Open-circuit fault diagnosis method for switching tube of mine NPC three-level inverter

The inverter of the motor drive system in the mine hoist and belt conveyor mostly adopts neutral point clamped (NPC) three-level inverter. This inverter has a large number of switching tubes and high running frequency. Switching the working state of the switching tubes at high frequency in a short time and in complex working environment are prone to open-circuit fault. The fault signal has non-stationary characteristics. The existing fault diagnosis method for switching tube of NPC three-level inverter has the problems of difficult fault feature extraction, large calculation amount, and low fault accuracy. In order to solve the above problems, an open-circuit fault diagnosis method for switching tube of mine NPC three-level inverter based on probabilistic neural network (PNN) is proposed. Firstly, the phase voltage signals of three-phase of inverter are collected by oscilloscope. The phase voltage signals are processed by denoising and normalization. Secondly, the three-phase voltage is converted into two-phase rotating (d-q) coordinate system voltage by Clark transform and Park transform. The d-axis voltage is decomposed into multiple intrinsic mode function (IMF) using empirical mode decomposition (EMD). For different open-circuit faults, the variance contribution rate of each IMF is calculated. The variance contribution rates of the second, third and eighth IMF differ greatly. The three IMF represent different open-circuit faults. The mean, mean square and variance of the second, third and eighth IMF are calculated as the open-circuit fault feature vector of the inverter switching tube. Finally, the feature vector is input into the PNN for training and classification. The open-circuit fault diagnosis of the NPC three-level invert switching tube is realized. The experimental results show that the fault diagnosis method based on PNN has higher fault diagnosis accuracy than the fault diagnosis method based on CNN and SVM, and the average fault diagnosis accuracy reaches 97.75%.