Fault detection and isolation of inverter based on FFT and neural network

The paper presents a approach based on FFT and neural network to detect and isolate in inverter. The positive sequence symmetrical component of the inverter output is obtained by windowing FFT, and the concept of spectral residual and relative spectral residual are presented. Firstly, the fundamental residual spectral is computed by FFT with fixed width window function, and the switch fault occurred in the inverter is detected. Secondly, an simple judge strategy for locating the faulty bridge with switch fault is proposed by using spectral residual and its phase, and the inverter bridge with switch fault are positioned. Thirdly, the fault switch is isolated using neural network. The simulation results show that the method can detect and isolate the fault effectively.