Hybrid Learning Model-Based Inter-turn Short Circuit Fault Diagnosis of PMSM

This paper investigates a novel model framework that integrates supervised and unsupervised learning. Based on U-shaped network(U-net) and stacked autoencoder model (SAE), we propose a hybrid model for permanent magnet motor (PMSM) inter-turn fault diagnosis. Generally, small deep networks cannot satisfy complex diagnosis requirements, and larger scale network with high complexity is adverse to use in occasions with low computational resource. the proposed model uses unsupervised learning to assist network parameter optimisation, which makes the small supervised model prone to find the best performance parameters. In this paper, the verification of the proposed model is verified by the 3-phase current data set of PMSM collected under different loads.