Open-Circuit Fault Diagnosis Based on Deep Learning for Four-Level Active Neutral-Point Clamped Inverters

In this paper, a deep learning-based fault diagnosis has been proposed for improving reliability when detect faulty switches of four-level active neutral point clamped (ANPC) inverters. The proposed method detects an open-circuit fault in a single switching device by training a 1-D Convolutional Neural Network (1-D CNN) to extract the periodic feature of pole voltage, which can be done without adding any algorithm for feature extraction. Due to the increased number of switches and conduction paths compared to a two-level inverter, the other method requires complex mathematical equations, threshold settings, and theoretical knowledge of the topology. However, the proposed data-driven deep learning-based method can be made simpler and more robust under different system conditions without using mathematical equations. Simulation results have validated the feasibility of the proposed method that show the diagnostic average accuracy rate reached 99.53 % with conditions that were untrained in the deep learning model.