Rotating Machinery Fault Diagnosis Based on Spatial-Temporal GCN
Chenyang Li, Lingfei Mo, Ruqiang Yan
Abstract
Multi-sensor can provide more comprehensive and accurate information for mechanical fault diagnosis. Aiming at the weak ability of traditional artificial intelligence (AI) models to model multi-sensor signals, a method of fault diagnosis is proposed based on Spatial-Temporal Graph Convolution Network (ST-GCN) in this paper. The multi-sensor data is firstly modeled as a multivariate temporal graph. The relationship between different sensors expressed as graph topology is established adaptively using node features. Then, the spatial and temporal features are learnt simultaneously by a designed ST-GCN model. Finally, the fault type is inferred by a softmax classifier based on the entire graph representation. The diagnosis model is testified on bearing and gearbox and the results indicate effectiveness of extracting the information of multi-sensor and enhancing diagnosis performance.