基于数据融合与GRU的航空发动机剩余寿命预测  

A Remaining Useful Life Prediction of Aero-Engine Based on Data Fusion and GRU

作　　者：李路云 王海瑞[1] 朱贵富[2] LI Luyun;WANG Hairui;ZHU Guifu(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Information Construction Management Center,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]昆明理工大学信息化建设管理中心,昆明650500

出　　处：《空军工程大学学报》2022年第6期33-41,共9页Journal of Air Force Engineering University

基　　金：国家自然科学基金(61863016)。

摘　　要：针对深度学习构造复合健康指标可解释性差,预测结果难以量化发动机剩余寿命预测中的不确定性问题,提出一种基于数据融合与门控循环单元(GRU)的航空发动机剩余寿命预测方法。首先,将多源传感器数据加权融合构造一维复合健康指标;然后,利用Bootstrap方法对一维复合健康指标进行有放回抽样,获取n组发动机退化特征样本;最后,利用一维复合健康指标和n组发动机退化特征样本构建“n+1”个基于GRU的剩余寿命预测模型,实现对航空发动机剩余寿命的区间预测。为证明所提方法的可行性和优越性,采用涡扇发动机退化数据集(C-MAPSS)的数据进行实验,得到的均方根误差为15.8254,评分函数值为344.2105。结果表明,该方法不仅能获得较好的预测效果,还能有效解决深度学习在发动机剩余寿命预测中存在的缺陷。Aimed at the problems that interpretable ability of composite health indicators constructed by deep learning is poor,and the prediction results are difficult to quantify the uncertainty of engine remaining life,an aero-engine remaining life prediction method based on data fusion and gate recurrent unit(GRU)is proposed.Firstly,the one-dimensional composite health index is constructed by weighted fusion of multi-source sensor data.And then,the Bootstrap method is utilized for putting the samples back on the one-dimensional composite health index,obtaining the engine degradation characteristic samples of n-groups.Finally,“n+1”remaining life prediction models based on GRU are constructed by using one-dimensional composite health indicators and the n-groups of engine degradation characteristic samples,the prediction interval of engine remaining life is quantified.In order to prove the feasibility and superiority of the proposed method,the turbofan engine degradation dataset(C-MAPSS)is used for the experiment,and the root mean square error obtained is 15.8254,the score function value is 344.2105.The results show that this method can not only achieve the better prediction results,but also can effectively solve the defects of deep learning in engine remaining life prediction.

关 键 词：航空发动机 寿命预测 维纳过程 多源传感器数据融合 自助法 门控循环单元 

分 类 号：V263.5[航空宇航科学与技术—航空宇航制造工程]