第36卷第2期2023年2月传感技术学报CHINESEJOUＲNALOFSENSOＲSANDACTUATOＲSVol.36No.2Feb．2023项目来源:浙江省自然科学基金项目(LQ19F010012);国家科技重大专项项目(2016ZX05017－003)收稿日期:2022－03－26修改日期:2022－05－17DeformationPositionDetectionAlgorithmofOuterCasinginDouble-LayerPipeBaseonGANNetwork*CHENYourong1，JINHeli2，HUANGHua3，XUFei3，ＲENTiaojuan1，WANGKe1*(1．CollegeofInformationScienceandTechnology，ZhejiangShurenUniversity，HangzhouZhejiang310015，China;2．SchoolofComputerScienceandArtificialIntelligence，ChangzhouUniversity，ChangzhouJiangsu213164，China;3．PetroleumEngineeringInstitute，ZhongyuanOilfieldCompany，SINOPEC，PuyangHe’nan457001，China)Abstract:Itisdifficulttodetectthedeformationpositionofoutercasingindouble-layerpipeinPuguangGasfield．Tosolvetheprob-lem，eddycurrentdetectionsignalofelectromagneticflawdetectorisusedandadeformationpositiondetectionalgorithm(DPDA)ofout-ercasingindouble-layerpipebasedonGANnetworkisproposed．InDPDA，theeddycurrentdetectionsignalsarepreprocessedandconvertedintoeddycurrentimages，andadeformationpositiondetectionmodelisconstructed，includingthegenerationsub-networkandthediscriminationsub-network．Thenthelossfunctionsofthesetwosub-networkswithreconstructionloss，hiddenlossandconfrontationlossareproposedandtheweightcoefficientsofthelossfunctionsareoptimizedbyusingreptilesearchalgorithm．Thedeformationposi-tiondetectionmodelisobtainedbytraining，andthedeformationpixelsarepreliminarilydetectedaccordingtothedeformationscore．Fi-nally，accordingtothecharacteristicsofeddycurrentimages，acorrectionmethodisdesignedtocorrecttheabnormalresultsandobtainthedeformationpositiondetectionresults．TheexperimentalresultsshowthatDPDAcanaccuratelydetectthedeformationpositionoftheoutercasingindouble-layerstring．Itimprovesthedetectionaccuracyandreducesthemisseddetectionrateanderrordetectionrate．Itisbetterthanautoencoder(AE)，Pyramidstructurenetworkmodel(PSNM)andcatswarmoptimizationalgorithm(CSOA)．Keywords:eddycurrentsignal;deformatio...
