Development of variable-fidelity based selective design and flow analysis methods for practical aerodynamic design optimization

Abstract

There have been many attempts to apply computational design methods to engineering problems in industry. However, solving an engeering design problem requires very high computational costs, even when applying computational analysis with state-of-the-art parallel computing. The main objective of the current study is to develop an efficient design methodology to reduce the required costs for both finding an optimal design solution and the computational flow analysis for the function evaluation. In this thesis, efficient design method and flow analysis are considered. Efficient global optimization starts with a small number of initial samples to construct a response surface, and additional number of samples are required during a design process. Variable-fidelity (VF) flow analysis uses both high-fidelity (HF) and low-fidelity (LF) flow analyses to evaluate aerodynamic function values according to design variables. To improve their efficiencies, three methodologies are introduced in the current study. First, dynamic fidelity indicators (DFIs) are developed to estimate necessary fidelity of the flow analysis for the additional sample points, and a corresponding selective VF design method is developed. Second, a tightly coupled Euler/panel method is used for HF flow analysis. It reduces the computational cost by decreasing domain size of computational fluid dynamics by using panel method. Third, a nonlinear vortex lattice method (NVLM) is developed for LF flow analysis. The NVLM with table look-up can simulate nonlinear aerodynamic characteristics, such as transonic and viscous flows, which classical VLM cannot simulate. To accurately calcualte the effective angle of attack for table look-up, an equivalent lattice approach is developed. Finally, a VF aerodynamic design composed of the selective VF design method, tightly coupled Euler/panel method, and the NVLM is developed. It is applied to the design problem of reducing the drag force of a wing in transonic flow. As a result, the design optimization four times faster than the conventional design method using only single-fidelity flow analysis can be achieved. Design efficiency can be further improved by using an LF-based response surface and by fine-tuning threshold parameters in the DFIs.