[sberg]

I've tried to run some 3D simulations with a Reynolds stress model lately, but they will not converge. I've played around with the settings and variables but the simulations remains unstable. However, with the k-epsilon model the simulations converges nicely. Have anyone else had succees with the RSM, or am I doing something utterly wrong?

[hjasak]

It should work without problems. If you are running something relatively small and simple I wouldn't mind having a look, just to make sure RSTM is still fine.

Hrv

[sberg]

It's a small case where two square ducts, each of them with dimensions 0.1*0.1*1 (m), are joined together in a 90 degree bend. Air is flowing through at a Reynolds number of 100,000. The turbulence intensity, and the turbulent length scales compared to the outer dimensions, are both assumed to be 10 %.

The kEpsilon and RNGkEpsilon works well on this geometry, but I cannot obtain a solution with the LRR or LaunderGibsonRSTM models.

[sberg]

[hjasak]

Steve,

I don't think you managed to attach the file.

Hrv

Hi,

Im a beginner att openFOAM but I have tried the turbmodels you mension and I have had a similar problem which i got thru by having a epsilon value 3 times as big as when I used the kepsilon models.

/erik

[sberg]

I'll try again:-)

Here is a link to the case: http://home.chello.no/~sberg/bend.tar.gz


Steve

[hjasak]

I don't see a problem: it converges monotonically. This is what I did:
- run 100 iterations with k-epsilon
- calculated Reynolds stress using R
- restarted with Launder-Reece-Rodi RSTM.

Converges smoothly with no trouble. I can either give in normal stress only at the inlet of the full equilibrium R from the k-epsilon model.

A few comments:
- your numerics is rubbish: you should try using second order
- if you try to start up LRR from the beginning it blows up because of epsilon. You should really start k-epsilon and restart; otherwise you will have fun trying to set up under-relaxation coefficients
- at the end of the residual graph you start hitting solver convergence tolerances: hence the plateau
- I can make this even better using my fixedMeanValue outlet b.c.: makes it prettier close to the outlet



Hrv

[sberg]

Hi Hrvoje

I didn't now about the "R" function.

When I followed you steps the solution converged without problem.

Thank you

Steve

[Daniele111]

Hi
I try to use LRR for my simulation. I'm simulating an atmospheric boundary layer, but my result with it are completely different from the same case with k-eps model. Post my case.
Thanks

[CHARLES]

Originally Posted by hjasak

A few comments:
- your numerics is rubbish: you should try using second order
Hrv

Hrvoje,
Could you please provide an fvSchemes and fvSolution that aren't "rubbish"?
I'm not very well-versed in numerical methods (yet!) and would like to get a better idea of what 'good' settings for LRR are.

Thanks!