Inter-turn Short Circuit Fault Detection for PMSMs Based on Co-Simulation Residuals

This paper presents a novel framework for detecting inter-turn short-circuit (ITSC) faults for permanent magnet synchronous motors (PMSMs). A high-fidelity co-simulation model integrating PSpice for circuit-level motor dynamics and Simulink for field-oriented control is developed, enabling accurate ITSC fault injection and residual generation. Control and angular position residuals are derived by transfer function analysis, enhancing fault-induced harmonic features. A position-based feature extraction network is proposed to address spectral smearing under variable-speed conditions. The framework employs speed-synchronized resampling, short-time Fourier transform (STFT), and an RNN-convolutional architecture for spatiotemporal feature fusion. Experimental validation demonstrates significant improvements: residual-based detection achieves 89.4% fault detection accuracy (vs. 73.2% with raw signals), while position-based features elevate classification accuracy to 94.4% (vs. 71.4% with traditional methods). The results validate the effectiveness of model-informed residuals and position-based learning in early ITSC fault diagnosis.