基于人工神经网络的LVDT输出特性预测方法(IJEM-V8-N4-3)

I.J.EngineeringandManufacturing,2018,4,21-28PublishedOnlineJuly2018inMECS()DOI:10.5815/ijem.2018.04.03Availableonlineat’sInstitute,Dehradun-248197,IndiaReceived:27March2018;Accepted:16May2018;Published:08July2018AbstractThispaperpresentsanovelapproachfortrainingandoutputpredictionofdataofaLinearvariabledifferentialtransformer(LVDT).LVDTisacommonlyuseddeviceusedinlaboratoriesformeasuringlineardisplacementsinspecificsituations.ThisarticleconsidersapplicationofArtificialNeuralNetworks(ANNs)forlearningandoutputestimationofLVDT.Real-timeexperimentswereconductedandresultswerecollectedfortrainingofANNs.TheRegressionresultsandoutputsverifiedthelearningandpredictioncapabilityofANNs.IndexTerms:Artificialneuralnetwork,LVDT,Matlab,Simulink,Meansquareerror,Regression.©2018PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience.1.IntroductionALinearvariabledifferentialtransformer(LVDT)isanelectromechanicaltransducerwhichiscapableofmeasuringverysmalllinearmovements[1].Thetransducerconvertsmechanicalmotionofthesystemintoelectricalsignalswhichcanbeeasilyrecordedandanalysed[2].LVDTcanbeconsideredasadifferentialtransformerhavingoneprimarycoilandtwosecondarycoilsconnectedsuchthattheinducedvoltagesare180°outofphase[3].Theassemblycomprisesofacylindricalcorewhichcanmovelinearlybetweenthetwocoils.Theoutputsignalproducedindicatesdirectionofcoremovementsfromthecentreposition[4].Thesedeviceshavecapabilitiesofcontact-lesssensing,toleranceagainstradiations,infiniteresolution,goodlinearityandarecosteffective[5].ResearchershavebeenshowingkeeninterestinstudyingandanalysingperformanceofLVDT’sduetotheirvastengineeringapplications.Inamulti-objectivestudybySanthosh&Roy[6],theauthorsaimedatextendingthelinearrangeofLVDTalongwitheliminatingitsdependencyonphysicalparametersandworkingtemperature.Thestudyfurtheraddedanartificialneuralnetworkblockincascadefor*Correspondingauthor.Tel.:[phone]E-mailaddress:[e-mail]22ArtificialNeuralNetworksbasedApproachforPredictingLVDTOutputCharacteristicdataconversion.Liuetal.[7]constructedanon-machinelearningmeasuringsystemhavinganair-bearingcapacitiveLVDTcontactsensormountedonadesktopmachinetool.Theproposedsystemwascapableofdecodingthedigitalsignalsoflinearencodersandalsoacquirestheanalogsignalofcontactsensor.InanarticlebyMeydan&Healey[8]alinearvariabletransducerhavingametallicglassribbonasthecorematerialratherthannickel-ironmaterialhasbeenused.Theauthorsinvestigatedthesuitabilityofmetallicglassribbonoverotherconventionalmaterialintermsofexcitationmagnitudeandfrequency.Tianetal.[9]proposedanequivalentmagneticcircuitforasolenoidtypeLVDT.Theauthorsconsideredmagneticcircuittheorytocalculateitsmagneticreluctance,mutualinductance,outputvoltageandsensitivity.Muhammad&Umar[10]developedasmallscaleLVDTtodetectlevelofdifferentfluids.Thefluidswhichwereconsideredforanalysiswerewater,petroleumandgasoline.Theresultsshowedthattransducersworkswithgoodprecisionandhashighsensitivityforallthreefluids.NomenclatureANNArtificialneuralnetworkLVDTLinearvariabledifferentialtransformer2.SystemDescriptionThesetupincludesaLVDTmountedonapanelprovidedwithacapabilityoffinemovementofcorewiththehelpofaleadscrew.TheleadscrewisfurthercoupledwithadialgaugewiththehelpofathumbandwheelarrangementasshowninFig.1.Thesetupincludesalownoisecarrierfrequencysignalamplifierof5KHz,ademodulator,3.5inchdigitalLEDindicatorandanI.C.regulatedpowersupplyhousedinawoodenbox.AcompletesetofsystemspecificationarehighlightedinTable1.ThecircuitdiagramofLVDTsetupisshowninFig.2.Fig.1.LVDTSetupFig.2.CircuitdiagramofLVDTSetupArtificialNeuralNetworksbasedApproachforPredictingLVDTOutputCharacteristic23Table1.SpecificationsofLVDTSetupSpecificationvalueCoredisplacement+/-10mmCarrierfrequency5KHzCarriervoltage1.0Voltr.m.sLVDToutput190mVoltDemodulatoroutput1.5VoltD.C.Power220Volt/50HzThestudyconsideredthedatasamplescollectedbyreal-timeexperiments.Theresultsareobtainedbyrotatingtheleadscrewthrough1mmeachtimeandnotingdownthemodulatedoutputvoltage.Asetof25suchreadingwereobtainedandstoredin.MATfileforfurthertrainingofArtificialneuralnetworks(ANNs).3.ApplicationofANNforOutputPredictionNeuralnetworksarecomputationalmodelswhichareinspiredbybiologicalneuronspresentinhumanbrainandareusedforprocessinginformation[11-12].Thesenetworksarewidelyusedinmachinelearning,speechrecognition,computervision,medicines,textprocessingprocesses[13-18].AbasicarchitectureofANNisshowninFig.3.ANNarchitecturecomprisesofvariousnodesandbiaseswhoseweightscanbeadjustedduringthelearningprocess.Thesystemcomprisesofthreedifferentlayersnamelyinputlayer,hiddenlayerandoutputlayer.Alltheinputsarefedtothenetworkthroughinputlayer.Theseinputsareprocessedinthehiddenlayersandfurthersuppliedtooutputlayerasoutput[19].Fig.3.ANNArchitectureThe