PMSM Fault diagnosis based on Chebyshev-BiGRU with Cross-Modal Multiscale Transformer Fusion

In this paper, a fault diagnosis method based on Chebyshev-Bidirectional Gated Recurrent Unit (BiGRU) with Cross-Modal Multiscale Transformer Fusion (CBiGRU-CMMTF) model is propoesd for Permanent magnet synchronous motor (PMSM). To simultaneously address the challenges of multimodal feature extraction, CBiGRU-CMMTF innovatively integrates the Chebyshev Network and Bidirectional Gated Recurrent Units (CBiGRU), followed by a cross-model multiscale Transformer fusion (CMMTF) module. First, for CBiGRU-CMMTF model, Chebyshev Network with adjacency matrices constructed from K-nearest neighbor is applied to extract spatial fault features. Critical dependencies in temporal fault features are captured through BiGRU with the GlobalAttention mechanism. The CMMTF module enables comprehensively spatiotemporal feature integration through multimodal alignment, interaction, and dynamic aggregation. Then, the fault diagnosis based on CBiGRU-CMMTF model is implemented. Simulation results on PMSM datasets validates the effectiveness of the proposed model.