[Alasir]

Hi.

I'm a student and have most worked with RANS simulations, but are now starting work more with LES.

One thing that I would like to understand better is the concept of mesh independence. As far as I understand, because LES simulates smaller and smaller vortexes the more you refine the mesh, you will never achieve a mesh independence in a strict sense.

So I am wondering what strategies are commonly used around this problem when verifying your simulations? Do you keep refining it until its "close enough"? Do you aim for a spesific "mesh to scale ratio"?

Or have I perhaps missunderstood completely and this is not a issue at all?

[ykanani]

Hello,
yes I believe you are correct to some degree. However, it is a very broad question and hard to answer it shortly (even with my limited knowledge) .

I think by assuming that LES is used in a appropriate way inline with assumptions made in the LES formulation (e.g. having fine enough grid that the subgrid scale falls in the inertial sub range), one can expect reaching the grid independent solution. But perhaps as you mentioned, the variation of the parameter that you are monitoring may get very small but it may not remain constant.

Also, the other way is to compare the results with the experiments and see if your predictions are "close enough".

I believe it is typically easier for LES predictions to be grid independent in unbounded flow where the assumption of the sub-grid scale being homogeneous and isotropic is valid. On the other hand, my experience shows that you may need to get to the DNS-like accuracy near walls if you are for example trying to simulate transition to turbulence or turbulence itself in the boundary layer flows. My understanding is that for the wall bounded flows, there is no way for LES to "model" the effect of near wall structures (which play the most important role in the turbulent boundary layer flow) unless you resolve the grid well enough in all directions to be able to resolve and capture the near wall structures.


There are some studies on the grid point requirement for LES simulation (specially wall-bounded flows).
For example, you can take a look at this report by H. Choi AND P. Moin: https://pdfs.semanticscholar.org/b16...47ebacb062.pdf


I know it may not have answered your question completely, but I thought it may be a helpful starting point for you if I share my very limited knowledge on this.

Regards,
Yousef