[chivaee]

Hello CFD guys,

I'm trying to generate turbulent channel flow using wall resolved LES simulation on a collocated block structured, multi-grid accelerated finite volume code. I defined periodic boundaries on the horizontal plane, symmetry on top and no slip wall (no model) for the bottom and started with a log profile with some random noise (10-50% magnitude). The problem is the fluctuations are damped rather quickly. u_tau=1, domain: 0.5pi x 1 x pi with 48*48*48 grids (uniform in horizontal direction and stretched in the vertical with the first grid in yplus=0.25).

This could partly make sense as the initial fluctuations are not correlated etc. but there is also a suspicion that since the code is is not pseudo-spectral (unlike many available codes) and is collocated, it's highly dissipative and also not kinetic energy conserving.

What is your opinion? Do you have any idea how I should generate the correct turbulence filed?

My code has also the problem that the finer the mesh is, the more dissipative it becomes, isn't it strange?

[FMDenaro]

Originally Posted by chivaee

Hello CFD guys,

I'm trying to generate turbulent channel flow using wall resolved LES simulation on a collocated block structured, multi-grid accelerated finite volume code. I defined periodic boundaries on the horizontal plane, symmetry on top and no slip wall (no model) for the bottom and started with a log profile with some random noise (10-50% magnitude). The problem is the fluctuations are damped rather quickly. u_tau=1, domain: 0.5pi x 1 x pi with 48*48*48 grids (uniform in horizontal direction and stretched in the vertical with the first grid in yplus=0.25).

This could partly make sense as the initial fluctuations are not correlated etc. but there is also a suspicion that since the code is is not pseudo-spectral (unlike many available codes) and is collocated, it's highly dissipative and also not kinetic energy conserving.

What is your opinion? Do you have any idea how I should generate the correct turbulence filed?

My code has also the problem that the finer the mesh is, the more dissipative it becomes, isn't it strange?

The first issue is that LES is a 3D simulation, you must solve the entire 3D channel, no simmetry at half-height ... you are forcing a false solution

[cfdnewbie]

Originally Posted by chivaee

Hello CFD guys,


My code has also the problem that the finer the mesh is, the more dissipative it becomes, isn't it strange?

How did you determine that? Is the code correctly verified?

[chivaee]

Originally Posted by cfdnewbie

How did you determine that? Is the code correctly verified?

Well the code is verified before by other researchers. Here I just observed if I refine the mesh, it pretty quickly becomes laminar by damping the initial perturbations ...

[chivaee]

Originally Posted by FMDenaro

The first issue is that LES is a 3D simulation, you must solve the entire 3D channel, no simmetry at half-height ... you are forcing a false solution

That's a good point and I agree that LES is in fact 3D, but does it mean you are never allowed to use symmetry BC in LES? Cause I've seen quite a few papers (mainly atmospheric boundary layer simulations) that have set the top BC as a symmetric surface (I guess they've presumed that the top of the domain can be treated like a laminar flow)

[flotus1]

There are circumstances when a symmetry boundary condition can be applied in LES. The main prerequisite is that the flow is laminar next to the boundary.
For a turbulent channel flow, this condition is not satisfied.

[FMDenaro]

in channel flow with LES the condition of simmetry is simply wrong and drives the flow to dissipate artificially the energy ... such a condition can be used in RANS simulation where the variable which is solved has a steady statistical meaning.

Furthermore, I suggest to run the full 3D case with and without SGS model, on the same grid, to assess the real effect of the model...