A Study Of Aerodynamic Shape Optimization Based On Kriging Surrogate Models

Based on the Kriging surrogate model, The Efficient Global Optimization (EGO) algorithmapplied for optimization is studied in this thesis. First, based on the Simple Genetic Algorithms (SGA),the Genetic Algorithms (GAs) with the consideration of survival stress is developed, by combiningwith the adaptive multi-point crossover, adaptive crossover and mutation rate. As a result, AdaptiveGenetic Algorithms (AGA) is proposed, which improves SGA’s efficiency and capability of globalsearch. The relationship between hyper parameters and the model’s characteristics in the process ofconstructing a Kriging model is studied. Combining with the Expected Improvement (EI) function,EGO algorithm is presented based on the developed AGA and Kriging model. Furthermore, a newtype of multi-points updating strategy is presented for EGO in a parallel manner. Severalrepresentative numerical examples are selected to validate the performance of parallelized EGOpresented. The results show that the parallelized EGO with multi-points updating strategy is moreefficient and accurate than AGA and classical EGO as well. By constructing an automatic workplatform based on API function library, the developed parallelized EGO is applied for optimizing atransonic airfoil with the target of minimizing the drag coefficient. The results show that the dragcoefficient of the optimized airfoil is decreased by16.23%comparing to the original airfoil, whichindicates that the drag reduction is clear. Then, the method is applied for optimizing a flow deflectorwhich is used to control the stall of an airfoil under large attack angle condition. The optimized resultshows that the stall angle of attack is improved by almost7degree delay and the airflow separation onupper surface has been under control. Consequently, the practicability of present method in anengineering environment is well demonstrated.