基于深度神经网络的超声速民机机翼结构设计  

Wing structural design of supersonic civil aircraft based on deep neural network

作　　者：牛芳淦 马文圆 杨超[1] 王宇[1] 尹海莲[1] NIU Fanggan;MA Wenyuan;YANG Chao;WANG Yu;YIN Hailian(Advanced Design Technology of Aircraft National Defense Key Subject Laboratory,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学飞行器设计先进技术国防重点学科实验室,南京210016

出　　处：《机械强度》2025年第4期122-130,共9页Journal of Mechanical Strength

基　　金：国家自然科学基金项目(12032011)。

摘　　要：目前对超声速民机机翼的研究主要侧重于低声爆设计技术和超声速减阻技术,针对机翼结构设计的研究相对较少。因此,提出了一种面向超声速民机初步设计阶段机翼结构设计的多级优化方法,包括机翼结构布局参数化建模、结构尺寸优化有限元模型的自动生成、深度神经网络代理模型的搭建与训练,以及基于深度神经网络代理模型进行优化求解。分析结果表明,提出的优化策略能够对超声速民机机翼结构进行良好的快速设计,深度神经网络模型相比于传统代理模型具有更高的预测精度,提高了机翼结构初步设计的效率。At present,the research on supersonic civil aircraft wings mainly focuses on the low sonic boom design and supersonic drag reduction technologies.There are relatively few studies on the wing structural design.Therefore,a multilevel optimization method for the wing structural design in the preliminary design stage of supersonic civil aircrafts was proposed.It included the parametric modeling of the wing structural layout,the automatic generation of the finite element model for the structural size optimization,construction and training of a surrogate model for the deep neural network.And the optimization was solved based on the deep neural network.The analysis results show that the proposed optimization strategy could quickly design the wing structure of the supersonic civil aircraft.The deep neural network model has higher prediction accuracy than the traditional surrogate model.Thus,the proposed approach can improve the efficiency of the preliminary design for wing structure.

关 键 词：超声速民机 参数化 深度神经网络 代理模型 结构设计 

分 类 号：V221[航空宇航科学与技术—飞行器设计]