio
defined in Eq. 7) is chosen to be placed on 2, for the sake of comparison with the PCA
results.
As an illustration, the ICA was implemented for the processing of the raw EC
signatures of a sound rivet and a rivet featuring a 5mm long defect buried in plate 2.
The results are compared to the raw EC signatures and to the PCA results in Figure 2.
One can note that 2 is mainly relative to the defect, and that 1 is mainly relative to the
rivet alone, in S-mode or P-mode.

3. Detection results using PCA and ICA

The PCA and ICA based defect detection methods were implemented using the EC
signals provided by the inspection of 10 flawed rivets in plate 2, 10 flawed rivets in
plate 3 and 10 sound rivets, in S-mode and in P-mode, at f2=1200Hz and f3=600Hz.
The PCA and ICA were carried out using each flawed rivet signature separately
(PCA2, ICA). The PCA was also carried out on a reference sound rivet signature
obtained by the averaging of 10 sound rivet signatures (PCA1), so that the
determination of the rivet source 1 is less influenced by the presence of the defect.
Then, the detection ratio U was computed as expressed in Eq. (7), in each detection case,
as a function of the defect length. The results obtained for defects in plate 2 are
presented in Figure 4, and defects in plate 3 in Figure 5. In each case, detection ratio
U = 0dB is relative to the defect detection method applied to the reference sound rivet.
In addition, a detection threshold was computed. This threshold corresponds to the

131

A. Tassin et al. / A Comparative Study of Source Separation Techniques

highest value of U z 0dB (wrongly) obtained for the inspection of sound rivet samples.
Above this threshold, defects are correctly detected without false alarm.
14

4.5
4

S-mode EC acquisitions
Defects in plate 2

PCA 1

12

PCA 2

P-mode EC acquisitions
Defects in plate 2

PCA 1

3.5
PCA 2

10
Detection ratio in dB

Detection ratio in dB

3
ICA
2.5
2
1.5

6
ICA
4

0
-0.5
0

Detection threshold

Detection threshold

0.5

4
5
6
Defect length in mm

-2
0

10

4
5
6
Defect length in mm

10

Figure 4. Detection ratio versus defect length for PCA and ICA based detections of defects buried in the 2nd
plate. Acquisition carried out at f2=1200Hz in S-mode (left) and P-mode (right).

The detection results obtained in plate 2 (Figure 4) show that defects longer than
1mm are detected without false alarm. However, one can note that the detection ratio
decreases for the longer defects. This feature can be attributed to a poor verification of
the linearity assumptions for long defects (i.e. a reduced efficiency of the source
separation methods) and to the distance between the 2 pickup coils (12mm) which is
not optimum for defects longer than 5mm. Finally, one can note that best detection
results are obtained with the PCA applied on P-mode EC signals, which enables a
higher rejection of the rivet signature.
1.8
1.6

12

S-mode EC acquisition
Defects in plate 3

PCA 1

1.4

10

P-mode EC acquisitions
Defects in plate 3

PCA 2

PCA 2

PCA 1

Detection ratio in dB

Detection ratio in dB

1.2
ICA

0.8
0.6
0.4

4
ICA

Detection threshold
2

Detection threshold
0

0.2
-2

0
-0.2
0

4
5
6
Defect length in mm

10

-4
0

4
5
6
Defect length in mm

10

Figure 5. Detection ratio versus defect length for PCA and ICA based detections of defects buried in the 3nd
plate. Acquisitions carried out at f3=600Hz in S-mode (left) and P-mode (right).

Same conclusions can be derived from the results obtained in plate 3 (Figure 5), except
that defects longer than 2mm can be detected without false alarm in S-mode, and longer
than 5mm in P-mode. The ICA is inadequate for the detection of defects in P-mode and
less efficient than PCAs in S-mode. These results can be explained by the high source
rejection possibilities provided by the PCA [1,2] when estimating the source of low

132

A. Tassin et al. / A Comparative Study of Source Separation Techniques

energy. This feature is not operative with the ICA because of the whitening step
expressed in Eq. (9). Also, one can note that the results obtained in PCA1 and PCA2
are quite similar: indeed, the directions of the first eigenvectors are quite identical in
both approaches, since the presence of the deeply buried defects hardly modifies the
EC signatures of the rivets.
Finally, the presented results exhibit some scattering of the data relative to small
defects and sound rivets, which leads to non-detection or false alarms. This feature can
be attributed to the poor accuracy of the machining of both the used mockup and sensor
array, and to the possible presence of air gaps between the lap joint plates, since they
are not fastened with real rivets.

Conclusion

In this paper the authors have presented and compared the implementation of two
source separation techniques (PCA and ICA) used for the detection enhancement of
buried defects growing next to the rivets. The methods were implemented on the EC
signals provided by a specific eddy current sensor dedicated to the rapid inspection of
riveted lap-joints. The sensor was used in scanning mode and in punctual mode, in the
case of buried defects placed in the 2nd or 3rd plate of a riveted lap-joint mockup. The
obtained results show that the sensor could be optimized so that all defect lengths can
be considered with equal detectability. Also, the PCA appears to be more efficient than
the ICA, thanks to its ability to reject the source of higher energy (rivet) when
estimating the source of lower energy (defect). Further works will focus on the
extension of the signal processing technique to 2D EC signals, in order to detect defects
of any orientation, and to reject the possible edge effect occurring in the inspection of
rivets placed near the edges of the lap-joint. Finally, a multi-frequency approach will be
developed in order to extend the detection technique to a classification scheme.

Acknowledgements

This work was supported by grants from Rgion Ile-de-France in the framework of the
competitiveness cluster SYSTEM@TIC PARIS-REGION (Digital Production project).

References
[1]
[2]
[3]
[4]

T. Jolliffe, Principal Component Analysis, Springer Verlag, New York, 2002


S. Hermosilla-Lara, P.-Y. Joubert, D. Placko, Identification of thermal and optical effects for the
detection of open-cracks in photothermal non destructive testing, Eur. J. Appl. Phys. 24, 223-229
(2003)
Y. Le Diraison, P.-Y. Joubert, Multi-frequency Eddy Current Imaging for the detection of buried cracks
in aeronautical structures, in proceedings of ENDE 2007, 19-21 June 2007, Cardiff, United Kingdom.
A. Hyvrinen, J. Karhunen, E. Oja, Independent Component Analysis, John Wiley and sons, 2001

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-133

133

Fundamental Feature Extraction Methods


for the Analysis of Eddy Current Data
Jeremy S. KNOPPa,1 and John C. ALDRINb
a

Air Force Research Laboratory, USA


b
Computational Tools ,USA

Abstract. Features are investigated in eddy current data that are sensitive to
corrosion and fatigue cracks in airframe structures while invariant to other NDE
noise factors. To investigate subsurface corrosion characterization at the faying
surface, a series of eddy current studies were performed using an analytical model
for varying total subsurface thickness loss and percentage of the thickness loss
occurring in each layer.
Results for the simulated studies are presented
demonstrating a novel feature for corrosion characterization using first and second
order derivatives of the impedance response with respect to frequency. For
characterization of subsurface cracks around fastener holes in structures,
numerical simulations and experimental studies are presented. Unique features in
the measurement response in circumferential direction were found to be sensitive
to subsurface cracks around fastener holes and invariant to irregular geometric
factors such as fastener fit and probe tilt. Multifrequency eddy current data
combined with circumferential (spatial) measurement features were found to be
promising for characterizing subsurface cracks in terms of length and depth.
Keywords: eddy current, feature extraction, material loss, cracks

Introduction
The detection and characterization of subsurface cracks and corrosion in multi-layer
airframe structures is a common practical problem and also a challenging research
problem in NDE. Eddy current and ultrasonic methods have been applied with some
success for detecting and quantifying damage. Reliable second layer crack detection
using eddy current is still in its infancy [1,2]. Dual frequency eddy current techniques
have been shown to be effective in detecting second layer corrosion even in cases
1
Air Force Research Laboratory (AFRL-MLLP), Wright-Patterson AFB, OH
Jeremy.Knopp@WPAFB.AF.MIL

45433, USA,

Email:

134

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

where there is a variable air gap between layers [3]. Pulsed eddy current with giant
magnetoresistive (GMR) sensors in conjunction with advanced data processing has also
been used to detect and estimate some properties of subsurface corrosion [4].
Likewise, GMR sensors and advanced feature extraction techniques have been
employed to detect subsurface cracks around fastener holes [5,6]. On-board in-situ
eddy current sensors have also been considered [7].
Although an asymmetric response observed for a hole feature in an eddy current
image from 2D raster scan data is typically used to distinguish crack and no crack
conditions, there are a variety of potential sources of coherent noise features that
produce similar asymmetric responses. In particular, misdrilled holes cause the
fasteners to be skewed and irregular gaps to exist between the fastener and the hole.
Variability in probe lift-off related to the scanning system hardware alignment or part
surface conditions can cause variability in the signal. The consistency of windings in
eddy current probes due to difficulty in manufacturing can also be a source for
measurement asymmetry. Lastly, the response from adjacent fastener sites can mask an
asymmetric response due to a crack. Clearly, the presence of coherent noise can
increase false call rates, limit crack detectability, and ultimately decrease the prospects
for crack sizing. Thus, there is a need to develop advanced data analysis approaches
and find reliable features that are sensitive to the crack condition yet invariant to such
coherent noise signals also present in real data.
The use of invariant features has been shown to be valuable for other problems in
eddy current nondestructive evaluation. A basic approach to reduce sensitivity to
liftoff during EC measurement of surface breaking cracks concerns adjusting the phase
during calibration so as to isolate liftoff to the horizontal measurement component
while a threshold is applied to the vertical measurement component in order to make a
call. Advanced feature extraction methods have also been developed to address a wide
variety of problems. For example, feature extraction methods were developed to
address unknown permeability variation through an invariance transformation of flux
density measurements incorporating radial basis functions [8]. The performance of
neural network classifiers were found to significantly benefit from the use of such
invariant signal features.
This paper focuses primarily on two concepts related to feature extraction in eddy
current NDE. The first concept concerns the use of multifrequency analysis of eddy
current data. In particular, a novel feature involving the second derivative of the
reactance component of the impedance change with respect to frequency is
investigated. The second concept concerns the use of a feature related to
circumferential gradients in the reactance component at a particular radius from the
center of an embedded cylinder. These two concepts are essentially applied to estimate
parameters related to the aspect ratio of subsurface cracks at fastener holes.

1. Frequency Domain Feature Extraction for Characterization of


Thickness and Depth of Material Loss
A series of simulations using the analytical solution for layered media were performed
varying total subsurface thickness loss, and depth of the thickness loss as illustrated in
Figure 1. Several different features were discovered that can be used to determine total
thickness loss and depth. Figure 2(a) shows the real part of the change in impedance as

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

135

a function of frequency for a variety of thicknesses and depths. At approximately 1.5


KHz, bands form that provide a good measure of total thickness loss. This measure
is also insensitive to the depth of the loss. Figure 2(b) shows a similar plot of the
imaginary part of the change in impedance as a function of frequency. For a particular
thickness loss, the imaginary component decreases as the depth of the air gap increases.
Figure 3 shows the second derivative of the imaginary component of the change in
impedance as a function of frequency. Again, bands are visible that provide a good
measure of total thickness loss as observed for the real part of the change in impedance.
Inner Diameter = 2.438 mm
Outer Diameter = 4.42 mm
30% IACS

Probe Height = 2.54 mm

3.175 mm
3.175 mm

b
Figure 1. Diagram of two layer system with material loss at the faying surface.

Figure 2. (a) Resistance and (b) reactance component for varying percent material loss (6%,8%,10%) and
percent of material loss in 2nd layer (0%, 50%, and 100%).

6%
8%
10%

Figure 3. Second derivative of reactance component for varying percent material loss (6%,8%,10%) and
percent of material loss in 2nd layer (0%, 25%, 50%, 75%, 100%).

136

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

2. Spatial Domain Feature Extraction for Improved Crack Detection


Recently, the difficult problem of distinguishing crack responses from other non-flaw
asymmetric features, gaps between the fastener and hole, probe liftoff variation, and
probe skew, was investigated [9,10]. The fruit of this investigation was a promising
feature that is invariant to several irregular non-flaw conditions listed in the
introduction to this paper. Earlier work proposed a method to determine optimal probe
location for crack detection around a fastener hole [9]. Once the probe location is
determined, data is collected, and if necessary, interpolated at a fixed radius around the
center of the fastener site. Figure 4(a) shows the imaginary component of the change in
impedance as a function of the polar angle around the fastener hole. The noticeable
Gaussian shaped response is the feature that is used for crack detection. To obtain a
measure of this localized crack feature, an approach was developed using a fit of a
characteristic function

f T

A cos T  B exp  DT 2  C

(1)

through nonlinear least squares estimation where the localized Gaussian response, B,
can be used as a crack measure separate from the sinusoidal noise feature as shown in
Figure 4(a). A model-based optimization approach was implemented to evaluate the
best signal processing algorithm design to distinguish crack size. In addition,
experimental studies were performed to further explore the reliability of this feature in
the presence of experimental noise and adjacent holes in close proximity. Through the
development of an automated algorithm to quantify this feature, results in Figure 4(b)
for the experimental study demonstrate the ability to detect small cracks around
fasteners while maintaining a low false call rate [10].
Although the proposed circumferential feature extraction methodology is
beneficial for distinguishing crack features in the presence of certain asymmetric hole
features, liftoff and probe tilt, other complex features of aircraft structures can hinder
the direct application of the approach. For general inspections, the presence of adjacent
fastener sites and part edges in close proximity can hinder the detection of cracks
located at certain angular locations around the hole of interest. In addition, they can
also contribute to the eddy current response in crack regions and thus decrease
-3

0.600

a=2
a=1
a=0
a=2
a=1
a=0

0.590
0.580

mm,
mm,
mm,
mm,
mm,
mm,

10

no tilt
no tilt
no tilt
w/tilt
w/tilt
w/tilt

Aluminum
Titanium
8
6
B (:)

0.570

x 10

0.560

0.550

2
0.540

0
0.530
0.520

-180

-135

-90

-45
0
45
T (degrees)

90

135

180

-2
0

0.5

1.5
a (mm)

2.5

Figure 4. (a) Gaussian response due to crack shown in the presence of coherent noise and (b) simulation
results show relationship between crack size and B parameter.

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

137

(
4

y (mm)

(a)

10

2
20

1
0

30
20

40

60
0.156

(b)
y (mm)

80

100
x (mm)

120

140

0.131

160

180
0.15

0.169

0.1

10

0.05
20

0
-0.05

30
20

40

60

80

100
x (mm)

120

140

160

180

Figure 5. Image plots of (a) original measurement data for in-phase component (Vx) and corresponding (b)
processed data with hole and edge feature extraction.

prospects for sizing. There is a clear need for model-based feature extraction schemes
that also compensate for adjacent fastener sites and part edges. New feature extraction
methods were developed to address fitting approximate models to data associated with
geometric part features including adjacent fastener sites and panel edges. The solution
strategy focuses on three steps: 1) a heuristic approach using a physical understanding
of the sources of greatest error, 2) a least-square estimation approach to solving for the
polynomial response quickly and accurately, and 3) an iterative approach to improve
model solutions for overlapping fastener site and part edge regions. Figure 5(a)
displays an image plot of the original measurement data for the in-phase component
(Vx) containing 10 fastener sites (9 of titanium, 1 of steel). Figure 5(b) displays an
image plot of the processed data with both hole and part edge feature extraction. For
this specimen with three cracks located around fastener sites, the crack features are
clearly observed. A complete automated process performs this feature extraction
algorithm in approximate 60 sec for a 10 hole panel, providing far greater accuracy and
a 10X improvement in speed over prior experience with direct global estimation
methods [11].

3. Model-based Data Analysis in Spatial and Frequency Domains for the Fastener
Crack Problem
In order to fully characterize sub-surface corner cracks at fastener sites in multilayer
structures, additional measures with sensitivity to all important crack dimensions are
needed. In particular, the sizing of corner cracks initiating from the near surface of the
second layer shown in Figure 6 requires values in the measurement data that are well
correlated to varying crack length (a) and crack depth (b). Multifrequency eddy current
methods have been investigated for many NDE applications that require the
characterization of various damage states from part characteristics of varying depth
[12]. Multifrequency eddy current data have also been used with imaging [13] and
inversion methods [14,15] to improve the detection and characterization of cracks and
corrosion conditions. In this work, a model-based feature extraction method of the
eddy current response in the spatial domain is coupled with multifrequency data
analysis for improved sensitivity to corner crack size and aspect ratio. To obtain
measures with sensitivity to the localized crack dimensions, the approach uses a fit of a
characteristic function for a fastener site and radial crack:

138

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

Figure 6. Diagram of fastener crack problem with varying notch length and depth: (a) a = 1.27 mm, b =
1.27 mm, (b) a = 2.54 mm, b = 1.27 mm, (c) a = 1.27 mm, b = 2.54 mm, (d) a = 2.54 mm, b = 2.54 mm.

f T , r, f

A r, f cos T  B r, f exp  D r, f T 2  C r, f

(2)

for varying radial location and frequency. The localized Gaussian response, B, is
evaluated as function of frequency and radial extent from the fastener site center and
used as a crack measure.
To investigate the viability of this approach, simulated studies in VIC-3D [16]
were performed. Cracks were modeled as notches of finite width with a quarter ellipse
profile. The crack length and depth were each varied over two levels in the study (1.27
mm and 2.54 mm) as shown in Figure 6. The frequency ranged from 50 Hz to 2000
kHz. Additional details on the probe model and multilayer structure properties can be
found in prior work [10].
Figure 7 presents the simulated results for the local crack characteristic response,
B, as a function of radial location and frequency for the four combinations of varying
notch length and depth: (a) a = 1.27 mm, b = 1.27 mm, (b) a = 2.54 mm, b = 1.27
mm, (c) a = 1.27 mm, b = 2.54 mm, (d) a = 2.54 mm, b = 2.54 mm. The response as
function of the radial direction provides a characteristic valley and peak moving away
from the hole center. The magnitude of these local minima and maxima as a function
of frequency were estimated using interpolation and investigated further for sensitivity
to crack length and depth. Figure 8 presents (a) the minimum response and (b) the
maximum response as a function of the frequency and the four combinations of notch
length and depth. The minimum and maximum response measures are partially
correlated to the cross-sectional area of the corner crack (=Sab/4). Both the minimum
and maximum responses also exhibit some varying sensitivity to the crack length and
depth with respect to frequency. Processing these minimum and maximum responses
can help distinguish the crack length and depth parameters.
For example,
normalization of the maximum response with respect to the minimum response [as in
Figure 8(c) at lower frequencies] and the maximum response at 500 Hz [as in Figure
8(d) at higher frequencies] provides the means to distinguish the two levels of crack
depth. A combination of these feature measures can thus be used to help distinguish
between these classes of crack length and depth and provide promise for accurate crack
sizing.

139

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

(a)

a = 1.27 mm, b = 1.27 mm


2000
0.05

1800

a = 2.54 mm, b = 1.27 mm

(b)

2000

1600

1600
1400

1200

freq (Hz)

1400
freq (Hz)

0.05

1800

1000

1200

1000

800

800

600

600
-0.05

-0.05

400

400

200

200
0

10

15

r (mm)

a = 1.27 mm, b = 2.54 mm

(c)

15

a = 2.54 mm, b = 2.54 mm

(d)

2000

2000

0.05

1800

0.05

1800

1600

1600
1400

1200

freq (Hz)

1400
freq (Hz)

10
r (mm)

1000

1200

1000

800

800

600

600
-0.05

-0.05

400

400

200

200
0

10

15

r (mm)

10

15

r (mm)

Figure 7. Local crack characteristic response as a function of radial location and frequency for varying
notch length and depth: (a) a = 1.27 mm, b = 1.27 mm, (b) a = 2.54 mm, b = 1.27 mm, (c) a = 1.27 mm, b
= 2.54 mm, (d) a = 2.54 mm, b = 2.54 mm.
0

0.07
a=1.27 mm,
a=2.54 mm,
a=1.27 mm,
a=2.54 mm,

-0.01

b=1.27
b=1.27
b=2.54
b=2.54

-0.04
-0.05

0.04

0.03

0.02

-0.06

0.01

-0.07

(a)
500

1000
frequency (Hz)

1500

2000

1.6
b=1.27
b=1.27
b=2.54
b=2.54

500

0.8

0.6

a=1.27 mm,
a=2.54 mm,
a=1.27 mm,
a=2.54 mm,

Xmax / Xmax(f=500 Hz)

1.2
-Xmax / Xmin

(b)
0

1000
frequency (Hz)

1500

2000

6
a=1.27 mm,
a=2.54 mm,
a=1.27 mm,
a=2.54 mm,

1.4

b=1.27
b=1.27
b=2.54
b=2.54

0.4

0.2
0

b=1.27
b=1.27
b=2.54
b=2.54

0.05

-0.03

Xmax (Ohms)

Xmin (Ohms)

-0.02

-0.08
0

a=1.27 mm,
a=2.54 mm,
a=1.27 mm,
a=2.54 mm,

0.06

(c)
500

1000
frequency (Hz)

1500

2000

(d)
0

500

1000
frequency (Hz)

1500

2000

Figure 8. Calculated response with respect to frequency for varying notch dimensions representing (a) the
miniumum response, (b) the maximum response, (c) the ratio of maximum to minimum responses, (d) the
normalized maximum response (with respect to maximum response at 500 Hz).

140

J.S. Knopp and J.C. Aldrin / Fundamental Feature Extraction Methods

4. Conclusions And Recommendations


Features have been demonstrated to quantify subsurface material loss, detect
subsurface cracks around fastener sites, and estimate parameters related to subsurface
cracks around fastener sites. Experimental studies will be conducted to study the
impact of measurement noise on second derivative features.

5. Acknowledgements
Funding was provided by the Air Force Office of Scientific Research. The authors
thank Dr. Harold Sabbagh and Dr. Matthew Golis for the input they provided.

References
[1] D. Hagemaier and G. Kark, Eddy Current Detction of Short Cracks Under Installed Fasteners, Mat
Eval, Vol 55 (1), pp. 25-30, (1997).
[2] J..S. Knopp, J. C. Aldrin, E. Lindgren, and C. Annis, Investigation of a Model-Assisted Approach to
Probability of Detection Evaluation, Rev. Prog. Quant. Nondestr, Vol 26, pp. 11775-1782, (2007).
[3] J.G. Thompson, Subsurface Corrosion Detection in Aircraft Lap Splices Using a Dual Frequency Eddy
Current Inpsection Technique, Materials Evaluation, Vol 51 (12), pp. 1398-1401, (1993).
[4] Y.A. Plotnikov, W. J. Bantz, and J. P. Hansen, Enhanced Corrosion Detection in Airframe Structures
Using Pulsed Eddy current and Advanced Processing, Mat Eval, Vol 65 (4), pp. 403-410. (2007).
[5] B. Wincheski , J. Simpson, M. Namkung, D. Perey, E. Scales, and R. Louie, Development of Giant
Magnetoresistive Inspection System for Detection of Deep Fatigue Cracks Under Airframe Fasteners,
Rev. Prog. Quant. Nondestr, Vol 21, pp. 1007-1014, (2002).
[6] N.V. Nair, V. Melapudi, H. Jimenez, X. Liu, Y. Deng, Z. Zeng, L. Udpa, T. Moran, and S. Udpa, A
GMR Based Eddy Current System for NDE of Aircraft Structures, IEEE Transactions on Magnetics,
Vol 42 (10), pp. 3312-3314
[7] N. Goldfine, V. Zilberstein, A. Washabaugh, D. Schlicker, I. Shay, and D. Grundy, Eddy Current
Sensor Networks for Aircraft Fatigue Monitoring, Mat Eval, Vol 61 (7), pp. 852-858, (2003).
[8] S. Mandayam, L. Udpa, S. Udpa, and W. Lord, Invariance Transformations for Magnetic Flux
Leakage Signals, IEEE Transactions on Magnetics, Vol 32 (3), pp. 1577-1580, (1996).
[9] J. S. Knopp and J. C. Aldrin, Numerical Studies of Eddy Current NDE for Small Crack Detection
around Fasteners in Multi-Layer Structures, Rev. Prog. Quant. Nondestr. Eval., Vol 24, pp. 417-424,
(2005).
[10] J. C. Aldrin and J. S. Knopp, Crack Characterization Method with Invariance to Noise Features for
Eddy Current Inspection of Fastener Sites, Journal of Nondestructive Evaluation, 25 (4), pp. 165-181,
(2006).
[11] J. C. Aldrin and J. S. Knopp, Case Study for New Feature Extraction Algorithms Automated Data
Classification, and Model-Assisted Probability of Detection Evaluation, Rev. Prog. Quant. Nondestr.
Eval. 26, pp. 257-264, (2007).
[12] H. L. Libby, Introduction to Electromagnetic Nondestructive Test Methods, John Wiley, (1971).
[13] T. Chady, M. Enokizono, and R. Sikora, Crack Detection and Recognition Using an Eddy Current
Differential Probe, IEEE Transactions on Magnetics, Vol. 35 (3), pp. 1849-1852, (1999).
[14] J. C. Aldrin, H. Sabbagh, E. Sabbagh, R. Murphy, M. Concordia, D. Judd, E. Lindgren, and J. Knopp,
Methodology Using Inverse Methods for Pit Characterization in Multilayer Structures, Rev. Prog.
Quant. Nondestr. Eval. Vol. 25, pp. 767-774, (2006).
[15] N.C. Haywood, and J. R. Bowler, Eddy Current Imaging of Buried Cracks by Inverting Field Data,
IEEE Transaction on Magnetics, Vol 28 (2), pp. 1336-1339, (1992).
[16] R. Murphy, H. Sabbagh, A. Chan, and E. Sabbagh, A Volume-Integral Code for Electromagnetic
Nondestructive Evaluation, Proceedings of the 13th Annual Review of Progress in Applied
Computational Electromagnetics, (1997).

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-141

141

Inversion of Potential Drop Data for the


Reconstruction of Crack Depth Profiles
Giuseppe SPOSITO a, Peter CAWLEY a and Peter B. NAGY a,b
Non-Destructive Testing Group, Department of Mechanical Engineering, Imperial
College London, London SW7 2AZ, UK
b
Department of Aerospace Engineering and Engineering Mechanics, University of
Cincinnati, Cincinnati, OH 45221

Abstract. This paper considers the inverse problem of estimating the depth profile
of an unknown defect from measurements of transfer resistance. The results of
both finite-element analyses and experimental tests on specimens with EDM
notches of various shapes and sizes were used to develop a simple inversion
algorithm that allows a good reconstruction of the depth profile. A synthetic
focusing technique is applied which improves the quality of the reconstruction.
Keywords. Potential Drop, inverse problem, crack sizing, focusing

1. Introduction
Estimating the shape and size of a defect is a problem of major interest in many
industrial applications, since the depth of a crack is often a key parameter in
calculations of structural integrity. The potential field created on the surface of a
testpiece by the injection of direct or alternating currents for Potential Drop (PD)
measurements has been obtained analytically for simple geometries in the absence of
defects [1]; previous studies have also shown that a simple three-dimensional Finite
Element (FE) model is able to give an accurate solution to the direct problem of
predicting the response of a probe to a surface-breaking defect of known geometry [2].
Most previous work on the inverse problem of using values of transfer resistance
measured at a number of different locations to calculate the depth profile of an
unknown defect have assumed a priori knowledge of the defect shape, or made use of
parameters to be evaluated heuristically [3-6]; crack gauges are commercially available
that assume the defect has a semi-circular form. However, this assumption is not
always correct. The aim of the present study is to develop an inversion technique of
more general validity.

2. Experimental Setup Used for Measurements


For the present study a linear array probe having the dimensions shown in Figure 1 was
manufactured, its dimensions being chosen so that if current is injected at the outer pins
the distribution across the measurement electrodes is fairly uniform. The 5 mm spacing
between the lines of measurement electrodes was chosen to give reasonably high

142

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

60
5
j=-2
j=-1
j=0
j=1
j=2

Figure 1. Geometry of probe (dimensions in mm)

sensitivity to cracks (the fractional increase in resistance due to a defect is roughly


proportional to the defect depth as a fraction of the electrode spacing) while not being
overly sensitive to errors in the pin positioning.
However, the whole system is reciprocal and it is much more practical to multiplex
the healthy input currents (~130 mA) than the very small measured voltages (order 1
V). Therefore when running the tests it was decided to measure the voltage across the
two outer electrodes of Figure 1 and inject current across pairs of inner electrodes on
opposite sides of the defect position in turn.
The excitation signal is generated by the internal oscillator of a lock-in amplifier
and driven by a differential circuit: the symmetry of the signal is ensured by two
scaling amplifiers whose gain, initially set as equal and opposite, is automatically
adjusted to compensate for any differences between the contact resistances of the
electrodes. A multiplexer then routes the signal to one of the pairs of inner electrodes at
a time. The voltage difference measured at the outer electrodes is amplified by a lownoise preamplifier, and the values are read on the same lock-in amplifier used to
generate the excitation signal. The instruments are remotely controlled via a simple
LabView routine.
Relatively small currents (as little as 130 mA) could be used, as opposed to the
large currents (up to the order of 100 A) often required in commercially available
DCPD systems, thanks to the high sensitivity of the lock-in amplifier used for the
experiments, which is capable of measuring signals of the order of a nV. However, it
must be mentioned that the signal to be measured, i.e. the voltage difference between
the two sensing electrodes (differential signal), is typically much smaller than the
voltage at either electrode (common signal), and therefore, in addition to the
differential driving mentioned earlier, a very high Common Mode Rejection Ratio
(CMRR) is required; the preamplifier used for the experiments had a CMRR in excess
of 110 dB, which means that the differential part of the signal is amplified by a factor
at least 3 105 times more than the common signal.
The electronic noise introduces oscillations of less than 0.1% in the measured
signal, and it is negligible compared to the variations due to the uncertainty in the
positioning of the probe.

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

143

Figure 2. Current distribution along the centreline of the array probe (x=0) for focusing with 1, 3 and 5 pairs
of electrodes.

3. Focusing
The thin line in Figure 2 shows the current distribution across the centreline of the
array when a unit current I0 is injected between the two central electrodes. The wide
lateral spread of current causes an averaging effect that would smear rapid variations
in the depth profile of a defect.
The array of Figure 1 gives the possibility of producing focused currents. If a
negative current was applied to the pairs of electrodes j=1 in Figure 1 adjacent to the
central pair (j=0) which was driven with the positive current that yielded the thin-line
distribution of Figure 2, it would be expected that part of the wide lobe could be
cancelled out, thus squeezing the current into the centre of the array. The thick and
shadowed curves of Figure 2 show the resulting current distribution if this concept is
applied to three and five pairs of electrodes respectively: the peak is considerably
narrower, as desired; as a side effect the current at the centre is reduced. The optimum
weightings of currents applied to the pairs are different depending on the number of
pairs used, and they are also a (weak) function of frequency. The distributions shown in
Figure 2 are for low frequency (DC) and are obtained by injecting I1=-0.33 at the
electrode pairs j=1 for 3-pair focusing, or I1=-0.39 at j=1 and I2=+0.04 at j=2 for
five pairs. These values, scaled to the positive unit current I0 injected at the central pair,
are those that minimise the value of the integral
J x ( y)
dy ,
x (0)

(1)

where Jx is the axial current density at the centreline of the array, as plotted in Figure 2,
and the integral is calculated over the centreline along the entire width of the specimen.

144

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

12

Transfer resistance [P:]

11

10

8
PR0 (Baseline)
PR1 (Rectangle)
PR2 (Circular arc)
PR4 (Triangle)

6
0

10

15

20

25

30

35

40

45

50

Distance from edge [mm]

Figure 3. FE predictions (lines) and measurements (points) at 10.3 Hz on a notch-free specimen and on
specimens with 10-mm long, 3-mm deep notches of different shapes.

In practice the focusing was done synthetically: measurements were taken by


applying the same current to each electrode pair in turn, and the focused results were
computed later by combining the weighted data. This procedure is analogous to SAFT
(Synthetic Aperture Focusing Technique), commonly used in ultrasound and radar.

4. Notch Profile Reconstruction Using Focused Array


Tests were run on 150-mm long, 50-mm wide, 10-mm thick blocks of SS304. In
addition to experimental measurements, FE simulations were run using the commercial
FE software Abaqus and the simple quasi-three-dimensional model presented in [2].
Both experimental and numerical results on a notch-free specimen (baseline) and on
three specimens with EDM notches of different shapes (rectangular, triangular, circular
arc) but identical length and maximum depth are shown in Figure 3. It should be noted
that no focusing has been applied at this stage. Three array positions were required to
cover the full block width, which explains the three groups of points in the graph for
each specimen; some variation would be expected across the array even in an infinitely
wide plate, because the outer electrodes are fixed and only directly in line with the
central pair of the array. This effect is predicted satisfactorily by the FE model, as is the
much larger variation towards the edges of the specimen; both effects are removed
once results relative to the baseline are considered.
As an example, the reconstructed profile of the triangular notch is shown in Figure
4. At each measuring point the estimated depth is given by
d

V  V0
,
V

(2)

145

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

3.5
1 pair

3.0

Estimated depth [mm]

3 pairs
5 pairs

2.5

Real profile
2.0
1.5
1.0
0.5
0.0
-0.5
0

10

15

20

25

30

35

40

45

50

Distance from edge [mm]


Figure 4. Reconstructed profiles of a 10-mm long, 3-mm deep triangular notch using 1, 3 and 5 pairs of
electrodes to focus currents. FE predictions (solid lines) and measurements (points) compared with the real
profile (dashed line).

where T is the block thickness and V and V0 are the voltages measured on the notched
specimen and on the baseline, respectively. This formula is easily derived assuming
that at low frequency the potential drop is inversely proportional to the remaining
thickness T-d seen by the current. It is therefore important that the lateral spreading of
the current be small, as achieved by focusing. The results of Figure 4 show that
focusing the currents synthetically does in fact yield a very good reconstruction of the
depth profile already for three pairs. Note that the same number of electrodes must be
used for both V and V0 if focusing is used.
4.0
1 pair

3.5

Estimated depth [mm]

3 pairs
3.0

5 pairs
Real profile

2.5
2.0
1.5
1.0
0.5
0.0
-0.5
0

10

15

20

25

30

35

40

45

50

Distance from edge [mm]


Figure 5. Reconstructed profiles of a 10-mm long, 3-mm deep rectangular notch. FE predictions (solid lines)
and measurements (points) compared with the real profile (dashed line).

146

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

4.0
1 pair

3.5

Estimated depth [mm]

3 pairs
3.0

5 pairs

2.5

Real profile

2.0
1.5
1.0
0.5
0.0
-0.5
0

10

15

20

25

30

35

40

45

50

Distance from edge [mm]

Figure 6. Reconstructed profiles of a 10-mm long, 3-mm deep circular arc notch. FE predictions (solid lines)
and measurements (points) compared with the real profile (dashed line).

The reconstruction of the rectangular notch (see Figure 5) is less satisfactory:


focusing, while sharpening the representation of the step at the extremities of the notch,
causes an overestimation of the maximum depth. Defects of such shape, however, are
very unlikely to occur in practice; in fact, previous studies on this subject often made
the assumption that cracks have a semi-circular or semi-elliptical shape (see for
example [3] and [6]). A circular arc is therefore a fairer approximation of a real crack.
The results for this notch are shown in Figure 6: as for the triangular notch, the
reconstruction is very good if focusing with three or five pairs.

5. Conclusions
A FE model was used to determine a simple formula for the inversion of Potential Drop
measurements. This was subsequently applied to experimental data obtained with an
array probe on specimens with EDM notches of various shapes, and it has been shown
to give a good reconstruction of the notch depth profiles. The improvement in the
results introduced by synthetic focusing of the injected currents has also been
discussed.
Future work will be aimed at exploring the range of applicability of the formula
used for the estimation of crack depth, by extending the study to notches of different
aspect ratios and blocks of different geometry.

References
[1]

N. Bowler, Analytical Solution for the Electric Field in a Half-Space Conductor Due to Alternating
Current Injected at the Surface, J. Appl. Phys. 95(1), 344-348 (2004)

G. Sposito et al. / Inversion of Potential Drop Data for the Reconstruction

[2]
[3]
[4]
[5]
[6]

147

G. Sposito, P. Cawley, and P. B. Nagy, Crack Profile Reconstruction by Means of Potential Drop
Measurements, Review of Progress in Quantitative Nondestructive Evaluation 26A, 733-740 (2006)
D. H. Michael, R. T. Waechter, and R. Collins, The Measurement of Surface Cracks in Metals by
Using AC Electric Fields, Proc. R. Soc. Lond. A 381, 139-157 (1982)
M. P. Connolly, D. H. Michael, and R. Collins, The Inversion of Surface Potential Measurements to
Determine Crack Size and Shape, J. Appl. Phys. 64(5), 2638-2647 (1988)
M. McIver, Characterization of Surface-Breaking Cracks in Metal Sheets by Using AC Electric
Fields, Proc. R. Soc. Lond. A 421, 179-194 (1989)
K. Ikeda, M. Yoshimi, and C. Miki, Electrical potential drop method for evaluating crack depth, Int.
J. Fracture 47, 25-38 (1991)

148

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-148

Automatic classification of defects with


the review of an appropriate feature
extraction
Alicia ROMERO RAMIREZ, Neil PEARSON and Dr. J. S.D. MASON
Swansea University, SA28PP, UK. Email: 436912@swansea.ac.uk

Abstract. A novel method for the automatic classification of defects using


magnetic flux leakage inspection is presented. A technique based on geometric
measures to distinguish between different defects due to petro-chemical tank
corrosion is presented. In order to characterize a defect, a process of feature
extraction is proposed. Principal component analysis is then used to select the most
powerful set of features.
The performance is compared using two different methods: k-nearest
neighbor and support vector machine. The results show an accuracy of 91% with
which automatic classification is possible on unseen test examples on steel plates.

Introduction
Petro-chemical storage tanks are important objects in the industrial sector. The servicelife of a storage tank is between 20-40 years, although in some cases this life is reduced
to 2-3 years due to failures within the storage tank [1]. The importance of predicting a
possible tank failure for the industry is not only the value of the tank content, which in
the case of petroleum tank is very high, but also the environmental damage that the
tank failure can cause and consequently, fines the industry would potentially face.
The main cause of failure of steel storage tanks is corrosion [2]. Magnetic flux
leakage (MFL) can reliably detect metal loss due to corrosion, permitting a quantitative
evaluation of the size of the defect [3]. MFL tools are equipped with sensors to collect
information about the state of the floor tank. It is suggested in [4] that a deeper
understanding of the shape of the MFL signals could be beneficial to find more
effective type-of-defect separation methods. Once a defect is detected, its shape and
size is not easy to predict. This is the topic of this paper.
In recent years there has been much interest in the development of automatic
classification of defect patterns and many are using neural networks, for example [5-9].
Some successful work in the use of ultrasonic approaches is reported in [5,8,9] and
automatic defect classification using MFL is reported in [7] for pipe welds achieving a
success classification rate of 71% across 3 classes namely spheres, parallelepipeds and
cylinders. In [6] eddy current signals are used for defect classification in aluminium
plates with a reported error rate of 10%.

A. Romero Ramirez et al. / Automatic Classication of Defects

1.

149

Data acquisition

MFL testing machines require a mapping of magnetic fields onto flaw geometry. The
machine used for the development of this work, measures magnetic flux signals and
converts them into estimates of percentage of volumetric loss. A discussion about the
reliability of this method can be found in [3]. Each prospective or potential defect is
represented by a matrix of data ([m x n]), in which the geometry of the defect is not
easily determined. In very simple terms the defect classification could be in terms of
the number of rows and the number of columns ([m x n]) for which the signal exceeds
some threshold. The percentage of metal-loss could be a suitable integration of these
values. However we show here that more useful classification is possible.

2.

Feature extraction and feature transformation

Defect characterization and pattern recognition are essential for the development of an
automatic classifier. The goal is to highlight similarities between defects from the same
class and to draw attention to the differences between defects from different classes.
A number of geometrically derived features are considered. These include: relation
between length and width, position of the maximum and minimum, angle of the slopes,
gradient of the channel with the maximum, mean values, area from the top view,
volume and length relations. In total, the number of extracted features per defect is 50.
By examining a number of examples and studying the different classes, there have been
identified that lead to good classification performance. Clearly, there is significant
redundancy across the set of 50 chosen features. Hence in an attempt to reduce this
redundancy, principal component analysis (PCA) [11] is used. Figure 1 shows the
results of the PCA analysis on a preliminary training data set. Note subsequent
experimental results are performed on data test sets that do not direct overlap with this
PCA set. In Figure 1 it is shown that by selecting the 30 most significant components
the percentage of information lost is below 1%.

Figure 1. % of information versus PCA components retained. Experiment on preliminary data set.

150

3.

A. Romero Ramirez et al. / Automatic Classication of Defects

Classification

The overall objective of the work is to determine that an incoming signal belongs to a
specific category from among a finite set of possibilities. This is referred to as
identification and is a 1-from-n classification task. Alternatively verification can be
considered; this is a 1-from-2 class task and can be regarded as a special case of
identification.
The choice of classes form a closed set was made taking into consideration the
shape of real defects appearing in a petro-chemical storage tank. Due to the width of
the platforms (between 6-12 mm) undercutting defects are unlikely to appear [15].
Consequently the shapes of our defects are restricted. Defects are assumed to have a
profile similar to the ones shown in Figure 2, namely pipe, conical and lake.

Figure 2. Different defect profiles.

4.

Experimental work

The assessment of the classifier is reported in two different ways. The first uses a
multi-class classifier with three classes and the second uses three binary classifiers (one
per class). The arrangements are shown in Figure 3. In both cases a signal to test
gives rise to three scores (Ps, Pc, Pl) and these are subjected to a single decision
threshold. The relative performance is shown on a detection error tradeoff (DET) curve
[14] as the decision threshold is varied. In addition accuracy scores are given.

Signal to test

Signal to test

Multi class
classifier

Probability of being a pipe defect. (Ps)


Probability of being a conical defect. (Pc)
Probability of being a lake defect (Pl)

Pipe classifier

Probability of being a pipe defect. (Ps)


Probability of NOT being a pipe defect.

Conical
classifier

Probability of being a conical defect. (Pc)


Probability of NOT being a conical
defect.

Lake classifier

Probability of being a lake defect (Pl)


Probability of NOT being a lake defect

Figure 3. Experimental setup.

151

A. Romero Ramirez et al. / Automatic Classication of Defects

DET curves are frequently used to measure the verification performance of a


classification algorithm. Verification addresses a two-class problem where the expected
answer is either true or false. Here we focus on verification so that the assessment is
independent of the number of classes. For each trial, the score is thresholded and a
decision is made. The DET curve shows the variation of this threshold. Such an
approach gives a robust indication of performance. Figure 4 presents the scheme of a
feature vector entering the classifier which produces a score which is them threshold.
Moving the threshold changes the distribution of the error rates as indicated in the DET
plots.

Threshold
2.5

Classifier

SCORE

1.5

Density

Feature
Vector

0.5

0
1

0.5

0.5
Critical values

1.5

Figure 4. A feature vector enters the classifier which produces a score which is then thresholded. The 2
distributions represent actual in-class and out-of-class scores.

Due to the difficulty of having real data from the field with accurate ground truth,
emulated corrosion has been used. Thirty defects in total representing three classes
(pipe, conical, lake) were created in steel plates. The profiles and sizes are shown in
Table 1.
The data set contains 816 records of a total of 30 defects classified as: pipe (121),
conical (330), lake (365). It is worth noting that the recordings were made on 7
different dates. The reason for having fewer samples of pipe defects is due to a
limitation of the scanning tool, which captures only defects with a volume-loss larger
than 20%. The data set is divided into two groups: a) training (408 recordings), b)
testing (408 recordings). There is no data overlap in between the two.

Table 1. Description of emulated defects.


Profile

Type

Thickness of
plate

Pipe

6 [mm]

Conical

6 [mm]

Lake

6 [mm]

Diameters [mm] x maximum depth [mm]


2.0 x
2.0
2.0 x
4.0
10 x
1.2
17.3 x
4.2
20 x
1.0
20 x
2.0

3.0 x
2.0
3.0 x
4.0
12.1 x
1.8
18.2 x
4.8
30 x
1.0
30 x
2.0

4.0 x
2.0
4.0 x
4.0
13.7 x
2.4
10 x
1.2
40 x
1.0
40 x
2.0

5.0 x
2.0
5.0 x
4.0
15.1 x
3.0
13.7 x
2.4
50 x
1.0
50 x
2.0

6.0 x
2.0
6.0 x
4.0
16.3 x
3.6
16.3 x
3.6
100 x
1.0
100 x
2.0

152

A. Romero Ramirez et al. / Automatic Classication of Defects

Performances with SVM

Performances with KNN

Table 2. Experimental results for K-nearest neighbour and support vector machine classifiers. The
SVM with three verification classifiers seem to give the best results.

Accuracy using
multiclass
classifier: 85.75 %
Accuracy using three
binary classifiers:
90.25 %

Accuracy using
multiclass
classifier: 87.46 %
Accuracy using three
binary classifiers:
91.44 %

The experimental results in Table 2 show the accuracy and the DET curves using
two different classifiers, namely K-nearest neighbour and support vector machine. In
both cases there is a discontinuous profile which corresponds to the multiclass
experimental setup and a continuous profile that correspond to the three binary
classifiers experimental setup. The number of scores used per profile is 1224 scores.

A. Romero Ramirez et al. / Automatic Classication of Defects

5.

153

Conclusion

Classification of emulated corrosion based on the proposed approach has been


successfully achieved with accuracy over a 90% when classifying pipe, conical and
lake defects. The final goal of this work is the characterization of real defects appearing
in storage tank floors. Here a simplification of the problem based on emulated
corrosion is presented. The feasibility of using emulated corrosion as data to train the
classifier for real inspections is currently being tested.

Acknowledgements
The work is funded by European Social Funds in collaboration with Silverwing Ltd.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]

M.L. MEDVEDENA and T.D. TIAM. Classification of corrosion damage in Steel Storage Tanks.
Chemical and Petroleum Engineering.Vol.34. Nos 9-10. 1998
GEYER W.B Handbook of storage Tank Systems: Codes, Regulations and Designs.
K.REBER, A.BELANGER. Reliability of Flaw Size Calculation based on Magnetic Flux Leakage
Inspection of Pipelines. ECNDT 2006
Till SCHMITTE. Modelling of Magnetic Flux Leakage Measurements of Steel Pipes. ECNDT 2006
Oleg KARPASH, Maksym KARPASH, Valentine MYNDJUK. Development of Automatic Neural
Network Classifier of Defects Detected by Ultrasonic Means. ECNDT 2006
Adam DOCEKAL. Signal Preprocessing Methods for Automated Analysis of Eddy Current Signatures
during Manual Inspection. ECNDT 2006
A.A.CARVALHO. MFL signals and artificial neural networks applied to detection and classification of
pipe weld defects. NDT &E International. June 2005.
J.B. SANTOS. Automatic defects classification-a contribution. NDT &E International.. June 2000.
A.MASNATA. Neurual network classification of flaws detected by. ultrasonic means.NDT & E
International. October 1995
K. MANDALY, D.L. ANTHERTON.A study of magnetic flux-leakage signals. July 1998
LINDSAY I SMITH. Principal component analysis. February 2002.
V.N Vapnik, Statistical Learning Theory. Adress: New York: Wiley, 1998.
C.J.C BURGES. Discov., vol2, no 2, pp.1-47, 1998.
NIST, DET-Curve Plotting software for use with MATLAB. Software available at
http://www.nist.gov/speech/tools
http://www.corrosion-doctors.org/Forms-pitting/Pitting.html

154

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-154

Microwave Nondestructive Detection of Longitudinal


Cracks in Pipe with U-bend and Prediction of its Location
by Signal Processing
a

Kavoos ABBASIa,1 , Satoshi ITOa , Hidetoshi HASHIZUMEa


Dept. of Quantum Science and Energy Engineering, Tohoku University, Sendai, Japan

Abstract. A circular electromagnetic TE11-mode is used for detection of


longitudinal crack in a stainless steel pipe. To show the capability of microwave
nondestructive testing for detection of cracks in large and complex piping systems
such as steam generator (SG) tube, inspected pipe including Ubend are examined
in this study. The crack location is determined through time of flight (TOF) of the
reflected wave and group velocity of electromagnetic signal. The TOF is evaluated
in the time domain via Inverse Fast Fourier Transform (IFFT) of the wave
spectrum. Then, two different methods of signal processing are applied to obtain
TOF more accurately.
Keywords. Microwave, longitudinal crack, Time of flight (TOF)

Introduction
Steam generator (SG) tubing in pressurized water reactors (PWRs) is subject to a
variety of degradation processes that can lead to tube cracking, wall-thinning, and
potential leakage or tube rupture [1]. In order to avoid the tube failure, the piping
systems of the reactors must be routinely inspected to guarantee the safety of
operations. Hence detection of defect in complex piping systems in early stage of
degradation with a high-speed and high-accuracy inspection method is serious matter in
the operation of pressurized water reactor. At present, the typical methods for detecting
defects are eddy current testing (ECT) and ultrasonic testing (UT). Even though these
methods have high accuracy, they need point-by-point inspection and therefore, they
are time/costconsuming for the inspection of long pipes [2, 3]. Hence it is desirable to
develop another high-speed technique for crack detection. One of the most promising
techniques which might reduce the time/cost of long-range inspection is the NDT
method using microwaves. The experimental results show that, due to propagation of
an electromagnetic wave above cutoff frequency in waveguides without significant
attenuation loss it is possible to inspect lengthy pipes [4, 5]. In our previous studies,
detection of circumferential and longitudinal cracks by using TM01 and TE11 modes
were experimentally verified in a short straight tube [6, 7].
In this study to show the potential of this method for inspection of large and
complex piping system, the microwave NDT method is applied to detect a longitudinal
crack in a long pipe including U-bend. The obtained experimental results are analyzed
by two different signal processing methods.
1Corresponding

Author: Kavoos Abbasi, Tohoku University, Aramaki-Aza-Aoba 6-6-01-2, Sendai, Japan;


E-mail: abbasi@karma.qse.tohoku.ac.jp

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

155

1. Background and Experimental Setup

This study is carried out to detect longitudinal crack in a pipe with U-bend by
using circular TE11-mode which is suitable for detection of longitudinal crack in the
pipe. The TE-modes are characterized by fields with Ez = 0 and Hz z 0, where z
indicates the direction of propagation (In the case of cylindrical pipe, z corresponds to
the axial direction of the pipe). When an electromagnetic wave is propagated in the
pipe, an electric surface current with a circumferential component is produced at the
inner surface of the pipe. Once this surface current flows in the pipe with a
longitudinal crack, the crack prevents proper flow of the surface current. Consequently,
part of the incident wave is reflected, which has information about the crack.
A schematic diagram of the experimental setup is given in Fig.1. The network
analyzer is used as the generator of the electromagnetic wave. The generated wave
passes through the mode converter via the coaxial line. The mode converter is formed
by joining the rectangular waveguide (C-band) to the circular waveguide in order to
transform the rectangular TE-modes to the circular TE- and TM-modes.

pipe 2

Crack

Matched load
Plunger position

Tapered
waveguide
F

110.9
mm

Circular
waveguide

C
B
69.5 mm
D
298.8 mm
U-bend

126 mm

pipe 1

85 63 A

SPE C TR U M AN A L YZ E R

9 kH z - 26.5 G H z

65 mm

Rectangular
waveguide

Mode
converter

Coaxial line

Network Analyzer
10 MHz ~ 40 GHz

Fig.1. Experimental system for detection of axial crack


The first dominant circular TE-mode is TE11-mode which can be resonated in the
system by moving the plunger in the circular waveguide. The conditions to resonate or
to damp the mode are given by the following equations:

156

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

m
(for resonating)
2
2m  1
(for damping)
Og .
4

Og .

(1)
(2)

where l is the distance between plunger and center of rectangular waveguide and m is
an integer. The Og is the wave length of the electromagnetic wave as is guided in the
circular waveguide.
Group velocity of the wave, which is an important parameter for calculating the
crack locations, is also given by the following equation.

vg

PRH R

f
1  c
f

(3)

wherein c, PR, HR are light velocity, relative permeability and relative permittivity,
respectively.
The inspected pipe made of SUS-304 has inner and outer diameters of 34 mm and
38 mm respectively. As seen from Fig.1, the inspected pipe is made of three parts, two
straight pipes and U-bend pipe. The length of pipe 1, pipe 2 and U-bend is 3000 mm,
800 mm and 1200 mm respectively. The crack is a longitudinal slit whose length is 40
mm and width is 0.3 mm and made by cutting blade from the outer surface of the pipe
to the inner surface. The position of the created crack is indicated by point F in Fig.1.
To reduce environmental noise and to absorb the transmitted wave, a matched load
which is made of paraffin and graphite is mounted at the end of the pipe. The
electromagnetic modes can be resonated or reduced by changing the plunger position
from 90 mm (minimum position) to 180 mm (maximum position). The experiment is
performed for pipes without fluid in them.

2. Results and Discussions


2.1 Evaluation of results at plunger position of 90 ~180 mm
Two experiments are carried out for two crack positions CF = 4200 mm and 4600
mm separately, where CF is the length from point C to point F in Fig.1. The TE11-mode
is the first and dominant circular mode in any circular waveguide. The cutoff frequency
of this mode in the inspected pipe is 5.17 GHz. The next circular mode is TM01-mode
with cutoff frequency of 6.755 GHz. To avoid generation of TM01-mode in the system,
frequency range of 5.2 ~ 6 GHz is chosen in this experiment.
Figures 2(a) and (b) show the results for the crack located at 4180 mm in
frequency and time domain, respectively when the plunger position is changed from 90
mm to 180 mm. Fig.3 shows the same results for a crack located at 4580 mm. These
results are obtained by subtraction of reflection coefficient signal () of the pipe with
and without crack. The color bar in each figure shows these differences of two signals
(). Reflection coefficient is defined as the ratio of the reflected signal voltage to the
incident signal voltage as the following equation.

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

Vreflected
Vincident

157

(4)

-3

|'*| [x10 ]

As observed from Fig.2 and Fig.3, reaches the maximum value in several
plunger positions. This means that the TE11-mode is resonated at those plunger
positions. Dashed-line indicates TOFs obtained by the calculation at frequency of
6 GHz.

Calculated time
at 6 GHz

(a)

(b)

-3

|'*| [x10 ]

Fig.2. Difference of reflection coefficient () at different plunger positions


for the crack located at 4180 mm (a) frequency domain (b) time domain

Calculated time
at 6 GHz

(a)

(b)

Fig.3. Difference of reflection coefficient () at different plunger positions


for the crack located at 4580 mm (a) frequency domain (b) time domain

158

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

2.2 Evaluation of results at one plunger position

0.10

0.10

0.08

0.08

0.06

0.06

0.04

0.04

0.02

0.02

'*

'*

The obtained results in Fig.2(b) and Fig.3(b), indicated that large response of the
electromagnetic waves are occurred in several plunger positions such as 90 mm, 125
mm and 155 mm (almost center of area with large intensity). Herein, As an example,
the result when plunger located at 125 mm is investigated. Figure 4 displays in
frequency domain and Fig.5 displays in time domain for two crack locations. The
signals shown in Fig.5 are obtained by taking the IFFT (Inverse Fast Fourier transform)
of the signal shown in Fig.4.

0.00

0.00

-0.02

-0.02

-0.04

-0.04

-0.06

-0.06

-0.08

-0.08

-0.10
5.2

5.4

5.6

5.8

-0.10
5.2

6.0

5.4

5.6

5.8

6.0

Frequency [GHz]

Frequency [GHz]

(b

(a

1.0

0.5

0.5
-3

' * [x10 ]

1.0

-3

' * [x10 ]

Fig.4. Difference of reflection coefficient () in frequency domain for plunger


position of 125 mm and (a) crack located at 4180 mm (b) crack located at 4580 mm

0.0

0.0

-0.5

-0.5

-1.0

-1.0
0

20

40

60

80

Time [ns]

100

120

140

20

40

60

80

100

120

140

Time [ns]

(a)
(b)
Fig.5. Difference of reflection coefficient () in time domain for plunger position
of 125 mm and (a) crack located at 4180 mm (b) crack located at 4580 mm

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

159

In Fig.2 (b) and Fig.3 (b), the shortest times are measured as TOFs (Time of
Flight) for the each crack location since the highest frequency (6 GHz) in the operation
frequency range has the shortest time of flight. In order to calibrate the experimental
result, the TOF is calculated at a frequency of 6 GHz. The obtained TOFs from the
calculations for the two aforementioned crack locations, 4180 mm and 4580 mm, are
62 ns and 67.2 ns, respectively. The calculated TOFs are denoted in Fig.5 (a) and Fig.5
(b) by solid arrows.
Figure 6 is obtained by a applying Hilbert transform filter upon the signal shown in
Fig.5. As is seen from this figure, the response of electromagnetic waves to the crack
becomes clearer than before and noise as displayed in each graph (from 0 to ~60 ns) is
mostly canceled out. The solid arrows in Fig. 6 indicate the calculated TOFs. One
might take for granted that the two sharp peaks with large amplitudes in Figs.6 (a) and
(b) with their related times (75 ns and 80 ns) can be considered as the TOF of the
electromagnetic wave for each crack location. However, it is not a correct assumption
for the maximum frequency (6 GHz) in the frequency range considered has the shortest
TOF. Later on, we will show that the TOF is actually the time when starts to
increase rather than the time related to those large peaks.
Hence, in the next step another method is applied to the signal shown in Fig.5 so as
to tell which time should be considered as the real TOF of the crack. In this method a
cut-off value is defined with regard to the maximum value of as given by the
following equation.
Cut off value = E (') max
where the threshold factor (E) is changed from 0.05 to 0.3 (such values are
extracted from the experimental error). The results obtained via such a signal
processing are shown in Fig.7 (a) and Fig.7 (b). In this analysis the signals are cut off
until the response of electromagnetic waves becomes larger than the cut-off value.
As observed from these figures, for two crack locations, the TOF remains almost
constant when the threshold value changes from 0.1 to 0.3. The obtained TOFs by this
method for two crack locations are 66.45 ns and 71. ns. There is also a small
discrepancy between these TOFs and those calculated (62 ns and 67.2 ns). These
differences for two crack locations (4180 mm and 4580 mm) are 4.45 ns and 3.9 ns.
From Fig.4, it is seen that the amplitude of the signal in frequency range of about 5.9 ~
6 GHz is rather smaller than at other frequencies (5.2 ~ 5.9 GHz). This means that the
response of the electromagnetic wave within this frequency range is too small to be
detected. This is why the calculated time is a little different from the one predicted.

160

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

4.0
3.5

Calculated
time = 62 ns

Calculated
time = 67.2 ns

3.0
-6

_'*_ [x10 ]

-6

_'*| [x10 ]

2.5
2.0
1.5
1.0

0.5
0
0

20

40

60

80

100

120

0.0

140

20

Time [ns]

40

60

80

100

120

140

Time [ns]

(a)

(b)

80

80

70

70

60

60

Time [ns]

Time [ns]

Fig.6. Difference of reflection coefficient () in time domain for plunger


position of 125 mm after applying Hilbert transform and (a) crack located at 4180
mm (b) crack located at 4580 mm

50
40

Predicted TOF

30

50

30

20

20

10

10

0
0.05

0.10

0.15

0.20

0.25

0.30

Predicted TOF

40

0
0.05

0.10

0.15

0.20

0.25

0.30

(a)
(b)
Fig.7. Difference of reflection coefficient () in frequency domain for plunger
position of 125 mm and (a) crack located at 4180 mm (b) crack located at 4580 mm
The times as calculated at frequency of 5.9 GHz are 64.94 ns and 70.47 ns and
differences with those predicted are 1.51 ns and 0.63 ns. The calculated crack locations
at this frequency are 4173 mm and 4618 mm, which are closer to the predicted ones.
The predicted TOF and predicted crack location are summarized in Table 1.

K. Abbasi et al. / Microwave Nondestructive Detection of Longitudinal Cracks

161

Table 1: predicted TOFs and Crack location


Crack
locations

Frequency
range (GHz)

Calculated TOF
(ns)

Predicted
TOF(ns)

Predicted crack
location (mm)

4180

5.2 ~ 6

62

66.45

4517

4580

5.2 ~ 6

67.2

71.1

4871

4180

5.2 ~5.9

64.94

66.45

4173

4580

5.2 ~5.9

70.47

71.1

4618

3. Conclusions
In this study, a NDT method using high frequency electromagnetic waves is used
to detect a longitudinal crack in a piping system including U-bend. The results show
that the circular TE11-mode is a suitable mode to detect longitudinal cracks. To infer
information about the existence of a crack and its location, two different signal
processing methods are introduced. From our results the response of electromagnetic
waves to the crack is clearly exhibited. The signals in the time domain are obtained by
applying IFFT onto the signals in the frequency domain. TOFs follow quite accurately
by performing two different signal processing methods. Then, we show that by
knowing both TOF and group velocity of electromagnetic waves, the position of crack
can be determined. Although the open crack with depth as same as the thickness of test
pipe was detected in this study, however ,due to this fact that skin depth of microwave
in the inner surface of the pipe is too small (in microscale), the ID crack with depth of
micrometer can be detected by this technique. There was a limitation to make a crack
with width smaller than 0.3 mm, however we looking for a way to create crack with
smaller width. The detection of crack with small length will be investigated in the next
experiment.

References
[1]
[2]
[3]
[4]

[5]
[6]

[7]

P.E. MacDonald, V.N. Shah, L.W. Ward, P.G. Ellison, Steam generator tube failures. NUREG/CR6365, INEL-95/0393 (1996).
M. V. Brook, D. K. Ngoc, J. E. Eder, Ultrasonic Inspection of Steam Generator Tubing by Cylindrical
Guided Waves, Review of Progress in Quantitative Nondestructive Evaluation, 9 (1990), 243-249.
K. Sugawara, H. Hashizume, S. Kitagima., Development of NDT method using electromagnetic waves,
JSAEM Studies in Applied Electromagnetic and Mechanics 10 (2001), 313-316.
H. Hashizume, S. Kitajima, T. Shibata, Y. Uchigaki and K. Ogura, Fundamental study on NDT method
based on electromagnetic waves, ENDE2003, Saclay, Studies in Applied Electromagnetics and
Mechanics 24 (2003), 263-270.
H. Hashizume, T. Shibata and K. Yuki , Crack detection method using electromagnetic waves,
International Journal of Applied Electromagnetics and Mechanics 20 (2004), 171-178
K. Abbasi, S. Ito, H. Hashizume, K. Youki, Crack detection by using electromagnetic waves, EPRI 5th
International conference on NDE in relation to Structural Integrity For Nuclear And Pressurized
Components 2006.
K. Abbasi, S.ito, H.Hashizume, Microwave detection of longitudinal crack in straight pipe, ICONE 15,
Nagoya, 2007.

162

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-162

Some Experiences with Microwave


Investigation of Material Defects
Dagmar FAKTOROV
University of ilina, Faculty of Electrical Engineering, Slovak Republic

Abstract. The paper deals mainly with microwave measurement of deeper cracks in
metal samples while an attempt to describe the crack as a part of the microwave line is
set out. Information about the cracks properties is obtained from the reflected signal
directly measured and also by means of impedance measurement.
Keywords. Non-destructive testing, microwaves, crack depth, waveguides, rust

Introduction
The article is engaged in detection of cracks in metals from the microwave theory
standpoint, and so it tends to basic research. Nevertheless an example about more versatile
using of Maxwells equations is presented on a practical case at the detection of a defect
in metal, a smaller one comparing with the detectors wavelength. On the assumption
starting from the wave theory applied in microwave technique a proof is given that a
seeming unreability of the defect detected in such configuration is in a fact one case of
electromagnetic wave spreading out in the rectifying surroundings.
On the experiments with artificial defects it is demonstrated how general relations for
impedance can be used at the determining of the defect geometry.
On the basis of measurements also the influence of dielectric splits on the measuring
signal is quantitatively presented.

1. Theoretical basis and applied formulae


As to general approach to the problems, Maxwell equations provide the basis to solution
and for the experimental part we have chosen the waveguide technique making use of the
same theoretical basis. For the transversal electric field having a sinusoidal character with
the angular frequency we can write

2 E&

&
2E

2 E&

2 &
E = 0,
x 2 y 2 z 2 c 2
_______________________________________________
+

University of ilina, Univerzitn 8215/1, 010 26 ilina, Slovak Republic; E-mail: faktor@fel.uniza.sk.

(1)

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

163

2
where E& is the phasor vector of electric field intensity,
is the phase constant
=

c
for the transversal electromagnetic (TEM) waves and is the wavelength in free space.
On the assumption that the change of the E& in dependence on coordinate x has the form

&
2E
x

&
= 2 E

where =

(2)

2
is the propagation constant and g is the wavelength in the waveguide,
g

we get

2 E&
y 2

2 E& 2
+ 2 2 E& = 0 .
2

z
c

(3)

For experiments we use transversal electric (TE) waves and they are based on the
reflected signal from defects. Our measurements and calculations are based on this reality
exploiting the waveguide technique, where the complex reflection coefficient & can be
measured and it is given as

& =

E&
,
E& +

(4)

where E& + and E& are intensities of reflecting and incident waves, respectively. When
we take in account expressions of E& + and E& by means of propagation constant we
have

& = & 0 e j(0 + 2x ) ,

(5)

where 0 is the phase of & in the point x = 0 and & 0 is absolute value of & in the same
point. The incident and reflected wave create the standing wave. Standing wave ratio
(SWR) s

s=

&
E
min
&E

(6)

max

&
&.
can be measured and from the E
min position it is possible to determine the phase of
Seeing that & is a complex quantity we can determine complex impedance of defect Z&
like terminative impedance of waveguide in the component form

164

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

Z& = Z& 0

1 &

1 + & 2 & cos

+ jZ& 0

2 & sin
2
1 + & 2 & cos

(7)

where is the angle of & and Z& 0 is the characteristic impedance of waveguide. As all
quantities on the right hand of Eq. (7) are measurable, [1] Z& can be evaluated.

2. Experimental results
The experiments were carried out on the standard laboratory microwave equipment with
the connection in the schematic illustration in Figure 1.
As a source of microwave signal was used the reflex klystron modulated with 1 kHz
signal. The measurements were carried out on frequencies from the ranges X and G band
on the wave TE10 . The measured quantities were detected on the selective amplifier on
the end of the line. The switch enables measuring both SWR and direct reflections in the
same connection.
The measurements of standing wave ratio (SWR) in waveguide were taken with the
switch position to the open waveguide (OW). The SWR was measured for every depth at
each frequency by the standing wave detector.
CL

A CD
CL

SA

FM

KPS

CL

CL

sample

OW

CD
WRS
K
IM

FI

VA

MT

SWD
FC

CD

MSH

Figure 1. Experimental set up for inhomogenities measurement, K reflex klystron, KPS klystron power
supply, IM- impedance match, VA variable attenuator, MT magic T, A adapter, FM frequency meter, FI
ferrite isolator, CL coaxial line, FM frequency meter, WRS waveguide rotation changeover switch,
SWD standing wave ratio measurement line, FC ferrite circulator, CD crystal detector, OW open
waveguide, SA selective amplifier, MSH movable sample holder

Samples were made in such way to be as much as possible similar to the real crack and
simultaneously to provide a possibility for quantitative processing and evaluation. That is,
the samples were made from steel plates 5x4x1,5 [cm]. Their areas should have
overlapped the waveguide cross-section (22,5x10 [mm]) and in every sample there was
filed a slot with the width 1 mm and length 20 mm representing an artificial crack (the
depth of individual cracks are given in Figure 2). These samples were located in front of

165

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

SWR [-]

the empty waveguide without any other termination (open waveguide - OW) at the
distance of 1mm and their longitudinal sections were parallel to the longer side of the
waveguide.
The measured and calculated values are plotted in the Figure 2. The successive
curves for individual defects show quasiresonant course but in fact they represent values
of waveguide terminating impedance in the waveguidedefect contact position. It is
possible to assume, that individual samples with defects at particular frequencies behave
as a quarterwave transformers. The quarterwave transformer effect manifests itself at
g
.
individual frequencies at three multiple of
4
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0
0

f=10,20GHz
f=8,40GHz

8 10 12 14 16 18 20 22
deph of defect [mm]
f=10,03GHz
f=7,70GHz

f=9,61GHz
f=4,92GHz

f=9,20GHz

Figure 2. Dependence of SWR on the defect depth at seven frequencies

For the more complex assessment of the measured results from the point of view of
quantities with which the microwave technique operates the values of complex
impedance were calculated, Eq. (7) and plotted their dependences on the defect depth at
the frequency 9,23GHz, Figure 3.
An illustrative image about impedance course for the defect quarterwave transformer
affords Figure 3, where closed curves belongs to the defect depths

g
4

f=9,23GHz
600

Im{Z}

400

200

0
0

200

400

600

Re{Z}

Figure 3. Lissajouse curve of various depths of defects

, and 3

g
4

166

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

From the Figure 2 it can be seen declining SWR amplitude for n = 1 what shows
that the defect behaves as a loss waveguide section. To point this fact we carried out
another measurement on a purpose-built sample. This sample was manufactured from
two steel plates. The arrangement was adapted for setting desired different lengths widths
and depths of the artificial crack. Thereafter for one series of measurement there was
sideways delimited width and the depths were set for every measuring. Individual depths
could be set accurate to 0,01mm. This arrangement was used at measurements, the results
of which are plotted in Figure 4 and Figure 5.
In order to show to what extends the defect depth can influence the reflected signal
amplitude we took two measurements, Figure 4, where
f=10,1 GHz

100

amplitude [a.u.]

80
60
40
20
1

0
0

20

40

60

80

100

120

140

crack depth [mm]

Figure 4. Dependence of the reflected signal amplitude on the depth of the crack

curve 1: the reflected signal measurement in the lossless waveguide,


curve 2: the reflected signal measurement on the sample with adjustable depth.
The reflected signal was measured through the ferrite circulator, Figure 1 and
measurement were carried out for such position of the piston in the lossless waveguide
which were identical with the corresponding crack depths. The comparison of the both
measurements is in the Figure 4. From this graph it is possible to form a conception about
the decreasing amplitude of the reflected signal at the determining of the crack depth with

(2n + 1)

distant maxima.
4
For the reason of more complex evaluation of the defect character as a special
waveguide section we also followed the shift of the SWR minimum with the enlarging
defect depth. The corresponding values of the complex impedance were calculated from
Eq. (7) and the dependence of complex impedance imaginary component and SWR shift
on the depth of defect is in the Figure 5. From the point of view of microwave theory the
presented results bring an additional proof of the fact that for the defect investigation the
microwave method can be used as well as a tried and tested microwave practice.

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

167

f=10,1 GHz
500

100
0
-100 0

10

20

30

40

50
-5

-300
-500

minimum shift [mm]

Im{Z} []

300

posun minima [mm]

10

-10

crack depth
hbka defektu
[mm][mm]
Im {Z}

posun minima

Figure 5. Dependence of complex impedance imaginary component and standing wave minimum shift on the
depth of defect

With the open waveguide it could be possible to obtain information about the defect
orientation. Changing the angle between the waveguide Hplane and the straight line
passing along the defect we measured the reflected signal amplitude and the dependence
of signal amplitude on angle of defect rotation is in the Figure 6.

amplitude [a.u.]

f=9,23 GHz

90
70
50
0

50

100

angle of rotation []

Figure 6. Dependence of signal amplitude on angle of defect rotation

At defects detection it is necessary to admit that an older crack is partly or wholly


filled with rust or another deposit and also can be covered with paint, rust or with their
combination. Additional crack filling can also be water or various water solutions. These
materials signify from the microwave defectoscopy point of view dielectrics which will
have an influence on the defect viewed as a part of the microwave network.
We followed these conditions experimentally and the obtained results are in Figure 7.
In the Figure 7 there are demonstrated courses of the reflected signal from the defect
gradually being filled with rust layers (width of defect - 1mm, depth of defect - 10mm).
In the Figure 7 we also present for a comparison the curve of the reflected signal course
from the empty crack (curve e). The curve f represents the course of the reflected
signal from the defect filled with pertinax, the curve g from the empty defect covered
with a paint, the curve h from the defect filled with water and the curve i from the
defect filled with paint.

168

D. Faktorov / Some Experiences with Microwave Investigation of Material Defects

f = 9,3 GHz
100

80

80

amplitude
[a.u.]
amplitda [a.u.]

amplitude [a.u.]

amplitda [a.u.]

f = 9,3 GHz
100

60
40
20

60
40
20
0

0
-10

-5

10

-10

-5

10

polohaposition
sondy [mm]
probe
[mm]

poloha
sondy
[mm] [mm]
probe
position

Figure 7. Dependence of reflected signal amplitude from defect gradually filled with the rust layers (a one
layer, b two layers, c three layers, d defect filled with rust, e empty defect) and form the presence
different dielectrics in the volume and on the defect surface

3. Conclusions
Our goal was to find an interface of practical testing knowledge with the theory which is
at disposal in microwave domain. The acquired experiences can be summarized in
several points indicating possibilities of microwave NDT: 1. to find out the defect (with a
waveguide or a coaxial probe), 2. to determine the defect orientation, 3. to obtain
information about the defect width, [2] and about presence dielectrics, 4 to determine the
defect depth (according to the defect impedance), 5. to fix the defect depth utilizing the
quarter-wave transformer effect and the attenuating characteristics, [2].
It is worth also saying that microwaves offer additional possibilities, with regard to
expanding their utilization as well sensibility and accuracy. These goals can be achieved
by using higher frequencies (around 100 GHz) and more sophisticated techniques (e.g.
cavity resonators).

Acknowledgement
The author would like to thank MSc. Pavol irko director of High School for Agriculture
and Fishing in Moovce for technical help at realization of experiments.

References
[1]
[2]

M. Pastorino, A. Massa, S. Caorsi, A Global Optimization Technique for Microwave Nondestructive


Evaluation, IEEE, Transaction on Instrumentation and Measurement, 51, (2002), 666-673.
D. Faktorov: Using of Microwaves at Investigation of Solid Materials Inhomogenities, Conference
proceeding APCNDT 2006, 12th Asia - Pacific Conference on Non-Destructive Testing Auckland,
New Zealand, http://www.ndt.net/article/apcndt2006/index.htm.

Modeling

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-171

171

Effect of Crack Closure on Quantitative


ECT Inspection of Closed Fatigue Cracks
Zhenmao CHEN a1, Noritaka YUSA b Kenzo MIYA b and Hideaki TOKUMA c
a
School of Aerospace, Xian Jiaotong University, China,
b
International Institute of Universality, Japan
c
Nuclear Power Engineering Department, Tokyo Electric Power Company, Japan

Abstract. In this paper, the feasibility to apply a closed fatigue crack as a


substitute of a Stress Corrosion Crack (SCC) is investigated aiming at
applications to ECT crack sizing. Several testpieces of closed fatigue
cracks are fabricated, and ECT signals are measured after selected
bending loads being applied to close the crack. From the measured
signals, the crack profiles are reconstructed by using a deterministic
inversion technique, and the sizing results are compared with the true
crack profiles in order to evaluate the effect of crack closure. The results
reveal that the closing load does not give significant influence on the
crack sizing precision. Therefore, to simulate SCC with a fatigue crack
closed by 3 point bending testing is not a suitable way from the view
point of ECT inversion.
Keywords. Fatigue crack, Crack sizing, Eddy Current Testing, Crack Closure

1. Introduction
In practical NDT applications such as in Performance Demonstration (PD)
activities, various kinds of TestPieces (TP) are necessary for calibrations, inspector
training and etc. TP of artificial Stress Corrosion Crack (SCC) with known profile is
indispensable in PD because the SCC is a major concern of many critical mechanical
structures, e.g. a nuclear power plant. The fabrication of SCC, however, is high cost
and time consuming due to difficulties to control the crack initiation and propagation.
Recently, a strategy using closed Fatigue Crack (FC) as a substitute of SCC is
proposed in the ultrasonic NDT applications because the signal features of an SCC and
some closed FCs are similar [1]. This strategy is promising not only because of the low
fabrication cost, but also because the profile of a fatigue crack is much easier to be
controlled in the fabrication procedure. In several papers the crack closure effect on the
UT signals has been studied [2]-[5]. The influence of crack closure on the ECT signals
is also investigated by some researchers [6],[7],[8]. The feasibility to apply closed FC
as a substitute of SCC in ECT inspection, however, is not clarified yet especially for
the quantitative ECT inspections.
In this paper, effect of crack closure on the quantitative ECT inspection is studied
___________________________________
1 Corresponding author, MOE Key Laboratory for Strength and Vibration, Xian Jiaotong University, 28
West Xianning Road, Xian, 710049, China, Tel/Fax: 86-29-82663973, E-mail: chenzm@mail.xjtu.edu.cn.

172

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

by inspecting and sizing fatigue cracks in different closure state that is adjusted by
using a mechanical loading system. 6 fatigue crack TPs of two sizes are fabricated
through 3 point bending fatigue testing in different loading conditions. Bending loads
from 13kN to 31kN are applied to the TPs to change the closure state of cracks after the
cracks are introduced. A pluspoint ECT sensor is adopted to inspect the TPs that are
unloaded from the test machine after selected maximum bending load being applied.
From the detected ECT signals, the crack sizes are reconstructed based on a
deterministic ECT inversion technique. All the TPs are destructed after detailed ECT
inspections to observe the true crack profiles. The correlations between the ECT signals,
the sizing results and the bending loads are analyzed through comparing the observed
and reconstructed crack information. The results show that closure states of fatigue
cracks do not give significant influence on the ECT signals and consequently, the
sizing precision. In other words, application of closed FCs to replace SCC is not
suitable in the quantitative ECT inspection in view of that the signal of an actual SCC
is much smaller than that of an FC of similar size.

2. Fatigue crack TPs and ECT inspection system


Two kinds of fatigue crack TPs are fabricated with type SUS316 austenitic
stainless steel. The fatigue cracks in one kind of them are introduced with a relative
larger strain range (tension-tension) and smaller number of loading cycles. The sizes of
these TPs are 400 mm in length, 120 mm in width, and 15mm in thickness. The
maximum bending load and the loading range are 29 kN and 25 kN respectively. The
fatigue tests are terminated at 40,000 (TP L1), 50,000 (TP L2) and 60,000 (TP L3) of
loading cycles respectively. For another kind of TPs, relative smaller loading range and
large number of loading cycles are selected. The loading cycles used for these TPs are
0.52 million (TP S2), 0.88 million (TP S3) and 1.0 million (TP S3) respectively. The
final sizes of the second kind of TPs are 200 mm in length, 100 mm in width and 8 mm
in thickness.
Figure 1 shows the loading system for both the fatigue testing and applying closing
loads to the TPs. In all TPs, the fatigue cracks can be confirmed through visual
observation. In Fig.2, a zoom up view of a fatigue crack initiated at the two ends of the
initial slit is given. The initial EDM slit of 0.5 mm depth, which is introduced for
guiding the location of cracking, is removed by grinder machining after fatigue testing.

Fig.1 Jigs for closing loading and fatigue testing Fig.2 Zoom up of a fatigue crack

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

173

A pluspoint sensor of relative large size is selected for the ECT inspection (C scan).
The sensor is scanned around the crack with a range of 40 mm in width and 60 mm in
length. Three frequencies, 10 kHz, 20 kHz and 50 kHz, are chosen for the ECT
inspection. Figure 3 shows a flowchart of the ECT testing system. After plastic
deformation is introduced to the TP by applying bending load of selected maximum
value for closing the crack, the TPs are unloaded and set to the scanner for ECT and
TOFD (Time of Flight Diffraction, an UT method) inspections. Both the ECT signals
and the position information are inputted to a computer through the A/D converter. The
structure and size of the Pluspoint sensor are shown in Fig.4. In order to correspond
crack depths in wide range, a relative large sensor size is adopted.
Table 1 List of TPs
TP Number

Loading Cycles

TP Thickness (mm)

TP Size (mm)

L1
L2
L3
S2
S3
S4

60,000
50,000
40,000
1,000,000
880,000
520,000

15
15
15
8
8
8

400/120/15
400/120/15
400/120/15
200 /100 /8
200 /100 /8
200 /100 /8

7.5

Side View
side
view

PC

aect2000s

12.5

front
view
Front View

A/D converter
2.5

2.5

7.5

7.5

12.5

12.5

stage contoller
View
tTop
op view

XYZ stage

Fig.3 Flowchart of the inspection system

Fig.4 Structure of the Plus-point sensor

(a) SCC image


(b) FC image (before loading) (c) FC image (after loading)
Fig.5 Comparisons of TOFD images of SCC and closed FC cracks

174

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

Figure 5 shows a comparison of TOFD (Time of Flight Diffraction, an UT


method) inspection results (B scope images) for an artificial SCC TP and the fatigue
crack TP L1. Due to bridging between the crack surfaces, many echoes appear inside
the crack region in case of the SCC TP (Fig.5 a). For the fatigue crack, however, there
is no such echo recognized before closing load being applied (Fig.5 b). After applying
plastic deformation to close the crack, however, similar echoes appear inside the
fatigue crack region. These results demonstrate that to simulate SCC with FC is valid in
case of UT (TOFD) inspection.

3. Inspection Results of Eddy Current Testing


Figure 6 depicts a typical ECT C scan signal for the fatigue crack TP L1 to show
the quality (high S/N ratio) of inspection signals. In Fig.7 and Fig.8, the ECT signals
measured at each loading step are compared for the TP L1 and TP S3 respectively.
Only signals along the crack line are given in the figures. For TP L1, loads of 7
conditions (10kN, 20kN, 22kN, 24kN, 26kN, 28kN and 30kN) are selected to adjust
the closure state of crack. ECT signals are measured before and after each loading step.
There is no significant change observed in the measured signals for the TP L1.
Figure 8 shows results of the small TP S3 in which a crack has been introduced by
880,000 cycles of fatigue testing. The signal changes due to closing loads are also not
significant though it is bigger than the case of TP L1. From the signals, it is difficult to
extract a simple correlation between the loading value and the change of signal
amplitude. Similar observations are also found for other TPs.

Fig.6 A typical C scan signal of ECT inspection for a fatigue crack

4. Sizing of fatigue cracks with a deterministic optimization strategy


To evaluate the effect of crack closure on the sizing precision, the ECT signals for
fatigue cracks at different loading conditions are applied to reconstruct the crack
profiles for each TP. A deterministic inversion technique the Conjugate Gradient (CG)
method is employed in the sizing procedure [10, 11]. The database approach for fast
simulation of crack signals is adopted to calculate the crack signals in the inverse
analysis [12]. In the sizing procedure, the crack is supposed nonconductive and in a
rectangular or a semi-elliptic shape respectively.

175

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks
0.9

0.35
noload
18kN
20kN
22kN

0.3

0.7

24kN
26kN
28kN
30kN

0.6

0.2
0.15
0.1

0.5
0.4
0.3

0.05

0.2

0.1

-0.05
-0.1
-30

noload
18kN
20kN
22kN

0.8

Signal (V)

Signal (V)

0.25

24kN
26kN
28kN
30kN

-20

-10

10

20

30

-0.1
-30

-20

-10

10

20

30

x (mm)

x (mm)

(a) Real
(b) Imaginary
Fig.7 Comparison of ECT signals at different closing loads (TP L1)
2

noload
12kN
14kN
15kN

1.5

noload
12kN
14kN
15kN

2.5

Signal (V)

Signal (V)

0.5

1.5
1
0.5

0
0

-0.5
-30

-20

-10

x (mm)

10

20

30

-0.5
-30

-20

-10

10

20

30

x (mm)

(a) Real parts


b) Imaginary parts
Fig.8 Comparison of ECT signals measured after different closing loading (TP S3)
As example of sizing results, Fig.9 shows reconstructed crack depths for TP L1
under different conditions of closing loads. The two block lines in the figure denote
results using the rectangular and semi-elliptic crack model respectively. The true crack
depth and length, which are obtained through destructive observation, are depicted in
the figures as dot lines. One can read that the reconstructed crack profiles are in a
satisfactory agreement with the measured ones for all closing loads. The rectangular
crack model gives better predictions for crack depth, while the crack lengths obtained
by using the semi-elliptic crack model are in better agreement with the true value.
Figure 10 gives the comparison of the measured crack signals and the simulated
signals due to the reconstructed crack for TP L1. The signals are in good agreement. In
Fig.11 and Fig.12, the sizing results for another TP (S2) are presented. Similar to the
results of TP L1, the predicted crack depth and length are near the true crack profile
and do not show simple dependence on the load values.
In Fig.13 and Fig.14, reconstruction results for all the 6 TPs and load values are
summarized. Figure 13 shows a comparison of results for crack depth. The error
between the true and reconstructed crack depth is less than 20%. For crack length, the
sizing precision is much better (Fig.14). These results verify that treating FCs as a
nonconductive notch is reasonable for FC sizing. The results also demonstrate that the
CG method is efficient for profile reconstruction of FCs.
It is well known that the signal of an SCC is much smaller than that of a fatigue
crack of similar size [7], [9]. The results of this study, however, reveal that the closure
state of an FC does not significantly influence the ECT signals. Therefore, we have to
conclude that an FC closed by simple 3 point bending testing is not appropriate to
simulate SCC in applications of crack sizing.

176

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

There are several reasons can be listed to interpret why the ECT signals do not
decrease significantly, e.g., residual stress introduced by the 3 point bending testing
may actually not be in a state closing the whole crack, or simply due to oxidation at the
crack surfaces. Further studies are necessary to clarify these questions [13].
5. Concluding Remarks
From the research work described above, the following conclusions are obtained.
1) The nonconductive crack model and the deterministic inverse analysis strategy
are feasible for the reconstruction of fatigue cracks. The sizing error is
acceptable for the crack depth and is much better for sizing of the crack length.
2) The maximum load of 3 point bending does not significantly influence the
crack signals and the consequent crack sizing results.
3) The fatigue crack closed by applying plastic deformation with 3 point bending
testing is not suitable to simulate SCC in the ECT crack sizing application.
Acknowledgements
This work was supported in part by the National Natural Science Foundation and
National Basic Research Program of China through Grand No.50677049,
No.2006CB601206 and No. 2007CB707702, and Program for New Century Excellent
Talents in University.
Rectangular

Rectangular

Elliptic

Crack Length (mm)

Crack depth (mm)

Elliptic

12

2.5

1.5
1

0.5

10
8
6
4
2
0

0
0

18

20

22

24

26

29

30

31

18

20

22

24

26

29

30

31

Load (kN)

Load (kN)

(a) Results for crack depth


(b) Results for crack length
Fig.9 Reconstructed crack profiles for the TP L2
80

80

Measured (real)
Reconstructed (real)
Measured (imag)
Reconstructed (imag)

60

40

Signal (V)

Signal (V)

40

20

20

-20

-20

-40

-40

-60
-0.03

Measured (real)
Reconstructed (real)
Measured (imag)
Reconstructed (imag)

60

-0.02

-0.01

x (m)

0.01

0.02

0.03

-60
-0.03

-0.02

-0.01

0.01

0.02

x (m)

(a) Rectangular crack model


(b) Semi-elliptic crack model
Fig.10 Comparison of true and reconstructed crack signals (TP L2)

0.03

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

Rectangular

Rectangular

elliptic

elliptic

40

35

C ra ck d epth (m m )

C rack leng th (m m )

5
4
3
2
1

30
25
20
15
10
5
0

0
0

12

14

15

12

14

15

Load (kN)

Load (kN)

(a) Results for crack depth


(b) Results for crack length
Fig.11 Reconstructed crack profiles for different closing load (TP S2)
0.5
Reconstructed(squre Re)
Reconstructed(squre Im)
Reconstructed(ellipse Re)
Reconstructed(ellipse Im)

0.4

Measured(Re)
Measured (Im)

Signal (mV)

0.3

0.2

0.1

-0.1

-0.2
-0.03

-0.02

-0.01

0.01

0.02

0.03

x (m)

Fig.12 Comparison of true crack signals and that due to reconstructed crack

Reconstructed (mm)

elliptic
Rect.

True (mm)

Fig.13 Reconstruction results for crack depth for all TPs and load values

177

178

Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks

Reconstructed (mm)

elliptic

Rect.

True (mm)

Fig.14 Reconstruction results for crack length for all TPs and load values
References
[1] Savahn PH, Hogberg K, Defect simulation for interdendritic stress corrosion cracks in Alloy 182 welds,
CD-ROM Proceedings of the 16th World Conference on NDT (available at NDT.net).
[2] Clark R, Dover WD, Bond LJ. The effect of crack closure on the reliability of NDT predictions of crack
size. NDT&E international, 20(1987), 269-275.
[3] Salam Akanda MA, Saka M. Relationship between closure stress of small fatigue crack and ultrasonic
response. J. Nondestr. Eval. 23(2004), 37-47.
[4] Buck O, Thompson RB, Rehbein DK. The interaction of ultrasound with contacting asperities:
applications to crack closure and fatigue crack growth. Journal of Nondestructive Evaluation, 4(1984),
203-212.
[5] Mihara T, Nomura S, Akino M, Yamanaka K. Relationship between crack opening behavior and crack
top scattering and diffraction of longitudinal waves. Materials Evaluation, 62(2004), 943-947.
[6] Kurokawa M, Kamimura T, Fukui S. Relationship between electric properties and width of cracks of
Inconel alloy. In: Proceedings of the 13th International Conference on NDE in the Nuclear and Pressure
Vessel Industries, Kyoto, Japan, 1995, 261-265.
[7] Villone F, Harfield N. Simulation of the effects of current leakage across thin cracks. Electromagnetic
Nondestructive Evaluation (IV). S.S. Udpa, T. Takagi, J. Pavo and R. Albanese (Eds.). IOS Press
(1999), 79-86.
[8] Badics Z, Matsumoto Y, Aoki K, Nakayasu F, Kurokawa A, Finite element models of stress corrosion
cracks (SCC) in 3-D eddy current NDE problems. Nondestructive Testing of Materials. R. Collins, W.D.
Dover, J.R. Bowler and K. Miya (Eds.). IOS Press (1993), 21-29.
[9] Ohshima, K., Hashimoto, M., Research on numerical analysis modeling of SCC on eddy current testing.
Journal of the JSAEM, 10 (2002), 384-388.
[10] Chen Z and K.Miya, ECT inversion using a knowledge based forward solver, Journal of Nondestructive
Evaluation, 17(1998), 167-175.
[11] Yusa N, Chen Z, Miya K, Uchimoto T, Takagi T. Large-scale parallel computation for the
reconstruction of natural stress corrosion cracks from eddy current testing signals, NDT&E international,
36(2003), 449-459.
[12] Chen Z, Miya K. and Kurokawa M., Rapid prediction of eddy current testing signals using A-Phi
method and database, NDT&E International, 32(1999), 29-36.
[13] Yusa N., Perrin S., Mizuno K., Chen Z. and Miya K. , Eddy current inspection of closed fatigue and
stress corrosion cracks, Meas. Sci. Technol. 18(2007), 3403-3408.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-179

179

Toward the Reconstruction of Stress


Corrosion Cracks Using Benchmark Eddy
Currents Signals
Maxim MOROZOV a, Guglielmo RUBINACCI b, Antonello TAMBURRINOc,11,
Salvatore VENTRE c and Fabio VILLONE c
a
CREATE Consortium, Naples, Italy
b
Ass. EURATOM/ENEA/CREATE, DIEL,
Universit degli Studi di Napoli Federico II, Italy
c
Ass. EURATOM/ENEA/CREATE,DAEIMI, Universit degli Studi di Cassino, Italy

Abstract. This paper concerns numerical modelling of stress corrosion cracks in


Inconel600 plates using experimental data of benchmark eddy current
measurements. The problem is considered from a broad perspective, in view of its
integration with algorithms for solving the inverse problem. The accuracy and
efficiency have been particularly considered. The computational technique applied
for modeling the crack is based on an integral formulation of the eddy current
problem. The cracks are treated as either penetrable volumetric or zero-thickness
(surface) defects enabling some electric current flowing through them. Real stress
corrosion crack have been considered in this study where measured and simulated
eddy current signals due to the cracks are presented.
Keywords. Eddy currents, nondestructive evaluation, stress corrosion cracks,
numerical simulation.

Introduction
This paper is in the framework of Electromagnetic Non-Destructive Evaluation
(ENDE) of real defects by means of conventional Eddy Current (EC) instrumentation.
The main objective of the present work has been to validate an original method for
modeling the electromagnetic response of stress corrosion cracks (SCC) which have
certain conductivity and therefore enable some electric current flowing through them
[2]. The numeric method is based on an integral formulation in terms of a twocomponent current density vector potential expanded over edge-elements [3]. The
exploitation of superposition and a proper choice of the current density degrees of
freedom gives rise to a very efficient numerical implementation. As discussed in [4], a
fast numerical method for the forward problem is essential to reconstruct the defect
with an iterative method minimizing the discrepancy between the simulated and
measured EC signals. Experimental data comprising eddy current responses to several
SCC flaws and fatigue cracks (FC) in Inconel600 plates, as well as the respective crack
profiles found by destructive metallographic examination, have been offered as a
1

Corresponding Author: Antonello Tamburrino, DAEIMI, Universit degli Studi di Cassino, V. G. di Biasio
43, Cassino, 03043-Italy; E-mail: tamburrino@unicas.it

180

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

benchmark to the scientific community by a research team of International Institute of


Universality, Tokyo, Japan [5]. The experimental data are available for various EC
probes, however in this paper we focus on the modeling of SCC flaws from signals
obtained with an absolute type pancake coil. Reconstruction of thin FC flaws has been
discussed in [6]. Reconstruction of SCC flaws presents actual challenge due to their
volumetric nature, complex geometry and conductivity distribution within a crack [7,
8].

1. Numerical method: Forward analysis for partially conducting cracks


The numerical method consists in an integral formulation of the eddy currents problem
in terms of a two-component electric vector potential [9]. This integral formulation
allows us to discretise only the conducting domains where the eddy currents are
induced and automatically enforcing regularity conditions at infinity. In addition, the
introduction of the electric vector potential T, such that the eddy currents density is
J=uT, ensures that J is solenoidal and the choice of the two-component gauge
minimizes the number of discrete unknowns required. The equations to be solved are
the standard eddy current equations in the frequency domain. The electric field is:
E = jZAM

(1)

where M is the scalar electric potential and A is the magnetic vector potential given by:
A(x, t )

P0
4S

J ( x' , t )
dV '  A 0 (x, t )
x  x'
VC

(2)

where P0 is the magnetic permeability of the vacuum, VC is the conducting domain and
A0 is the contribution due to the external current density. The integral equation is, then,
obtained by combining (1) and (2) with the constitutive equation KJ E in VC , K being
the electrical resistivity.
From the numerical point of view, the formulation is solved using finite elements:
a mesh of VC is given, and an edge element basis functions Nk is introduced for T:
T

I N
k

u Nk

(3)

It is worth noting that the choice of edge elements allows us to enforce the right
continuity conditions of the various electromagnetic quantities and to impose easily
both for the gauge and boundary conditions. The numerical model is, finally, obtained
by imposing the constitutive relationship in weak form by using the Galerkin approach:

Vc

u N k (KJ  jZ A)dV

0 N k

(4)

The term involving the electric scalar potential gives no contribution thanks to the
solenoidality of the test function. Finally, using (2) we obtain:

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

ZI=U

181

(5)

Where, Z=R+jZL, I = ^Ik`, U = ^Uk` and


L ij

R ij

P0
4S

u N i ( x) u N j ( x ' )
x  x'

VC VC

dV dV '

K u N j dV

(7)

 u N i jZ A 0 dV

(8)

uN

VC

Ui

(6)

VC

In modeling SCC, it is important to exploit that the defect occupies, usually, a small
volume VD of the conducting domain VC. Therefore, the eddy current density J undergo
a local perturbation G J in a neighborhood of VD requiring, from the computational
viewpoint, a small and local mesh in a neighborhood of VD.
Let K, K0 and 'K be the total, background and perturbation resistances,
respectively: K=K0+'K: and let R, 'R and R0 be the corresponding matrices defined in
(7). From (5), it follows that the eddy current perturbation can be numerically
computed by solving:

Z 0  'R GI
where Z 0

'RI 0

R 0  jZL , I 0 k

(9)

u N

J 0 dV and J0 is the unperturbed current

VC

density that, for canonical geometries, can also be computed analytically (see [10] for
the analytical calculation in planar geometries). The impedance variation due to the
presence of the flaw can be expressed as
GZ = UTGI/Is2

(10)

where U is defined in (8) and Is is the impressed current flowing in the excitation coil.
There are situations that may be encountered in practice where one dimension of
the defect is negligible. This type of defects is termed as zero-thickness (surface)
defects. Equations (9) and (10) are still valid but 'R, involving an integral extended to
the volume VD of the defect, must be defined in the limit of zero thickness defect [11].
Specifically, let d and 6D be the thickness and the surface of the defect. In the limit
do0, matrix 'R (and, consequently GI ) is vanishing if 'K is finite. On the other hand,
if 'K=+f we have a perfectly insulating defect. Therefore, we assume that do0 and
d'K=const.
Under this condition, it follows that [11]
'R ij o d'K ( u Ti n )( u T j n ) dS
6D

(11)

182

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

Finally, we assume that the discretization is such that 6D is the union of facets of the
finite elements mesh. Finally, the numerical model (9), regardless the type of anomaly
that can be either volumetric or surface, can be significantly improved in term of
efficiency by introducing the concept of tentative region, that an a priori known region
VT where the defect is contained (VDVT) and the Woodburys algorithm [1], [4].

2. Experimental data and samples


The studied EC test specimens represent Inconel600 plates with stress corrosion cracks
[5]. The electric resistivity K0 of Inconel600 is assumed to be 1 P:m and its relative
magnetic permeability Pr | 1. The dimensions of the specimens (mm) are given in
Figure 1. SCC was produced into the plates by loading the plate with three-point
bending and immersion into polythionic acid solution. SCC fabrication conditions are
given in Table 1. EC signals due to the SCC were obtained with an absolute type
pancake coil probe, shown in Figure 2. The excitation frequency of EC testing with the
pancake coil was 100 kHz and its lift-off above the surface of a sample was 1 mm.
After EC testing, crack profiles were found by destructive metallographic examination.
A cross section of specimen SCC4 is shown in Figure 3. Since by conditions of the
benchmark test neither the excitation current flowing through the coil, nor the phase
shift and amplification of the measured signals are known, the measured results have
been calibrated with respect to an artificial notch produced by Electrical Discharge
Machining (EDM).

Figure 1. Specimen layout (dimensions in mm)

Figure 2. Absolute pancake coil (dimensions in mm)

Table 1. SCC flaws fabrication conditions


Specimen

Duration
(hours)

SCC1
SCC3
SCC4
SCC5

75
50
50
165

Crack
Length
(mm)
14
12
21
29

Figure 3. Metallographic cross section of specimen SCC4

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

183

Figure 4. Measured and simulated eddy current signal obtained by a pancake coil due to an EDM notch. 2D
indicates that the crack was considered as a zero-thickness defect whereas 3D stands for volumetric model

Figure 5. Dependence of crack signal on crack resistivity (signals have been normalized to the absolute value
of the crack signal when the crack resistivity is infinite)

The EDM notch has a rectangular profile of 10 mm in length, 0.3 mm in width, and 5.0
mm in depth. The calibration process consists in finding a magnitude scaling factor and
an appropriate phase shift at which the numeric simulation result is in agreement with
the measured signal for the EDM notch. Then, the same scale factor and phase shift
found by calibration is applied to measured signals for the SCC flaws.

3. Results of SCC flaws modeling


The volumetric cracks search regions have been modelled as regular parallelepipeds
with width of 0.3 mm for EDM notch and 1.5 mm for SCC flaws. In all the cases a
flaw comprised two mesh elements in the transversal direction. Results of simulation of
EC signals for volumetric cracks model, hereafter termed as 3D, have been compared
with corresponding signals obtained by simulating cracks as zero-thickness defects
(width | 0, see [6]), hereafter termed as 2D. The calibration signal (represented as the
real and imaginary components) due to the EDM notch is shown in Figure 4, with a
magnitude scaling factor and an appropriate phase shift being applied to the measured
result in order to bring it to agreement with the simulated signal. Error of 9.4% between
volumetric flaw simulation (3D) and measured signal occurs due to background noise
in the measured signal (low-pass filtering was applied to reduce this noise).

184

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

(a)

(d)

(b)

(e)

(c)

(f)

Figure 6. Crack profiles in specimens SCC1 (a) and SCC4 (d) and respective EC signals: real (b) and
imaginary (c) parts for SCC1, real (e) and imaginary (f) parts for SCC4

Discrepancy between zero-thickness flaw simulation (2D) and measured signal is big
which might be due to significant thickness of the EDM notch (0.3 mm). The
magnitude scale factor and phase shift found by calibration on the basis of 3D
simulation are maintained when numerically reconstructing the SCC flaws. However,
in contrast to completely non-conducting EDM notch, natural SCC flaws have certain
conductivity and therefore enable some electric current flow across their surface [2,7,8].
Consequently, partial conductivity of a crack, denoted by grey facets, should be
introduced when reconstructing fatigue cracks. Dependence of a crack signal on the
crack resistivity is shown in Figure 5, with the response values being normalised to the
absolute value of the crack signal when the crack resistivity is infinite. The cracks
resistivity Kc = 2.5 P:m corresponds to the optimum correlation between SCC4 flaw
profile and EC signal (Figure 6 e, f). Metallographic profiles of various SCC flaws and
the respective numerically reconstructed profiles, as well as comparison of the
corresponding measured and simulated EC signals are given in Figures 6-7. The figures
representing SCC flaws profiles (Figures 6a, 6d, 7a) show discretisation of crack search

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

185

(a)

(b)

(c)

Figure 7. Profile of SCC5 flaw (a) and the respective EC signals: real part (b) and imaginary part (c)

regions in the crack plane, where white elements correspond to undamaged material,
light grey elements belong to the crack and have resistivity Kc = 2.5 P:m, dark grey
elements belong to the crack and have zero conductivity. The thick line approximately
denotes cracks metallographic boundaries. The value Kc = 2.5 P:m follows from the
experience on previous SCC whereas the perfectly insulating elements (dark grey
elements) have been individuated by a search and trial approach.

4. Conclusions and outlook


A numerical method has been presented for simulating EC signals of partially
conductive volumetric cracks in conductive materials. The method enables fast and
accurate reconstruction of fatigue cracks on the basis of measured signals obtained with
conventional EC instrumentation. Moreover, experimental tests show that a SCC can
be modeled as a region of appropriate resistivity that, eventually, may be spatially
varying.
The future development will address:
x reconstruction of defects using signals obtained with EC probes of more complex
arrangement, such as uniform EC probe [7];
x integration of the numerical model with an inversion algorithm for automated crack
reconstruction.

Acknowledgements
The experimental data have been kindly provided by Dr. Noritaka Yusa of International
Institute of Universality, Tokyo, Japan.
This work was supported in part by the Italian Ministry of University (MIUR) under a
Program for the Development of Research of National Interest (PRIN grant #

186

M. Morozov et al. / Toward the Reconstruction of Stress Corrosion Cracks

2004095237) and in part by the CREATE consortium, Italy.

References
[1]

M. Morozov, G.Rubinacci, A.Tamburrino, S.Ventre, Numerical Models of Volumetric Insulating


Cracks in Eddy-Current Testing With Experimental Validation, IEEE Trans. Mag., Vol. 42, No. 5,
May 2006, pp. 1568-1576
[2] N. Yusa, Z. Chen, K. Miya, T. Uchimoto, T. Takagi, Large-Scale Parallel Computation For The
Reconstruction Of Natural Stress Corrosion Cracks From Eddy Current Testing Signals, NDT&E Intnl.
36 (2003) pp. 44959
[3] R. Albanese, G. Rubinacci, A. Tamburrino, F. Villone, Phenomenological approaches based on an
integral formulation for forward and inverse problems in eddy current testing, Int. J. of Applied
Electromagnetics and Mechanics, Vol. 12, No. 3-4/2000, pp. 115-137
[4] R. Albanese, G. Rubinacci, F. Villone, An integral computational model for crack simulation and
detection via eddy currents, J. of Comp. Phys., Vol. 152, pp. 736-755 (1999)
[5] N. Yusa, L. Janousek, Z. Chen, K. Miya. Diagnostics of stress corrosion and fatigue cracks using
benchmark signals, Materials Letters 59 (2005), 3656-3659
[6] M. Morozov, G. Rubinacci, S. Ventre, F. Villone, Reconstruction of Fatigue Cracks Using Benchmark
Eddy Currents Signals, in ENDE, Electromagnetic Non-destructive Evaluation (X), S. Takahashi and
H. Kikuchi (Eds.), pp. 267-274, IOS Press, 2007.
[7] N. Yusa, L. Janousek, M. Rebican, Z. Chen, K. Miya, N. Dohi, N. Chigusa and Y. Matsumoto,
Caution When Applying Eddy Current Inversion To Stress Corrosion Cracking, Nucl. Eng. Des. 236
(2006) pp. 211-221
[8] N. Yusa, H: Huang, K. Miya, Numerical Evaluation of The Ill-Posedness of Eddy Current Problems to
Size Real Cracks, NDT&E Intnl. 40 (2007) pp. 185-191
[9] R. Albanese and G. Rubinacci, Finite element methods for the solution of 3D eddy current problems
in Advances in Imaging and Electron Physics, Peter W. Hawkes (ed.), vol. 102, (Academic Press), 1998,
pp. 1-86.
[10] C.V. Dodd, W.E. Deeds, Analytical Solution to Eddy-Current Probe-Coil Problems, J. Appl. Phys.,
Vol. 39, No. 6, 1968, pp. 2829-2838
[11] F. Villone, Simulation of Thin Cracks with Finite Resistivity in Eddy Current Testing, IEEE Trans.
on Magnetics, vol. 36, no. 4, July 2000.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-187

187

Evaluation of Subsurface Cracks In


Riveted Aluminium Joints Using Industrial
Eddy Current Instrumentation
Maxim MOROZOV a1 , Guglielmo RUBINACCI b, 1, Antonello TAMBURRINO c and
Salvatore VENTRE c
a
CREATE Consortium, Naples, Italy
b
Ass. EURATOM/ENEA/CREATE, DIEL,
Universit degli Studi di Napoli Federico II, Italy
c
Ass. EURATOM/ENEA/CREATE,DAEIMI, Universit degli Studi di Cassino, Italy

Abstract. This paper concerns the efficient numerical modeling of subsurface


flaws emanating from a fastener hole in riveted aluminum joints on the basis of
eddy current (EC) signals measured with industrial instrumentation. Modeling of
EC signals from modern industrial instrumentation is a difficult task because of the
typical complex (in term of geometry and materials) configuration of the probes.
The computational technique we applied is based on an integral formulation of the
eddy current problem in the presence of magnetic (the core of the probe) materials.
Measured and simulated eddy current signals due to the second-layer artificial
flaws of various profiles are compared.
Keywords. Eddy currents, nondestructive evaluation, rivet joint, ferromagnetic
core, numerical simulation.

Introduction
Fast and reliable evaluation of minute cracks deeply buried beneath rivet heads in lapjoints is an important issue of the aircraft maintenance process. This paper concerns
Electromagnetic Non-Destructive Evaluation (ENDE) of subsurface flaws in aluminum
sandwiches on the basis of measured signals obtained with industrial Eddy Current
(EC) instrumentation.
This paper addresses the numerical simulation of the response due to a crack. The
main objective is to experimentally validate an original method for modeling EC
signals arising from industrial instrumentation. This is a challenging problem because
industrial EC probes present higher sensitivity at the price of a complex structure in
terms of geometry and materials. Moreover, due to the skin effect, the inspection for
subsurface defects in aluminum must be conducted at low excitation frequencies in the
range of few kHz [1, 2]. At our knowledge, this work is one of the first attempt
available in literature to model an industrial EC probe.
The numeric method is based on an integral formulation in terms of a twocomponent current density vector potential expanded over edge-elements [3]. The
1
Corresponding Author: Guglielmo Rubinacci, Dipartimento di Ingegneria Elettrica, Universit degli Studi
di Napoli Federico II, Via Claudio, 21 80125, Napoli, Italy; E-mail: rubinacci@unina.it

188

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

exploitation of superposition and a proper choice of the current density degrees of


freedom gives rise to a very efficient numerical implementation.

1. Numerical method
1.1. Thin insulating crack
The present method consists in an integral formulation of the eddy currents problem in
terms of a two-component electric vector potential [3]. This approach has a number of
advantages as follows. Using an integral formulation allows us to discretise only the
conducting domains where the eddy currents are induced, automatically enforcing
regularity conditions at infinity. The introduction of the electric vector potential T,
such that the eddy currents density is J=uT, ensures that J is solenoidal. The choice
of the two-component gauge minimizes the number of discrete unknowns required.
The equations to be solved are the standard eddy current equations in the frequency
domain. The electric field is:
E = jZA M

(1)

where M is the scalar electric potential and A is the magnetic vector potential given by:

J ( x' )
P0
dV '  A 0 (x)

4S V x  x'

A ( x)

(2)

where P0 is the magnetic permeability of the vacuum, Vc is the conducting domain and
A0 is the contribution of the external coil currents. From the numerical point of view,
the formulation is solved using finite elements: a mesh of Vc is given, and an edge
element basis functions Nk is introduced for T:

I N
k

u Nk

(3)

On the one hand, the choice of edge elements allows us to enforce the right continuity
conditions of the various electromagnetic quantities; on the other hand, their properties
are fully exploited both for the gauge and boundary conditions imposition. The electric
constitutive equation is imposed in weak form using Galerkin approach:

Vc

u N k (KJ  jZ A)dV

0 N k

(4)

where K is the resistivity. The term involving the electric scalar potential gives no
contribution thanks to the solenoidality of the test function. Using (2) we have:

(R+jZL) I = jZQi
where I = ^Ik`, Q = ^Qk`, i=^ik`is the vector of the external coil currents and

(5)

Lij

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

189

u N i ( x ) u N j ( x' )
P0
dV dV '

4S V V
x  x'

(6)

Rij

uN

K u N j dV

Vc

Qik

1
u N i A 0 k dV
ik Vc

(7)

(8)

Supposing that the crack has a negligible thickness, it can be schematised as a surface
(not necessarily planar), discretised via a set of finite element facets (defect pixels),
where the normal component of the current density must vanish. In order to reduce the
computational load, and exploiting linearity, we use superposition: the total current
density is the sum of the solution computed in absence of crack (unperturbed solution
J0) plus the perturbation GJ due to the presence of the defect (J = J0 + GJ). In
particular, on the insulating crack surface, since the total current density normal
component must be zero, we impose that
J n = 0 GJ n = J0 n

(9)

where n is the unit normal to the crack. This approach offers the great advantage that J0
can be calculated either analytically, or numerically using the scheme described above
on a mesh that does not depend on the crack geometry. Conversely, when solving for GJ
the mesh must account for the crack only, so that the mesh refinement is required only
close to the crack, regardless of the position of the exciting source. Due to the properties
of edge elements, the set GG of perturbation currents crossing the crack facets (that must
be equal the unperturbed currents G0) can be written as [4]:

GG = P GI

(10)

where GI are the coefficients of the expansion of GJ in terms of edge elements, and P is
a (m,n) sub-matrix of the edge-facet incidence matrix with coefficients 0, +1 or -1. The
degrees of freedom of the edge element expansions are in fact related to the line
integrals of T along the edges, and the circulation of T along a closed line gives the total
current (flux of uT) linked with the line. We then make a change of variables [4]:

GI = K GX + P+ GG

(11)

where K is a (n,n-m) matrix given by an orthonormal basis for the null space of P, P+ is
the pseudo-inverse of P, and GX is a new set of unknowns, providing no net current
flowing through the crack. Galerkins procedure in terms of these new variables yields:
KTZK GX = KTZ P+ G0

(12)

190

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

solving which we have GX, and hence GI from (11).


Knowing GI it is possible to compute the reaction field. In particular, the impedance
changes of the probe coils are given by

jZ
[(Q k )T GI l ik  (Q l )T GI k il ] ,
2ik il

GZ kl

(13)

where Qk and is the kth column of Q, ik and il are unitary currents for all k and l.

1.2. Treatment of magnetic materials


Many conventional EC probes contain ferromagnetic cores. Usually the magnetic fields
encountered in EC testing are low and linearity of the magnetic material can be
assumed, resulting in a significant simplification of numerical simulation. Applying
superposition and using conventions defined in (3), (6), (7) and (8), the problem at
steady current becomes [5]:
(R+jZL) I + jZF M =jZQi

(14)

1  F m
P 0
D  E M  F T I
Fm

Ni

(15)

where M=^Mk` is magnetization vector, Fm=Pr-1 is the magnetic susceptibility


(constant after linearization), i=^ik`is the vector of the external coil currents and:
Fij

Eij

P0
4S

u Tj (x) Pi u (x  x' )

Vc Vm

P0 Pi Pj dV 
Vm

Dij

Pi P j dV

N ik

1
ik

Vm

Vm

x  x'

P0
4S

wVmi wVmj

dV dV '

(16)

Pi (x) n Pj (x' ) n'


x  x'

dS dS '
(17)

(18)

Pi B 0k dV

(19)

in which wVmi is the surface bounding the element where the shape function Pi is
located and B 0k is the magnetic induction produced in the vacuum by the k-th coil. In
particular, the magnetization vector is supposed to be piecewise constant, so that Pks
are unit vector pulse functions: M = k Mk Pk
Hereafter

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

191

(R+jZL*) I = jZQ*i

(20)

L*= L + F S FT,

(21)

Q*= Q + F S N,

(22)

where

Fm
1  F m P 0

Fm
D 
E

1  F m P 0

1

(23)

In this way the problem is put in the same form as in section 1.1, with the only
difference that L* must be considered instead of L and Q* instead of Q. Hence, the
techniques described above can be used also in this case. If linearization cannot be
applied, it is possible to compute the field variations due to defect presence as the
numerical difference between solutions obtained for unperturbed and perturbed case
using the same mesh (same shape and weight functions). In spite of that the
perturbations are substantially smaller than the whole solution, this method allow to
obtain results of good quality, because using the same mesh systematic errors are
eliminated [6,7]. For linear problems, the method is obviously equivalent to
superposition. We finally notice that, in the presence of magnetic materials, the
impedance changes of the probe coils are given by

GZ kl

jZ
[(Q*k )T GI l ik  (Q*l )T GI k il ]
2ik il

(24)

2. Experimental setup and samples


The samples are riveted aluminium-to-aluminium two layer sandwiches with
countersink fastener holes. The plates are 200u200 mm2 large and 2 mm thick. The
fastener hole has a diameter of 4 mm. The layers of the sandwiches are electrically
insulated and this is taken into account into the numerical model by meshing separately
the two layers. The samples contain Electric Discharge Machined (EDM) notches
emanating from the fastener hole. The inspection cases studied represented cracks lying
in the second layer under rivet head. The EDM notches have rectangular profiles of
0.15 mm in width, 2 mm in depth (passing through the second layer plate) and length
of 5 and 3 mm.
In order to detect subsurface flaws, a sliding probe for fastener inspection was used
at low excitation frequency of 1kHz. The probe is a dual element reflection probe with
a complex ferrite magnetic circuit [1, 2]. The EC inspection was performed by robotic
scanning of samples along straight lines containing both the fastener hole and an EDM
flaw with the sensitive axis of the sliding probe being oriented along the scanning path.
The experimental setup is shown in Figure1. EC signals from the sliding probe were
measured by Phasec2d EC instrument [8]. Both the robotic arm and EC instrument
were controlled by a PC via RS-232 interface.

192

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

3. Results and discussion


Figure 2 shows a fragment of the finite elements mesh and the crack search region,
where white facets have zero conductivity (3 mm long flaw) and black facets represent
undamaged material.
The EC instrument Phasec2d measures EC signals in relative units. In order to
convert measured signals to Ohms the instrument output has been calibrated with
respect to difference between signals from probe placed on a defect-free aluminum
sandwich and probe away from a conductor.
Figure 3 represents measured and calculated EC response to a fastener hole.From
the measurements we have subtracted the signal coming from the probe in absence of
the specimen. Figures 4 and 5 represent flaw contributions of the EC responses to flaws
emanating from a fastener hole in the second layer. The flaw contribution signal is
defined as the difference between the signal coming from a hole with a rivet and the
signal for the same structure including a flaw. The flaw contribution, obtained as a
difference between two signals of the same order of magnitude, is thus quite sensitive
to noise. Moreover, the measurements setup is critical because of the need of using
relatively low frequencies (1kHz). Nonetheless, there is a reasonably good agreement
between experimental results and simulations. Since the probe contains a ferromagnetic
core, a possible source of discrepancies is the estimate of the relative permeability of its
core. Moreover, another critical parameter is the lift-off. Both parameters have been
assessed on the basis of a comparison of the measurements with the numerical model
of the reference design of the probe. A deeper analysis is therefore needed in view of a
possibly more accurate simulation of the response of the probe. Finally, another source
of discrepancies is the finite elements mesh discretization that, especially for the
unperturbed solution, requires a large number of unknowns because of the probe
geometry and materials. In this work, taking into account the constraints for a standard
PC (CPU time and memory), we have optimized the mesh for a total of 8842 unknowns
for computing the unperturbed field and 9498 unknowns for computing the effect due
to the defect.

4. Conclusions and outlook


A numerical method has been presented for simulating EC signals of thin cracks deeply
buried beneath rivet heads in aluminum lap-joints. The method enables accurate
modeling of artificial flaws on the basis of measured signals obtained with industrial
EC instrumentation. The future development will address implementation of an
inversion algorithm for automated crack reconstruction.

Acknowledgements
The technical data of the sliding probe have been kindly provided by Hocking NDT
Ltd.
This work was supported in part by the Italian Ministry of University (MIUR) under a
Program for the Development of Research of National Interest (PRIN grant #
2004095237) and in part by the CREATE consortium, Italy.

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

RS-232

193

Robotic
Arm

Sliding
Probe

PC

RS-232
Phasec2D

Figure 1. Measurement setup.

Figure 2. The finite elements mesh (gray) and crack search region:
white facets have zero conductivity (flaw) and black facets represent undamaged material

Figure 3. Response to a fastener hole with rivet, measurements vs. simulation: left - real component and right
- imaginary component

194

M. Morozov et al. / Evaluation of Subsurface Cracks in Riveted Aluminium Joints

Figure 4. Measured and simulated crack contribution to an EC signal due to 100% deep (passing) 5 mm long
EDM notch emanating in the subsurface layer from a fastener hole.

Figure 5. Measured and simulated crack contribution to an EC signal due to 100% deep (passing) 3 mm long
EDM notch emanating in the subsurface layer from a fastener hole.

References
[1]
[2]
[3]

[4]
[5]

[6]
[7]

[8]

J. Hansen and N. Thorpe, Low frequency eddy current inspection, Conference Proceedings, NDT
2003, Worcester, UK, BINDT, pp. 147-154
J. Hansen, Back to basics: The eddy current inspection method, Parts 1-4, Insight - Non-Destructive
Testing and Condition Monitoring, Vol. 46, No. 5-8 (2004)
R. Albanese, G. Rubinacci, A. Tamburrino, F. Villone, Phenomenological approaches based on an
integral formulation for forward and inverse problems in eddy current testing, Int. J. of Applied
Electromagnetics and Mechanics, Vol. 12, No. 3-4/2000, pp. 115-137
R. Albanese, G. Rubinacci, F. Villone, An integral computational model for crack simulation and
detection via eddy currents, J. of Comp. Phys., Vol. 152 (1999), pp. 736-755
R. Albanese, G. Rubinacci, F. Villone, Crack simulation in the presence of linear ferromagnetic
materials using an integral formulation, Electromagnetic Nondestructive Evaluation (V), J. Pavo et al.
(Eds.), pp. 16-21, IOS press, 2001.
R. Albanese, R. Fresa, R. Martone, Accurate computation of electromagnetic fields in the presence of
conducting and magnetic material, Int. J. Applied Electromag. and Mech., Vol. 6, 1995, pp. 73-88.
R. Albanese, R. Fresa, G. Rubinacci, Assessment of the Accuracy of Electromagnetic Feld
Calculations for Non Destructive Testing, Electromagnetic Nondestructive Evauation, T. Takagi et al.
(Eds.), IOS Press, 1997, pp. 17-22.
Eddy
Current
Probes
&
Accessories
Catalogue,
GE
Inspection
Technologies,
http://www.geinspectiontechnologies.com/download/products/ec/GEIT-50016EN_ec-probe-hi.pdf

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-195

195

Design of a System for the Long Defects


Detection with Advanced Methods for
Eddy-currents Analysis1
E. Cardelli b , A. Faba b , A. Formisano c , R. Martone c , F.C. Morabito e , M. Papais a ,
R. Specogna a , A. Tamburrino d , F. Trevisan a , S. Ventre d , M. Versaci e
a
University of Udine, via delle Scienze 208, Udine, Italy
b
University of Perugia, via G. Duranti 67, Perugia, Italy
c nd
2 University of Napoli, via Roma 29, Aversa, Italy
d
University of Cassino, via Di Biasio 43, Cassino, Italy
e
University of Reggio Calabria, via Graziella Feo di Vito, Reggio Calabria, Italy
Abstract. The aim of the paper is to highlight some of the innovative methodologies, techniques and systems for non-destructive electromagnetic testing, which
have been developed in the framework of the AMDE project (Applications of Methods of Diagnostics Electromagnetic) partially funded by the Italian Ministry of
University and Research.
In particular, we will present the feasibility design of a suitable excitingreceiving coils conguration able to detect long defects by means of eddy-currents.
To solve the forward eddy-current problem, advanced analysis tools have been developed and validated. In this paper, we will also introduce the approach for numerical simulations in the detection of the surface defects.
Keywords. Non destructive testing, eddy-currents, discrete approaches.

Introduction
The strong international competition forced industrial companies to change dramatically
the manufacturing processes, in order to reduce the overall manufacturing time. One of
the conditions to be satised for this goal is the capability to detect very quickly the
product non-conformities with respect to the assumed standards.
For these reasons, there is a remarkable interest in the techniques for the surface
defects detection during the hot mill rolling process of the steel bars (bars with circular cross-section and diameter that ranges from 8 to 80 mm, a longitudinal speed that
changes from 5 to 100 m/s and a temperature from 800 to 1200 C).
The capability to detect these defects permits a fast and straightforward quality assessment of the product and provides the possibility to reduce those non-conformities
due to a wrong set-up of the manufacturing process parameters. The defects considered
1 This work has been supported by the Italian Ministry of Education - Scientic Program PRIN 2004-2006
AMDE (Application of Methods of Diagnostics of Electromagnetics).

196

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

have a depth ranging from 0.1 mm to 2 mm and, even though they have quite different
shapes and sizes, they generally correspond to an interruption of the material continuity
(also from the electrical point of view) and lay along an almost radial direction. Two
main categories of surface defects can be considered depending on their axial length L:
the short defects, with L ranging from 1 mm to 20 mm and the long defects with L
from a meter to tens of meters. In any case, the defect width is much smaller than the
two other dimensions. Short defects can be easily detected using a differential method
in which the signal, after the noise reduction, is compared to a similar signal taken few
centimeters away along the rolling direction. On the contrary, so far, no practical solution
has been found as regards the detection of long defects, for which a differential approach
is not suitable.
The motivation of this paper is to develop the feasibility design of an excitingreceiving coils conguration able to detect the long defects. The numerical simulations
have been performed with a Discrete Geometric Approach [1], [2] based on the so called
A formulation described in [3], [6] and modied in order to represent the effect of
source currents in an integral way. As second tool for numerical simulations and comparisons we used an integral formulation [5].

1. Geometrical model of the detector


The geometry of the detector consists of a conducting AISI 310 steel bar, modeled as
a conducting cylinder Dc . The radius of the bar is 17 mm and the conductivity is =
1.236106S/m. A longitudinal perfectly insulating defect is assumed, 0.5 mm deep from
the surface of the cylinder and 0.2 mm thick. A pair of source coils Ds (30 mm inner
Ds

long
defect

Dc

receiving
coils

Figure 1. Geometric model of the detection system. It consists in a pair of transmission coils coaxial with the
steel bar and 12 evenly spaced circular receiving coils.

radius, 39 mm outer radius, 1.5 mm height, 7 turns each) feeded by a sinusoidal current
of I = 200 mA per turn with a frequency of f = 100 kHz. They are connected in
counter series and the axial distance between the two coils is 30 mm, see Fig. 1 and 2.
A set of 12 evenly spaced circular receiving coils (3 mm inner radius, 6.5 mm outer
radius, 6 mm height, 400 turns, lift-off 15 mm) with axis directed as the radii of the
bar, are considered. Increasing the number of receiving coils the spatial resolution will

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

197

be improved but it is also more difcult to dispose the coils without modifying their
geometric characteristics.
y

D1

6mm

0.5

mm

30

il #
co

15

9mm

coil #
5

6mm

m
14m

30

3
coil #

m
2m

0.

D2

6mm
14mm
30mm

15

1.5mm

100mm
17mm

15mm

a)

b)

Figure 2. Design of the detection system geometry.

2. Solution of the eddy-current problem


2.1. Discrete Geometric Approach (GAME code)
In order to solve the eddy current problem we resort to the so-called Discrete Geometric Approach[1], [2]. The domain of interest D of the eddy-current problem, has been
partitioned into a source region Ds , consisting in a pair of
current driven coils, and in
a passive conductive region Dc . The complement of Dc Ds in D represents the air
region Da . A pair of interlocked cell complexes is introduced in D, [1]. The cell complex is obtained meshing the model domain. The primal complex is simplicial with inner
oriented cells such as nodes n, edges e, faces f , volumes v (v are tetrahedra).
The dual cell complex is obtained from the primal, according to the barycentric
subdivision, with outer oriented cells such as dual volumes n
, dual faces e, dual edges
f, dual nodes v. For example, a dual node v is the barycenter of the tetrahedron v, a dual
edge f is line drawn from the barycenter of f joining the two dual nodes v , v in the
tetrahedra v  , v  on both sides of f ; with this notation, the one-to-one correspondence
between a cell and its dual is underlined.
The interconnections between cells of the primal complex, are dened by the usual
connectivity matrices G between pairs (e, n), C between pairs (f, e), D between pairs
(v, f ). Similarly, the corresponding matrices for the dual complex are GT (the minus
sign is due to the assumption that a dual volume n
is oriented by the outward normal,
while a node n is oriented as a sink) between pairs (
n, e), CT between pairs (
e, f) and
DT between pairs (f, v). With respect to these cell complexes, we recall the algebraic
equations governing the Discrete Geometric Approach [3], [6], formulated in terms of
the array A of the circulations of the magnetic vector potential along primal edges e of
D and in terms of the array of scalar potential associated with primal nodes n of Dc .
We obtain:

198

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

(CT C A)e = (Is )e


e D Dc
(CT C A)e + i( Ac )e + i(G )e = 0
e Dc
i(GT Ac )n + i(GT G )n = 0
n Dc ,

(1)

where array Ac is the sub-array of A, associated with primal edges in Dc and Is is the
array of the source currents crossing dual faces in Ds . With notation (x)k , we mean
the k-th row of array x, where k = {e, n} is the label of edge e or of node n. Finally
and are matrices representing the discrete counterparts of reluctivity and Ohms
constitutive relations respectively; dim() = F , F being the number of faces in D, and
dim() = Ec , Ec being the number of edges in Dc .
2.2. Construction of the constitutive matrices
We will construct the constitutive matrices and using the Discrete Hodge technique
based on Whitneys maps, described in [7]. We will consider the elementary case of a
single tetrahedron, assuming reluctance and conductivity element-wise constants.
For a mesh of tetrahedra, we will add the contributions element by element.
2.2.1. Reluctance matrix
Reluctance matrix relates the magnetic uxes k on primal faces fk with the magneto
motive forces (m.m.f.s) Fi on dual
 edges fi . We use Whitneys map [1] to express the
magnetic ux density eld b = k wkf k , where wkf is the vector proxy of the Whitneys function associated to face fk . Because of the Gauss Magnetic Law D = 0, the
eld b is elementwise constant [4], and using the pointwise material law h = b, we may
compute Fi as

Fi =

fi

b =

4


wkf (p) fi k ,

(2)

k=1

where fi is the dual edge vector associated with edge fi and p is any point in the considered tetrahedron. Then, the entry vik of a possible reluctance matrix v for tetrahedron
v is vik = wkf (p) fi .
2.2.2. Conductance matrix
The conductivity matrix links the electro-motive forces (e.m.f.s) Uj , with the currents
Ii on
dual faces ei . Using the Whitneys map, we may express the electric eld e as
e = j wje Uj , where wje is the vector proxy of the Whitneys function associated to
edge ej . It is an afne eld and from j = e, we obtain the following expression for Ii

Ii =

e =
e
i

6


wje (mi ) ei Uj ,

(3)

j=1

where ei is the area vector associated with ei and mi is the center of mass of face ei .
Finally, the entry vij of the conductance matrix v for tetrahedron v is vij = wje (mi )
ei . The obtained matrix is non-symmetric, but its possible to demonstrate that, if the
Whitneys functions are evaluated in the barycenter v of the tetrahedron, the matrix
becomes symmetric [6].

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

199

2.2.3. Integral representation of sources


Thanks to the linearity of media, we can express the array A as A = Ar +As , where As
is the array of circulations of the contribution to the magnetic vector potential produced
by the source currents in Ds and Ar is the array of circulations of the contribution to the
magnetic vector potential due to the eddy-currents in Dc . Therefore we have that
(CT C As )e = (Is )e (CT C Ar )e = 0 e Ds
(CT C As )e = 0
(CT C Ar )e = (I)e e Dc

(4)

holds, where I is the array of eddy currents crossing f in Dc . Each entry (As )i of the
array As can be pre-computed as (As )i = ei As dl, where ei is a primal edge in D
and As is the magnetic vector potential due to the known source current density in Ds .
In our case, we have a stranded circular coils and As can be computed in closed form in
terms of the elliptic integrals of the rst and second kind [9].
In this way, we can rewrite the system (1) by removing the source currents from its
right hand side, obtaining
(CT C Ar )e = 0
e D Dc
(CT C Ar )e + i( Acr )e + i(G )e = v
e Dc
i(GT Acr )n + i(GT G )n = w
n Dc ,

(5)

where v = i( Ac s )e and w = i(GT v)e . The system (5) is singular and, to solve
it, we rely on CG method without gauge condition.
2.2.4. Calculation of the induced voltage
For the calculation of the induced voltage we will sub-divide the coil in a series of M
sub-coils. The voltage induced at the terminals of the i-th sub-coil can be determined by:

Ui = ji = jNi

A dl,
ci

where ci is the circumference coaxial with the coil and passing trough the barycenter of
the considered sub-coil. For the calculation of the integral we use the Biot-Savarts law:
A(P ) = As (P ) +

0
4


Dc

J(P  )
dV.
|P P  |

2.3. Integral formulation (CARIDDI code)


As second tool for numerical simulations and comparisons we used an integral formulation described in [5], [11], [12], and [13]. Assuming non-magnetic conductors and time
harmonic elds, the integral formulation of eddy currents is governed by the following
equations:
0
J(P ) = j 4

J(P  )
dP 
Vc |P P  |

jAs ,

(6)

200

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

dS = 0 , J n
= 0 on Vc ,
Jn

(7)

where J is the current density, As is the vector potential dened in Section 2.2.3 produced
by the excitations sources, is the electric scalar potential, is an arbitrary closed
is the outward normal on Vc , and is the
surface in Vc , is the angular frequency and n
resistivity of the conducting domain Vc . The method offers the advantages to discretize
only the conducting domains where the eddy currents are induced, and to automatically
enforce the regularity conditions at innity. Moreover, if we introduce the electric vector
= 0), then only Equation 6
potential T, such that J = T (and ( T) n
is to be imposed. The numerical formulation is obtained
n by expanding T in terms of
edge-elements based shape functions Nk , as T(r) = k=1 Ik Nk (r) where Nk satises
. The uniqueness of the electric vector potential T is achieved by imposing the
Nk n
two-component gauge condition by means of the tree-cotree decomposition of the nite
element mesh [11]. Imposing Equations 6 in weak form by the means of the Galerkins
approach, we obtain:

Nk (J(P ) + j
Vc

0
4


Vc

J(P  )
dP  + jAs + )dP = 0, Nk .
|P P  |

The numerical system is expressed in the form:


(R + jL)I = U,

(8)

where I = {Ik }, U = {Uk } and


0
Lij =
4

Vc

Ni (x) Nj (x )
dV dV  ,
|x x |

Vc


Rij =

Ni Nj dV,
Vc


Ui =

Ni jA0 dV.
Vc

Let us assume that a perfectly insulating crack with negligible thickness is present in
the conductor. It can be schematized through the following condition: J n = 0 on
d , where d is the surface representing the crack and n
is the normal to the crack. To
reduce the computational load and enhance the accuracy we can apply the compensation
method where the total current density J is written as the sum of the solution computed
in absence of the defect (the so-called unperturbed solution J0 ) and of the perturbation
J (the so-called perturbed solution) due to the presence of the defect. In terms of J we
have:
J n = J0 n.

(9)

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

201

The unperturbed solution J0 can be computed using a mesh that does not depend on the
crack geometry, whereas the perturbed solution J requires only a local and rened mesh
in a neighborhood of the crack (see [12] for details). Once I has been computed, the
impedance change due to the aw is given by
Z = UT I/Is2 ,

(10)

where Is is the impressed current. Finally, we mention that this approach has been extended to 3D volumetric defects in [13].

3. Numerical results
When detecting long defects, a reference signal for each coil is not available, therefore is
not possible to use a differential detection system. To have an estimation of the expected
voltage variations in the coils due to the presence of the defect, we computed the voltage
variations U = Ud U0 .
4

1.5

x 10

Integral Formulation
Diffrential Formulation

| U| [V]

0.5

6
coil index

10

12

Figure 3. Voltage variation on each of the 12 receiving coils. The numerical results obtained with the GAME
and CARIDDI codes are in a good agreement each other.

To this aim we need to solve a pair of eddy-current problems with the GAME code
(Geometrical Approach for Maxwell Equations) [8] with the A formulation and the
integral representation of sources. The defect has been modeled as a volume discretized
with a collection of tetrahedra. The unstructured mesh used consists of 505k tetrahedral
elements, yielding 630k DoF.
We apply also the CARIDDI code [5] implementing the integral formulation to the
system under test, splitting the current into a perturbed and an unperturbed solution. Due
to the symmetry of the excitation and pickup coils with respect to the conductive region,
we can reduce memory storage and computational time by discretizing only one-eighth

202

E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods

of the steel cylinder. The mesh used for the perturbed solution is localized near the aw,
allowing for both an increase of accuracy and a reduction of the numbers of unknowns.
The compared results are shown in Fig. 3.

4. Conclusions
The paper exploited two numerical approaches tailored to solve a non-destructive eddycurrent testing problem for the long defect detection during the hot steel-bar production.
The two numerical approaches are the Discrete Geometric Approach (GAME code) and
the integral formulation (CARIDDI code).
The numerical results are in agreement each other and demonstrate that the two
formulations can be considered as useful tools for the numerical modeling and design of
eddy-current diagnostics devices.

References
[1]

Bossavit, How weak is the Weak Solution in nite elements methods?, IEEE Trans. Mag. Vol 34, No. 5,
1998, pp. 24292432.
[2] E. Tonti, Algebraic topology and computational electromagnetism, 4-th International Workshop on Electric and Magnetic Fields, Marseille (Fr) 1215 May, pp. 284-294, 1988.
[3] F. Trevisan, 3-D Eddy Current Analysis With the Cell Method for NDE Problems, IEEE Trans., Vol. 40,
No. 2, 2004, pp. 13141317.
[4] F. Trevisan, L. Kettunen, Geometric interpretation of discrete approaches to solving Magnetostatics,
Vol. 40, No. 2, March 2004, pp. 361-365.
[5] R. Albanese, G. Rubinacci, Integral Formulation for 3D Eddy Current Computation using Edge Elements, IEE Proceedings, vol. 135, Part A, n. 5, pp. 457462, 1988.
[6] R. Specogna, F. Trevisan, Discrete constitutive equations in A geometric eddy-currents formulation,
IEEE Trans. on Magn., Vol. 41, No. 4, 2005, pp. 12591263.
[7] T. Tarhasaari, L. Kettunen, A. Bossavit, Some realizations of a discrete Hodge operator: a reinterpretation of nite element techniques, IEEE Trans. Mag. Vol. 35, 1999, pp. 1494-1497.
[8] R. Specogna, F. Trevisan, The Geometric Approach to solve Maxwells Equations (G.A.M.E.) code
http://www.quickgame.org, copyright 2003-2007.
[9] E. Durand, Magnetostatique, Paris: Masson & C. 1968.
[10] E. Cardelli, A. Faba, R. Specogna, F. Trevisan, Image Reconstruction of Defects in Metallic Plates
Using a Multi-Frequency Detector System and a Discrete Geometric Approach IEEE Transaction on
Magnetics, vol. 42, n. 4, 2007, pp. 18571860.
[11] R. Albanese, G. Rubinacci, Finite element methods for the solution of 3D eddy current problems, Advances in Imaging and Electron Physics, vol. 102, pp. 1-86, 1998.
[12] R. Albanese, G. Rubinacci, F. Villone, An integral computational model for crack simulation and detection via eddy currents, J. of Comp. Phys., Vol. 152 (1999), pp. 736-755
[13] M. Morozov, G. Rubinacci, A. Tamburrino, and S. Ventre, Numerical Models with Experimental Validation of Volumetric Insulating Cracks in Eddy Current Testing, IEEE Trans. on Magnetics, vol. 42, no.
5, pp. 1568-1576, 2006.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-203

203

Theory of Four-Point Alternating Current


Potential Drop Measurements on a
Layered Conductive Half-Space
Nicola BOWLER 1 , and John R. BOWLER
Center for Nondestructive Evaluation, Iowa State University, USA
Abstract. An analytic expression describing the complex voltage measured between the pickup points of a four-point probe, in contact with the surface of a
layered metal half-space, is derived. The driving current is assumed to be timeharmonic. A Greens function formulation leads to a result in which the voltage is
expressed as a sum of two terms. One represents the potential drop due to a homogeneous half-space conductor with properties the same as those of the surface
layer. The other represents the effect of the substrate and is expressed as a Hankel
transform. A scheme for numerical evaluation of this term, by truncating the range
of integration, is presented. An example calculation is given.
Keywords. Four-point, alternating current, potential drop, layered metal half-space

Introduction
Potential-drop measurements using a four-point probe are commonly used to determine
bulk and surface conductivity of metals and semiconductors [1,2], and for crack sizing,
and several commercial instruments are available for these purposes. Four-point methods
rely on using either direct current or very low frequency alternating current for which
the potential drop is essentially real, being in phase with the applied current. In this
regime, the effect of permeability and conductivity separate. This means that the method
is suitable for measuring the conductivity of both non-ferromagnetic and ferromagnetic
materials, contrasting with eddy-current (EC) measurements in which the conductivity
and permeability are not easily separated at typical EC operating frequencies, restricting
EC conductivity measurements to non-ferromagnetic metals.
The alternating current potential drop (ACPD) technique has been analyzed extensively for crack sizing measurements, under the assumption that the current injection
points are sufciently far apart that the applied current density at the crack is uniform
[3,4]. In recent years, the four-point ACPD method of nondestructive evaluation has been
developed both in the context of materials property measurements [5,6] and for crack
sizing [7].
1 Corresponding Author: Center for Nondestructive Evaluation, 279 Applied Sciences Complex II, 1915
Scholl Road, Ames, IA 50011-3042, USA; E-mail: nbowler@iastate.edu.

204

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

The work presented here is motivated by the need to determine the prole of conductivity and permeability in a metal as a function of depth from the surface, in applications such as nondestructive determination of the case depth in surface-hardened steels
or, at higher frequencies, surface residual stress in aircraft engine alloys. It is anticipated
that ACPD measurements provide greater sensitivity in measuring changes in electromagnetic material properties as a function of depth, when compared with EC testing
[8,9], because injected current has a component perpendicular to the conductor surface,
whereas induced eddy currents always ow parallel to the surface of an unawed specimen. ACPD measurements also offer greater sensitivity for this application than direct
current potential drop measurements [10] since the skin effect provides a mechanism for
the current to be concentrated at a particular depth of interest such as the transition region between a surface-hardened layer and substrate in case-hardened steel. In contrast
with earlier work, the work reported here does not rely on the assumption of uniform
current density but, rather, an exact analytical solution for the measured potential drop is
obtained on the assumption that the test-piece is signicantly larger than the largest separation of the probe points. An example calculation shows how the ACPD voltage varies
as a function of the depth of a surface layer, relative to that of a homogeneous test-piece.

1. Analysis
In a four-point measurement, two current electrodes and two voltage electrodes are used.
Typically they are arranged in a straight line, or on the vertices of a rectangle, and contact
with the specimen is made using spring-loaded pins. The potential drop is measured
between the voltage electrodes. The potential drop between the pickup points, v, may be
written as the sum of four terms; the potential at each of the two measurement points due
to the sources at the current injection and extraction points. With reference to Figure 1,
v = v1 v2 =


I 
f (22 ) f (21) f (12 ) + f (11) ,
21

(1)

where I is current amplitude and f () depends on frequency and the variation of conductivity and permeability of the test-piece as a function of depth below the surface.
In this article, an analytic expression is derived for the ACPD voltage measured by
a four-point probe with points located at arbitrary positions on the surface of a planar

v2 r
v1 r
B
@

B
@

11 
B 22
@
 
B
12 @
@ B
r

21@ B
+I
@B
@Br
I
v = v 1 v2
Figure 1. Plan view of the four electrode points on a conductor surface.

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

Air
Region 1 (Layer)

1 , 1

Region 2 (Substrate) 2 , 2

205

z=0
z=d

?
z

Figure 2. Cross-section of the metal test-piece.

conductor, with a surface layer of depth d. The conductivity and permeability of each
region are denoted j and j respectively, j = 1, 2, with region 1 being the surface layer
and region 2 being the substrate, Figure 2.
Assuming the eld varies as the real part of eit , f () for a homogeneous halfspace conductor with parameters the same as those of region j is given by [11],
fhs,j () =

eikj
+ ikj [E1 (ikj ) + ln ] ,

where kj2 = ij j .

(2)

Corresponding expressions for plates of various thickness relative to the probe dimensions are given in Ref. [11].
In its present form, Eq. (2) does not take into account inductive pick-up in the closed
loop formed by the pick-up pins, the conductor, and associated wiring. This contribution
to the measured voltage is seen in the imaginary part (in quadrature with the applied
current). From the practical point of view it is important to minimize the inductive pickup by making the pick-up loop physically as small as possible, otherwise it dominates
the signal as frequency increases [5]. An additional analytic term can be derived that
theoretically accounts for the inductive pick-up [5] but, in this article, we are concerned
with the effect of the surface layer on the voltage measured between the pick-up pins and
do not deal with the inductive term explicitly.
1.1. Formulation and Solution
In this work we follow the formulation of Ref. [11], in which we solve the electromagnetic eld problem initially assuming a single current injection or extraction point. In this
way cylindrical symmetry can be exploited. The result for current injection and extraction at two separate points is then obtained by superposition. Under certain assumptions
[11], the magnetic eld H is transverse magnetic (TM) with respect to the direction of
the normal to the conductor surface (
z ) and hence can be expressed as
H = [
z ]

(3)

where is the TM potential. With this formulation, the potential v at a point Q1 relative to that at another point Q2 , both in the plane z = 0, is found from the following
expression [11],


 

1 

.
(4)
v=
1 z Q1
z Q2
The solution is formulated in terms of the Greens function G(r, r ) for the structure,
from which the TM potential is obtained using the following relationship [11].

206

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

(r) =

I2
t

G(r , r)
z 


z  =0

(5)

The potential drop v then follows from relation (4). In order to determine G(r, r ), express the layered half-space kernel in the following cylindrically-symmetric form,

1


z, z  )J0 () d,
(6)
G(,
G(r, r ) =
2 0
which is a solution of
(2 + kj2 )G(r, r ) = (r r ).

(7)

At the interface between the layer, region 1, and the substrate, region 2, the continuity of
H at z = d implies that is continuous there. The continuity of the tangential electric
eld implies that 1
z is also continuous at z = d. These conditions on also apply
to G. In addition we retain the requirements from the homogeneous half-space problem
that G(r, r ) = 0 at z = 0 and that the remote dipole eld vanishes (this occurs when
the combined effect of current injection and extraction points is considered).

z, z  ) assumes that the source point (denoted


The following general solution for G(,
by the primed co-ordinates) is located within region 1. The solution vanishes at z = 0.


1
1 |zz  |
1 (z+z  )
1 (zz  )
e
,1

[1
+
A()]
e
+
A()e

z, z  ) = 21
(8)
G(,
1
2 z1 z 
,
2
21 B()e

with j = 2 ij j and the root with a positive real part is taken. Applying the
interface conditions gives
A() =


e21 d
(e21 z 1),
1 + e21 d

z  < d,

(9)

where
=

1 /1 2 /2
.
1 /1 + 2 /2

(10)

Substituting for A() from (9) into expression (8) gives




z, z  ) = 1 e1 |zz | e1 (z+z ) +
G(,
21


4e21 d

sinh(1 z) sinh(1 z ) .
1 + e21 d

(11)

Now note that the rst two terms in (11) are simply those forming the Greens function
for a half-space, Ghs,j (r, r ), with parameters j and j denoted by the subscript j.
Hence
G(r, r ) = Ghs,1 (r, r ) + V (r, r )
where, from (6)

(12)

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

V (r, r ) =

1
e21 d
sinh(1 z) sinh(1 z  )J0 () d.
1 (1 + e21 d )

Now, referring to Eq. (3.26) in Ref. [11] and from Eq. (5) with relation (2),

I

[fhs,1 () + flayer ()] ,
=
z z=0
2

207

(13)

(14)

where

flayer () = 2
0

1
e21 d
J0 () d
(1 + e21 d )

(15)

since the operator 2t introduces the factor 2 in transform space. The potential drop
between two points may now be computed by substituting (14) into (4).
1.2. Limiting Cases
In this section it is shown that the solution obtained here reduces as anticipated in limiting cases. First, if regions 1 and 2 have identical parameters, vanishes and only fhs,1
remains in Eq. (14), as expected. Similarly, as d , e21 d 0 and only fhs,1
remains in Eq. (14). Considering the case d 0, note rst that [12]

1
J0 ()d.
(16)
fhs,1 () =

0
Then see that, as d 0,

1
(1 2 2 1 )
J0 ()d
flayer ()
2 0

1
= fhs,1 () + fhs,2 ().
2
Substituting this relation into (14) it is seen that

I 1

=
fhs,2 (),
as d 0,
z z=0
2 2

(17)
(18)

(19)

which leads to the replacement of 1 by 2 in (4), as required. Finally, note that putting
2 = 0 and 2 = 0 gives a representation for a plate in air. Putting 2 = 0 makes
= 1 and




I
1 1 + e21 d

=
(20)
J0 ()d.
z z=0
2 0 1 e21 d
Eq. (20) is in agreement with an expression that can be obtained from the result specifically derived for a metal plate in Ref. [13, Eq. (30)]. It is interesting to see that (20) is
obtained by putting 2 = 0, regardless of the value of 2 . This reects the nature of the
TM excitation of the test piece; a low conductivity substrate will give the same v regardless of whether or not the substrate is magnetic. In order to observe the magnetic state of
the substrate, it is necessary to excite the transverse electric mode, as in EC testing.

208

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

2. Numerical Evaluation
The functions in Eq. (2) are readily available in computational packages such as Matlab,
but evaluation of flayer (), Eq. (15), requires more work. An efcient and accurate numerical evaluation of the integral can be performed by articially truncating the domain
of the function flayer () by setting flayer () = 0 for a and hence recasting the integral as a sum [14]. The value of the parameter a is chosen to be sufciently large for this
to be a reasonable approximation. A factor of 10 greater than the probe length is usually
sufcient. In detail, for an integral of the form

f
()Jn ()d,
(21)
f () =
0

including Eq. (15) with


e21 d
1
f
() = 2 2
(1 + e21 d )

(22)

and n = 0, it is assumed that the function f () can be written as the following summation:
f () =

Al f (l )Jn (l ),

(23)

n=1

where Jn (z) is the Bessel function of the rst kind of order n and Jn (l a) = 0. The
coefcients Al are determined by applying the Hankel transform to Eq. (21) and truncating the domain of f () so that f () = 0, a. Next, f () as given in Eq. (23) is
substituted into the resulting integrand to give
f
(m ) =

Al f
(l )

l=1

Jn (l )Jn (m ) d,

(24)

for a particular = m . Now apply standard integral Eq. 11.4.5 of Ref. [15] to determine
a
a2 
2
[Jn (l a)] lm ,
Jn (l )Jn (m )d =
(25)
2
0
where Jn (z) = dJn (z)/dz and lm = 1 for l = m and 0 otherwise is the Kronecker
delta function. Hence, from Eq. (24), Al = (2/a2 )[Jn (l a)]2 and substituting Al into
Eq. (23) gives
f () =

Jn (l )
2 

f (l )
2.
2
a
[Jn (l a)]

(26)

l=1

In the particular case of interest here, following the above procedure allows Eq. (15) to
be written as follows.


4  1
e21 d
J0 (l )
.
(27)
flayer () = 2
a
2l (1 + e21 d ) [J1 (l a)]2
l=1

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

209

since J0 (z) = J1 (z), [15, Eq. 9.1.28]. The set of zeros of the function J0 (z) are used
to determine the set of l that satisfy J0 (l a) = 0. Once values of l are known, the
terms in Eq. (27) may be computed and summed.

3. Example Calculation and Discussion


As an example consider a four-point probe, with co-linear arrangement of the points,
placed on the surface of a layered test-piece with 1 = 36 MS/m (aluminum) and 2
= 25 MS/m (a typical aluminum alloy). The separation between the current injection
and extraction points is 40 mm. That between the pickup points is 35 mm. In Figure 3
normalized relative voltage, V, dened
V =

v vhs,1
,
vhs,1

(28)

is plotted for surface layer thickness d = 1, 2 and 3 mm. v is computed using expressions
(14) and (15), whereas vhs,1 is computed using (16). From Figure 3 it is clear that the real
part of the signal shows greater relative change, due to the layer, than the imaginary part.
V 0 as frequency increases due to the layer appearing more like a half-space as the
electromagnetic penetration depth decreases. In the case of thicker layers, V 0 at
lower frequencies than for thinner layers. On this basis, broadband potential drop measurements, coupled with model-based interpretation, hold promise for determining the
depth of an interface at which there is a change in electrical conductivity. In general, V
is enhanced if the conductivity contrast between the layer and the substrate is increased.
Acknowledgement
This work was supported by the NSF Industry/University Cooperative Research Program.
References
[1] R. K. Stanley, P. O. Moore, and P. McIntire (eds.), Nondestructive Testing Handbook, 2nd ed., vol. 9,
Special Nondestructive Testing Methods. American Society for Nondestructive Testing, Columbus, OH,
1995.
[2] D. K. Schroder, Semiconductor Material and Device Characterization, Wiley, New York, 1998.
[3] D. H. Michael, R. T. Waechter and R. Collins, The measurement of surface cracks in metals by using
AC electric elds, Proc. Roy. Soc. Lond. Ser. A 381 (1982), 139-157.
[4] A. M. Lewis, D. H. Michael, M. C. Lugg and R. Collins, Thin-skin electromagnetic elds around
surface-breaking cracks in metals, J. Appl. Phys. 64 (1988), 3777-3784.
[5] N. Bowler and Y. Huang, Model-based characterization of homogeneous netal plates using four-point
alternating current potential drop measurements, IEEE Trans. Mag. 41 (2005), 2102-2110.
[6] V. A. Mitrofanov, Problems of the theory of the electric potential method of nondestructive inspection
using alternating current, Russ. J. Nondestructive Testing 34 (1998), 183-189.
[7] G. Sposito, F. Simonetti, P. Cawley, and P. B. Nagy, Potential drop spectroscopy for characterization
of complex defects, CP820, Review of Quantitative Nondestructive Evaluation Vol. 25, ed. by D. O.
Thompson and D. E. Chimenti, American Institute of Physics 2006.
[8] F. Yu and P. B. Nagy, Simple analytical approximations for eddy current proling of the near-surface
residual stress in shot-peened metals, J. Appl. Phys. 96 (2004), 1257-1266.

210

N. Bowler and J.R. Bowler / Theory of 4-Point Alternating Current Potential Drop Measurements

Real Normalized Relative Voltage

0.35

0.25
0.2
0.15
0.1
0.05
0
0.05 0
10
0.1

Imaginary Normalized Relative Voltage

d = 1 mm
d = 2 mm
d = 3 mm

0.3

10

10

10

10

0.08
0.06
0.04
0.02
0
0.02 0
10

10

10
frequency (Hz)

10

10

Figure 3. Calculated normalized relative potential drop V, Eq. (28), for a co-linear four-point probe on a
metal half space, 2 = 25 MS/m, with a surface layer of thickness d and 1 = 36 MS/m.
[9] Y. Shen, C. Lee, C. C. H. Lo, N. Nakagawa and A. M. Frishman, Conductivity prole determination by
eddy current for shot-peened superalloy surfaces toward residual stress assessment, J. Appl. Phys. 101
(2007), 014907.
[10] F. Takeo, K. Nakajima, T. Baba, Y. Aonahata and M. Saka, Arrangement of probes for measuring case
depth by means of four-point probes, Advances in nondestructive evaluation, Key engineering materials
Vols. 270-273 82-88, Part 1-3 2004.
[11] J. R. Bowler and N. Bowler, Theory of four-point alternating current potential drop measurements on
conductive plates, Proc. R. Soc. A 463 (2007), 817-836.
[12] N. Bowler, Analytical solution for the electric eld in a half space conductor due to alternating current
injected at the surface, J. Appl. Phys. 95 (2004), 344-348.
[13] N. Bowler, Electric eld due to alternating current injected at the surface of a metal plate, J. Appl. Phys.
96 (2004), 4607-4613.
[14] T. P. Theoudoulidis and E. E. Kriezis, Eddy Current Canonical Problems, Tech Science Press, Forsyth
GA, 2006.
[15] Handbook of Mathematical Functions with Formulas, Graphs and Mathematical Tables, edited by M.
Abramowitz and I. A. Stegun, Dover, New York, 1972.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-211

211

Integration of tilted coil models in a


volume integral method for realistic
simulations of eddy current inspections
Theodoros THEODOULIDIS a,1 , and Gregoire PICHENOT b
a
University of West Macedonia, Energy Department, Greece
b
CEA-LIST, CEA, Saclay 91191 Gif-sur-Yvette, France
Abstract. Tilted coil models are incorporated in a Volume Integral Method code for
realistic simulations of eddy current inspections. The coils have either cylindrical
or rectangular shapes and the congurations involve a layered conductor system
with an arbitrary number of layers. Emphasis is put on the rapid calculation of the
incident elds by modifying existing integral expressions to equivalent Fourier-type
series expressions. In this paper we present only the cylindrical coil conguration.
Theoretical results for the crack signal of a titled coil are veried by experimental
measurements.
Keywords. Eddy currents, analytical modelling, tilted coils, Volume Integral
Method

Introduction
In the application of the Volume Integral Method for the simulation of eddy current defect
inspections an important part of the solution is the accurate calculation of the incident
(0)
electromagnetic eld, that is the term Ek (r) in Equation (1), describing the numerical
discretization scheme for the calculation of the electric eld in the volume of a defect
embedded in a multilayered conductive half-space [1]
(0)

Ek (r) = Ek (r) j0

N


(ee) (r, r )[l (r )]El (r )dr


G
kl

(1)

l=1
(0)

Here Ek (r) is the electric eld with the defect present, Ek (r) is the electric eld with
(ee) (r, r ) is the electric-electric dyadic Greens functhe defect absent (incident eld), G
kl
tion dened as the eld response to a unit point source, l refers to the conductivity of
the host layer, (r ) refers to the conductivity distribution of the defect and primed and
non-primed position vectors refer to source and eld cells respectively.
Within the framework of a collaborative project between the University of West
Macedonia, Greece and CEA-LIST, France, the incident electromagnetic eld as well as
1 Corresponding Author: Theodoros Theodoulidis, University of West Macedonia, Energy Department,
Bakola & Sialvera, 50100 Kozani, Greece; E-mail: theodoul@uowm.gr.

212

T. Theodoulidis and G. Pichenot / Integration of Tilted Coil Models in a Volume Integral Method

the coil self- and mutual-impedances for uncracked conductors have been calculated in
an efcient and rapid way for arbitrary cylindrical and rectangular coil orientations. The
model, in turn, has been successfully included into the CIVA software, which is a powerful multi-technique platform for simulating NDT industrial congurations. In this work
we outline the method for calculating the electromagnetic eld from a tilted cylindrical
coil located above a layered conductive and/or magnetic half-space. The existing integral
expressions are modied to equivalent series expressions thus providing a way for rapid
calculations and ease of implementation without any sacrice on accuracy of results. The
extension of the class of coils used in the simulations and the corresponding decrease of
computation time in the Volume Integral Method calculations are the main achievements
of this project. Results for the eddy current response of tilted coils using VIM have also
been previously demonstrated in [2].

1. Analysis
Consider Figure 1 which shows a cylindrical coil located above a layered conductor.
The coil is excited by a time harmonic current varying as the real part of I exp(jt).
The coil is tilted by an angle around the y-axis, rotated by an angle around the zaxis and moved so that its center lies at (x0 , y0 ). Both and angles are positive for
counterclockwise rotation. The layered conductor can have an arbitrary number of layers
nl , each layer t having conductivity t , relative magnetic permeability t and thickness
ct . In order for the eld to assume a double sum than a double integral expression, the
solution domain for the boundary value problem is truncated in both x and y directions.
Thus, the solution domain extends from 0 to hx in the x-direction and from 0 to hy in
the y-direction. These boundaries are chosen to be far from the coil and they are perfect
electric insulators, i.e. Bz = 0.

Figure 1. Tilted coil conguration above a layered conductor.

The analysis of the electromagnetic eld problem is based on the use of potentials.
The truncated domain is divided into the conductor layers and the above air region. The

T. Theodoulidis and G. Pichenot / Integration of Tilted Coil Models in a Volume Integral Method

213

solution then takes the form of series expansions in each region and the expansion coefcients found from the continuity conditions governing the eld at the interfaces between
each region (layer). In the air-region above the conductor upper surface, the magnetic
eld can be expressed as the gradient of a scalar potential B = where satises the
Laplace equation. The potential can be considered as the superposition of the isolated
coil potential (s) and the potential originating from the eddy currents in the conductor
(ec) . The expressions for these two potentials are then written as:

(s)

(x, y, z) =

(s)

sin(ui x) sin(vj y)eij z Cij

0 z l0

(2)

i=1 j=1
(ec)

(x, y, z) =

sin(ui x) sin(vj y)eij z Dij

(ec)

z0

(3)

i=1 j=1

The magnetic ux density in the conductor layers (z < 0) can be written using the

transverse electric (TE) part of the second order vector potential as B = W a z


with scalar potential Wa satisfying either the Laplace or Helmholtz scalar equations
according to the conductivity of the specic layer. The general expression for a particular
layer is:
Wa (x, y, z) =



(a)
(a)
sin(ui x) sin(vj y) eij z Cij + eij z Dij

(4)

i=1 j=1

where ui = i/hx , vj = j/hy (j is index), 2ij = u2i + vj2 , k 2 = jr 0 (j is the


2
= 2ij + k 2 . The electric eld in a conductive layer can be derived
imaginary unit), ij
. Expressions for the potentials satisfy the continuity conditions
from E = j Wa z
of the eld at the interface planes. A recursive use of these conditions [3], can be used
(ec)
for expressing the unknown coefcient Dij in air and the unknown nl coefcients
(a)

(a)

(s)

Cij and Dij in the conductor layers in terms of the source coefcient Cij , which is
(ec)

considered to be known. The Dij , which is needed for calculations of the magnetic
eld in air as well as for self- and mutual-impedances, can be written in the general
(ec)
(s)
form as Dij = Rij Cij where Rij is a reection coefcient that depends solely on the
characteristics of the layered conductor [3].
1.1. Self and Mutual Impedance Change
The magnetic eld in all regions as well as the electric eld in the conductor can be
calculated from the expressions that relate B and E to W. The general expression for
the impedance change caused by the presence of the layered conductor can be derived
by using a reciprocity relation [4].
Z =
(s)


jhx hy  
(s) (s)
ij Cij Cij Rij
20 I 2 i=1 j=1

(5)

where Cij represents the source coefcients characterizing the isolated coil and Rij
represents the reection coefcient characterizing the contribution of the eddy current
density induced in the layered conductor.

214

T. Theodoulidis and G. Pichenot / Integration of Tilted Coil Models in a Volume Integral Method

In case the mutual impedance change between two coils is sought, Equation 5 takes
the form
Z12 =


jhx hy  
(s)
(s)
ij Cij,driver Cij,pickup Rij
20 I 2 i=1 j=1

(6)

where now there are two source coefcients, one for the driver and one for the pickup
coil.
1.2. The Source Coefcient for a Tilted Cylindrical Coil
The method of calculating the source coefcient Cij is similar to the one presented in [4]
and involves an integration over the coil surface and a superposition over the coil crosssection. The resulting expression for the coil of Figure 1 is

0 i0 2 eij d ej(+ui xd +vj yd ) N(a ) ej(+ui xd vj yd ) N(b )
(s)

Cij =
(7)
hx hy ij +ej(ui xd +vj yd ) N(b ) ej(ui xd vj yd ) N(a )
where d = l0 + r2 sin || + (l/2) cos is the coil center height and
N() = sin (l/2) M(r1 , r2 )/ 3

(8)

a = (ui cos + vj sin ) sin jij cos

(9)

b = (ui cos vj sin ) sin jij cos

(10)

where i0 = N I/[(r2 r1 )l] is the coil current density with N denoting the number of
 r
wire turns and M(r1 , r2 ) = r12 xI1 (x)dx with I1 (x) denoting the modied Bessel
function of order 1.

2. Results
Code was written in Matlab to compute the impedance change of the tilted coil above
the layered conductor, the mutual impedance between combinations of cylindrical and
rectangular coils, the magnetic eld in air both from the isolated coil and the change
due to the conductor and the incident electric and magnetic eld in the conductor layers.
The eld can be computed very rapidly in a 3D grid (volume area), a 2D grid (surface),
a 1D grid (line) by utilizing the vector capabilities of Matlab. In all of the examined
cases, appropriate dimensions hx and hy for the truncated domain as well as appropriate
number of terms for the double series were used. Typical values were hx = hy = 20
max(r2 , r2 cos() + l/2 sin() so that whatever the coil dimensions and tilt, the coil is
not located close to the boundaries and also Ni = Nj = 200 which is much larger than
the rather small number of terms that is normally required.
Theoretical results, produced by using CIVA, were rst compared to results from
the exact double integral expressions in [4] for the coil and conductor (plate) data that
are given in Table 1. Accuracy was veried for the eld and impedance in case of an
uncracked conductor. Figure 2 shows amplitude of the eddy current density induced on
the top surface of the plate for three tilt angles. Computation time for such plots is less
than 1 sec.

T. Theodoulidis and G. Pichenot / Integration of Tilted Coil Models in a Volume Integral Method

215

Table 1. Coil and testpiece parameters for the results in Figure 4.

y [mm]

Coil

Testpiece

r1
r2

1.0 mm
1.75 mm

nl
c1

2
1.55 mm

l
N
l0

2.0 mm
422
0.1 mm

1
2
r1

1.02 MS/m
0.0
1

r2

5
5

0
x [mm]

5
5

0
x [mm]

5
5

0
x [mm]

Figure 2. Eddy current amplitude contours for tilt angles = 0o , 45o and 90o .

The theoretical results were also compared to experimental measurements conducted by CEA and involved two position scans of a surface EDM notch in an Inconel
plate. The notch dimensions were 0.1 7.0 1.24mm and the coil and plate data are
those given in Table 1. The coil tilt was = 90o and the coil rotation was = 90o
as shown in Figure 3. The measurements were done at 100kHz using an Agilent 4194A

Figure 3. Screenshot from CIVA showing the coil and testpiece conguration for the results in Figure 4.

impedance analyzer. The coils calculated inductance of 0.315mH was veried by the
measured value of 0.314mH. In the rst position scan the coil is moved along the x-axis
and in the second position scan it is moved along the y-axis. Since the EDM notch is

216

T. Theodoulidis and G. Pichenot / Integration of Tilted Coil Models in a Volume Integral Method

R/X

X/X

located along the y-axis, in the rst position scan the coil is moved across the notch while
in the second position scan it is moved along the notch. The results for the resistive and
reactive part of the impedance change due to the crack are shown in Figure 4. In both
cases the agreement is very good.
0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

0.1

0.1

0.2

0.2

0.3

0.3

0.4
6

0
x [mm]

0.4
6

0
y [mm]

Figure 4. Comparison of theoretical (lines) and experimental results (circles) for the resistive () and inductive
part () of the impedance change as a function of position. On the left the coil is moved across the notch while
on the right the coil is moved along the notch. Simulations were done with CIVA.

3. Conclusions
We have incorporated further capabilities in a Volume Integral code by introducing tilted
coils above a layered conductor system in absolute, differential and driver-pickup modes.
We are now able to compute in a rapid manner the incident electromagnetic eld, self
and mutual impedances and most importantly defect signals produced by tilted coils.

References
[1] S. Paillard, G. Pichenot., M. Lambert and H. Voillaume, Eddy current modelling for inspection of riveted
structures in aeronautics, Electromagnetic Nondestructive Evaluation, Japan, 2006.
[2] J.C. Aldrin and J.S Knopp, Crack characterization method with invariance to noise features for eddy
current inspection of fastener sites, J. Nondestr. Eval. 25 (2006), 165181.
[3] W.C. Chew, Waves and Fields in Inhomogeneous Media, Van Nostrand Reinhold, New York, 1990.
[4] T.P. Theodoulidis and E.E. Kriezis, Eddy current canonical problems (with applications to nondestructive evaluation, TechScience Press, 2006.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-217

217

Modeling of Flawed Riveted Structures


for EC Inspection in Aeronautics
S. PAILLARD a , G. PICHENOT a Y. CHOUA b Y. LE BIHAN b M. LAMBERT c
H. VOILLAUME d and N. DOMINGUEZ e
a

CEA, LIST, Gif sur Yvette, F-91191, France


LGEP, CNRS UMR 8507, Supelec, Univ Paris Sud, Univ PM Curie-P6, France
c
L2S (CNRS-Supelec-UPS), 3 rue Joliot-Curie, 91192 Gif-sur-Yvette, France
d
EADS Innovation Works, NDI & SHM, 12 rue Pasteur, 92152 Suresnes, France
e
EADS Innovation Works, NDI & SHM, 23 bv Victor Hugo, 31770 Colomiers, France
b

Abstract. Within the framework of a collaborative project between CEA and


EADS, a semi-analytical model based on a Volume Integral Method (VIM) has
been developed so as to simulate the Eddy Current (EC) inspection of riveted structures in aeronautics. The modeling is the one of a layered planar structure with a
aw located nearby a fastener. The VIM involving dyadic Greens functions is considered, a fastener and a aw being introduced as a variation of conductivity in a
stack of slabs. This semi-analytical approach is compared to a Finite-Element one
(FE) developed by LGEP and is validated with experimental data acquired on an
aeronautical conguration.
Keywords. Eddy Current Testing, Aeronautic inspection, Flawed fastener

Introduction

This contribution describes recent achievements about the simulation of EC inspections


of awed fastened structures. The model developed is mainly based on a volume integral
formulation employing the dyadic Greens formalism [1]. This model, which can quickly
predict the response of an EC probe, is implemented in the NDT software CIVA ([3]).
This multi-modality platform for industrial NDT also handles various eddy current testing congurations as is sketched in Figure 1. In a previous work, a multi-layer model has
been investigated [2], and results compared with those provided by a nite-element approach and with experimental data. Here, we intend to extend it to the case of a aw near
a rivet in a like structure. The paper goes as follows: A awed rivet model is presented,
a method of moments is developed and its results are compared with those provided by
a nite-element method and by one based on a potential formulation, experimental data
being used for further, in-depth validation.

218

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

1.a: Bobbin coil placed


inside a conducting tube

1.b: Bobbin coil placed on a


conguration dened by CAD

1.c: Ferrite core placed on a fastened


structure

Figure 1. Representation of several congurations in the CIVA user interface

1. The semi-analytical model


1.1. Theoretical Formulation
The conguration is the following one: a N -layer slab, each layer numbered i and of
conductivity i is sandwiched between two air half-spaces numbered 0 and N + 1, all
materials being linear, isotropic, and non magnetic (with permeability 0 ). The slab is
affected by a defect of volume and conductivity (r) crossing one or several layers (as
depicted in Figure 2.a). One is denoted with index m (resp. n) the rst (resp. last) layer
affected by the 
defect (m < n), the latter being divided into as many layers as necessary
n
such as =
k=m k (in the case of a rivet traversing the N layers, m = 1 and
n = N ). A time-harmonic source (circular frequency and implied time-dependence
exp (jt)), e.g., a coil probe, is set in the upper half-space. A vector domain integral
formulation of the electric eld Ek (r) in the layer k in such a conguration is obtained
by application of the Greens theorem to the diffusive vector wave equation and is given
by
(0)

Ek (r) = Ek (r) j0

n

l=m

(ee)

Gkl (r, r) [l (r)] El (r) dr

r k

(1)
where
is the primary eld in the layer k and Gkl (r, r) the electric-electric
dyadic Greens functions dened as the eld response to an unit point source and solution
of
(0)
Ek (r)

(ee)

(ee)

(ee)

Gkl (r, r) kk2 Gkl (r, r) = kl I(r r).

(2)

In the above equations k, l denote the index of the layer of the observation r and of the
source r point, respectively, I is the unit dyad, and kl stands for the Kronecker delta. kl
is the wave number in the lth layer dened as kl2 = j0 l . The Greens dyad satises
appropriate boundary conditions at the interfaces between the different layers in the same
way as the electric elds do. The response of the probe is given by its impedance variation
and is obtained via the reciprocity theorem, where I0 is the feeding current of the probe,
as

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

I02 Z

n

l=m

(0)

[l (r)] El (r) El (r)dr.

219

(3)

1.2. Numerical developments


Once the model has been chosen and the equations established, the numerical formulation can be implemented. Equation (1) is discretized by means of a Galerkins version of
the method of moments where the contrast zone is divided into Ncell parallelepipeded
voxels. The voxels are chosen in order to have an homogeneous conductivity inside each
one, the electric eld being assumed constant-valued within a given voxel as well. This
approach yields a linear system (4)

(0)
Em
Gm,m Gm,n
Em
.
. .
.
. = I
.. . . . .. ..
.
(0)
Gn,m Gn,n
En
En

(4)

where Gi,i are the electromagnetic self-coupling terms of the ith region of the divided
rivet onto itself and where Gi,j are the mutual coupling terms of the j th over the ith . An
example is given for a three-layered slab (n = 3 and m = 1) in Figure 2. The awed
rivet is here divided into three parts, each one entirely contained within a single layer
of conductivity k with k {1, 2, 3}. The self-coupling terms Gi,i with i {1, 2, 3}
and the mutual-coupling terms Gi,j with (i, j) {1, 2, 3} and i = j are represented in
Figure 2.b.

2.a: Separate parts of the calculation zone


along the Z-axis

2.b: Green s dyads for a three layer slab


conguration

Figure 2. Example of a awed rivet in a three-layered slab

To propose a semi-analytical awed rivet model, three main improvements have


been made as follows.
The multi-layer model: it takes into account a contrast zone which is crossing
several layers like a through-wall hole in a multi-layer slab (the model validation
is detailed in [4]). The self and mutual coupling terms of the Greens functions
are written in explicit analytical fashion and implemented to reconstruct the entire
matrix of equation (4).
The rivet model: considers what happens when a stack of slabs is fastened. The
model takes into account the typical rivet shape, i.e., its at cone-shaped head. It
is considered via the calculation of the volume ratio in every cell of the discretiza-

220

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

tion zone. A weighting coefcient is introduced in each cell based on the volume
occupied by the head rivet compared to the one occupied by the slab (it is equal
to one if the cell is entirely in the rivet or in the aw, and zero if not).
The awed rivet model: extends the rivet model to the case of a aw within a
fastened structure. The rivet and the aw are included in the same calculation
zone (see Figures 3.b and 3.c) with a volume ratio matrix so as to t the geometry
at best (an example is shown in Figure 3.a).

3.a: Contrast Matrix

3.b: Discretization in the


X-Y-plan

3.c: Discretization along the


Z-axis

Figure 3. Discretization of the full calculation zone

In the applications aimed at, the typical size of the domain might be more than ten
skin-depths at the frequency of operation. Few voxels per skin-depths are needed, which
leads to a large number of voxels and to a too large linear system to invert (the memory
2
size can be estimated as O (9 Ncell
)). Taking into account the convolution structure of
the integral equation (1) with respect to the two lateral directions via appropriate fast
Fourier transforms, an iterative solution of the system enables us to treat large defects by
4/3
reducing the memory size to O (9 Ncell ).
2. Comparison to a published example (Zeng et al., ACES07)
2.1. Congurations
In [5], the authors successfully compared their approach based on a potential formulation
with a nite-element approach on two congurations (one is depicted in Figure 4 and the
second is similar to it without aw). It consists of a plate with a through-wall cylindrical
hole with (and without) a aw located nearby, sketched in Figure 4.b. The EC probe is
moved along the surface, on a line passing by the diameter of the hole and along the
length of the breaking-surface aw, above the fastener assembly (Figure 4.a).
2.2. Results
There is a good agreement in both amplitude and phase (Figure 5.a and 5.b) between
CIVA and the potential formulation (which, in addition, as already said, was successfully
compared with the nite-element method). These congurations involved a breakingsurface aw, a cylindrical hole and a unique slab. These elements are still simple compared with respect to those in aeronautics. However, the good results achieved in these
simple congurations have allowed us to validate the awed rivet model for a more realistic aeronautical conguration with the following characteristics: a multi-layer conguration, a buried aw with a thin opening, a fastener shape (with conical head), and a
ferrite-core probe.

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

4.a: View of the conguration with CIVA

221

4.b: Details of the hole and the aw

Figure 4. Conguration for the data comparison

(

5.a: Amplitude of the signals


CIVA, Potential Formulation)

(

5.b: Phase of the signals


CIVA, and Potential Formulation)

Figure 5. Comparison between CIVA results and those of the potential formulation [5].

3. Calibration on the rivet


In most industrial applications, the measured EC signal is calibrated over a reference
conguration. The issue is to cope with discrepancies of scales, the size of the rivet
and the one of the aw differing by orders of magnitude. So, to be sensitive to the aw
response, a calibration of the signals simulated vs. the experimental one is seen as a
preliminary set to the validation of the awed rivet model.
3.1. Conguration of the calibration
The reference conguration for the calibration is a rivet without aw within a multi-layer
slab as is depicted in Figure 6. It consists of a layered planar structure with a fastener
hole: three aluminum layers (2.5 mm and 4 mm thick) with a conductivity of 17 MS/m
are traversed by a borehole with head diameter of 12 mm and body diameter of 6.35 mm
(see details in Figures 6.a and 6.b). The EC probe is moved on a line along the diameter
of the rivet. The experiments have been performed with a ferrite-cored probe operating
at 1.6 kHz (detailed in [6] and [7]). The probe has an inner radius of 3.74 mm, an outer
radius of 7.325 mm, a lift-off of 0.09 mm and a thickness of 3.46 mm with 926 turns.
The semi-analytical and nite-element results the latter being obtained from a code

222

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

6.a: View of the conguration with CIVA

6.b: Details of the rivet

Figure 6. Conguration for the calibration

developed at LGEP [8] are calibrated with respect to the complex value associated to
the maximum amplitude of the experimentally recorded borehole signal.
3.2. Results of the calibration
The impedance variations measured in the impedance plane calibrated on the rivet signal
with the semi-analytical model, with the nite-element code and the experimental data
are displayed in Figure 7.a with squares ( ), circles () and (), respectively .

7.a: Impedance plane diagram

7.b: Real and imaginary parts


Figure 7. Calibration

A good agreement between the semi-analytical model and the nite-element one is
obtained. Nevertheless, the two simulated signals do not accurately t the experimental one when the probe is placed right above the borehole. An explanation for this dis-

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

223

crepency is that the EC probe in this conguration has a 3D ferrite core (cylindrical core
with slots) which was not modeled using CIVA and the FE code. The disagreement between experimental and simulation data could be due to this specic probe as shown in
[6].
4. Validation on experimental data: Flawed Rivet
4.1. Conguration of the experimental validation
Now, the model is validated in the case of a aw near a rivet in a structure similar to the
one described in section 3.1 (see Figure 8.a). The conguration (slab, borehole dimensions, probe) is the same as at the calibration case save the addition of an EDM notch of
200 m-width, 5 mm-length and 4 mm-depth in the second layer as shown in Figure 8.b.
The probe is moved on a line along the diameter of the rivet and along the length of the
aw.

8.a: View of the conguration with CIVA

8.b: Details of the rivet and the aw

Figure 8. Flawed structure conguration

4.2. Results of the experimental validation


In the following all the results are calibrated (see section 3 for the calibration procedure).
In Figure 9.a the experimental data and the results obtained from the semi-analytical
model are compared using an impedance plane representation whereas in Figure 9.b
the latter and the nite-element results are presented. The aw signal is obtained by
subtracting the response of the probe from the awed rivet from the one over the rivet
without aw. Due to the small amplitude of the aw signal and to the incertitude of the
experimental signal, the result of the aw signal is here only compared to the Finite
element result (Figure 9.c). It appears that the aw signal has the same shape in the
impedance plane for the semi-analytical and the nite-element results, with a discrepancy
in amplitude better than 30 %.
5. Conclusion
The integration of the awed rivet model in the CIVA platform is in progress. The approach has been successfully compared with different methods (potential formulation

224

S. Paillard et al. / Modeling of Flawed Riveted Structures for EC Inspection in Aeronautics

9.a: Lissajous curves


( Experimental data,  CIVA)

9.b: Lissajous curves


( CIVA, FE)

9.c: Flaw response


( CIVA, FE)

Figure 9. Lissajous curves of the rivet and the aw

and nite-element method) for different congurations. This semi-analytical awed rivet
model is validated with experimental data on an aeronautical conguration, with good
agreement. Among the questions still open, how to better account for the discrepancy
between the sizes of the rivet and of the aw appears to be one of the most compelling
one.
Acknowledgements
This work is supported by the Paris le-de-France Rgion.
References
[1]
[2]
[3]
[4]

[5]

[6]
[7]
[8]

Chew W. C., Waves and Fields in Inhomogeneous Media, Van Nostrand Reinhold, New York, 1990.
Paillard S., Pichenot G., Lambert M., and Voillaume H., Eddy current modeling for inspection of riveted
structures in aeronautics, 11th International Workshop on Electromagnetic NDE, Iwate, 2006.
Le Ber L., Calmon P., Sollier T., Mahaut S., and Benoist P., Advances of simulation and expertise
capabilities in CIVA platform, in Review of Progress in QNDE 25, 2006, pp. 684-691.
Paillard S., Pichenot G., Lambert M.,Voillaume H. and Dominguez N., A 3D model for eddy current
inspection in aeronautics: application to riveted structures, Review of Progress in QNDE 26, 2006, pp.
265-272.
Zeng Z., Liu X., Deng Y., Udpa L., Knopp J. S., and Steffes G., Reduced magnetic vector potential
and electric scalar potential formulation for eddy current modeling, Review of Progress in ACE, Verona,
2007, pp. 773-777.
Buvat F., Pichenot G., Prmel D., Lesselier D., Lambert M., and Voillaume H., Eddy current modeling
of ferrite-cored probes, Review of Progress in QNDE 24, 2005, pp. 463-470.
Pichenot G., Buvat, F., Maillot V., and Voillaume H., Eddy current modeling for non destructive testing,
16th World Conf. on NDT, Montreal, 2004.
Choua Y., Santandrea L., Le Bihan Y., Marchand C., Thin Crack Modeling in ECT with Combined
Potential Formulations, IEEE Transactions on Magnetics, 43, 2007, pp. 1789-11792

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-225

225

Numerical Modeling of Eddy Current


Nondestructive Evaluation of
Ferromagnetic Tubes via an Integral
Equation Approach
Anastassios SKARLATOS a , Grgoire PICHENOT a , Dominique LESSELIER b ,
Marc LAMBERT b and Bernard DUCHNE b
a
CEA-LIST, 91191 Gif-sur-Yvette
b
Dpartement de Recherche en Electromagntisme - Laboratoire des Signaux et
Systmes (CNRS-Suplec-Univ Paris-Sud), 91192 Gif-sur-Yvette
Abstract. In this contribution a Volume Integral Equation (VIE) formulation for
the modeling of eddy current Nondestructive Evaluation (NDE) of ferromagnetic
tubes is proposed. The method is well suited for the simulation of NDE applications
involving moving probes including Remote Field Eddy Current Effect (RFEC)
probes. The proposed algorithm has been integrated in the CIVA platform.
Keywords. Ferromagnetic tubes, Integral Equations, Remote Field Eddy Current
Effect

1. Introduction
Volume Integral Equation (VIE) - based models are well established and successfully
applied in the industry for nonmagnetic tubes [1,2]. Such models have also been already
developed for the modeling of ferrite-core eddy-current probes [3]; however, the formulation of the problem in the more general ferromagnetic case deserves more studies. The
presence of material defects in this case results in a local variation of the magnetic permeability in addition to that of the conductivity met in nonmagnetic materials. This in
turn requires a system of two integral equations, one for the electric eld and one for the
magnetic eld, in order to obtain a well determined problem. This system involves the
full family of the Greens dyads.
The integral equations are solved numerically using the Method of Moments
(MoM). For the latter, only the aw region needs to be discretized. The matrix produced
by the discretization of the integral equations depends only upon the geometries of the
tube and the aw and upon the frequency. Hence, once the Greens dyads have been calculated and the MoM matrix has been constructed, the response of the aw to a given excitation can be obtained by a simple multiplication of the discretized primary eld vector
(which calculation is very fast) with the inverse of the above matrix. Thus, the method is
very efcient, particularly for the simulation of Nondestructive Evaluation (NDE) tech-

226

A. Skarlatos et al. / Numerical Modeling of Eddy Current Nondestructive Evaluation

niques involving moving coils, as the most important part of the computational burden
relies on the computation of the Greens dyads that have only to be computed once.
In the following, the VIE formulation for the ferromagnetic problem is presented
briey and validated using experimental data obtained by means of the Remote Field
Eddy Current (RFEC) technique. The discussed model has already been integrated in an
expertise software for the simulation of NDE applications, i.e., the CIVA platform [4].

2. The Integral Equation Formulation


Let us consider a ferromagnetic tube with conductivity b2 and permeability b2 . Index
2 denotes the second layer (the tube wall) of a cylindrically layered medium. The rst
and the third layers correspond to the inner and outer regions, respectively. The presence
of a material defect inside the tube wall causes a local variation of conductivity, and/or
permeability, depending upon the type of aw. In NDE applications, we usually deal
with aws caused by corrosion or material damage (like cracks) so that a simultaneous
change of both parameters is the most interesting case. Let 2 ( r ) = 2 ( r ) b2 and
2 ( r ) = 2 ( r ) b2 stand for the variation of conductivity and permeability inside
the tube wall with respect to their background values in a non-damaged zone. The electric
and magnetic elds can be expressed in the following form (the time convention e jt
is implied) [5]:

(ee)
2 ( r ) = E
inc ( r )+jb2 G
2 ( r  ) dV 
E
r, r  ) 2 ( r  ) E
2
22 (

+j

Vf
(em)

G22

2 ( r  ) dV  ,
( r, r  ) 2 ( r  ) H

(1)

Vf

and


(me)
inc

2 ( r  ) dV 

H2 ( r ) = H2 ( r )+ G22 ( r, r  ) 2 ( r  ) E

Vf
(mm)
2 ( r  ) dV  ,
+ 2 b2 G22 ( r, r  ) 2 ( r  ) H

(2)

Vf

inc , H
inc denote the
where the integration extents over the support of the aw Vf . E
2
2
b2
b2
electric and magnetic elds in the absence of the aw, and b2 = b2 j
j
(ab)

is the complex permittivity of the medium. The dyadic Greens functions G 22 ( r, r  ),


(a, b = e, m) are dened as the eld response of unit point current sources of electric
and magnetic types. Their denition is given in [3,6].
The system of equations (1),(2) provides a complete description of the problem. It
can be solved numerically by means of the method of moments. Let us notice that the
eddy current inspection of nonmagnetic conducting tubes can be deduced from the above
formulation as a subcase. Once the elds inside the damaged region have been obtained
by solving the above-mentioned system, the variation of the probe mutual impedance,
which is actually the physical quantity that can be measured, can be found by applying
the reciprocity theorem:

A. Skarlatos et al. / Numerical Modeling of Eddy Current Nondestructive Evaluation

Z12

227

 
1
inc ( r  ) E
2 ( r  )
2 ( r  ) E
=
Rx
I1 I2
Vf


2 ( r  ) H
inc ( r  ) dV  ,
+ j2 ( r  ) H
Rx

(3)

where I1 and I2 are the electric currents owing in the driving and receiving coils, re inc ( r ) and H
inc ( r ) denote the electric and magnetic elds induced in
spectively, and E
Rx
Rx
the unperturbed material by the receiving coil (Rx) when operating in the transmission
mode.

3. Validation
The above integral formulation has been validated by comparing the theoretical results
with two experimental data sets. The rst experimental set-up consists of a RFEC probe
moving inside a steel tube with inner and outer diameters of 14 mm and 18 mm, respectively. The operating frequency is set to 250 Hz. At this frequency, the measured values
of conductivity and relative permeability are 6.25 MS/m and 210, respectively. Figure 1
displays the layout of the probe and some representative results obtained for a 3 mm
wide 70% deep external groove and a 5 mm through-hole. The number of grid cells used
for the discretization of the integral equation is 200 for the groove and 1089 for the hole,
which results in respective computation times of 23 min and 31 min on a Pentium 4
workstation operating at 3.6 GHz with 1 GB RAM, for 161 different probe positions.
It must be noticed that, for both cases, the major part of the computation time for both
cases is dedicated to the calculation of the dyads and to the inversion of the matrices. In
the case of the through-hole, for instance, the CPU time needed for the computation of
the primary eld is ca. 3 min, whereas the rest of the time (28 min) is the dedicated to
the calculation of the Greens dyads and the matrix inversion. Duplicating the number
of probe positions, the difference in the calculation time is less than one minute. The
number of probe positions has thus a very small impact on the total calculation time. Furthermore, the simulation of a new scan does not require the recalculation and inversion
of the matrix.
In the second experiment, the tube has inner and outer diameters of 23.3 mm and
31.9 mm, respectively. The frequency of operation is 150 Hz. The values of the tube
conductivity and permeability at this frequency are 3.5 MS/m and 100, respectively. The
probe conguration and the results for a 3D defect (notch) are displayed in Figure 2. Both
experimental and FEM simulation results for this given conguration are courtesy of
Chen et al. [7]. 567 grid cells are used for the discretization of this problem and the CPU
time reaches 13 min for 200 probe positions, 40 s being dedicated to the computation of
the primary eld.
Tab.1,2 compare the measured values of the probe signals amplitude and phase to
those obtained by means of the VIE model. In general, simulation and measurements are
in good agreement. Some deviations in phase observed in the cases of 3D aws (hole
and notch) can be attributed to the side effects of the aw fabrication procedure. This is
evidenced in Tab.3 which displays the values measured (in the rst experiment) for two
identical holes produced with different fabrication techniques (one is electro-eroded and
the other is drilled). It can be observed that the aw fabrication procedure has a non-

228

A. Skarlatos et al. / Numerical Modeling of Eddy Current Nondestructive Evaluation

Table 1. Comparison of the amplitude and phase of the measured signal to that obtained by means of the VIE
model (2D aws).
Flaw

Amplitude
VIE

External Groove 20% (1st exp.)

Phase

Meas.

4.4 mV

VIE

Meas.

4.1 mV

-43

-41

External Groove 60% (1st exp.)


External Groove 70% (1st exp.)

21.9 mV
30 mV

21 mV
30 mV

-20.6
-17

-24.1
-21

External Groove 50% (2nd exp.)

18 mV

16 mV

11

15

3 mm

35 mm

35 mm

5 mm

6.8 mm
12.8 mm

(a)
15
VIE
Measurements

5
0

10
15
15

VIE
Measurements

Im{VR} (mV)

Im{VR} (mV)

10

1
VIE : 30 mV, 17
Meas: 30 mV, 21
10

2 VIE : 5.0 mV, 0


Meas: 4.5 mV, 15
0
Re{VR} (mV)

10

15

(b)

0
Re{VR} (mV)

(c)

Figure 1. (a) Probe layout in a tube with a hole. Comparison of the simulation and experimental results for:
(b) a 3 mm wide 70% deep external groove, and (c) a 5 mm through-hole. The results are calibrated by using
the results obtained for a 3 mm wide 40% deep external groove

negligible impact on the measurements. Furthermore, the Lissajous curves calculated by


the VIE model and the FEM t very well with each other and both present the same
deviation in phase from the experimental curve (cf. Figure 2). As the two results are
obtained by means of two methods quite different from one another, it seems probable
that the deviation from the experimental curve is due to the local change of the magnetic
permeability in the vicinity of the aw caused by manufacturing.
Table 2. Comparison of the amplitude and phase of the measured signal to the ones obtained by means of the
VIE model and FEM (3D aws).
Flaw

Amplitude
VIE

Through Hole (1st exp.)


External Notch 50% (2nd exp.)

4.5 mV
2.4 mV

Phase

FEM

Meas.

VIE

FEM

Meas.

2.7 mV

5 mV
3.8 mV

15
14

11.9

0
29

A. Skarlatos et al. / Numerical Modeling of Eddy Current Nondestructive Evaluation


15 mm

229

5 mm

2
1.5

VIE
FEM
Measurements

114 mm

8 mm

Im{V } (mV)

1
0.5
0

0.5
1

VIE : 2.4 mV, 14

1.5 FEM : 2.7 mV, 12


Meas: 3.8 mV, 29
2
2
1

19.5 mm

15.5 mm

(a)

0
Re{VR} (mV)

(b)

Figure 2. (a) Probe layout, and (b) results for a 5 mm wide, 60 and 50% deep external notch (b). Both
simulation and experimental results are calibrated by using the results obtained for a 5 mm wide 20% deep
external groove.
Table 3. Comparison of the measured signals for two identical through-holes made with different techniques.
Flaw

Amplitude

Phase

Electro-Eroded Hole (1st exp.)


Drilled Hole (1st exp.)

5 mV
4.2 mV

0
6

4. Conclusion
A Volume Integral Equation formulation for the inspection of ferromagnetic tubes has
been proposed. The results of the model have been validated using experimental data.
A good agreement between theoretical and experimental results has been observed. The
presented model is particularly efcient for simulation of NDE applications involving
moving probes. It has been already integrated in the CIVA platform and it will be available with the next release (CIVA 9.0).

Acknowledgments
The authors would like to thank Prof. Chen and his colleagues from the International
Institute of Universality in Tokyo, and Prof. Takagi from the Institute of Fluid Science
in Sendai, Japan, for kindly supplying their experimental and FEM simulation results for
the validation of our model.

References
[1]

[2]

V. Monebhurrun, D. Lesselier, and B. Duchne, Evaluation of a 3-D bounded defect in the wall of a
metal tube at eddy current frequencies: the direct problem", J. Electromagn. Waves Appl., vol. 12, pp.
315-347, 1998.
G. Pichenot, D. Prmel, T. Sollier, and V. Maillot, Development of a 3D electromagnetic model for
eddy current tubing inspection: Application to steam generator tubing", Rev. Quant. Nondestr. Eval., vol.
16, pp. 79-100, 2005.

230
[3]

[4]
[5]

[6]
[7]

A. Skarlatos et al. / Numerical Modeling of Eddy Current Nondestructive Evaluation

J. R. Bowler, L. D. Sabbagh, and H. A. Sabbagh, A theoretical and computational model of eddycurrent probes incorporating volume integral and conjugate gradient methods", IEEE Trans. Magn., vol.
25, no. 3, pp. 2650-2664, 1989.
CIVA 9.0, State of the art simulation platform for NDE, 2007, www-civa.cea.fr.
A. Skarlatos, G. Pichenot, D. Lesselier, M. Lambert, and B. Duchne Remote eld effect modeling
via an integral equation approach, 5me Confrence Europenne sur les Mthodes Numriques en
Electromagntisme, Lille, pp. 139-140, 2006.
W. C. Chew, Waves and elds in inhomogeneous media. New York: IEEE Press, 1995.
M. Rebican, Z. Chen, N. Yusa, K. Miya, T. Uchimoto, and T. Takagi, Investigation of numerical precision of 3-D RFECT signal simulations", IEEE Trans. Magn., vol. 41, no. 5, pp. 1968-1971, 2005.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-231

231

Design of Reflection Type Pulsed Eddy


Current Nondestructive Testing
a

Young-Kil SHIN , Dong-Myung CHOI and Hee-Sung JUNG


School of Electronic and Information Eng., Kunsan National University,
San 68, Miryong-Dong, Kunsan, Chonbuk, 573-701, Korea
b
R&D Institute, Sae-An Engineering Corporation,
Byucksan Digital Valley II, Gasan, Geumcheon, Seoul, 153-803, Korea

Abstract. Reflection type shielded send-receive probe is designed for accurate


measurement of specimen thickness by pulsed eddy current testing and when
evaluating material conductivity, effects of pulse width on the signal are
investigated. Results show that the best sensitivity to thickness is achieved when
ferrite shields for both coils are used and when the exciter coil is located inside the
sensor coil. Pulse width study suggests that the shorter pulse width is desired if the
peak amplitude is used to evaluate the material conductivity, while the longer
pulse width is needed if the peak time is used for the same purpose.
Keywords. Pulsed eddy current, Reflection probe, Pulse width

1. Introduction
Thickness and material property measurements without contacting the test object is the
main advantage of eddy current testing over the other testing methods. Among eddy
current testing methods, the pulsed eddy current (PEC) testing is expected to be rich of
information and to have deeper penetration than conventional eddy current testing. This
is because a pulse current can be transformed into infinite train of harmonically related
sinusoidal waveforms so that it has wideband frequency [1-3]. There are two types of
PEC testing, one is through transmission method and the other is reflection method.
Both methods are known to be sensitive to thickness variation of test specimen. When
the far side of test object is not accessible, the reflection type PEC testing has to be
used. In this paper, a shielded send-receive type reflection probe is designed by using a
self-written numerical analysis code and their performance to evaluate the thickness
and conductivity of test object is investigated. Also, effects of pulse width to PEC
signals are investigated when material conductivity is evaluated. The time taken to
reach the peak value of the step response is first investigated and it is used as the pulse
width in the PEC testing to produce the maximum PEC signal.

2. Numerical method and test design models


Pulse coil current induces eddy currents in a specimen and their magnitude change
continuously with time. In such a case, a transient analysis is required to predict their

232

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

behavior so that the backward difference method is used for temporal analysis. For the
spatial modeling, the finite element method is used [4,5]. An axisymmetric modeling
code is written and used for the prediction of PEC signals.
The governing equation for PEC testing is
1

u u A
P

where

Js V

P , V , J s , A are

wA
.
wt

(1)

permeability, conductivity, source current density vector,

magnetic vector potential, respectively. In this work, Eq. (1) is written using a
cylindrical coordinate system. Applying the finite element formulation for the space,
the following type of matrix equation is obtained [6].
wA

> S @^ A`  >C @ wt ^Q`

(2)

To treat time, the backward difference method is used where all the values are
evaluated at a new time, t n1
wA

wt

n 1

n
t  't , and the time derivative term is expressed as

^ A`n1  ^ A`n

(3)

't

where ^A`n is the magnetic potential evaluated at time t n .


Rewriting Eq. (2) by using Eq. (3), the following recurrence relation is obtained
and the magnetic potential at any time step can be calculated.
1

n 1
't >C @  > S @ ^ A`

^Q`n1 

1
n
C @^ A`
>
't

(4)

The test signal in PEC testing is the electromotive force induced in the sensor coil
so that it can be calculated as follows.

Vemf

^ A`n1  ^ A`n
't

2S rc

(5)

where rc is the centroidal radius of a coil.


Four probe design models that use copper and ferrite shields, as shown in Figure 1,
are first tested. Figure 1 shows cross section of axis symmetric test model probes and
the vertical line is the axis of symmetry. The exciter and sensor coils are denoted by
E and S, respectively. Later, all the shields in model 1 and model 4 are changed to
ferrite and tested for thickness evaluation of copper and inconel 600 plates.

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

233


(a) Model 1

(b) Model 2


(c) Model 3

(d) Model 4

Figure 1. Four test models of reflection type PEC probe using copper and ferrite shields [unit = mm]

3. Selection of probe design for thickness evaluation


Figure 2 shows PEC signals obtained by model 1, peak amplitude and peak time
variations as a function of thickness. Figures 3, 4 and 5 show same kind of results
obtained by models 2, 3 and 4. As the thickness increases, numerical modeling results
show that the peak value decreases and the time taken to reach the peak amplitude
(hereafter, the peak time) increases at the low thickness, but decreases when the
thickness exceeds a certain limit. When the coil spacing is increased from 3 mm
(D=10) to 6mm (D=20), the peak value decreases but the peak time increases.
Comparing results from model 1 and 4, and model 2 and 3, we can find that the higher
peak value appears when the exciter coil is inside the sensor coil. Meanwhile, if we
compare results from model 3 and 4 in which the exciter is inside the sensor, we can
find that the higher peak value is obtained when the exciter coil is shielded by copper
while the sensor coil is shielded by ferrite.

234

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

(a) Signal (coil spacing = 3mm)

(b) Peak amplitude variation

(c) Peak time variation

Figure 2. Thickness variation results obtained by probe model 1

G
(a) Signal (coil spacing = 3mm)

(b) Peak amplitude variation

(c) Peak time variation

Figure 3. Thickness variation results obtained by probe model 2

G
(a) Signal (coil spacing = 3mm)

(b) Peak amplitude variation

(c) Peak time variation

Figure 4. Thickness variation results obtained by probe model 3


(a) Signal (coil spacing = 3mm)

(b) Peak amplitude variation

(c) Peak time variation

Figure 5. Thickness variation results obtained by probe model 4

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

235

Table 1. Sensitivity to thickness variation when copper and ferrite shields are used

Model 1
Model 2
Model 3
Model 4

Copper
D=3mm
D=6mm
53.3 %
59.78 %
51.6 %
59.4 %
46.2 %
51.15 %
56.16 %
64.76 %

Inconel 600
D=3mm
D=6mm
40.7 %
47 %
44.96 %
51.7 %
40 %
47.4 %
44.7 %
52 %

In a practical sense, however, the probe sensitivity to thickness variation seems


more important than peak value changes. This sensitivity is calculated by dividing peak
value changes due to thickness variation from 1.8 mm to 4.5 mm by the biggest peak
value, that is, when the thickness is 1.8 mm. Results are summarized in Table 1.
According to this table, model 4 gives the best sensitivity and the better sensitivity can
be achieved as the coil spacing is wider and as the conductivity of the test material is
higher. This result seems reasonable because direct influence from the exciter would be
suppressed if the coil spacing increases (although PEC signal itself is reduced) and the
influence of eddy currents in the test material that reflect the condition (thickness, in
this case) of test specimen would be increased if the conductivity is higher.
If eddy currents are induced in the copper shield and when the conductivity of test
material is very low, the influence of eddy currents in the shielding material may
appear in the PEC signal. To avoid such unwanted signals, models using ferrite shields
for both coils are tested. The results are shown in Figure 6 and they are much better
than those from previous models. Table 2 shows that the sensitivity, as well as the PEC
signal, is also much higher.





(a) Signal (coil spacing = 3mm)

(b) Peak amplitude variation

(c) Peak time variation

Figure 6. Thickness variation results obtained by a probe that uses only ferrite shields (Exciter inside)


Table 2. Sensitivity to thickness variation when only ferrite shields are used

Sensor Inside
Exciter Inside

Copper
D=3mm D=6mm
74.3 %
74.2 %
74.3 %
74.2 %

Inconel 600
D=3mm D=6mm
44.3 %
60.2 %
50.96 %
60.2 %

236

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

4. Effects of pulse width variation in the evaluation of conductivity


PEC signal variation due to material conductivity is investigated in conjunction with
the pulse width variation. First, the response from step input current is investigated to
decide the pulse width. As shown in Figure 7(a), peak amplitudes of step responses that
are from various conductive materials appear at different times. The time taken to reach
the peak amplitude in the step response is investigated and summarized in Table 3. The
peak time increases as conductivity increases. Figure 7(b), (c) show PEC signals
obtained by using different pulse widths. In Figure 7(b), the peak time of tungsten is
used as the pulse width and that of copper is used in Figure 7(c). As the conductivity
increases, the peak amplitude gets reduced and the peak time increases. When the pulse
width is long enough, signal passes the peak point and reduces somewhat before the
steep drop caused by the pulse-off. If the pulse width is too short, the peak amplitude is
less than that of step response. Peak values and peak times from various conductors are
investigated and effects of pulse width on them are shown in Figure 8. The steeper
slope would give better signal resolution. The sensitivity to conductivity variation is
also investigated and summarized in Table 4. According to these investigations, the
pulse width needs to be shorter if the peak value is used to evaluate the conductivity,
while the longer pulse width is desired if the peak time is used for the same purpose.


(b) Pulse width = 180 s

(a) Step responses

(c) Pulse width = 520 s

Figure 7. Step responses and PEC signals obtained by model 4, except that only ferrite shields are used.
Table 3. Peak time of step responses to various conductivities and thickness when ferrite shields are used

Thickness
1.8 mm
4.5 mm

Copper
520 s
950 s

(a) Peak value variation

Aluminum
390 s
720 s

Tungsten
180 s
340 s

(b) Peak time variation

Figure 8. Variation of peak value and peak time due to conductivity changes.

Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing

237

Table 4. Sensitivity to conductivity variation when only ferrite shields are used.

Plate Thickness

1.8 mm
Sensor Inside
4.5 mm
1.8 mm
Exciter Inside
4.5 mm

Pulse Width

Peak Value Sensitivity

Peak Time Sensitivity

520 s
180 s
950 s
340 s
520 s
180 s
950 s
340 s

63.4 %
79.7 %
63.9 %
81.0 %
64.1 %
80.1 %
63.9 %
81.0 %

65.3 %
37.0 %
63.8 %
27.6 %
66.6 %
29.1 %
65.2 %
26.0 %

5. Summary
In this paper, numerical modeling of reflection type PEC testing is performed. Results
show that the best sensitivity to thickness variation can be achieved when exciter coil is
located inside the sensor coil and both are shielded by ferrite. Effects of pulse width for
conductivity evaluation are also studied by monitoring the peak times of step responses.
Results suggest that the pulse width needs to be shorter if the peak amplitude is used to
evaluate the conductivity, while the longer pulse width is desired if the peak time is
used for the same purpose.

Acknowledgement
This work was supported by the Korea Science and Engineering Foundation (KOSEF)
grant funded by the Korea government (MOST) (No.2007-00467).

References
[1]
[2]
[3]
[4]
[5]
[6]

C. J. Renken, The use of a personal computer to extract information from pulsed eddy current tests,
Materials Evaluation, Vol. 3, pp. 356-360, 2001.
M. S. Safizadeh, B. A. Lepine, D. S. Forsyth, and A. Fahr, Time-frequency analysis of pulsed eddy
current signals, J. NDE, Vol. 20, No. 2, pp. 73-86, 2001.
Gui Yun Tian and Ali Sophian, Reduction of lift-off effects for pulsed eddy current NDT, NDT&E
International, Vol. 38, pp.319-324, 2005.
R. Ludwig and X. W. Dai, The numerical and analytical modeling of pulsed eddy currents in a
conducting half-space, IEEE Trans. Mag., Vol. 26, No. 1, pp. 299-307, 1990.
Xiao-wei Dai, Reinhold Ludwig, and R. Palanisamy, "Numerical simulation of pulsed eddy-current
nondestructive testing phenomena," IEEE Trans. Mag., Vol. 26, No. 6, pp. 3089-3096, 1990.
Young-Kil Shin, Numerical modeling of probe velocity effects for electromagnetic NDE, Ph. D.
Dissertation, Iowa State University, U.S.A., 1992.

This page intentionally left blank

Applications

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-241

241

Noninvasive Characterization of BjorkShiley Convexo-Concave Prosthetic Heart


Valves using an Electromagnetic Method
Raimond GRIMBERG1, Shiu C. CHAN2, Adriana SAVIN1,
Lalita UDPA2, Satish S.UDPA2
1

National Institute of R&D for Technical Physics, Iasi, Romania


2
Michigan State University, East Lansing, USA

Abstract. Prosthetic heart valves Bjork-Shiley Convexo-Concave type were


intense implanted during long time, being reported cases of breaking of one
component of the valves which had lead to deceases. This paper presents a new
method for noninvasive electromagnetic method for this type of valve, using an
eddy current transducer with orthogonal coils. In vitro experiments had shown that
discontinuities of outlet strut with depths equal or bigger than 0.4mm can be
detected with a probability of detection of 86.4% and in the case of discontinuities
with depth equal or bigger than 0.6mm with POD of 97%.
Keywords: eddy current examination, nondestructive evaluation, Bjrk-Shiley
convexo-concave prosthetic heart valves

1. Introduction
Current prosthetic heart valve devices are subject to strict regulatory ([1],[2],[3])
and engineering controls in design evaluations, manufacturing quality control and in
vitro verifications, and must demonstrate satisfactory results in both animal and human
trials.
The BSCC heart valve comprises of a flange (orifice ring), an inlet strut, an outlet
strut, a disc occluder and a woven Teflon fabric sewing ring for implantation. The
flange and the inlet strut are manufactured as an integral unit from a cobalt-based
Haynes-25 alloy bar stock. The outlet strut is formed from a wire of the same alloy and
joined to the flange by TIG welding. The occluder disc is composed of an outer
Pyrolite coating over a graphite core (Figure 1).
The failure mechanism of the BSCC heart valve is not yet completely understood,
many reports [1] have suggested material fatigue as the key cause of the fractures.
There is, therefore, a necessity and considerable interest in developing techniques to
detect cracks and single-leg separation (SLS) failures before a complete outlet strut
fracture (OSF) occurs.
Many detection approaches have been proposed. They generally belong to one of
the three categories: high speed and energy cineradiography [4]; acoustic approaches
[5]-[7]; electromagnetic approaches [6], [7].

242

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

Figure 1. BSCC convexo-concave heart valve

This paper proposes a novel electromagnetic method for noninvasive in vivo


detection of fatigue cracks in the outlet strut. In this method, an absolute send-receiver
transducer [8], [9] with orthogonal coils for excitation and reception is used.
Characterization of the detection system with heart valve replicas has demonstrated a
high system sensitivity to crack size as small as 0.4 mm. Meanwhile, in vitro testing of
32 BSCC heart valves with various outlet strut conditions using the proposed method
has demonstrated 100% accuracy in indicating SLS failures.

2. Test Specimens
Two types of test specimens were used in the studies:
 Replica valves made from Haynes-25 alloy with the tolerances of a 27 mm BSCC
valve. On the outlet strut of each specimen, there is either no crack or an EDM
notch with a depth of one of 0.2, 0.4 or 0.6 mm and 0.3mm width. The replicas do
not contain occluder discs (Figure 2a). This is done without loss of accuracy
since the contribution of conductivity from the disc is small, as is its influence on
the results of the inspection.
 BSCC heart valves with either intact or single-leg separated outlet struts (Figure
2b). The SLS cases consist of both manufactured and naturally occurring defects.
In the manufactured SLS cases, a laser was used to sever one of the outlet strut
legs. The occluder discs are present in these valves.

b
Figure 2. BSCC valves: a) replicas; b) real valves

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

243

3. Fatigue Growth Rate


The relationship between the fatigue growth rate per cycle, da/dN, and stress intensity
range 'K is given by Paris law [10]

da
dN

C 'K

For the Haynes-25 alloy, da/dN was also experimentally determined using alloy
wire samples with a diameter of 1.2 mm, immersed in Ringers lactate solution at 370C.
Figure 3 shows a comparison between the experimental and theoretical results.
According to the data from this figure, for a normal pulse of 72pulses/minut, if
initially in OS a crack with 0.6mm depth exists; it will grow until OS fractures in 3
months. If it can be determined with a good probability of detection, an interval of time,
long enough for the preparation of the operation of explant-implant will be obtained.

Figure 3 Comparison of experimental and theoretical results

4. Experimental Setup
A test setup had been constructed to investigate the proposed approach involving the
use of orthogonal coils and to demonstrate the methods ability to accurately detect and
characterize discontinuities on outlet struts of BSCC heart valves.
For each test, the heart valve is rotated 400 clockwise and tilted back 600 from the
vertical, with the outlet side of the valve facing up towards the transducer.
The transducer is made from two orthogonal coils wounded on an insulator support.
It is placed approximately 70 mm away from the outlet strut of the valve. This
particular valve orientation and test configuration resembles the situation when the
transducer is placed on the chest surface of the patient.
During the heart valve inspection, the relative position and distance between the
transducer and the heart valve continuously vary due to the cardiac cycle. In order to

244

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

investigate the effects of this movement on the performance of the proposed orthogonal
coils detection system, a cinegram of an unidentified BSCC heart valve patient was
obtained and the in situ movement of the heart valve was traced from a series of
fluoroscopic images [11].
The obtained travel path (Figure 4) was then used to program a motorized stage to
simulate the relative movement between the transducer and the outlet strut (the
superposition of the displacing of a point from OS along horizontal and vertical
direction).
In the in vitro test setup, the heart valve was held at a fixed location by a silicon
holder while the transducer was mounted on and moved by the motorized stage.

Figure 4. Trajectory of a point on the outlet strut during a cardiac cycle

The experimental setup is shown in Figure 5.

Figure 5. Experimental setup: a)basic scheme; b) transducer - valve assembly; c) photo of equipments

The proposed detection system uses an absolute send-receiver electromagnetic


transducer (remarkable sensitivity and self-nulling property) [8], [9]. The emission coil
is 120 mm in diameter and has 300 turns, was excited with a 60 kHz, 40Vrms sinusoidal

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

245

signal (the impedance of emission coil is 2.8k: at 40kHz). The reception coil is
wounded in the plane orthogonal to the emission coil and also has 300 turns.
The motorized stages simulating the relative in vivo valve-transducer movement
had a repetitive frequency of 1.2 Hz, corresponding to a 72 cycles/minute cardiac
rhythm. The data acquisition frequency was 100 samples/second.
Figures 6a and 6b show the phase dependency of the received signal for testing the
replica valves with no defect, and those with one of 0.2, 0.4 or 0.6 mm deep EDM
notch, respectively.

b
Figure 6 Experimental measurements for replica valves: a) real component; b) imaginary component

5. Experimental Results
In the proposed detection system, the power spectrum is obtained for the imaginary
component of the reception coil signal. The amplitude of the first harmonic of this
power spectrum is then used as the indicator of BSCC valve quality state.
The power spectrum of a signal U is defined as

246

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

1
fft Im U conj fft Im U
N

where N is the number of samples and fft is the Fast Fourier Transform function. The
power spectrum contains N harmonics.
5.1. Determination of Probability of Detection (POD)
One hundred measurements were taken on BSCC valve replicas with intact outlet struts
as well as those with EDM slots of 0.2, 0.4 and 0.6 mm deep.
The recorded signals were processed using the procedure described above.
Examples of the processing scheme outputs are shown in Figure 7 for 12 of the
measurements.
From the measured data, the lower-bound for the probability of detection with a
95% confidence level [12] was determined for each slot depth. The results are
summarized in Table 1.
Although all the 0.6 mm cases were correctly classified, the POD was less than
100% due to the imposed 95% confidence level and the fact that the number of
measurement was 100.
Table 1. The lower-bound probability of detection for a 95% confidence level for various slot depths
Slot depth [mm]

Probability of Detection (POD)[%]

Probability of Missing Flaw (PMF)[%]

0.2

41.3

58.7

0.4

84.6

15.4

0.6

97

Figure 7. Experimental results on BSCC replica valves

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

247

6. In Vitro Testing with BSCC Heart Valves


Double-blind in vitro tests using 32 BSCC heart valves were also performed. The test
samples consisted of 12 valves with intact outlet struts, 10 valves with manufactured
SLSs and 10 explanted valves with SLSs. The data collected are processed and the
results are presented in Figure 8.
In this figure, only the amplitude of the first harmonic of a signals power
spectrum is considered. The decision threshold was obtained using the 3V law [10]: the
mean plus three times the standard deviation of all the good valve data points:

Threshold

P g.v.  3V g.v.

Any valve with a data point located above this threshold was considered an SLS
valve. Using this technique, the proposed detection method achieved a 100% correct
classification with all the BSCC heart valves samples.

Figure 8 Results of the double-blinded in vitro test

7. Conclusion
A novel electromagnetic method has been developed for noninvasive detection of
cracks in the outlet strut of BSCC prosthetic heart valves. With the valve replicas, this
detection method has demonstrated a POD of 86.4% for a 0.4 mm deep crack, and a
POD of 97% for a 0.6 mm deep slot in the strut. In vitro tests effectuated through
described method have allowed the correct classification of the valves (good valves and
valves with defects)

248

R. Grimberg et al. / Noninvasive Characterization of BSCC Prosthetic Heart Valves

References
[1]

P. Hedger, Important updated information for physicians about patients with Bjrk-Shiley convexoconcave heart valves Dear Doctor letters, Irvine, CA: Shiley Inc. March/April, 1993.
[2] ISO 5840:1996 (E) Cardio-vascular implants cardiac valve prostheses
[3] EN 12006-1:1999 Non-active surgical implants Particular requirements for cardiac and vascular
implants: Part I. Heart valve substitutes
[4] W.W. ONeil, J.G. Chandler, G.T. OConnor, Radiographic detection of strut separation in BSCC
valves, New Engl. J. Med, 333, (1995), pp. 414-419
[5] J.W. Candy, H.E. Jones, Classification of Prosthetic Heart Valve Sounds: A Parametric Approach, J.
Acoustic Soc. of America, 97, 6, (1995), pp. 3675-3687
[6] S. Udpa, New electromagnetic methods for the evaluation of prosthetic heart valve, J. Appl. Phys., 90,
(2002), pp. 1-5
[7] S.C. Chan, R. Clifford, S. Majunar, N. Nair, S. Ramakrishnan, Y. Li, P. Ramuhalli, L. Udpa, S. Udpa,
Novel Methods for detecting fractures in prosthetic heart valves, INSIGHT, 47, (2005), pp.15-19
[8] E. Radu, R. Grimberg, A. Savin, O. Mihalache, Modeling the operation of the eddy current transducer
with orthogonal coils in the presence of material discontinuities, Sensors and Actuators, A, 59, (1997),
pp. 201-204
[9] R. Grimberg, A. Savin, E. Radu, O. Mihalache, Nondestructive Evaluation of the Severity of
Discontinuities in Flat Conductive Materials Using the Eddy Current Transducer with Orthogonal Coils,
IEEE Trans on Mag. 36, 1, (2000), pp. 299-307
[10] J. Lemaitre, J.L. Cheboche, Mechanics of Solids Materials, Cambridge University Press, 1990
[11] S C Chan, R Clifford, S Majumdar, N Nair, S Ramakrishnan, Y Li, P Ramuhalli, L Udpa, S S Udpa,
Novel Methods for detecting fractures in prosthetic heart valves, INSIGHT, 47, 15-19
[12] R.C. McMaster, R.C.P. McIntire, M.L.Master, Nondestructive testing handbook (2nd Ed), 4
Electromagnetic testing, American Society for Nondestructive Testing, London, 1986.

This paper is supported by Romanian Ministry of Education and Research Research of Excellence Program, Contract no. 6110/2005 SINERMAT and CNCSIS
Grant no.586/2006.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-249

249

Remote field eddy current control using


rotating magnetic field transducer.
Application to pressure tubes examination
Adriana SAVINa, Lalita UDPAb, Rozina STEIGMANNa,
Alina BRUMAc, Raimond GRIMBERGa1, Satish S.UDPAb
a

National Institute of R&D for Technical Physics, Iasi, Romania


b
Michigan State University, East Lansing, USA
c
Faculty of Physics, Al.I Cuza University, Iasi, Romania

Abstract. In this paper we propose to investigate the possibility of obtaining the


remote field effect for the transducer with rotating magnetic field and to use this
for detecting the artificially discontinuities practiced on un-irradiated pressure
tubes samples.
Keywords. Remote field effect, eddy current transducer with rotating magnetic
field, propagator matrix

Introduction
The remote field eddy current (RFEC) technique was originally developed for the
nondestructive examination of ferromagnetic tubes [1]. Relatively recent RFEC has
started to be used for examination of magnetic pipes and tubes [2] and was applied at
the evaluation of metallic parts, too [3]. The RFEC technique shows few distinctive
characteristics compared to those of conventional eddy current examination: they are
equally sensitive to ID and OD defects; has insensitivity to probe wobble or variable
lift-off, defect indications in the signal always appears double and with the same
strength.
In this paper we propose to investigate the possibility of obtaining remote field
effect for the transducer with rotating magnetic field [3] and to use it for detecting the
artificial discontinuities practiced on un-irradiated pressure tube samples.

1. Theoretical Aspects
The state equations for a cylindrically layered medium can be derived from Maxwells
equation, which are

1
Corresponding Author: Raimond Grimberg, National Institute of R&D for Technical Physics, 47
D.Mangeron Blvd., Iasi, 700050, ROMANIA, grimberg@phys-iasi.ro

250

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

u E

jZP H

u H

 jZH E

(1)

using a cylindrical coordinate system U , I , z , the system (1) can be written as


Ez
Ez

E
d EI
I
U U
Hz
d U Hz


HI
H I

(2)

where U U is a 4x4 matrix having the form

U U

jk n
z
ZPU

0


jZH 


jk z 2

ZP

jk z n

ZPU

 jZP 

ZHU

jZP 

jk z 2
ZH

jk z n

ZHU

1


jk z n
jn 2

ZHU 2

0
0

(3)

The eq. (2) is a equation with own vectors, having 4 linear independently solutions.
The form of own vectors can be determined if the expressions of the field created in
free space by the emission part of the transducer with rotating magnetic field are
inserted in eq.(2)
f

jZP0 f
k
n
j nI  k z
EU U , I , z
dk z e z jA1 z H n(1) ' k U U  jA2 2 H n(1) k U U
2
k
k
4S n f f
U
U
U

(4)
f f
kz n

jZP0
1
j nI  k z z
(1)
(1)

U
'
U
EI U , I , z
dk
e
A
H
k
A
H
k



U
U
1 n
2
n
2
z
4S 2 n f f
kU
k U U

f f
jZP0
Ez U , I , z
dkz e j nI kz z A1H n(1) kU U
4S 2 n f f
where the prime sign represents the derivation of Bessel function

A1
f

I0 J n kU R R
1  e jnS  e

2S
3

Lz
L
sin c k z z f ;
2
2
5S
 jn
 jn 23S
3

e
e

A2

j 43S
e

L
2 I 0 sin k z z nf J n k U r0 dr0
2 0

4S
 jn
 jn S3

3

e
e

I0 the amplitude of the three phase current; Lz the dimension of one of emission coli
after z axis; 2R dimension after U direction; k U

Z 2 P0H 0  k z 2 , with Im(kU)>0.

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

251

The components of magnetic field created by the transducer in vacuum are immediately
obtained taking into account that
1
(5)
u E (r )
H (r )
jZP0
Inserting (4) and (5) in (3) we obtain the eigenvectors

v1

J n (kU U )

nk

 2 z J n (k U U )
kU U

,
0

jZH

J n '(k U U )

k
U

v3

H n (1) (k U U )

 nk z H (1) (k U )
U
n
2
kU U

, v4
0

jZH

(1)

H n '(k U U )
k U

v2

jZP

J n '(k U U )

kU

J
k
(
U
)
n
U

nk z
 2 J n (kU U )
k
U

(6)

jZP

(1)

H n '(k U U )
kU

(1)
U
H
(
k
)
n
U

nk z

(1)
 2 H n (k U U )
k U U

> v1 , v2 , v3 , v4 @

(7)

In the case in which the source is into a layered cylindrical medium, the fields in all
media can be calculated using the method of propagator matrix [4].

Ez ( U )
E (U )
I

H z (U )

H I ( U )

Ez ( U )
E (U )
1
I

v( U )v ( U ')
H z (U )

H I ( U )

Ez ( U )
E (U )
I

P( U , U ')
H z (U )

HI ( U )

(8)

1

P( U , U ') v( U )v ( U ')
Now, equipped with the propagator, we can solve the transmission and reflection
fields through a cylindrical layered medium.
The field in air, created by the emission part of the transducer with rotating
magnetic field was previously calculated (Eqs. (18) and (19) from [5]), the expressions
of dyadic Greens function for layered cylindrical medium (Eqs. (22), (28) and (30)
from [5]) have been developed, also.
To solve the forward problem, a discretization with the moments method in pointmatching variant has been used. The numerical code was developed in Matlab 7.0.
The distance between the center of transducers emission part and those of the
(
reception part is D, so that for the calculus of the dyadic Greens function G21 , the
supplementary factor e

jk z D

must be inserted.

252

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

2. Numerical Simulations

The transducer with rotating magnetic field which works in RFEC conditions is made
from 3 orthogonal coils making 2S/3 angle between them and supplied with a triphased current system.
The pressure tubes, on which the experimental measurements were made, are
confectioned from Zr 2.5%Nb alloy having inner diameter 103mm, outer diameter
111.4mm and electrical conductivity 1.89x106S/m.
In figures 1 a and b we present the responses of transducer, obtained by solving
forward problem for the case D=0 for a slot with 6x0.2x0.2mm placed on internal
surface, and respective on external surface of pressure tube.

b
Figure 1. The response of the transducer for D=0; 6x0.2x0.2mm slot
a) slot is made on internal surface; b) slot is made on external surface

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

253

The condition in which the transducer functions in RFEC mode is obtained


modifying D. This is obtained, in the case of pressure tubes with dimensions and
electromagnetic properties mentioned above, if the distance D becomes equal to 1.8 x
outer diameter of pressure tube. This situation is presented in figure2.

b
Figure 2. The response of the transducer for D=1.8x outer diameter; slot dimensions 6x0.2x0.2mm
a) the slot is made on internal surface; b) the slot is made on external surface

The analysis of the data presented in Figure 2 shows the presence of two
pronounced peaks, the distance between them being a little bigger than D. The peaks
have equal amplitude, being indifferent by the position of the discontinuities (on inner
surface, respective on outer surface). It must be mentioned the existence of a signal
placed at the middle of the distance between two peaks and having amplitude relatively
small. This supplementary peak is due, probable, to the shape of the source [5].

254

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

3. Experimental Results

In the procedures for eddy current examination of pressure tubes from PHWR reactors,
CANDU type, the minimum detectable defect must be circumferential or axial slots
with 6x0.2x0.2mm practiced on inner and outer surface of the tubes.
With the transducer with rotating magnetic field functioning in RFEC mode and
the equipment described in [3], experimental measurements were effectuated. The
optimal conditions are: frequency 47 kHz, D=200mm.
In Figure 3 a) we present the response of the equipment for one axial slot described
in figures caption and in figure 3 b is presented the picture of the same slot.

a)

b)
Figure 3. The signals, remote field type, delivered by a 6x0.157x0.189mm axial slot practiced on external
surface of pressure tube sample: a) signals; b) Photo of discontinuity.

In Figure 4 we present the same features for a circumferential slot.


The obtained experimental results confirm the righteousness of the theoretical
model developed above, for the case of circumferential slot, the central peak, unusual

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

255

in the case of axi-symmetric excitation, becomes well marked. This is due probable to
the source.

b
Figure 4. The signals, remote field type, delivered by a 6x0.19x0.19mm circumferential slot practiced on
internal surface of pressure tube sample. a) signals; b) photo of discontinuity.

4. Conclusions

A method for calculation of the field generated by the eddy current transducer with
rotating magnetic field using propagator matrix method was developed. The
effectuated calculi allow a simpler solving of the forward problem, the optimal distance
emissionreception which assures the functioning in RFEC mode being determined.
The experimental measurements are in good concordance with the numerical calculus,
which confirm the righteous of model.

256

A. Savin et al. / Remote Field Eddy Current Control Using Rotating Magnetic Field Transducer

5. References
[1]
[2]
[3]

[4]
[5]

T.R. Schmidt, The remote field eddy current inspection technique, Materials Evaluations, 8, (1984),
225-230
D.L. Atherton, Remote Field Eddy Current Inspection, IEEE Trans on Magnetics, 31, 6, (1995), 41424147
R. Grimberg, L. Udpa, A. Savin, R. Steigmann, S.S. Udpa, Inner-Eddy-Current Transducer with
Rotating Magnetic Field: Experimental Results, Application to Nondestructive Examination of Pressure
Tubes in PHWR Nuclear Power Plants, Research in Nondestructive Evaluation, Springer-Verlag, New
York, LLC, vol 16, issue 2, (2005), 65-78
W.C. Chew, Waves and Fields in Inhomogeneous Media, Von Nostrand Reinhold, NY, Chapter 3 and 7,
1995
R. Grimberg, L. Udpa, A. Savin, R. Steigmann , S.S. Udpa, Inner-Eddy-Current Transducer with
Rotating Magnetic Field: Theoretical Model, Forward Problem, Research in Nondestructive Evaluation,
Springer-Verlag, New York, LLC, vol 16, issue 2, (2005), 79-100

This paper is supported by Romanian Ministry of Education and Research - Research


of Excellence Program, Contract no. 6110/2005 SINERMAT, Contract No.49/2006
ROLIGHT and Nucleus Program, Contract No. PN 06 - 38 01 03.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-257

257

Lifetime Prediction of Pressure Tubes in


PHWR Nuclear Power Plants using Eddy
Current Data
Raimond GRIMBERG, Adriana SAVIN, Rozina STEIGMANN,
Aurel ANDREESCU, Nicoleta IFTIMIE, Marius Mihai CAZACU
National Institute of R&D for Technical Physics, Iasi, Romania

Abstract: In this paper is presented a model for lifetime prediction using Markov
hidden chains method, starting from the results of eddy current nondestructive
evaluation using rotating magnetic field transducer. The model is trained with
previous experience in pressure tubes examination and allows the determination of
probability that the tubes shall be found in different imposed states.

Keywords: life time prediction, pressure tubes, degradation state.

Introduction
Pressure tubes (PT) assure the cooling of the fuel channels in nuclear power plant
CANDU type. During the service of nuclear power plants, it appears hydrogen due to
the zirconium corrosion, which is absorbed by the material of manufactured tubes, Zr2.5%Nb alloy. The absorbed hydrogen forms the zirconium hydrides which are
decreasing the materials resistance. Under the influence of hydride, the incipient cracks
in pressure tubes can develop in unstable and uncontrolled forms, being named
Delayed Hydrogen Cracking (DHC) phenomena. The statistically analysis is made for
the existent data at the IAEA Vienna and indicates a slowly evolution to a dangerous
state systems. To describe the degradation processes we use Markov processes [1]. A
Markov process is a stochastic one with the properties that the given value of X(t), at
time W, where W > t, are independent of the values of X(u), u < t. For N states, the
probabilities of transition between all possible states pairs are given by

P
11
P
21
.
P
N 1

12

22

.
P

N2

P
1N

... P
2N

.
.
... P
NN
...

(1)

258

R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants

1. Markov hidden chains model


Evaluating the lifetime for pressure tubes represents a probabilistic model on basis of
continue Markov chains. The system consists in pressure tubes of CNE CANDU type.
A Markov chain is governed by the transition matrix P, where Pjk(t) elements is the
probability of transition of the system from the current state j to state k:
Pjk t

P X t

k X 0

j ; j, k S , t ! 0

(2)

There are defined the following possible states in which we can find the PT system
in function of the degree of deterioration; we define: OK- state in which PT is not
degraded; D1 - state in which PT are in slowly degradation state; depth limit of defect is
0.15mm; D2 : - state in which PT has degradation in relatively major degree; the depth
limit of defect is 0.5mm; F1- PT system can be in critical degradation state; the depth
limit of defect is 1.6mm [2], case in which the tube is replaced. The system states D
are included in two categories: states with defects detected at control effectuated D1d
and D2d and states with defects undetected at control effectuated D1u and D2u. The
overall failure model is presented in figure 1

Figure 1. Overall failure/maintenance model,

The transition matrix P for our model is

OK D
1u

1
0

0 1  q1

0
0
q
0
2
q
0
3
1
0

0
1

0
q

0
0

0
0

0
0

1
0

0
1 q

0
0

0
0

1 q

1d

2u

2d

*
OK

0
0

(3)

Further, we introduce the probabilities that degraded states are detected by the
inspection:

R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants

259

x
x

q1 - probability that state D1 of the system is detected;


q2 - probability that a degraded failure is detected by the inspection; it is
unknown in advance if the state D1 was reached;
x q3 -probability that a degraded failure D2 is detected by the inspection; it is
known in advance that the state D1 was reached.
The dotted lines of Figure 1 indicate transitions at the end of the test interval.

2. Experimental results
The model for prediction uses so previous experience, it means the data obtained by a
inspection of high number of pressure tubes (10.000), specific data for the examination
method, equipment, analysis team, characterized by probability of detection and
evaluation, as well as the results of current inspection. The method was developed for
the case of eddy current examination of pressure tubes using a transducer with rotating
magnetic field [3] (Figure 2a) and the adequate measurement system [4], Figure 2b

Figure 2 Experimental set up: a) the transducer with rotating magnetic field; b) the control equipment

2.1. Probability of detection for discontinuities with depth of cca 0.15mm.


In this scope it was used the sample ARG 1 from Argentina in frame of a Contract
between NIRDTP Iasi and IAEA Vienna [5]. It was used the unirradiated PT samples
with artificial flaws of different types, positions and geometrical dimensions made by
EDM. The inspection defects are noted with #1, #2, #3 and #4 with dimensions and
orientations given bellow.
Table 1: Flaw details in ARG 1 sample
Flaw #

Location and Orientation

Length (mm)

Width (mm)

Depth (mm)

Characteristics

ID, axial

6.1

0.3

0.136

calibration slot

OD, axial

6.15

0.4

0.14

calibration slot

ID, circumferential

6.2

0.3

0.152

calibration slot

OD, circumferential

6.6

0.4

0.16

calibration slot

The distribution for the estimated defect severity with depth ~0.15 mm is presented
in figure 3a using a method to solve inverse problem for EC [1]. The location of defect
on internal, respective external surface of the tube is established from the phase

260

R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants

information, this is reducing the number of discretization cells and is increasing the
robustness of inversion algorithm.
Considering the probability distribution for defects of log normal type (continue
curve by figure 3) and integrating in range 0.1 0.3, it was a detected probability of
46% with reliability coefficient 95%.
2.2. Probability of detection for discontinuities with depth of ~ 0.5mm.
The inspection defects are noted with #1 and #4 with dimensions and orientations
given of KOR 1 sample.
Table 2 Flaw details in KOR 1 sample
Flaw
#

Location and
Orientation

Length
(mm)

Width
(mm)

Depth
(mm)

Characteristics

OD, axial

6.0

0.3

0.47

Short notch deeper than calibration slot

OD, circumferential

6.0

0.3

0.41

Short notch deeper than calibration slot

The histogram and log normal distribution for estimations concerning the defect
with 0.5 mm depth is presented in figure 3b.

Figure 3 The estimated distribution for flaw: a) with depth 0.15mm; b) with depth 0.47mm

Using the method described above, the probability of detection and estimation for
the discontinuity with depth ~0.5mm is 70% with a reliability coefficient of 95%.
The probability of detection for discontinuities with depth of ~1.63mm was 98%
with reliability coefficient 95%.
Using the model of life/maintenance time prediction described above, the
probabilities that a PT examined by eddy current with the transducer with rotating
magnetic field, shall be in one of states D1u (there are undetected discontinuities with
0.15mm depth), respective D2u (there are undetected discontinuities with depth equal or
smaller than 0.5mm), has been determined.
The results obtained through simulation, based on the existent statistic about the
states of the PT from CANDU nuclear power plants and on the probabilities of
detection, experimental determined on EDM slots practiced on unirradiated pressure
tube samples are presented in the next figures. The figures 4 and 5 show the probability
to detect a pressure tube to be found in the degradation state (D1u or D2u) in function of

R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants

261

time range between the inspections. The probabilities are referred to m*day of PT with
the reactor at full day exploitation.

Figure 4 The probability for state D


1u

Figure 5 The probability for state D


2u

We can see that in time range between 300-400 days is minimum time up at the
inspection following. Hence, it results that in this range the chance as a flaw is
undetected is minim.
The decreasing of the probabilities that the PT shall be in one of states D1u and D2u
is not due, evidently, to the decreasing of the number of flaws, but to the fact that, these
flaws are evolution in time; these are rather in the states D1d and D2d.
The input parameters for the estimation of model parameters are listed in table 3.
The statistical data was taken on 10,000 pressure tubes in range time 10-12 years
by IAEA Vienna.
Table 3: Inputs to parameter estimation
Parameter definition

Parameter

Value

Number sets

10 000

Length of pressure tubes

62 000

Number of tests/inspections 1989-2002

nTF

11.3

Number of tests/inspections 1991-2002

nTD

Number of days, 1989-2002

N1

4661

Number of days, 1991-2002

N2

3909

262

R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants

Length of test/inspection interval

360

Number of observations in state D1, (i.e. transitions from D1u to D1d)

ND1

187

Number of observations in D2 when it was not known that state was


degraded, (i.e. transitions from D2u)

ND2a

238

Number of observations in D2 when it was known that state was


degraded, (i.e. transitions from D1d)

ND2b

20

Number of observations in F1

NF1

83

Probability of detecting D1 failure for coefficient of 95%

q1

0.48

Probability of detecting D2 failure at test (state D1n not detected


previously) for coefficient of 95%

q2

0.70

Probability of detecting D2 failure at test (state D1 already detected)

q3

0.98

Rate of detecting D2 in additional inspections; (assuming on the


average two additional inspections within each interval T)

(2/T)* q3

From here results a conclusion with high practical importance namely that for the
aim of obtaining a maximum POD for the discontinuities of PT, these must be
examined at an interval of 300-350 days, when the probability of existence of the states
D1u and D2u are minim. Upper 600 days in the range time between two inspections the
undetected probability of the flaws can be dangerous for reactor running.

3. Conclusions
It was elaborate a Markov model for the life time prediction for the pressure tubes. The
initial values with which we have to work are given in table 3.
We developed a soft using MATLAB 7.0 programmer which uses a series of
apriority knowledge which appear as accumulated experience in the exploitation of
nuclear reactors PHWR type. Another data are obtained by nondestructive testing for
pressure tubes at annual outage. This model has more simplifications for interpretation
of the results and to use for maintenance optimization.
References
[1]
[2]
[3]
[4]

[5]

Raimond Grimberg et al. Emerging Technologies in Non-destructive Testing, ETNDN Fourth


International Conference, April 2 4, 2007, Stuttgart, Germany;
Periodic Inspection of CANDU Nuclear Power Plant Components, CAN/CSA N285.4, (1994)
R. Grimberg, Lalita Udpa, Adriana Savin, Rozina Steigmann, S. Udpa, Inner Eddy Current Transducer
With Rotating Magnetic Field. Theoretical Model Forward Problem, Research In Nondestructive
Evaluation, Springer-Verlag New York, Llc, Vol 16, Issue 2, (2005), 79-100
R. Grimberg, Lalita Udpa, Adriana Savin, Rozina Steigmann, Satish S. Udpa, Inner Eddy Current
Transducer With Rotating Magnetic Field; Experimental Results: Application To Nondestructive
Examination Of Pressure Tubes In Phwr Nuclear Power Plants, Research In Nondestructive Evaluation,
Springer-Verlag New York, Llc, Vol 16, Issue 2, (2005), 65-78,
IAEA TECDOC 2005 , Report for CRP Meeting Chalk River Canada - October 2005

This paper is supported by Romanian Ministry of Education and Research Research of Excellence Program, Contract no. 6110/2005 SINERMAT and Nucleus
Program, Contract No. PN 06 - 38 01 03.

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-263

263

Electromagnetic Non-Destructive
Evaluation of Reinforced Concrete Rebars
Maxim MOROZOV a, Guglielmo RUBINACCI b, Antonello TAMBURRINO c and
Salvatore VENTRE c, 1
a
CREATE Consortium, Naples, Italy
b
Ass. EURATOM/ENEA/CREATE, DIEL,
Universit degli Studi di Napoli Federico II, Italy
c
Ass. EURATOM/ENEA/CREATE,DAEIMI, Universit degli Studi di Cassino, Italy

Abstract. This paper concerns quantitative imaging, consisting in finding location,


direction and size, of concrete rebars by means of eddy current measurements and
an innovative inversion method. Owing to an accurate numerical modelling of the
probe-rebar interaction, an analysis of issues to be considered for the quantitative
imaging of rebars is conducted, which leads to significant simplifications in the
numerical model and enables developing an accurate and computationally efficient
imaging method. Results of experimental testing demonstrate favourable validity
of the proposed numerical model.
Keywords. Nondestructive evaluation, eddy currents, numerical simulation,
reinforced concrete rebars

Introduction
This paper presents critical considerations for application of an innovative inversion
method to the problem of quantitative eddy current (EC) imaging of reinforced
concrete rebars and optimisation of an apposite EC probe. The EC imaging is aimed to
determine location, direction and size of rebars. From a broader perspective,
electromagnetic imaging of concrete rebars is attracting a growing interest because of
the need of monitoring the health of existing structures that may become unsafe or
collapse when, due to corrosion or damage of reinforcement bars (rebars), they cannot
longer support the tensile load how it was originally designed. Several sensing methods
have been developed, among them we mention techniques based on impedance probes
[1]-[4], static magnetic field measurements, such as residual magnetic flux density and
magnetic flux leakage [5], microwave tomography [6, 7] and polarization resistance
based techniques [8].
The respective EC commercial instrumentation (the pachometer) makes use of
calibration technique performed for a set of various rebar dimensions, which is
ineffective for complex configurations, such as crossed bars etc. Our method benefits
from an accurate numerical modelling of the EC probe interaction with rebars. As a
result the possibility of relevant simplifications in the numerical model is outlined,
1
Corresponding Author: Salvatore Ventre, Universit degli Studi di Cassino, Via G. Di Biasio 43, 03043 Cassino (FR), Italy; E-mail: ventre@unicas.it

264

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

allowing the development of an accurate and computationally efficient method. The


respective numerical model is based on a integral formulation requiring the
discretization of only the conducting and/or magnetic regions such as the iron rebars
and eventual ferrite core of an EC probe. An efficient numerical model is essential to
develop a quantitative imaging system in view of an EC probe design as well as of
imaging procedures based on experimental data. The key feature of an inversion
method suitable for EC imaging is the maximally possible independence on the
geometrical configuration. It must be capable to work properly in non a priori known
configurations consisting of, for instance, either a single or more rebars eventually
interacting, non parallel and located at different depth. The choice is made of an
algorithm based on a monotonicity property of the unknown-data operator recently
developed by the authors. The monotonicity property, that is characteristic of systems
governed by elliptic PDEs, has been exploited to develop an inversion algorithm in
eddy current testing of conductors and it gives rise to a fast imaging algorithm having a
computational cost associated to the solution of a number of elliptic forward problems
proportional to the number of voxels used to discretize the unknowns.

1. Numerical method
1.1. Forward analysis
An appropriate and effective numerical model is essential to develop a quantitative
imaging system. It is essential to design the probe and during the processing (imaging
procedure) of the experimental data. The modeling of the interaction between the
rebars and the probe is a complex issue. First, the ferromagnetic material has a
nonlinear characteristic, second the skin depth in the iron is significantly smaller
already at relatively low frequencies. The first issue, potentially requiring a nonlinear
numerical model, can be disregarded by noting that in the typical inspection cases the
magnetic flux density produced by the inducing probe is low enough so that nonlinear
effects can be neglected. The second issue asks for a refined mesh in the outermost
layer of the rebars, where the fields decay rapidly to zero along distances of the order
of the skin-depth. In an integral formulation, such as the one presented in the following,
this can be conveniently taken into account by discretizing only this outermost layer
(having a thickness of the order of the skin-depth) of the rebars, thus saving elements,
unknowns and in ultimate analysis, computational time and resources. In the following
we neglect the electrical conductivity of the concrete. This is possible because the
typical values for the electrical conductivity are fractions of S/m, thus giving a skindepth much larger (the skin-depth at 100kHz and 1S/m is 159cm) than the typical sizes
of interest (up to 40cm) of the problem.
The numerical model considered in this work, and here briefly described, is based
on a integral formulation [9-11] requiring the discretization of only the conducting
and/or magnetic regions such as the rebars (made of iron) and the magnetic core of the
probe array. As mentioned, for frequencies such that the skin-depth is much smaller
that the rebars diameter, it is sufficient to discretize only an outermost layer of the
rebars for a thickness of the order of the skin depth. In this formulation, the eddy
currents are represented in the finite dimensional space of curl of edge element shape
functions Nks: J=uT where the electric vector potential T is expanded as linear
combination of Nks. A two-component gauge condition [9] guarantees the uniqueness

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

265

of the vector potential. The magnetization vector is numerically approximated as a


piecewise constant vector function (hereafter Pk stays for the k-th shape function used
to represent the magnetization M). The field equations in the magneto-quasi-stationary
limit (time harmonic operations) are automatically solved once that the electric field is
expressed as E  jZA  I where the magnetic vector potential is calculated by
means of the Biot-Savart law. Finally, the numerical model is obtained by applying the
Galerkin method to the constitutive equations J E and M ( P r  1) P 0 P r B :

[M  kB]dV

0, k

(1)

Vm

u N

(KJ  jZA)dV

0, k

(2)

Vc

By doing this, we obtain the following algebraic linear system:

R  jZL I  jZFM
k

1

DE MF I

(3)

(4)

where k P r  1 / P 0 P r , I is the column vector of the complex coefficients of the


expansion of the current density J in terms of the shape functions u N k s, M is the
column vector of the complex coefficients of the expansion of the magnetization in
terms of the shape functions Pks, and:
Rij

Lij

Vc

u N i V 1 u N j dV

P0

4S

Vc Vc

P0
4S

1

r  r ' u N i r u N j r ' dV dV '


1

r  r ' Pn ,i r Pn , j r ' dSdS '

Eij

Dij 

Fij

P0
4S

Uj

 jZ u N j A 0 dV

Dij

Wj

Vc Vm

wVm ,i wVm , j

Vm

Pi P j dV

3

r  r ' u N i r P j r ' u r  r ' dV dV '

Vc

Vm

(5)

B 0 dV ,

(6)

(7)

(8)
(9)
(10)
(11)

266

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

being A0 and B0 the vector potential and the magnetic flux density produced by the
sources in the free-space, Vc and Vm the conductive and magnetic domains,
respectively, and Vm,j the volume of the j-th element of the finite element mesh
discretizing the magnetic domain Vm.
After some manipulations, it can be possible to prove (see [12]) that the impedance
variation due to the presence of the rebar can be expressed as follows:

GZ

jZ
J A 0 d V  M B 0 dV
2
i Vc
Vm

(12)

where i is the complex amplitude representing the current circulating in the inducing
coil. Similarly, when a coil array is used as a probe it can be proved that

GZ kj

jZ
2 ik i j

J j A 0k  J k A 0j dV  M j B 0k  M k B 0j dV
Vc

Vm

(13)

where iD (D=j, k) is the complex amplitude of the current circulating in the k-th
inducing coil, JD, M D , A D0 and B D0 are the eddy current, the magnetization, the freespace vector potential and magnetic flux density when only the current iD
circulating.

is

1.2. Inverse problem


A key feature of an inversion method suitable for this class of problems is the
capability to be, as much as possible, independent on the geometrical configuration. It
must be capable to work properly in non a priori known configurations consisting of,
for instance, either a single or more rebars eventually interacting, non parallel and
located at different depth.
The choice is fall on an algorithm based on a monotonicity property (see below) of
the unknown-data operator recently developed by the authors. The monotonicity
property, that is characteristic of systems governed by elliptic PDEs, has been first
exploited to develop an inversion algorithm in the framework of Electrical Resistance
Tomography [13], and then extended to Eddy Current Testing of conductors [14] and it
gives rise to a fast imaging algorithm having a computational cost associated to the
solution of a number of (elliptic) forward problems proportional to the number of
voxels used to discretize the unknowns.
In Electrical Resistance Tomography the data processed by the inversion algorithm
is the NN resistance matrix made by the self and mutual resistances between pairs of
N electrodes (plus one grounded electrode) located on the boundary of the conductive
specimen under test. In Eddy Current Testing of conductors the measured quantity is,
similarly, the impedance matrix (self and mutual impedances between pairs of coil used
to probe the conductive specimen) measured at several frequencies. Here, in the
framework of rebars inspection, the measured quantity is still the impedance matrix but
measured at large enough frequencies. We notice that by increasing the frequency, the
magnetic flux density inside the ferromagnetic rebar vanishes. Therefore, as long as the

267

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

displacement current is negligible, the magnetic flux density is solution of a proper


magneto-static (elliptic) problem and, therefore,
ZV
Im m
Z

Z large
o LVm

(14)

is the measured impedance matrix, and LV

where ZV

is the inductance matrix

corresponding to the magnetostatic problem with the constraint b=0 in the


ferromagnetic region Vm. Thanks to the fact that LV arises from a magnetostatic
m

problem, it can be shown that


DE DD G L D  G L D is positive semi - definite
D

(15)

where G L D L )  L D , L ) is the free-space inductance matrix and L D ( L D ) is


k

the inductance matrix related to the magnetostatic problem corresponding to the


constraints b=0 in DD (DE).
By reversing (15) and setting DE=Dk and DD=Vm, we have that

G L V  G L D is not positive semi - definite Dk  Vm


m

(16)

where Dk has the role of test volume and G L D is numerically computed ( G LV is


m

measured experimentally). By properly varying the position and size of the test domain
Dk, and by applying (16) to each different test domain, we can retrieve the size and
position of each single rebar. Moreover, (16) holds regardless the fact that the rebars
may interact or not and regardless the mutual position of the rebars. In this way we
overcome the detection and sizing by means of calibration charts and assuming a priori
known geometrical configurations. The first imaging algorithm based on these concepts
been presented in [1].
Finally, we highlight again that G LV and G L D are NN matrices, where N is the
m

number of coils used to probe the rebars. Increasing the number of measurement coils
(up to a certain extent), makes the test more effective due to the increase of the
information content of the measurement.

2. EC probe optimisation
The iron bars under test had diameter of nearly 19 mm. The electrical conductivity V of
the steel bars was found to be 4.61 MS/m, and their relative magnetic permeability Pr
was determined to be 85 [2]. At the same time it was confirmed that (i) nonlinear
effects are negligible and that (ii) Pr is constant at the frequencies of interest.

268

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

In order to produce the inductance matrix necessary for solution of the inverse problem
the eddy-current testing of steel bars used in reinforced concrete has been conducted
with an array of tree coaxial induction coils which ensures a significant mutual
coupling. The diameters of the coils as well as the excitation frequencies have been
numerically optimized in order to produce maximum relative response due to an iron
bar with respect to the intrinsic impedance of coils in absence of a test piece.
Figure 1 shows dependence of the EC response on the coil diameter at the
excitation frequency of 100 kHz for different couples of the excitation/measuring coils,
for various lift-offs of the array above the iron bar. According to (14) we are interested
in the imaginary part of the EC response. The optimum diameter of the most inner coil
appears to be 65 mm, with the successive coils having diameters 70 mm and 75 mm.
The height of the coils is 5 mm.

Figure 1. Optimisation of coil diameter for inspection of an iron bar of diameter 19 mm. Target lift-off is 20
mm, excitation frequency = 100 kHz. The EC response is represented by the imaginary part.

Figure 2. Optimisation of the excitation frequency for inspection of an iron bar of diameter 19 mm, coaxial
coils array of inner diameter 65 mm, lift-off 20 mm. The EC response is represented as the real and
imaginary parts.

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

269

Figure 3. Measured and simulated eddy current responses to an iron bar of diameter 19 mm, obtained with
coaxial coils array of inner diameter 65 mm, lift-off 20 mm, excitation frequency = 100 kHz. The EC
response is represented by the imaginary part.

Figure 2 shows dependence of the EC response on the excitation frequency for all
the coupled pairs of coils with the optimized diameter of the most inner coil (65 mm),
with the lift-off of the array above the iron bar being 20 mm. The optimum excitation
frequency is 100 kHz. The impedance of the coil has been measured with an LF
(bandwidth 5 Hz - 13 MHz) impedance analyzer HP-4192A at the optimum excitation
frequency of 100 kHz. The lateral scanning of the iron bar with a coil array of the
optimized size at excitation frequency of 100kHz has been conducted automatically by
a Mitsubishi 6-axis Melfa robot RV-1A. The bar has been placed in the horizontal
plane and the coil has been moved across the bar at various lift-offs from 1mm to

270

M. Morozov et al. / Electromagnetic Non-Destructive Evaluation of Reinforced Concrete Rebars

40mm with step of 1mm. Figure 3 shows the comparison of measured and calculated
signal contribution due to the iron bar, represented as the inductive part. Although there
is a discrepancy for the auto inductance variation, there is a good agreement for the
signals obtained due to the mutual coupling (12, 13, 23).

3. Conclusions and outlook


This work has been focused on quantitative rebar detection and sizing for arbitrary
geometrical configurations. The main contribution of the study are the following: (i) a
critical analysis of the issues that must be considered for the quantitative imaging of
reinforced concrete rebars; (ii) optimisation of an EC probe for rebars imaging; (iii)
experimental validation of the numerical model and its underlying hypothesis. It has
been found that the measured EC response due to an iron bar is stronger and more
accurate with respect to simulation in case of mutual coupling among the probe coils
rather than in case of auto-inductance.
Future work will address assembling together the numerical model with the
inversion algorithm for processing experimental data collected in a realistic setting.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]

G. Rubinacci, A. Tamburrino, S. Ventre, Concrete rebars inspection by eddy current testing,


International Journal of Applied Electromagnetics and Mechanics, Vol. 25 (2007), pp. 333339
M. de Magistris, M. Morozov, G. Rubinacci, A. Tamburrino, S. Ventre, Electromagnetic Inspection of
Concrete Rebars, COMPEL, Vol. 26, No. 2 (2007), pp. 389-398
G. Miller P. Gaydecski, S.Quek, B. T. Fernandes and M. A. M. Zaid, Detection and imaging of surface
corrosion on steel reinforcing bars using a phase-sensitive inductive sensor intended for use with
concrete, NDT&E International, Vol. 36 (2003), pp. 19-26.
T. Chady, R. Sikora, S. Gratkowski, S. Wojtowicz and S. Nagata Eddy current inspection of
reinforcement bars in concrete structures, Proceedings of The Tenth International Workshop on
Electromagnetic Nondestructive Evaluation, Michigan State University, June 1-2 2004.
J. Makar and R. Desnoyers, Magnetic field techniques for the inspection of steel under concrete
cover, NDE&E International, Vol. 34 (2001), pp. 445-456.
R. Zoughi, S. Gray and P.S. Novak Microwave non-destructive detection of rebars in concrete slabs,
Materials Evaluation, Vol. 49, no. 11 (1991), pp 1385-88.
Ch. Pichot and P. Trouillet, Diagnosis of reinforced structures: an active microwave imaging system,
Bridge evaluation, repair and rehabilitation; Proceedings of the NATO Advanced Research Workshop
on Bridge evaluation, repair and rehabilitation, Baltimore, Maryland, US, April 30 May 2 1990.
Law D.W., J. Cairns, S.G. Millard, J.H. Bungey (2004) Measurement of loss of steel from reinforcing
bars in concrete using linear polarization resistance measurements, NDT&E International, 37 381-388.
R. Albanese and G. Rubinacci, Finite element methods for the solution of 3D eddy current problems,
in Advances in Imaging and Electron Physics, Vol. 102 (1998), Academic Press, pp. 1-86
R. Albanese, G. Rubinacci, F. Villone, Crack simulation in the presence of linear ferromagnetic
materials using an integral formulation, in Electromagnetic Nondestructive Evaluation V (2001), J.
Pavo et al. Eds. , IOS press, pp. 16-21
R. Albanese, G. Rubinacci, A. Tamburrino, F. Villone, Phenomenological approaches based on an
integral formulation for forward and inverse problems in eddy current testing, Int. J. of Applied
Electromagnetics and Mechanics, Vol. 12, No. 3-4 (2000), pp. 115-137
G. Rubinacci, A. Tamburrino, S. Ventre, An efficient numerical model for a magnetic core eddy
current probe, accepted for publication on IEEE Trans. on Magnetics, 2008.
Tamburrino A., G. Rubinacci, A new non-iterative inversion method for Electrical Resistance
Tomography, Inverse Problems, Vol. 18 (2002), pp. 1809-29
A. Tamburrino, G. Rubinacci, Fast methods for quantitative eddy-current tomography of conductive
materials, IEEE Trans. Mag., Vol. 42, No. 8 (2006), pp. 2017-2028

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-271

271

Advanced Probe with Array of Pick-up


Coils for Improved Crack Evaluation in
Eddy-Current Non-Destructive Testing
Ladislav JANOUSEK a,1, Klara CAPOVA a, Noritaka YUSA b and Kenzo MIYA b
a
DEBE, FEL, University of Zilina, Zilina, Slovak Republic
b
IIU Corp., Tokyo, Japan
Abstract. The paper proposes a new probe for enhancing crack evaluation in
eddy-current non-destructive testing. The probe consists of one exciting coil and
two identical pick-up coils. The pick-up coils are situated at different locations
relative to the position of the exciting coil for gaining signals using various depth
profiles of eddy current density. The two signals are linearly superposed and a
certain indication of cracks depth is extracted from the resulting signal. Numerical
as well as experimental results demonstrate the effectiveness of the new probe in
the evaluation of a cracks depth. In addition, cracks much deeper than the
standard depth of penetration can be evaluated using the new probe.
Keywords. Non-destructive testing, eddy-currents, pick-up coil array, signal
superposition, cracks depth evaluation

Introduction
Scheduled in-service inspection (ISI) is necessary for the maintenance of structural
components in many industrial fields. When a defect is found during the inspection, the
cracked component can usually stay in service; however, it must be assured that the
dimensions of the crack will not exceed size limitations until the next scheduled ISI [1].
Frequently, ultrasonic-based methods are used for the sizing. However, such methods
are quite inefficient when inspecting cracks in certain structures, e.g. welds [2], where
electromagnetic methods are reported to provide good detection sensitivity [3].
One of the conventional electromagnetic methods utilized for the inspection of
conductive materials is eddy-current non-destructive testing (ECT). ECT signals are
integral values and do not carry explicit information about the cracks dimensions.
Several papers proposed to use numerical inversions for sizing [4]. However, the illposedness of the problem is not yet fully revealed [5]. Further improvements in eddycurrent non-destructive evaluation are therefore still necessary.
The authors have proposed to utilize various distributions of eddy currents during
the inspection for enhancing crack evaluation by ECT [6], [7]. An ECT probe
composed of several exciting coils and one pick-up coil has been employed. The
present paper proposes a new design of an ECT probe. Only one exciting coil drives
the eddy currents in a test-piece and two spatially distributed pick-up coils sense crack
signals. The signals are further processed to extract indication about a cracks depth.
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; Email: janousek@fel.uniza.sk

272

L. Janousek et al. / Advanced Probe with Array of Pick-Up Coils for Improved Crack Evaluation

1. Proposal of a New ECT Probe

exciter
Exciting coil:
- height: 30 mm
- length: 30 mm
- width: 10 mm
- winding thickness: 1 mm

21

pick-up 1

Pick-up coils:
- inner diameter: 1 mm
- outer diameter: 3 mm
- width: 1 mm

pick-up 2
9
Figure 1. Design of a new ECT probe

Figure 1 displays the design of a new ECT probe. The probe consists of one rectangular
exciting coil positioned tangentially relative to the surface of a tested object. Two
identical pancake pick-up coils located at different positions from the exciting coil
sense the signals. The two pick-up coils are situated 21 mm and 30 mm away from the
centre of exciting coil along its axis, respectively. The spatial distribution of the pickup coils assures that the two sensed signals of the same crack are obtained with
different depth profiles of eddy currents. The configuration and the dimensions of the
probe have been designed for an inspection frequency of 50 kHz.
The two signals are linearly superposed based on:

Re = C1 Re1 C2 Re2 , Im = C1 Im1 C2 Im2 ,

(1)

where Re1, Re2 are the real parts of the complex signals for the pick-up coils 1 and 2,
respectively; Im1, Im2 are the imaginary parts of the complex signals for the pick-up
coils 1 and 2, respectively; and C1, C2 are arbitrary numbers defining a ratio of the
superposition = C1 C2 . The numbers C1, C2 are changed in such a way that the ratio
is increased from zero to infinity. The resulting superposed complex crack signal (Re,
Im) is evaluated in respect to the value of the ratio.

2. Numerical and Experimental Results


A plate specimen made of stainless steel SUS316L is inspected in this study. The
electromagnetic characteristics of the material include a conductivity of = 1.4 MS/m
and a relative permeability of r = 1. The thickness of the specimen is 25 mm. An
electro-discharge machined (EDM) notch of rectangular shape with a length of

L. Janousek et al. / Advanced Probe with Array of Pick-Up Coils for Improved Crack Evaluation

273

lc = 40 mm, a width of wc = 0.5 mm and variable depth models the crack. The EDM
notch of a variable depth dc ranging from 0 to 25 mm with a step of 1 mm is
numerically inspected with the probe. The probe scans right over the crack along its
length; the winding of the exciting coil is perpendicular to the cracks length and the
pick-up coils sense the crack signals just over the crack along its length. The lift-off of
the probe is 1 mm. Frequency of 50 kHz is adopted in the inspection.
A three dimensional finite element (FEM) and boundary element (BEM) hybrid
method based upon A-V formulation is used for the numerical analysis. The governing
equations of the A-V formulation for the low frequency eddy-current problems are as
follows:

2 A =
V ,

(2)

A

V = 0
t

(3)

in the conductor region and

2 A = J 0

(4)

in the air region outside the conductor. A denotes the magnetic vector potential, V is the
electric scalar potential, J0 is the vector of the exciting current density, is the electric
conductivity and is the magnetic permeability.
Dependences of the crack signals amplitude and its phase on cracks depth
obtained by the two pick-up coils are shown in Fig. 2a) and 2b), respectively.
6

-35

pick-up 1
pick-up 2

-40
-45
-50

phase [degree]

amplitude [mV]

3
2

-55
-60
-65
-70
-75

-80

pick-up 1
pick-up 2

0
0

10
15
crack depth [mm]

a)

-85
20

25

10
15
crack depth [mm]

b)

Figure 2. Dependences of crack signals on crack depth for the two pick-up coils

20

25

274

L. Janousek et al. / Advanced Probe with Array of Pick-Up Coils for Improved Crack Evaluation

100

dc=10mm
dc=12mm
dc=15mm
dc=20mm

80

40

ratio
[a.u.]
ratio
[-]

phase [degree]

60

20
0
-20

5
4
3
2

-40
1

-60

simulation
experiment

-80
0

10

12

14

ratio[a.u.]
[-]
ratio

Figure 3. Dependences of the superposed crack


signal phase on the ratio of superposition for the
four cracks with depths of dc = 10, 12, 15 20 mm

10

15

20

25

crack depth [mm]

Figure 4. Dependences of the ratio of superposition


on crack depth, comparison of the numerical and the
experimental results

It can be observed that the dependences are different for the two pick-up coils due
to different depth profiles of eddy current density in the vicinity of each pick-up coil.
The two signals obtained by the two pick-up coils are linearly superposed based on (1)
for each particular depth of the crack. The phase of the superposed signal is extracted
for each value of the ratio. Fig. 3 displays four dependences of the superposed signal
phase on the ratio of superposition for the crack with depths of
dc = 10, 12, 15 and 20 mm.
It can be seen that the crack signals rotate almost 180 when increasing the ratio
and their rotation depend on cracks depth. Thus, a unique feature is extracted from the
characteristics. It is a certain ratio value where the crack signal rotates in half the angle
of its overall rotation. Dependence for the extracted feature is shown in Fig. 4. As it can
be seen, the dependence provides clear indication about cracks depth. In addition, the
dependence is almost linear and thus, cracks much deeper than the standard depth of
penetration ( = 1.9 mm in this case) can also be unambiguously evaluated. The width
of crack does not influence the gained dependence; however it is affected by the length
of crack. The length of crack can be estimated in advance from certain features of the
signals and than appropriate dependence of the ratio on cracks depth for actual crack
length should be used.
Four EDM notches introduced in a 25 mm thick SUS316L plate are experimentally
inspected by the probe. The notches measure lc = 40 mm in length, wc = 0.5 mm in
width and dc = 10, 12, 15, 20 mm in depth. All the parameters of the inspection and
consecutive processing are the same as ones used in the numerical investigations. The
extracted feature values of the ratio for the four cracks are shown in Fig. 4 along with
the simulated results. Quite good correspondence between the numerical results and the
experimental ones can be observed.

L. Janousek et al. / Advanced Probe with Array of Pick-Up Coils for Improved Crack Evaluation

275

3. Conclusion
The paper proposed a new probe for enhancing sizing ability in eddy-current nondestructive testing. Configuration of the probe assures that two signals of the same
crack are obtained using different depth profiles of eddy current density. The two
signals of each crack were linearly superposed and a feature value of the ratio of
superposition was extracted from the resulting signal for a corresponding crack. It was
shown that the value provides clear indication about cracks depth. Moreover, cracks
much deeper than the standard depth of penetration can be sized using the new probe.

Acknowledgment
This work has been partially supported by a grant VEGA No. 1/2053/05 of the Slovak
Ministry of Education.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]

ASME: Boiler and pressure vessel code, section XI, Rules for in-service inspection of nuclear power
plant components, 2001.
W. Cheng et al.: Ultrasonic and eddy current testing of defects in Inconel welding metals, Proceedings
of the 12 MAGMA conference, Oita, Japan, 2003, 187-190.
N. Yusa et al.: Application of eddy current inversion technique to the sizing of defects in Inconel welds,
Nuclear Engineering and Design 235 (2005), 1469-1480.
B.A. Auld and J.C Moulder: Review of advances in quantitative eddy current nondestructive evaluation,
Journal of Nondestructive Evaluation 18 (1999), 3-36.
N. Yusa et al.: Caution when applying eddy current inversion to stress corrosion cracking, Nuclear
Engineering and Design 236 (2006), 211-221.
L. Janousek et al.: Excitation with phase shifted fields enhancing evaluation of deep cracks in eddycurrent testing, NDT&E International 38 (2005), 508-515.
L. Janousek et al.: Utilization of two-directional AC current distribution for enhancing sizing ability of
electromagnetic nondestructive testing methods, NDT&E International 39 (2006), 542-546.

276

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-276

Observation of Stress Loaded


Ferromagnetic Samples Using Remanent
Flux Leakage Method
a

Tomasz CHADY a,1, Grzegorz PSUJ a and Ryszard SIKORA a


Department of Electrical and Computer Engineering, Szczecin University of
Technology,
ul. Sikorskiego 37, 70-313 Szczecin, Poland

Abstract. In this paper the results of remanent flux leakage inspections of the
stress loaded ferromagnetic samples are presented. Two different magnetizing
methods and several configurations of the transducers measuring unit are used in
order to compare and chose the optimal one and to enhance the performance of the
whole system. Absolute and differential GMR sensors are used as the pick-up
elements. All measurements were done using specimens made of a low carbon
steel SS400, tensile deformed in the longitudinal direction.
Key Words. Remanent flux leakage method, GMR elements, magnetic materials

Introduction
Remanent flux leakage method is frequently used to evaluate structure of magnetic
materials [1]. It is based on a detection of leakage fields, caused by changes of the
reluctance of ferromagnetic materials, which were magnetized in a DC field. The
excitation unit can be either made of a permanent magnet or a coil driven by a DC
current. The leakage magnetic fields are measured by scanning the surface of the
specimen with a magnetic field sensor. The collected information can be used to detect
and evaluate defects in stress loaded materials. The usage of the MFL technique for the
defects detection and evaluation of stress degradation has already been researched [2].
The results of the evaluation of the stress degradation stage using MFL technique were
presented in [3]. During the experiments the objects were magnetized in a uniform DC
field before the measurement and the leakage field was measured by absolute GMR
element. In many practical applications this type of the measuring method is not an
easy task due to a complicated geometrical shape of real test objects. The alternative
solution in this case can be use of an integrated local magnetizing coil. One of the
purposes of this paper is to broaden the scope of the tests and to compare the results
obtained using two different magnetizing methods (using the uniform DC field and the
local magnetizing coil).
In the literature, mostly little notice has been given to the configuration of the
magnetic field sensing device. In most cases a single element such as a Hall, a flux set
1
Corresponding Author: Szczecin University of Technology, Department of Electrical and Computer
Engineering, ul. Sikorskiego 37, 70-313 Szczecin, Poland; E-mail: tchady@ps.pl.

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

277

Figure 1. A view of a transducer with integrated excitation unit: 1 pick-up section; 2 rod ferrite core; 3
excitation coil; 4 sample

or a GMR is used as the field sensor. In such cases a background signal can have an
essential influence on the measurement results. An absolute sensor is prone to noises
while gradiometer provides low amplitude in case of signals with slow changes. In this
paper several differential and absolute configurations of Giant Magnetoresistive
(GMR) magnetic field sensors will be compared in order to minimize the impact of the
background signals and to maximize the effectiveness of the measurements.

Concepts of the Measuring Transducers


Several configuration of transducers were investigated in order to find the optimal
construction of the measuring unit and enhance performance of the system. Two
methods of measurements were considered. In the first method tested samples were
magnetized by a uniform DC field before measurements. Next, the specimens surface
was scanned with the pick-up transducer. In order to overcome problems with uniform
magnetization of a test sample we propose to integrate a magnetization section with the
pick-up element (Figure 1). Such construction allows to magnetize only the area, where
the measurement will be taken in the next step. The magnetizing unit consists of a coil
wounded on a rod ferrite core. The whole magnetization section is placed in a front of
the pick-up transducer. The dimensions of the magnetizing coil and its distance from
the pick-up element was optimized to achieve a high sensitivity to defects and a low
direct influence of the magnetizing unit. An absolute (NVE AAH001-02) and a
differential (NVE ABH000-01) GMR elements were used as the pick-up transducers.
The AAH-series sensor is made of high sensitivity GMR elements and is very suitable
to measure low magnetic fields. The ABH-series sensor is extremely sensitive to the

Figure 2. Block diagram of the measurement system

278

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

Figure 3. The dimensions of the transducers: a) P_ABS and P_GRAD, b) P_DIFF1, c) P_DIFF2

gradient of magnetic field. More information about the GMR elements used in the
transducers can be found in [4]. In order to select the optimal construction four
transducers with different configuration of the pick-up sections were evaluated
(Figure 2): P_ABS transducer consisting a single absolute GMR, P_GRAD
transducer with a gradiometer GMR, P_DIFF1 transducer consisting of two
differentially connected absolute GMR placed one by another and P_DIFF2
transducer consisting of two differentially connected absolute GMR placed one over
another. The dimensions of the magnetizing unit and its distance from the pick-up
element, as well as the distance between two GMR elements in transducers P_DIFF1
and P_DIFF2 are presented in Figure 3. The dimensions of the GMR elements can be
found in [4].

Description of the Experiments


A. Preliminary Evaluation of Transducer Performance
To verify a spatial resolution and sensitivity of all described transducers, two
preliminary experiments were carried out (Figure 4.). In the first one a test sample
consisting of two thin (thickness less than 0.1 mm) metal strips placed within a distance
of 1 and 2 mm from each other (Figure 4a) was used. During the measurements
transducers were moved crosswise the strips in steps of 0.1 mm. The lift-off distance
was set to 1 mm. Measured signals are plotted in Figure 5. The results give the
opportunity to compare the performance of the transducers. One can observe that the
P_GRAD transducer (Figure 5d) has greater spatial resolution than the P_ABS one
(Figure 5a). The highest signal to noise ratio was achieved for the transducer P_DIFF2
(Figure 5c).
Because of the problem with magnetizing of two metal strips to equal degree a
second preliminary experiment was introduced. During this experiment a magnetic

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

279

Figure 4. Visualization of the first (a) and the second (b) preliminary experiment

Figure 5. Results of the preliminary tests with samples consisting of two metal strips with distance between
each other equal to 1 mm (continuous line -) and 2 mm (dashed line --): a) P_ABS transducer; b) P_DIFF1
transducer; c) P_DIFF2 transducer; d) P_GRAD transducer

Figure 6. Results of the preliminary tests with sample consisting of copper wire carrying a DC current of 500
mA: P_ABS transducer (continuous line ); P_DIFF1 transducer (dashed line ); P_DIFF2 transducer
(dotted line ); P_GRAD transducer (dash-dotted line )

280

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

Figure 7. Results of the preliminary tests with sample consisting of two copper wires carrying a DC current
of 500 mA with distance from each other equal 1 mm (continuous line -) and 2 mm (dashed line --) and 5 mm
(dotted line ): a) P_ABS transducer; b) P_DIFF1 transducer; c) P_DIFF2 transducer; d) P_GRAD transducer

field generated by long copper wires (diameter 0.18 mm) carrying a DC current of 500
mA was measured. The measurements were done in steps of 0.1 mm for single wire
and two wires placed within the distance of 1, 2 and 5 mm from each other. The lift-off
distance was set at 0.5 mm. Analyzing obtained results (Figure 6 and Figure 7) one can
come to similar conclusions as in the case of the first preliminary experiment. The
greatest sensitivity was achieved for the P_ABS transducer. Using P_GRAD transducer
it is possible to distinguish two wires placed in the distance of 1 mm between each
other. The P_DIFF2 transducer generates nonzero signal only in the case of magnetic
inhomogenity, which results in optimal use of A/D converters dynamic range.
Comparing P_DIFF1 with P_DIFF2 transducer one can see that the response of the
P_DIFF1 is more complicated than the P_DIFF2.
B. Results of the Measurements
Final measurements were done using seven planar specimens made of the SS400 low
carbon steel, tensile deformed in the longitudinal direction. The samples SS400-1,
SS400-2 and SS400-3 are loaded with the maximum stress of 103.2 MPa, 206.4 MPa
and 300 MPa respectively, which is lower than the yield stress (350 MPa). The sample
SS400-4 was tensile deformed exactly up to the yield point. More details about the test
samples and the measuring system can be found in [3] and [5]. In case of the first
measuring method the leakage flux from samples was observed after magnetization in
uniform field. In case of the transducer with the integrated magnetization section the
excitation coil was driven by 1 A DC current. The lift-off distance was about 0.5 mm
from the sample surface independently of the surface roughness. The sensitivity axis of
the pick-up element was parallel to the loading direction of the sample. The
measurements were done along the x-axis (the longitudinal direction) and the y-axis in
steps of 1 mm. The scanning area was 120 mm x 27 mm.

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

281

Figure 8. Results of measurements achieved using P_GRAD transducer for sample: a) SS400-3; b) SS400-4;
c) SS400-5; column A uniform magnetization of a test sample using a solenoid; column B local
magnetization utilizing coil driven by a DC current.

Figure 9. Results of measurements obtained for sample loaded below materials yield point: a) P_ABS
transducer; b) P_DIFF1 transducer; c) P_DIFF2 transducer; d) P_GRAD transducer; column A uniform
magnetization of a test sample using a solenoid; column B local magnetization utilizing coil driven by a DC
current.

First P_GRAD transducer was used in order to evaluate the stage of degradation of
tested samples (Figure 8). Samples SS400-4 and SS400-5 were loaded with a
maximum stress value respectively equal to and grater than the yield stress, which
resulted in occurrence of Lder bands almost in all scanned area. In order to investigate
an early stage of defects forming sample SS400-3 was selected for further
examination.
The results obtained for the sample SS400-3 are shown in Figure 9. The P_GRAD
presents an extreme high spatial resolution, but the sensitivity is low. Especially low
sensitivity was observed in the case of the magnetization section integrated with the

282

T. Chady et al. / Observation of Stress Loaded Ferromagnetic Samples

transducer. Moreover the interpretation of differential signal is complicated. The


highest sensitivity to the defect was observed in case of the P_ABS transducer,
however it measures also unwanted background signal. Connecting two absolute GMR
elements in the differential way it is possible to achieve higher sensitivity than for the
P_GRAD and a higher spatial resolution than for the P_ABS. The P_DIFF2 transducer
presents also a low dependence on an external interfering magnetic field. Using local
magnetizing coil results in obtaining lower sensitivity to defects, however the problem
with uniform magnetization of a tested object is avoided. Regarding to the
interpretation of measured signals, the best results were obtained for the P_DIFF2
transducer.
Conclusions
According to achieved results (Figure 9), one can observe that the proposed differential
transducers help to enhance the flux variations caused by defects. Additionally
P_DIFF2 allows for simple interpretation of the signals and has the ability to reduce
influences of the external fields.
The transducer with integrated magnetic sections can be more easily applied in
practical cases. It presents the opportunity to magnetize each measuring part of the test
object in the same way. The difference between both magnetizing methods can be seen
especially in Figure 9a-c. When the sample is magnetized prior the measurements (for
example in solenoid), disturbances of the signal can be observed not only in the
defects area, but in its direct neighborhood too (column A in Figure 9a-c). The effect
is significantly smaller in case of the local magnetizing coil (column B in Figure 9a-c).
In the future works authors would like to propose integration of a demagnetizing
section with the transducer in order to overcome a problem with unknown magnetic
history of the specimen.

Acknowledgements
This work was supported in part by the State Committee for Scientific Research,
Poland, under the Grant no: 3T10A 017 30 (2006-2009).

References
[1]
[2]
[3]
[4]
[5]

Y. Tsuchida, et al, Evaluation of Strain Distribution of Austenitic Stainless Steels by Measuring


Remanent Magnetization, Electromagnetic Nondestructive Evaluation, IOS Press, 2005, pp 151-158
V. Babbar, L. Clapham, Residual Magnetic Flux Leakage: A Possible Tool for Studying Pipeline
Defects, Journal of Nondestructive Evaluation, Vol.22, No. 4, 2003, pp.118-125
T. Chady, Evaluation of Stress Loaded Steel Samples Using GMR Magnetic Field Sensor, IEEE
Sensors Journal, Vol. 2, No. 5, October 2002, pp. 488-493.
GMR Sensor Application Notes, Nonvolatile Electronics, INC. [Online], Available:
http://wwww.nve.com
T.Chady, et. al, Computerized System for Complex Electromagnetic Nondestructive Evaluation, XIII
International Symp. on Theoretical Electrical Engineering, ISTET05, Lwow, Ukrain, pp. 333-336

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-283

283

Evaluation of Complex Multifrequency


Eddy Current Transducer Designed For
Precise Flaw Depth Measurements
Tomasz CHADY, Piotr BANIUKIEWICZ, Ryszard SIKORA, Grzegorz PSUJ
Szczecin University of Technology, al Piastow 17, 70-310 Szczecin, Poland

Abstract. In this paper the authors propose an eddy current transducer dedicated to
the precise flaw depth measurements. The transducer consists of two probes with
C-shaped ferrite cores. The frequencies of the excitation currents differ for both
probes. A flux generated by the big probe penetrates the material deeper than a
flux from the small probe. Therefore, the bigger probe is sensitive to deep defects
while the small probe is affected more by surface defects in the material. The
results of measurements are presented.
Keywords. Eddy current method, nondestructive testing

Introduction
The rapid industry growth that has been done recently encourages new demands for
safety of people and natural environment. Most of the cracks that can be observed in
metals are close to notch. An eddy current (EC) method is one of the most popular
electric NDT methods due to its simple hardware implementation, rapid scanning and
contact-less inspection. The method is sensitive to various types of defects, mainly the
surface and subsurface discontinuities with small dimensions and different shapes. This
is crucial because all critical tensions, caused by fatigue of material, are concentrated
on the surface of material. In the EC system, the probability of crack detection is
closely related to the construction of the probe. Various probes have been proposed in
literature [1]. Most of them offer high sensitivity that enables them to detect cracks
located deep under the surface of the specimen. However, in particular applications it
is necessary not only to detect the flaw but also to measure its depth precisely. In the
case of deep flaws in thick material, most of widely used eddy current transducers are
not capable to fulfill such kind of requirements [2]. Therefore, works on the new
transducer started in 2003 [3]. The developed transducer achieves a very good depth
resolution. It is dedicated to precise measuring the depth of the notches that may appear
in thick materials.

1. Description of the Sensor and the Measuring System


The proposed transducer is differential one and it is different from the other probe
described in the literature [4]. The transducer, simplified view of which is shown in the
Figure 1, is composed of two probes: a small probe (SP) and a big probe (BP).

284

T. Chady et al. / Evaluation of Complex Multifrequency Eddy Current Transducer

Figure 1. Simplified view of the eddy current differential transducer and the measuring system

The probes consist of the same elements and differ only in size. Both probes
contain a c-shaped ferrite core. The distance between columns of a big core (FCBP) is
20 mm and in case of a small probe (FCSP) is 10 mm. Each probe consists of two
exciting coils wound around the ferrite core. The coils are connected in series and
supplied with the alternating current. The output signals are taken from two measuring
coils wound around neighboring columns of the cores. The transducer is embedded in a
plastic case and submerged in epoxy resin to prevent any mechanical damages. A flux
generated by the big probe penetrates the material deeper than a flux from the small
probe. Therefore, the bigger probe is sensitive to deep defects while the small probe is
affected more by surface defects in the material. Selection of the excitation signals
parameters, which allow us to maximize an accuracy of depth estimation, is very
complicated. Distribution of the eddy currents in the specimen depends on the
excitation currents ratio as well as on their frequencies. In this paper, the influence of
the frequency of the excitation signals is analyzed. The big probe is under stronger
influence of deep flaws than the small probe. The main task of the SP is to compensate
the contribution that is brought in the output signal by the surface cracks. In this way,
the transducer is affected more by the deeper located part of the crack, what results in a
better depth system resolution. It is also possible to calibrate the transducer, to reach an
equilibrium state, by adjusting the excitation current in the SP. The state of equilibrium
is accomplished when the amplitude of output differential signal is close to zero for an
unflawed specimen. The correct choice of excitation parameters is crucial. The most
important parameters are frequency and amplitude of excitation currents. Too high
amplitude of current causes that the output signal from the small probe is much greater
than the big probe signal. Consequently, the small probe becomes dominating what, in
result, rapidly decreases the depth resolution of the transducer.
The transducer is powered by two function generators through the power
amplifiers. The generators are coupled and triggered simultaneously from the computer.
Thus, the phase shift between the excitations of the probes is constant and it can be
easily adjusted. The output signals obtained from the search coils are amplified by
measurement amplifiers and then filtered by band-pass filters to remove unwanted
components and noise. It is done this way that, the frequency component descended
from the BP excitation has to be removed from the SP output and vice versa. This
guarantees that the RMS values are computed only from frequency components

T. Chady et al. / Evaluation of Complex Multifrequency Eddy Current Transducer

285

contained in the excitation signals for both probes separately. Next, both signals are
converted to DC value using two true RMS/DC converters. An integrated circuit
AD637 has been used as the RMS/DC converter. A single pole Sallen-key filter
configuration has been chosen to measure low frequency signals, what gives us a good
compromise between the conversion errors and setting time. Then, the RMS values of
measured signals are subtracted using differential input operational amplifier. Finally,
the output signal from the transducer is acquired by a high resolution A/D converter.
As one can see, all electronic circuits used in the system are available in the form of
integrated circuits. A high speed A/D converter is not required in this application due to
a DC output signal from the system. Consequently, the whole measuring system can be
realized as a small mobile system.

2. Measurement Results
In order to verify usefulness of the proposed transducer, a set of experiments was
carried out. The test specimens used for the experiments are 20 mm thick aluminum
plates, which have flaws in the form of EDM notches. The flaw depth is from 7 mm to
18 mm (Figure 2). The transducer was moved over the tested specimen along a straight
line parallel to the longer edge of the plate in steps of 1 mm. The lift-off was measured
to be 0.5 mm. The performance of the transducer was evaluated using coefficient
defined by:

U MAX18  U MAX16
U MAX18

100%

(1)

where UMAX16 and UMAX18 stand for the maximum of relative signal obtained for inner
flaws with the depth of 16 mm and 18 mm respectively. At the beginning of each
measurement the probe was moved towards the unflawed area of the test sample. The
amplitude of the excitation current in the small probe was adjusted in order to achieve
the state of equilibrium. Such adjustments have been done for each excitation
frequency of SP. The parameters of the excitation signal for the big probe were not
changed during the measurements. The excitation frequency of BP was 80 Hz, whereas
the current amplitude was set to 100 mA. Next, the transducer was moved over the
specimen with step of 1mm in order to obtain the signals for two flaws 16 mm and 18
mm deep.

Figure 2. View of the test specimen made of aluminum

286

T. Chady et al. / Evaluation of Complex Multifrequency Eddy Current Transducer


f =160Hz

f =240Hz

SP

40
0

U /U

10
U /U

U /U

16mm
18mm

5
0

20

40 60
x [mm]

80

100

20

f =480Hz

20

100

20

40 60
x [mm]

80

100

20

40 60
x [mm]

80

100

20

80

100

f =2000Hz
SP

U /U

U /U0

40 60
x [mm]

20

50

100

SP

100

80

f =1600Hz

80

100

10

10

50

40 60
x [mm]

80

SP

SP

40 60
x [mm]

20

f =1200Hz

20

20

f =960Hz

20

20

50

30

U /U

U /U0

U /U

10

SP

U /U

80

40

20

50

40 60
x [mm]

20

f =640Hz

SP

30

10

SP

15

f =320Hz

SP

10

20

40 60
x [mm]

80

100

0
20
40

20

40 60
x [mm]

Figure 3. Relative signals obtained for flaws 16 mm and 18 mm deep and various excitation frequencies of
small probe

Maximizing the coefficient k, the optimal parameters of the SP excitation were


obtained. For those parameters the transducer achieves the best depth resolution what
means that two inner flaws of similar depths are easily distinguishable. Figure 3 shows
selected signals obtained for flaws of depth 16 mm and 18 mm for various excitation
frequencies. A strong dependence between the excitation frequency of the probe SP
and the depth resolution achieved by the transducer can be observed in the Figure 4.
The best results were obtained when the SP was excited with the current frequency
640 Hz. Negative peaks, that occur for higher frequencies, result from the growth of the
sensitivity of the SP together with its excitation frequency.

Figure 4. The depth resolution of the transducer versus excitation frequency

287

Umax/U0

T. Chady et al. / Evaluation of Complex Multifrequency Eddy Current Transducer

D [mm]

Figure 5. Maxima of the relative signals obtained for chosen fSP versus crack depth

Figure 5 confirms that the highest depth resolution of the transducer is obtained for
the optimal frequency 640 Hz. The nearly linear dependence between maximum of the
signal and crack depth occurs for lower frequencies.

Conclusions
The proposed transducer achieves very good depth resolution. The experiments
performed on thick material show that two inner flaws of similar depths are fully
distinguishable. The performance of the transducer is closely related with excitation
frequency of the small probe. The best resolution, in the case of flaws 16 mm and 18
mm deep, has been achieved for frequency 640 Hz. Due to a strong influence of
excitation current of the SP on the output signal, the state of equilibrium can be set very
precisely. It is also possible to use the transducer as a depth discriminator. This can be
accomplished by equilibrating the transducer for flaw of specified depth.

Acknowledgments
This study was supported by State Committee for Scientific Research, Poland, Grant no.
3T10A01730 (2006-2009).

References
[1]
[2]
[3]
[4]

T. Chady, M. Enokizono, T. Todaka, Y. Tsuchida, R. Sikora, A Family of Matrix Type Sensors for
Detection of Slight Flaws in Conducting Plates, IEEE Transactions on Magnetics, 35(5), 3655-3657,
1999.
F. Thollon et al., Numerical and Experimental Study of Eddy Current Probes in NDT of Structures with
Deep Flaws, NDT&E International, 28(2), 97-102, 1995.
T.Chady: private email correspondence 2003-07-10, 2003-08-07.
L.Janousek et al., Excitation with phase shifted fields-enhancing evaluation of deep cracks in eddycurrent testing, NDT&E International, Vol.38, 508-515, 2005.

288

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.
doi:10.3233/978-1-58603-896-0-288

Comparative Study of Coil Arrangements


for the EC Testing of Small Surface
Breaking Defects
Cyril RAVAT, a,b,1 , Yann LE BIHAN a , Pierre-Yves JOUBERT b and
Claude MARCHAND a
a

LGEP/SPEE Labs; CNRS UMR8507; SUPELEC; Univ Pierre et Marie Curie-P6;


Univ Paris Sud-P11; 91192 GIF-SUR-YVETTE FRANCE
b
SATIE, ENS Cachan, CNRS, UniverSud, 61 av. du President Wilson, 94235 CACHAN
Cedex FRANCE
Abstract. The increasing need of both reliability and speed during inspection operations requires to develop new inspection devices, such as EC multisensor arrays.
In this paper, the authors consider the detection of small surface breaking defects
ranging from 0.1x0.1x0.1 mm3 to 0.8x0.1x0.4 mm3 . A low-cost multilayer PCB
multicoil array was implemented using different transmit/receive strategies and the
detection performances were exhaustively quantied and compared, thanks to the
computation of Receiver Operating Characteristics. Two strategies allow the smallest defects to be detected without false alarm.
Keywords. Eddy current, non destructive testing, multicoil array, receiver operating
characteristic

Introduction
Eddy current (EC) sensors are widely used for non-destructive evaluation on electrically
conducting materials, since they are sensitive to defects such as fatigue cracks, inclusion
or corrosion, and easy to implement in industrial applications.
Moreover, the use of multicoil arrays permit to increase the rapidity and the reliability of the detection and is particularly well suited to the detection of surface breaking
defects. Nevertheless, the strategy of operations of the sensor has to be optimized, in
order to maximize the sensitivity, regardless of the operating conditions and the defect
orientation [1].
In this paper, an exhaustive and qualitative study of a 3-coil 1-D array, viewed as
an elementary structure of a larger array, is proposed. After the denition of the chosen measurement strategies and the short description of the implemented experimental
set-up, a quantitative comparison of the detection performances is carried out thanks to
Receiver Operating Characteristic (ROC) curves.
1 Corresponding

Author: cyril.ravat@lgep.supelec.fr

C. Ravat et al. / Comparative Study of Coil Arrangements for the EC Testing

289

1. EC Measurement Strategies
The basic principle of the EC method is to induce an EC ow within the inspected material (transmission), and to measure the resulting magnetic eld at the surface of the
sample (reception), which is relative to the integrity of the material [2]. Actually, the
EC signal is the ratio between the electromotive force generated at the terminals of the
receiver (R) device, and the excitation current feeding the transmitter (T) device.
Since coils can be used either as transmitter, receiver or transmitter-receiver (T&R),
the following transmission-reception strategies can be dened for a 3-coil 1-D array, as
depicted in Figure 1:
T/R: one T&R coil. In this case, the EC signal is actually the coil impedance.
This strategy is widely used since it is the simplest to implement, despite a low
sensitivity.
TR: one T coil and one adjacent R coil. This strategy is the basic structure of
separate function EC measurements. The measurement is absolute, and thus quite
sensitive to lift-off and tilt.
RTR: one middle T coil and two R side coils. Voltages at each R coil terminals
are subtracted: the measurement is differential.
TRT+: one middle R coil and two T side coils connected so that the measured
magnetic uxes are added in the R coil. The measured signal is the superposition
of two shifted TR ones.
TRT-: same as TRT+ but measured magnetic uxes are subtracted in the R coil.
This strategy can be considered as differential (the EC signal should be null in
absence of defect), though not the reception but the transmission is differential.
T/R

TR

TR

RTR

TRT+
T

TRT
R

Figure 1. The ve T/R strategies

These ve 3-coil 1-D strategies constitute all the possible relevant strategies using
three adjacent coils in line and can be considered as elementary structures for the design
of larger 1-D or 2-D coil arrays.

2. Experimental Set-up
The three coils used in this study are 8-layer at square coils, realized thanks to conventional PCB technology (Figure 2). Each coil features a 3x3 mm2 surface and a 1 mm
thickness. The inductance and the resistance of each coil are about 2 H and 3.5 , respectively.

290

C. Ravat et al. / Comparative Study of Coil Arrangements for the EC Testing

The sensor array was implemented for the exhaustive inspection of a nickel based alloy mock-up featuring a magnetic permeability = 4107 H.m1 and an electrical conductivity = 0.76 MS.m1 . 30 surface rectilinear defects were machined in the mockup. Their dimensions are 0.1, 0.2, 0.4, 0.6 or 0.8 mm in length, 0.1 mm in width and 0.1,
0.2 or 0.4 mm in depth. Since the signal amplitude and shape highly depend on the defect orientation, all the defects are available for two directions, parallel to the array main
orientation (horizontal defect) and perpendicular (vertical defect).
The measurements were carried out thanks to a HP4192A impedance analyzer and
a PC-controlled 3-axis robot modifying the position of the sensor. The computer which
controls synchronously both devices also acquires the data. In this study, the experimental set-up was adjusted so that the sensor array operates in the best possible conditions,
i.e. tilt and lift-off noises were controlled and kept negligible. As the transmitter current frequency modies the EC intensity and the penetration depth, it highly affects the
detection performances. The used EC frequencies range from 500 kHz to 6 MHz.
Figure 3 shows EC images obtained at 3 MHz for the 5 T-R strategies, in the case of
three different defects (largest horizontal and vertical defects, and smallest defect). Spatial sampling step is 0.2 mm in both directions. All strategies allow the largest defects to
be visualized with a sufcient signal to noise ratio, independently from their orientation.
However, the smallest defect is hardly seen except for the RTR and TRT- strategies.

Figure 2. Eight 8-layer 3x3 mm2 at coils in line (only 3 adjacent coils are actually used)

3. Receiver Operating Characteristic


The detection performances of the multicoil array depend on the used transmissionreception strategy and EC frequency. In order to quantify and compare these performances, a detection algorithm was built and used in ROC curve representations.
The detection algorithm determines whether, for a given threshold, an EC image
contains a defect response or not. The defect response is basically dened as a set of
adjacent pixels featuring an EC amplitude higher than the threshold. The detection matrix
for the i-th defect reads
Di = (EC_Imagei > t) S
where designates the convolution operator, t is the threshold and S is a square matrix
of ones. EC_Imagei is the matrix of the EC amplitude for the i-th defect. The dimension
of S was experimentally adjusted in order to optimize the detection performances, and
was xed to 10 pixels (2 mm). The detection diagnostic denoted (i, t) is then expressed
by

C. Ravat et al. / Comparative Study of Coil Arrangements for the EC Testing

291

Figure 3. EC images (EC signal signed modulus) obtained at 3 MHz with the largest horizontal defect (0.8x0.1x0.4 mm3 , rst line), the largest vertical defect (second line) and the smallest defect
(0.1x0.1x0.1 mm3 , third line); from left to right: T/R, TR, RTR, TRT+, TRT-; scales are in mm.

!
(i, t) =

1
0

if (x, y) : Di (x, y) = 1
else

i varies from 1 to 30 and t ranges from tmin to tmax dened by




tmin = min min (EC_Imagei (x, y))
i
 x,y

tmax = max max (EC_Imagei (x, y))
i

x,y

For each strategy and frequency, the EC images of the 30 defects are used to compute a 30-level discretized probability of detection (POD), depending on the threshold t,
expressed by
POD(t) =

30
1 
(i, t)
30
i=1

Moreover, 30 EC images of defect-free areas are used to compute a 30-level discretized


probability of false alarm (PFA), for the same threshold values. The ROC curve is the
parametric curve which plots POD versus PFA for the different threshold values [3].
ROC curves were compared using the minimal distance from the curve to the maximum efciency point [4], which stands for no false alarm (PFA = 0) and all defects correctly detected (POD = 1).
Figure 4 presents the best curves obtained for each strategy, considering all defects (a), only horizontal ones (b) and only vertical ones (c). The TRT- and RTR strate-

292

C. Ravat et al. / Comparative Study of Coil Arrangements for the EC Testing

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

T/R (500kHz)
TR (1MHz)
RTR (4MHz)
TRT+ (1MHz)
TRT (4MHz)
0.2 0.4 0.6 0.8
Probability of false alarm

Probability of detection

Probability of detection

Probability of detection

gies are the most efcient and allow surface defects as small as 0.1x0.1x0.1 mm3 to be
detected. Indeed, if a ROC curve passes by the maximum efciency point, then at least
one threshold allows all defects to be detected without false alarm. Figure 4 also shows
that horizontal defects are far better detected than vertical ones, since a defect which is
parallel to the array main orientation perturbs more the EC ow.

T/R (500kHz)
TR (1MHz)
RTR (4MHz)
TRT+ (1MHz)
TRT (4MHz)
0.2 0.4 0.6 0.8
Probability of false alarm

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

T/R (500kHz)
TR (1MHz)
RTR (4MHz)
TRT+ (1MHz)
TRT (4MHz)
0.2 0.4 0.6 0.8
Probability of false alarm

Figure 4. Best ROC curves for each strategy, considering (a) all defects, (b) only horizontal defects, (c) only
vertical defects

In order to discriminate equidistant curves, the ratio of the largest threshold allowing all the defects to be detected to the lowest threshold allowing no false alarm to
be triggered, was also calculated. In the case of the two perfect curves for RTR and
TRT- strategies, this represents the space of choice of the correct threshold. The magnitude of this quantity quanties the separability between detections and false alarms. The
TRT- strategy has a better separability than the RTR strategy, and thus the TRT- strategy
is globally the most efcient of the ve implemented strategies.

Conclusion
In this paper, an elementary array of 3 coils in line is studied and 5 transmit/receive strategies were carried out for the detection of small surface breaking defects. Two strategies

C. Ravat et al. / Comparative Study of Coil Arrangements for the EC Testing

293

permit defects as small as 0.1x0.1x0.1 mm3 to be detected without any false alarm, and
the TRT- strategy is the most efcient. The inuence of the tilt noise was not presented in
this study, however same results were obtained for a lift-off up to 0.5 mm. The obtained
results are very promising and further work will focus on 2-D multicoil array using the
TRT- strategy, implemented in different orientations in order to maximize the sensitivity
for defects of any orientation.

References
[1]

[2]
[3]
[4]

P.-Y. Joubert and Y. Le Bihan, Eddy Current data fusion for the enhancement of defect detection in
complex metallic structures, International Journal of Applied Electromagnetics and Mechanics, 19 (2004),
647651.
H.L. Libby, Introduction to electromagnetic non-destructive test methods, Roberty Krieger Publishing
company, New York, 1979.
J.P. Egan, Signal detection theory and ROC analysis, Series in cognition and perception, New York,
Academic press, 1975.
D. Horn and W.R. Mayo, NDE reliability gains from combining eddy-current and ultrasonic testing,
NDT&E International, 33 (2000), 351362.

This page intentionally left blank

Electromagnetic Nondestructive Evaluation (XI)


A. Tamburrino et al. (Eds.)
IOS Press, 2008
2008 The authors and IOS Press. All rights reserved.

295

Author Index
Abbasi, K.
Abe, T.
Aldrin, J.C.
Altpeter, I.
Andreescu, A.
Ara, K.
Baniukiewicz, P.
Bennett, W.
Bowler, J.R.
Bowler, N.
Bruma, A.
Capova, K.
Cardelli, E.
Cawley, P.
Cazacu, M.M.
Chady, T.
Chan, S.C.
Chen, Z.
Choi, D.-M.
Choua, Y.
Davis, C.L.
Dixon, S.
Dobmann, G.
Dominguez, N.
Duchne, B.
Edwards, R.S.
Faba, A.
Faktorov, D.
Formisano, A.
Gotoh, Y.
Grimberg, R.
Hao, X.J.
Hashizume, H.
Hopkins, P.
Horikawa, N.
Hbschen, G.
Iftimie, N.
Ike, H.
Ito, S.
Janousek, L.
Jayakumar, T.
Joubert, P.-Y.
Jung, H.-S.

154
62
133
18, 54
257
37, 42
283
90, 98
203
203
249
271
195
141
257
276, 283
241
171
231
217
86
78
18, 54
217
225
78
195
162
195
26
241, 249, 257
86
154
90, 98
62
54
257
62
154
271
70
117, 125, 288
231

Kamada, Y.
Kern, R.
Kikuchi, H.
Knopp, J.S.
Kobayashi, S.
Kopp, M.
Koshika, T.
Lambert, M.
Le Bihan, Y.
Le Diraison, Y.
Lesselier, D.
Liu, T.
Marchand, C.
Martone, R.
Mason, J.S.D.
Matsukawa, J.
Miya, K.
Morabito, F.C.
Morozov, M.
Morris, P.F.
Nagy, P.B.
Paillard, S.
Papais, M.
Pearson, N.
Peyton, A.J.
Pichenot, G.
Pirani, A.
Psuj, G.
Rabung, M.
Raj, B.
Rajkumar, K.V.
Rao, B.P.C.
Ravat, C.
Ricci, M.
Romero Ramirez, A.
Rubinacci, G.
Sasi, B.
Sato, T.
Savin, A.
Scruby, C.B.
Shaw, B.A.
Shin, Y.-K.
Sikora, R.

37, 42
18
37, 42
133
37, 42, 46
54
42
217, 225
217, 288
117, 125
225
42
288
195
148
62
3, 171, 271
195
179, 187, 263
86
141
217
195
148
86
211, 217, 225
109
276, 283
54
70
70
70
288
109
148
179, 187, 263
70
62
241, 249, 257
10
90, 98
231
276, 283

296

Skarlatos, A.
Specogna, R.
Sposito, G.
Steigmann, R.
Strangwood, M.
Takagawa, T.
Takagi, T.
Takahashi, N.
Takahashi, S.
Tamburrino, A.
Tassin, A.
Theodoulidis, T.
Tian, G.Y.
Tokuma, H.

225
195
141
249, 257
86
62
62
26
37, 42, 46
109, 179, 187,
195, 263
125
211
78
171

Trevisan, F.
195
Uchimoto, T.
62
Udpa, L.
241, 249
Udpa, S.S.
241, 249
Vaidhianathasamy, M.
90, 98
Vaidyanathan, S.
70
Ventre, S.
109, 179, 187, 195, 263
Versaci, M.
195
Villone, F.
179
Voillaume, H.
217
Wilson, J.
78
Wolter, B.
18
Yin, W.
86
Yusa, N.
171, 271

This page intentionally left blank

This page intentionally left blank