Fault Diagnosis of Rotating Machinery Based on LWKConv-DRSN-FPN
In order to solve the problem that the traditional fault diagnosis methods of rotating machinery are difficult to deal with the strong noise interference and their low diagnostic accuracy,a new fault diagnosis method combining Laplace wavelet kernel convolutional layer(LWKConv),deep residual shrinkage networks(DRSN)and feature pyramid networks(FPN)is proposed.Specifically,based on the DRSN model structure,the LWKConv is constructed to extract the mutation impact features caused by faults from multi-scale by updating the scale parameter and translation parameter.FPN is introduced to fuse the deep and shallow features to strengthen the use of the shallow details of the model,and realize the fault diagnosis of rotating machinery.The research shows that the diagnostic accuracy of the proposed LWKConv-DRSN-FPN method on bearing and gear dataset can reach 100%at most,especially under the condition of strong noise interference of-4 dB,the diagnostic accuracy rate can reach 97.75%.The proposed method can effectively extract mutation impact features,and has good versatility and the ability to resist strong noise interference.
万方数据
维普
