Fault Diagnosis for Power Converters based on Random Forests and Feature Transformation

Abstract

Power converters have been key enablers of many energy conversion fields, and it is a trend to apply artificial intelligent (AI) technology to power converters to improve stability. A novel fault diagnosis method based on the combination of random forests (RFs) and feature transformation is proposed in this paper. Firstly, the three-phase AC fault currents of three-phase PWM rectifier are analyzed as examples. Secondly, the feature transformation, a novel current trajectories slopes based method, is adopted to transform the fault currents data. With the help of feature transformation, the fault diagnosis classifier can obtain a good load adaptability. And then the RFs based method, a data-driven method, is employed to train the fault diagnosis classifier with the fault current trajectories slopes data. Finally, the proposed method is carried out on an three-phase PWM rectifier system, which can detect and locate the open-circuit faults of IGBTs.