结合机器学习的SA湍流模型闭合系数修正
Closure coefficient modification of SA turbulence model combined with machine learning
徐向阳 1胡冠男 2王良军 3朱文浩 4张武1
作者信息
- 1. 上海大学力学与工程科学学院,上海 200444
- 2. 上海大学计算机工程与科学学院,上海 200444
- 3. 上海大学信息化工作办公室,上海 200444
- 4. 上海大学计算机工程与科学学院,上海 200444;上海大学信息化工作办公室,上海 200444
- 折叠
摘要
将修正Morris分类筛选法与极端梯度提升(extreme gradient boosting,XG-Boost)相结合,在计算流体动力学(computational fluid dynamics,CFD)数据驱动下,用于SA(Spalart-Allmaras)湍流模型闭合系数的修正.利用分类筛选法有效缩小闭合系数研究范围,同时依据XGBoost方法在小规模数据集下取得精度较高的拟合模型,有效提升系数修正效率.在三维DLR-F6-WB构型下进行了数值实验,实验结果显示利用本方法能够在三维复杂模型上基于小样本数据进行系数修正,修正后的升阻力系数计算精度得到了显著提升.
Abstract
This paper presents a combined approach integrating the modified Morris clas-sification and screening method with extreme gradient boosting(XGBoost),driven by computational fluid dynamics(CFD)data.The methodology is applied to modify the closure coefficient of the Spalart-Allmaras(SA)turbulence model.The utilization of the classification and screening method effectively narrows the research scope of the closure coefficient.Using the XGBoost method,a highly accurate fitting model can be obtained even with a small-scale data set,leading to effective improvements in the efficiency of coef-ficient modification.Employing this method,numerical experiments are conducted for the flow over the three-dimensional(3D)DLR-F6-WB configuration.The experimental results demonstrate the method's capability to rectify coefficients on complex 3D models based on small sample data.Consequently,the accuracy of the modified lift-drag coefficients has been significantly improved.
关键词
SA(Spalart-Allmaras)湍流模型/敏感度/极端梯度提升(extreme/gradient/boost-ing,XGBoost)/线性回归/系数修正Key words
Spalart-Allmaras(SA)turbulence model/sensitivity/extreme gradient boosting(XGBoost)/linear regression/coefficient modification引用本文复制引用
基金项目
国家自然科学基金重大研究计划重点项目(91630206)
出版年
2024
NETL
NSTL
万方数据