基于交叉注意力的无监督域适应轴承故障诊断  

Unsupervised domain adaptive bearing fault diagnosis based on cross-attention

作　　者：汪振鹏 朱晓娟[1] WANG Zhenpeng;ZHU Xiaojuan(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan Anhui 232001)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《宁夏师范学院学报》2024年第7期58-71,共14页Journal of Ningxia Normal University

基　　金：安徽高校自然科学研究重点项目(KJ2020A0300)。

摘　　要：针对轴承在变工况场景下,振动数据存在分布差异,故障标记样本不足的跨域故障诊断问题,提出基于交叉注意力的无监督域适应轴承故障诊断方法.首先,使用交叉注意力模块对齐跨域间局部相似样本,并提取源域和目标域的故障数据特征;其次,使用基于Wasserstein距离的域对齐策略,以对齐全局边缘特征分布;最后,对无标签的目标工况样本,提出伪标签生成方案,进一步对齐细粒度的类别信息.实验结果表明,所提方法具有较高的诊断精度,在变工况场景下更具优势.In response to the issue of cross-domain fault diagnosis where vibration data distribution varies and fault labeled samples are insufficient under variable working conditions,an unsupervised domain adaptive bearing fault diagnosis method based on cross-attention was proposed.Firstly,a cross-domain module is used to align locally similar samples across domains and extract fault data features from both the source and target domains.Secondly,a domain alignment strategy based on the Wasserstein distance is employed to align the global marginal feature distribution.Finally,for the unlabeled target condition samples,a pseudo-label generation scheme is proposed to further align fine-grained class information.Experimental results demonstrate that the proposed method has high diagnostic accuracy and is more advantageous under variable working conditions.

关 键 词：轴承 故障诊断 迁移学习 域适应 交叉注意力机制 

分 类 号：TH133.3[机械工程—机械制造及自动化]