difference between kOmegaSST (OF-1.7.x) and kOmegaSST_lowRe (OF-1.6-ext)
I know there are already two threads existing where the kOmegaSST_lowRe model from OF-1.5-dev is discussed. Anyway, my issue is not how to set the boundary condition or how to compile the model...
Recently I had a closer look at both the kOmegaSST (OF-1.7.x) and kOmegaSST_lowRe (OF-1.6-ext) and try to figure out the differences and thus the need for the kOmegaSST_lowRe model.
Both versions calculate the near-wall omega correctly as (below fromOF-1.7.x omegaVis)
00207 scalar omegaVis = 6.0*nuw[faceI]/(beta1_*sqr(y[faceI]));
The only difference is, that in OF-1.7.x omega is calculated as the magnitude of omegaLog and omegaVis, which for y+ approx 1 is omegaVis.
The production term G is calculated as (below fromOF-1.7.x)
00211 G[faceCellI] =
00212 (nutw[faceI] + nuw[faceI])
which seems to be exactly the same than in OF-1.6-ext.
Now, the main difference I encountered is the averaging of G and omega in case of multiple boundary faces. In OF-1.7.x there is no averaging (only TODO note), whereas in OF-1.6-ext the averaging is performed.
I have no clue why its not yet implemented in OF-1.7.x (seems to be fairly easy) but in general it shouldnt affect the result that much.
To conclude, you can ONLY use the kOmegaSST_lowRe (OF-1.6-ext) model in case you are sure that you have a wall resolving grid everywhere in your domain (straight forward to test). On the other hand you can use the kOmegaSST (OF-1.7.x) model more general for both, wall resolving grid everywhere in the domain and grids, where the gridpoints normal to the wall are sometimes in the log-layer and sometimes in the viscous sublayer.
I think I dont have to mention that both models are no low-Re RANS models in the sense that they don't employ any near-wall damping terms...
I would appreciate any comments, especially if I missed anything there or misunderstood parts!
do you have an answer to my question, also located in this thread
I do not understand if the computation of the production term is also valid for low Reynolds turbulence models.
My problem is the computation of u_tau in the production term, which is according to the log law, namely
u_tau = (c_mu)^(1/4) * sqrt(k).
Do you have an idea?
not too much time at the moment, but I just had a quick look again at omegaWallFunctionFvPatchScalarField.C (OF-2.0.1).
As far as I understand, the idea of this specific wall function is to have reasonable results in case you resolve your boundary layer and in case you have your first cell in the log-law region.
As it is mentioned in the other thread, the production term is correctly calculated for the log-law region (I didn't verify that again, but assume its correct as you state there).
Now, if the viscous layer is resolved and the first cell is at y^+ approx 1, the turbulence production should be very very small (practically zero), right?! This is accomplished by using the log-law formulation for G as well: In the viscous sublayer, turbulent kinetic energy k is very small and since G ~ sqrt(k), the production term will be very small in case y^+ approx. 1. (as it should be)
You can find the same "trick" in the omega boundary conditions: omegaLog ~ sqrt(k) and thus omega = omegaVis and omegaLog is very small in case of y^+ approx. 1.
EDIT: The above explanation is specifically for the kOmegaSST RANS model in conjunction with omegaWallFunction. In case you use a "real" low-Re RANS model, you won't use any wall functions of course.
Thank you for the explanation, Florian. That makes sense!
|All times are GMT -4. The time now is 23:25.|