文档名：深度子领域自适应网络电机滚动轴承跨工况故障诊断
摘要：针对实际生产中旋转机械工况变化引起状态监测数据分布差异及获取待诊断样本标签困难问题,提出多尺度子领域自适应模型(MSDAM)的跨工况下滚动轴承故障诊断方法.首先,以原始振动信号作为输入,无需信号预处理及人工特征参数提取;其次,搭建多尺度卷积神经网络将已知标签样本和待诊断样本特征迁移到同一子空间,捕获具有细粒度信息的多尺度公共特征;然后,以不同的故障类型来划分相关子域,并通过局部最大均值距离(LMMD)来完成子域的适配,有效削弱不同工况同类故障特征的分布差异;最后,在三个数据集的多个迁移任务上进行试验验证.结果证明,所提MSDAM的跨工况故障诊断性能优于关注全局领域适配的迁移学习方法.
Abstract：Inreal-worldindustrialscenarios,duetovariationsinworkingconditions,thevibrationdatagatheredfrommechanicalequipmentpresentsdifferentprobabilitydistributions.Moreover,thegatheredvibrationdataisusuallyunlabeled.Therefore,theunlabeledvibrationdatamaybemisclassifiedbythetraineddiagnosticsourcemodels.Deeptransferlearningwiththeadaptationofmarginalorconditionaldistributioncanextendtheknowledgeofthetrainedmodelsinsourcedomainstoanovelbutitisrelatedtoatargetdomains.However,theonlyreductioninmarginaldistributiondiscrepancylimitsthegeneralizationabilityandperformanceofthediagnosticmodel.Besides,themosteffectivediagnosticmodelthatmatchesbothmarginalandconditionaldistributionsfailtotakefulladvantageofcomplexandsensitivefeaturesinthedataset.Toaddresstheseissues,amultiscalesubdomainadaptivemodel(MSDAM)forrollingbearingsfaultdiagnosisundercross-workingconditionsisproposed.Firstly,inordertoeliminatesignalpre-processingandmanualfeatureextractiondependencies,theoriginalvibrationsignalisusedastheinputoftheproposedMSDAM.Secondly,amultiscaleconvolutionalneuralnetworkisbuilttocapturemultiscalecommonfeatureswithfine-grainedinformationandtransferthefeaturesoflabeledsourcedataandunlabeledtargetdatatothesamesubspace.Then,thepredictionresultsofthesourceclassifierareusedasthesoftlabelsfortheunlabeledtargetdata,andtherelevantsubdomainsaredividedaccordingtodifferentfaulttypes.Finally,thelocalmaximummeandistance(LMMD)methodisemployedtoaligntheconditionaldistributionsofthesubdomainsinthecommonfeaturespacetoeffectivelyreducethedistributiondiscrepancyofsimilarfaultcharacteristicsunderdifferentworkingconditions.Thelocationselectionexperimentofthesubdomainadaptationlayershowsthatthesubdomainadaptationlayerissetbeforetheclassificationlayer,whichcanbetterguidetheoptimizationoftheentirenetworkandmaintainfeaturedistributionmatching.TheresultsonthePaderbornbearingdatasetshowthattheaveragediagnosticaccuracyofthemultiscalemodelisimprovedby7.98%comparedtothesingle-scalemodel.Despitesacrificingextratrainingtime,theproposedMSDAMhasstrongertransferability.TofurtherverifytheeffectivenessoftheproposedMSDAM,atotalof12transferlearningtasksarecarriedoutontheCWRUbearingdataset,Paderbornbearingdataset,andlaboratory-builtdatasetsrespectively.TheexperimentalresultsindicatethattheproposedMSDAMhashigheraccuracyinfaultdiagnosisacrossworkingconditionsthanthedeepadaptationmethodssuchasdomainadaptivenetworks(DAN)anddomain-adversarialneuralnetworks(DANN).Inaddition,t-SNEvisualizationofthesubdomainadaptationlayerrepresentationsindicatesthattheproposedMSDAMcanaccuratelyaligntherelevantsubdomaindistributions.Thefollowingconclusionscanbedrawnthroughtheanalysisofexperimentalresults:(1)TheproposedMSDAMcanextractmorefine-grainedcommonfaultfeaturesandalignrelevantsubdomainsbasedonLMMDcomparedwithDANandDANN.(2)TheproposedMSDAMonlyrequirestheoriginalvibrationsignalasinput,eliminatingtheneedforsignalpreprocessingandmanualfeatureextraction.(3)AhighcrossdomaindiagnosticaccuracyonallthreedatasetsprovesthestronggeneralizationabilityoftheproposedMSDAM.
作者：宋向金 孙文举 刘国海 赵文祥 王照伟Author：SongXiangjin SunWenju LiuGuohai ZhaoWenxiang WangZhaowei
作者单位：江苏大学电气信息工程学院镇江212013
刊名：电工技术学报
Journal：TransactionsofChinaElectrotechnicalSociety
年，卷(期)：2024, 39(1)
分类号：TM307
关键词：轴承故障诊断 子领域自适应 迁移学习 软标签学习
Keywords：Bearingfaultdiagnosis subdomainadaptation transferlearning softlabellearning
机标分类号：TP391.41TH133.33TP27
在线出版日期：2024年1月18日
基金项目：国家自然科学基金，国家自然科学基金，江苏省自然科学基金，江苏省双创项目深度子领域自适应网络电机滚动轴承跨工况故障诊断[
期刊论文] 电工技术学报--2024, 39(1)宋向金 孙文举 刘国海 赵文祥 王照伟针对实际生产中旋转机械工况变化引起状态监测数据分布差异及获取待诊断样本标签困难问题,提出多尺度子领域自适应模型(MSDAM)的跨工况下滚动轴承故障诊断方法.首先,以原始振动信号作为输入,无需信号预处理及人工特征参...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
深度子领域自适应网络电机滚动轴承跨工况故障诊断 Across Working Conditions Fault Diagnosis for Motor Rolling Bearing Based on Deep Subdomain Adaption Network
深度子领域自适应网络电机滚动轴承跨工况故障诊断.pdf
- 文件大小:
- 14.06 MB
- 下载次数:
- 60
-
高速下载
|
|
