内对抗指导无监督多域适配网络的轴承协同故障诊断  

Internal adversarial guided unsupervised multi-domain adaptation network for collaborative fault diagnosis of bearing

作　　者：邵海东 陈星恺 曹鸿儒 姜洪开[2] SHAO HaiDong;CHEN XingKai;CAO HongRu;JIANG HongKai(College of Mechanical and Vehicle Engineering,Hunan University,Changsha 410082,China;School of Civil Aviation,Northwestern Polytechnical University,Xi'an 710072,China)

机构地区：[1]湖南大学机械与运载工程学院,长沙410082 [2]西北工业大学民航学院,西安710072

出　　处：《中国科学：技术科学》2023年第7期1229-1240,共12页Scientia Sinica(Technologica)

基　　金：国家重点研发计划项目(编号:2020YFB1712100);国家自然科学基金面上项目(批准号:52275104);湖南省自然科学基金优秀青年项目(编号:2021JJ20017)资助。

摘　　要：现有的轴承无监督跨域智能故障诊断研究多数是基于单源域自适应开展的,这将导致实际场景中多个拥有充足且多样化诊断信息的源域未能得到协同利用.如何更好地从多个源域中提取到故障轴承的共性特征并融合多源域知识协同诊断是主要挑战.针对上述问题,本文提出了一种内对抗指导的无监督多域适配网络;构造了一个内对抗模块以计算多源域对抗损失,并联合多组源域-目标域自适应子网指导提取多源域和目标域间的共性特征,增强知识覆盖能力;设计了一个多子网协同决策模块,利用多组源域-目标域的对抗损失和分布差异损失计算置信分数,以辅助多子网分类器做出更佳的融合决策,提升协同故障诊断的准确率.基于恒转速和变转速工况的轴承故障数据构造了多个无监督多源域迁移诊断任务,对比实验结果表明所提方法的优越性和鲁棒性.Most of the current research on unsupervised cross-domain intelligent fault diagnosis of bearings is based on single-source domain adaptiveness,which fails to simultaneously use multiple source domains with adequate and diverse diagnostic data in practical application scenarios.The main difficulty in diagnosing bearing faults is how to more effectively extract shared characteristics of defective bearings from various source domains and combine multi-source domain knowledge for collaborative diagnosis.A proposed intra-adversarial guided unsupervised multi-domain adaptation network(IAG-MDAN)aims to address these issues.In particular,an intra-adversarial module is first constructed to determine the multi-source domain adversarial loss,and multi-sets of sources and target domain adaptive subnetworks are combined to guide the extraction of common features between multi-source and target domains,which improve the knowledge coverage.In addition,a multi-subnet collaborative decision module is designed to calculate confidence scores using the adversarial loss and distribution difference loss of multiple source-target domains,which assists the multi-subnet classifier in making better fusion decisions and improving the accuracy of collaborative fault diagnosis.Several unsupervised multi-source domain migration diagnosis tasks are performed using faulty bearing datasets under both constant and variable speed conditions,and the comparative experimental results show the superiority and robustness of the proposed method.

关 键 词：轴承故障诊断 迁移学习 无监督多域自适应 内对抗损失 多子网协同决策 

分 类 号：TH133.3[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]