ResNet-18-Based Interturn Short-Circuit Fault Diagnosis of PMSMs With Consideration of Speed and Current Loop Bandwidths

A residual network (ResNet)-18-based model for the online interturn short-circuit (ITSC) fault diagnosis of permanent magnet synchronous machines (PMSMs) widely employed in traction systems of electric vehicles and locomotives is proposed in this article, which is a data-driven solution and robust to bandwidth variation in both speed and current controllers. It is found that the extracted fault features at the early stage of the ITSC are usually quite small and will be significantly influenced by the bandwidths of both speed and current controllers, which will potentially lead to a misjudgment of fault degree. Hence, an ResNet-18 deep learning model for ITSC fault is designed in this article, which has considered the variation of bandwidths of speed and current controllers by improving the input of the network. It can accurately detect and classify the fault degrees concerning adjacent levels of insulation deterioration thanks to its high performance in feature representation. Multivariate signals including speed,
$dq$
-axis currents, and reference voltages are employed as inputs with five channels to train the proposed diagnostic model, which can be easily measured from the PMSM drive system. The diagnosis results of ITSC fault on a prototype PMSM show a high average accuracy being up to 97.81%.