工程翼型气动特性数据挖掘与建模

Modeling and data mining of engineering airfoil aerodynamic characteristics

以伊利诺伊大学香槟分校UIUC的工程翼型库为研究对象,首先通过几何数据直接对比以及基于型函数/类函数变换(CST)的参数化方法实现了重复翼型和数据异常翼型的清洗;接着,对CST参数的取值分布规律进行分析发现其近似呈正态分布,对CST参数之间的关联规律进行挖掘得到翼型参数之间的相关性,并对CST参数进行聚类分析,其结果基本与工程翼型的分类一致;进一步,采用级差分析方法、SOM自组织映射方法和Apriori方法分析了CST参数与典型工况翼型气动特性之间的关系。其中,级差分析方法给出了各CST参数对气动特性影响的显著程度,SOM和Apriori方法则分析了CST参数和气动特性之间的相关性;最后,分别使用支持向量机(SVM)和深度神经网络(DNN)构建了CST参数与典型工况下气动特性之间的预测模型,在拟合和泛化能力方面,深度神经网络模型明显优于SVM模型。本文所得到的数据挖掘及建模结果可为工程翼型气动特性分析与设计提供支撑。

The data mining&knowledge discovery technology is utilized for the UIUC(University of Illinois at Urbana-Champaign)airfoil database in this study.The CST(class function/shape function transformation)method is used for the parametric geometry representation of airfoils,and those airfoils with false data or deficient geometric data can be identified by comparing the CST-reconstructed profile with the original one.The distribution of the CST parameters is analyzed and appears to be a normal distribution.The correlation analysis of the CST parameters shows that some parameters are weakly correlated.The results of the CST parameter clustering analysis are basically consistent with the airfoil classification in engineering.Moreover,the extremum difference method,SOM(self-organized mapping),and Apriori algorithm are applied for knowledge mining of the influence of CST parameters on the aerodynamic characteristics of airfoils under typical conditions.More specifically,the extremum difference method is used to obtain the significance of each CST parameter affecting the aerodynamic characteristics,while the SOM and Apriori algorithm are applied to analyze the correlation between the CST parameters and the aerodynamic characteristics.Finally,the SVM(support vector machine)and DNN(deep neural network)are separately used to construct prediction models linking the aerodynamic characteristics with the CST parameters under typical operating conditions.It is found that the DNN model performs significantly better than the SVM model in terms of the fitting and generalization capability.These mined knowledge can provide supports for the aerodynamic characteristics analysis and design of engineering airfoils.