Open-Circuit Fault Diagnosis For Power Modules Based on Light Gradient Boosting Machine

This paper proposes a diagnosis method to detect the open-circuit faults of the switching devices in power converters of electric vehicle charging piles based on time-domain feature extraction and Light Gradient Boosting Machine (LightGBM). We find the resonant capacitor voltage is sensitive to the switch faults that occur in the rear-stage resonant converter. Therefore, we measure this voltage together with the three-phase currents of the input side and the output voltage of the DC side. This original signal with five channels is then divided into segments by a sliding window to increase the sampling number but also avoid the diagnostic lag due to long sequence calculation. A series of time-domain features, e.g. mean value, are extracted for each segment to train the fault classifier based on the high efficient LightGBM. Simulation and experiment show that the proposed method can recognize eleven different types of opencircuit faults in either single or multiple switches within half of the fundamental period, achieving a precision rate of 99.94%.