[Artur]

Hi All,

I have some concerns about how one should specify the omega BC at the wall when using the kklOmega transition model.

In a standard k-omega formulation (low-Re) we would fix the wall and cell centre value of boundary cells using this formulation:
http://www.cfd-online.com/Wiki/Near-...k-omega_models
which is indeed what happens in omegaWallFunctionFvPatchScalarField.C:

scalar omegaVis = 6.0*nuw[faceI]/(beta1_*sqr(y[faceI]));
scalar omegaLog = sqrt(k[cellI])/(Cmu25*kappa_*y[faceI]);
omega[cellI] += w*sqrt(sqr(omegaVis) + sqr(omegaLog));

For a flat plate the same effect may be achieved by evaluating this expression for a known wall distance and using a fixedValue BC instead of the omegaWallFunction one.

But, the omegaWallFunction BC does not work for the kkl model, as it tries to access the G field which doesn't exist:

--> FOAM FATAL ERROR: 
    request for volScalarField::DimensionedInternalField kkLOmega:G from objectRegistry region0 failed
    available objects of type volScalarField::DimensionedInternalField are
(long list of objects)

Now the problem: if I try to use the fixedValue approach, I get very bad results (see attachment for the T3A case). If, however, I follow the recommendation given here (see sec. 3 on p. 6):
http://www.engineeringmechanics.cz/pdf/20_5_379.pdf
and use zeroGradient the results are very good. Wall values of omega obtained are several orders of magnitudes different than what the wall formulation would suggest, however.

Can someone please explain why this model requires a different omega BC than the standard k-omega formulation? Would be great if you can point to the lines in the cause that make the difference.

All the best,

Artur

P.S. my mesh is just an orthogonal, hexahedral, 2D mesh with nearly identical parameters as used in the paper by Furst; tried various 2nd and 1st order schemes, linear solvers, etc., all give consistently the same end result.

[malv83]

The choice of that boundary condition allows the RANS model to characterize relatively low-frequency fluctuations present in the fully turbulent boundary layer.

Now, if you are still using the k-kl-omega model, I want to tell you that there is already a new model in the market, the k-v2-omega model. It is a new model based on the k-kl-omega. I recommend everybody still using the k-kl-omega model to switch to the new k-v2-omega model developed by Walters group.

There are a few problems with the k-kl-omega model in the farfield. One of them is the growth of Laminar Kinetic energy when separation occurs. Lopez and Walters have a paper (have not been published yet) correcting this issue:

Maurin Lopez. D. K. Walters. “A recommended correction to the k-kl-omega transition sensitive eddy-viscosity model”. Journal of Fluid Engineering.

This correction has to be made to the 2008 k-kl-omega model from now on.

Now, Lopez and Walters also developed a new transitional model (k-omega-v2) as an alternative to the k-kl-omega one. This new model has more capabilities (it is more reliable) than the k-kl-omega model, especially in the farfield computations. Fortunately the paper for this new model is already publish.

Maurin Lopez. D. K. Walters. “Prediction of transitional and fully turbulent free shear flows using an alternative to the laminar kinetic energy approach”. Journal of Turbulence, Vol 17, Iss. 3, 2016.

If you see the papers, you will immediately see how the k-kl-omega model is not good for free shear flows, and how the new model corrects all those issues. From now on, k-kl-omega users have to start using the new k-omega-v2 model.