[JohnMartinGodo]

Hi
Does anyone have experience with the overset mesh functionality in the 1706 and 1712 releases from openfoam.com? I am running a simulation of an oscillating 3D wing, with an inner oscillating mesh comprising the wing and a stationary background mesh. I started out with a 2D simulation of an oscillating airfoil and found that this worked fine. After extending to 3D I am having trouble with the overset interpolation taking an immense amount of time. Roughly speaking it seems as if 95-99% of the time OpenFOAM hangs on the step of interpolating velocities between the inner and the outer meshes. I have tried both inverseDistance and trackingInverseDistance as overset interpolation methods, but there seems to be little difference in interpolation time. Have anyone experienced similar behavior and found a way of coping with it?

I am running the simulation in parallel on 20 cores, and I suspect that some performance decrease might come from inter-processor communication. Does anyone have experience with how the choice of decomposition method affects the performance of the overset mesh interpolation?

I am relatively new in OpenFOAM, so please forgive me if I am asking a trivial question.

[omdemircan]

Hi there,

I have conducted a somewhat simple test on this some week ago myself and you're right, the overRhoPimpleDyMFoam in OpenFOAM v1706 and v1712 is very, very slow.

The mesh I used for rhoPimpleFoam and overRhoPimpleDyMFoam is attached.
mesh.png
Top one is the mesh (27300 structured cells) for rhoPimpleFoam and the bottom one for overRhoPimpleFoam (total of 28200 structured cells). For your info, the red mesh is the overlapping mesh. You can see that the interpolated region is exactly the same, as I intended to also test the effect of interpolation on the solution.

As for the results, I have conducted analyses in 2D, both serial and parallel. I didn't give the overlap mesh any motion, meaning both analyses are static, no dynamic motion is present.

For the serial analyses (note that I used same deltaT, same turb model, same everything), the overset solver is over 10 times slower than rhoPimpleFoam. When I used 56 processors, both analyses run faster. However, this time the overset solver is relatively over 40 times (yes 40 times!) slower than rhoPimpleFoam. Same would certainly happen for 3D analyses, even greater slowdown can be encountered.

Other than solution-time, your solution is also affected regardless of how your overlapping mesh is in coherence with your background mesh, as I have tested. Not only your solution slows down, but the fluctuations of pressure and velocity in the flow field are dissipated.

[JohnMartinGodo]

Hi Omdemircan
Thanks for sharing your experiences on this! My testing has been with overPimpleDyMFoam (incompressible single-phase), but other than that it seems as if we are struggling with somehow the same challenge.

I have looked a bit more closely through my setup since my post above, and found at least one important change which can speed up the simulation. My case is symmetric, and in the first runs I had set all the boundary conditions on the outer boundaries of the overset region as the overset type. This means that a lot of cells on the symmetry plane were attempted interpolated between the inner and the outer regions. This plane cuts through the wing and hence comprises an area with a lot of small cells. After changing the boundary condition on the inner mesh symmetry plane to symmetryPlane the simulation ran much faster. The overset interpolation procedure still represents a significant part of the total run time, so any advice on how to speed this up would be highly appreciated.

[sepehr_s]

Hi
i simluate an airfoil in pitching motion. mesh movement is ok but force coeffs have fluctuation.
is there any way to remove this fluctuations?
thanks

[JohnMartinGodo]

Originally Posted by sepehr_s

Hi
i simluate an airfoil in pitching motion. mesh movement is ok but force coeffs have fluctuation.
is there any way to remove this fluctuations?
thanks

Hi sepehr_s
I have experienced high-frequency fluctuating force coefficients, possibly similar to yours. These were oscillating with a period of 2-3 time steps, and plots can be seen in the attached image. After trying a lot of different things it turned out that the problem was that the motion input file used in the motion function "tabulated6DoFMotion" was written with too long time steps and too low precision. It was written with only 10 000 steps per oscillation and six decimal places for positions and angles. The low resolution in time and the round-off errors due to few decimal places probably messed up numerical approximations to derivatives of position in the motion solver. I increased number of steps per oscillation to 1 000 000, making sure it is at least an order of magnitude finer in time than the simulation using it, and increased the number of decimal places to twelve. After this all the noise disappeared

[sepehr_s]

Hi JohnMartinGodo
thanks for rapid reply
i apply your comments to my case but fluctuations in result not disappeared. i also choose low amplitude and frequency in airfoil motion but fluctuations remain.
(check picture please)
any help please.

[mAlletto]

I would guess the oscillation in the forces come from the interpolation of the velocity in the overset region: Usually the interpolation is not mass conservative. This leads to local violation of the continuity which has to be balanced by the pressure equation.



I think this results in oscillation of the pressure and hence also of the forces. Did you try with a finer mesh?

[fsifsi]

Hi sepehr_s
Could you post the fvSchemses and fvSolution in your setting ?. I think the fluctuation comes from these setting.

[fsifsi]

Hi,
can you post the fvSchemes and fvSolution ?

[bennn]

Hi all,


I've carried out a couple of tests, and here are the settings that will speed up the overset simulation:


- Use a hierarchical decomposition. scotch was very slow for me.
- use inverseDistance with no other setting : the default bounding box is best
- do not use oversetInterpolationRequired


I could achieve a 10x speedup thanks to these settings, going from 100% to 10% of fluid simulation time.

[bentkj]

Originally Posted by bennn

Hi all,


I've carried out a couple of tests, and here are the settings that will speed up the overset simulation:


- Use a hierarchical decomposition. scotch was very slow for me.
- use inverseDistance with no other setting : the default bounding box is best
- do not use oversetInterpolationRequired


I could achieve a 10x speedup thanks to these settings, going from 100% to 10% of fluid simulation time.

Is there a reason why you do not advise using oversetInterpolationRequired? It seems as though this is needed for additional equations like turbulence modelling. Would you still be able run turbulent cases without oversetInterpolationRequired?

[minh khang]

Why dont they minimize the overlapping area in the interpolation?

[Hussamsababha]

which decomposition method are you using. I find scotch extremely slower than others. Try changing the decomposition and see what you get !