摘要
航空发动机技术是衡量一个国家科技水平和工业实力的重要标志,健康状态监测和剩余使用寿命(remaining useful life,RUL)预测技术是航空发动机安全服役、经济运行的重要保障.针对航空发动机RUL预测精度较低、不确定性难以量化的问题,本文提出了一种数据驱动的航空发动机RUL区间预测方法.首先,在ConvJANET框架下构建新的卷积-卷积循环-全连接结构的深度学习模型,逐层提取航空发动机监测数据中的退化特征;其次,利用极大似然思想指导神经网络模型的优化求解,并基于损失函数形式变化的策略训练模型,实现对航空发动机RUL的高精度预测与不确定性量化.将所提出的方法用于分析航空发动机退化数据集,结果表明,对比传统基于蒙特卡洛的方法,本文提出的方法具有更高的RUL预测准确率和更好的置信区间预测性能.
The technology of aircraft engines is a crucial symbol for measuring the national level of science and technology and industrial strength.For the safe operation of aircraft engines,the technology of health condition monitoring and RUL prediction is a major assurance.A data-driven RUL interval prediction method for an engine is proposed to address the issues of low precision and complex quantification of uncertainty.First,a deep learning model on ConvJANET is built using a convolution-recurrent-fullconnection structure,and the engine's degradation features engine are then extracted layer by layer.Second,the precision of the neural network model optimization and the strategy training model based on the changing loss function are guided by the maximum likelihood theory to realize the high-precision prediction and uncertainty quantification of aviation engine RUL.The proposed method is applied to the analysis of the aircraft engine degradation dataset,and the results demonstrate that the suggested method has higher RUL prediction accuracy and better confidence interval prediction performance than the conventional Monte Carlo-based method.
作者
苗永浩
李晨辉
石惠芳
林京
MIAO YongHao;LI ChenHui;SHI HuiFang;LIN Jing(School of Reliability and Systems Engineering,Beihang University,Beijing 100191,China)
出处
《中国科学：技术科学》
EI
CSCD
北大核心
2023年第7期1189-1201,共13页
Scientia Sinica(Technologica)
基金
国家重点研发计划(批准号:2021YFB2500604)资助项目。
关键词
航空发动机
剩余寿命预测
卷积循环神经网络
极大似然估计
不确定性量化
raft engine
remaining useful life prediction
convolutional recurrent neural network
maximum likelihood estimation
uncertainty quantification
