A Data-Model Linkage Framework for Eccentricity Fault Diagnosis of Permanent Magnet Synchronous Motor

As a typical mechanical fault in electric machines, rotor eccentricity in permanent magnet synchronous motor (PMSM) can cause distortion in the air-gap magnetic field, leading to nonlinear fluctuations in stator current signals. This paper investigates the coupling mechanism between eccentricity faults and current signals by constructing a dynamic model of PMSM with eccentricity fault. An improved winding function method is employed to achieve high-precision modeling of the motor's air-gap spatial distribution, based on which a dynamic model under eccentric conditions is developed to generate fault data. To address the limited availability of fault samples in real-world applications, a data-model linkage framework is developed by combining virtual simulation data and physical experiment data. Domain adaptation based on Maximum Mean Discrepancy is used to align the feature distributions between the simulation (source) and experimental (target) domains. In addition, a margin-aware regularization term is introduced to enhance model discriminability under highly imbalanced data conditions. Finally, the proposed method is validated on a self-developed PMSM eccentricity fault test platform. Experimental results demonstrate that, even when eccentric fault samples constitute only 4.8% of the total dataset, the method achieves an average diagnostic accuracy of 92.5%, confirming its robustness and practical application potential.