A Novel Variable Convolution Kernel Design according to Time- frequency Resolution Altering in Bearing Fault Diagnosis

Ordinary neural networks have achieved relatively high accuracy for bearing failure classification. However, they usually use the same size of convolution kernels to convolve the time-frequency transformed images of the bearing features. These types of kernels may not reflect the time-frequency features of wavelet transformed time-frequency maps because the time-frequency images are not consistent everywhere. To solve this problem, this paper proposes a new method for planning convolution kernels. Starting from analyzing the size of the time-frequency window corresponding to the wavelet basis function of wavelet transform, then considering the inexact theorem of wavelet transform and the actual situation, finally variable convolution kernels are designed to convolve time-frequency maps to achieve bearing fault classification. Finally, it is shown through datasets and practical experiments that the proposed method outperforms other newly proposed classification methods in terms of achieving higher accuracy in less time.