基于多分支CNN与改进级联森林的故障诊断

打开文本图片集

中图分类号：TP391 文献标志码：A

Abstract：Infaultdiagnosis，deeplearningmodelssuchasConvolutionalNeuralNetworks（CNN）andDepForesthavedemonstrated outstandingperformance，atractingsignificantaention.However，single-branchCNsextractliitedfaultfeatures，ndthemultigrainedscainginDeepForestrequiresredesigningandajustingparametersfordiferentdatasets.Thispaperproposesahbriddep learningmodelthatcombinesamulti-branchCNNwithanimprovedcascadeforest.Firstly，amulti-branchCNwithdiferentcnvolutionalkerelsisigdtoactveeauresiaalllisfopes.codlyincEtG dient Boosting（XGBost）handlesnonlineardatabeterthanrandomforest，onerandomforestinthecascadeforest isreplacedwith XGBoost.Thispartialeplacementleveragestheadvantagesofdiferentalgoris，tiizigtheodel'soverallpeformaneFinally，ahybriddplearning modelcombines themulti-branch CNNandtheimprovedCascadeForest.Experimentsconductedonthree bearingdatasets and one rotor dataset demonstrate the proposed model's strong effctiveness in fault diagnosis.

Keywords：faultdiagnosis;CNN;cascade forest;XGBoost

对轴承和齿轮准确的故障诊断是保证设备安全运行的基础，对工业领域的持续发展至关重要[1-3]

随着技术的发展，利用计算机和人工智能诊断机械故障成为一个热门话题[4]。（剩余13882字）