[LuReym]

Hello,

I am running a 2D steady flow through a straight duct: the bottom wall is a no-slip wavy wall (wavelength/domain length = 0.05-0.01 and wave height/wavelength=0.025) and the opposite top wall is free slip.
Other settings are:
M=0.1 to 0.8
Reynolds number based on domain length is 2e+06
Inlet total pressure imposed
Outlet static pressure=0

later, I will be adding pressure gradients to the flow (by changing the shape of the top free slip wall)
--> I use SST turbulence model to predict separation accuratley later on, with pressure gradients.

My problem is the following:
I ran several meshes to try and satisfy the y+=1 condition near the no slip wavy wall (meshes with wall distance from 1e-06m to 1e-03m)
I manage to get y+between 0.1 and 1 for M=0.1 with a wall distance of order 1e-03 but when I increase to M=0.8, I get y+ four orders of magnitude greater at the wall.

Should I refine the mesh near the wall by four orders of magnitude for M=0.8 (with the risk that high AR cells might give convergence issues)?

Also, I don't know if this interferes with y+:
I have selected auto timescale and first order resolution turbulence numerics, because these are fairly simple geometry and flow, so I don't expect anything "exotic" to happen.

Any hint is welcome.
LuReym

[ssss]

Turbulence length scale as something like Re ^ 3, so if you increase speed by 8 you will need to reduce more or less by 512 your smaller elements to capture the phenomena.

It would be just perfect if at M=0.8 you could maintain the same resolution as at M=0.1 isn't it?

[LuReym]

Hi ssss

Thanks for your reply!

I moved the inlet boundary further upstream with free slip walls in the upstream section. It turns out the inlet was having influence on my region of interest downstream in the duct. And I keep Re constant by modifying viscosity as I modify inlet Mach.
This gives much more consistent results... however :

From my results and the physics, "leading edge" (understand : the point where the upstream free slip wall joins the downstream no slip wavy wall) y plus is nearly impossible to keep around 1 whilst keeping it 1 elsewhere, because shear is far more important at the leading edge.
Is that correct? If so, what are the consequences of having y+>(>) 1 at the leading edge on the downstream boundary layer?

Lucie

[ghorrocks]

Have you shown that you need y+ = ~1 in your simulation anyway? It may work just fine with wall functions with a much coarser mesh. This will then be far easier to solve.

[LuReym]

@ ghorrocks

I thought SST turbulence model required y+ ~ 1 in order to predict separation correctly. Can CFX take larger y+ with this turbulence model and still predict separation? Note: I will be introducing pressure gradient later into the straight duct, that might induce separation and I wish to evaluate this.

I do get a problem with such a fine mesh at the wall, which is probably due to the high AR of those cells: residual max RMS near the wall is ~= 4E-4 and oscillate between that value and 1E-4.

Does one usually find a compromise between cell AR and y+ ?

Lucie

[ghorrocks]

I thought SST turbulence model required y+ ~ 1 in order to predict separation correctly.

It is not as simple as that. For separations in adverse pressure gradients usually, yes; but for separations off corners definitely no. Have a look at a turbulence modelling textbook for further details.

In your case I recommend you try coarse and fine meshes to see if it makes a difference in your case.

Yes, very high aspect ratio is a problem so you should refine along the surface as well as perpendicular to it. This becomes pretty expensive so that is why you do it only if you really need to.