Professional Documents
Culture Documents
EVALUATION (XI)
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.
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
LEGAL NOTICE
The publisher is not responsible for the use which might be made of the following information.
PRINTED IN THE NETHERLANDS
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.
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
viii
M. Versaci
F. Villone
J. Wilson
L. Xin
N. Yusa
Z. Zeng
ix
Organizing Committees
International Committee
Chairman
Members
Organizing Committee
Local Committee
Phil Anderson
Jeremy Hall
Eugene Melikhov
John Snyder
Paul Williams
Stan Zurek
xi
List of Participants
Mr. Kavoos Abbasi
Tohoku University, Japan
abbasi@karma.qse.tohoku.ac.jp
xii
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
Prof. Li Luming
Tsinghua University, China
lilm@tsinghua.edu
xiii
xiv
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
18
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
70
78
86
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
117
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
148
154
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
203
211
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
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
276
283
288
Author Index
295
Invited Speakers
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
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.
Projection principle
Connection principle
System
s
Component
Equipments
Reflection
Reflection of
maintenance
technology on
planning
Maintenanology
humanities
maintenance
scheme= Flow
of
maintenance
activities
A-B-C-D-E-F
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
Inspection plans
Maintenance methods :
Decomposition testing :
Nondestructive testing :
Function tests :
Condition monitoring :
Corrective
Maintenance
Safety
Safety Operation
Operation
Maintenance
Engineering
Maintenance Science
Maintenance Engineering &
Technology
Maintenance Sociology
M ethodology of optimization
Codes and Standards
development
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
Casing
Impeller
Blades
5.00E-07
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
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)
0.01
U (v)
3.00E-08
-4.00E-08
Time(s)
0.01
-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)
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.
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
5.00E-03
1.00E-02
1.50E-02
2.00E-02
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
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.
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
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
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.
Technique
development
Service
suppliers
Product
development
Equipment
suppliers
11
Asset owners
and operators
Research
Product
Manufacturers
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.
12
230
Acceptance
Standardisation
Commercialisation
Product
development
100
Further research
&
development
Trials &
evaluation
Research
Pioneers
1
Technology development
121
-30
Scientific push
Industrial pull
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.
13
Proven
Method
Pre-Production
Technology
Assessment
and Proving
Development
Production
Implementation
Idea
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
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
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
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
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
Other disciplines
Figure 4. NDE research and linked engineering disciplines
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
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
19
20
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.
1600
1600
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
200
400
600
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.
lamellar graphite
GJL
transition zone
vermicular graphite
GJV
22
23
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
200
150
100
50
Adj.R = 0.941
1 = 19.0 MPa
-50
-50
50
100
150
200
250
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
Fig.11. Comparison of nitrating hardening depth measured by 3MA and Nht 700 (Vickers) versus optical
result
24
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.
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
25
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
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).
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 2H e H g .
(3)
28
22
52.5
.5
.5
16 20
19
dc-yoke
dc-exciting coil (3A)
29
33
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
b (Hmin, Bmin)
Dw
Dw=0.5
1.6
1200
B (T)
1000
1.2
800
600
0.8
400
200
0.4
0
2000
4000
6000
H (A/m)
8000
differential Pd
1200
1000
800
600
400
200
0
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
'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).
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
Depth Cd:1mm
:1mm
Bx search coil D
30
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
0.5
1
1.5
Position [mm]
Crack position
(Cw=0.01)
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
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
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
10
position z [ mm]
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
1 .5
2.5
2 2
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
15
with fin
20
10
0
-8
-6
-4
-2
10
without fin
4
6
8
position z [mm]
Bx [x10-4 T]
Bx [x10-4 T]
20
0
0
20
40
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).
Magnetic Materials
37
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
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
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
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].
40
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.
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]
42
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
2. Experimental Procedure
43
Excitation coil
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
480
32
MBN energy
rms voltage
400
28
320
24
240
20
160
16
80
44
12
0
10
20
30
40
50
45
(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
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
26
MBN Energy
rms voltage
380
(d) 40 %
14
285
DBTT (K)
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
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.
(1)
48
S. Takahashi and S. Kobayashi / A Bridge Between NDE and Charpy Impact Testing
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)
(3)
(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
(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.
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%
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
50
U0
DBTT, T (K)
280
200
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.
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
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
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
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
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.
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
55
1. Material Selection
Figure 1. Neutron fluence in a German NPP of the last generation as function of lifetime
56
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
57
58
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).
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)
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
59
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
receiver coil
poleshoes of electromagnet
60
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.
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
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)
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]
62
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
63
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
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.
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
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
65
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
/ CIPGVKEHNWZFGPUKV[
)
(% CUAECUV
(% CUAECUV
(% CUAECUV
+
+
/ CIPGVKE(KGNF
1 G
.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
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
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
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
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]
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
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
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.
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
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
74
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
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.
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
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.
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
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.
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.
80
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
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
82
(a)
(b)
83
Frequency - kHz
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
84
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.
[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
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
87
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
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
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
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
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
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
93
94
300
Gear S1
(d)
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.
30
(a)
20
25
95
Tooth-T3-L
Tooth-T1-1
15
10
5
0
-100
-50
50
100
96
200
(d)
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.
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
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.
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.
100
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.
101
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
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.
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
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.
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.
Inverse Problems
109
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
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
(1)
where J, J 0 L2div VC , J
A : vr o
P0
4S
J0
0 in VC J n
vr '
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
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
VCLOC
u N k 'KJ 0dV
N k .
(4)
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
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)
x n 'x n
(12)
(13)
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.
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
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 %
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
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]
117
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.
118
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.
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:
s21
0
SS t = .
..
0
0
s22
..
.
0
0
0
. . ..
. .
s2r
119
(2)
t1 0 0
0 t2 0
T = QT DT = QT .
.. . . ..
..
. .
.
0
0 tr
with QT QtT = Ir
(3)
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
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
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.
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-
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.
122
Table 1. Optimum separation frequencies in each plate, and corresponding penetration depth relative to a
90phase-shift of the EC density.
Plate
f2 = 1200
3.9
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
0.990
0.994
0.996
0.984
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
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.
125
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
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.
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.
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)
MM T
TSS T T T
d 2V 2
R 1 1
0
d 22V 22
RT .
(4)
128
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).
129
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
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.
kurt ( s)
E[(s s ) 4 ] 3( E[ s s 2 ]) 2 .
(8)
130
~
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)
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.
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
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
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]
133
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
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.
135
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
f T
A cosT 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.
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
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
Ar, f cosT 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
(a)
1800
(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)
(c)
15
(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,
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
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).
141
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.
142
60
5
j=-2
j=-1
j=0
j=1
j=2
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
12
11
10
8
PR0 (Baseline)
PR1 (Rectangle)
PR2 (Circular arc)
PR4 (Triangle)
6
0
10
15
20
25
30
35
40
45
50
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.
V V0
,
V
(2)
145
3.5
1 pair
3.0
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
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
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
146
4.0
1 pair
3.5
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
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).
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)
[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
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%.
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.
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.
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.
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
Pipe classifier
Conical
classifier
Lake classifier
151
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.
Type
Thickness of
plate
Pipe
6 [mm]
Conical
6 [mm]
Lake
6 [mm]
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
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.
5.
153
Conclusion
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
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
155
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
156
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.
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 ]
Calculated time
at 6 GHz
(a)
(b)
158
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 ]
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
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
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]
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.
161
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
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.
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)
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
(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
Z& = Z& 0
1 &
+ 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
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
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}
, and 3
g
4
166
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
Figure 4. Dependence of the reflected signal amplitude on the depth of the crack
(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.
167
f=10,1 GHz
500
100
0
-100 0
10
20
30
40
50
-5
-300
-500
Im{Z} []
300
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 []
168
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]
Modeling
171
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.
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
174
Z. Chen et al. / Effect of Crack Closure on Quantitative ECT Inspection of Closed Fatigue Cracks
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)
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
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)
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)
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)
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.
179
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
(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:
ZI=U
181
(5)
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
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
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].
Duration
(hours)
SCC1
SCC3
SCC4
SCC5
75
50
50
165
Crack
Length
(mm)
14
12
21
29
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.
184
(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
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.
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
References
[1]
187
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
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
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
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.
(14)
1 F m
P 0
D E M F T I
Fm
Ni
(15)
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)
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
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)
192
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.
RS-232
193
Robotic
Arm
Sliding
Probe
PC
RS-232
Phasec2D
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
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
195
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].
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)
198
E. Cardelli et al. / Design of a System for the Long Defects Detection with Advanced Methods
(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
(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 |
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 |
(8)
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.
203
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
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).
[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)
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
(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.
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
0.35
0.25
0.2
0.15
0.1
0.05
0
0.05 0
10
0.1
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.
211
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
(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.
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)
0 z l0
(2)
i=1 j=1
(ec)
(x, y, z) =
(ec)
z0
(3)
i=1 j=1
The magnetic ux density in the conductor layers (z < 0) can be written using the
(a)
(a)
sin(ui x) sin(vj y) eij z Cij + eij z Dij
(4)
i=1 j=1
(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)
(9)
(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
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.
217
Introduction
218
Ek (r) = Ek (r) j0
n
l=m
(ee)
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)
(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
I02 Z
n
l=m
(0)
219
(3)
(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.
220
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).
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.
221
(
(
Figure 5. Comparison between CIVA results and those of the potential formulation [5].
222
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 .
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-
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.
224
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
225
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
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].
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)
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
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
Phase
Meas.
4.4 mV
VIE
Meas.
4.1 mV
-43
-41
21.9 mV
30 mV
21 mV
30 mV
-20.6
-17
-24.1
-21
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
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
Amplitude
VIE
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
229
5 mm
2
1.5
VIE
FEM
Measurements
114 mm
8 mm
Im{V } (mV)
1
0.5
0
0.5
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
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]
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.
231
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.
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)
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
(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
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)
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]
234
Y.-K. Shin et al. / Design of Reection Type Pulsed Eddy Current Nondestructive Testing
G
(a) Signal (coil spacing = 3mm)
G
(a) Signal (coil spacing = 3mm)
(a) Signal (coil spacing = 3mm)
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 %
(a) Signal (coil spacing = 3mm)
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
(b) Pulse width = 180 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
Aluminum
390 s
720 s
Tungsten
180 s
340 s
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
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.
Applications
241
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
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
243
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.
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
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 5. Experimental setup: a)basic scheme; b) transducer - valve assembly; c) photo of equipments
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
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]
0.2
41.3
58.7
0.4
84.6
15.4
0.6
97
247
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.
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
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.
249
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)
(2)
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
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
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.
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
257
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.
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
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
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
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
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
OD, circumferential
6.0
0.3
0.41
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.
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
62 000
nTF
11.3
nTD
N1
4661
N2
3909
262
R. Grimberg et al. / Lifetime Prediction of Pressure Tubes in PHWR Nuclear Power Plants
360
ND1
187
ND2a
238
ND2b
20
Number of observations in F1
NF1
83
q1
0.48
q2
0.70
q3
0.98
(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]
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.
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
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
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
265
[M kB]dV
0, k
(1)
Vm
u N
(KJ jZA)dV
0, k
(2)
Vc
R jZLI jZFM
k
1
DE MF I
(3)
(4)
Lij
Vc
u N i V 1 u N j dV
P0
4S
Vc Vc
P0
4S
1
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
Vc
Vm
(5)
B 0 dV ,
(6)
(7)
(8)
(9)
(10)
(11)
266
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
267
Z large
o LVm
(14)
where ZV
(15)
(16)
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
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.
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
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).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
271
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
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:
(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.
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)
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
10
15
20
25
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
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.
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.
278
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].
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
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.
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
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]
283
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.
284
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
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.
286
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
287
Umax/U0
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
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
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
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
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)
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
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
292
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
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.
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