Fault diagnosis of inverter open circuit with improved residual network
An improved two-dimensional convolutional neural network optimization method for fault diagnosis is proposed to address the problems of low accuracy and poor robustness of the traditional open-circuit fault diagnosis method for three-phase voltage source inverters.The method first introduces a new data preprocessing method to convert the original time-domain voltage signal data into two-dimensional grayscale images by Markov transition fields(MTF),which effectively preserves the spatio-temporal relationships of the features.Then,a parallel attention mechanism is proposed to filter the features extracted by the ResNet18 feature extraction layer of the convolutional neural network for channel and spatial features,respectively,and complete the effective feature fusion.Finally,the fused features are obtained as fault classification results by ResNet18 fully connected layer and output layer.The experimental results show that the proposed improved fault diagnosis method can improve the diag-nosis accuracy to 99.80%,and maintain more than 90%classification accuracy under different noise conditions,which verifies that the method can effectively improve the performance and robustness of inverter open-circuit fault diagnosis.
万方数据
维普
