导航

ACTA AERONAUTICAET ASTRONAUTICA SINICA ›› 2022, Vol. 43 ›› Issue (10): 527352-527352.doi: 10.7527/S1000-6893.2022.27352

• Fluid Mechanics and Flight Mechanics • Previous Articles     Next Articles

Intelligent computation of vacuum plume

CAI Guobiao1,2, ZHANG Baiyi1,2, HE Bijiao1,2, WENG Huiyan1,2, LIU Lihui1,2   

  1. 1. School of Astronautics, Beihang University, Beijing 100191, China;
    2. Key Laboratory of Spacecraft Design Optimization and Dynamic Simulation Technologies, Ministry of Education, Beijing 100191, China
  • Received:2022-04-30 Revised:2022-05-31 Published:2022-06-17
  • Supported by:
    National Natural Science Foundation of China (51676010)

Abstract: 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.

Key words: vacuum plume, artificial intelligence, deep learning, neural networks, direct simulation Monte Carlo

CLC Number: