Iterative Algorithm-Induced Deep-Unfolding Networks for Interpretable Fault Detection of Permanent Magnet Synchronous Motor

Fault detection in permanent magnet synchronous motors (PMSMs) is essential for ensuring their operational reliability. However, existing high-accuracy algorithms, especially those based on neural networks, frequently face challenges such as poor interpretability and decreased accuracy in noisy environments. To address these issues, this paper proposes an interpretable fault detection method for PMSMs based on sparse representation theory, termed IAIUNet-SRC. The proposed low-dimensional over-relaxation ADMM algorithm (LOADMM) effectively reduces computational complexity by incorporating over-relaxation techniques and matrix inversion lemmas, thereby avoiding direct matrix inversion operations and enhancing convergence efficiency. Building upon LOADMM, the iterative algorithm-induced deep-unfolding network (IAIUNet) translates the iterative process of LOADMM into a layer-wise neural network structure, embedding learnable parameters to adaptively optimise performance. This design inherently preserves the interpretability of the optimisation process. Experimental results demonstrate that under noisy conditions, IAIUNet-SRC achieves a fault detection accuracy of 98.87%, representing a 2.51% improvement over the baseline method ADMM-SRC, while reducing computation time by 80%.