简介:Thechaoticcharacteristicsoftimeseriesoffivepartialdischarge(PD)patternsinoil-paperinsulationarestudied.TheresultsverifyobviouschaoticcharacteristicofthetimeseriesofdischargesignalsandthefactthatPDisachaoticprocess.Thesetimeserieshavedistinctivefeatures,andthechaoticattractorsobtainedfromtimeseriesdifferedgreatlyfromeachotherbyshapesinthephasespace,sotheycouldbeusedtoqualitativelyidentifythePDpatterns.Thephasespaceparametersareselected,thenthechaoticcharacteristicquantitiescanbeextracted.ThesequantitiescouldquantificationallycharacterizethePDpatterns.TheeffectsonpatternrecognitionofPRPDandCAPDarecomparedbyusingtheneuralnetworkofradialbasisfunction.Theresultsshowthatbothofthetworecognitionmethodsworkwellandhavetheirrespectiveadvantages.Then,boththestatisticaloperatorsunderPRPDmodeandthechaoticcharacteristicquantitiesunderCAPDmodeareselectedcomprehensivelyastheinputvectorsofneuralnetwork,andthePDpatternrecognitionaccuracyistherebygreatlyimproved.
简介:Plasmasourceperformanceparameters,includingplasmaejectiondensityandvelocity,greatlyaffecttheoperationofashort-conduction-timeplasmaopeningswitch(POS).Inthispaper,theplasmasourceusedinthePOSofQiangguangIgeneratorischosenasthestudyobject.AtfirstthePOSworkingprocessisanalyzed.TheresultshowsthattheopeningperformanceofthePOScanbeimprovedbyincreasingtheplasmaejectionvelocityanddecreasingtheplasmadensity.Theinfluenceofthecableplasmagunstructureandnumberontheplasmaejectionparametersisexperimentallyinvestigatedwithtwochargecollectors.Finallyasemi-empiricalmodelisproposedtodescribetheexperimentalphenomenon.
简介:Anewtime-resolvedshifteddualtransmissiongratingspectrometer(SDTGS)isdesignedandfabricatedinthiswork.ThisSDTGSusesanewshifteddualtransmissiongrating(SDTG)asitsdispersivecomponent,whichhastwosubtransmissiongratingswithdifferentlinedensities,of2000lines/mmand5000lines/mm.TheaxesofthetwosubtransmissiongratingsinSDTGarehorizontallyandverticallyshiftedacertaindistancetomeasureabroadrangeof0.1–5keVtime-resolvedX-rayspectra.TheSDTGhasbeencalibratedwithasoftX-raybeamofthesynchrotronradiationfacilityanditsdiffractionefficiencyisalsomeasured.ThedesignedSDTGScantakefulluseofthespaceonarecordpanelandimprovetheprecisionformeasuringspatialandtemporalspectrumsimultaneously.ItwillbeapromisingapplicationforaccuratediagnosisofthesoftX-rayspectrumininertialconfinementfusion.
简介:Oneofthetechnicalbottlenecksoftraditionallaser-inducedbreakdownspectroscopy(LIBS)isthedifficultyinquantitativedetectioncausedbythematrixeffect.Totroubleshootthisproblem,thispaperinvestigatedacombinationoftime-resolvedLIBSandconvolutionalneuralnetworks(CNNs)toimproveKdeterminationinsoil.Thetime-resolvedLIBScontainedtheinformationofbothwavelengthandtimedimension.Thespectraofwavelengthdimensionshowedthecharacteristicemissionlinesofelements,andthoseoftimedimensionpresentedtheplasmadecaytrend.Theone-dimensionaldataofLIBSintensityfromtheemissionlineat766.49nmwereextractedandcorrelatedwiththeKconcentration,showingapoorcorrelationofR^2c=0.0967,whichiscausedbythematrixeffectofheterogeneoussoil.Forthewavelengthdimension,thetwo-dimensionaldataoftraditionalintegratedLIBSwereextractedandanalyzedbyanartificialneuralnetwork(ANN),showingR^2v=0.6318andtherootmeansquareerrorofvalidation(RMSEV)=0.6234.Forthetimedimension,thetwo-dimensionaldataoftime-decayLIBSwereextractedandanalyzedbyANN,showingR^2v=0.7366andRMSEV=0.7855.Thesehigherdeterminationcoefficientsrevealthatboththenon-KemissionlinesofwavelengthdimensionandthespectraldecayoftimedimensioncouldassistinquantitativedetectionofK.However,duetolimitedcalibrationsamples,thetwo-dimensionalmodelspresentedover-fitting.Thethree-dimensionaldataoftime-resolvedLIBSwereanalyzedbyCNNs,whichextractedandintegratedtheinformationofboththewavelengthandtimedimension,showingtheR^2v=0.9968andRMSEV=0.0785.CNNanalysisoftime-resolvedLIBSiscapableofimprovingthedeterminationofKinsoil.