基于随机森林的飞行载荷代理模型分析方法  

Surrogate model for flight load analysis based on random forest

作　　者：李海泉 陈小前[1,3] 左林玄[2] 赵卓林 LI Haiquan;CHEN Xiaoqian;ZUO Linxuan;ZHAO Zhuolin(School of Aerospace Science and Engineering,National University of Defense Technology,Changsha 410073,China;Shenyang Aircraft Design Institute,Aviation Industry Corporation of China,Shenyang 110035,China;National Innovation Institute of Defense Technology,Chinese Academy of Military Science,Beijing 100071,China)

机构地区：[1]国防科技大学航天科学与工程学院,长沙410073 [2]中国航空工业集团有限公司沈阳飞机设计研究所,沈阳110035 [3]中国人民解放军军事科学院国防科技创新研究院,北京100071

出　　处：《航空学报》2022年第3期309-318,共10页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(61903349)。

摘　　要：飞行载荷分析(FLC)通常需消耗大量的计算资源和时间,提升飞行载荷计算效率对于缩短研发周期、提高设计质量具有重要意义。本文研究了数据驱动的机器学习代理模型飞行载荷分析方法,由于随机森林(RF)代理模型具有学习效率较高、泛化能力较强、可避免过拟合、参数可解释、变量敏感度分析等优点,使其十分切合飞行载荷分析,具有重要的应用潜力和前景。文中基于传统的飞行载荷分析方法,采用NASTRAN的SQL144载荷分析框架获得训练随机森林代理模型的样本数据,然后以高度、马赫数、过载、俯仰角加速度等作为输入参数构建了飞机对称机动载荷预测代理模型。采用建立的模型预测了算例飞机其他工况机翼翼根、平尾翼根的剪力、弯矩、扭矩,通过对预测结果评估校验,证实了模型具有较高的精度,可以大幅提升飞行载荷分析效率,并能够分析飞行载荷对各状态参数的敏感度,为高效全面地分析飞行载荷提供了新的思路。Flight Load Computation(FLC)consumes considerable computational time and resources,and promoting FLC efficiency is highly significant to shorten the development cycle and improve the design quality of aircraft.This paper focuses on a data-driven machine learning surrogate model for load analysis based on the Random Forest(RF),which fits the load analysis with important potential and application prospect in this field because of its advantages such as high learning efficiency,strong generalization ability,overfitting avoidance,parameter interpretability,and variable sensitivity analysis.The training data for the RF surrogate model using conventional load analysis methods and the SQL144 framework of NASTRAN are generated,and parameters such as height,Mach number,overload,and pitch angular acceleration serve as the input variables to establish the surrogate model of load prediction in the case of symmetric maneuver.The proposed model is used to predict other conditions and the shear force,bending moment,and torque of roots of wings and horizontal tail.The predicting results are evaluated and verified,demonstrating high accuracy of the RF surrogate model and its ability to drastically promote the FLC efficiency and conduct sensitivity analysis of the flight load to state parameters.This study provides a new approach to the efficient and comprehensive flight load analysis.

关 键 词：飞行载荷 代理模型 机器学习 随机森林 对称机动 

分 类 号：V215.1[航空宇航科学与技术—航空宇航推进理论与工程]