Fault Diagnosis Method for Rotating Machinery Based on MSCNN-MGAT

Abstract

Currently, the field of rotating machinery fault diagnosis still faces the following problems: the inability to simultaneously focus on the performance patterns of fault features at different scales, the lack of description for complex structural relationships among features, and poor real-time performance. To address these challenges, we propose a novel fault diagnosis method based on multi-scale convolutional neural networks and multi-head graph attention networks (MSCNNs-MGATs). By combining multiscale convolutional network and multigraph attention network (GAT), the method is the first to simultaneously address the issues of multiscale feature extraction and modeling of complex relationships among features. It constructs a complete fault diagnosis framework from signal to graph structure. A large number of comparative experiments demonstrate that our method performs well in various complex industrial scenarios, achieving an accuracy of up to 98% with extremely low latency.