Hi all,
I'm simulating a fan located in a free space (just like a fan used in house), model description:
1. fan with 8 blades, I just mesh one blade and define periodic interface.
2. the interface between fan and ambient is "Frozen Rotor"
3. rotating at 2000 RPM, fan diameter 12 inch
4. no initial speed defined . The outside ambient 6 sides are all "opening".
5. Advection scheme=upwind, Turbelence model=SST
but the run time convergence curve will become almost horizontal at 1e-2 at iteration 50 , and will last the same situation to over iteration 300. Is it normal for fan simulation?Does it need thousands iterations to converge?
<font color="Blue">
Note: As I stop it at iteration 310 and view the result, the flow stream line and vector looks resonable</font>

Thanks.

Hi,

What version of CFX are you using? CFX4 will need a lot of iterations to converge for complicated simulations, but CFX5 should be converged for most steady state models by 100 iterations.

If it is not converging there is probably a problem with your model setup. This is especially so with upwind differencing, this should converge very quickly.

Glenn Horrocks

Hi Nepal/Glen, I think I am having a similar situation. I am modeling disc with ducts (thermal energy+flow) inside (0.436 m outer dia and 30 ducts, duct radii from 0.116 to 0.218 m) and I have used periodic interfaces to simulate only two ducts and also openings at the boundaries. Angular vel is 450 rpm. RMS sets in at 1e-3 while peak RMS climb up to 8e-2 in the radial direction!. The convergence curve for RMS (average) fluctuates between 9e-4 to 2e-3, almost sinousoidal as of the iteration number 60. I have obtained 'initial' solutions using laminar model and then I have used that solution for k-e but that doesn't change the value of RMS. I also did mesh refinement in some regions. I am using CFX5.

C.

Hi, Glenn Horrocks, Thanks for your response! I use CFX5.7.1 . I'm wondering what wrong it is for my model. I've tested many combinations of settings to test it, but still NO-GO! And I'll still work on it...

Hi,

You will find that using laminar flow for a turbulent simulation will usually make the convergence worse, not better. Therefore using a laminar simulation as an initial condition for a turbulent simulation is not a good approach.

A better approach to get an initial condition is to use upwind differencing and/or the zero equation turbulence model.

Glenn Horrocks

Thanks for the advise Glenn. I have been looking for information in convergence of rotating fluids. I found one reference speaking of a convergence level of 2.5e-5 in RMS (rotating fluids) and 1e-4 in multiphase rotating flows (Tucker, Numerical precision and Diss errors in rotating flows, Intl J for Numerical Methods for Heat&Fluid Flow)... however I can not get such convergence levels. Any other idea is welcomed.

what you said is very much correct but i wish to add a little bit, when you have a laminar intial guess and if you enable k-e or k-w model and run the calculations this usually creates problems than helping in convergence,

but

to use laminar solution as initial guess a better approach (which i have tested so many times) is to switch off the calculations for flow variables while you do calculations for k and epsilon, for starting few iterations , till you see the residuals for k and e have fallen sufficiently , then one can enable whole solution and go on iterating. this way one can use the laminar case as initial guess.

Hi,

If the simulation is turbulent then you will often have difficulties in getting a laminar model to converge. Thus I do not recommend this as a good method of getting an initial guess for a turbulent model.

A far better way of getting an initial guess for a turbulent simulation is to use the zero equation turbulence model. It is very stable and easy to converge, and is more physically realistic than using a laminar model.

Regards, Glenn Horrocks