摘要
针对迁移学习中源域,目标域数据分布差异大,传统学习中边缘分布与条件分布贡献动态变化难以适应的问题,提出了一种基于深度动态域适应的轴承故障诊断方法。在域适应层引入动态分布自适应方法,通过域分类器分别进行边缘分布对齐和条件分布对齐,在根据平衡因子动态衡量条件分布和边缘分布对域的贡献度,进行动态域适应。通过对凯斯西储大学和江南大学轴承数据集变工况下的迁移诊断试验及对比分析,有效地提高了跨域诊断的精度,验证了所提方法的有效性与卓越性。
Aiming at the problem that the data distributions in source domain and target domain in transfer learning are very different,and it is difficult to adapt to the dynamic changes of edge distribution and conditional distribution in traditional learning,a bearing fault diagnosis method based on deep dynamic domain adaptation was proposed.In the domain adaptation layer,a dynamic distribution adaptation method was introduced,and edge distribution alignment and conditional distribution alignment were performed by domain classifiers,and dynamic domain adaptation was performed by dynamically measuring the contribution of conditional distribution and edge distribution to the domain according to a balance factor.Through the migration diagnosis test and comparative analysis of the bearing data sets of Case Western Reserve University and Jiangnan University under variable working conditions,the accuracy of cross-domain diagnosis was effectively improved,and the effectiveness and excellence of the proposed method were verified.
作者
王军辉
雷文平
刘华杰
魏李军
韩东洋
WANG Junhui;LEI Wenping;LIU Huajie;WEI Lijun;HAN Dongyang(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)
出处
《振动与冲击》
EI
CSCD
北大核心
2023年第14期245-250,共6页
Journal of Vibration and Shock
基金
河南省高等学校精密仪器制造技术与工程重点学科开放实验室开放基金(PMTE201301A)。
关键词
迁移学习
故障诊断
动态域适应
贡献度
transfer learning
fault diagnosis
dynamic domain adaptation
contribution
