Multimodal Domain Generalization for Inter-Turn Short-Circuit Fault Diagnosis of Permanent Magnet Synchronous Motors

The current characteristics of inter-turn short-circuit (ITSC) faults in permanent magnet synchronous motor (PMSM) typically exhibit continuous evolution characteristics with changes in operating conditions. However, industrial data acquisition usually only covers limited and discrete operating conditions, making it difficult to fully characterize fault evolution information across operating conditions. To address this issue, this article proposes a multimodal domain generalization diagnostic framework. The proposed framework constructs feature representations from three physical perspectives, namely the time domain, frequency domain, and time–frequency domain, based on the same stator current signal. By introducing an operating-condition evolution prior, the proposed framework extends conventional modeling approaches that treat operating conditions as discrete domains to a continuous evolution representation, thereby recovering the progressive trajectories of fault features in the latent space. In addition, semantic-guided multimodal feature fusion and a hierarchical domain alignment mechanism are jointly employed to mitigate modality heterogeneity and cross-condition distribution shifts. Experimental results on both private and public datasets demonstrate that the proposed method maintains stable and accurate diagnostic performance under unseen operating conditions, validating its effectiveness for cross-condition ITSC fault diagnosis in PMSM.