基于改进多尺度卷积网络的轴承故障诊断研究

打开文本图片集

中图分类号：TH133.3 文献标识码：A 文章编号：2096-4706（2025）07-0179-07

Abstract： In this paper，Improved Multi-Scale Convolutional Networksbearing fault diagnosis method is proposed to solvethe problemsofConvolutionalNeuralNtworkincomplexevironments，suchassytobedisturbed，difculttoetract rich fault featuresfromfixedreceptivefeldandlowdiagnosisaccuracy.Firstly，theoriginalvibrationsignal ispreproced. Secondly，theconvolutionkerelsofdifrentreceptivefieldsareusedtoextractmulti-salefeaturestoeffectivelyapture diversifedfaultinformation.TrdlytheSelf-AentionMechanismisintroducedtoenablethemodeltodynamicallalculate andadjustthe weight ofeach position inthe feature map，and adaptivelyenhance the keyfault features.Finaly，the fully conected layer isused toclasifytheextracted features toachieveaccurate diagnosis.Theexperimentalresultsshowthatthe diagnosis accuracy of the method on the public dataset reaches about 98 % ，and it shows good anti-noise and generalization ability underdifferent signal-to-noise ratio conditions.

Keywords：Multi-ScaleConvolutionalNetworks;featureextraction;Self-AtentionMechanism;bearing fault diagnosis

0 引言

随着装备制造业的发展，轴承性能直接影响设备表现[1]。（剩余10638字）