Multi-Connected Temporal Encoding Graph Neural Network for Complex Industrial Fault Diagnosis

In modern industry, the complex system scale is large and the process variables are highly coupled. Traditional fault diagnosis methods are inherently challenged in their ability to effectively integrate information, leading to inappropriate feature representations and inaccurate diagnosis outcomes. To address these issues, this paper proposes a Multi-Connected Temporal Encoding Graph Neural Network (MCTEGNN). For multi-connected graph construction, by additionally considering the correlation between different sensors at different timestamps and connecting sensors between all timestamps, a multi-connected graph can be formed to achieve comprehensive dependency modeling. Multi-connected graph convolution captures local dependencies by utilizing moving windows and time pooling, and then learns advanced features to use the moving pool GNN. The proposed method improves the problem of incomplete modeling caused by low efficiency in capturing data relationships. The experimental results on the three dataset demonstrate the effectiveness of the proposed MCTEGNN for faults diagnosis