跨工况轴承退化评估任务可迁移性分析  

Transferability Analysis for the Cross-operating-condition Bearing Degradation Estimation Tasks

作　　者：肖俊玲 杨正伟 王旭 冯辅周 XIAO Junling;YANG Zhengwei;WANG Xu;FENG Fuzhou(Rocket Force Engineering University,Xi’an 710025,China;Mechanical Engineering Department,Tsinghua University,Beijing 100084,China;Army Academy of Armored Forces,Beijing 100072,China)

机构地区：[1]火箭军工程大学,陕西西安710025 [2]清华大学机械工程系,北京100084 [3]陆军装甲兵学院,北京100072

出　　处：《装甲兵学报》2022年第4期92-96,103,共6页Journal of Armored Forces

摘　　要：由于实际中很难收集到完备工况下有标注的全寿命数据,基于迁移学习(Transfer Learning,TL)方法的跨工况轴承剩余使用寿命(Remaining Useful Life,RUL)估计近年来成为热点问题,但当迁移学习的假设无效时,即源和目标域之间没有相似的隐式结构时,必然会导致负迁移。鉴于此,提出了一种用于跨工况轴承退化估计的可迁移性分析方法,采用连续小波变换(Continuous Wavelet Transform,CWT)、Rényi熵以及动态时间规整(Dynamic Time Warping,DTW)作为跨域数据相似度测量方法,并在PRONOSTIA平台上验证了该方法的有效性。所提方法能够为不同工况间轴承数据可迁移性的定性分析提供理论指导,有效缓解模型负迁移问题,进而保证跨工况轴承退化知识迁移的可靠性。Due to the difficulty of collecting enough marked run-to-failure data under complete operating conditions in practice,the estimation of Remaining Useful Life(RUL)of bearings across operating conditions based on Transfer Learning(TL)method has become a hot issue in recent years.However,when the transfer assumptions are not valid,that is,no similar implicit structure between the source and target domains,it will inevitably lead to negative transfer.To amend this issue,a transferability analysis method for the cross-operating-condition bearing RUL estimation tasks is proposed.Continuous Wavelet Transform(CWT),Rényi entropy and Dynamic Time Warping(DTW)are used as cross-domain data similarity measurement methods,and the effectiveness of this method is verified on PRONOSTIA platform.The proposed method can provide theoretical guidance for the qualitative analysis of bearing data transferability between different working conditions,effectively alleviate the negative transfer problem of the model,and then ensure the reliability of bearing degradation knowledge transfer across operating conditions.

关 键 词：轴承 剩余使用寿命(RUL) 连续小波变换(CWT) Rényi熵 动态时间规整(DTW) 

分 类 号：TH133.3[机械工程—机械制造及自动化] TB114.3[理学—概率论与数理统计]