Efficient Global Optimization using a Multi-point and Multi-objective Infill Sampling Criteria

Abstract

An efficient optimization framework for the practical application of aerodynamic design is presented in this study. A multi-point and multi-objective approach infill sampling criteria (MPMO ISC) is proposed to enhance efficient global optimization (EGO) performances and best utilize parallel computing resources. A new infill sampling strategy is devised on consideration of efficient way to infill the design space is appropriate tradeoff between exploitation and exploration. It can be solved with the multi-objective approach that one objective is to search the region with high uncertainty and the other is to find the optimum point. A strategy of MPMO ISC is to determine the multiple additional points for training of surrogate model by picking the several populations on the pareto-frontier set. The EGO design framework containing the Kriging model as surrogate model, the non-dominated sorting genetic algorithm as optimizer and the MPMO ISC method is validated using the analytic function to prove accuracy and efficiency of the framework. The proposed design framework is applied for the aerodynamic design of airfoil which objective is to minimize drag of RAE 2822 airfoil, maintaining the reference lift. A total 10 design variables are used to provide variations to the airfoil shape. Consequently, an optimal airfoil shape is obtained and 25% reduction of drag is observed.