Airfoil optimizations require costly high-fidelity solvers. By replacing the expensive solver with a cheap to evaluate surrogate model genetic optimization becomes feasible and a global optimum can be obtained. This thesis investigates applicability of surrogate model based optimization for airfoil optimization and provides recommendations regarding parameterization and options for building surrogate models.

Building a surrogate model is done by firstly evaluating an initial sampling with the high-fidelity solver. A surrogate model is fitted through the training data. An iterative process follows that adds sequential sampling points and refits the surrogate model until a stopping criterion is met. The quality of the surrogate model is determined through error analysis.

CST, NURBS and Bezier-Parsec parameterization are investigated for surrogate model building. To build an effective, consistent and efficient surrogate models CST or NURBS without weights is recommended. Sequential sampling plans should be created with LOLA-Voronoi or generalized probability of feasibility. For the surrogate model Kriging is recommended with a minimum sample size of 10 times the amount of parameters.