nondestructive damage detection scheme for steel bridges · there is a glowing need for bui;yin...
TRANSCRIPT
Nondestructive damage detection scheme forsteel bridges
著者 Beskhyroun Sherif, Mikami Shuichi, OshimaToshiyuki
著者別名 三上 修一, 大島 俊之journal orpublication title
Journal of Applied Mechanics
number 9page range 63-74year 2006-08URL http://id.nii.ac.jp/1450/00007372/
Journal ofApplied Mechanics Vol.9 (August 2006) JSCE
NoRdestructive damage detection scheEne for stee] bridges
SherifBeskhyroun", Shuichi Mikami"", and Toshiyuki Oshirna"""
t-,, p7 bixse-mV)/*, =-UL5s-**, j<blRZ""*
*GraduateStudentDeptLofCivilEng.,KitamiInstituteofTechnology,(165Koen-cho,Kitami,090-8507,Japan)
""Assooiate Professor, Dept ofCivil Eng., Kitami Institute ofTechnology, (l65 Koenttcho, }(lltami, 090-8507, Japan)
"*"Professor,DeptofCivilEng.,KitamiInstituteofTechnology,(165Koen-cho,Kitami,090-8507,Japan)
This paper pregE)nts structural damage deteodon method based on changes in Transfer Function Esdrnate
(ITFE) for detecting damage, piedicting its location and monitoring the gtovvth in damage, This method
assumes that the dispiacement or the acoeleratjon response tinie histories at various locatioms along the
strzicture both before and afier damage aie ayadable for daiTiage assessment rftrese Tesponses ae used to
estimate'IFE.ThechangeofTM]betweenthebaselinestateandthecurrents{ateisthenusedtoidentify
thelocationofpossibledamageinthestnJcture.Tiiernethodisappliedtotheexperimentalandnumerical
data obtained fiorn a railway steel bridge that is no ionger in service seveial damage scenados weie
irxtroduoed to the main girder of the test structure. Resutts show the method can be used to detect aie
damage, determine tire exact location and measuie the giowth in damage with very good accuiacy. The
use ofpiezoelectric actuatDrs as a local excitation sow'ce fbr large structuies such as steel bridges is also
presentecL Experimental and numetfical results show that the proposed approach may be suooessfu11y
irnplemented on-line to detect tire darnage and to locate regions where clamage occuned,
Kej, PP'bizty: Damage detection; Modal paiameters; Health monitoring; Tmsfer Function
1. Introduodon
There is a glowing need for bui;yin monitoring systeims for civil
engineering infiastmctures due to problems such as increasing naffic
loads and rising costs of maintenance and repair. in the past two
decades, a significant amount of effbrts hEvve been disected towards the
development of structural health monitoring (SHM) and non-
destrLictive damage detection methods to rr}anage civil structures more
ethcientiy. Numerous papers aie Ewailable in the technical literature
related to non-clestmctive damage deteedon and evaluadoA SHM, and
ins{rumerrtation schemes, Significant effbrts hEasie also been fbcused on
developing data collection piocedures and damage detections schemes.
The term SHM has gaiiied wide acceptance in the past decade as a
meantomonitorastnictuteandprovideanearlywemningofanunsafe
condition using real tirne data. The goal of SHM and other so called
`smart s{nictuies' technologies and concepts is to clevelop
`multifunctional' structures, i.e. struc tures which provide functionality in
otirer aieas besides the pitimary focus of canying opeiational static,
dynamic and fl]tiguc loads with the ultimate objective of providing an
enhanced level of system perforrnanoe. In addition to SI-ilvl, a broad
mnge of smart technologies are under development at universities,
sensor and actuator oompanies, and aerospace system manufactures. In
recent yeais, there has also been a rer}ewed interest in tire dannage
diagnosis and health imonitoiing of existing highway biidges using
vibration based demnage identification techniques. Most vibration-based
damage detection theories ancl piactioes aie formulated based on tiie
assumption that imlure or deterioration would primauly aifect the
sdflhess and therefore, afibct the modaI characterishos oftire dynamic
response ofthe structurei'5). lfthis kind ofchanges can be detected and
classified, this measure ean be further implemented for a bridge
monitoring system to indicate the conditioza or damage, or remaining
capacity ofthe stnictuies. It can also bo used to evaluate the seismic
behayior of tite structures. Howeveg conventionally defined modal
paraineers hEyve been shown to be mildly sensitive in tire detection of
various types ofbridge damages. Furtheiinore, the modal pasameters of
conventional modal testing such as fiequencies ancl modar damping aie
gEobal parameters, which cannot locaie the damages`{}, Reseaich etlbrts
have been made to detect stiuctuial damage directiy fiom dynamic
response ineasurements in the dme domaipa e.g. the rmdom decrement
techniqueD, or fiom fivquency response functions (FTU )8}, Also, sonie
damage detection methods haye beqn proposed to detect damago using
system identification techniques9` iC)), ln this paper, an algorithm based on
changes in TFE is presented. The algorithrn is used to detect danage,
1ocate its position and monitor the increase in damage using only tlie
measuted data without the need for any moclal identifiearkon or
nurnerical models. The method is applied to the experimenmi and
-1 5-
numerical data extiacted fiom a railwEry steel bridge afier inducing
sorr}e defects to its members. The damage was intrDduced to the bridge
through the release of some bo1ts f}vm some stifil]ners located on the
webofthemaingirderofthebridge.AfimegoalofacompreheRsive
bridge management sygU)m is to have a se1fimonitoring bridge where
sensors fl ed measured responses (accelerations, strains, ete.) into a local
computer. This computer woula in turn, apply adamage identilication
algorithm to this data to determine if the bridgo has significantly
deterioratedtobepointwhereusersafetymaybejeopardized.ThelcK)al
computer could then contact a oentrul monitDhng flicility fyia ceIlular
phone) to notify are appropriate maintenarice or safety oficiE"s of the
bridge'scurrentcondition.ifsuchamonitoringsystemistobepractical,
it wil1 have to identify the dynamic propenies of the stnJ{ ture fiom
ambieng mhc-inducod vibia!ion or using another contrDlled excitation
technique. The ambient vibTation has the advantage of being
inexpensive and convenient sinoe no equipment is needod to erccite the
structure. The servioe neecls rK)t be intermpted to use this technique.
One dilliculty with deterrnining dynainic parameters of a structufe
undergoing ambient vibrations is that the forcing fuiction is not
piecisely charamemi Moclai testing has some weaknesses as well.
One of these is the high level of noise, as compared to the signals.
Another diiificulty of using ambient vibmion data to implement
damage identification method presented in this study is to find two
equal excitation forces (before and afier damage), as require(l in this
metlK}d ln this paper, the implementation ofpiezoelectric actuatDrsii'i3)
as a local excitaion souioe for Iargc stivctures such as stee1 bridgos is
presented. The advantages of using piezDeleetric mmrs instead of
shakers hammers or ambient vibrations wM be discussed in detai]s in
thefoliowingchapters.
2,Theoreticaldescription
A novel method is proposed herein for detecting damage, iderrtifying
its location and monitoring the increase in damage using TFE. This
method assumes that the displacement or the aooeleration response
time histories at various locations along the structure both before
and afier damage are availabJe for damage assessment Tliese
responses are used to estiniate TT?E, 'Ilie change of"IFE between
the baseline state and the current state ls then used to identify the
location of posslble darnage in the stnigture. The exchation forces
usedfortheundamagedanddamagedstiu{tuTemusthemiethesame
amplitude, location and wayeform in order to ensure that the
changes in 'IFE data are mainly due to clamage. In order to
overDome the pioblem of the lirnited number of identlfiecl modal
parameters, TFE infbrmadon estimated fiom the various accelerometer
readings at al1 fiequencies in the measurement range and notjust the
modal fiequencies wM be compared before and after damage using the
proposeci methocl, ln order to identify the darnage witli rr}ore
conddence, every measuing channel will be used onoe as a reference
for other channels which wM create large sets of data. These sets of
data can then be analyzed using statisdcal proceduses to deterrnine the
darnage location with more confidence, as wrn be explaif}ed in details in
this seczion Transfer functions are mathematical functions used to
chmerize the inputoutput re1aionships of linear systems which can
be deseribed by the following relaionship:
Y(f)=H(f)xX(f) (1)or
HCf)==Y(f)/XU) (2)whereH(f)isatmasferfurictio4Y(f)isthefiniteFouriertransforrn
(FFT) of the measmement signal y(t) ar}d X(f) is the F[Fr of the
referencesignalx(t),inEq.(1),itcaribesoenthatY(Z,DwMalwaysbe
the input signal .X(f) multiplied by the transfer finiction H(f), for
every X(f). We can imagine that the transfer funodon H (f) is the
object that is modifying the source signal X(f). in this paper, the
tiansfler function wM be estimaied between the structure's response at a
rneasuring point x (t) reladve to a reference response r(t) assuming
that the re$ponse x (t) represents the input foree that is related to the
responser(t)bythetransferfunction.lnmiscase,thetransferfuncnion
represents a re1aion between the stmctuie's responses at two different
measuringpointsxandr.
Let 71r (f) denote the [IIFE which re1ates a response x (t) to a
reference response r (t ). Sinoe every channel will be used as a
reference for other channel& 7J. (f) wru Tepresent the 'IFE which
relatesaresponser(t)toareferencerespc)nsex(t),ThereladveTIIl
betweenx and r can then be defined as:
Rxr (f) == 71tr (f)m7;= (f)・ (3)
Rxr(.f)representstherelaivemovernent(response)betweenxandrin
the fiequency domain. ifequal foroes aie used to exche the undamaged
struc ture a number of times then it is assumed that the same re1aive
responsg, Rr(f), wM be obtained each time. On the other hanct if
damageoocursat(ornear)thelocationofxorr(orboth),thenthevalue
ofl<,, (D wilI in turn change. The ahsolute difference in absolute value
ofRxrU)beforeandaflerdamagecanthenbedefinedas:
Dxr(f)=Rxr(f)l'IR:r(f)i (4)
where the asterisk denotes the damaged stmcture. When the change in
re1adve ma Q,r (f), is measured at different fiequencips on the
measurementrmgefrom"tojX,,amatrix[D,]canbefbrrnulatedas
Dr =
Dlr(fi) Dlr(h) "'""
D2r(JXi) D2r(L!) ・・・・・・・
Dnr(A) Dnr(fiz) '"""
Dlr(f)n)
D2r(.f;n)
Dnr(f;n)r
(5)
wherenrepieserrtsthenumberofineasuringpointsandrrepmsentsthe
number of reference channel. ln matrix [Dll, every column represents
the changes in P. (f) at diflermt measuring channels but at the same
fieciuency value. Each measuring channe1 wM be used as a reference
for the other channels (r = i: n), Thereibre, the manix [D,] wM be
formulated n drrerent thnes (3D matrix), The tomi change in the
re1ahve TFE in the {iequency range ofl17 to.1:, can be esthnated fivm the
sumofrowsofmatrix[Dilas:
-fi 6-
STr =
]li: Dir (f)
fZ D2r (f)
f
(6)
£ Dnr (f)
frwheief==ri:thandr==1:n.
The surn of the changes in the relative "IFE over difflerent fiequencies
using different refeie2ices can be used as the indicator of ddinage
ooc urTence. in other words, the fiist damage indicator is calculated from
thesumof{ST,}overdfferentreferenoesas:
r'Ilhis indicatc,r is used to cletect the (x)currence of damage and monitor
the gtowth in damage; however it was fbund to be a weak indicatnr of
damage localizatlon. A nurnber of statislcal decision making
approaches will be cmployed to determine the lcwation of dainage. The
first smp in this procedute is the selec tion of the maximum change in
relative TFE at each bequency tine (the maximum value in each
columnofmatrix,IDil)anddiscardingallotherchangesinrelativeTFE
measuied at other nodes. For example in matrix [Dg (Eq, (5)), if op
(.1;) is the maximum value in the first colum then this value wru be
usecl as Mbr(]7) and other values in this column wM be discarded 'Ihe
sarne prooess is applied to the dffarent columns in matrix [Ddi to
fomiulate the matrix of maximum changes of re1aive TI E at diffk:rent
fiequencies pafi. It skould be noted that iJVI- is a 3D mahix where
eachvalueofr(r==1:n)forrnulatesonemauix;
Mr =
o oM3r(fi)
o
o
oM2r(h)
o o
o
o .". o o .m o O -- M3r(tht)M4r(ll)-- O
o oiJ
.(8)
In order to monitor the fiequency ofdamage detection at any node, a
new matiix ua is fbnnulated. The matrix consists of O's at the
undamaged locations and l's at the damaged locations. At each
fiequency line (each colurnn of M,,), damage location will be
iepresentedby1.Forinstance,inthematrix[L,],anurneiicalvalueof1
isinseitedatthelocationsoflvab,C17),M!b・C]S).,.ete,asfoliows:
Lr =
OOO ---b OO 1 O .."". O
1 O O ,,"". 1
O O 1 .".". O
o o o .-". oiJ
The summaion of maximum changes in relahve tlFE is calculated
from the sum ofthe rows ofmatrix [M,] using diffk rerit referenoes. At
each value of}; the sum ofroxvs ofmatrix [][Y[A will result in one vec tor.
Therefore,ndifferentvectorscanbeobtained;
SMr "
: Mlr(f)
f:A42r(f)
f
:Mnr(f)f
r
(9)
(lO)
ln acoordance with previous proceduTes, the total number of times of
detecting the damage (nurnber of frequency 1ines at which damage is
detectecl) at different nodes is calculated from matiix [Ltr1 as:
SLr =
ZLir(f)
fZ L2r(f)
f
2Lnr(f)f
(11)
r.At each value ofr, {SL} repiesents the nLunber offiequency lines at
which damage is detectod at each node. Asstmiing that the co11ection of
the damage indice$, {SZ}, {SM,}, and {SLtr}, iepresents a sample
pQpuladon of a normady distribtAed random variablq. nonmalized
damage localization indicators aie obtained as tbllows:
{S7]i}-fi1rSTNr =" (!2) alr
where fihJ and ai,・ represent the mean and standard devladon of the
elements in vectDr {ST.}, respectively. It should be noted that for each
value of r, new values of fiir and 6i, are esdmated. Similarly, the
noirnalized vec tors {SMN,} and {SLN,} aie formulated as:
sMN,={SMr}-hr, a3) 'cr2r
{SL,,}- th,.SLNr =
03r(l4)
where l72r and op,j represent respectively the mean and standard
deviation of the elements in vector {SM,.} and fi3,・ and oti,i represent
respecuively the mean and standard deviation ofthe eiements in vector
{SLi.},Inordertoreducetheeffectofnoiseormeasurementerrors,a
thi'eshold level has to be defined. in yectors {STN,}, {SIVIN,} and
{SLN,}, values smaller, than the threshold ]evel will be discaitded. In
summary, damage localization indicatois SM and SL are not
normalized, no threshold is used and damage 1oeations can be
predicted at the maximum values or pealcs, On the other hana STTNI,
-1 7-
SMN and SItiN aie normalized and damage is predicted at the
locations where their values exceed the threshold level. rllhe
advantagesanddrawbacksofeachapproachwillbediscussed.
3.Itailwaysteelbridge:descriptionandexpeimentalsetup
TheexperimentaiwotkinthisreseaiChwasperfbnnedonamiSway
steel bridge that is no longer in service, The bridge was mmoved from
its servioe location severul years ago and is now supported on two
wooden blocks, as shown in Fig. I. rlhe bridge consists of two steel
plate girders and two stee1 stringers support the train rails, Loads from
the stringers are transferred to the plate girders by fioor beams located at
various intervals. The bridge dimensions and layout are shown in Fig. 2.
The multi-1ayer piezoeleetric actuator was used for local erccitation. The
actuator force amplitude was 2oo N. Although this foree amplitude
seems to be very smab compared to the sliaker fbive or ambient
vibraioa it was enough to excite the web of the main girder at the
pesition of the fartliest accelerometer. Two amators were used fbr
excihng the web of the main girder in the horizonni direction. The
actuators were located at the upper part on the web of the main girder
Gig, 2). The excitation foroes used for the undarr)aged and damaged
structure are random, equal in amplitude and have the same vibration
waveform but the excitation force does not need to be measured. The
main advantages of using pie2oelecnic mmrs than using
conventional excitation metIiods such as clynamic shakers, or ambient
vibrationcanbesurnmaiizedasfollows:
- Actuatorsizeisverysmallandcaribehandledeasily.Therefore,it
can be fixed to any structural elerrient and nmotely operated for
continuoushealthmonitDringofthestnicture.
- rlhe' trafiic over the bridge does not need to be intenupted as the
caseofusingdynarnicshakers.
- Piezoelectric actuator ean exche the structure at high fieqLiericies,
typically 1-900 Hz thus activating the higher modes of the
su[tame・
- Largenumbersofvibrationdatacanbereoordedinashortimeas
thesamplingrateinthecaseofusingamrcanreach2kHz
- The same excitation force (equal magnitude and the same
wa)veform) cari be procluoed for exciting both the undamaged and
damaged structure, which is needed for applying damage
identilication technique studied in this paper.
- Undesiredvibiationsinducedfromwindtrailicoranyothersource
can be ayoided since the vibration data induced fiom the actuatDrs
can be generated at any desired fune.
Fig,1Photoofthebridge
ft
geeAccelerometerUnits:mm:
ooDamagg-1!!]og:tt/LqnstrF
-dmActuator
l
.-l・t4.>.5
t1't1
I"
rts
e..t6>
Il
o`<i;'or'{2>
orRHe
(3'>
8ts
C7)le'
l
V--4"
.cg)
Roky8!!-"30u6Lt3uiOL5L7LwO-48tw-L5i32s-OL252!-50Lt7gyJ50
-120J HI.120Q.t.1300"un1400 1510 1200 119Q-E.T.l.!91!70
I
';'ti"yxg .stgo Sl9t9-O /
19610
.t.llOg".1200=us1510 1400 1300 1200 /430 /- 360
JJZ20
Elevation
ggm
rl
o.-wor-I
:se
R:
sN
1
l L'v"t
-
I.==---E:'- ZIII![--- I!l3-- P------"L m------.--
'i
JilllE'iiEEEiE's,'ikrffiv
tii41 :
'lt.ttt
I' 't"ttt:・==--・i l
,l!'-i-"I=-'' ,1 :
"'r-'-=---It -t・--- tIT--=-:--.----It.--T-.k-- - -----;---Tl v---t r--tt"-f-. 't'---.n.-.mt--.h =-tt , t '
=X::: 'Kt l il
±#ntE.±lj
×=i .t:
l・: .l. ・I :
'-.S---..---"' l'tits--.-.--.-w--:'I=t--q
-・・"---tt'-.-------)' 'i-:''t''tl'
-- ----r--TIL"HL--L-----t. I
t ---- -'.--Li.-U--I÷ -, , , ,
'i'tt' llilt4
t:'
t'
tmi.. lt'
'tt.t' x
i4-ySt '
-'---i--IIIIEI-- ---------- -t-- -N- - --."Th- r-------"- +-----"--?5YYL80 2930"-: 2910 L 290U970 l 2900 2910 2930 580
lpt--'tn
19610
PlanFig.2Bridge1ayoutandmaindimensions
-18-
Eight aocelerometers were used to measutie the acoeleration response in
the hoiizontal disection. One accelerometer was mounted at the
goometricalcenterofgrayityofeachpartetofthemaingirder,asshown
in Fig. 2. For this stucly, 20-second time histories were sarnpled at a rate
of 1600 Hz, producing 32000 time points. A manix of baseline
undamaged data sets were recorded befbre damage was introduced to
the structure. For each damage case, five sqparatc ime histories were
recorded.Alloftheconnoctionsofdifferentelementsofthebridgeaie
riveted and no damage could be intiDduoed to these connections. Only
two angles (1ook like stilifcmers) aie bolted to the web ofmain girder.
Therefoie, it was deeided to remove the bo1ts one by one from the two
sdffeners to introduce damage to the main girder.
4,Damageidentificationresults
4.1RemovingoneboltnearchannelS
The fiist level ofdamage was introduced to the bridge by removing
the first bolt from the top ofthe right stitfener (near chaniiel 5), as shown
in Fig. 2. TF[E is calculated at each measuring channel fi'om the
acc£Ieration time histDry data using MArlLAB Standa'd andMATLAB Signaa Piecessing rlbo1box'` i5). Hanning window ofsize
256 is applied to the time signals to minirnize leakage. ln this techniquq
the signaE (acceleration data) is divided into overlapping sections (5CYV6
overlap) of the spectaed winclow length (256) and windows eacli
section using the Hanning window function. In such case, the TFE can
bemeasuredat129bequencylinesinthefiequencyrangeof1-8ooI-lz
(fiequency step = 8oo"21256). IFE at channel 5 using the iesponse at
chainel8asaieference(71tDand'IlEatchannel8usingtheresponse
at chaniiel 5 as a reference (7hs) for the undamaged stmctuie aie sliown
in Fig. 3 (aj. The area between the two curves yepresents the reEative
TFE between chdimel 5 and 8 (Rs,D, as estimated from Eq, (3). The
absoltnc values of Rss are shown in Fig, 3 (b). Siinilarly for the
damaged strvcnm, 7Xs 7tz.s are shown in Fig. 4 (a) and the absolute
values ofR ss is shown in Fig. 4 (b). The change in the relanve TFE at
channel 5 due to the removai of one bo]t fi'om the sdffener near this
channel can be observed by comparing Figs, 3 (b) and 4 (b). The
absolute dffotence between these two figures (Eq. (4)) is showp in Hg.
5. In this figure, it is clearly indicated how the estimated change in R s,e
dqpends on the i}equency range. Foi' example, the estimated changes in
1ine are re1aively high in the fiequency ranges of501oo }-iz and 760
800 Hz compared to the changes at the fiecluency ranges of O-50 Hz
and 350-450 I+{z The fiequency ianges that show higher changes in Rs"
(better indication of ddinage) aie randomly disnibuted in the total
fiequency range from 1-800 I-Iz Therefore, it was decided to use TFE
data in the total measuTement range withont the need to identify the best
fiequency range in which TIFII has to be usod in the proposed
algorithm, The values ofl]ts?v si)own in Fig. 5 represent the 5th ivvv in
the 8th matrix [Ds] in Eq, (5). Every row in this matrix repi'esents the
changesestimatedatonechannelusingchannel8asarefeience.Every
new reference constructs a new matrix and thereby generates the 3d
matrix[D,g,"IIhefollowingstepiscenstiuctingthematrixofmaximum
changes [M,] (Eq, (8)) and the conesponding matrix [L,l (Eq, (9)) and
5(a) Undammged
eL/ge
afmvorE-
v-ut.
,gtse
ge
sgB2
oi
-,・A" Isf:'
-io- a"l
-d5・- -
1-20- -
.. .,e"
fti:
l/ :
-k
v-
'25 6i -- rt6o
- brd
+r W-
:
I..-J=-. 200
-1 - -s.
,.,t・-x!'ih
y
/ik
?-
:,," A
-L.V, 1
vAi, //iAidtv/9"V`{
d
tt-tbt / T
i
tt l- tnt t t tt300 400 500" Frequency(Hz)
{b) Undamagedri4;.-...TTm.M.mnt.1 ...1 1
''ifi2i---1---1- ・. {"- i ..
':'i'loL- i. ..-" I- ., l ,',,,BI--・・ ----.・-''r-J ,U'"
e, v'.6Lnfi..-.t," ".1-},.. i, i. ,i i !ll],・l'/l,
4 - }-- i.-.t.± .( ・;,'--;{ti',,tt. t.
2x?.",,,i., ,/12,f.,gllis.t,/r,./E,l",,s l・ f,・//f,ki-, t.
o
r・ 'i,,
t,ii
l
-. ./ ..J
'- TFE(S,B)
]----TFE(8,5)・
6oo 7eo 6oo
-: - l
.)
ll,・・
-1/
trE..-1.,i
-tl
fi/l
i'iil//1'tl'i
' '""'tf'tt"
'"";';"""""'
'i:-'"">t'-'
-i,,i 1,
・ll
l]
l
O 100 200 300 400 50D 600 700 800 Frequeney(Hz)
Fig.3TFEclataandtheassociatedre1aiveTl:Ebetwaenchannels5and8
forundamagedstructue
(a} IBoltRemovedatCharmel5 5,---r'--T---r '+ -mr'-"-- ' :"'' '1u ' i
::・s・,. ・l 1 n ,tlli ktn+ i, ,iii' tyY,,J;¥ 'ptAiA,tsvi y" vv, ifi f2
-isl-YIV- 'i"TiL' I , i-:-:-L:i--・f・dL-,,,is,j
lt .2oLt:i..1".:...i .l..i.rfi;-T.FEI...{s・s).il
o aDo 2oo 3oo 4oo seo 6oo 7oo BoD Freqvenay(Hz)
af ,4I(bl
mE a2;-
rr 16・} tol---
LlFlg 8i-'
/S' 6I-,
:,:ge 4--t/.,t.
sss j'`12{ x's}/,t
o
i
'tt':'{"t'
:,fa"ittt{"]・k"i
iJ・'i.・t
l・
!i,
'e
.l・
///.
il
tt "i'
・i"!
."lt1.'
ii'
//T・,i,,
l-'y'l{・,
1BoitRemovedatChamel5';"''fi'F"'L' '1''= : '
llI''i i- ,iti1.-.1 tu t.,t'I lt -tt U tt r- tt ni
t ,/-4tii "' ; 'M1' ll//1'i'11 'l''
'. 'x
/t.x' ;,f-,',・ )pmIStt"XiS{" f?/..{・-
li-. t,
l'`'njt'//' tl'
i
.i [ 1 il
'i-/E:i'i' `'- L g',・,, -
't・ ;11"//'iJ・
t t-'L t :・,/il・.L ・iEr/IE,/t・,':.
,,.7・1. ,;. ,xliltl.:,,//...it・L.:.ir,';.
I
;,
:
'1
1
lg-
11,L/l,,l..i・
,IILIIil
O 100 200 300 40D 5DO 600 700 BOO FreqL}eney(Hz)
Fig.4TFEdataaidtheassociatedrelativeTFEbetsmenchannels5and8
aflerremoving1boltnearchanltel5
then stimining up the rows of each 2d matrix to estimate the damage
indicatDrs {SM,,} and {SI.,.}, respectively. Figs. 6 (a) and (b) show the
tesulhag values of {SM,.} and {SL.}, iespoctive)y, At each reference
-19-
number, the estirrlaled values in each vectDr are drawn using waterfall
cuives. Although the values at the measuring channels are discrete, it
was decided to use cohtinuous 1ine to connect them instead of using
bars in order to enharioe the visualization ofthe results. In Fig. 6 (a), the
maximum reading is indicated accurately at ehannel 5 using various
reference channels (exccpt channel 5). However, the aDcuracEy of
detecting the damage at channel 5 depends slightiy on the usecl
reference. It should be noted that when one channe! is used as a
reference, it cannot be used to detect thg damage at its lcmation at the
same tirne. For ercample, the reariing at (thannel 7 at the refuence
*number 7 equal O. 'lhis is simply because Ita=R. = O (Eq. (3)) and
hence Q,, = O (Eq. (4)). ThereforE; when the channel near the darnage
Iocation is used as a refenmce, it will always produce ptse positive
readings at other channels. Thus, damage at one location can be
piedicted using at most (n-1) ieferences. {Sle} is used to estimate the
tota1 number of fiequency lines at which damagp is detectecL In this
study, the tomi ficquency range fiom 1-800 Hz is divided to 129 Iines
(see Fig. 5). As indicated in Fig. 6 (b), damage at channe} 5 is detected
at about 5e fiequencry lines oxit (}f 129 lines that were used tD estirriate
TFE data. This index is useful for indicating the confiderioe ofdetecting
darriage at a certain location, In Figs. 6 (a) and (b), although the
maxirnum ieading exists at channel 5 using various refenmceg the
ieadings at the Lmdamaged locations sometimes clegrade the aoguracy
oflocating damage. It was, therefbng ciccided to create new damage
localiza!ion indicators that can locate the damage more aecmeIy, The
proposed damage localization indicators piesented in Eqs. (12-14) aie
nomiali2ed to akow for better comparison, Furthermore, athreshold
ievel is definod to eliminate the ieadings at the undamagod locations
thatusuallyresultsfromthepiesenceofnoiseormeasurementemors.in
this study, the threShold level equds one. The values of {SrlTNI,},
{SMN,} and {SLN,} indices below the threshold level are related to
undarnaged cases and the values above (or eqtval to) the threshold level
identjfy a potentialy damaged elemerit The main task is then selec ting
anadequatethresholdlevelinordertodefinetheiealdamagedelements,
Thisthiesholdlevelcanbeconsideieclasadiscrimiriatinglevel.ifthis
acceptar!ce criterion is placed at a too high level (thresliold = 1.5), some
damages wM be unreveated. At a proper revel (threshold = 1), clear
discriminaion will result lf the acoeptance criterion is too low
(thTeshold= O.5), seve{al blse alarrns wM resute Theiefbre, an adecluate
level must be deteimined alowing clear diserimination in this study,
using a tiueshold levet equal 1,O for the test structure and its FEM has
yieldedthemostaccurateresults.However,thebestthresholdleverfbra
diderent structure in diffl)rent circumstances may diffhr depending on
many factors such as the type of the stiucture, level of' noise,
experimenmi variadong environmenmi changeg damage location and
darr,age size. in order to goneralize the thseshold !ove! apptoach based
on the proposed methed the e{foct of these famrs on more
sophisticated structuies needs to be studied, Using ari accurate FEM of
the strucrture to investigate sffveml Emenarios of damagg the effect of
noise and the effecrt of environmenul changes can be a useful tool to
detenrnine a good thieShold level. The resulting values 6f {S'IIN,},
{SMN,} and {SI-N,} for mis case of damage are shown in Figs. 6 (c),
-8Eeo`' 7=Eei.pq
T4Eg'ii,
o
1BottRemoveclatChame15
o loo 2oo 3eo 4oo seo 6oo 7oo Boo Frequenqy(HZ)
Fig.5'Iheabsolutedifirermiceinthere1aive'IIFI]mmdhannels5and8
atlerremoving1boltnearcinanrre15
(a) ChangeinTFEMwhiReferenceMethod
・- '1 A ...'i'''"l.il g ..,'t.F--"tl.t
i・, i:,:O?.<i{il:,tl.IIIIas)・gema'tJSA,es "tsg- kte, ,,
3`・ 5 if(f '4ReferenceNumber 2 (2 3 channetnumber
pa) ChangeinTFEMwhi・ReferenceMetbod
..T ..'r t'.. .1 J dg
E
o
(c} Changeinl"FEMuitiReferenceMethod .'11 lx g( xk. I,&s,;zx3t¥.
1 ChannelnumberFig.6Damageidendicationiesultsafierremoving1bo!tnearcihannel5
(d) and (e), iespectively. Damage is located very aocrmiy at channel
5 without any fa1se pc sitive readings at every refeience channel erccept
the reference nurnber 7, using damage indicatDr SIIN. Damage
indicator SMN shows better results where the darnage is located
-20-
(e> ChangeinTFEMultiReferenceMethod
''' t
i if3iiii'ssi,,
1 channetnumberFig7(Cont)Damageidenthcationiesultsafterremoving2bo1tsnear
chaniiel5
ChangeinTFEMultiRefereneeMethed 3500 3oooY-- J--L・ "-
l, 2500-- ---L ±g 2ooot-- -in-'r-n
la isooL-" L- -.
stli-
slt 1. J i'-,L ]d-.r:HmllT-rl/
[ 1- !・ L 1 . ,- 44l l li d
[lil i:,,llg:dg.:e-, l3l
--v---2Betts [de3Bolts l-4BLg!t!ES
tOOO"'-.."- ±50 : l
-"-' nt ;'-- l- M-
,ii :tlf
;1O-----".L.
12:
1
l i
:1
measurement errois, environmenmi or operational loads fiom the
changes attributed with darnage. Because ofmis need, the experiment
was peftbrrned five tirrres on the undarnaged strucJture prior to the
introduction of any darnage, Four diffbient combinations of TFE data
obmined fiDm the unclamagod stnic ture arre used to estimate the values
ofSSI'. Fbr examplq the fourtii set of' data is used to estimac R in Eq.
+(4) and the first set of data is used to estirnat£ R ,then the resulting
values ofSS;I' are piotted in Fig. 8 as indicated by the legond Undm 1.
Similarly, the remaining sets Undam 2, Undam 3 and Undam 4 are
estirnated and plotted in the sarne figure, The values of SST was
determined using TFE data in the fiequency rango of 1-8oo Hz The
tomi change in TFE ranged fii[)rn about 4oo to 600 dB. "Ihe estimated
changes in TFE aie obviously due to the presenoe of noise and
measuiement errors. The upper limit crf this mnge can be used・ as a
thresholdforthedamageindicatorSSI".ItisthenassumedthatifSST
ercceeds the threshold limig this will inclicate the oocurrence of damage.
In this study, the estirr}ated threshold is based only on changes due to
noise or measurement errors, However, in order to determine a more
praodcal threshold more data are noeded to aecount for the changes in
TFE causod by changes in temperature over different seasons or ffom
opeiaional loads. When the fiN lovel of darnago was introduced to the
bridge, the values ofSSI' incteased at most ofchannels to aiound lOOO
dB and increased at the damErge lecalion to aiound 15oo dB, exceeding
the thTeshold limit at ak channels. The increase in SST at the
undamaged locations is due to the fact that darTiage at one location
changes the iesponse, and hence Tl E data, significantly at the close
seiisors locations and sligluly at the more far sensors.
Since a serious damage to a structure is usually the iesult of the
grovvth of Iess serious damage, it is important to haye the abMty to
measure the growth in damage. We need to be able to monitor this
growth fbr the purpose of bridge maintenaiice, The iesulting darnage
indicator values of the darriage indicator SSI' fbr four levels of actual
damage - removing one to four bolts aie shown Fig. 8 and indicated by
the legends 1 Bolt thiough 4 Bolts, respectively. It is clearly indicated in
this figtire how the values crf' SST increase with inciease in the damage
level. Unfbrtunately, the damage severity carmet be ideruified
quantitatively, Howeve4 for the same darnage locations but difR:rent
levels ofdarr}age compaTed with Fig, 8, the amplitude levels are higlier
L--L-L---"-Lm-
345678 Channel Number
Fig,8Monitoringthegrowthinclamagenea!'channel5
for the cases of moie severe damage, which can represent the darnage
seventytosomeext£nt
4.4Multipledamage
MoststudiesconcemingcrackdetectiondealwithasinglecrackThe
case of multiple cracks has not received the same degree of attemion.
The problem of detection of iocation of a number cif fl]mlts in a
component sitnultaneously is much moie involved and complex than
thecaseofasinglecracklnthepresentdamagecase,eclualarnourrtsof
darnage were introduoed to the two sddeners 1ocated on the wel) crfthe
main girder (Fig. 2). Channel 5 is locatecl l5 cm from the cerrterfine of
the first darnaged siffener while channel 3, the nearest sensor to the
second darr}age locatio4 is located 53 cm tfom the centerline of the
second damaged stiiiflener. As explained previously, the maxirnum
change at each frequency line will be selectecl (the maximum value in
each column in Eq. (5)) whiCh indicates damage at this node, Therefore,
damage is usually detected at only one location at each i}equency line
unless the same maximum value exists at more than one node which
rarely oocurs, Howeveq the maxinrium change at another fiequency
line may indicate the damage at the second location Moieoveg using
multiple refeience channels can be usefUl in this case; as one reference
may give aocuiate results in detegting the darTiage at one location while
another reference mEty detect the damage at the second location. This
underrines the importance of using al1 the fiequency lines in the tctal
measurement rango as well as using each measuring channel as a
ieibience. The first level ofdamage is introduced by removing one bolt
ffom the top of each sitiEl:nen The predicted results using difierent
ddinage indicators are shown in Fig. 9. Damage indicators SM and SL
show the predicted iesults without normalizing the data or using
threshold as in case of using STN, SMN and SLN. It is clearly
indicated in Fig, 9 hgw the selection ofthe referenoe channel atllects on
the results. in case of using damago indicanrs S'IIN, SMN and SM,
darriage indicator SEN showed the most aocurate results while STN
showed the least accurate one. The growth in damage could also be
monitoieci accurately using daTnage indicator SSI', as skown in Fig. 1O.
rlhe darriage level gFaduaky incieased at the two locations by removing
moie bo1ts fivm the two stilreners. SST values at channel 5 aie higher
than its values at channel 3 for most of the daTnage levels because
channel5isclosertothedamagelocation.
-21-
accurately using al the reil)mmces. Damage indicatpr SI.N has shown
the Ieast aoctrate results and some false positive readings arrpeaied at
channel 4 at references number 3, 6 and 8. If damage lcx)ation is
piedigted at one channel using ad the refererioes (excqpt the same
channel), theri the identified clamage locations using this channei as a
referunoe should be igr)oied
42RemovingtvvoboksnearchannelS
The amourrt of damage inereased in this case by retnoving one more
bolt fiom the top crf' the riglrt stifflener (Fig. ). Damago identification
results are shown in Fig. 7. Similar results aie obtained in this case
compared to the iesults of the previous case (removing one bo10 and
canbesummarizedasfoHows:
- DamageindicatorSMindicatedthedarnagelecationatchannet5
more c;early than the previous case of darriage as the indicator
values at channel 5 inc eased and the values at the undamaged
locations decreased.
- The number of fiequency lines at whch damage is detected
accurately at channel 5 increased signblcantiy in this case as
determined by the clamage indicator SL Damago is detec ted at the
correct location at approxirriately 1oo fiequency lines out of 1op
fieclLiericy liries (Fig. 7 fo)), achieving one of the most important
obje[tives of arry damage icientification algorithms which is to
asoertainwithccmlidenceifdamageispmsentornotL
- Normalized damage indimmr SIN, SMN and SITkl identified
damage location at channel 5 using ak the refeiences without arry
blse poshive Teadings (Figs, 7 (oj, (d), and (e)).
- Damage location is ideritified accurately using various damago
indicatorsregardlessfffthelocationoftherefeiencechannel.
<d) ChangeinTFEMwhiReferenceMethod .. r '1-ls li Iiiec,K,,., ,,iiglSII.lipt'".'lii,il -
(e> ChangeinTFEMuttiReterenceMethod
.・lkk
igxll(keg/iX
S ChsTmelmmber Fig6(Cont)Darnagoidenthcationiesultsafierremoving1boltnear
cliannel5
(a)
g il IiD:,, J, ..i.il/lmamuS
5x... ,. ,.....,.....,....
4 x./, .... ..
' 3 '; .-'". '.ReterenceNumber 2 ;tYt"i 3
ChangeinTFEMwhiRefefenceMethed
'''te.I
.-:eFI1li.-Ji" :i'
・tt-fill
.,.pte-
esewrv5 #',
4 ChaTmeirHJmbeT
i・l' 1 "1
l 1"
ll il[i [I .,v}wS・"' 1
"" B 7
(b> ChangeinTFEMwhiReterenceMethed
"'tdi ...t.・T-
ll:'II:.:l
i k/¥k'lf.1#,au.ss-
{c) ChangeinTFEMultiReiereneeMethod .. i' 'l
1 channelnumber
{d} ChangeinTFEMultiReferenceMethod
" "ix
ik.,,l. iWXN
L
'
I
'8
Fig7DEur}ageidmdicationresultsaftei'mmoving2bo1tsftearchannel5
43MonitoringthegrowthindamageatchannelS
As discussed previously, darriage indicatDr SST (Eq. (7)) wru be
usod to detect the occurrerioe of darriagst and monitor the increase in
darnage. TheTeilore, it is important to define a threshold of the toml
chango in TIFE that classhies the changes in TFE due to noise,
nm22pt-
ffvv
:
.8v--c
ge8
200
too
o 8
Reierenae Number
(a) ehangeinTFEMu#iReferenceMetbod
' '''' t t; -1 /,.,." I',,
i........l..il::,t//,.l..)..,t . t
7ny,liEilpa,7/,ig.sge£igee pa/.eqs,ggi,i, ,
2 .,..glA'"""3 4 112 channeirtumber
8
{b) ChangeinTFEMuvaReferenceMetbod
nt '" t " pl1,] co .-', 'I ii$Ool¥ili/1:es.#.iixei3zif,, tr#,
6, gi;?l.urig';'l///,・:,'tS}'IT'Xk) :,.,,,t/rit],/;.l.ii2.,,.,.,,//i';l÷l・'i,・/il"isi/,.・'・l;・t",fs・l
ReterenceNumber432rkl;l ....{.l.,il/1Iiiil'gii.:?Stu'l'l;{U"g6"k7 s
112 charuvelnumber
{c) ChangeinTFEMuttiReferenceMethod
' ' '1 t"'
l/ ium Yx
1-L 2 1 Chanrtelmmber (d) ChangeinTFEMuTtlReferenceMethod
'b ii .,li,Iititiicirt$'
1 channelnumberFig9Damageiderimacationiesultsafierrcmovinglboltnearchannels3
artd5
(e) ChangeinTFEMultiReferenceMethod
i = ''i co .t./l !t t ts4・1 ・・, i'"'.i Jeeeeqigiii ,iitn,
Fige9(C(mL)Darnageidenthcationiesultsaheriemoving1boltnear
channels3and5
400o ttH+ 1350o L -
l3000 ・-
2500V・ +
ChangeinTFEMultiRefereneeMethoci
gg
t' ! --------ptt.,2001- ---T-- .-F-uJi
:b 11500--!--・r---t---1--- - ;aooot--・l---1'L.- -l u.;n.-[ L.1...
1lIl1 11 l50
Tl--[/---l.-.m
a t
r"1-N-1-''
'
±
t-
T
I
l
4-
1
t- --.-:- -:- l 1
t- Lt
--tt"-"ttt lt---Ttttt tt tifi lig::g::Ii tl -・it"- Undam 3・t tlin-e.--undam4
, ・l-IBolt :.i. .""2Bolts ' S]-3Bolts tt":-:"k'--,'-xJ4Bofts I d-tLt--t ttnu tt '
;}
5.Numericaldata
The finite elemem model of the bridgo is cieated using Structural
Analysis Piogtam, SAP2(X)OiG). Main gipdems and cross bearns are
simuiated by shell elements (Fig, 11), The FE model contains 1878
e
-t
IJIilt oL.--.L.-.=-r..in--..i.--j." "
rt 2345678 Channel Number
Fig,1OMonitoringthegrovvthindamagenearctiaimels3and5
rl . ca)t t 't .w ' ,ut.Eiglua pfi・.'es' .de,,srkfwh .t"l'. tnk,fsli'"av
rd.n're' ・:..,nt'',,t'"'"mant.,,rtth di See"'ff. .,,,,.:Es cak,em.i."...mag . .,
'sii#. ik":lj,}sfii}ee, ee ge
kes.,・:..ve .. si ,,..
n ttt l' w.E di'
' 't
Fig,l1Numer:icalmodeloftliebridge
tt di
'2'
yme" un- ..
;y-ff v
v
-swl ees
lttt t
shell elements and 10508 active degrees of fieedom, The weight
densltyofstee1isassumedtobe77kixlhn3ar}dthemodulusofelasticity
of steel is assumed to be 206 GPa Acoeleiation response in the
horizonim direction is measured at eight nodes (N1:N8), as showii in
Fig. 12. 20-second time histories weie colleeted at O.OO0625 second
incretnents, pioducing 32eOO time points to simulate the experimental
data The main objoutive of using the numetical dala is to assess
diffeienteffectsratherthannoiseormeasurementemoisontheaecuiap>t
of the obtained results from the studied damage identification mediod,
Moreoverl the numerical model caii be used to evaluate tiie
performanceofthedamageideruificationmethodtodetectdifferent
-23-
(,}
iulr-----・-L..
Units: mmMeasur!ng pomt
I,L-ff] .1 'nE-. t.-.
57SO
Imae #.-
''t'" e7--'
iT-----・-・・.tr.t..
.S'.H--..
s1''""- J--- D2 D3 i,e DS I
L.-.-s --
1 Nl g N2'
N: 1N4 ' : 'iNs N6' ' }--
N7NB si]J - r
ll
P..a,m.a. g.e locations
J.1350uaJ!gE9.L1200 loSo 11so 13so 5750
19800
Fig.12[IhelocationofmeasuTingpointsanddamagedareasinthenumericalmocic1
(a) ChangeinTFEMuhiReferenceMethotitypes of damage at dfferent locations. Three cases of darnage are
introduoed to the numedcal model by reducing the thickness of one
shell element by 1(Y),6 (1.1 mm). Damaged locations are indicated by
the shadowed ateas D2, D3 and D5 (Fig 12). Mode shapes in the
fiequencyrangeof1-140HzcouldbeidentiiiedusingSAP2000,TFE
in this frequency range is estirriated using the same window size of256
and the same sampling rate of1600, ptoducing 22 fiecpiency lines. The
predicted results using darnage indimmms SL and SLIF(l fbr the thiee
damage locations D2, D3 and D5 are shown in Figs. 13, 14 and 15,
respec)tively. For damage locations D2 and D3, damage is detegted at
the nearest measuring point at about 20 fiequency lines 6nt of 22 Iines
using ab the referenc£s. On the ether hand damage location D5 is
derm at 1ess fiequency Iines and some false positive readings are
predic ted (using SL only) thougli the numerical data doe$ not contain
any noise or measurement errors. The main iwason involved here is that
at each frequency line, the clynamic response at dilferent measuimg
1ocations represents one opeiaional mode shape. lfthis mode contains a
node (i,e, zerD ampEtude) at the damage location, this mE"i reduce the
chanceofpredictingdamageatthislocatlonandhenceproduceatalse
positive reading. It is, again, iecommended that TFE be used in the tomi
measurementrangetoieducethee{fectofundesiredfiequency1ines.
6. Comparing the proposed method with a proviously reported
method
In this seedog the proposed method wM be compare[l to a
previously mpoited rnethodM by the author, The previous method is
based on changes in Power Spo[ tral I)ensity (PSD) caused by darriage.
For more details abont this mettiod, the reader is refetTed to the cited
ieference. The main differences between the two mediods can be
summarized as follows:
- The previous method estirnates PSD magnimdes from the
accelera!ion clata at dfferent measuring channels. Theri use the
changes in PSD rnagnitude to identjfy darriagp. The proposed
method uses the re1ahve TFE between two measuring points to
identify damage.
- Different statistjcal proceduies aie used in the two methods to
identify the damage.
- PSD method shows damage identilication results at each
measuring channel using various damage lcKalization irKlicamrs.
The proposed method shows damage idemification results at each
channel versus the reference channel that is usecl to calculate TFE
data
. .. ." di .--・-]-J[:-:---sl}, IGIo:,i,tdeel ty,it,i
' 5 ''i 4XRefererKeNumber 3' 2 fi.ee,,pt.gs'""''4'' s 6
1t2 chamelru"nber
x
X>N
"X>N
XI N>N
.> N]
Xl
78
{b) ChangeinTFEMultiReferenceMethod
.s .nl , i ..irlll/.:'Nii,x"
i[pt ee5giti Sx&
1 channelnumberFig.13DamageidenthcationresuksfordamagelocationD2using
nurnedcal clata
Qg 2o
i 10foif 8
8 7
(a)
6N5
ReterenceNumber
x
-es
ChangeinTFEMwhlReferenceMetltod
t"e} .'1 N ..'1 -・.Nh-'
,:J;il-r'"i <l,s,
t NIN N ..a' ・lt,S., 1. .t ,e ,tinv N
tt tt"t N ," ,ast.
N43t.evN gelv ,,ee"
2' Nes 1i '2
'xFJ N ''"N> N ' XN N Nts N,. Sx pttwtxNF)1,
N ・lt .t.,,. N 'A';. ,N xl
as.'ESs'.4's'6 '7 8
3 ChaTmelnurnber
Fig.14DarnageicleritificationiesultsfoirdamagelocationD3using
nurneical data
ThetwomethcKlsaresimilarinthefollowingrespects:
- Both methcKis use the structure's response without the noed to
measuretheexcitationforoe.
- Both methods can use the structure response in the fiequency
domain in the tota measured frequency range.
-24-
(b) ChangeinTFEMultiReferenceMethod
" f' ., l' .'e )l i '"'1'''lj'li't' ii 'ig!eq"ilsi..:ag"" ': ,
Fig.14(Cont)DarnageideniificationresultsfordamagelocationD3using
numerical data
(a) ChangeinTFEMulfiReierenceMetbod
tt 1 v)I. : t. 1 ..ti-r'].F-?k:,., l16 iO:7, Sliig:iiillll`,fi$wa.geiim'iil,lilil ],i.,ik;" ',
,,.,,.,,,,,,,1`32r:!:?i, i'tt/esk,llii(iiiiiliiilif..,6,,utY""B
<b> ChangeinTFEMultiReterenceMethoci {・ 1,1 T-
gi .twpteqgits,
d ChannelnurnbeT Fig,1SDamageidaidicationresultsfhr'darr)agelocationD5using
numedcal data
Fig. 16 Shcrws PSD method iesults after mmoving one bolt fiom the
stifilener near channel 5 using damage lcrcalization indicator 1. The vase
positive reading at channel 8 may degrade the aocuiacy of identifying
damage location. The same remark is also observed for the case of
double clamago after removing one bolt fiom two sdfft)ners near
channels 3 and 5, as shown in Fig, I7, When PSD is applied after
mmoving two bolts fiem the sdffeneis near channels 3 and 5, the
predicted iesults at ciiannel S is much bigger than that at channel 3
whichieducetheconfidenceofpiedigtingdamageatthislocationGig,
18). The corresponding tesults of this case using damage localization
indicator SMN aie shown in Fig. 19, The previous drawbacks ofPSD
method are avoided in the proposed method by showing the results at
ChangeinPSDMethod 3000
r2500------------・--- -・ -------, g2eoor-- ---r-'''-'" ' ' -[ g Rlsoo`''"n'J Tr-・ -' - ., g'
g"::o,l:.1:IlJ:ILlI [[ ''.'-.I
oL'-.m.-.".-Trrr-."--nv-.- -..r-u,.,.M.-...l/
1234 5678 ChamelNumber
Fig,16Damageidenthcationresuitsafterremoving1boltnearchalinel5
usingPSDmethod
2500T------・--7
2oeoY------- gi ll g 15ool----T- ." ts ・ gioool・・・---''-
ig seol-- ・- '---
' i otmo-.gem l2
Fig. I7 Dannage id
rge
figg
ChangeinPSDMethod-'T-'
-imm- 4 ChartielNumber
-ee--- -・
678-ee --
ll't
l
j
--l
.1
entificationiesultsafterremoving1boltnearchenmels3
and5usingPSDmethod
sooo.--・・.h++.T.-9.if9.m"g.P±"-P-SD..nyfeLthoHd.
:4sool・-...... .. .. i4oool '' "' i3500:-・ ----------- ・- '3000b-・・・・--- -----. ・.. i2500p- .... .. -.m . -. sf:::llI J11JIlIJi1I .'"
lioooFT-U'--rr""-''n" 'seoL- :"T.-.- oL-r-].N-Hee.I.JI -..-..fi
A23456 ChametNurnber
.1.t' '-
.J.." . ・L
78
--i,
i'i
I
-I
l
-l
i'
--l--l
Fig.18Damageiderrtificationiesultsafierrerriaving2bo1tsrbeearchannels3
atrd5usingPSDmethod
Change in TFE Multi Reference Method i' co ..It -l ・ L ., S`i,g :¥lk'ii.i'ag".,..
32 -,e"....g. .,-・S''6 7
ReferenceNumber 1'・'"23 t
Fig.19Dariiageid
ard5msingtireimoposedmethod
b
:j
8
¢hannel numberentificationiesultsaferremoving2boltsnearchannels3
-25-
each location versus the reference channel that is used to estimate the
TFE dan ln this casq the piedic ted darnage location may be deteeted
(ta1) times which inciease the confidence of detecting damage at
vaiioms locations.
7, Conclusions
'lhis paper piesents a novelty detectioi? technique fbr stnictuial
darnage diagnosis using r[FIi data TFE is calculated from the
acceleration response at every channel reladve to a referenee channel
and every channel is usod as a refuence to other charmels to create a
large number of infotmation. Moreover, TFE rmagnitude at each
fiequency value is used in the proposed method. rlhis accumulated
information is then used for staistical procedures to identify the damage
location with high confiderice. The propos"ecl method encompasses the
first three steps of the process of darnage detection - existeneq
localization and monitoring the growth in darnage based only on the
measuied data without the need for any rncKlal ideritification or
numerical models. The technique pnesentod here may adow some
progressinin-servicemonitoringofstee1bridgeswheretheacceleration
data can be transfleired via wireless methocls and the piezDelecnic
agtuators can be usecl fbr local excitation. "Ihe proposed approach is
demonstrated using experirnenul and numerical data obmined fiom a
railway stee1 bridge. The new method shows very high accuracy in
prediedng damage location and monitoring grewth in damage. The
high detection performance, combined with the sirnple computation
structure and the easy implementation could lead to a promising
real-time damage detftction system
Acknowiedgement
This research is strppoitecl by the Grant-inidLids for Sciemific
Researclt IVlinistry of Education. The authors wish to er<press their
gratitudefbrthissupporkSpecialthanksareowedtoDr.Yamazaki,Mr.
Tsubota and Mr. Ol(Ltyama (Kitami lnstitute of Teclinology) fbr the
plppalationoftheexperimenmisomp.
References
1) Doeblmg S. W., C. R. Farrar, M B. Ptimq and D. W. Shevit2
Dan2czge lalentifcation cmd Hlaalth Mbniton'ug ofSimtctzazTl and
Adlecimical Ebtstems .17om enatzges in thetr va'bration
Choracteriyties, A Literattue Reviexv, Los Alamos National
Laboratory Repo4 LA-13070- MS, 1996.
2) FarTar C, R et al., llb-zamic Chnracteitzation and Damcrge
Detectt'on in the L40 Bhagel ber the Rio Gunal,, A Literatute
Review, Los Alamos Naional Laboratory Report, L.A-12767- MS,
1994.3) Farrar C. R. and D. A Jauregui, Damc{ge Detection Aigorithnts
oplied to ibcperimental and Nztmerical Mbdel Datafrom the P40
BriLige, bos Alamos Nadonal Labormry Repeg LA-12979-MS,
1op6.
4) Sampaio R P. C., Maia N. M M. and Silva J. M M, Damage
detection using the fiecluency-rcsponse-function curvature mediod,
JOto72cti(2flSOunciandVib,zuto4226(5),pp.1on9-1042,1ep9.
5) Peetefs B., Maeck j. and De Roeck G., Vibration-based darnage
detection in civil engineering: excitaion sources and temperature
effects,5h7tzrtMtneriirzLsandSh'z`cttoies,IO,pp.51&527,2oo1.
6) S.T. Xue, N. Fajhani, Z.B. Wei and IIS, Tang Comparison of
var:iaions ofnatural frequencies for wooden stiuctmal models with
and without damages part 2, shaking tal)le based ms
Proceedi',zgs ofthe intemational co}zlarence on Simctzaul Hlealth
IT 1 Mbniton'ng and helligent mpasnacrtme, China, pp 1101-1I04,
2006.
7) J. C. S, Ya!ig J. Cheri and N. G. Dagalas Damage detection in
otshore stnsctums by the random deeiement technique, Jbuinal of
l!hei:gy Resote'ees 71?chnolqgr,; Ameiican Society of Mechanical
Engheers 1os, 38-42, 1984.
8) R G FIesch and K Kemichler, Bridge inspection by dynamictests
and calculaions dynamic investigations of Lavent bridge,
woJikshop on 5b`trctanti ,Skyety Et,ahaitibn Btzsed on ,Sl;tytem
lalantpt ation mprcxurhes (H. G. Natke and J. T. P. Yao, editors),
433459, Lambrechtf Pfal4 Germairy: Vieweg & Sons, 1988,
9) S. F. Masd R K. Miller; A F. Saud and T. K Caughe>i,
Identilication of nonlinear vibraing strums Joumal offmplied
A`lechcmics54923-929:PartI-forrnulation1987.
1O) H G. Natke and J. T, P. Yao, System identification methods for
fault detection and diagnosis intematioual Coizlerence on
Sb'uettazzl &gZ,ty and Reliability, American Soeiety of Civil
Enghws New Yori<, 1387-1393, 1990,
1l) Oshima T., Yamazaki T., Onishi K. and Mikami S,, Stucly on
damage evaluation oijoint in stee1 membor by using local vibraion
excitatiog (In JEg)anese), Jozenal c2f'mplied Mlachanics J/SCI4
Vol.5,pp,837-gn6,2oo2.
I2) Beskhyroun S., Oshima T., Mikami S., Yamazaki T,, Damage
detection and 1wh1ization on strvgtura1 connec tions using vibration
based damage jdentilication methoda .lotornd of Aicptied
A4i3ehanics, .(SCa Vol.6, pp. 1055-1064, 2oo3.
13) Beskliyroun S., rVlikarni S., Oshirna T, and Yamazaki T., Modified
damage identification algorithrn based on vibration measurements,
.lozfi'nal (2L4lpplied Adechanics .ISCEI Vol.7, pp. 97-107, 2004.
14) MATLAB Rtltarence (htide, The Math Woss Inc., NaicK MA,
2oo3.
15) MATLAB Clser's ([}ttidb, The Math Works, lnc., Nadcks MA,
2oo3,
1ol SAP2000, lhtegrated FVnite Nement cznd Dest'gn ofSbrttctzcres,
Analysis RefeTence Manual, Computers and Structuies, lnc,
Betkeley, CA, USA, 1op5.
17) Beskhyroun S., Mikarni S., OsliimaT. and Tsubota Y,, Structural
damage idcmtification algorithm based on changes in power
spectral density, Jbumcal ofAFzplied Mlachanies, Jsw Vol,8, pp.
73-84,2oo5,
(Received: Apdi 13, 2006)
-26-