航空学报 > 2024, Vol. 45 Issue (15): 129579-129579   doi: 10.7527/S1000-6893.2023.29579

基于Liutex的数据驱动湍流模型修正

龙家俊1, 刘陈飘1, 秦飞1, 张加乐2, 徐圣冠3, 高宜胜1()   

  1. 1.南京航空航天大学 航空宇航学院,南京 210016
    2.厦门大学 航空航天学院,厦门 361102
    3.南京工业大学 机械与动力工程学院,南京 211816
  • 收稿日期:2023-09-13 修回日期:2023-09-28 接受日期:2023-10-17 出版日期:2023-10-25 发布日期:2023-10-24
  • 通讯作者: 高宜胜 E-mail:gaoyisheng@nuaa.edu.cn
  • 基金资助:
    国家自然科学基金(12102185);江苏省高校优势学科建设工程(2018年);南京航空航天大学校创新计划项目(xcxjh20220116)

Liutex based data-driven turbulence model correction

Jiajun LONG1, Chenpiao LIU1, Fei QIN1, Jiale ZHANG2, Shengguan XU3, Yisheng GAO1()   

  1. 1.College of Aerospace Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing  210016,China
    2.School of Aerospace Engineering,Xiamen University,Xiamen  361102,China
    3.School of Mechanical and Power Engineering,Nanjing Tech University,Nanjing  211816,China
  • Received:2023-09-13 Revised:2023-09-28 Accepted:2023-10-17 Online:2023-10-25 Published:2023-10-24
  • Contact: Yisheng GAO E-mail:gaoyisheng@nuaa.edu.cn
  • Supported by:
    National Natural Science Foundation of China(12102185);A Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institution(2018);Nanjing University of Aeronautics and Astronautics Innovation Program(xcxjh20220116)

摘要:

现有广泛应用的湍流模型对于大分离流动问题的计算结果往往与实验结果存在着较大偏差,这主要是由于大分离流动中存在的强逆压梯度导致湍流模型的基本假设不再成立。为了提高大分离流动问题的计算精度,需要对湍流模型进行修正。采用流场反演和数据驱动相结合的方法,提出了基于Liutex的Spalart-Allmaras (S-A)一方程模型生成项的神经网络修正方法。首先采用基于离散伴随的流场反演获得S-A一方程模型生成项的修正系数;由机器特征选择方法表明在翼型附近大分离流动区域本文选取的Liutex变量相对于修正系数具有最高的相关性,适合作为神经网络输入;再将Liutex等变量作为输入构造神经网络近似S-A模型生成项修正系数,建立生成项修正系数的神经网络结构,通过S809和S814翼型分离流动结果验证了该修正方法能够有效改善大分离流动的计算精度。

关键词: 湍流模型, 流场反演, 神经网络, 分离流动, Liutex方法

Abstract:

The current widely used turbulence models often exhibit significant discrepancies with experimental results for largely separated flows primarily due to the strong adverse pressure gradients present in the flows, which violate the fundamental assumptions of turbulence models. To enhance the computational accuracy for largely separated flow problems, it is necessary to modify the turbulence models. A combined approach of field inversion and data-driven methods is proposed to introduce a neural network for the correction of the Spalart-Allmaras (S-A) one-equation model based on Liutex. Initially, correction coefficients for the S-A one-equation model generation terms are obtained through field inversion with discrete adjoint. Feature selection indicates that Liutex exhibits the highest correlation with the correction coefficients for the S-A one-equation model generation terms and is suitable as an input for the neural network. Subsequently, Liutex and other variables are used as inputs to construct a neural network to approximate the correction coefficients for the S-A model generation terms, establishing the neural network structure for the generation term correction coefficients. This correction method is validated through separated flow results for the S809 and S814 airfoils, demonstrating the ability to significantly improve the computational accuracy of largely separated flows.

Key words: turbulence model, field inversion, neural network, separated flows, Liutex method

中图分类号: