Deep Graph Convolutional Neural Network for Fault Diagnosis of Complex Industrial Processes

As industrial production processes become increasingly complex and production scales continue to expand, ensuring the safety of operational processes has evolved into an exceptionally challenging task. Industrial process data variables are interconnected and coupled, how to effectively extract the correlation information between variables has become the key to improve the fault diagnosis accuracy. To address these challenges, this paper proposes a deep graph convolutional neural network (DGCN) for fault diagnosis in complex industrial processes. Firstly, the fault data in the time domain is analysed by wavelet transform to convert the time domain data into time-frequency data. And the graph used to represent the potential interactions of industrial processes is constructed according to the Pearson correlation coefficient. At the same time, considering the weak connection between some nodes, appropriate threshold sparse edges are used to improve the anti-interference ability of the model. Secondly, a graph convolution network (GCN) is utilised for information transfer and aggregation between nodes, and the feature representation of the nodes is learnt through multi-layer graph convolution operations. Finally, the softmax classifier transforms the learned feature representation of each fault into a probability distribution to complete the fault diagnosis task. Experimental results on the Tennessee Eastman (TE) process show that the DGCN model exhibits good fault diagnosis.