摘要
针对不同工况下轴承监测数据分布差异性大且带有标签数据稀缺的问题,提出一种基于多尺度残差子域适应的轴承故障诊断方法。首先,构建多尺度残差网络,提取源域与目标域的可迁移特征;其次,通过局部中心距差异对齐源域和目标域中同一类别的相关子域分布;最后,将局部中心距差异和分类器损失作为目标优化函数,完成模型训练,实现目标域数据的状态识别,并通过实验表明,该方法在目标域数据无标签的情况下准确率可以达到98%以上,满足工程应用的实际需求。
Aiming at the problem that bearing monitoring data distribution was different in different working conditions and label data was scarce,a bearing fault diagnosis method based on multi-scale residual sub-domain adaptation was proposed.Firstly,a multi-scale residual network was constructed to extract the transferable features of source domain and target domain.Secondly,the distribution of related subdomains of the same category in the source domain and the target domain was aligned by the local central moment discrepancy.Finally,the local central moment discrepancy and classifier loss were used as the objective optimization function to complete the model training and realize the status identification of the target domain data.Experiments showed that the accuracy of this method could reach more than 98%when the target domain data was unlabeled,which could meet the actual needs of engineering applications.
作者
刘晶
宁森
徐伟杰
盛译瑶
季海鹏
LIU Jing;NING Sen;XU Weijie;SHENG Yiyao;JI Haipeng(School of Artificial Intelligence,Hebei University of Technology,Tianjin 300400,China;Hebei Data Driven Industrial Intelligent Engineering Research Center,Tianjin 300400,China;Tianjin Development Zone Jingnuo Data Technology Co.,Ltd,Tianjin 300400,China;CITIC Dicastal Co.,Ltd,Qinhuangdao 066000,China;University of Queensland,Queensland 4072,Australia;College of Materials Science and Engineering,Hebei University of Technology,Tianjin 300400,China)
出处
《郑州大学学报（理学版）》
CAS
北大核心
2023年第5期39-46,共8页
Journal of Zhengzhou University:Natural Science Edition
基金
京津冀基础研究合作专项项目(E2021203250)
天津市人工智能重大专项(19ZXZNGX00040)。
关键词
轴承故障诊断
子域适应
多尺度残差网络
局部中心距差异
状态识别
bearing fault diagnosis
sub-domain adaptation
multi-scale residual network
local central moment discrepancy
status identification
