融合TA-TCN和迁移学习的滚动轴承寿命预测  

Rolling Bearing Life Prediction Combining TA-TCN and Transfer Learning

作　　者：车鲁阳 冷子文 付惠琛 张佳佳 高军伟[1,2] CHE Luyang;LENG Ziwen;FU Huichen;ZHANG Jiajia;GAO Junwei(School of Automation,Qingdao University,Qingdao 266071,China;Shandong Provincial Key Laboratory of Industrial Control Technology,Qingdao University,Qingdao 266071,China;Shandong Provincial Special Equipment Inspection and Research Institute,Rizhao 276826,China)

机构地区：[1]青岛大学自动化学院,青岛266071 [2]青岛大学山东省工业控制技术重点实验室,青岛266071 [3]山东省特种设备检验科学研究院,日照276826

出　　处：《组合机床与自动化加工技术》2024年第3期147-151,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：山东省自然科学基金资助项目(ZR2019MF063)。

摘　　要：针对在实际工业生产中,滚动轴承由于数据量少导致剩余寿命预测的准确度不高的问题,提出了一种时序注意力(temporal attention, TA)优化的时间卷积神经网络(time convolutional networks, TCN)与迁移学习相结合的剩余寿命预测方法。首先,通过互补集合经验模态分解(complementary ensemble empirical mode decomposition, CEEMD)将原始特征向量分解为一组子序列分量,突出特征信号、降低噪声干扰;然后,将子序列分量输入搭建好的TCN模型并添加TA进行优化,深度挖掘深度特征与退化曲线关系;最后,引入迁移学习,利用源域数据进行训练和少量目标域数据进行参数微调,得到目标网络模型。经实例验证,所提模型的稳定性、预测精度相对于其它对比模型有所提升,且在异工况条件下依然有着良好的预测能力。Aiming at the problem that the accuracy of remaining life prediction of rolling bearings is not high due to the small amount of data in actual industrial production,a time convolutional neural network(TCN)optimized by temporal attention(TA)and transfer learning is proposed.Firstly,the original eigenvector is decomposed into a set of subsequence components by empirical mode decomposition of complementary sets(CEEMD)to highlight the eigensignal and reduce noise interference.Then,the subsequence components are input into the built TCN model and TA is added for optimization,and the relationship between the depth features and the degradation curve is deeply explored.Finally,transfer learning is introduced to obtain the target network model by using the source domain data for training and a small amount of target domain data for parameter fine-tuning.After example verification,the stability and prediction accuracy of the proposed model are improved compared with other comparison models,and it still has good prediction ability under different working conditions.

关 键 词：滚动轴承 寿命预测 互补集合经验模态分解 时序注意力 时间卷积神经网络 迁移学习 

分 类 号：TH133.3[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]