Towards efficient high-dimensiona I aerodynamic shape optimization: Surrogate modeling via gradient-enhanced kriging

This paper presents a novel surrogate modeling method that can be potentially used for higher-dimensional aerodynamic shape optimization based on high-fidelity Computational Fluid Dynamics (CFD) methods. The key idea of this method is to use gradient to enhance the prediction of a kriging model. The gradient can be efficiently computed by adjoint method. For give number of samples points, the prediction accuracy of the surrogate model can be improved. In turn, the total computational cost of building a surrogate model at a give accuracy level can be reduced. The presented method is validated by analytical function and demonstrated for building the surrogate model for the object function of an airfoil inverse design problem, with 20 design variables. It is preliminarily shown that gradient-enhanced kriging is promising for high-dimensional aerodynamic problems.