基于FDR数据的发动机基线建模及优化  

Engine Baseline Modeling and Optimization Based on FDR Data

作　　者：刘佳欢 白杰[1] 刘帅[1] 王文进 LIU Jia-huan;BAI Jie;LIU Shuai;WANG Wen-Jin(Key Laboratory for Civil Airworthiness Certification Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学民航航空器适航审定技术重点实验室,天津300300

出　　处：《计算机仿真》2025年第2期40-45,77,共7页Computer Simulation

基　　金：科技部国家重点研发计划项目-航班运行安全预警与辅助决策技术(2022YFC3002502)。

摘　　要：研究了如何利用FDR数据建立高精度的发动机基线模型,提出了一种结合移动均值滤波和一阶滞后滤波算法的稳态点筛选方法和发动机稳态判断标准,并利用MLP神经网络建立了高精度的发动机基线模型。首先,对FDR数据进行了分析,提取发动机相关参数。其次,根据发动机稳态状态的判断标准,对比了不同的噪声滤波方法及热稳态滤波对稳态点筛选数量的影响。最后,利用300次飞行的FDR数据进行实验,搭建MLP神经网络并进行训练和验证。实验结果表明,移动均值滤波结合热稳态滤波相对传统方法能够提取更多高质量的稳态点,使用MLP神经网络建立的基线模型预测精度优于传统线性回归模型。In this paper,the flight data recorder(FDR)is used to establish a high-precision engine baseline model.A steady-state point screening method and engine steady-state judgment standard combining moving average filtering and first-order hysteresis filtering algorithms are proposed,and a high-precision engine baseline model is established by using a multilayer perceptron neural network.First,the FDR data was analyzed to extract the relevant parameters of the engine.Secondly,according to the judgment criteria of the steady state of the engine,the influence of different noise filtering methods and thermal steady-state filtering on the number of steady-state point screenings was compared.Finally,using the FDR data of 300 flights,the MLP neural network was constructed,trained and verified.The experimental results show that the moving average filtering combined with thermal steady-state filtering can extract more high-quality steady-state points than the traditional method,and the prediction accuracy of the baseline model established by the MLP neural network is better than that of the traditional numerical regression model.

关 键 词：数据 航空发动机 稳态点 基线模型 

分 类 号：V231.3[航空宇航科学与技术—航空宇航推进理论与工程]