Bearing Fault Diagnosis Based on Fourier Convolution Kolmogorov-Arnold Networks

Rolling bearings are critical yet vulnerable components in rotary machinery. Its malfunction hinders industrial production and equipment operation. Despite the significant success of Multilayer Perceptron (MLP) and Deep Learning (DL) in bearing fault diagnosis, the issue of poor model interpretability remains. Recent Kolmogorov-Arnold Networks (KAN) springs up, which enhances the interpretability of DL models. This paper explores the application of KAN in bearing fault diagnosis and proposes an improved method for one-dimensional convolutional Fourier KAN layer. Experimental results demonstrate that the improved one-dimensional convolutional Fourier KAN model performs well in fault diagnosis.