基于深度降阶模型的翼型气动反设计方法  

Method of Aerodynamic Inverse Design for Airfoils based on the Deeply Reduced-Order Model

作　　者：闫若鹏 黄典贵[1] YAN Ruo-peng;HUANG Dian-gui(College of Energy and Power Engineering,University of Shanghai for Science and Technology,Shanghai,China,200093)

机构地区：[1]上海理工大学能源与动力工程学院,上海200093

出　　处：《热能动力工程》2021年第1期17-23,共7页Journal of Engineering for Thermal Energy and Power

摘　　要：借鉴深度学习中通过卷积核对连续数据进行特征提取的方法,提出一种翼型气动反设计初步模型。将125组NACA标准翼型作为反问题的学习样本,并随机选取8组非样本NACA翼型检验反设计模型的预测功能。模型以翼型表面压力系数分布作为输入,将翼型几何形状作为输出,通过多层数据卷积的思想,最终找到由翼型表面压力分布到翼型几何的隐式对应关系。结果表明:学习样本内的125组翼型几何数据均方差可保持在1×10^(-4)以下,而样本外的8组翼型几何数据均方差可保持在1×10^(-4)以下,说明本模型具备一定的反设计预测精度。By referring to the deep learning method,which uses convolution kernel to extract the features of continuous data,this article presents a preliminary model for aerodynamic inverse design.Firstly,125 standard NACA airfoils are taken as the learning samples of the inverse problem,and then 8 non-sample NACA airfoils are randomly selected to check the prediction function of the model.In the model,surface pressure coefficient distribution of the airfoil is taken as the input,while geometry of the airfoil is taken as the output.Finally,the corresponding relationship between the surface pressure distribution and the geometry of the airfoil is found by the function of multi-convolution.The result shows that the mean square error of the 125 sets of airfoil geometry data in the training sample can be kept below 1×10^(-4),and that of the 8 sets of airfoil geometry data outside the sample can be kept below 1×10^(-4),indicating the proposed model has a certain inverse design precision.

关 键 词：特征提取 翼型气动反设计 表面压力系数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]