Multi-objective tree-structured Parzen estimator optimized Res-Net for ITSC fault diagnosis of PMSM

Abstract With the advancement of artificial intelligence technology, fault diagnosis methods based on deep learning have been extensively studied due to their ability to automatically extract fault latent features and develop end-to-end diagnostic models. However, the existing methods focus on achieving high accuracy while neglecting model complexity. Therefore, this paper proposes an inter-turn short circuit fault diagnosis method of permanent magnet synchronous motor using data-level fusion and multi-objective tree-structured Parzen estimator (MOTPE) optimized Res-Net. In this method, the original three-phase current signals are fused into a new modal signal through Clarke transform at the data layer. Based on an improved Res-Net18, hyperparameters are optimized using MOTPE to achieve high-performance and lightweight model design. Experiments have validated the fault diagnosis model that integrates current signal fusion and MOTPE optimization, achieving an accuracy of 99.62%, with the best noise robustness and the lowest model complexity. Compared with single-objective tree-structured Parzen estimator, multi-objective Non-Dominated Sorting Genetic Algorithm-III, and Random algorithms, MOTPE not only maintains high accuracy but also achieves lower computational costs and a lightweight network structure.