A multi-level classification method for fault diagnosis of grid-connected inverter

Abstract

This article proposes a multi-level classification fault diagnosis method for the open-circuit fault of power semiconductor; signal-stuck, gain-variation, and zero-offset of the current sensor. First, the average value, range value of current, and the mean value of absolute current signal are used to divide all faults into five major types of faults; then for power semiconductor open-circuit fault, fast Fourier transform (FFT) is used to extract the fault characteristics; The range and average value of current are selected as the fault characteristics for gain-variation and zero-offset fault, respectively. Three ELM classifiers are constructed to classify the faults for the above three types of faults. The proposed method is verified on the fault data collected in experimental platform, and achieves extremely high diagnostic accuracy. Moreover, this method can still perform well even when the gain coefficient and zero-offset of sensor change, showing its high robustness characteristics.