[Karpfen]

Hello all,

i'm very inexperienced with OpenFoam and tried to do a mesh convergence study for a 2D-simulation with the simpleFoam solver. The geometry is pretty simple and the mesh is created with blockMesh.
The perhaps not so easy features of the case are a rotating wall resulting in backflow at the outlet-patches (or more precise mostly outlet-patches due to this).
The Reynolds number is pretty high at ~100,000.
Of main interest are the forces at two stationary walls.
I used the kOmegaSST turbulence model.


My problem:

Basically, there is no clear picture or pattern. The forces vary greatly with the different solutions both undershoot and overshooting the results of a CFX-simulation done for cthe sake of comparisons at times by half an order of magnitude and even changing the sign.

Basically, i could not say at all which choice of resolution, boundary-layer grid spacing and other parameters would be fitting for this case.


Comparison to CFX-results:

I had the opportunity to to the same simulation in Ansys CFX. There it did not show the same behavior. The CFX results seem to be "converging" with increasing resolution.
CFX seems to be really insensitive for this case as changing discretization schemes and turbulence models did not change the results significantly.

While the resulting forces of CFX and OpenFOAM are of comparable magnitude for some parameter combinations, i did not manage to get OpenFOAM-results which are particularly close to the CFX-results.

What i tried: resolutions, residuals, RAS-models, some discretization schemes

I tried different spacings at the wall boundary layers: Uniform spacing, half and quarter the grid spacing at the wall compared to the biggest grid spacing.
I looked at 4 different resolutions. The coarsest just a little bit over the smallest resolution where the simulation still converges. Up to the finest mesh the resolution in each direction doubled two times.
Just to get an idea: The smallest mesh had ~5500 cells and converged after a couple of seconds, the biggest had ~90,000 and needed ~10 hours. (1 core always)

I systematically looked at different target residuals and the picture there is pretty clear and the forces not changing significantly from the residual 1e-5 to 1e-7.

I tried the k-omega and the k-epsilon turbulence model for some resolutions. The resulting forces were noticably different most of the time.

I did not get the solution to converge with some other discretization schemes.



Does anybody have any advice or has made similar observations for some cases?
Since i'm new to OpenFOAM, but i don't really know what to try to get a clear picture/understanding of the problem.

Further increasing the resolution is always a possibility, but for a 2D-case the simulation already took pretty long with the finer meshes and CFX does not seem to care even with the coarser grids.

Thank you and kind regards,
Karsten