Rolling Bearing Fault Diagnosis Using CNN-based Attention Modules and Gated Recurrent Unit

Abstract

Rolling bearings are of great importance to rotating machinery. However, in real operating conditions, rolling bearings are damaged chronically by complex factors like nonuniform workload, which leads to occurrences of faults. Thus, there is necessity to recognize the bearing faults in advance. Algorithms based on deep learning (DL) have excellent feature auto-learning ability. However, many DL methods pay few attentions on channel-wise information, with less satisfactory in the results. Therefore, this paper proposed an end-to-end model named CAT-GRU. This method is composed of convolutional neural networks (CNN), channel attention mechanism (CAM) and gated recurrent units (GRU) to early recognize the bearing faults. In this structure, the CNN encodes the abstract features by multi convolution operations from the raw one-dimensional vibration data. The CAM then obtains the channel-wise fault significances from the output of CNN and weights each channel to highlight more fault-relevant features. Next, the following GRU captures time-related features before the final fault type classification. On motor bearing dataset from Case Western Reserve University, the CAT-GRU shows excellent fault diagnosis performance.