基于域迁移的滚动轴承故障诊断研究  

Rolling Bearing Fault Diagnosis Based on Domain Transform

作　　者：曹梦婷 谷玉海[1] 王红军[1] 徐小力[1] CAO Mengting;GU Yuhai;WANG Hongjun;XU Xiaoli(Key Laboratory of Modern Measurement&Control Technology Ministry of Education,Beijing Information Science&Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学现代测控技术教育部重点实验室,北京100192

出　　处：《机械设计与制造》2025年第4期269-273,共5页Machinery Design & Manufacture

基　　金：国家自然科学基金(51975058);促进高校内涵发展—学科建设专项资助项目(5112011015)。

摘　　要：目前基于深度学习的滚动轴承故障诊断方法已经在机械设备领域得到了广泛的学习,而进行深度学习训练需要海量数据样本,针对深度学习方法在这一方面的不足,这里提出一种基于域迁移学习的滚动轴承故障诊断方法,能够在小样本数据量的前提下依旧对滚动轴承进行故障诊断并取得良好的诊断结果。首先,根据一维卷积神经网络和长短期记忆网络构造一个域迁移深度学习网络,将获得的源域数据与目标域数据作为输入,其次,经过网络训练之后,对提取出的故障特征分类。实验结果证明,在小样本数据量的前提下,采用的方法和基于无迁移的深度学习故障诊断方法相比,故障特征的分类精度更高,提高了故障诊断的正确率。At present,deep learning-based fault diagnosis methods for rolling bearings have been extensively learned in the field of machinery and equipment.Deep learning training requires massive data samples,in view of the shortcomings of deep learning methods in this respect,it proposes a domain transfer learning based the rolling bearing fault diagnosis method,this method can still diagnose the rolling bearing fault and obtain a good diagnosis result under the premise of a small sample amount of data.First,it choose to construct a domain migration deep learning network based on the one-dimensional convolutional neural net⁃work and the long short-term memory network,taking the obtained source domain data and target domain data as input.Sec⁃ondly,after network training,classify the extracted fault features.The experimental results prove that,under the premise of small sample data,the method adopted in this paper has higher classification accuracy of fault features than the deep learning fault di⁃agnosis method based on no migration,and improves the accuracy of fault diagnosis.

关 键 词：故障诊断 域迁移 一维卷积神经网络 长短期记忆网络 

分 类 号：TH16[机械工程—机械制造及自动化] TP206.3[自动化与计算机技术—检测技术与自动化装置] TH133.3[自动化与计算机技术—控制科学与工程]