Digital Twin Based Fault Diagnosis of Motor with Spatio-Temporal Graph Convolutional Networks

Artificial intelligence (AI)-driven fault diagnosis is essential for the safe and efficient operation of motor. However, in practical industrial settings, it is challenging to get high-quality labeled datasets for model training. This paper proposes a fault diagnosis method using digital twin and graph convolutional networks, consisting of a virtual-real fusion module and a spatio-temporal graph convolutional networks module. The fusion module uses a Wasserstein generative adversarial network to improve virtual data quality, addressing the disparity between virtual and physical spaces. The spatio-temporal graph convo-lutional networks module, built with CNN and GCN, captures spatio-temporal features. Experiments show improved diagnostic performance with limited physical fault data, advancing AI engineering applications.