基于Transformer-GRU并行网络的滚动轴承剩余寿命预测  

Remaining Useful Life Prediction of Rolling Bearings Based on Transformer-GRU Parallel Network

作　　者：唐贵基[1,2] 刘叔杭 陈锦鹏 徐振丽 田寅初 徐鑫怡 TANG Guiji;LIU Shuhang;CHEN Jinpeng;XU Zhenli;TIAN Yinchu;XU Xinyi(Department of Mechanical Engineering,North China Electric Power University,Baoding Hebei 071003,China;Hebei Provincial Key Laboratory of Health Maintenance and Failure Prevention of Electrical Machinery Equipment,North China Electric Power University,Baoding Hebei 071003,China)

机构地区：[1]华北电力大学机械工程系,河北保定071003 [2]华北电力大学,河北省电力机械装备健康维护与失效预防重点实验室,河北保定071003

出　　处：《机床与液压》2024年第19期188-195,共8页Machine Tool & Hydraulics

基　　金：国家自然科学基金面上项目(52177042);河北省自然科学基金项目(E2020502031)。

摘　　要：为有效描述滚动轴承性能退化趋势和准确预测其剩余寿命,提出一种基于多域特征融合的Transformer-GRU并行网络的滚动轴承剩余寿命预测方法。建立评价指标对滚动轴承振动信号的时域、频域和时频域等多域特征进行筛选,得到评分高的敏感特征,获得退化特征集。利用自编码对退化特征集进行降维,减少数据复杂度和冗余度,得到滚动轴承的退化曲线。最后,利用Transformer-GRU并行网络进行剩余寿命预测,并将该方法运用到公开的轴承数据集分析中。结果表明:Transformer-GRU并行网络不仅可以高效准确地捕捉输入序列中的长期依赖关系,还能更好地处理时间序列之间的特征;该方法可以有效地预测滚动轴承剩余寿命,相比LSTM、GRU等经典方法更具优越性和泛化性。In order to effectively describe the performance degradation trend of rolling bearing and accurately predict its remaining useful life,a method for predicting the remaining useful life of rolling bearings based on multi-domain characteristics fusion in Transformer-GRU parallel network was proposed.An evaluation index was established to screen the sensitive features for the time domain,frequency domain,and time-frequency domain of the rolling bearing vibration signal,and the sensitive features with high scores were obtained and the degraded feature set was obtained.The degradation feature set dimension was reduced by using self-coding to reduce data complexity and redundancy,and the degradation curve of rolling bearing was obtained.Finally,the remaining useful life prediction was carried out using Transformer-GRU parallel network,and the method was applied to the analysis of the public bearing dataset.The results show that the Transformer-GRU parallel network can not only capture long-term dependencies in input sequences efficiently and accurately,but also process features between time series better.Compared with LSTM,GRU and other classical methods,the proposed method can effectively predict the remaining useful life of rolling bearings.

关 键 词：滚动轴承 剩余寿命预测 多域特征融合 TRANSFORMER GRU 

分 类 号：TH133.33[机械工程—机械制造及自动化]