k-epsilon boundary condition in openFoam
 

I am trying to validate k-epsilon model on openFoam for a simple turbulent channel via the pisoFoam solver. The boundary condition being used for k and epsilon is the kqRWallFunction. While the previous forum posts suggest that kqRWallFunction simply applies a zero gradient condition for k at the wall, I'm not seeing dk/dy=0 at the wall for the solutions I am getting. In fact, for the converged solution, there seems to be a very steep gradient in k at the wall, and k(y=0) seems to be non-zero. My first grid point at the wall is at around y^+=1.

I am still not very good at reading the source code, and I found scare documentation for this. So does anyone know what exactly the kqRWallFunction wall BC means ? I'm guessing it is some sort of log-law based wall-function, and maybe my first grid point needs to be outside the viscous region.

Any insight on this would be great.

Amitabh

You are right about your first grid cell. If you got y+ around wall, you should not use wall functions. How did you calculate y+? (See this thread: http://www.cfd-online.com/Forums/ope...-testcase.html )

You can find the wall functions here:
src/turbulenceModels/incompressible/RAS/derivedFvPatchFields/wallFunctions

Of special interest is the epsilonWallFunction, have a look at epsilonWallFunctionFvPatchScalarField.C , the member function updateCoeffs (line 175 and further). Here you can see what is actually being calculated (G is the production term).

Thanks Bernhard, this was very helpful. The source code seems to indeed use a log-law based wall function for epsilon. I clearly need to use a larger grid size near the wall.

I am basically imposing pressure gradient in the flow, so it is then easy to calculate u_tau from dP/dx and channel half-width H. I calculate l^+ as nu/u_tau. Let me know if this does not sound right.

thanks !

Amitabh

You could also choose to switch to a Low-Re RANS-model, but than you may have to do some implementation yourself. It depends a bit on your needs.

I am not 100% sure about your way of calculating y+, but you can easily check it afterwards with the utilities provided in the thread I linked above. You may want to just calculate y* if you decide to coarsen your mesh.

I had a follow-up question. It seems to me that (for high-Re RANS) there should be a slip velocity at the wall if y+ of the first grid point is large. The slip velocity can be obtained from the log-law, i.e. U=u_tau*(log(y+)/kappa+B_i). Alternatively dU/dy=u_tau/(kappa*y) can be imposed.

For all the example problems I have seen, U=0 is imposed at the wall, which seems wrong.

Again, any insight on this would be great.

regards

Amitabh

p.s.: seems like I'll need to start writing my own boundary condition routines soon

Dear Amitabh,

did you finally managed with k-epsilon? I did a periodic channel flow simulation with different turbulence models. I tried low Re k-epsilon model, but did not manage to get the right velocity gradient at the wall, no matter what wall treatment I'm using. My wall+ is around 0.8.

In any case my wall gradient is under estimated, it is around 2/3 of the DNS database I'm using. If I use v2f instead, everything is fine.

Thanks,
Lilla

I did indeed manage to figure out the low Re k-epsilon last year -- it seemed to be working ok. I can send you the files if you can somehow send me your email, since there are upload limits here.

Thank you very much! My email is kapalilla@gmail.com

Hi

Is it possible for you to send me the low Re k epsilon model files too? I am trying a similar exercise in validating the model for channel flow with periodic inlet and outlet. My email id is manan.lalit@gmail.com. Thanks.