Comparison of Adaptive Design Space Exploration Methods Applied to S-Duct CFD Simulation

Abstract

Design space exploration is an important step in the design process that enables a qualitative and quantitative understanding of physical phenomena, tradeoffs, and constraints. Additionally, design space exploration can be used to create surrogate models for later design and optimization tasks. For large design spaces that are explored with computationally expensive tools such as CFD or FEM, the design space exploration method must be efficient to obtain a good representation of the trade space with relatively few sampled design points. Several design space exploration methodologies and approaches are described in the literature, including a priori sampling approaches, adaptive methods, and multi-fidelity strategies. The main goal of the present paper is to conduct a comparison of some of these methodologies to understand their exploration behavior and efficiency when used with high fidelity and computationally expensive simulation tools. This goal is achieved by applying the methods to the same test case: a design space exploration of the internal aerodynamics of s-duct geometries to assess pressure loss performance. The aerodynamic analysis is conducted with a viscous 3-D CFD solver that allows multi-fidelity results to be easily obtained by changing the mesh density.