CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Turbulence model convergence problem (

Andrew August 14, 2008 08:32

Turbulence model convergence problem

I am modeling a buoyancy drive flow in a concentric cylinder annulus. For very high temperature difference, i am assuming the flow to be turbulent and used SST and Reynolds stress BSL model. I set all the initial turbulence conditions to automatic. Upon solving, i only get convergence to 10^-5 using RMS criteria at 1500 iterations. Rate of convergence is very low. Can anyone tell me why is that so? Have i set anything wrong?


Andrew August 14, 2008 08:33

Re: Turbulence model convergence problem
Note: i am running a steady state simulation... could it be the solution is actually not steady?

John S. August 14, 2008 09:23

Re: Turbulence model convergence problem
1e-5 is decent for residual convergence. I would check your timescale. If your timescale is too small convergence can take a long time to achieve. In general, the Reynolds Stress models should be run with a lower timescale.

Andrew August 14, 2008 11:51

Re: Turbulence model convergence problem
my residuals were all stuck fluctuating at only 10^-3. I am using SST models with turbulence buoyancy on. I don't have any idea what time scale factor or what time scale control to use.

Ricardo August 14, 2008 15:10

Re: Turbulence model convergence problem
One criterion is the Courant Number, even though you get convergence, with high courrant number, is better the value around 1, the MAx value, what results in a RMS courant number samller arounf o(10^-1). RElated to SST model, consider the y+ value, it is below 2 as recomended? And has the fact that your case don't have a steady state solution, and you 'll see oscilations of residue with a aplitude and frequence nearly constant, after some interations. This is a indication, not a certain, that your problem has a periodic behavior.

Andrew August 14, 2008 18:25

Re: Turbulence model convergence problem
Yes, my y+ value is below 2. I can't get convergence while i turned on buoyancy turbulence option in cfx. I turned it off, the the solution was able to get up to 10^4 RMS and the results seems fine. I am also not sure whether to turn on turbulence buoyancy. Also, i am not sure of using the full buoyancy model with varying density or the bousinnesq model assuming constant density. Any ideas?

Ricardo August 14, 2008 19:45

Re: Turbulence model convergence problem
generally, the buoyancy option for turbulence overprecict the eddy viscosity and kill the dynamic behavior. Por exmplae, if you have a bubble colum, the plume with buoyancy can be well-behaved, and when you turn off, the dynamic and instabilities of the flow arise. The model is not so good yet in my opnion. but depends on the case. Related to density, depends on the case again. Variable density is hard to converge, requiring small time steps... generaly, and boussineq model, well, is there the thermal gradient strong? If not, boussinesq is a good approach, if is strong or very steep,no! But As you problme is in steady state, begin with contant density and turbulence on bouyance tirned off. Oce converged, use this solution as initial guess to a density variable simulation and/or boussinesq. It is a good a approach to complicated problmes, a initial simulation more simple and step-by-step, effort the model with more features until you reach the desired solution. I have a multiphase problem that operation takes 1.5 hours! I begun with a transient simulation durig 35seconds and I've used the last time step as initial guess to a steady state with a pseudo elapsed time aroud 2 hours - the time wich I wated! Try similarsome thing . from simple to complex models until you have the good slution. Good luck!

Glenn Horrocks August 17, 2008 19:35

Re: Turbulence model convergence problem
Have you looked at this page?

Glenn Horrocks

All times are GMT -4. The time now is 17:44.