应用数据驱动算法提高涡粘模型分离流动模拟精度  

Applying data driven algorithm to promote prediction accuracy of separation boundary simulation with eddy viscosity model

作　　者：顾忠华[1] 颜培刚[1] 刘泮宏 王祥锋[1] Zhong-hua GU;Pei-gang YAN;Pan-hong LIU;Xiang-feng WANG(School of Energy Science and Engineering,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学能源科学与工程学院,哈尔滨150001

出　　处：《吉林大学学报（工学版）》2022年第11期2532-2541,共10页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(51776047)。

摘　　要：针对雷诺平均N-S方程中涡粘模型对分离流动预测精度这一关键问题,开发了基于人工智能数据驱动机器学习算法,克服了传统涡粘模型对分离边界层流动预测过于依赖经验参数等问题,并提高了数值模拟精度。根据影响湍流演化的物理机理,通过将涡粘模型计算分离流动的计算结果作为基准,选取多个由平均流状态表征的变量作为输入变量,以高阶雷诺应力模型计算结果构建高保真度数据库,并将雷诺应力进行分解的6个变量作为输出变量,建立基于随机森林回归的由基准流场到高保真度数据的映射关系和预测模型。结果表明:预测模型对分离流动的预测精度都有明显提高。Aiming at the key problem of prediction accuracy of eddy viscosity model in Reynolds averaged N-S equations for separation flow,a data-driven machine learning algorithm based on artificial intelligence is developed to overcome the problem that the traditional eddy viscosity model relying too much on empirical parameters for the separation boundary layer flow prediction,and the accuracy of numerical simulation is improved. According to the physical mechanism affecting turbulence evolution,the eddy viscosity model calculation results of separation flow are taken as the baseline, several variables characterized by the average flow state are selected as the input variables,the high fidelity database is constructed based on the calculation results of high-order Reynolds stress model,and the six variables decomposed from Reynolds stress are taken as output variables,the mapping relationship and prediction model from baseline flow field data to high fidelity data constructed by random forest regression algorithm are established. The results show that,the prediction accuracy of the model for the separation flow is significantly improved.

关 键 词：动力机械及工程 数据驱动算法 涡粘模型 雷诺应力 分离流动 

分 类 号：V231.3[航空宇航科学与技术—航空宇航推进理论与工程]