Bearing fault diagnosis method based on a multi-head graph attention network. (1st July 2022)

• Record Type: Journal Article
• Authors: Jiang, Li; Li, Xingjie; Wu, Lin; Li, Yibing
• Is Part Of: Measurement science & technology. Volume 33:Number 7(2022)
• Publication Date: 2022-07-01
• Subjects: bearing -- fault diagnosis -- multi-head graph attention network -- dynamic time warping; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac56f1
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: The bearing is the core component of mechanical equipment, and attention has been paid to its health monitoring and fault diagnosis. Bearing fault diagnosis technology based on deep learning has been widely developed because of its powerful feature learning and fault classification ability. However, the traditional deep learning-based bearing fault diagnosis methods fail in mining the relationship between signals explicitly, which is beneficial to fault classification. Therefore, this paper proposes a new method based on a multi-head graph attention network (MHGAT) for bearing fault diagnosis. Firstly, it employs dynamic time warping to transform the original vibration signals into graph data with topological structure, so as to exploit the intrinsic structural information of the independent samples. Next, the graph data is input into the MHGAT, and the weights of neighbor nodes are learned automatically. Then, the MHGAT extracts the discriminative features from different scales and aggregates them into an enhanced, new feature representation of graph nodes through the multi-head attention mechanism. Finally, the enhanced, new features are fed into the SoftMax classifier for bearing fault diagnosis. The effectiveness of the proposed method is examined by two bearing datasets. The superiority of the proposed method is verified by comparison to traditional deep learning diagnosis models.