In the field of bearing fault diagnosis,many convolutional models with excellent performance face challenges in industrial applications due to deployment cost constraints.This paper aims to develop a lightweight diagnostic method with reduced parameters.We investigate the feasibility of using depthwise separable convolution to construct a lightweight bearing fault diagnosis model,thereby proposing a strategy to compress the parameters of the convolutional network while ensuring diagnostic accuracy.The effectiveness of the proposed method is validated on both publicly available and custom vibration signal datasets.The results demonstrate that compressing convolutional models using depthwise separable convolution allows for lightweight requirements while maintaining a high diagnostic accuracy(96.20±2.81%).
维普
万方数据