A Semi-supervised Fault Diagnosis Method Based on Graph Convolution for Few-shot Fault Diagnosis

In practical bearing fault diagnosis, labeled fault data are difficult to obtain, and limited samples will lead to training overfitting. To address the above problems, a semi-supervised fault diagnosis method based on graph convolution is proposed. Firstly, the KNN graph construction method based on Euclidean distance (ED-KNN) is used to achieve label propagation. Then, a graph convolutional network framework based on dot product attention mechanism (GPGAT) was constructed to enhance the weights of high similarity nodes and diagnose bearing faults. The proposed method is validated on a public bearing dataset. The results show that the proposed method can make full use of very few labeled samples for fault diagnosis. Compared with other state-of-the-art methods, the proposed method achieves better diagnosis performance.