Fault Diagnosis for Multi-Phase Inverters: A Telescoping Modulus Approach With Convolutional Neural Networks

The modern power system provides an unstable working environment, which accelerates inverter failures. This brief proposes an intelligent fault diagnosis method for inverters via telescoping modulus convolutional neural networks (TMCNN). To be specific, a telescoping modulus (TM) approach is first designed to augment the output current modulus by the telescoping length, and the telescoping function then synthesizes the samples. The synthetic samples are finally trained through convolutional neural networks (CNN), and online monitoring data are recognized with trained models. Experimental results on 5-level inverter platforms validate the satisfying generalization capability and high accuracy of the proposed method.