Surrogate Model In Airfoil Design With Multi-Output Gaussian Process Regression

Wings are the important component of the airplane to generate lift to take off and are also thecore research subject of airfoil design. The quality of airfoil design will directly affect the subsequentdesign of aircrafts. Traditional airfoil design methods including wind tunnel and CFD computation areboth time-consuming, costly and highly rely on the experience of designers. To improve airfoil designefficiency and reduce costs, urrogate models are introduced to predict aerodynamic performance. Inthis thesis, we use the multiple output Gaussian process (MOGP) regression to build surrogate model..MOGP uses the convolution of the smooth kernel functions and Gaussian base processes to modeleach output, and captures the correlation between multiple outputs by sharing the common baseGaussian processes.In this thesis, a group of NACA series airfoils and supercritical airfoils are selected. We use CFDsimulation to generate the corresponding aerodynamic performance for these airfoils using CSTparameterization method. Aerodynamic performance evaluation uses the airfoil shape parameters asinputs to predict the aerodynamic performance parameters, such as lift coefficients, drag coefficientsand moment coefficients. Therefore, the output aerodynamic performance parameters are oftencorrelated with each other. we build up the surrogate model based on MOGP and verify the significantcorrelations between lift coefficient, drag coefficient and moment coefficient using the experimentaldata. We compare our model with kriging and other two neural networks, BP and RBF. Results showthat the MOGP can efficiently improve the prediction accuracy compared with other alternativeswhen the outputs are significantly correlated. Work in this thesis is a great complement to the existingsurrogate models in the area of airfoil design.