Novel Semi-Supervised Multi-Scale Graph Convolution Autoencoders for Industrial Fault Diagnosis

This paper presents a novel semi-supervised multi-scale graph convolution autoencoder network for industrial fault diagnosis. The proposed model aims to address the challenge of limited labeled data in industrial processes by leveraging both labeled and unlabeled data. The model consists of a multi-scale feature extraction network, a multi-scale graph structure learning module, and a mix graph convolution network. Two types of models, MFP-GCAN and PMF-GCAN, are proposed, each with different optimization methods (frozen and cross optimization). The models are evaluated on industrial process data, demonstrating superior performance in fault diagnosis compared to existing methods. The results show that the proposed semi-supervised models effectively utilize unlabeled data to improve fault diagnosis accuracy.