The aerodynamic performance of vehicles which experience uncommon aerodynamic phenomena, such as high angles of attack or nonlinear effects, may not be accurately captured by typical design tools. More complex analysis tools, such as Euler or NavierStokes CFD, may be required to accurately predict vehicle behavior. Unfortunately, increased analysis complexity carries with it increased analysis cost, and this increased cost may quickly become excessive. Even surrogate modeling of these tools may be infeasible if the design space has many dimensions. Two promising techniques for reducing the cost of developing surrogate models of Euler-level CFD results have been identified: contour-based sampling and multi-fidelity modeling. In this effort, those techniques are applied to a simple two-dimensional test problem to assess the degree to which they reduce the cost of developing accurate surrogate models. Contour-based sampling was found to significantly reduce the number of samples required when only a certain portion of the response is of interest. When combined with the contour-based sampling technique, multi-fidelity techniques did not produce significant improvements for this small test problem; however, multi-fidelity techniques may still offer benefits for larger-scale applications.
Nomenclature
CFD= computational fluid dynamics C m = pitching moment coefficient OML = outer mold line = angle of attack, in degrees