Abstract：In order to realize the location of the inverter circuit fault quickly and accurately, an open circuit fault diagnosis method of the switching devices in an inverter circuit based on BP neural network was proposed to reduce the downtime and to improve the operation efficiency. MATLAB was used to model and simulate the inverter circuit, and the fault signal characteristics were extracted directly from the output voltage waveforms.According to the fault features and the diagnostic target, a three-layer neural network model was established, and the numbers of neurons and the transfer functions were determined. The fault characteristic signals were used as the input of BP neural network, and the neural network was trained by Levenberg Marquardt algorithm. The trained fault diagnosis model could be realized to diagnosis the faults of the inverter circuit. The simulation results showed that the direct waveform sampling was simple, and the faults of one or two transistors could be located simultaneously and accurately. The diagnosis accuracy of the proposed fault diagnosis model was high.

Keyword：fault diagnosis;open circuit;inverter;BP neural network;