不均衡样本下轴承故障的LSGAN-SwinTransformer诊断方法

打开文本图片集

关键词：故障诊断；滚动轴承；不均衡样本；最小二乘生成对抗网络；Swin Transformer 中图分类号：TH165.3；TH133.33 文献标志码：A DOI：10.16385/j.cnki.issn.1004-4523.202308023

Abstract：Aiming atthe problems bearings workingincomplex environments，where faultdata aredificult toobtain in large quantitiesandtheseriousimbalancebetweentherationormaldataandfaultdataresultingininsuficienti-depthmodeltraining andlowdiagnosticacuracy，abearing fault diagnosis methodbasedonLSGAN-Swin Transformeris proposed.Theleast-squares generativeadversarialnetwork isutilizedtoexpandtheimbalancedorlackbearingdataset，andthewindowedself-atentivenet work is introducedforbearingfaultstate identificationTheproposed methodisvalidatedbyusingtwodatesets，andcompared with SGANand WGANrespectively.It isdemonstrated thatLSGANgenerates data training models with higheraccracy.The proposed Swin Transformer（Swin-T） model is compared withCNN，AlexNet and SqueezeNet undersmallsampleconditions， and the accuracy is improved by 34.85% ， 13.45% ，and 12.95% ，respectively.The classification effect the model is evaluated byt-SNEvisualizationndtheresultsshowthat theLSGAN-Swin-Tmodelcanstillmettherequirementsinfaultdiagosiset terwhenthenumber trainingsamplesissmall，whichprovidesanewideafortheresearchbearingfaultdiagnosisundernbalanced data.

Keywords：faultdiagnosis；rolingbearings；unbalancedsample；leastsquaregenerativeadversarialnetwork；shiftedwindows transformer（Swin Transformer）

滚动轴承在各类机械设备中具有重要作用，在实际工程应用中，由于设备运行数据监测过程的工作环境复杂，信号一般呈现非平稳及非线性的特点，且部分状态数据难以大量采集，正常数据样本通常远远大于故障数据样本，数据之间比例严重失衡。（剩余19110字）