[Leander]

Hello FOAMers,

I am using RAS kEpsilon for an Atmospheric Boundary Layer simulation in order to calculate the mean wind speeds above a mountainous area. My domain is 3km long, 3km wide and 1.5km high. I first checked whether the model gives reasonable results and apparently it reproduces the measurement results from the Askervein Hill Project quite well.
Currently I am trying to run a mesh convergence study. I wrote a script that runs the simulation over and over again each time with a finer mesh. However, for some mesh configurations, the solution suddenly diverges.
I picked such a configuration that does not work and attached it as a .zip. Unfortunately the mesh was too large to upload, so I only have a screenshot of it.
When I run simpleFoam (Build : _f815a12b-20210902 OPENFOAM=2106) the solution diverges, and at the 16th time step I get a floating point exception.

The mesh seems to be okay. checkMesh prints "Mesh OK."

Do you have an idea what is wrong here? Thank you in advance.

Best regards,
Leander

[matejfor]

HI, you said it is a refined mesh, and you have 5 layers in the Z direction? I wonder how the coarse mesh looks like.

Your discretisation is pretty tight, can you post the log file with this FPO crash and checkMesh log as well?

[Kummi]

Why don't you try boundary layer mesh?

[Leander]

Originally Posted by matejfor

HI, you said it is a refined mesh, and you have 5 layers in the Z direction? I wonder how the coarse mesh looks like.

Actually I started with a very coarse mesh (I guess 25x25x5). I then increased the discretisation in x and y to 29 respectively but I haven't increased the z discretisation yet. Since it is the start of a convergence study, I do not expect the results to be correct at this coarse discretisation. Do you think such a coarse mesh might be the reason for divergence?

Originally Posted by matejfor

Your discretisation is pretty tight, can you post the log file with this FPO crash and checkMesh log as well?

I have uploaded the two logs. Sorry about the lengthy log for simpleFoam, I am solving one time step at a time.

[Leander]

Originally Posted by Kummi

Why don't you try boundary layer mesh?

Hi Kummi, what is the advantage of using a boundary layer mesh over what I am doing right now?

[matejfor]

There is one thing I do not understand. YOu are using simpleFoam, you run the case for a single iteration and you stop the case and then you restart the case running again a single iteration. Is there any reason for such an unusual solution strategy? Looking at the simpleFoam log, at iteration 10 your max velocity is too high which leads to excessive gradients and unphysical values of k and epsilon feeding momentum equation with too high Laplacian leading to continuity equation divergence.


What happens if you let your simpleFoam do it's work for a couple of hundreds of iterations?

[Kummi]

Hello Leander,
I thought mesh should be the issue. I havent looked any of your files yet. Boundary layer mesh grow from the surface and used to analyze high gradients of temperature or momentum normal to a surface.
Hope there is no error during compilation. Try to post your error message here.
Irrespective of the mesh, have you tried to locate the error in your problem ? It's a good practice to run your problem in debug mode which locates your error exactly. This helps you in future to understand the code easily. Check this link,
https://sites.google.com/site/foamgu...-debug-version
Thank you

[Leander]

Hi Matej, I was solving step by step in order to monitor the convergence. I was doing it with a small script that postprocesses after every time step and decides whether the solution has converged. Since you asked, I did a bit of research and found that this can be done much simpler by using convergence criteria in openfoam. I switched to this method now. Unfortunately, letting openfoam run for ~500 timesteps does not stop it from diverging at timestep ~10, however the solution runs much faster now. Thanks for the advice!

Hi Kumar: I have attached a working minimal example that reproduces the error. It contains the mesh so if you like, you can check it yourself. It also contains the log and the error message. Thank you for the advice to switch to debug mode. However I feel uncomfortable rebuilding openfoam since I fear messing up things and ending up with no working installation of openfoam.

I added a screenshot to show where the divergence starts. Depending on the mesh size it always seems to start at the top boundary. The fact that it starts at the edge towards the inlet might be a coincidence here.


Here is my definition of the top boundary

U:
top
{
type fixedShearStress;
tau (0.390796574 0 0);
value uniform (0 0 0);
}

p:
top
{
type zeroGradient;
}

k:
top
{
type zeroGradient;
}

epsilon:
top
{
type zeroGradient;
}

nut:
top
{
type calculated;
value uniform 0;
}