基于神经网络和模态分析的桥梁损伤识别

华中科技大学硕士学位论文基于神经网络和模态分析的桥梁损伤识别姓名：黄鹏申请学位级别：硕士专业：桥梁与隧道工程指导教师：张海龙20061030ANSYSVBBPBPBPVBBPBPIIABSTRACTWiththerapiddevelopmentofcommunicationindustryinourcountry,densityofhighwaysnetworkimprovesconstantlyandthelargespanbridgeisemergingconstantly.Butstructuraldamagessuchascrackingandagingetc.tosomedegreeduringthebridgestructures’servicelifeoccurfromtimetotime.Someinvisibledamagesdonotalerttheengineersintimeandmayresultindisastrousconsequencesandultimatelycausesubstantiallossoflifeandproperty.Therefore,itisimportanttocommandthehealthstatusofbridgeincommonlyusingstateintimeandtodetectdamagesattheearliestpossiblestage.Onthebasisofcollectionandanalysisofthedataaboutbridgestructuralhealthmonitoringsystem,structuraldamageidentificationandartificialneuralnetworks,vibrationmodalanalysistheoryisintegratedwithBPneuralnetworkssoastodetectbridgedamagewiththehelpofANSYSandaprogramwrittenwithVB.Theresultshowsthemethodworkswell.Firstly,severaldamageidentificationmethodsusedtoidentifythebridgestructuraldamagetheoryarediscussedandanalyzed.Varioustheoreticalmethods,formulaanduseareintroducedmainlyfromthestatic,dynamicrecognition,andartificialintelligence.Secondly,thebasictheoryofBPneuralnetworksisdiscussedandthebasictenetsofdamageidentificationusingBPneuralnetworksareanalyzed.AnewvisualprogramiscompiledwithVBsoftware.Andonesimplesupportedrectanglesteelbeammodelispresentedascalculatedexamplefortheabove-mentionedmethodapplication.Thirdly,thedamageidentificationmethodbasedoncurvaturemodalshapeandflexibilitycurvaturearediscussedinthispaper.Therelationshipbetweenstructuraldamageandmodalchangesisexpoundedandathree-spancontinuousbeammodeisusedtostudyIIIonthedamageidentificationmethodbasedonvibrationmodaltheory.Bothsingleandmultipledamageidentificationarestudied.Finally,onthebackgroundoftheprojectofLujingRoadOverpassBridgeinGuangzhou,thetechniquebasedoncurvaturemodalshapeanalysistheoryandBPneuralnetworksisusedtoidentifythedamagedpositionandextentlikelyoccurring.Theresultshowstheeffectivenessofthismethodusedinbridgestructures.TheresultindicatesthatthemethodbasedonBPneuralnetworksandvibrationmodalanalysistheorycandetectnotonlythedamagepositionbutthedamagedegreeofconcretebridges.Themethodwillhavegoodapplicationperspective.Itisimportanttoforecastthehealthstatusofbridgestructures.Keywords:DamageidentificationBridgestructureBPneuralnetworksVibrationmodalanalysisCurvaturemodalshapeFlexibilitycurvature_____111.1[1]1.219671246[3]1983628(Connecticut)Mianus[4]1999161020502[5]structuralhealthmonitoringSHM4[6]1.3[5][7]31.3.1,,,Alemang[8]MAC(ModalAssuranceCriterion)MACMACMoller[9]MACWest[10]MACMACRatcliffe[11]LaplacePandey[12]Stubbs[13][14]Ratcliffe[15][16]4[17]Shi[18]MSECR(modalstrainenergychangeratio)Cornwell[19]10%[20]()1.3.2,,,,[21]Cawley[22]/ijdwdw(ID)FEM(ID-Ai)(ID-X)()()minIDAiIDX---iJuneja[23](contrastmaximization)Stubbs[24][22]Hearn[25]i5Pandey[26]2[27]3[20]Baruch[28]Zimmerman[29][30]Doebling[31]Hemez[32][33][34],1.3.32090BPPurduVenkatasubramnianChan[35]1989BP6Wu[36]Worden[35]BP20Choi[35]BP[37]BP[38]RBF[39]BPRBF[40][41][42][43][44]1.3.4Holland[45]19751986GoldbergSamtani101990Deb[46][47](1)7(2)(3)1.41,,2,,,,38,,4,,1.5123BPBPVBBP45ANSYSVBBP922.1(,)()NNGA2.210(SI)()[48]2.3[35]()112.3.1()N[]{}[]{}[]{}{}0MxCxKx++=ggg(2-1)2-1[][]()[]{}20KMwF-=(2-2)[]K[]M[]KΔ[]MΔ2w[]F2wΔ[]DF(2-2)[][]()()[][]()()[][](){}220KKMMwwFDF+Δ-+Δ+Δ+=(2-3)[]MΔ(2-3)[][]()()[]()[][](){}220KKMwwFF+Δ-+Δ+Δ=(2-4)12[][][][][][]2TTKMFFwFFΔΔ=2-5{}iFi123N{}[]{}{}[]{}2TiiiTiiKMFFwFFΔΔ=2-6[]nKΔn2-6{}[]{}{}[]{}2TininiTiiKMFFwFFΔΔ=∑2-72-7n{}[]{}{}[]{}2TiniiTiiKMFFwFFΔΔ=2-8()nnii,22awwΔ=Δ2-9nan[][]nnnKKaΔ={}[]{}{}[]{}2TniniiTiiKMaFFwFFΔ=2-102-102iwΔ2jwΔ{}[]{}{}[]{}{}[]{}{}[]{}22TiniTiiiTjjnjTjjKMKMFFFFwwFFFFΔ=Δ2-11132-11122.3.212341[49]ModalAssuranceCriteria(MACI)CoordinateModalAssuranceCriteria(COMACI)11(MACI)20,(,)0,0,TijMACijTTiijjIFFFFFF⎡⎤⎣⎦=×2-120FFji,;14),(jiMACI0110(COMACI)21220,0,111()mmmCOMACjkjkjkjkjjjIkFFFF-===⎛⎞⎡⎤=×⎜⎟⎢⎥⎝⎠⎣⎦∑∑∑2-13km)(kICOMAC01k[50]()MACICOMACI20EFF=-2-140ΦFE315{}{}xuxYF∂=∂2-15{}xY{}uFx[51]xYxekjkjYae=akjkjejjY41/r221dxyd=r2-16rxy2-1616iijijijijjillzzzc1),1(,),1(,21-+-+-==r2-17jic,jiz,ij1-ilili-1iHCCH-=02-18C0Ce[17]()()()()re/22xhxhxvxhxvxu=′′=∂∂≈∂∂=2-19uxv1/rh(x)2-192.3.31[52]F171211nTTijrrrrFfFWFffw-=⎡⎤===⎣⎦∑2-20r;1,2nFfff⎡⎤=⎣⎦LrfM),,2,1(1nrMrTrL==ff22212,,ndiag⎡⎤Ω=⎣⎦Lijfji,()()⎪⎪⎩⎪⎪⎨⎧=≠=∑∑==nrirrnrjrirrijjijif1221211fwffw(2-20)0ijFFFf⎡⎤Δ=-=Δ⎣⎦2-21F0FjfΔFΔjjfΔijijffΔ=Δmax2-2222-20[]nfffFL,,21=2-23jj18{}{}∑==njjff12-24jf′jf′′lfffjjj1--=′j2,3,n-12-252112lffffjjjj+-+-=′′j2,3,n-12-26l;n[53]2.3.4[][][]1211nTiiiiKFFw-==∑2-27[][][]KKK-=Δ02-28[]KΔ19[]KΔ2.42.4.1HopfloldKohononChenShan(1995)Garrett(1993)BP202.4.270JohnHolland1,2N3,,m,n,,GA,3N4213N(),,5,,6352.52233.1(A.N.N)40PittsMcCulloch(1)(2)(3)(4)3.2BP3.2.1BP1985RumelhartBPBP23BP3.2.2BPBPBPBP{(X,Y)XYX}nh(1)hn≤≤LhhF(h)W24BPiio=iipjjipinetow=∑(3-1)()pjjpjofnet=(3-2)pjiwijpiojpjojfBP(1)01(2)(3)3-311()()kkk-kk-jjijiiOfIfOw==∑(3-3)kjOkjkjI1k-kijw1k-ikj(4)111(1)()k-kk-kkkijijjjttdOwawh-+=-(3-4)k=m()()kmmjjijdOyfI=-km11()(