0% found this document useful (0 votes)
47 views23 pagesNDT & E Vol7 Jan2001-Guided Waves-Remote Field-Composites
Uploaded by
Rafael FerraresiCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
47 views23 pagesNDT & E Vol7 Jan2001-Guided Waves-Remote Field-Composites
Uploaded by
Rafael FerraresiCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 23
NDT&E
international
Volume 34 Number 1 January 2001
“ARTICLES. =
‘The effect of dispersion on long-range inspection using ultrasonic guided waves
P. Wilcox, M. Lowe & P. Cawley
Microtopographic inspection of thermoplastic rubber shoe's sole. The influence of surface roughness
on sole to leather gluing
M.F.M. Costa & V. Pinho
Remote field eddy current technique applied to non-magnetic steam generator tubes
7 Hi Fukutorni, T. Takagi & M. Nishikawa
25 Feature extraction of machine sound using wavelet and its application in fault diagnosis
lin
BE irene of compacta mated cern vng veri concent
8. Legendre, J. Goyette & D. Massicotte
39 Feasibility of detecting embedded cracks in concrete structures by reflection seismology
Y.-F. Chang, C.-Y. Wang & C-H. Hsieh
49° Detection of delamination
¥. Ju, M, Saka & H. Abé
57 Society Journals Digest
63 JSNDI Abstracts
67 Calendar
ae
IC packages using the phase of microwaves
The NDT&E International homepage: http:/www.elsevier.comMlocate/ndteint
Structural integrity Network Website: http://www elsevier.com/locate/sinet
The Elsevier Science Catalogue on the Internet: htip:/www.elsevier.nI: gopher.elsevier.n
CONTENTS. tis journat is part of ContentsDirect. the free alerting sevice which sends tables of
Contents by email for Elsevier Science books and journals. You can register for
‘CoutentsDirect online at: www.elsevier-nMlocate/contentsdirect
NDTBE International has establishedan oficial association withthe Japanese Society for Non-Destructive Inspection (SND) forthe
Publication of selections of their matenalin English. Papers translated rom the Journal of JSNDI are identiied by a footnote ELSEVIER
NDTsE
international
NDTA&E tnemstional 4 (2001) 1-9
oma
The effect of dispersion on long-range inspection
using ultrasonic guided waves
P. Wilcox*, M. Lowe, P. Cawley
NDE Laboratory. Deparment of Mechanical Enginecrin, Imperial College, Edibtion Read. Landon SW7 2X. UR
5 20H recive
Abstract
The dispersion of ultrasonic puided waves causes wave packets to spread out in space and time as they propagate through sete Th
that ean be obtained in a long-range guided wave inspection system. 4 technique ts presented for quickly predicting the
ne of spreading of a dispersive wave-packet a it propagates. Its shown thal the duration of a wave packet increases fi
rpaation distance. tis also shown thatthe duration of a wave-packet after a given prepagation distance can be minimised by ep
1um resolvable distance (MRD) is defined that enables
‘olution atainable st different operating points. Some conclusions are made canceming
© 2001 Eevier
tits te eso
the input signal. A
taade between th
co
ess parameter called a
eins
Keguonds:Dpern, Gunked waves: Lami waves: Long-range testing
rein frm M March 2000; accep 3 Api 2048)
direct comparison te
re resolution oF vai
vence Lid. All rights reserved
1. Background
LA, Guided waives and non-destructive inspection
Much work has been published on the use of Lamb waves
and other guided waves for inspection purposes and a
comprehensive review of applications may be found in
Ref. [1]. Very broudly speaking, the use of guided waves,
fir non-destructive inspection purposes falls into two cat
egories depending on the distance of propagation. Firstly,
there are short-range applications, where guided waves are
{wed to obtain information about a specimen that eannot be
really obtained by more conventional means. These areas
include the det ion of the elastic properties of ma-
{erils 12.3] the detection of defects near to intertaces such
asin the inspection of adhesive joints [4] and air coupled
Ultrasonic inspection of thin specimens [5]. In these cases,
sensitivity is of key importance and generally this is the
smain criterion for selecting a suitable guided wave mode.
‘The elect of dispersion is relatively unimportant as the
Propagation distances are small
‘This paper is concemed with the second area of guided
wave applications where the propagation distance is large
‘These include the detection of delaminations in rolled steel
5 Carepeating stor. Tel: +4 0171-596.7227; fan: 44-0171 580
10
‘mal addres: p.vikon isc uk (P, Wilson).
[6.7] and composites [8]. pipeline [9-11] and plate [121
inspection, In long-range applications. the ain is to inspect
large areas of a structure rapidly
1.2, Long-range inspection using guided waves
In long-range guided wave testing applications. the
‘guided waves are excited by a shor burst of energy (the
nput signal) applied by a suitable transducer at one lovattion
‘on a structure, The excitation causes a packet of guided
waves (the wave-packet) to propagate away from the t
ducer into the surrounding structure. Then either the same
transducer oF a second transducer is used t0 detect signals
caused by reflections of energy in the wave-packet from
surrounding structural features or defects.
‘The problems associated with the use guided Wi
inspection purposes are well decun
multiple modes of guided wave propagation are possible
‘most structures and these modes are generally dispersive
(ie. their velocities are frequency-dependent). In order t0
‘obtain useful data from a guided wave inspection system. it
is necessary to selectively excite and detect a single guided
‘wave mode while suppressing coherent noise due to otly
modes of guided wave propagation. For this reason, the
design of the transducer and the input signal are tailored
so that the excitation energy is targeted at a single poi
fon a suitable guided wave mode at a suitable Irequency.
‘This point is called the operating point.
ed J13]. In summary,
(93 N@SIOL- se frome mater © 2001 Eever Science Lid. AU igh reverse,
PH. s963-sooscony00024-4 P. Wileox tol /NDTRE
@ ©
CGreyscale indicates out-of plane surface
Internationa 34 (2001) 1-9
‘isplacement due te wave-packet
whte incieates where |
envelope of wave packet
€ Is <-20 dB relative to
& ek ple mht
3 ‘stonce
5
5 Tne (i) cy
o>
g
Zo =
:
o - Time (is)
g(a Nom
sisal isa 5- Ws ws ww
Coil postion camnd
Fig 2 Mowat FEC wgnats as function ofthe dsance between the excite an detector very numerical predictions. The thd Hine WD.NI reeset the
fans nth ae fn clement and the fourth ne (3D. epesents hone fromthe edpehsd ite ements fa Onesies the ay nthe
ttm oe tween the nara remote Held zones, the tendencies that sigral aces decrease ae il
pho
imdels are set: one is For a comparison between the exper-
menial measurements {13] and numerical predictions. and
the her is for the applications to SG tubes. The straight
tuhes are fabricated from brass and nickel-hase superalloy.
‘which malistic SG tubes are made from. Fach probe consists
ff an exciter und a detector coil, which have the sme
dimensions, Table | summarizes the important dimensions,
‘material propenties, and test conditions.
In Fig, 2. the induced voltage changes are plot
etector moves with the exciter fixed. to show the agreement
betweun the experimental measurements and the numerical
predivtinas of the axisyrametric and three-dimensional
Serions in. model I. This good agreement implies the
same degree of accuracy in predicting the eddy-current
response due t a defect oF a crack, By the four-way symi-
Inetry only a quarter of the volume is discretized in thre
imensional problems, and Table 2 summarizes. the
‘computational costs. This finite-clement code is considered
tweful for three-dimensional problems.
das the
‘the?
Gotan eons te fie clement cod in tree diersiema mets
Moet | Moe! 2
ade ase Ege bacd Faye based
Heer 24192 2216
Nas 2R280 asi
taps. wos: a0)
Tite 71.706 65.246 59687
emery)" 20 200 200
Urine” 16 nis ne
The compstion i carcd out wit double precision
The eampute i fom Viel Technology with DEC alphs A216
nine ck espe S35 MUL The acceleration factor the ICCG
fet aren pak hse clement and 10? in edge bused elem
Comyepeee ance 6 = 10
ni (b) Inthe tranition re tet te sharp hops
Similar to Fig
field are taken as the detector is pulled av
stationary exciter at 200 KHz (skin depth: 1.13 mm). Fig.
3 shows the comparison in model 2 between the axisym
metric version and thr jonal version with the
cedge-based elements, and the
‘agreement in the SG tube model, At this frequer
skin depth approximately corresponds tothe tube Wa
ness. the log magnitude falls rapidly with a near straight line
in the direct field (0 to 0.9 diameters from the exciten. and
dips are abserved: the phase promptly drops in the san
field. This can be expla the decline of
rmgnetic flux direetly transmitted from the exeiter ix mich
faster than that outside the tube in the ax
Therefore, indirect magnetic flux is dom the remote
Id zone inside the tube. To comprehend the REEC
phenomena, it is also worth mentioning that a Poymting
vector real energy plot shows that the transition zone
‘occurs where outwardly directed energy from the exciter
meets inwardly directed energy from the outside tube in
Fig. 4. Although it is known that the distance between
fan exciter and a transition zone, in general. is 2
2.5 diameters, it appears in a diameter from the exciter
in this case
‘The signal traces associated with a circling circumfere
tial outer defect, which are now-through, 2 mm wide and
206 deep through the wall, are shown in Fig. 5 and are
‘computed by axisymmetric version. this figure, the indi
‘cations occur when the exciter and detector pass the defcets.
‘A circumferential defect gives us two approximately same
high peaks in magnitude with the exciter and detector
passages. In this model. the same solution ix obtained
‘using the node-based finite elements. The voltage ch
ddue to the defects mentioned above are plotted in Fig. 6.
following impedance plane analysis form of ECT. The
maximum value and its phase changes due to the defect
esults are also in excellent 2» 1H. Pabutomt etal. / NDT&E International 4 (2001) 17-28
o $18 2 28 Ww a AD
Ca positon ro
(a) Magnitude
Fig.3, Numeral prediction of RFEC yignls as a funtion ofthe dance between the excita detector
nt ECE
CLE E oh
mee 7
08 Wis 0 sw WO as
al position am
(hy Phase
200.
[ Sourside watt
Fig 4, Anisyinmewre Hte-clemem prediction of Poynting vector ral energy ditestons ur the RFEC pisonens in the new-ningnetie whe
are detined as:
IM= yav? raw, o
2 1aw "(Av/av,) x 18000, “
Where AV, and AV, are the real and imaginary parts of the
changes, respectively
To detect outer defects (ODs) with higher sensitivity
‘maximum changes dc to two 40% deep and 1 mm wide
— tao |
Moyne | E
2 Pie
0
Atcrnmare (ates pose |
SO 0 010 awn 5
Cal postion
ig. 5, Sipmaltaces asexiat with a cicumcntiel outer dee
wean 0% Joep throug the wall at 2UD He with eal
The cig of the wo peaks ate she sae eth in magni ant
lnscad of the poke mation the deft ves in the numeral
Fig. 6 Trajectories ofthe voltage change de to
‘Plane. The defect considered s the sine hat a Fis
otects (an inner defect (ID) and an OD) were predicted
with various coil intervals and excitation frequencies. Fig.
7 shows the ratio of the OD to ID in magnitude and differ
tences of the phase changes between them, It can be seen that
the signal changes of the OD are discernibly higher than
those of the ID when detector is in and around the transition
zone. a shown in Fig. 4. To he exact, the volume of the (0
defects is slightly different but the difference can he
neglected in this comparison, Irespective wf preferable
high sensitivity, duc to noises linked 10 fift-off variations
(oF magnetic inhomogeneities, the use of the transition zone
is generally avoided. However, SG tubes ane non-magnetic
'
Imaenary pat V)
oon
soto anid? soato?
eal pc)
eect compen —_ = ;
Vs oan
f Ps roar} tS
0 1s 20 28 wah as wo 15 20 28 + 38 a as M0
Co eter {Cot nevi
_ Moris te
Fig. 7. Magitade tes and hae changes of maximum signals of an ID and an OD, which are 2m wide and OF% deep. aS. withthe various
‘change are dete s/f | and argtS)~ art), respectively, The OD is deste with highs snsis ty when
aie
ely cel ae eo
afl ooo A
z T | la Gw
Bae !
x10”
vw ML fll i o
7040" 60 100" 6a 40." 38 aoe hehe ne
od pater) ate deh ha
e
i? (a) Transition zone
kul!
0 x0 100 060 4) 20
Detect depth
(by Remote Held sone
Ds. hvough detect ar the ODs, 2m wid, 200 EW
31 ODsane detected to be ro serie than 1D. ae hep
the ptase changes donot depend on te depth andthe phase ome
Fhe RIC signals va fonction ofthe detest depth du othe cute
‘At 1Stamton ane) and XY mente el one, ay There sa advan
twhrn- detect dhs.) With sppronsmtey equ senstin ity to he de
the mason mnduds remit consan
lex 2 1. Fakatcmi ee NDT&E International 242001) 17-28
i [ete {Toot T)
Win
i 9. RFEC signals ava function ofthe detect depth due wo ine ough. andor pty, 2m ent ad wih, a 200 KH with a coi geval 150m
The same chart
and have smooth surfaces inside, and the there is no need to
lake the velocity effect by the probe motion system into
account. Therefore, putting a detector in the transition
zone is expected to be useful in terms of the detection of
small outer defects such as eracks. As the tube wall becomes
thicker, the same characteristies are observed at a lower
frequeney where the wall thickness to the skin depth ratio
is of the order of one.
The RFEC signals obtained by means of the axisym-
metric and the three-dimensional models are compared at
200 KHz for defects with various depths and for two loca:
tions of the detector: in the typical remote field and trans
tion zone, The defects are 2 mm wide. and 20, 40,60, 80 and
100% deep. Fig. 8 shows the maximum changes in magni
wide due to the defects as a function of the defect depth. The
revults demonstrate good agreetent and itis distinctive that
the correlation between the changes and depths is approxi-
‘mately linear although the changes due to the outer defects
from typical ECT exponentially decline, not in direct
proportion to their depths. Unlike Fig. 8b), the signal
changes due to the ODS are larger than those of the IDs in
Fig. 8(a) a the frequency as the wall thickness to the skin
depth ratio is ofthe order of one. As for the phase, while the
changes reflect the depths in Fig, 8, the changes approxi=
‘mately remain constant in Fig, 8(b). This means setting the
detector in the transition zone makes it possible 0 know
tect depths from acquired signals as well as typical
FCT. For practical purposes, RFEC signals must. be
predicted from arbitrarily shaped defects in three-dimen
‘sional problems. The same tendency observed in Fig. 8 is
‘observed with pit-type defects of 2 mm length and width. as
shown in Fig. 9
4. Summary
For an improvement in detectability of outer cracks, the
es of the RFEC techniques that stems from the
throgh-transmission effect was used for the inspection of
non-magnetic tubes. The experimentally validated finite-
Ht code solving an electromagnetic governing equi
fx are verve on the condition thatthe detectors fied in the transition zone asin the case the cncuaferctial deere
tion constructed with a magnetic vector and an electric
scalar potential way used in the axisymmetric and three
dimensional problems, to evaluate the detectability. Both
hhumerical predictions were compared for the purpose of
accuracy of the numerical simulations tested in the medel
simulating the SG tube inspection. Sine they ane
{ent agreement. the use of the simulation is possible and
what is more, valid to study the phenomena and_petfor
mance under the test. From the results of the numerical
simulation, the common RFEC tee re expected to
detect outer cracks as sensitively as inner eracks when they.
sto the non-magnetic tubes such as st
ator tubes in nuclear power plants, Moreover. this paper
demonstrates that when putting a detector in a transition
zone, signals to an outer defect ate higher than those ta
inner defect with the same depth and volume, and the depths,
are classified by their signal phase informatio
usually used in ECT.
um gener
wit is
Acknowledgements,
This study was supported in part hy the Research
‘Committee on the Nondestructive Evaluation Technology
by Eddy Current Testing of the Japan Society of Applied
Flectromagnetics and Mechanics through a grant from 5
PWR utilities and Nuclear Engineering Ltd. andthe
Grant-in-Aid for COE Research (11CE20N03). ‘The autho
‘would like to thank Prof Hiroshi Hoshikawa and Dr Kiyoshi
Koyama of Nihon University in Japan for their advice and
cencouragement for this work
References
UT Sehmigt TR. The remone tld ey cums napecion tet
Mater Eval 194202) 225-0,
12} Hovpland SM. Fundamental alysis he rx te ky cma
fet IEEE Teans Mag 199632141195 211
ond W. Sun YS. Udpt $8. Nath S.A fie stent stely of he
toe fick edly current eld 1FEE Trans Map TORR 1) 496-8
141 LinF1Y.Sun YS. Application o 00min extn AD en a
sally cre fft computation. HEE Trans Mie 19912642 8k 1. Fao ol / NDTEE Interational 48 (2001) 17-25 n
[3] BeghC. beats WE: Analytica solutions wo eddy-current prabe-ct
blr 3 Appl Phy 1968106) 280-38.
1A Paensany R Lord W. Fite element modeling of electromagnetic
[NDT phemxnen. WHEE Trans Mag 1979:15(6 1479-8
IM) Mi, Loed WA tat element model for thee dimensional ody
‘orert NDT phenomena IEFL Trans Mag 198521652539 43,
IM Tog, Hosunwao M. Fuku H, Kurokawa M, Miya K. Tsuboi
Taka M, Tay. Serizawa T, Harads Y, Okano, Mra 8
Benchnuk made fed curen testing for Seam generator tbe:
‘xg and numerical analysis, Int} Appl Eleewomagn Mater
Iwpts99-02
[0 Fukubou HL, Takap, Tani, Hashimoto M, Numerical evaluation of
CT impedance signal de We minute cracks IEFE Trane Mig.
134) 2124-6
Ho} Fukotons Ht Huang H. Takagi T. Tani J. Mlentitction of crack
epths fem eddy cure esting. signal IEEE Trams Mag
1996:34572893-6.
111} TakapsT, Uesha M. Miya K:Elecuomagnetic NDE rescare tiv
ies in JSAEM In: Takagi T. Bowler JR. Yoshida Y- eur Ee
tromagnetic rondetrctive ealtion 108 Press, 197 p96,
112] Talae! T, Huang H, Fukuomi H, Tani J. Numerical evaluation of
cenelaion between crack size nd ey current testing vial by a
‘ry fis simulator IEEE Trans Mag 1K: 343) 298
113] Kee M, oats K Nishikawa Mt High sensitive remote tel ey
cater esting by esi dl exciting co, Collins, Doser WD.
Baws JR. Biya K editors. Nondestructive testing of material, 10S
Pros, 1997p. 14582 ELSEVIER
NDTsE
international
INDIE: Ineratonal 4 (2001) 31-97
Ultrasonic NDE of composite material structures using wavelet
coefficients
S. Legendre*, J. Goyette, D. Massicotte
drogen Research Insite, Universit du Quebec Trois Rviéres. CP. 500, Tue Rvidren Quoc, Cama GA SH
Received 20 November 199
tract
A-wavelet-bused merbod is propased o perform the analysic of NI
owner nate
Is. The non-homnogenus nature of such
serials induces a very high level of structura
the imerprctation of the NDE. signals, By combining the time domain and the classical Fourier analysis
Simultaneously specteal representation and temporal order of the signal decomposition components, To construct a C-sca
‘rave irnsform of the A-scan signals, we propose a selection process ofthe wavelet coefficients followed by a imerpret
ied in evn form 2 May 2000; acspted 16 May 2000
tultrsonic signals received during the inspection of
se which greatly com
the wavelet 1
son pravides
rage fom the
Insel on a windowing process in the time-frequency domain. The proposed NDE method is tested on eryogenic glawepory hydragen
rir samples. © 2001 Elsevier Science Ld. All ights reserved
Keown Cnuenie mera; Sina proces
rosie evaluation: Wavelet ansform
troduction
In its liquid farm, hydrogen is of
‘composite material reservoirs, These reservoirs can be
subjected high mechanical and thermal constraints that
could create flaws having a potential of weakening the res
oir structure [1]. Non-lestructive evaluation (NDE) of
fomentional reserveirs is offen made through a C-scan
analysis using ultrasonic bulk-waves as a probe; the signal
‘uupling hetween the ultrasonic transducer and the piece
funder inspection is made by immersing it into a water
funk, Due t© the non-homogenous nature of reinforced
‘composite materials, new inspection techniques need to be
Akyeloped in order to be able 10 perform the analysis of
IDE ultrasonic signals that can be highly complex. For
example, some peak overlaps are created by scattering
‘Phenomenon anal the resulting background noise, also called
sinetural noise, is high enough to bury the meaningful
reflection echoes in a signal having a large number of
frequency components. Consequently. the process of flaw
tection is not simple and a well-adapted kind of signal
analysis is required to interpret the ultrasonic inspection
signals. Since. in this case a meaningful Fourier transform
en stored in cryogenic
Comapeing ab Tel: +1937 SAT: fx: 1-819-376-5219.
mal address: aye pendrwerugeugucbec ca (S, Leper)
espe yoetiovpe quchocca J. Goyette, diel messcote@ ut
lunitecen(D, Maceo.
analysis could he dificult to perfarm due toa possible shill
ing ofthe peak frequency. some processing methods based
con the principle of quasi frequency diversity (QED). st!
the well-known split spectrum processing (SSP) and its de
rivatives have been proposed [24] These itering metho
allow a significant reduetion of the strututal noise in
signals by applying a process of noise dees
permits a coherent information detection inthe
domain
Some new kinds of processing methods based
frequency analysis, such as the wavelet
bbeen proposed recently «0 treat similar problem [5-8]. By
combining the time domain and the clas aly
sis, these methods provide simultancous spectral represen:
tation and temporal order of the signal decomposition
components [9]. The min advantage of the wavelet trans
form-bssed analysis method isthat they ae suitable for peak
detection problems in highly noigy envisonment [8]. The
wavelet transform prod
wavelet coethvient sets or wavelet levels
signal, separating the different frequency components While
keeping its temporal representation. In this paper, we
propose a method of A-scan ultrasonic signal analysis
tuased on a selection provess of coeliciens, alo called
feature extraction, provided by the wavelet transform,
Once sclected and interpreted these features allow Us t
construct a C-sean image of the inspected structure, The
proposed NDE method is tested on cryogenic glasslepony
es diffore
representations: —
‘ofan analyzed
OH 4ST see fot mace © 2001 Eee Sctene Li A igs reserved
PI: Soves-no9scany00029.5 reservoir samples in which varying size circular defects are
In this paper, the following points are discussed: (i)
problem formulation; (i) presentation of the algorithm
used fo compute the wavelet transform of the ultrasonic
NDE signals: (it) description of the proposed imaging tech-
nique based on the wavelet coefficients: (iy) the experimen-
tal set-up used to inspect the composite hydrogen reservoirs:
and (v) experimental results of the proposed NDE method.
obtained on a glassepoxy sample. Finally, we will conclude
in the last section
2 m formulation of NDE signal processing
TThe main goat of the analysis of the signal received by the
ultrasonic receptor in a NDE process is to obtain infor
tion about the material inteprity, We therefore need to detect
1 possible reflection induced by a defect in the structure in
signal that itis corrupted by the structural noise generated
Inainly by the reinforeement of the composite: material
From a mathematical point of view, one can define the
ved by the transducer
‘consists in a sum of components — the front surface ech
— the hottom or defect echo, further called target echo —
the background oF structural noise [10]. The material
‘evaluation consists in detecting the target echo and in deter
ining i the bottom or a defect generates it, This can be
summed up by the following equation:
=e + ein + MN) o
where (2) is the received signal, e,(7) is the surface echo,
‘(2 isthe target echo and (0) is the electrical and structura
noise, The analysis aims to extract information on the target
echo, especially on its origin,
‘A Fouricr-transform:tased method of analysis could be
Aifficult to perform due to a possible shifting of the mean
ingful signal component frequencies and inefficient because
‘would neglect the time domain representation of the
signals which can be important in that case. To deal with
these drawhacks. we use a wavelet-transform-based method
‘of analysis. The concept of such a multi-scale analysis
method allows us to extract simultaneously time and
Wequency information. enabling temporal windowing
process impossible in the Fourier domain, Due to the wave
let transform ability to adapt the window size of the
processed signal [6]. one can easily separate coherent infor-
‘mation and noise without needing a complex windowing
Step. as il would be required by Fourier analysis. In addition,
while the ultrasonic NDE signal analysis requires temporal
information, one can consider that is a priori knowledge
provided by the physical nature of the evaluated material
3. Orthogonal wavelet analysis,
In this section, a synthesis of the basic concepts of the
2 S. Legendre otal. PNDTEE Itermatonal 34 (2001) 31-37
onhogonal wavelet transform used to calculate the waved
Coefficients of the A-scan NDE signals is proposed. A wav
let 4AA) is a function of zero average that is dikated wi
scale parameter s and translated by u [9]
ato= eof 2) é
1
o= fro
for
where the symbol * denotes the complex conjugate. Of
thogonal wavelets dilated ya scale "cary sig vat
ton atthe scale 2 Tn that eave the enthonormal ask
(Gt) rte a
wy
dened by
L finan
y= 7
(0 Yh :
form = 1,2,...,Nandj = 1,2.....J with NV = 2. The signal
{FO can be decomposed over this ha
so-
(fia ol
and the partial sum of wavelet cove
DY adtatt ol
‘can be interpreted as an approximation of the signal /(0).
¢ wavelet transform of the ultrasonic signals
computed by using a fast orthogonal wavelet transform!
algorithm described in [9] and implemented in WaveLaty
[IL The main advantage of the otthogonal wavelet trans
form is to avoid a redundaney characterizing the continuo
‘one and which provides lots of wavelet coeflicients carrying
the same information. This algorithm is based! on the fille
bank theory — once the function /(2) is digitized,
ing sequence fin) is decomposed in two si
d(x), of half its size using filtering and sub.
dure. We denote the approximation of f(a the scale 2! by
the column vector d, Further the approximation d, will ke
referred as a wavelet level, each wavelet level represen
Uifferent Frequency band
coefficient set. Finally. the wavelet representation of ft
consists of a concatenation of all the wavelet levels [1I}
to form the vector W,
we dime wi)
aida yd! aati]
with dimta,
specific wavelet coetficient belo
NPY tor j
ng. to the approximati 2
y
oe
ia
ay
6)
CO}
nat
orm,
Labs
ying,
filter
ssult-
) and
2Iby
il be
A
‘Legendre tal J NDTE Ieratonal $6 (2001) 51-37
~~
|
.
5
|
Fig. 1. Cotman water yre C511
of fn) atthe seate 2
1sk=Nr2.
‘The choice of the mother wavelet, a Coifman wavelet
also called Coiflet, was made in considering the shape of
the peak to detect in an ultrasonic signal [6]. As shown in
Fig. 1, we use a 18-point C-5 wavelet to perform our signal
analysis [11,12]. An example of the orthogonal wavelet
transfo Ultrasonic A-sean signal is given in Fig. 2.
tly the five highest wavelet levels are showin the original
Signal isin the lower part of the figure. Considering that the
tiiginal discrete signal contains N= 1024 samples. the
highest and lowest wavelet approximations contain S12
an! 16 samples, respectively.
wl located at kis referred as dh).
st
|
Wavelet level
rT a ee
|
Amplitude (mV)
ot
02 Oa 08
) x0"
Time (rs
fie 2 Wowk levels ot an unone Asean ina generated ina peed
fhny comes suse compe woe oil ina inthe ower pat
te ere
4, tmage generation using wavelet coeffi
4.1. Feature extraction
“The ability of data compression of the wavelet transform
allows one to use a restricted number of wavelet coefficients
find possibly a restricted number of wavelet levels to char
ultrasonic signal
jcterize each component of an A-sea
This result is based on the preponderant discriminating
power of the main coefficients that constitute a set of dis
Priminating parameters, Once processed, these discrimi:
hating parameters produce & set of features, which
characterizes the reflection data. The determination of this
ct of features, ideally independent from each other, is called
feature extraction, TWO important assumptions allow ws 10)
fimit the number of wavelet coetficients and wavelet levels
to be considered during the feature extraction step:
Lin a recorded signal, one can consider the highest
frequency components formed from the first wavelet
fevels, as representing the noise-not necessary the stu
tural noise. As a result, wavelet coefficients contained in
them can be ignored.
‘The wavelet coefficients contained in the last wavelet
tevels representing the lowest frequency components of
the signal are not significant in our analysis meth, Such
‘an assumption is allowed by considering that the Follow-
ing analysis and image generation steps a the
time-frequency domain and do not require any
‘wavelet transformation,
In the wavelet representation of a signal. Wf, the simpli
fications induced by taking both of these assumptions inte
‘account are done by a procedure of wavelet levels cli
tion [6.8]. Ifwe use H and to represent the thresholds used
tocliminate the non-significant wavelet tevels. only levels at
the scale 2! fork! = j = L are kept. Applying the followin
procedure modifies the vector dj, the thresholds H and Late
Empirically determined. nevertheless they depend on the
both reflection peak and sampling frequencies
0 e
for = booed and
wy = (0dydy_ dimN=N
‘The following processi 5 in selecting cook
ficients in each kept wavelet level d) tor H= j= L. We
‘assume that the signal Features consist of a timited
Gf highest modulus wavelet coefficients within exch retained
revel. Thus. @ simple selection procedure that sets to zero the
‘wavelet coefficients that are not taken into aceount allows us
to complete the preponderant coefficient extraction step [SI
We represent by M, the number of selected coetticients in u S Legere ef. /NDTE&E tovermational $4 (2001) 1-37
Wavset oe!
Time
a ae _
») “tee rel 10"
i Foot extraction from ulsonic ign Hf = 3.L= Sand, =
fo j = 3.4.5 ey (8) (delet fee region, aa (by region with defect
wavelet level dat scale 2! and d the resulting vector. An
example of & Wavelet-irinsfonm-hased feature extraction
process applied on A-scan ultrasonic signal reflected
from a dfect-ree glasvepony region and from a rexion
with defect, respectively is presented in Fig 3, the surface
and bottom reflections ae indicated by “¥” and
spestively
4.2, Cscan image generation
(Once the feature extraction procedure has been applied on
an A-scan signal, sets of some non-zero wavelet coefficients
belonging to one or several wavelet levels d, atthe scale
for H = j= Lare created. These coefficients translate the
lifferent components (Eq. (1)) of the analyzed signal in the
time-frequency domain by parting the frequencies in separate
‘bandwidths while keeping their temporal onder. ‘The coherent
information in the signal consists in both front surface and
bottom oF flaw reflections (Eq. (1D), the evaluation of the
structure is given by the origin — bottom or flaw surface
— of the target echo, To extract this information, we have
developed a procedure of feature interpretation that allows us
1 quantify the amplitude of the bottom reflection, the corre
sponding value is being called the characterization parameter
The proposed method of feature interpretation is based on
the simple principle that a defect-free region induces
nnevessarily @ bottom refleetion signal af @ constant temporal
location defined by the physical characteristics of the
material (thickness and mean propagation veloc
On the other hand, if no bottom surface
detected, then the inspected region potentially conti
a8 defect. The principle of this metlidl consists in appl
ing a window on each kept wavelet level, in such wa
that the window allows us 10 examine the presence
not of a bottom peak reflection, This procedure requi
ection
the adjustment of parameters defining the window si
for each wavelet level. This can be done easily. un
consideration on the physical characteristics of 1
inypected structure — the arrival time of
echo is dictated by the thickness of the
‘or empirically during a calibration step on a elk
known part spect
structure regarding the position of the main eoefticiens
in the corresponding A-sean signal w:
The following three steps deseribe the complete proces
of image generation:
bowton
ucture —
~ in terms of in
1. For each A-seam signal r,(0) for = ToocNy amd
TeovaNe Where the indices + and y indicate the spl
coordinates of the A LN, and Ny ane th
numbers of rows and fines of the C-scan respective
(Go simplity the notation indexes, » will be dropped
the following steps
© generation of a discrete signal (n) of the signal rf
© calelation ofthe wavelet representation of 1) a
production of Wr according 1 Ey. (6)
© production of d, according Eqs. (7) an (9) for H
jst,
© windowing on for 4 = j = Land determination d
the wavelet coeflcient mean, Mj. in the Wind
according 0
1 es
wore DM «0
whore w) detnes the sizeof the window applied on
wavelet level at the
© calculation of the char
the signal 7,0.
s [s>] ‘ a
This summation of the main wavelet coeflicient
retained at different frequency level (H