Grid for LES
 

Hello,

I'm trying to proceed to a Large Eddy Simulation (with the Smagorinsky model for the SGS model) in a diffuser but I don't know exactly how to choose the spatial discretization (on the boundary layer and on the main internal flow). Indeed, I would to know how to choose the spatial discretization in order to ensure that the cut-off frequency (then the filter) is enough small to obtain good results.

Thanks in advance for your help,

Andre.

Re: Grid for LES
 

(1). Looks like that you are worrying about the mesh resolution and the accuracy of the solution. (2). I don't fool around with this big white elephant, LES. I think it requires a lot of resources. (3). You can try a couple of variations, a uniform mesh (body aligned mesh) and a non-uniform mesh (with finer mesh near the wall). (4). At the end of the calculation (I have no idea about how many time steps are needed, but the longer the better.), first try to define your "good results". Then, based on this criterion, find out which one of the results is a good result. (5). I don't know whether you have to match the transient solution or not. (assuming that the diffuser flow is oscillating, or separating.) (6). For non-separated diffuser flows, you probably can get much better results using Reynolds averaged equations with turbulence models. (For LES,I am no expert in this area.)

Re: Grid for LES
 

Based on my experience with LES (a Ph.D. dissertation) I agree with John completely that you should be using RANS rather than LES. It sounds like you are trying to get a solution to an engineering flow. In general, LES is only accurate at scales slightly larger than DNS resolution (especially in wall-bounded flows) or as VLES in geophysical applications where the boundary layer effects can be modelled very simply. See some of the discussion under the heading 3D-2D LES (164) that has been written previously.

Re: Grid for LES
 

Basically, what you want is to ensure that you are not having any accumulation of turbulent kinetic energy at the scale of your mesh cell (usually, as you can see in the Smagorinsky model, the cubic root of the cell volume is used as an order of magnitude). If your mesh is sufficiently dense in the right areas, the SGS model should account for the energy cascade towards scales smaller than your grid resolution. However, be aware that the Smagorinksy model is over-dissipative. A good way to find out if you are not accumulating energy is to carry out an FFT (Fast Fourier Transform) of your solution and see if you have a peak near the length scale of your mesh which would denote energy accumulation. Simple FFT routines are often available in ready-to-use form in libraries on most computer platforms. Hope this helps.