Open-Circuit Fault Diagnosis for Three-Phase Inverters Using a Step-GA-BP Algorithm

Power converters based on power semiconductor switches (e.g. IGBTs) are widely used in power grids, ships, unmanned vehicles, etc. Open-circuit faults of the switches, which are featured with concealment, must be diagnosed in time to prevent serious losses. In this paper, a step-GA-BP algorithm for the open-circuit fault diagnosis is proposed, and then validated using a typical three-phase two-level inverter. A Simulink model of the inverter is built, by which the output phase voltages are simulated and collected as signal samples to train a neural network diagnosis model. The conventional BP neural network is combined with the genetic algorithm (i.e. the step-GA-BP algorithm) to optimize the training process. The proposed algorithm achieves high diagnosis accuracy and shows advantages over the compared algorithms, in addition, the effects of duration and number of the signal samples are analyzed.