A framework for surrogate-based aerodynamic optimization

Kurzfassung

At DLR, an optimization framework combining different CFD solvers, Design of Experiment methods, optimization algorithms, and various surrogate modeling methodologies with sample refinement strategies for efficient surrogate-based global optimization is under development. Several Kriging predictors are used because of their ability to approximate multi-dimensional, highly-nonlinear functions. In order to find global optima accurately, the surrogate model is adaptively refined based on the Kriging error and the Expected Improvement Function. With this hybrid refinement strategy, only a few initial samples need to be evaluated, which improves the performance of the overall optimization process. Additionally, the strategy of running a local optimizer starting from the “optimum” found on the surrogate model is investigated in order to further improve the efficiency and accuracy of the framework. Two test cases indicate that the developed framework combined with hybrid strategy is more efficient.