Page 96 - 2023年第54卷第5期
P. 96
50(6):687 - 698.
[32] 任秋兵,沈扬,李明超,等.水工建筑物安全监控深度分析模型及其优化研究[J].水利学报,2021,52
(1):71 - 80.
[33] 陈俊杰.耦合 BIM的长距离输水渠道无人机巡检与险情智能图像识别研究[D].天津:天津大学,2020.
[34] RENQ,WANGG,LIM,etal.Predictionofrockcompressivestrengthusingmachinelearningalgorithmsbased
onspectrumanalysisofgeologicalhammer[J].GeotechnicalandGeologicalEngineering,2019,37(1):475 - 489.
[35] HANS,LIH,LIM,etal.Measuringrocksurfacestrengthbasedonspectrogramswithdeepconvolutionalnet
works [J].Computers& Geosciences,2019,133:104312.
[36] HANS,LIH,LIM,etal.A deeplearningbasedmethodforthenon - destructivemeasuringofrockstrength
throughhammeringsound [J].AppliedSciences,2019,9(17):3484.
[37] LUX,XUY,TIANY,etal.Adeeplearningapproachtorapidregionalpost - eventseismicdamageassessment
usingtime - frequencydistributionsofgroundmotions[J].EarthquakeEngineering& StructuralDynamics,2021,
50(6):1612 - 1627.
[38] LIAOW,CHENX,LUX,etal.Deeptransferlearningandtime - frequencycharacteristics - basedidentification
methodforstructuralseismicresponse [J].FrontiersinBuiltEnvironment,2021,7(10).doi:10.3389?fbuil.
2021.627058.
[39] HEK,ZHANG X, REN S, etal.Deep residuallearningforimagerecognition[C]??IEEE Conferenceon
ComputerVisionandPatternRecognition ,2016.
[40] SZEGEDYC,IOFFES,VANHOUCKE V,etal.Inception - v4, Inception - resnetandtheimpactofresidual
connectionsonlearning [C]??ProceedingsoftheAAAIConferenceonArtificialIntelligence,2017.
IntelligentidentificationandanalysisofPCCPwirebrokensignalin
waterdiversionprojectusingprototypetesting
ZHANGYe,YUANSimin,LIYanlong,WENLifeng,SIZheng,SUNKaiyu
(StateKeyLaboratoryofEco - hydraulicsinNorthwestAridRegion,Xi’anUniversityofTechnology,Xi’an 710048,China)
Abstract:Inwaterdiversionprojects,thebreakageofsteelwirescaneasilyleadtostructuralandfunctionalfail
ureofPrestressedConcreteCylinderPipes (PCCP).Thisstudyaimstoanalyzesignalcharacteristicsandidentify
thetypeofwirebreakageusingintelligentlearningmodels.Intheresearch,Aprototypetestforwirebreakagewas
conductedonanembeddedPCCPwithaninnerdiameterof3.4metersandalengthof5meters.Real - timemonito
ringwascarriedoutusingadistributedopticalfibersensortodetectcuttingwire,corrosionwire,andimpactnoise
signals.BasedonShort - timeFourierTransform(STFT)anddeeplearningmodels,thewirebreakagesignalswere
reconstructed.Finally ,awirebreakagesignalrecognitionmodelwasestablishedusingasupportvectormachine.
BasedonthereconstructionofsignalsusingInception - ResNet - v2 ,thelowestandhighestaccuracyofthewire
breakagerecognitionmodelare92.9% and100%,respectively.Theeffectivenessofthesignalreconstructionwas
alsodemonstratedusing t - SNE.Thisstudyhasachievedeffectiverecognitionofwirebreakagetypesbycombining
differentintelligentlearningmethods.Itprovidesnewmethodsforlong - term wirebreakagemonitoringandearly
warninganalysisinPCCPoperation.
Keywords:prestressedconcretecylinderpipes(PCCP);deeplearning;knowledgetransfer;short - timeFourier
transform ;wirebrokensignal;intelligentidentification
(责任编辑:韩 昆)
— 5 9 —
8