Fault diagnosis of inverter-fed motor drives based on neural network

Aiming at more switch faults occurred in the inverter and difficulties in isolating the faults using conventional diagnosis technique, this paper investigates the use of neural network for fault diagnosis in inverter-fed motor drive systems. The inverter voltage signal is acquired, and the relative spectral residue is obtained by analyzing the spectral of the inverter signal. All these residues can be viewed as fault features and sent to the network for training. Observing the outputs of the network, the faults can be isolated. The proposed technique can isolate the open-switch and short-switch faults occurred in the inverter. Simulation result shows that the method is effective and the diagnosis is accurate. The study lays a theoretical foundation for fault tolerant drive system.