Data-based incipient actuator fault detection and diagnosis for three-phase PWM voltage source inverter

Abstract

A new incipient fault detection and diagnosis (FDD) method is proposed for three-phase PWM voltage source inverters, based on the combination of wavelet and principal component analysis (PCA). Firstly, a mathematical inverter model in the dq coordinate system is derived, based on which, a typical incipient actuator fault modeled as an exponential function is injected into the system. Considering that there exist highly-coupled strong-gain noises in the inverters, wavelet analysis is used for denoising, and then PCA is used for online FDD. Simulation results conducted on an inverter show that the proposed method is sensitive to this type of incipient actuator fault in the inverter, and has better performance than the traditional PCA based FDD method.