Aerodynamic design prediction using surrogate-based modeling in genetic algorithm architecture

A new optimization algorithm is proposed and tested for four different test cases: benchmark test functions, multi-element airfoil optimization in subsonic flow, active flow control parameter optimization in transonic flow, and a direct shape optimization of an airfoil in transonic flow. The new algorithm emphasizes the use of an aerodynamic design prediction based on a global and a local surrogate modelling in genetic algorithm structure. A global response surface approximation is modelled by using low-order polynomials. A local response surface approximation is constructed by using neural networks. For all the demonstration problems considered herein, remarkable reductions in the computational times have been accomplished. The new approach significantly decreases the required CFD calls by approximately 50% in multi-element airfoil optimization; more than 70% in active flow control parameter optimization; and approximately 50% in direct shape optimization problem.