| 摘要: |
| 随着运载火箭研发模式转变，快速迭代和协同优化设计成为主要发展方向，这就要求作为小回路论证中重要一环的气动特性计算能够实现在线输出数据，亟需研究一种快速计算气动特性的代理模型，代替耗时的CFD计算和风洞试验参与到总体优化设计中。综合比较多种快速计算途径，选择高斯基Kriging插值和BP神经网络两种方法构建代理模型。使用脚本控制的Cart3D软件生成数值试验样本，样本点精度与Fluent软件计算误差小于14%。通过样本点训练、内参优化和加点策略，最终获得相对误差小于10%的代理模型，能够实现给定外形参数在线秒级输出气动数据，极大地推动了气动计算在总体论证中的作用。 |
| 关键词: 运载火箭 总体优化 气动计算 代理模型 |
| DOI： |
|
| 基金项目:装备发展部领域基金（6140246030216HT19001） |
|
| Research on Aerodynamic Surrogate Modeling for Launch Vehicle Collaborative Optimization |
| SHEN Dan,PENG Bo,LI Zhouyang,GONG Yukun,LI Pingqi |
| (Beijing Institute of Astronautical Systems Engineering, Beijing 100076, China) |
| Abstract: |
| With the transformation of launch vehicle R&D mode, rapid iteration and collaborative optimization design have become a major trend for the research. This requires that aerodynamic characteristics calculation, which is an important part of the conceptual design, can supply real-time data online. Therefore, it is necessary to develop a surrogate model that can calculate aerodynamic characteristics rapidly, in place of time-consuming CFD methods and wind tunnel experiments for overall design optimization. This paper compared a variety of calculation methods comprehensively, and chose two methods：Gaussian Kriging interpolation and BPNN, to build a surrogate model. A script program was used to drive Cart3D to generate a set of numerical experiment samples. The relative error between the sample value and the result obtained by Fluent was less than 14%.Through sample point training, internal parameter optimization, and point addition strategies, a surrogate model with a relative error of less than 10% was finally obtained. It can output real-time aerodynamic characteristics data for a certain shape, which is significant for aerodynamic design to play its role in the overall design of launch vehicle. |
| Key words: Launch vehicle Overall optimization Aerodynamic calculation Surrogate model |