电子测量与仪器学报

刘应东,刘韬,李华,王廷轩.变工况轴承的联合分布适应迁移故障诊断[J].电子测量与仪器学报,2021,35(5):69-75

变工况轴承的联合分布适应迁移故障诊断

Transfer fault diagnosis of bearings under variable workingconditions based on joint distribution adaptation

DOI：

英文关键词:joint distribution adaptation variable working conditions transfer learning fault diagnosis maximum mean discrepancy

基金项目:国家自然科学基金（52065030,51875272）、国家重点研发计划（2018YFB1306103）项目资助

作者	单位
刘应东	1.昆明理工大学机电工程学院
刘韬	1.昆明理工大学机电工程学院
李华	1.昆明理工大学机电工程学院
王廷轩	1.昆明理工大学机电工程学院

Author	Institution
Liu Yingdong	1.Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology
Liu Tao	1.Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology
Li Hua	1.Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology
Wang Tingxuan	1.Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology

摘要点击次数: 445

全文下载次数: 4

中文摘要:

针对传统的机器学习算法在变工况条件下的轴承故障分类中诊断率低的问题,提出了基于联合分布适应( JDA)算法与 K-最近邻(KNN)分类算法相结合的轴承故障诊断方法。首先该方法通过提取不同工况下的轴承故障信号的时域特征分别作为源域样本和目标域样本,并通过 Fisher 线性判别分析(FLDA)方法计算各个特征所占权重。然后将权重较大的特征组成的特征向量通过 JDA 方法进行联合分布适配,即通过核函数将源域样本和目标域样本映射到低维潜在空间,以最大均值差异 (MMD)距离为度量标准,同时减小源域和目标域样本的边缘分布和条件分布差异。最后将适配完的源域和目标域样本分别作为训练集和测试集,通过 KNN 分类器进行模式识别,最终实现在变工况条件下的轴承故障诊断分类。通过仿真分析和实验验证,所用方法相较于主成分分析(PCA)、核主成分分析(KPCA)传统机器学习方法以及 TCA 迁移学习方法,显著提高了变工况条件下的轴承故障诊断精度。

英文摘要:

For the low diagnostic rate of traditional machine learning algorithm in bearing fault classification under variable working conditions, this paper proposed a bearing fault diagnosis method based on the combination of joint distribution adaptation ( JDA) algorithm and K-nearest neighbor ( KNN) classification algorithm. Firstly, the time domain features of bearing fault signals under different working conditions are extracted as source domain samples and target domain samples respectively, then calculating the weight of each feature by FLDA method. The feature vectors composed of features with higher weights to adapting joint distribution by JDA method, that is, the source domain samples and target domain samples are mapped to the low-dimensional potential space by kernel function, and the maximum mean discrepancy (MMD) distance is taken as the measurement standard to reduce the marginal distribution and conditional distribution differences between the source domain samples and the target domain samples. Finally, the mapped source domain and target domain samples are used as training data and test data respectively, and the model identification is implemented by KNN classifier, and the bearing fault diagnosis classification under variable conditions is achieved. Compared with the method of PCA, KPCA and TCA, through simulation analysis and experimental verification, the method proposed in this paper significantly improves the accuracy of bearing fault diagnosis under variable working conditions.

查看全文查看/发表评论下载PDF阅读器