[cryabroad]

Hi Guanjiang,

I'm using nutUSpaldingWallFunction on solid walls, which is the one I found the most robust. Again, I'm not doing very rigorous internal flow simulations, so this should not be regarded as the only choice. In fact, I know for a fact that if the mesh is fine enough, one should be able to NOT use any wall functions and get a good result.

If you meant the SGS model, I'm using the standard Smagorinsky model. Although for internal flows maybe WALE is a better choice.

Ruiyan

[guanjiang.chen]

Originally Posted by cryabroad

Hi Guanjiang,

I'm using nutUSpaldingWallFunction on solid walls, which is the one I found the most robust. Again, I'm not doing very rigorous internal flow simulations, so this should not be regarded as the only choice. In fact, I know for a fact that if the mesh is fine enough, one should be able to NOT use any wall functions and get a good result.

If you meant the SGS model, I'm using the standard Smagorinsky model. Although for internal flows maybe WALE is a better choice.

Ruiyan

Hi Ruiyan,

Thank you for your reply!
You mentioned the WALE model might be better, do you have some references? I am simulating the flow past a cylinder. The result of dynamic smagorinsky and WALE are close. The standard smagorinsky did not work well. Some papers recommend dynamic smagorinsky, while others recommend WALE.

Regards,
Guanjiang

[Santiago]

Originally Posted by guanjiang.chen

Hi Ruiyan,

Thank you for your reply!
You mentioned the WALE model might be better, do you have some references? I am simulating the flow past a cylinder. The result of dynamic smagorinsky and WALE are close. The standard smagorinsky did not work well. Some papers recommend dynamic smagorinsky, while others recommend WALE.

Regards,
Guanjiang

I'd suggest you to avoid taking conclusions on turbulence models using studies where OpenFOAM has been used as the tool for making such an assessment. Many of these studies are biased, and sometimes even irreproducible (missing details such as filters used, schemes, grid spacing/convergence study, Co or dt, amount of numerical dissipation, type of averaging procedure in post processing, convergence issues, incomplete code, etc.) Oftentimes is difficult to discern whether the authors got a reasonable result for the wrong reasons (i.e: numerical dissipation of the order of physical disippation, uDNS being presented as iLES, etc.)

I suggest you instead to do the assessment yourself, in a case you can run without wall models, having in mind the correct grid spacings for wall-resolved LES, and post process the results. Don't stick only with the mean values of the primitive fields, also focus on getting the flow invariants. Examples: retro-calculating the Reynolds number, or Nusselt, the Strouhal number of a wake, getting the -5/3 slope for the inertial part of the energy spectrum, near-unity relation between production and dissipation near walls (at y+~11), calculating the cross correlation component of the forcings/stresses, etc. In general, in archetipal flows (and non archetipal) there are invariants that you must recover from the simulation, that depend of turbulence.

I tell you this as someone who wrote an article comparing solvers and LES models, using OpenFOAM (yeah, I'm a hypocrite, I know).

[guanjiang.chen]

Originally Posted by Santiago

I'd suggest you to avoid taking conclusions on turbulence models using studies where OpenFOAM has been used as the tool for making such an assessment. Many of these studies are biased, and sometimes even irreproducible (missing details such as filters used, schemes, grid spacing/convergence study, Co or dt, amount of numerical dissipation, type of averaging procedure in post processing, convergence issues, incomplete code, etc.) Oftentimes is difficult to discern whether the authors got a reasonable result for the wrong reasons (i.e: numerical dissipation of the order of physical disippation, uDNS being presented as iLES, etc.)

I suggest you instead to do the assessment yourself, in a case you can run without wall models, having in mind the correct grid spacings for wall-resolved LES, and post process the results. Don't stick only with the mean values of the primitive fields, also focus on getting the flow invariants. Examples: retro-calculating the Reynolds number, or Nusselt, the Strouhal number of a wake, getting the -5/3 slope for the inertial part of the energy spectrum, near-unity relation between production and dissipation near walls (at y+~11), calculating the cross correlation component of the forcings/stresses, etc. In general, in archetipal flows (and non archetipal) there are invariants that you must recover from the simulation, that depend of turbulence.

I tell you this as someone who wrote an article comparing solvers and LES models, using OpenFOAM (yeah, I'm a hypocrite, I know).

Hi Santiago,

Thank you for your suggestions! Those are also what I am doing now. I will try my best to do that.

Regards,
Guanjiang