基于注意力残差深度网络的滚动轴承故障诊断方法研究

Research on method of rolling-bearing fault diagnosis based on attention residual depth network

针对现有滚动轴承故障智能诊断方法大多通过增加网络模型的复杂度以提高特征提取能力,从而导致模型参数量剧增、计算效率低等问题,文中提出了一种基于注意力残差网络的滚动轴承故障诊断方法。采用复合降噪方法减小噪声干扰与特征提取难度,将压缩与激励网络注意力机制引入残差网络模型,从而进一步提高诊断准确率与计算效率。最后,分别采用滚动轴承公共数据集及滚动轴承加速寿命试验对所提方法进行了验证。

Currently,most methods of intelligent rolling-bearing fault diagnosis enhance the complexity of network models in order to improve the capability of feature extraction,which leads to substantial increase in model parameters and decreased com-putational efficiency.In this article,efforts are made to present a method of rolling-bearing fault diagnosis based on the attention residual network.The composite noise-reduction techniques are introduced,so as to minimize noise interference and facilitate fea-ture extraction;the compression and excitation network attention mechanisms are integrated into the residual network models,which further enhances both diagnostic accuracy and computational efficiency.Finally,through the rolling bearing's common datasets and the accelerated life test,it is verified that this method is practical and effective.