Variable-Fidelity Aerodynamic Design Using Gradient-Enhanced Kriging Surrogate Model with Regression

The purpose of the current work is to develop efficient and yet accurate design optimization framework using variable-fidelity aerodynamic analysis. The basic idea of the variable-fidelity method is to maximize the efficiency of the analysis while maintaining the accuracy of the high-fidelity analysis. It performs a small number of high-fidelity analysis only when is needed for the function evaluations which are not accurate by lower-fidelity analysis. To explore a large design space with relatively many design variables, an efficient global optimization (EGO) is selected. However, the number of function evaluations for the global search process is often too expensive to directly carry out aerodynamic analysis using computational fluid dynamics (CFD). The Kriging surrogate model is introduced as an efficient alternative. To facilitate the variable-fidelity analysis, a corresponding variablefidelity Kriging model is developed. Gradient data are directly utilized to improve the accuracy of the Kriging model and reduces considerably the total number of function evaluations. The variable-fidelity and gradient-enhanced Kriging model is constructed and regression effect is also taken into account to mitigate the errors from the low-fidelity analysis to better predict the trend of the function of interest. The validity of the proposed Kriging model is validated for the analytic function with varying number of samples. Finally, practical design applications of both two-dimensional RAE2822 airfoil is carried out using the proposed surrogate model-based EGO design framework.