Comparison of infill sampling criteria in kriging-based aerodynamic optimization

Kriging-based optimization methods are of great interest for aerodynamic design optimization where high-fidelity thus timeconsuming computational fluid dynamics (CFD) are used. In the framework of kriging-based optimization, a core technique called sampling infill criteria (also called adaptive sampling) is used both to search the design space and to refine the surrogate models. Among all the infill criteria, expected improvement (EI) is the favorite one while others are also still in use. However, there is little research compared these criteria when kriging-based optimization method is applied in the aerodynamic design optimization. The following paper addresses this issue by investigating 6 types of infill criteria in the drag minimization problem of the RAE2822 airfoil with different number of design variables. For each infill criterion and each number of design variables, the optimization process is repeated 30 times, and the averaged value and standard deviation are compared. The results show that, some criteria can be comparable to EI or even slight better than EI, while some criteria are much worse.