Bearing fault diagnosis based on data missing and feature shift suppression strategy

Abstract

To mitigate the impact of fault iconic feature shift and feature missing due to missing data values on bearing fault diagnosis, this paper proposes a fault diagnosis method based on a spatial frequency filter and a Multi-Scale feature recombination calibration network (MSRCN). First, the fault features are converted into frequency band features and feature enhancement is realized using Mel filters to weaken the effect of fault feature offset. Then, the spatial calibration module (SC) in the MSRCN is utilized to further improve the fault feature distribution and eliminate the fault feature offset problem. Next, to solve the fault feature missing problem, the remaining fault features are sampled by multi-scale reorganization using MSRCN to obtain new fault features, which overcomes the effect of fault feature missing on fault diagnosis. Finally, experiments are conducted on CWRU and XJTU-SY rolling bearing datasets to verify that the algorithm can effectively solve the fault feature offset and missing problem. Meanwhile, the experimental results prove that the algorithm proposed in this paper can realize high-precision fault diagnosis.