
[Sponsors] 
October 9, 2000, 07:47 
realizable k,emodel

#1 
Guest
Posts: n/a

Hi collegues, what experiences do you have with the realizable k,emodel? I wonder if someone compared its results with those of the RNG model.


October 9, 2000, 11:27 
Re: realizable k,emodel

#2 
Guest
Posts: n/a

Hi, I just finished a case with the ke realizable. The case is a strong swirl flow. Comparison between RNG and realizable show that there are no obvious differences.


October 9, 2000, 12:01 
Swirl factor

#3 
Guest
Posts: n/a

Hi I'm using RNG ke model with swirl factor (default value) for a swirling flow. My question is how to make a good estimation for the swirl factor. Thank you Phil


October 9, 2000, 13:06 
Re: Swirl factor

#4 
Guest
Posts: n/a

Could you detail about swirl factor? I choose RNG ke model with swirl dominated flow. However, the results are different from experimental those. By the way, I have try several other models. I find no good model to compare with experimental data. Do you have good idea?


October 9, 2000, 14:56 
Re: Swirl factor

#5 
Guest
Posts: n/a

I use the RNG Model with the default swirl factor. I use the Two layer zonal model but I accept that my grid at the wall adjacent zones is in some parts a little bit to coarse (y+ in the order of 5). I am using a unstructered 3 dim tethyb grid that I have generated in Gambit. I think for these swirling flows, in my geometry the swirl is induced by the geometry, need a fine mesh also in the core flow, because the flow fluctuations are very high hence the flow quantities change in these regions as well. My results are in good agreement with the experimental results when I simulate the isothermal flow. Unfortunately the errors are higher when I solve the nonisothermal case with heat transfer between the walls and the fluid. However the right trend is represented by my simulation, so if I compare different geometries of this heat exchanger the comparison between the results are right. So if you want to discuss some details you can also write an email
Regards Phil 

October 10, 2000, 04:06 
Re: realizable k,emodel

#6 
Guest
Posts: n/a

I've investigated the flows in continuous casting tundishes with the ke, RLZke and the RNGke model. I found that there are only small differences between the results of the ke and RLZke model, whereas the RNGke model estimates the turbulence of the flow field in a quite different way.


October 10, 2000, 04:10 
Re: realizable k,emodel

#7 
Guest
Posts: n/a

Interesting. Which calculation is closest to experiments?


October 10, 2000, 04:26 
Re: realizable k,emodel

#8 
Guest
Posts: n/a

Unfortunately, it isn't possible to state that one models fits best the turbulence in ALL flow situations. However i found, that the RNGkee model estimates the turbulence in complex flows better than the other models.


October 10, 2000, 04:31 
Re: realizable k,emodel

#9 
Guest
Posts: n/a

Of course not in all situations. Which was the best in the tundish?


October 10, 2000, 04:35 
Re: realizable k,emodel

#10 
Guest
Posts: n/a

RNGke (in tundishs with flow modifying devices) and ke (in tundishs without such devices).


October 10, 2000, 08:51 
Re: realizable k,emodel

#11 
Guest
Posts: n/a

Have you published something about the different results using kepsilon and RNGkepsilon models in complex flows, especially in swirling flows. I have also had good results with RNGkepsilon. Regards Phil


October 10, 2000, 09:08 
Re: realizable k,emodel

#12 
Guest
Posts: n/a

I'm working on a publication.


October 10, 2000, 14:04 
Re: Swirl factor

#13 
Guest
Posts: n/a

I also use RNG model. At most cases I use standard wall. Due to a big difference I try to use Two Layer to compare results. However, there is still a big errors. Because of memory problem it is impossible to lead a mesh to less than y+ of 5.
However, in non swirl flow, I test both RNGTwo Layer and standard model. There is a little difference. I do not know whether the swirl flow leads a bigger difference compared with experimental data or the model leads errors regards 

October 11, 2000, 01:33 
Re: realizable k,emodel

#14 
Guest
Posts: n/a

I recently ran some cases comparing various turbulence models for a NACA0012 airfoil, at 4 deg angle of attack. Using standard wall functions, the realizable model compared better with experimental results than the RNG, for both lift and drag. (RSM was better again).


October 12, 2000, 11:04 
Re: Swirl factor

#15 
Guest
Posts: n/a

Hi those who are interested in swirling flows
Although the RNG model tends to improve the results for weakly swirling flows, there's alimit to what it can do, since the RNG model is after all an isotropic eddy viscisoity based model and as such is less than adequate for swirling flows. If you have to deal with swirling flows most of the time, I would consider using the RSTM. 

October 12, 2000, 11:08 
Re: realizable k,emodel

#16 
Guest
Posts: n/a

I though I could contribute my two cents.
In general, it's fair to say that the RKE and the RNG model give comparable results. The RNG model has been found to "often" perform better than the RKE for swirling flows. The RKE model works better than the RNG for jets and impinging flows (airfoil or hydrofoil at incidence). I prefer the RKE. 

October 12, 2000, 13:11 
Re: Swirl factor

#17 
Guest
Posts: n/a

You are right, but the problem with RSM is that you need to solve 7 additional equations which might be not a problem if you really have good computational ressources, but for me it is only possible with isotropic eddy viscosity models. So I have to live with less accurate results! Anyway thank you very much for your help
Phil 

October 12, 2000, 14:07 
Re: Swirl factor

#18 
Guest
Posts: n/a

It's totally up to you to decide which turbulnece model to use. It depends on the purpose of you computation, computational resource, time constraints, etc. And that's why we offer a suite of turbulence models in FLUENT.
However, The RSTM is not as expensive as many people think. They think, as you did, in terms of the additional number of equations. In terms of memory requirement in FLUENT, the RSM requires only 20% more (a whole lot is already used for metrics, other solution variables, etc.). In terms of the CPU per iteration, it requires 60 % more at most. My point is that the RSTM is not "prohibitively" expensive as often blindedly quoted in the literature. 

October 13, 2000, 06:28 
I'l try to do it!!

#19 
Guest
Posts: n/a

Thank you for your advice. But should I start the simulation with my results I got with ke model?
Regards Phil 

October 13, 2000, 09:38 
Re: I'l try to do it!!

#20 
Guest
Posts: n/a

It's one good way of getting the RSTM sooution started. But you may start the RSTM solution from the start with less agressive underrelxation factors (e.g. 0.5 for turbulent quantities). The optimum solution strategy (solver settings, etc.) will depend upon the flow under consideration. But once you start trying the RSTM in FLUENT, you will be able to build your knowhow very quickly. One more thing I'd like to remind you is that you should choose highorder discretization (2ndorder at least) to have full benefits of using secondmoment turbulence closure. Firstorder upwind, for instance, can easily plague your results, no matter how good the turbulence model is. Good luck!


Thread Tools  
Display Modes  


Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Superlinear speedup in OpenFOAM 13  msrinath80  OpenFOAM Running, Solving & CFD  18  March 3, 2015 06:36 
about Subgridscale model  impecca  OpenFOAM Running, Solving & CFD  4  December 20, 2013 11:36 
Low Reynolds kepsilon model  YJZ  ANSYS  1  August 20, 2010 13:57 
LES and combustion model  Margherita Cadorin  CFX  0  October 29, 2008 06:24 
Advanced Turbulence Modeling in Fluent, Realizable kepsilon Model  Jonas Larsson  FLUENT  5  March 13, 2000 04:27 