Adaptive topology construction enhanced fault diagnosis of rotating machinery

Rotating machinery has been widely used in the industrial scenarios, and conducting fault diagnosis on it is of great significance for the safety and efficient production of equipment. As a promising method, graph convolutional network (GCN) can obtain richer features by extracting additional relevant information. Although GCN has been successfully applied to fault diagnosis of rotating machinery, most of the existing methods rely on specific criterion based on similarity and thresholds when constructing graphs. When the criterion is set inappropriately, it can lead to information loss and invalid graph topology, limiting further improvements in diagnostic performance. To address this issue, this paper proposed an adaptive topology construction enhanced fault diagnosis method. Specifically, graph topology is adaptively constructed in a learnable manner, and to avoid information loss, a non-parametric complete graph topology is also constructed. Features are extracted from the constructed graphs using graph convolution, and then the fused dual-view features are input into a classifier. Based on publicly available gear fault dataset, fault diagnosis experiments were conducted to validate the effectiveness of the proposed method. The proposed method achieves over $\mathbf{9 0 \%}$ diagnostic accuracy under different working conditions.