Efficient approach for CFD-based aerodynamic optimization using multi-stage surrogate model

In order tofurther improve the efficiency of CFD-based aerodynamic optimization with pre-constructed surrogate model, an efficient approach using multi-stage surrogate model is proposed in this study. In the proposed approach, the initial surrogate model is constructed with quite scattered samples produced by Maximin Latin Hypercube Design (LHD) which evenly spread in the entire design space. During the optimization procedure, optimization is carried out using genetic algorithm (GA) based on current surrogate model, and new samples which locate near to the real optimal solution are successively added as to improve the approximation accuracy of surrogate model until the optimization converged. Radial basis function (RBF) is adopted for both pre-constructed and multi-stage surrogate models. Compared with pre-constructed surrogate model, multi-stage surrogate model concentrates accuracy in the meaningful region where the real optimal solution probably exists instead of the global accuracy in the entire design space, moreover, the validate procedure using extra samples is not required for multi-stage surrogate model. Thus, the scale of samples required in the efficient approach using multi-stage surrogate model is much less than the pre-constructed and the efficiency of CFD-based aerodynamic optimization is also improved. A CFD-based airfoil aerodynamic optimization is employed to validate the efficient approach using multi-stage surrogate model. The optimization results demonstrate that the optimization efficiency is greatly improved due to the proposed approach using multi-stage surrogate model.