Deep Residual Shrinkage Network With Time-Frequency Features For Bearing Fault Diagnosis

Abstract

The rolling bearing is one of key components to guarantee the safe operation of rotating machinery widely applied in industry. However, rolling bearing working in the complex environment often leads to failure, which may destroy the stability of rotating machinery and cause potential personnel safety hazards. Therefore, the precise fault diagnosis of bearings is significant to industrial system. In this paper, we proposed a fault diagnosis model based on the deep residual shrinkage network (DRSN) using the continuous wavelet transform (CWT). Firstly, the one-dimensional time-domain vibration signals collected from bearings are transformed into two-dimensional time-frequency map by CWT as the inputs of the fault diagnosis model. Then, the structure of DSRN is adjusted to be used for the classification of two-dimensional fault time-frequency maps. Moreover, the DSRN integrates a soft threshold module in each residual unit to eliminate the redundant noises in fault samples. Last, we use the bearing data from Case Western Reserve University (CWRU) to verify the effectiveness of our proposed fault diagnosis model. The experimental results demonstrate that the proposed model exhibits good fault diagnosis ability compared with the other deep neural network model in the presence of strong noise.