Modifying an Unstructured Roe Solver for Large Eddy Simulation

Rob Watson,Paul G. Tucker,Zhong-Nan Wang,Xin Yuan
DOI: https://doi.org/10.2514/6.2014-1441
2014-01-01
Abstract:With the ever increasing availability of computing power, attention is being turned to large eddy simulation as a design tool at the component level. To maintain integration with pre-existing design processes based on solving the Reynolds Averaged Navier-Stokes equations, legacy RANS codes are being pressed into service with minimal adaptation to perform large eddy simulations. Often, these legacy RANS solvers are based around the approximate Riemann solver of Roe. It is shown here that the Roe scheme is not ideally suited to large eddy simulations, and may require excessive user care to extract worthwhile information — a process not always possible within design timeframes. A different discretisation scheme, which is shown to conserve kinetic energy, but which also shares some of the favourable structural behaviour of Roe’s is implemented, and is then tested on a range of geometries, both canonical and industrial. One of the advantages of this approach is its ability to eliminate the smoothing terms which have a strong deleterious affect on the LES performance of Roe. This allows explicit subgrid models to be sensibly applied. The new scheme, with explicit turbulence modelling, is found to outperform the Roe scheme for LES in all of the test cases.
What problem does this paper attempt to address?