Efficient Aerodynamic Optimization Using a Multiobjective Optimization Based Framework to Balance the Exploration and Exploitation

In many aerospace engineering design problems, objective function evaluations can be extremely computationally expensive, such as the optimal design of the aerodynamic shape of an airfoil using high-fidelity computational fluid dynamics (CFD) simulation. A widely used approach for dealing with expensive optimization is to use cheap global surrogate (approximation) models to substitute expensive simulation. The effective global optimization (EGO) based on Kriging model is a widely used approach for dealing with the expensive optimization problems. In the standard EGO and most Kriging based aerodynamic optimization application, one sampling point is determined for expensive simulation. To make best use of parallel computing resources, multi-point infill sampling criteria is need to improve the efficiency of aerodynamic shape optimization. In this paper, a recently developed multiobjective optimization based framework balance the global exploration and local exploitation in EGO, called EGO-MO, is introduced. It can generate multiple test solutions simultaneously to take the advantage of parallel computing. The EGO-MO is applied for the aerodynamic shape design of a transonic airfoil to minimize drag maintaining the reference lift. The class/shape transformation (CST) method is employed for the parameterization of airfoil. The open source code SU is adopted to perform the high-fidelity aerodynamic analysis of initial and infill sampling points. The comparison of EGO-MO and standard EGO for the transonic airfoil problem is presented. The investigation shows that the EGO-MO feature less iteration numbers and can give better optimal results.