Fault detection and diagnosis of inverter based on spectral estimation and neural network

Low reliability of variable frequency drive is due to switch faults occurred in the inverter. The paper investigates the use of spectral estimation and neural network for fault detection and diagnosis in inverter-fed motor drives. Comparing voltage signal of actual and ideal inverter without faults, an equation for estimating spectral residual is derived. The spectral residual of the signal is obtained in real time by using complex recursive least square algorithm. A simple fault decision strategy is proposed and fault occurred in the inverter is detected in real time. Being treated spectral residual properly, relative spectral residual is obtained. Multilayer perceptron neural network is used to isolate switch faults combing with spectral estimation. The simulation result shows that the method can detect and isolate the faults effectively.