CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Help with K-epsilon for 2D turbulence modelling (http://www.cfd-online.com/Forums/main/106757-help-k-epsilon-2d-turbulence-modelling.html)

Help with K-epsilon for 2D turbulence modelling

1 Attachment(s)
Hi,
I'm simulating a flow though an volute casing.
I'm in the initial stages of showing grid independence.
the k-epsilon model was working fine until the no. of elements was around 2.5 lakhs.
in the 4 lakh element model im not getting a converged solution.
the residuals for continuity are not stable now.
I'm using the simple algorithm, with 2nd order upwind diff(used 1st order too; but same problem) 0.5 relaxation. its a single-phase flow.

Please suggest ways to converge this solution.

 Orkun September 7, 2012 07:46

Re : Help with K-epsilon for 2D turbulence modelling

It seems to me that there are two possibilities : numerical diffusion problem or wrong near-wall treatment.

For the case of near-wall treatment, as you probably know, when you fine you mesh near wall, you decrease the y_plus value of your closest cell to wall. If y_plus values had decreased under 5 or 6, you need to use enhanced wall functions to have solution for viscous sub-layer.

I would think that there would be numerical diffusion problem if you had not the same issue with a higher order discretization scheme.

Orkun

 ztdep September 7, 2012 21:57

another possiblity:
change from single precesion to double precesion

I am getting a y+ value of around 3 to 4.
So I am running a simulation using enhanced wall treatment.

I will also run a double precision solution after this run!

another question, how do we find the value of velocity gradient at the wall?
since the velocity profile equations themselves are functions of friction velocity (which in turn is viscosity multiplied by velocity gradient at wall)

thanks again

 Orkun September 8, 2012 06:06

You can use wall shear stress to calculate velocity gradient, or you may calculate it by a UDF.