[Moreza7]

Hello,

in the most of the papers that I've read about LES, they've used structured grids for the simulation which is commonly a 2d mesh that is extruded along the spanwise direction. But in a real complex 3d geometry, it is not possible to always use a structured cubic grid and we should use unstructured polyhedral grids just like RANS or URANS.

The question is why unstructured(polyhedral) meshes are not commonly used in LES?

[LuckyTran]

They are used. You are just seeing a (good) bias due to the medium you are consuming. That is, you are reading papers published in academic journals where the CFD'ers have taken extra super duper care to use nicer meshes.

In industrial CFD, it's very common to see LES done on tets and soccer-ball meshes when there is not enough time to ensure high quality grids.

Blockbuster films with >$100 production budgets are only a tiny fraction of films produced annually worldwide. But these are the ones with the greatest visibility. Few people care about B-grade films.

[FMDenaro]

Yes, unstructured grid in LES are apparently not common if one searches references mainly in the journals used in accademy. However, some studies are reported since 20 years ago (K. Jansen) and reports can be found in CTR site.
Recently, some renewed interest in using FE (I mean DG approaches) cna be addressed.
Industrial applications must use prevalently unstructured grids owing to the complex geometries. But LES is still less common than other apporaches.


My opinion is that in using unstructured grids for LES there is still a theoretical lack in the understanding of the role of the implicit filter induced by the method and the consequent application of the test filtering required to use a dynamic modelling.

[sbaffini]

An additional aspect, related to what Filippo mentioned, is the fact that most LES "perspectives" (which is a very broad term including numerical and theoretical aspects that would be difficult to grasp for a novice or even properly qualify here) are somehow linked to a higher (>2) order discretization of the equations, which has been historically more difficult for unstructured grids.

The renowed interest in DG methods mentioned by Filippo (and SEM, I would add), which somehow simplify achieving a higher order discretization for unstructured grids, has also leaded to a larger spread of LES computations on such grids for academic works as well.

Which brings in another possible bias: you might be focusing too much on early LES works (say, before 2010).