Multiscale Graph Attention Framework under Multigraph Learning for Rotating Machinery Fault Diagnosis

To address the reliance on a single graph structure in graph-based fault diagnosis methods, this paper proposes a multiscale graph attention framework under multigraph learning for rotating machinery fault diagnosis. First, we introduce a multiscale feature extraction module to mine hierarchical features from raw vibration signals, enabling the model to capture both local and global fault-related patterns. To further enhance representation diversity, a multigraph learning strategy is employed to fuse complementary information derived from different graph construction methods. Moreover, a graph attention network is adopted to assign varying importance to graph nodes, allowing the model to focus on the most critical components for fault diagnosis. This approach not only improves feature representation but also enhances the generalization performance of the fault diagnosis framework. Comprehensive experiments based on real-world datasets validate the superiority of the proposed framework over existing methods.