航空涡扇引擎剩余使用寿命预测算法研究  

Remaining useful life prediction algorithm for aviation turbofan engine

作　　者：吴锐 马洁[1] 丁恺林 Wu Rui;Ma Jie;Ding Kailin(School of Automation,Beijing Information Science and Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学自动化学院

出　　处：《南京理工大学学报》2019年第6期708-714,共7页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61973041)

摘　　要：为改进故障预测技术,提出了全特征输入长短时记忆单元(AF-LSTM)算法。根据底层原始传感器数据自动学习更高级的抽象表示,并利用这些表示导出传感器数据,估计剩余使用寿命(RUL)。该文方法不依赖任何退化趋势假设,对噪声健壮,并能处理缺失值。在公开航空涡扇汽轮机引擎仿真数据集上的实验结果表明:该文方法的均方误差(MSE)指标明显优于多层感知器(MLP)模型,其时间分数(Score)指标优于带指数假设的线性回归(LR-EXP)模型。To improve failure prediction technology,a all features input long-short term memory(AF-LSTM)algorithm is proposed. This algorithm learns higher-level abstract representations from the underlying raw sensor data automatically,and uses these representations to derive sensor data and evaluate remaining useful life(RUL). This algorithm is independent of degradation trend assumptions and robust to noise,and can handle missing values. The experimental results of the open aviation turbofan engine simulation data set show that the mean square error(MSE)index of AF-LSTM is better than that of the multi-layer perception(MLP),and the score index of AF-LSTM is better than that of the linear regression-exponential(LR-EXP).

关 键 词：航空 涡扇引擎 剩余使用寿命 预测 全特征输入 长短时记忆单元 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]