No convergence after mesh refinement
 

Hi everyone,

I am using icoFoam in OF2.3 to calculate laminar flow in a 3d diffusor.
Everything works fine for a certain mesh resolution. But as soon as I refine the mesh, the calculation of p does not converge. The solver aborts calculations after 1001 Iterations.

I already tried reducing deltaT.
For the refinement I used refineMesh and also ICEM CFD.

Some specs of the case:
Re ~900
d_in =3mm
d_out ~30mm
2*alpha ~8°
u_in ~3m/s (uniform)
nu =35.4e-6

Does someone have an idea on how to solve this problem?

thanks
Andreas

Hi Andreas,

post your case and I can have a look. Without the case files, there is no way to find the problem.

kind regards
Florian

Hi Florian,

i attached all files that I think contain relevant information.
If something is missing let me know.

thanks in advance

Andreas

Are fvSchemes, fvSolution and checkmesh helpful?

Hi,

add at the end of your controlDict:

adjustTimeStep yes;

maxCo 0.3;

I am not sure if icoFaom can handle this.

There is smth wrong in U chang it to:
OUT
{
type zeroGradient;
}

Why you use scotch for decompose??? I prefer simple:

numberOfSubdomains 16;

method simple;

simpleCoeffs
{
n ( 2 2 4 );
delta 0.001;
}

The rest is ok.

I willcheck now the other files

Hi Florian,

the maxCo is always below 1. Do you think it might help if its below 0.3? And why is that?

I think icoFoam can't handle adjustTimeStep. I have a modified pisoFoam solver that can run with adjustTimeStep. I already tried running the case with this solver and turbulence switched off. The results are similar.

I will no try to thange my U file.

And on the topic of decompose, to be honest, I didn't have patience to understand how to decompose in a good way, so i picked the first method that worked without further specification.

Hi,

you have a lot of cells for a laminar case. Why do you need so many cells??

fvSchemes change this:

divSchemes
{
default none;
div(phi,U) Gauss linear;
}

find the new fvSoloution

can you give me your boundary file in /constant/polyMesh

kind regards
Florian

You don't need maxCo 0.3. You can alos choose 0.9 smth like this. But I would control this.

You can also choose backward for your ddtSchemes:

ddtSchemes
{
default backward;
}

With this you can handle higer Co-numbers.

----------------

decompose with simple is very easy to understand:

simpleCoeffs
{
n ( 2 2 4 );
delta 0.001;
}

means that you divide your domain in 2*2*4 subdomains. 2 in x, 2 in y and 4 in z. But that depands on your case. With simple you cannot do anything wrong. I do not know scotch.

In your U file for OUT you use zeroGradient with a value. In my opinion this makes no sense because it is zeroGradient and not fixedValue.

can you give me your boundary file?? Because the most things that can go wrong is in this file (and off course bc)

kind regards
Florian

Here is the boundary file.

The reason why I use so many cells:
First i set up the case in turbulence modeling.
The result was a flow separation in the diffusor, as expected.
Then someone mentioned, that at at Re=900 I should use a laminar solver to prevent mistakes by the turbulence model.
So i set up the case in icoFoam.
But in that case the flow wasn't attached to the wall at all. It just went right through the diffusor to the outflow.
We figured it might be due to low resolution.

But maybe that isn't the problem at all...

ok I thing I have it. It is in your boundary file
Inlet and outlet is a patch and not a wall.
INL
{
type patch;
nFaces 5824;
startFace 10757376;
}
OUT
{
type patch;
nFaces 5824;
startFace 10763200;
}

Pherhaps refineMesh have change this.

Try this.

But also change the other things, especially the decompose method.

That is correct. The value in combination with zeroGradient is a leftover from the tutorial files i copied. I think the value is ignored as soon as zeroGradient is chosen. I never had a problem with that in other cases.

Andreas

Hi,

you can also use pisoFoam for laminar flow. The only thing you have to do is change your turbulenceProperties file:

simulationType laminar;

That's all.
Then your turbulence->divDevReff(U) use only laminar viscosity. So there is no difference.

kind regard
Florian

Thank you so far.

I will change everything and see what happens.

Andreas

Ok. can be true, that OF ignores the value (0 0 0) for zeroGradient. I would delete it.

So I tried everything suggested.
Still 1001 Iterations on p.

Maybe lunch will bring new ideas.

Thanks anyway.

Regards
Andreas