Introduction. Keywords: decay, delamination, glulam, moment analysis, wave propagation, wavelet transform

Transcription

1 Holzforschung 7; op Feng Xu, Xiping Wng*, Mrko Teder nd Yunfei Liu* Acoustic impct testing nd wveform nlysis for dmge detection in glued lminted timer DOI./hf--7 Received Decemer, ; ccepted My, 7; previously pulished online xx Astrct: Delmintion nd decy re common structurl defects in old glued lminted timer (glulm) uildings, which, if left undetected, could cuse severe structurl dmge. This pper presents new dmge detection method for glulm inspection sed on moment nlysis nd wvelet trnsform (WT) of impct coustic signls. Acoustic signls were collected from glulm rch section removed from service through impct testing t vrious loctions. The presence nd positions of internl defects were preliminrily determined y pplying time centroid nd frequency centroid of the first moment. Acoustic signls were then decomposed y wvelet pcket trnsform (WPT) nd the energy of the su-nds ws clculted s chrcteristics of the response signls. The su-nds of 7 Hz nd 7 7 Hz were identified s the most discrimintive fetures tht re ssocited with decy nd delmintion nd therefore re indictive of the presence of delmintion or decy defects. A defect dignosis lgorithm ws tested for its ility to identify internl decy nd delmintion in glulm. The results show tht depth of delmintion in glulm memer cn e determined with resonle ccurcy. Keywords: decy, delmintion, glulm, moment nlysis, wve propgtion, wvelet trnsform Introduction Glued lminted timer (glulm) is widespred ll over the world nd extend modern wooden uilding cpilities. In glulm, timer lmintes re stcked nd glued together nd re le to crry more lods, for instnce, in *Corresponding uthors: Dr. Xiping Wng, USDA Forest Products Lortory, Mdison, WI, USA, e-mil: xwng@fs.fed.us; nd Dr. Yunfei Liu, College of Informtion Science nd Technology, Nnjing Forestry University, Nnjing 7, Chin, e-mil: lyf@njfu.com.cn Feng Xu: Nnjing Forestry University, Nnjing, Chin Mrko Teder: Estonin University of Life Sciences, Trtu, Estoni lrge-spn roofs. Glulm constructions re pprecited for their strength, durility nd verstility. Modern qulity ssessment methods re needed to ssure or extend the service life of glulm structures, especilly in the context of dhesive onding filures cusing delmintion in glulm elements. Delmintion usully results from improper onding, overloding during the service life of structure or y moisture chnges tht introduce stresses (Angst nd Mlo ; Bucur ; Neuenschwnder et l. ). Similr to solid-swn timers, decy cn lso occur in glulm memers s result of wter intrusion nd poor ventiltion. Dmge from delmintion nd decy is usully the result of slow processes nd ctstrophic filure cn e voided y regulr inspections nd erly detection. However, these types of deteriortions my tke plce hidden from view nd remin undetected. There hve een mny incidences where glulm structures filed nd cused significnt property losses nd endngered pulic sfety (Hnsson nd Lrsen ; Frühwld et l. 7; Divos ; Sitr ). In current prctice, structurl condition ssessment still lrgely relies on visul surfce inspection, which provides only limited informtion on the internl stte concerning the onding performnce of glulm memers (Ross et l. ; Ksl ). Stress wve timing mesurement hs proved to e effective in predicting the stiffness of structurl memers nd detecting moderte to severe internl rot or crcks in oth solid-swn timer nd glulm memers; however, it hs limited success in identifying erly stges of deteriortion nd locl onding filures (Ross et l. ; Senlik et l. ; Arrig et l. ; Teder nd Wng ; Wng et l. ). Methods of ultrsonic wve propgtion (perpendiculr to the grin) hve een pplied to glulm for ssessing the ond qulity nd resulted in etter results in terms of resolution nd sensitivity (Mev et l. ; Dill-Lnger et l. ; Birks et l. 7; Divos ; Snri et l. ; Xu nd Liu ). However, the pprtus nd operting procedures for ultrsonic inspection in lrge glulm memers re constrined y the low level of the energy source nd the coupling issues often encountered in the field (Bell ; Dill-Lnger et l. ; Senlik et l. ; Bucur ; Snri et l. ; Neuenschwnder et l. ). More dvnced methods need to e explored

2 F. Xu et l.: Impct testing for dmge detection in glulm nd developed for the field inspection of glulm components in modern uildings. Severl non-destructive testing techniques (stress wve timing, ultrsonic trnsmission, resistnce microdrilling) were lredy tested to ssess the physicl conditions of glulm rches removed from service uilding (Teder nd Wng ). The ojective of the present study ws to explore the possiilities nd limittions of n coustic impct testing method coupled with two dvnced wveform nlysis techniques, i.e. moment nlysis nd wvelet trnsform, to detect delmintion nd internl decy in glulm rches. Bckgrounds Moment nlysis is one of the most pplied signl processing methods y which mny signl fetures cn e otined. In generl, when signl is excited nd trnsmitted in flwless medium, most of its energy is typiclly locted in the eginning of the signl wveform. Reflection nd mode chnges of signl often come from the oundries nd mteril flws, cusing the skewing of the signl in the time domin. Therefore, moment nlysis should e le to chrcterize the dmge of mteril. Tiitt et l. (99) pplied time centroid (TC) nd frequency centroid (FC) ( form of the first moment) of the cousto-ultrsonic signls in lortory experiments s inputs of the neurl network clssifier for decy detection in rown-rotted Dougls fir glulm. A 79% efficiency ws found for detecting smll levels of decy nd % efficiency for the overll ccurte clssifiction of different degrees of decy. Byiss et l. () pplied dmge detection technique sed on the sttisticl moments of the energy density function of the virtion responses in the time-scle domin. The zeroth-order moment, known s the totl energy of the joint density function, ws computed t ech mesurement grid point for the pre-dmge nd post-dmge sttes nd pplied to locting structurl dmge in concrete plte model nd seprtely in steel plte girder of ridge. The roustness of the proposed dmge identifiction for rel-world pplictions hs een verified y mens of experimentl modl dt from full-scle ridge structure sujected to vrious dmge condition sttes nd rndom virtion excittions. The wvelet trnsform (WT) of signl is nother useful tool for time-frequency domin nlysis, which chrcterizes time-frequency vritions of the signl sund components. The WT pproch is suitle for nlyzing non-sttionry signls with trnsient nd time-vrying chrcteristics, in the course of which oth time nd frequency domins cn e chrcterized simultneously. In the menwhile, this technique hs een widely implemented for vrious unique purposes, such s de-noising of signls, informtion retrievl from noise-polluted signls, clssifiction nd pttern recognition pplictions (Mllt 999; Brenner ; Byiss et l. ), chrcteriztion of non-sttionry dynmic responses (Bsu nd Gupt 997; Xu nd Liu, ) nd identifiction of nonliner structurl dynmic systems (Kitd 99). The WT ppliction for dmge identifiction nd helth monitoring hs lso een reported. Liu et l. () presented crck detection method for fier-reinforced composite ems sed on continuous WT (CWT); the uthors found tht the crck loction could e determined successfully nd the crck depth could e estimted from the crck-reflection rtio. Ruck () pplied WTsed dmge detection technique on cntilever em, where the dmge ws present in the form of single notch t depths of, nd % of the em height. The conclusion ws tht the smllest detectle defect ws t % depth of the em height, nd defect loction cn only e determined y higher virtion modes. Umesh et l. (9) proposed n nlyticl method for locting nd quntifying the dmge in fixed ems with single crck y WT nlysis, nd demonstrted tht the method could locte nd quntify the dmge in ems or emlike structures without prior knowledge of the deflection corresponding to the undmged structure. Although the quoted ppers reported stisfctory results, they were mostly focused on numericl nlysis of idel models nd experimentl nlysis of isotropic mterils with idel nd single flws. Very little informtion is ville for the ppliction of WT nlysis in glulm structure inspection. The chllenge is tht the response signls in glulm memers re more complicted thn those in idel experimentl mterils. Mterils nd methods The glulm rch memer investigted here ws slvged from reserch uilding t the USDA Forest Products Lortory (FPL) tht ws deconstructed in the fll of, fter more thn 7 yers of service. Originlly uilt s test nd storge fcility, the uilding is considered to e the second oldest glulm structure in the United Sttes (US) (Rmmer et l. ). In Figure, I is n inside view of the uilding showing the solid glulm rches t the middle ys. The uilding ws m wide,.7 m high t the wlls nd. m high t the center. The hlves of the glulm rches were ent t the knee ut were stright elow nd ove the knee nd were held together t the pex with olted pltes. The rches in this structure represent

3 F. Xu et l.: Impct testing for dmge detection in glulm I decy nd delmintion due to moisture intrusion nd exposure to elevted tempertures in the 99 fire incident. A.7-m long section (mesured from the se nd long the inner fce of the rch) ws cut from this rch memer s specimen for n extensive coustic impct testing. II III B C D E F G H I J K the first genertion of oth construction dhesives nd glulm development in the US nd, therefore, the removed glulm rches were preserved s reserch specimens for vrious groups t FPL to evlute the durility nd residul strength of these ged rch memers (Teder nd Wng ; Rmmer et l. ). The solid glulm rches of rectngulr cross section were glued nd stcked with lmintions of.-cm mteril to constnt width of 9. cm, tpering from mximum depth of cm t the knee, ner the junction of wll nd roof, to lesser depths t the foundtion (. cm) nd the roof pek (. cm). Ech lyer ws mde of.-cm nd.-cm wide southern yellow pine lumer, with the edge joints stggered in lternte lmintions (Teesdle 9). The generl physicl conditions of ll the solid glulm rch memers removed from the uilding were determined through preliminry stress wve nd ultrsonic timing inspection (Teder nd Wng ). Glulm rch (Figure, II) contined wide rnge of I H H G G F F E E D D C C B B A I J Impct testing loction K J K L M N L M P Impct testing position c P L M N Figure : Glulm rch memer removed from service ws investigted y coustic impct testing. I Building, Forest Products Lortory, Mdison, Wisconsin, c. 9, considered to e the second oldest glulm structure in the US. The solid glulm rches re shown in the middle y. II Acoustic impct testing on glulm rch memer. () Photo of the glulm rch section cut from rch ; () pln view showing impct testing loctions long the glulm length (mesured long the inner fce of the rch); (c) cross-sectionl view showing three testing positions (P, nd ) long the width. III The glulm locks fter crosscutting t ech testing loction. Block A (not shown) ws severely decyed nd delminted nd fell into pieces following cutting. 7 9 Acoustic impct testing: Figure, II nd IIc show the loctions (long the length) nd positions (cross the width) where coustic signls were collected. The physicl condition of the rch section vried lengthwise; thus, the impct testing ws strted t the cross section cm from the decyed end (se), followed y -cm increment nd then y -cm increment. At ech of these loctions, coustic signls were collected t three positions, which were designted s P, nd (Figure, IIc). Positions P nd were ech cm wy from the edge nd position ws locted in the middle of the width. The impct tests were conducted through the depth of the glulm (perpendiculr to the lmintions), while the rch section ws plced in flt (Figure, II). Sensor proes (Fkopp Microsecond Meter sensors, Fkopp Enterprise, Ágflv, Hunry) were inserted into opposing fces of the glulm t the sme cross section nd sme position level. Acoustic wves were generted through hmmer impct. The response signls were collected vi dt cquisition crd NI connected to lptop, with smpling frequency of khz. Upon completion of coustic impct testing, the glulm specimen ws dissected t ech testing loction, resulting in glulm locks leled A N (Figure, III). Block A ws severely decyed nd delminted nd fell into pieces following cutting. Visul inspection ws conducted t ech opened cross section to identify decy, delmintion nd ny crcks tht could e oviously seen. Decy ws identified y detecting color chnges, loss of wood nd lck of proing resistnce in the pick test. Delmintion nd crcks were identified y oserving visile seprtion. A high-resolution digitl imge of ech cross section ws susequently otined to document the internl conditions of the glulm section. Signl processing nd nlysis: First, the time-energy centroid (TEC) nd frequency-energy centroid (FEC) of the coustic signls were otined through moment nlysis. The TEC indictes the time t which most of the signl energy hs een received. Mthemticlly, TEC (t E ) is the first moment of the squre of signl with respect to time divided y the zeroth moment with respect to time: t E = = N At i= i i N A i i where N is the numer of time smples, A i is the mplitude of the i th time step nd t i is the time t the i th time step. The FEC is defined s: f E = = N f S f i= i i N f S i i where f E denotes FEC, N f is the numer of frequencies in the spectrum nd S i nd f i re the mplitude nd frequency of the i th frequency of the spectrum, respectively. The centroid prmeters were exmined s potentil defect predictors through preliminry nlysis. The predicted defects were then compred with the ctul conditions of the cut-opened cross sections. Second, the coustic signls were decomposed y mens of wvelet pcket trnsform (WPT) nd the energy of the su-nds () ()

4 F. Xu et l.: Impct testing for dmge detection in glulm ws deduced s chrcteristics of the response signls. In mthemtics, WT is reking up of signl into shifted nd scled versions of mother wvelet or se function. This results in vrile sizes of window function nd mkes it possile to detect discontinuities nd rekdown points of dt tht other nlysis methods usully miss (Kim nd Melhem ). WPT cn e viewed s nturl extension of the WT, decomposing signl repetedly into successive low-frequency (LF) nd high-frequency (HF) components. The only difference etween WPT nd WT is tht WPT decomposes not only the pproximtion ut lso the detils t given level; therefore, it is more flexile nd hs wider se for the nlysis of signls (Mllt 999; Th et l. ). WPT lso enles multi-resolution dmge detection s it cn loclize multi-frequency nds in the time domin. In this pper, the response signls collected from the glulm rch section were decomposed into four levels y WPT with mother wvelet d (one of the wvelet functions of Duechies fmily), nd the wvelet pcket coefficients t the fourth level were reconstructed, respectively. The energy rtio (ER, defined s percentge of the energy of ech scle component decomposed) ws clculted s the feture vector to identify different types of sound sources. The chrcteristic differences of the response signls cn e compred y computing the ER of ech lef t the fourth level nd drwing time-frequency distriution mp nd wvelet pcket spectrum of the decomposed signl. Following the WPT of the response signls, the feture differences concerning the signls ssocited with typicl defects nd those ssocited with intct cross-sections were nlyzed, nd the su-nds tht cn serve s dignostic criterion for the presence of defects were identified. Bsed on the identified fetures, the possiilities for the presence of defects t ll testing loctions were nlyzed nd determined. Finlly, sed on the time-frequency distriution otined through the WPT, the trnsmission time differences were further nlyzed for the su-nd components of the signls from the dmged nd solid sections, nd pproximte velocity reltions were determined in terms of the su-nd components etween the signls. A defect dignosis lgorithm will then e proposed to predict the nture nd sizes of the defects sed on the trnsmission chrcteristics nd the geometric pth of the wve propgtion. Results nd discussion Time-energy nd frequency-energy centroids Figure shows the time-domin signls (), frequency spectr of the signls (), time-frequency distriutions (c), frequency su-nd distriution (d) nd the crosssectionl imges (e) for three representtive loctions (,, nd ; ll t P position) in the glulm rch section. The cut-opened cross sections reveled severe dmge (including decy nd delmintion) t loction (Figure, Ie), moderte decy nd delmintion t loction (Figure, IIe) nd intct condition t loction (Figure, IIIe). Both time-domin signls nd the frequency spectr show distinct differences mong these loctions. It ppers tht decy nd delmintion cused noticele decrese in the time-domin signl. In the frequency domin, the min frequency contents shifted to the LF region, when the glulm rch progressively deteriorted. Compring with the solid cross section (loction ), the HF content of the signl ws sent t loctions nd, nd the mplitude of the pek frequencies ws lso sustntilly reduced (Figure ). These oservtions re consistent with the results reported y Tiitt et l. (99) nd Senlik et l. (). Figure shows the distriution of TEC nd FEC of the response signls collected t loctions of the glulm rch section. The TEC vlue t the three positions (P, nd ) showed very little vrition throughout the entire glulm rch section. However, significnt chnge in the TEC vlue ws oserved long the length. TEC remined reltively constnt (.7. s) from loctions to, ut incresed to. nd. s t loctions nd, representing % nd % increse, respectively. Clerly, the dmges t loctions nd cused significnt dely in the rrivl of the ulk signl energy. On the other hnd, the FEC vlues show opposite trends long the length, with low frequencies ( khz nd less) oserved t loctions nd nd high frequencies (ove Hz) t the remining loctions. The sence of high frequencies t loctions nd is the indiction tht high-frequency components hd een ttenuted shrply, when coustic wves trnsmitted through the glulm cross section, which implies tht the lmintes t those loctions hd much lower density compred to the intct lmintes t other loctions, s result of the deteriortion confirmed y the cut-opened cross sections (Figure, Ie nd IIe). It is lso noted tht, from loction onwrds, the FEC vlue t position P ws consistently lower thn the vlues t positions nd, hint of possile erly stge of mteril deteriortion t P side reltive to positions nd. This could hve een cused y the high temperture exposure t tht side of the rch during the fire incident in 99 (Teder nd Wng ), evidenced y the cross-sectionl imges nd visul oservtion of the smll delmintion. Another interesting finding is tht the FEC of the response signls t loction 9 position nd loction positions P, nd fell shrply, ut the corresponding TEC siclly remined unchnged. A visul ssessment of the corresponding cross sections following cross-cutting reveled some smll delmintion t loction 9 (Figure ) nd longer delmintion t loction (Figure ). This indictes tht the FEC of the response signls is more sensitive to the presence of smll delmintion thn the TEC.

5 F. Xu et l.: Impct testing for dmge detection in glulm Amplitude c... Frequency nd (Hz)..... Amplitude. c Frequency nd (Hz) e.. Energy rtio (%).. d e. Energy rtio (%) c. Amplitude (V). d Amplitude Amplitude (V).. e Energy rtio (%) Amplitude (V). d Frequency nd (Hz) Figure : Fetures of the coustic signls received from loctions, nd nd the corresponding cross-sectionl imges of the glulm rch section. I Loction (with severe decy nd delmintion); II Loction (with moderte decy nd delmintion); III Loction (intct). () Time-domin response signl from testing point P; () frequency spectrum of the response signl; (c) time-frequency distriution; (d) frequency su-nd distriution; nd (e) cross-sectionl imges of the glulm rch section. A compritive nlysis of ll loctions clerly indicted tht vritions of oth the TEC nd FEC vlues reflected physicl conditions of the glulm rch section nd therefore re good predictors for detecting the presence of internl defects. However, from the time-domin nd frequency-domin nlyses of the non-sttionry coustic signls, it is difficult to scertin the nture of defects. Therefore, it is necessry to pply the WT method to further decode response signls nd extrct new fetures through time-frequency nlysis. Energy distriution of wvelet su-nds Figure shows the ERs of wvelet su-nds of the response signls t the testing loctions. The following

6 F. Xu et l.: Impct testing for dmge detection in glulm Frequency spectrum 7 No. No. Amplitude No. Figure : Three frequency spectr overlid showing the frequency shift from the higher to lower frequencies with incresing dmge (loction with severe decy nd delmintion; loction with moderte decy nd delmintion; nd loction intct). Figure : Cut-open cross-sectionl imges of the glulm rch section t loction 9 () nd loction (). Red lines indicte visully oserved delmintion. TEC (s) TEC (P) TEC () TEC () FEC (P) FEC () FEC () 7 9 Testing loction no. 7 Figure : Time-energy centroid (TEC) nd frequency-energy centroid (FEC) of the response signls t vrious testing loctions. discussion will e focused on three representtive signls from loctions, nd to illustrte the proposed method. The response signl received t loction (Figure, I) corresponds to the cross section tht contined severe defects (decy nd delmintion s illustrted in Figure, Ie). The distriution of frequency su-nds (Figure, Id) indictes tht the signl energy is primrily concentrted in the LF rnge. The comined energy for the two LF su-nds (LF: 7 Hz nd LF: 7 7 Hz) ccounts for % of the totl energy of the response signl. The ER of the LF su-nd reched.%. In contrst, the ERs of the middle-frequency (MF) su-nds (MF: 7 Hz nd MF: Hz) 7 FEC (Hz) re.% nd.%, respectively, nd the ER of the HF component (>. khz) ccounts for <%. The response signl received t loction (Figure, II) corresponds to the cross section tht contined moderte deteriortion (erly-to-moderte decy nd prtil delmintion, see Figure, IIe). The distriution of the frequency su-nds (Figure, IId) shows tht the signl energy is mostly concentrted in the MF rnge. The ERs of the su-nds LF ( 7 Hz) nd HF (>. khz) re.% nd.%, respectively. The ERs of the su-nds of 7 7 Hz, 7 Hz nd Hz re 7.%,.% nd.%, respectively. Compred to the response signl ssocited with severe defects (Figure, I), the distriution of the frequency su-nds presented in Figure, IId hs the following fetures: the energy of the LF component ( 7 Hz) dropped significntly, while the MF components incresed considerly. In ddition, the HF su-nd incresed s well. For the response signl without defects (Figure, III for loction ), the frequency su-nds hve very unique distriution. The ERs of the LF su-nds of 7 Hz nd 7 7 Hz re very low, only.% nd.7%, respectively. Menwhile, the ERs of the MF su-nds of 7 Hz nd Hz re reltively high,.% nd.%, respectively. The most importnt oservtion is tht the ER of the HF su-nd (>. khz) is the highest, reching.7%. The ERs of wvelet su-nds of the response signls otined from different loctions (long length direction)

7 F. Xu et l.: Impct testing for dmge detection in glulm 7 Energy rtio (%) 7 7 Testing loction no Su-nd (Hz) prtil delmintion t different depths t these loctions (Figure ). Anlysis of frequency su-nds indictes tht the internl condition of glulm memer could hve significnt impct on the signl energy distriution. When severe decy occurred, the energy of the response signl ws hevily concentrted on the first LF su-nd (LF: 7 Hz), while tht of the second LF su-nd (LF: 7 7 Hz) could e relted to erly decy or delmintion. HF su-nds of >. khz re indictive of sound timer; i.e. the higher the ER of the HF components, the etter is the qulity of the glulm memer. Consequently, it cn e sfely concluded tht the su-nd energy distriution of response signl from wvelet nlysis is n indictor of internl soundness. Energy rtio (%) Energy rtio (%) c Su-nd 7 Hz Su-nd 7 7 Hz Su-nd 7 Hz Su-nd Hz Su-nd Hz 7 Testing loction no. 7 Testing loction no. 9 nd different positions (P, nd ) re summrized in the r grphs in Figure. The high ERs of the LF nds t loctions nd re clerly ssocited with the severe-to-moderte deteriortion s reveled y the crosssectionl imges (Figure, Ie nd IIe). Loctions 9 nd hve slightly higher thn norml ERs of the LF components in (% increse in comprison with sound sections), indicting possile deteriortion in these res. A visul exmintion of the cross sections reveled some Su-nd (Hz) Figure : Energy rtios of wvelet su-nds of the response signls t testing loctions. () Position P; () position ; nd (c) position. Su-nd (Hz) Time-frequency spectrum nlysis Figure, Ic, IIc nd IIIc re the time-frequency spectr of the response signls collected t three representtive loctions, nd. The color indictes the mplitudes of the frequency components (color scle is presented in the secondry Y-xis). These figures illustrte oth intensity of the frequency components nd their distriution. For loction, which contined dvnced decy nd delmintion, energy distriution of the MF ( Hz) nd LF su-nds re explicitly presented in Figure, Ic. The receiving time of the response signl is t. s. After. s, the LF su-nds re still stle in trnsmission nd their mplitudes re siclly unchnged. The LF (7 7 Hz) component ppered regulrly t n intervl of. s (Δt ). In contrst, the HF su-nds re nerly not visile nd the MF components show n ovious ttenution. For loction, which hd moderte decy nd prtil delmintion, the response signl ws received t. s. The LF su-nd ( 7 Hz) is lmost nonexistent. The HF component ws oserved in the eginning, ut then dropped drmticlly fter. s nd ws hrdly visile. Menwhile, oth the LF (7 7 Hz) nd MF ( Hz) components were present stedily nd their mplitudes re unchnged. The MF component ppers regulrly t n intervl of out. s (Δt ). For the intct cross section t loction, the response signl ws received t.9 s, much erlier thn those t loctions nd. The frequencies of the signl re minly in the MF (7 Hz) nd HF (>. khz) sund regions, while LF ( 7 Hz) ws lmost missing. All frequency su-nds re stle with less ttenution

8 F. Xu et l.: Impct testing for dmge detection in glulm Wve initition point H min S in trnsmission, nd pper lmost t n equl intervl of. s (Δt ), which indictes tht the coustic wves propgte t velocity tht is governed y the mteril itself, independent of the signl frequency. Assuming tht the glulm memer is homogeneous ( hypotheticl ssumption), the velocity rtio mong loctions, nd cn e estimted sed on the time intervls of the ppernces of the su-nd components, which is pproximtely :: (/Δt :/Δt :/Δt ). This estimtion is consistent with the velocity rtios determined y the stress wve timing method reported y Teder nd Wng (). Nture nd sizes of defects S S H mx Wve reception point Figure 7: Wve propgtion pth cross the depth of glulm memer with n rtificil delmintion. S, S nd S represent three possile propgtion pths. S : reference pth with no defect; S : detour for delmintion locted in the middle of the em (with the mximum depth H mx =. S ); nd S : detour for delmintion locted ner the oundry (with the minimum depth H min =. S ). It is ovious from the time-frequency spectrum nlysis tht the response signl from loction is dominted y LF components, while the HF components nerly vnished. This mens tht lrge defects must hve een present in the wve propgtion pths. These defects sored so much of the HF energy components tht the HF wves simply disppered. For intct glulm, the wve propgtion pth cn e ssumed to e stright line cross the glulm depth, represented y the shortest distnce (S ) etween the inititing point nd the receiving point (Figure 7). In comprison, the perceived wve propgtion distnce (S or S ) t the cross section of loction with severe deficiencies (ssuming only delmintion without decy) should triple the reference distnce (S,. cm t loction ) ccording to the estimted velocity rtio. With tripled wve propgtion distnce, single delmintion my rnge from minimum depth of 9. cm, if it is close to the oundry (H min =. S ), to mximum depth of. cm, if it is locted in the middle (H mx =. S ) (Figure 7). In fct, the glulm rch ws only 9. cm wide, i.e. much less thn the predicted delmintion depth; therefore, the hypothesis of single delmintion should e rejected. This led to conclusion tht in ddition to delmintion, n dvnced decy occurred, which is consistent with the conclusion drwn previously. At loction, similr nlysis ws pplied nd it ws found tht in cse of delmintion it my rnge from. cm to 9.7 cm in depth, which is relistic scenrio. But the high velocity rtio could lso indicte scenrio of internl decy. In this cse, it is necessry to exmine the timefrequency distriution (Figure, IIc). As discussed ove, the response signl from severely decyed glulm exhiits cler su-nd of 7 Hz. If the response signl shows this LF component, the glulm memer must hve suffered significnt internl decy, nd the length of the supposed delmintion cnnot e determined. Conversely, in the sence of n internl decy, the delmintion size could e estimted. For loction, the time-frequency spectrum contined the 7 Hz component. It is concluded tht oth delmintion nd decy re present in this loction nd, thus, the delmintion size cnnot e determined. For loction, the size of the delmintion ws estimted to e in the rnge of cm. But the timefrequency distriution of the response signl contined lmost no LF component, i.e. the energy of the 7 Hz component only ccounted for.% of the totl signl energy (Figure ). It is therefore hypothesized tht the depth of delmintion should e etween.7 cm nd 7.7 cm. Exmintion of the cross section of loction confirmed tht the delmintion ws out. cm deep. The depth of delmintion t other loctions ws estimted y the sme method nd the results re listed in Tle. Most of the estimtions re generlly consistent with the mesured vlues except the ones t loction position nd loction position, where delmintion ws not oserved visully. The percentge errors re <% nd the minimum error ws only out %. The multiple delmintion prllel to ech other ws considered s single one. The clculted vlue of delmintion depth should e considered s the mximum or ccumultive vlue of the mjor delmintion ctully existing t ech cross section.

9 F. Xu et l.: Impct testing for dmge detection in glulm 9 Tle : Predicted nd mesured depths of delmintion in the glulm rch section. Loction Position H mx (cm) H min (cm) H m (cm) Prediction error (%) Visul ssessment of cross section No. P Severe decy + delm No. P Moderte decy + delm No. P Delmintion No. P Delmintion No. P Delmintion No. P.7... Delmintion.7. No visul sign of delm No. 7 P No. P Delmintion.. No visul sign of delm No. 9 P Delmintion No. P Delmintion Delmintion Delmintion No. P Delmintion Delmintion No. P Delmintion Delmintion No. P H mx, mximum depth of delmintion; H min, minimum depth of delmintion; H m, mesured depth of delmintion; prediction error ( H + H ) mx min = / Hm E %. Hm No delmintion oserved visully. Depth of delmintion could not e determined for decy existence in glulm rch section or invisile crcks. Conclusions TEC nd FEC of the first moment re good indictors of the presence of internl defects in glulm memers. Higher TEC nd smller FEC indicte greter proility of the presence of defects. FEC is more sensitive to smll defects thn TEC. There is link etween su-nd energy distriution of response signls nd internl defects of glulm memer. With incresing decy/delmintion, the dominting energy of the response signl ws grdully shifted from the HF nds to MF nd LF nds. The su-nd energy distriution is good indictor of internl dmges such s delmintion or decy in glulm memers. The energy distriutions of wvelet su-nds re more sensitive to defects thn the first moment prmeters. The LF su-nds of 7 Hz nd 7 7 Hz re

10 F. Xu et l.: Impct testing for dmge detection in glulm the most effective discrimintive fetures ssocited with internl dmge nd the HF nds re indictive of sound timer. In ddition, the ERs of the su-nds re relted to the extent of dmge. The higher the ER of the HF components, the etter is the qulity of the glulm memer. The locl time-frequency chrcteristics of the wvelet pcket cn e used to determine the velocity rtios reltive to the velocity in intct glulm. The nture of defects cn lso e preliminrily scertined y nlyzing the velocity rtio nd energy distriutions of the wvelet su-nds. In cses where only delmintion exist in the glulm em, the depth of delmintion could e estimted sed on the velocity rtio. Acknowledgments: This reserch ws conducted through coopertion etween the USDA Forest Service Forest Products Lortory (FPL), Nnjing Forestry University (NFU) nd the Estonin University of Life Sciences, nd ws supported in prt y the Priority Acdemic Progrm Development (PAPD) of Jingsu Higher Eduction Institutions, the NFU Innovtion Grnt for Outstnding PhD Disserttions (grnt 7) nd the Foundtion Archimedes Europen Socil Fund DoR progrmme. The uthors cknowledge the technicl support of the Engineering Mechnics nd Remote Sensing Lortory (EMRSL) t FPL. References Angst, V., Mlo, K.A. () Moisture induced stresses perpendiculr to the grin in glulm: review nd evlution of the reltive importnce of models nd prmeters. Holzforschung :9 7. Arrig, F., Esten, M., Iñiguez-Gonzlez, G., Bodill, I., Lin, D.F., Gonzlez-Snz, M. () Structurl ssessment of the timer structure of the Cs Grnde uilding in the Rel Cortijo de Sn Isidro, Arnjuez, Mdrid (Spin). In: Proceedings of the th Interntionl Nondestructive Testing nd Evlution of Wood Symposium, Mdison, Wisconsin, USA. pp.. Bsu, B.B., Gupt, V.K. (997) Non-sttionry seismic response of MDOF systems y wvelet trnsform. Erthquke Eng. Struc. :. Byiss, W.L., Hritos, N., Thelndersson, S. () Virtion-sed structurl dmge identifiction using wvelet trnsform. Mech. Syst. Signl Pr. :9. Bell, F.C. (). Overview of the use of ultrsonic technologies in reserch on wood properties. Wood Sci. Technol. :97. Birks, A.S., Green, R.E., Mclntire, P. Nondestructive Testing Hndook ( rd edition), Vol. 7. Ultrsonic Testing. Americn Society for Non-destructive Testing, Columus, OH, 7. Brenner, M.J. () Non-sttionry dynmics dt nlysis with wvelet-svd filtering. Mech. Syst. Signl Pr. 7:7 7. Bucur, V. Delmintion in Wood, Wood Products nd Wood-Bsed Composites. Springer, Berlin, Germny,. Dill-Lnger, G., Bernuer, W., Aicher, S. () Inspection of gluelines of glued-lminted timer y mens of ultrsonic testing. In: th Interntionl Symposium on Nondestructive Testing of Wood, Eerswlde, Germny. pp. 9. Divos, F. () Delmintion evlution of in-service glulm ems nd other structurl memers vi ultrsonics. In: Delmintion in Wood, Wood Products nd Wood-sed Composites. Ed. Bucur, Springer, Berlin, Germny. pp.. Frühwld, E., Tortti, T., Thelndersson, S., Serrno, E., Emilsson, A. Design of Sfe Timer Structures How Cn We Lern from Structurl Filures in Concrete, Steel nd Timer? Report TVBK-, ISSN Lund University, Lund, Sweden, 7. Hnsson, H., Lrsen, H.J. () Recent filures in glulm structures nd their cuses. Eng. Fil. Anl. :. Ksl, B. () In-situ ssessment of structurl timer: stteof-the-rt, chllenges nd future directions. Adv. Mt. Res. :. Kim, H., Melhem, H. () Dmge detection of structures y wvelet nlysis. Eng. Struct. :7. Kitd, Y. (99) Identifiction of nonliner structurl dynmic systems using wvelets. J. Eng. Mech.-ASCE :9. Liu, Y., Li, Z., Zhng, W. () Crck detection of fire reinforced composite ems sed on continuous wvelet trnsform. Nondestr. Testing Evl. :. Mev, E., Severin, I., Bondrenko, S., Gr Mev, R. () Acousticl methods for the investigtion of dhesively onded structures: review. Cn. J. Phys. :9. Mllt, S. A Wvelet Tour of Signl Processing, nd ed. Acdemic Press, London, UK, 999. Neuenschwnder, J., Snri, S.J., Schuetz, P., Widmnn, R., Vogel, M. () Delmintion detection in 9-yer-old glulm lock with scnning dry point-contct ultrsound. Holzforschung, 7: Rmmer, D.R., Mour, J.M., Ross, R.J. () Structurl performnce of the second oldest glued lminted structures in the United Sttes. In: Proceeding of the Structures Congress. April,, Boston, Msschusetts, USA. Structurl Engineering Institute of ASCE, Reston, VA, USA. pp.. Ross, R., Brshw, B.K., Wng, X., White, R.H., Pellerin, R.F. Wood nd Timer Condition Assessment Mnul. Forest Products Lortory, Mdison, WI, USA,, p. 7. Ruck, M. () Dmge detection in ems using wvelet trnsform on higher virtion modes. J. Theor. Appl. Mech. 9:99 7. Snri, S.J., Furrer, R., Neuenschwnder, J., Niemz, P., Sennhuser, U. () Air-coupled ultrsound inspection of glued lminted timer. Holzforschung :77 7. Senlik, C.A., Bell, F.C., O Dell, K., Reis, H. () Detection nd ssessment of wood decy in glulm ems using throughtrnsmission ultrsonic pproch. In: Proceedings of SPIE, Sensors nd Smrt Structures Technologies for Civil, Mechnicl, nd Aerospce Systems, Sn Diego, CA, USA, Vol. 9, 9W- to 9W- ( pges). Senlik, C.A., Bell, F.C., Reis, H. () Detection nd ssessment of wood decy in glulm ems using decy rte pproch. Insight- Non-Destructive Testing & Condition Monitoring :. Sitr, M. () Glulm em repir nd strengthening. Civil + Structurl Engineer, My feture rticle rticle//glulm-em-repir-nd-strengthening (ccessed on Decemer, ).

11 F. Xu et l.: Impct testing for dmge detection in glulm Th, R.M.M., Noureldin, A., Lucero, J.L., Bc, T.J. () Wvelet trnsform for structurl helth monitoring: compendium of uses nd fetures. Struct. Helth Monit. : 9. Teder, M., Wng, X. () Nondestructive evlution of 7-yer old glulm rch. In: Proceeding of the th Interntionl Nondestructive Testing nd Evlution of Wood Symposium, Mdison, Wisconsin, USA. pp. 7. Teesdle, L.V. (9) New uilding demonstrtes modern wood construction. ENR. April, 7. Tiitt, M., Bierncki, J.M., nd Bell, F.C. (99) Acousto-ultrsonic ssessment of internl decy in glulm ems. Wood Fier Sci. : 7. Umesh, P.K., Rvichndrn, R., Sivsurmnin, K. (9) Crck detection nd quntifiction in ems using wvelets. Comput. Aided Civil Infrstruct. Eng. :9 7. Wng, X., Teder, M., Wcker, J. () Condition ssessment of the timer structures of century-old industril uilding using non-destructive inspection procedure. Adv. Mt. Res. 77:. Xu, F., Liu, Y. () Feture extrction of coustic emission signl of plywood in loding y wvelet pckge trnsform. J. Nnjing Forestry Univ. (Nturl Science Edition) :. Xu, F., Liu, Y. () Feture extrction nd clssifiction method of coustic emission signls generted from plywood dmge sed on EMD-SVD. J. Bsic Sci. Eng. : 7.