基于RS-CLDNN的礦井提升機故障診斷方法研究基于RS-CLDNN的礦井提升機故障診斷方法研究

摘要：

為提高礦井提升機的可靠性和安全性，本文提出了一種基于RS-CLDNN的故障診斷方法。該方法采用不同的傳感器采集礦井提升機的運行數(shù)據(jù)，并通過RS算法進行特征選擇，選擇出最具有代表性的特征集。接著，使用CLDNN神經(jīng)網(wǎng)絡(luò)對特征集進行訓(xùn)練，得到一個準(zhǔn)確的故障診斷模型，能夠在礦井提升機運行過程中實時進行故障診斷和預(yù)測。最后，在MATLAB仿真平臺上進行了實驗驗證，結(jié)果表明本方法的故障診斷準(zhǔn)確率高達98.5%以上，能夠有效地提高礦井提升機的安全性和可靠性。

關(guān)鍵詞：礦井提升機；故障診斷；RS算法；CLDNN神經(jīng)網(wǎng)絡(luò)

Abstract:

Inordertoimprovethereliabilityandsafetyofminehoists,thispaperproposesafaultdiagnosismethodbasedonRS-CLDNN.Differentsensorsareusedtocollecttherunningdataofminehoists,andtheRSalgorithmisusedtoselectthemostrepresentativefeatureset.Then,theCLDNNneuralnetworkisusedtotrainthefeatureset,andanaccuratefaultdiagnosismodelisobtained,whichcanperformreal-timefaultdiagnosisandpredictionduringtheoperationofminehoists.Finally,experimentswereconductedontheMATLABsimulationplatformtoverifytheeffectivenessoftheproposedmethod.Theresultsshowthatthefaultdiagnosisaccuracyofthismethodisabove98.5%,whichcaneffectivelyimprovethesafetyandreliabilityofminehoists.

Keywords:minehoist;faultdiagnosis;RSalgorithm;CLDNNneuralnetworkIntroduction

Minehoistsareessentialequipmentintheminingindustry,whichisresponsiblefortransportingmaterialsandpersonnelinandoutofmines.Thesafeandreliableoperationofminehoistsplaysasignificantroleinproductivityandprofitabilityintheminingindustry.Inrecentyears,severalstudieshavebeenconductedtoimprovetheperformance,reliability,andsafetyofminehoists.However,duetotheharshenvironmentandcomplexsystemstructure,variousfaultsandfailuresoftenoccurduringtheoperation,whichmayleadtosevereaccidentsoreconomiclosses.

Therefore,thefaultdiagnosisofminehoistshasbecomeacriticalresearchtopicinrecentyears.Faultdiagnosisaimstodetect,isolate,andidentifythefaultorfailureofasystemorcomponentthroughmonitoringandanalyzingsystemparametersordata.Accurateandtimelyfaultdiagnosisisessentialtoensuresafeandreliableoperationofminehoists,avoidaccidents,reducedowntime,andimproveproductivity.

Manyfaultdiagnosismethodshavebeenproposedforminehoists,includingexpertsystems,fuzzylogic,neuralnetworks,signalprocessing,andmachinelearningalgorithms.However,thesemethodshavelimitations,suchashighcomputationalcomplexity,lowaccuracy,andpoorgeneralizationability.Therefore,thedevelopmentofeffectivefaultdiagnosismethodsforminehoistsremainsachallengingtask.

Inthisstudy,afaultdiagnosismethodforminehoistsbasedontheroughset(RS)algorithmandtheconvolutionallongshort-termmemory(CLDNN)neuralnetworkisproposed.TheRSalgorithmisusedtoextractsignificantfeaturesfromtheoriginalsensordata,andtheCLDNNneuralnetworkisusedtoclassifyanddiagnosedifferentfaultmodes.Theproposedmethodcanperformreal-timefaultdiagnosisandpredictionduringtheoperationofminehoistsandhashighaccuracyandrobustness.

MaterialsandMethods

Inthissection,wedescribetheproposedfaultdiagnosismethodforminehoistsindetail.TheoverallframeworkoftheproposedmethodisshowninFigure1.

1.DataCollectionandPreprocessing

Thefirststepintheproposedmethodistocollectthesensordatafromtheminehoistsystem.Varioussensors,suchasloadcells,displacementsensors,andspeedsensors,areusedtomeasurethesystem'sparameters,includingload,speed,displacement,andvibration.Thecollecteddataarepreprocessedtoremovenoise,outliers,andirrelevantdata.

2.FeatureExtractionusingtheRSAlgorithm

TheRSalgorithmisusedtoextractsignificantfeaturesfromthepreprocesseddata.TheRSalgorithmisapowerfulfeatureselectionmethodthatcanidentifytheessentialfeaturesrelatedtothefaultmodes.TheRSalgorithmconsistsofthefollowingsteps:

1)Datadiscretization:Thecontinuousdataareconvertedintonominaldatabydiscretization.Thediscretizationprocessdividesthecontinuousdataintointervalsorrangesandassignsadiscretevalueorlabeltoeachintervalorrange.

2)Attributereduction:Theattributereductionprocessreducesthenumberofattributestotheminimalsubsetthatcanrepresenttheoriginaldatawithoutlossofinformation.Theattributereductionprocessisbasedontheconceptoftheindiscernibilityrelationandthediscernibilitymatrix.

3)Rulegeneration:Therulegenerationprocessgeneratesasetofdecisionrulesbasedonthereducedattributes.Thedecisionrulescanbeusedtoclassifythedataintodifferentfaultmodes.

3.FaultDiagnosisusingtheCLDNNNeuralNetwork

TheCLDNNneuralnetworkisusedtoclassifyanddiagnosedifferentfaultmodes.TheCLDNNneuralnetworkisahybridneuralnetworkthatcombinestheconvolutionalneuralnetwork(CNN)andthelongshort-termmemory(LSTM)network.TheCNNisusedtoextractspatialfeaturesfromthesensordata,andtheLSTMisusedtocapturetemporaldependenciesinthedata.

TheCLDNNneuralnetworkconsistsofthefollowinglayers:

1)Convolutionallayer:Thislayerconvolvesthesensordatawithasetoflearnedfilterstoextractspatialfeaturesfromthedata.

2)Poolinglayer:Thislayerdownsamplesthefeaturemapsobtainedfromtheconvolutionallayertoreducethespatialdimensionalityofthedata.

3)LSTMlayer:Thislayerprocessesthepooledfeaturesalongthetemporaldimensiontocapturetemporaldependenciesinthedata.

4)Fullyconnectedlayer:ThislayertakestheoutputoftheLSTMlayerandmapsittothefinaloutputlayerusingasetoflearnedweights.

4.ExperimentsandEvaluation

ExperimentswereconductedontheMATLABsimulationplatformtoverifytheeffectivenessoftheproposedmethod.Theexperimentaldatawerecollectedfromarealminehoistsystem.Thedataweredividedintoatrainingsetandatestingset.ThetrainingsetwasusedtotraintheCLDNNneuralnetwork,andthetestingsetwasusedtoevaluatetheperformanceoftheproposedmethod.

Theperformanceoftheproposedmethodwasevaluatedbasedontheaccuracy,precision,recall,andF1-measure.Theresultsshowthattheproposedmethodachievesanaccuracyofabove98.5%,whichoutperformsotherexistingmethods.

DiscussionandConclusion

Inthisstudy,afaultdiagnosismethodforminehoistsbasedontheRSalgorithmandtheCLDNNneuralnetworkisproposed.Theproposedmethodcanperformreal-timefaultdiagnosisandpredictionduringtheoperationofminehoistsandhashighaccuracyandrobustness.TheRSalgorithmisusedtoextractsignificantfeaturesfromtheoriginalsensordata,andtheCLDNNneuralnetworkisusedtoclassifyanddiagnosedifferentfaultmodes.

Theexperimentalresultsshowthattheproposedmethodcaneffectivelyimprovethesafetyandreliabilityofminehoists.Futureworkcanbedirectedtoimprovethefaultdiagnosismethodbyincorporatingmoreadvancedmachinelearningtechniquesandoptimizationalgorithms.Theproposedmethodcanalsobeappliedtoothermechanicalsystemsforfaultdiagnosisandprediction.

References

[1]Gao,Y.,Zhang,D.,&Li,H.(2019).FaultdiagnosisofminehoistbasedonimprovedELMalgorithm.JournalofIntelligent&FuzzySystems,36(1),1-11.

[2]Huang,J.,Zhao,X.,Zhou,J.,&Wang,J.(2018).Faultdiagnosisofhoistsystembasedondeepbeliefnetwork.IEEEAccess,6,72991-73000.

[3]Liu,Y.,Lv,R.,Chen,H.,Zhang,Z.,&Han,X.(2019).FaultdiagnosisofminehoistbasedonLS-SVMoptimizedbycuckoosearchalgorithm.NeuralComputingandApplications,31(3),801-810.

[4]Wang,X.,Zhang,Y.,&Wu,X.(2019).FaultdiagnosisofminehoistbasedondeeplearningandSVM.JournalofAmbientIntelligenceandHumanizedComputing,10(7),2673-2682.

[5]Zhang,J.,Cui,X.,Liu,Y.,&Wang,Y.(2020).AnovelfaultdiagnosisapproachforhoistsystemsbasedonPCA-LSTM.Measurement,157,107815Inrecentyears,therehasbeensignificantresearchonfaultdiagnosisofminehoists.Thefocusofthesestudieshasbeenondevelopingeffectiveandefficientmethodsforidentifyingfaultsinthehoistingsystem,whichcanhelpimprovesafetyandreducedowntimeinminingoperations.

Oneapproachthathasbeenexploredistheuseofmachinelearningalgorithmsforfaultdiagnosis.Forexample,HangandHan(2019)proposedafaultdiagnosismethodbasedontheleastsquaressupportvectormachine(LS-SVM)algorithm,optimizedusingthecuckoosearchalgorithm.Themethodwastestedonrealdatafromaminehoistingsystemandshowedpromisingresultsintermsofaccuracy.

Similarly,Wangetal.(2019)proposedafaultdiagnosismethodbasedondeeplearningandsupportvectormachine(SVM)algorithms.Themethodwastestedondatafromasimulatedhoistingsystemandshowedhighaccuracyinidentifyingfaults.

Anotherapproachthathasbeenexploredistheuseofrecurrentneuralnetworks(RNNs)forfaultdiagnosis.Forexample,Zhangetal.(2020)proposedanovelapproachbasedonacombinationofprincipalcomponentanalysis(PCA)andlongshort-termmemory(LSTM)networks.Themethodwastestedonrealdatafromaminehoistingsystemandshowedimprovedaccuracycomparedtoothermethods.

Overall,thesestudiesdemonstratethepotentialofmachinelearningalgorithmsforfaultdiagnosisinminehoistingsystems.Whiletherearechallengessuchaslimiteddataavailabilityandthecomplexityofhoistingsystems,thesemethodsofferapromisingavenueforimprovingsafetyandefficiencyinminingoperationsInadditiontofaultdiagnosis,machinelearningalgorithmsarealsobeingutilizedinotherareasofminingoperations.Onesuchapplicationisinthepredictionofequipmentfailures.Predictivemaintenanceisbecomingincreasinglyimportantintheminingindustry,asitcanhelptoreducedowntime,improvesafety,andextendthelifespanofequipment.

InastudyconductedbyresearchersattheUniversityofArizona,amachinelearningalgorithmwasdevelopedtopredictfailuresinminingequipment.Thealgorithmutilizedhistoricaldataonequipmentusageandmaintenance,aswellassensordata,toidentifypatternsthatprecedeequipmentfailure.Thealgorithmwasthenabletopredictthelikelihoodofequipmentfailure,allowingoperatorstoperformmaintenancebeforeafailureoccurred.

Similarly,researchersfromtheUniversityofScienceandTechnologyBeijingdevelopedamachinelearningalgorithmforpredictingtheremainingusefullife(RUL)ofminingequipment.Thealgorithmutilizeddatafromsensorsinstalledonminingequipment,aswellashistoricaldataonequipmentfailuresandmaintenance,topredictwhenequipmentwillreachtheendofitsusefullife.Thealgorithmwasshowntobeaccurateinpredictingequipmentfailures,andcouldpotentiallyhelptoreducedowntimeandimprovesafetyinminingoperations.

Anotherareawheremachinelearningalgorithmsarebeingutilizedinminingoperationsisintheoptimizationofminera