[night-hawk]

Hello,


Can anyone what is wrong with my case setup. I am trying to reproduce result from this https://arc.aiaa.org/doi/10.2514/1.28497 Figure 4 a.


I have attached my case setup here https://workupload.com/file/uvjsyFbmaVS


Thank you

[wkernkamp]

I could not use your RAS library so used SST without LowRe.

The solution exploded after a few iterations (as probably it did for you.

To prevent explosion, I modified the following in system/controlDict:
maxCo 0.5;

The last four iterations show:

Cl: -0.0296027 -0.0295143 -8.84969e-05 0
Cl: -0.0295736 -0.0294852 -8.83705e-05 0
Cl: -0.0296012 -0.0295127 -8.84977e-05 0
Cl: -0.0295689 -0.0294805 -8.83692e-05 0

and

Cd: 0.0142826 0.0104069 0.00387571 0
Cd: 0.0142817 0.010406 0.00387571 0
Cd: 0.0142825 0.0104068 0.00387569 0
Cd: 0.0142816 0.0104059 0.0038757 0

So a slight unsteadiness remains. It has been a while, so you probably solved it already haha. Let me know what your findings were on this interesting study.

[night-hawk]

The main idea for this was to get a best numerical scheme for further simulations.
Unfortunately I was not get results better than you.

[wkernkamp]

1. The boundary layer mesh ended up in the wrong place.

I modified in Allrun.pre:

# runApplication surfaceFeatureEdges -angle 15 WhalefinBox.stl Whalefin.fms
runApplication surfaceFeatureEdges WhalefinBox.stl Whalefin.fms

also added runApplication reconstructPar at the end.

2. Took some known good solution settings

Made some changes to fvFiles in the system directory. I copied those from the OpenFOAM benchmark that we are running in "hardware".

Also modified system/controlDict to remove the maxCo parameter.

Now it runs and converges fast. On 30 cores, it takes 160 seconds.

3. Result much better

The last four iterations show Cl:

Cl: 0.759711 0.76118 -0.00146899 0
Cl: 0.759719 0.761188 -0.00146908 0
Cl: 0.759714 0.761184 -0.00146925 0
Cl: 0.759696 0.761165 -0.00146941 0

and Cd:

Cd: 0.120072 0.11106 0.00901263 0
Cd: 0.120064 0.111051 0.00901278 0
Cd: 0.120052 0.111039 0.00901319 0
Cd: 0.120036 0.111023 0.00901355 0

See interesting picture of the surface pressure.

[wkernkamp]

I forgot to mention that I also rotated the stl by 15 degrees. I think if you run what I sent, you will get the uninteresting zero degree angle of attack case. Sorry about that.

[wkernkamp]

I rotated the stl with:


surfaceTransformPoints -rotate-angle <(vector scalar)>
Rotate <angle> degrees about <vector> - eg, '((1 0 0) 45)'


In this case:
cp Whale_fin.stl Whale_fin0.stl

surfaceTransformPoints -rotate-angle '((0 0 -1) 15)' Whale_fin0.stl Whale_fin.stl


Had to keep the name as the original, otherwise the files in system need to be modified.

[wkernkamp]

I put it all together in an archive so that you can just run Allclean.pre after putting your file Whalefin.stl in the constant/triSurface directory. (Had to do that or this archive would be too large for attachment here.)


Once you are done running, you can use Allclean to remove all files generated by running.



I set the number of cores back to the 12 you had in system/decomposeParDict. Set this value no higher than the number of actual performance cores you have and not the number of threads.


If you have a six core processor, it will show 12 processes on windows due to two threads per core on x86. However, the best performance value would then be 6 and not 12 in decomposeParDict.


If you try to use efficiency cores in addition to performance cores, this will slow down all cores to the speed of the efficiency cores. The result may actually be a slowdown depending how far the efficiency cores are clocked down.