Open-circuit Fault Diagnosis Based on 1D-CNN for Three-phase Full-bridge Inverter

Abstract

The inverter is one of the most important parts of the motor driver system which also the weak link prone to failure. Its reliability and stability are critical to the smooth operation of the motor drive system and the entire system. By performing fault diagnosis on the detection of the inverter and appropriate treatment in time, the reliability of the motor drive system can be greatly improved. The traditional fault diagnosis method has the problems of low fault recognition rate and long recognition time. A novel fault diagnosis method based on one-dimensional convolutional neural network (1D-CNN) for three-phase full-bridge inverter is proposed, which combines the two parts of adaptive fault feature extraction and fault classification into one. The fast Fourier transform (FFT) is used to convert the three-phase current signals into an amplitude spectrum. The signal is divided into training sets and testing sets. The training sets is used to train the weights and bias of the 1D-CNN, and the testing sets is used to check the accuracy of the network classification. In the final experiment, the effectiveness of this network was proved by comparison with Multi-Layer Perceptron (MLP) and Stacked Denoising Auto Encoder (SDAE). Experiments show that the diagnostic accuracy of 1D-CNN is better than 99.21%, which is significantly higher than that of MLP and SDAE. 1D-CNN will have better development prospects in the sphere of fault diagnosis, which can set a good foundation for the Prognostics Health Management (PHM) technology of power electronic systems.