Research on bearing fault diagnosis method based on transformer neural network

Abstract

Abstract The Attention mechanism (AM) has been widely used for fault diagnosis and identifying the health of industrial equipment. Existing research has only used AM in combination with deep networks, or to replace certain components of these deep networks. This reliance on deep networks severely limits the feature extraction capability of AM. In this paper, a bearing fault diagnosis method is proposed based on a signal Transformer neural network (SiT) with pure AM. First, the raw one-dimensional vibration time-series signal is segmented and a new segmented learning strategy is introduced. Second, linear encoding and position encoding are performed on the segmented subsequences. Finally, the encoded subsequence is fed to the Transformer for feature extraction to achieve fault identification. The validity of the proposed method is verified using the Case Western Reserve University dataset and the self-priming centrifugal pump bearing dataset. Compared with other existing methods, the proposed method still achieves the highest average diagnostic accuracy without any data preprocessing. The results demonstrate that the proposed SiT based on pure AM can extract features and identify faults from the raw vibration signal, and has superior diagnostic performance.