基于类加权对抗网络的跨域旋转机械故障诊断  

Fault Diagnosis of Cross Domain Rotating Machinery Based on Class Weighted Adversarial Networks

作　　者：彭博 张毅[2] 蹇清平 于翔 PENG Bo;ZHANG Yi;JIAN Qing-ping;YU Xiang(Chengdu Vocational&Technical College of Industry,Equipment Manufacturing College,Sichuan Chengdu 610218,China;School of Mechatronic Engineering,Southwest Petroleum University,Sichuan Chengdu 610500,China)

机构地区：[1]成都工业职业技术学院,四川成都610218 [2]西南石油大学机电工程学院,四川成都610500

出　　处：《机械设计与制造》2022年第8期74-79,87,共7页Machinery Design & Manufacture

基　　金：四川省教育厅科研项目(18ZB0046)。

摘　　要：为解决边缘数据离群性问题,提出了一种基于类加权对抗网络的跨域旋转机械故障诊断方法。通过在源类别上附加类级权重,可以直观地表示源域和目标域之间的关系,有利于共享类别的条件对齐。进一步提出用于局部域自适应的类加权对抗网络,同时忽略源异常值,有效激励了正知识的转移,提升域自适应的效果。在CWRU数据集和一个列车转向架数据集上对该方法进行了实验,结果表明提出的方法可以有效地解决边缘数据离群性问题,提升知识迁移的效果从而提高故障诊断精度。In order to solve the problem of edge data outlier,a cross domain fault diagnosis method of rotating machinery based on class weighted adversarial network was put forward.By adding class level weights to the source category,the relationship between the source domain and the target domain could be intuitively expressed,which was conducive to the conditional alignment of shared categories.Furthermore,a class weighted adversarial network for local domain adaptation is proposed,which ignores the source outliers,effectively stimulates the transfer of positive knowledge and improves the effect of domain adaptation.Experiments were carried out on CWRU data set and a train bogie data set.The results show that the proposed method can effectively solve the problem of edge data outlier,improve the effect of knowledge transfer and improve the accuracy of fault diagnosis.

关 键 词：旋转机械 故障诊断 对抗网络 域自适应 

分 类 号：TH16[机械工程—机械制造及自动化] TH133.33[自动化与计算机技术—检测技术与自动化装置] TP277[自动化与计算机技术—控制科学与工程]