真空羽流智能化计算  

Intelligent computation of vacuum plume

作　　者：蔡国飙[1,2] 张百一 贺碧蛟 翁惠焱[1,2] 刘立辉 CAI Guobiao;ZHANG Baiyi;HE Bijiao;WENG Huiyan;LIU Lihui(School of Astronautics,Beihang University,Beijing 100191,China;Key Laboratory of Spacecraft Design Optimization and Dynamic Simulation Technologies,Ministry of Education,Beijing 100191,China)

机构地区：[1]北京航空航天大学宇航学院,北京100191 [2]航天器设计优化与动态模拟教育部重点实验室,北京100191

出　　处：《航空学报》2022年第10期102-114,共13页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(51676010)。

摘　　要：真空环境中,姿轨控发动机工作产生的高温高速真空羽流会对航天器产生气动力、气动热、污染、电磁干扰和视场干扰等效应,影响航天器正常工作甚至任务成败。因此,真空羽流及其效应评估和防护是航天领域的关键科学和工程问题。直接模拟蒙特卡洛(DSMC)方法是真空羽流数值模拟的主流方法,但DSMC是一种粒子模拟方法,非常耗时,严重制约了真空羽流及其效应的评估效率。提出了一种基于卷积神经网络的直接模拟蒙特卡洛(CNN-DSMC)方法。CNN-DSMC的训练集包括两个部分:将DSMC羽流数值模型的几何拓扑与边界条件信息作为训练集的输入,将DSMC数值模拟得到的羽流流场数据作为训练集的输出。将该训练集输入卷积神经网络进行训练,可得到高精度、高效率的真空羽流智能化计算模型,以此预测不同条件下的真空羽流流场。以月球探测器月面着陆过程中的真空羽流场为例,分别使用CNN-DSMC和DSMC数值模拟了在不同着陆高度条件下的真空羽流流场流速和密度。结果显示,两种方法结果基本一致,流场流速和密度的平均相对误差分别小于6.0%和8.8%。但与传统的DSMC方法相比,CNN-DSMC方法的计算速度提升至少4个量级,最大可达6个量级。因此,本文提出的CNN-DSMC方法在真空羽流数值模拟方面具有较强的应用潜力。The high-temperature and high-speed vacuum plume generated by the attitude orbit control engine in the vacuum causes aerodynamic forces, heat fluxes, contamination, electromagnetic interference, and visual interference to the spacecraft, affecting the spacecraft operation and even the success of missions. Therefore, the assessment of vacuum plumes and their effect is a critical scientific and engineering issue in aerospace. The Direct Simulation Monte Carlo(DSMC) method is generally utilized in the numerical simulation of vacuum plumes. However, being a particle simulation method, DSMC is time-consuming, and severely limits the efficiency of vacuum plume evaluation. In this study, we propose a Convolutional Neural Networks-based Direct Simulation Monte Carlo(CNN-DSMC) method. The geometric topology information, the boundary condition information, and the flow field data obtained from DSMC simulations are employed as the training set of the CNN-DSMC. Then, a highly accurate and efficient intelligent computational model of vacuum plumes is trained based on the convolutional neural network, which can be used to predict the vacuum plume flow field under different input conditions. In addition, the velocity and density of the vacuum plume during the lunar landing are computed by CNN-DSMC and DSMC simulation at different landing altitudes. The results obtained by the two methods are consistent, and the average relative errors of the flow velocity and density are smaller than 6.0% and 8.8%, respectively. Furthermore, the computational speed of the CNN-DSMC is improved by at least four orders of magnitude and up to six orders of magnitude compared with the conventional DSMC method. Overall, the CNN-DSMC proposed in this study offers promising applications for the numerical simulation of vacuum plumes.

关 键 词：真空羽流 人工智能 深度学习 神经网络 直接模拟蒙特卡洛 

分 类 号：V211.25[航空宇航科学与技术—航空宇航推进理论与工程]