Dynamic Adaptive Mesh Refinement for Wall Modeled LES of Complex Configurations

The objective of this proposal is the development and demonstration of a dynamic adaptive mesh refinement (AMR) capability for wall modeled large eddy simulations (WMLES) of complex configurations using arbitrary grid types including unstructured grids with hybrid element types. Our approach consists of linking legacy CFD codes to an octree mesh subdivision library, which is highly scalable and can handle all element types including tetrahedra, prisms, pyramids and hexahedra. Through the definition and implementation of a suitable API, our approach will enable dynamic AMR at scale on HPC systems with no internal changes to existing CFD codes.Although many WMLES applications can be considered as statistically steady or quasi-steady problems for which an optimized final static grid can be constructed either manually or through repeated application of AMR, the development of a fully dynamic AMR capability offers several advantages. Firstly, an efficient dynamic AMR capability enables more frequent mesh refinement passes during the simulation phase, opening up new possibilities in the use of refinement criteria, while at the same time retaining all the capabilities of quasi-steady AMR approaches. At the same time, the tightly coupled nature of the dynamic AMR library will provide a more streamlined and simpler to use workflow which can be run seamlessly on large scale HPC architectures including heterogeneous GPU hardware. Finally, we anticipate that truly dynamic AMR capabilities will be required for grand challenge problems and certification by analysis, which involve multiple time scales such as dynamic maneuvers and/or relative body motion, in addition to the unsteady nature of scale resolved turbulence eddies.