Application of wavelet kurtosis convolutional neural network model in rolling bearing fault diagnosis

Aiming at the problem of substantial noise in the actual industrial environment, where the intelligent diagnostic model exhibits weak anti-noise performance and the interpretability of the extracted features is not strong, the wavelet kurtosis convolutional neural network model integrating signal processing technology and deep learning neural networks has been proposed. Firstly, the standard CNN convolution kernel has been replaced with a wavelet basis function to form a wavelet convolution layer, and the input signal undergoes decomposition into various frequency components via the wavelet filter in this layer; Secondly, a kurtosis loss layer has been used to filter the high kurtosis value components that contain more fault features, so as to achieve the filtration of the noise and the accurate localisation of the fault signals; Lastly, the efficacy of the proposed model has been verified in the rolling bearing dataset and diagnostic model. The results showed that the model can reach 90% diagnostic accuracy under -4db Gaussian white noise interference, which was higher than other diagnostic models, and had good generalisation and noise resistance.