An interpretable gabor convolutional neural network inverse converter open-circuit fault hybrid diagnosis method

Abstract

The application belongs to the technical field of intelligent fault diagnosis of electronic equipment, and provides a hybrid diagnosis method for open-circuit fault of an invertor based on an interpretable Gabor convolutional neural network, which comprises the following steps: first, a current residual signal is obtained by establishing a hybrid logic dynamic model; then, the one-dimensional residual signal is converted into a two-dimensional image by using a Gram angle field; further, a double-channel Gabor convolutional neural network is constructed; a Gabor filter is used to replace the first layer of convolution to enhance the feature extraction capability; fault classification is realized; finally, a class activation mapping technology is introduced to visualize the diagnosis basis in the form of a heat map and improve the interpretability of the model. The application significantly improves the diagnosis accuracy and interpretability of the open-circuit fault of a T-type three-level inverter, avoids the dependence on threshold setting in the traditional method, and is suitable for intelligent fault diagnosis of complex systems.

Images