Deep learning networks have developed rapidly in rotating machinery fault identification over the past several years. However, due to the poor working conditions of the rotating machinery, deep learning networks often have difficulty effectively extracting critical information that characterizes faults. To overcome this challenge, this research presents a wavelet convolutional neural network with KAN (WCNN-KAN) for rotating machinery fault diagnosis.  Firstly, the signal is turned into wavelet time-frequency graphs, and fault features are extracted by improving wavelet convolution. Secondly, design a multi-stage characteristic fusion module and a feature purification module to extract significant characteristics. Finally, introduce KAN to further improve the diagnostic capability of WCNN-KAN. The effectiveness of WCNN-KAN is verified by bearing datasets.   Experimental results show that WCNN-KAN is superior to the existing advanced methods

rotating machinery,fault diagnosis,Kolmogorov-Arnold Networks,attention mechanism,wavelet convolution