关键词: 滚动轴承, 一维卷积神经网络, 注意力机制, 故障诊断

Abstract: Addressing the noise interference and complex operating conditions present in rolling bearings in practical applications,this paper proposes a fault diagnosis method based on attention mechanisms and one-dimensional convolutional neural network ( 1D-CNN ) . By adding noise to the rolling bearing data provided by Western Reserve University, we simulated the noise interference encountered in real-world applications. The study introduced four mainstream attention mechanism modules into the 1D-CNN model, including high-efficiency channel attention, convolutional block attention module,gated attention mechanism, and compared them with the traditional convolutional neural network ( CNN) model. Experimental results show that the 1D-CNN combined with attention mechanisms performs significantly better in fault diagnosis than the original CNN model. Specifically, the diagnostic accuracy of the original 1D-CNN is 96. 56%; the accuracy of the 1D-CNN with high-efficiency channel attention is 98. 44%;the accuracy of the 1D-CNN with convolutional block attention module is 99. 22%;the accuracy of the 1D-CNN with squeezed excitation network attention is 98. 75%;and the 1D-CNN with gated attention mechanism performs the best,achieving a diagnostic accuracy of 100%. The experimental results verify the effectiveness of attention mechanisms in enhancing the fault diagnosis performance of 1D-CNN models,indicating that this method has great potential for application in rolling bearing fault diagnosis.

Key words: rolling bearing, one-dimensional convolutional neural network, attention mechanism, fault diagnosis