![]() Distributed memory systems are handled using the MPI standard. The code is written in a combination of C++ and CUDA and can effectively target NVIDIA GPUs. ZEFR was developed in the Aerospace Computing Laboratory as a simple but high-performance CFD code for the purpose of developing new algorithms and applying them to useful test cases. The underlying numerical solver, ZEFR, is able to run on tensor-product elements using the DFR scheme. In this work, we outline the development of a high-order GPU-accelerated solver capable of running on moving overset grids with curved elements. romero2017direct proposed a simplified formulation of the FR approach, direct flux reconstruction (DFR), which exactly recovers the nodal DG version of FR. In 2007 Huynh proposed the flux reconstruction (FR) approach Huynh a unifying framework for high-order schemes for unstructured grids that encompasses both the nodal DG schemes of Hesthaven and Warburton HesthavenWarburton and, at least for a linear flux function, any SD scheme. Popular examples of high-order schemes for unstructured grids include the discontinuous Galerkin finite element method, first introduced by Reed and Hill reedhill1973, and spectral difference (SD) methods originally proposed under the moniker ‘staggered-grid Chebyshev multidomain methods’ by Kopriva and Kolias in 1996 kopriva96 and later popularized by Sun et al. Not only are they less dissipative, enabling the simulation of vortex-dominated flows with fewer degrees of freedom than a lower-order method, but they have also been shown to give remarkably good results when used to perform implicit large eddy simulations (ILES) and direct numerical simulations (DNS) pypeta pyfrstar. High order methods have a number of attractive attributes. OversetDG2011, without a high-order near body solver, vortical structures and other flow features incur high amounts of diffusion before (or while) passing into the high-order off-body solver, causing a large increase in error. As shown by Wissink HeliosStrand and by Nastase et al. For the best possible performance, accuracy, and applicability to a wide range of complex geometries, however, a higher-order unstructured near-body solver is preferable. Higher-order near-body solvers are available, but only for (multi-block) structured grids currently, such as with the OVERFLOW code using 5th order accurate finite differences pulliam2011 buning2016. In most production codes used for complex geometries, such as the CREATE-AV HELIOS software Helios, the near-body solvers are typically 2nd or 3rd order accurate in space. One of the many advantages of this approach, compared with say deforming grids, is that it is possible to employ multiple solvers: an unstructured near-body solver to allow for easier mesh generation around each body, and a high-order Cartesian off-body solver with adaptive mesh refinement (AMR) for speed and simplicity.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |