深层相关性对齐迁移学习的轴承故障诊断方法  

Bearing fault diagnosis method based on deep correlation alignment transfer learning

作　　者：窦唯[1,2] 王宝强 张宏利 Dou Wei;Wang Baoqiang;Zhang Hongli(Beijing Aerospace Propulsion Institute,Beijing,100076,China;Cryogenic Liquid Propulsion Technology Laboratory of China Aerospace Science and Technology Corporation,Beijing,100076,China;School of Mechatronic Engineering and Automation,Shanghai University,Shanghai,200444,China)

机构地区：[1]北京航天动力研究所,北京100076 [2]中国航天科技集团有限公司低温液体推进技术实验室,北京100076 [3]上海大学机电工程与自动化学院,上海200444

出　　处：《机械设计与制造工程》2023年第5期91-94,共4页Machine Design and Manufacturing Engineering

基　　金：国防技术基础科研项目(JSZL2019203A003)。

摘　　要：针对轴承实际工况复杂多变,现有的智能故障诊断模型和方法诊断效果不理想等问题,设计一种深层特征相关性对齐迁移学习的故障诊断方法。首先,对原始轴承振动信号预处理并将所获得的数据样本划分为训练集、迁移集和测试集;其次,建立一维卷积神经网络,采用训练集对网络模型进行初始化训练;再次,利用迁移集对微调模型进行迁移,提取源域和目标域的深层特征,在迭代过程中不断提高两域所提取特征之间的相关性;最后,使用测试集对所得到的轴承智能故障诊断模型的有效性进行验证。实验结果表明,相对于传统故障诊断方法,该方法可以提高模型的泛化能力,更好地完成实际工况下的轴承故障诊断任务。In view of the complex and changeable actual working conditions of the bearing and the poor diagnosis effect of the existing intelligent fault diagnosis models and methods,a fault diagnosis method based on deep feature correlation alignment transfer learning is designed.Firstly,the data samples obtained from the original bearing vibration signal preprocessing are divided into training set,migration set and test set.Secondly,a one-dimensional convolution neural network is established,and the training set is used to initialize the network model.Then the migration set is used to migrate the fine-tuning model to obtain the deep features of the source domain and the target domain,and constantly improved the correlation between the features extracted from the two domains in the iterative process.Finally,the test set is applied to verify the effectiveness of the bearing intelligent fault diagnosis model.Compared with the traditional fault diagnosis methods,the experimental results show that this method can improve the generalization ability of the model and better meet the task of bearing fault diagnosis under actual working conditions.

关 键 词：迁移学习 故障诊断 相关性对齐 小样本 深层特征 轴承 

分 类 号：TG156[金属学及工艺—热处理]