Output-based Grid Adaptation for Aerodynamics
Krzysztof Fidkowski, Massachusetts Institute of Technology
Solutions to complex aerodynamic systems generally contain multiple, possibly anisotropic, length scales arising from boundary layers, turbulent eddies, geometry, and singular features such as shocks and finite-angle corners. Finite-element computational meshes for these applications have to be constructed so as to adequately resolve these features, without making the resulting numerical process prohibitively expensive. In practice, the construction of such meshes usually relies on a “user in the loop.” That is, the mesh size at all or some stages of the solution process is dictated by a user who has experience or intuition about the behavior of the solution. For industry-level problems, the cost of this user input in terms of man-hours is often on par with or even exceeds the computational cost of the solution. Thus, for practical applications, an automatic mesh adaptation process may provide significant savings in the total time from geometry to solution.
While methods for automatically adapting meshes exist, most are either not accurate or not robust enough for use in industry-level applications. That is, the answers derived from these methods may not satisfy the engineering-required error tolerance, or the adaptation process may “break” and not produce a solution. This research therefore addresses the issues of both robustness and accuracy in a goal-oriented mesh adaptation method. The key ingredients for such a method are an output-based error estimator, an efficient adaptation strategy, and a robust mesher. This poster describes the new findings and contributions in all three of these areas, as well as the results of applying the adaptation method as a whole to practical engineering problems.
Abstract Author(s): Krzysztof Fidkowski