Time Graph Sub-Domain Adaption Adversarial for Fault Diagnosis

Abstract Graph Neural Network (GNN)-based fault diagnosis has attracted widespread attention. However, the existing measure metrics of domain distribution discrepancy always is single, the weight of different domains is unknown, and the scale of GNN-based feature extractor is unknown. To address these issues, a Time Graph Sub-Domain Adaption Adversarial for Fault Diagnosis (TGSDAA) is proposed in this paper. Firstly, a multi-scale time connection layer is proposed to extract the feature of the signal. Specifically, an Encode layer is proposed to construct the undirected graph. Next, a multi-receptive field cluster-graph convolution neural network is proposed to extract features of the graph. Finally, a sub-domain alignment with adversarial adaption is proposed to align different domains and achieve fault diagnosis. Two different datasets are used to verify the effectiveness of TGSDAA. The experimental results show that the average diagnosis accuracy of TGSDAA can improve 4% than other methods.