摘要
针对部分轴承故障样本稀少、源域和目标域数据存在分布差异、缺乏良好的跨域特征表示,从而导致分类性能退化、故障辨识精度低等问题,提出了一种基于迁移成分分析的域自适应轴承智能故障诊断方法。首先,建立了一个新型特征表示;然后,通过一个用特征提取方法得到的参数核实现域自适应,将数据投射到已学习的迁移成分,让源域和目标域样本在特征子空间上的最大均值差异最小化,得到了一个降维的特征子空间,由此显著地缩小了域分布间的距离,实现了从源域到目标域的跨域特征信息迁移;最后,通过实验对所提出的故障诊断方法的有效性进行了验证。研究结果表明:所提方法的最高分类精度达到95%,平均测试准确度达到81%,比常用分类方法的准确率提升了70%左右;所提算法可以减少域分布差异和标签噪声的影响,正确、有效地对小样本数据进行分类,检测出滚动轴承的健康状态。
In order to solve the problem of poor classification performance and low identification accuracy of fault caused by few bearing fault samples,discrepancy in data distribution between source domain and target domain,and the lack of good cross domain feature representation,a domain adaptive intelligent fault diagnosis method for bearings based on transfer component analysis(TCA)was proposed.A new feature representation was established,a parametric kernel obtained by feature extraction method was used to perform domain adaption.The data were projected onto learned transfer components,and the maximum mean difference between the source domain and the target domain samples in the feature subspace was minimized.Then a reduced dimension feature subspace was obtained.Thus,the distance between domain distributions was significantly reduced,feature information transfer was constructed from source domain to target domain.Finally,the effectiveness of the proposed fault diagnosis method was verified by experiments.The results indicate that the highest classification accuracy of the proposed method can reach 95%,the average test accuracy can reach 81%,which is about 70%higher than the accuracy of the common classification methods.The proposed algorithm can reduce the influence of domain distribution discrepancy and label noise,classify small sample data correctly and effectively.The health status of rolling bearing can be detected by the proposed methods.
作者
兰雨涛
胡超凡
金京
王衍学
LAN Yu-tao;HU Chao-fan;JIN Jing;WANG Yan-xue(School of Mechanical-Electronic and Vehicle Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;School of Mechanical and Electrical Engineering,Guilin University of Electronic Technology,Guilin 541004,China)
出处
《机电工程》
CAS
北大核心
2021年第5期521-527,共7页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(51875032,51965013)
北京市百千万人才工程资助项目(2020A26)
北京建筑大学市属高校基本科研业务费专项资金资助项目(X20159)。
关键词
滚动轴承
迁移成分分析
跨域
特征分类
故障诊断
rolling bearings
transfer component analysis(TCA)
cross domain
feature classification
fault diagnosis
