Experimental Research on Vibration Feature Recognition for Bearing Reliability Based on Improved Convolutional Neural Networks

In the past network model application, convolutional neural networks (CNN) have better performance in the recognition of vibration characteristics of electric spindle bearings, but due to their own structural limitations, the actual recognition is still misjudged, which affects the efficient operation of machine tools and product quality. To enhance the reliability of bearing fault identification, this paper selects motorized spindle bearings as the main research object. First, it analyzes the identification process and issues of traditional CNN regarding vibration characteristics. Subsequently, it innovatively proposes a Parallel Convolutional Neural Network-Attention Mechanism (ZOA-PCNN-AT) model improved by the Zebra Optimization Algorithm. This model uses the Zebra Optimization Algorithm to optimize the structure of the parallel convolutional neural network and combines an attention mechanism to enhance the ability to capture key features. Finally, comparative experiments verify the effectiveness of the proposed method.