PMSM High-Resistance Connection Fault Diagnosis via Deep Feature Fusion of Multisensor Signals

Permanent magnet synchronous motors (PMSMs) are critical components in electric vehicles, where early and accurate diagnosis of high-resistance connection (HRC) faults is essential for system reliability. This research paper proposes a novel fault diagnosis framework, the multichannel recurrence fusion vision transformer (MRF-ViT), which leverages a multichannel recursive fusion strategy for signal transformation and a lightweight vision transformer for classification. First, three-phase current signals are converted into two-dimensional spatial maps using a Modified Recurrence Plot (MRP) method. These maps are then adaptively fused across channels to enhance the representation of fault characteristics. Second, the fused images are processed by the MRF-ViT model, which integrates a squeeze-excitation (SE) attention module to efficiently capture both local and global fault features. Experimental results demonstrate that the proposed MRF-ViT framework achieves a superior classification accuracy of 99.75% for identifying HRC fault types and their severity, outperforming other deep learning models such as CNN, ResNet, ConvNeXt, and MobileNet. Furthermore, robustness tests under various noise conditions and validation on edge computing platforms confirm the method's practicality and potential for real-world industrial applications.