Interpretable Wavelet Kernel Network with Attention Mechanism

Planetary gearboxes have a wide range of applications in a variety of equipment, including helicopter reducers, wind turbines, and heavy trucks. Planetary gearbox is a key part that is prone to failure. With the development of deep learning, intelligent fault diagnosis technology has attracted more and more attention and has been successfully applied to gearbox fault diagnosis. Effective fault diagnosis can prevent the failure of planetary gearboxes from causing huge property losses. But the commonly used intelligent diagnostic methods face the problems of interpretability and transparency. This not only challenges the credibility of the decision itself, but also further limits the credibility of the decision. This paper proposes a wavelet kernel network weighted by convolutional block attention module (WKN-CBAM). This method not only retains the interpretability of the wavelet kernel network, but also enhances the performance of the network. The attention mechanism is added to improve the training efficiency and convergence speed, and the attention weight of network space is added to explain the attention distribution of the network in a certain frequency band. The model is tested on the planetary gearbox data set, and good results are obtained. The performance of the network is verified by six migration tasks of the planetary gearbox data set.