噪声干扰下基于PCA-SF的轴承故障诊断方法.pdfVIP

噪声干扰下基于PCA-SF的轴承故障诊断方法

摘要：机械故障诊断对降低维修成本和预防事故至关重要。振动信号监测是机械故障诊断中一种有效可行的方

法。然而，所采集故障信号往往容易受到其他设备噪声的干扰。因此，从受噪声干扰的监测信号中提取与故障相关的

周期脉冲是故障诊断的基础，也是难点。为解决此问题，提出一种基于主成分分析(PrincipalComponentAnalysis，PCA)

和稀疏滤波(SparseFiltering，SF)的机械故障特征提取方法。具体来说，首先利用PCA提取噪声干扰信号段的主成分，

然后利用SF从主成分中提取有效特征。为减小SF模型的过拟合问题，采用L1/2范数对其目标函数进行正则化约束。

最后，将提取的特征输入到Softmax分类器中进行故障识别。分别通过一组仿真和实验案例对所提PCA-SF方法的有

效性进行验证。实验结果表明，该方法不仅能准确实现故障分类，而且优于其他传统方法。

关键词：故障诊断；噪声干扰；主成分分析；稀疏滤波

FaultDiagnosisofBearingsBasedonPCA-SFunderNoiseInterference

Abstract:Mechanicalfaultdiagnosisisveryimportantforreducingmaintenancecostandpreventingaccidents.

Vibrationsignalmonitoringisaneffectiveandfeasiblemethodformechanicalfaultdiagnosis.However,thecollectedfault

signalscanbeinterferedeasilybytheexternalnoisefromtheotherequipment.Therefore,itisthekeyaswellasthe

difficultyoffaultdiagnosistoextracttheperiodicpulserelatedtothefaultsfromthenoiseinterferedmonitoringsignals.To

solvethisproblem,thispaperproposedamechanicalfaultfeatureextractionmethodbasedonPrincipalComponentAnalysis

(PCA)andSparseFiltering(SF).Specifically,PCAisusedtoextracttheprincipalcomponentsofnoiseinterferedsignal

segments.AndthenSFisusedtoextracteffectivefeaturesfromtheprincipalcomponents.Inordertoreducetheover-fitting

problemofSFmodel,theL1/2normisusedtoregularizetheobjectivefunction.Finally,theextractedfeaturesareinputinto

Softmaxclassifierforfaultidentification.TheeffectivenessoftheproposedPCA-SFmethodisverifiedbyasetof

simulationandexperimentalcases.Experimentalresultsshowthattheproposedmethodcanachieveaccuratefault

classification,andalsooutperformothertraditionalmethods.

Keywo