航空学报 > 2019, Vol. 40 Issue (1): 522480-522480   doi: 10.7527/S1000-6893.2018.22480

机器学习技术在气动优化中的应用

陈海昕, 邓凯文, 李润泽   

  1. 清华大学 航天航空学院, 北京 100084
  • 收稿日期:2018-06-26 修回日期:2018-07-17 出版日期:2019-01-15 发布日期:2018-08-16
  • 通讯作者: 陈海昕 E-mail:chenhaixin@tsinghua.edu.cn
  • 基金资助:
    清华大学自主科研计划(205Z22003)

Utilization of machine learning technology in aerodynamic optimization

CHEN Haixin, DENG Kaiwen, LI Runze   

  1. School of Aerospace, Tsinghua University, Beijing 100084, China
  • Received:2018-06-26 Revised:2018-07-17 Online:2019-01-15 Published:2018-08-16
  • Supported by:
    Tsinghua University Initiative Scientific Research Program (205Z22003)

摘要: 近年来优化设计在气动设计中发挥了越来越多的作用,但实用性和效率是制约其发挥作用的两大障碍。在大型客机超临界机翼设计中,通过"人在回路"(依靠人的经验在优化进行过程中实施必要干预)等努力,取得了较好的效果,机器学习技术逐步得到发展。提出了利用机器学习技术模拟人在优化过程中的合理行为和作用机制,以深层次利用信息和知识,改善优化的实用性和效率。梳理了机器学习技术在气动优化中应用的发展脉络,并结合工作实践介绍了机器学习在优化设计中的典型应用。进一步探讨了深度学习在气动优化中的可能应用方式。

关键词: 气动优化, 机器学习, 空气动力学设计, 计算流体力学(CFD), 深度学习

Abstract: While design optimization has been widely used in the aerodynamic design in the last decades, its incapability to efficiently obtain applicable solutions has seriously limited its utilization potential. The proposal and utilization of so-called "man-in-loop" methodology (to use expert's experience to interfere the optimization process) in supercritical wing design of commercial airliners has shown impressive performance improvement. Regarding the rapid development of machine learning technology in recent decades and to improve both the efficiency and applicability of the current aerodynamic optimization methods, this article proposes to leverage machine learning technology to imitate and substitute experts' rational interference inside the optimization loop to automatically further utilize essential information generated during optimization. For such purposes, this article reviews the history of the development and utilization of machine learning technology related to aerodynamic optimization and presents several typical cases based on authors' engineering practices. Furthermore, this article discusses possible utilization forms of deep learning technology in aerodynamic optimization.

Key words: aerodynamic optimization, machine learning, aerodynamic design, Computational Fluid Dynamics(CFD), deep learning

中图分类号: