[agungwp]

hi guys,
as we know that there are any several turbulence models to simulate physical problems. it couldn't use one turbulence model to simulate all of the problems. i hope you are could explain to me the differences between the turbulence models. such as k-epsilon standard and rng, k-omega standard and sst, spallart-allmaras.
thank you very much

[bramv101]

Uhm. This is basic CFD. get yourself a good handbook or usermanual for these type of things... Such questions are not likely to get answered on a forum

[teguhtf]

you can find it in CFD book!!!
it can be downloaded free CFD book on this site
http://avaxhome.ws/ebooks/science_bo...070016852.html
http://avaxhome.ws/ebooks/science_bo...540318003.html

[Jade M]

Good question. I have been wondering this myself.

SST is a good default choice

Sparlart-Allmaras: Eeconomical for large meshes. Performs poorly for 3D flows, free shear flows, flows with strong separation. Suitable for mildly complex (quasi-2D) external/internal flows and boundary layer flows under pressure gradient (e.g. airfoils, wings, airplane fuselages, missiles, ship hulls).

Standard k-epsilon : Robust. Widely used despite the known limitations of the model. Performs poorly for complex flows involving severe pressure gradient, separation, strong streamline curvature. Suitable for initial iterations, initial screening of alternative designs, and parametric studies.

RNG k-epsilon :

Suitable for complex shear flows involving rapid strain, moderate swirl, vortices, and locally transitional flows (e.g. boundary layer separation, massive separation, and vortex shedding behind bluff bodies, stall in wide-angle diffusers, room ventilation).

Realizable k-epsilon : Offers largely the same benefits and has similar applications as RNG. Possibly more accurate and easier to converge than RNG.

Standard k-omega : Superior performance for wall-bounded boundary layer, free shear, and low Reynolds number flows. Suitable for complex boundary layer flows under adverse pressure gradient and separation (external aerodynamics and turbomachinery). Can be used for transitional flows (though tends to predict early transition). Separation is typically predicted to be excessive and early.

SST k-omega : Offers similar benefits as standard k–ω. Dependency on wall distance makes this less suitable for free shear flows.

Reynolds Stress : Physically the most sound RANS model. Avoids isotropic eddy viscosity assumption. More CPU time and memory required. Tougher to converge due to close coupling of equations. Suitable for complex 3D flows with strong streamline curvature, strong swirl/rotation (e.g. curved duct, rotating flow passages, swirl combustors with very large inlet swirl, cyclones).

[LucasGasparino3110]

Well, that's a HUGE question... Honestly, I'd recommend Wilcox's book for a good description, and Fluent's user manual is pretty detailed as well. Still, here's a quick summary:


Most of these models are based on the Boussinesq assumption that turbulent stresses act as an additional dissipation term, which is isotropic. Essentialy, instead of having transport equations for the Reynolds stresses, which would result in at least 6 new equations, the usual models focus on creating empirical transport equations for scalars such as turbulence dissipation rate and kinetic energy. A few are actually algebraic, that means they don't rely on PDEs for the modelling. What happens then is that models with more terms are, in general, more capable of capturing complex behaviour such as secondary flows, reattachment of shear layers and boundary layer separation. For example, the standard k-e model stands for 2 transport equations: TKE (k) and turbulent energy dissipation (e). It's rather good enough for capturing free shear flows, but it's really poor near solid boundaries, and, for curved geometries, it has poor separation prediction. The SST model combines the simpler k-e with the k-o model through blending functions, the later model being better at dealing with near wall behaviour. The Spallart-Almaras model is a rather simplistic one equation model that focus on modelling airfoils at normal flight conditions. If stall is to be modelled, the SST model would be a better option.
In 3D cases, or cases with very strong swirling, the Boussinesq assumption fails, so the previous models are generally not recommended. In this cases, Reynolds Stress models such as the Omega-RSM are more adequate at capturing the inherent turbulence anisotropy. They are still RANS models tough, so they are quite limited. In practice, RSMs are VERY stiff, so much so that the SST model is the standard recommendation, unless you're studying cyclones or excessive vortex streching.
Finally, there's always the option of the scale-resolving models: those are based on length scale filtering, that allows large eddies to be resolved and small ones to be modeled. LES, DES and SAS models are focused on resolving flow structures accurately, so of course they are pretty ressource-intensive. They are inherently transient (although you CAN use SAS in a steady sim, I don't recommend it) and 3D (even though you can enable them in 2D, don't put too much trust in the answers in a quantitative sense), and results become very sensitive to small details, such as mesh structuredness, geometry representation and time-step size. For filtered models, the ONLY spatial discretisation that can be trusted is the Central Difference, as it introduces no extra dissipation. That means looking out for the Courant and Peclet numbers, in order to avoid spurious oscillations in the results. LES is for sure the most intensive model available, so I recommned reserving it for when you've got time to spare, a REALLY powerful computer and a case worth it (don't use it for first design evaluations for example, save it for the last one). DES and SAS are less demanding, as they actually blend LES and URANS approaches, with LES being triggered by the mesh size and turbulence parameters. So, make sure to mesh adequatedly the downstream region, in particular in the free stream shear layers.
To close these remarks, RANS and URANS models usually wall function options to better model boundary layers when the y+ value exceeds a certain criterion, and possess different sensitivity to this important parameter. In general, try to get smaller y+ possible/feasible when using more complex models, and definitely avoid Y+ > 300 for ANY case, if wall effects and results are important (Cd and wall heat transfer, for example). For filtered approaches, there is NO wall function, so you'll need this value to be at MOST 1; as well, you'll need at least 15 nodes inside the boundary layer regions to get a decent description, in particular of the viscous sublayer (this is why LES is so demanding for wall-bounded flows, even at low Re). Keep in mind that, for LES, it is a good idea to make more than one run with small differences between each, in general small variations to inlet conditions, so you can do some ensemble averaging and prove that your case is ergodic (statistically static for a certain time step).
I hope that helps... as everyone said, the subject is quite exhaustive, so if you really are into CFD, get those references mentioned and study, study, and study untill your brain burns out!
Some references:


Fluent Theory Guide
Turbulence Modelling for CFD (D. Wilcox)
Multiscale and Multiresolution approaches in Turbulence (P. Sagaut)
Basically, any book by P. Sagaut is great for LES references


Those are focus on modelling methods and techniques, but I'd suggest you to read a bit about turbulence itself, so the books by Yaglom and Monin are pretty much obligatory, as well as anything by Hunt and Vassilicos. Hinze and Kolmogorov are also mandatory, and P. Davies has an excellent introductory book in turbulence theory. Finally, books on chaos theory and dynamic systems evolution provide a nice and modern view on why turbulence is so complex.


Have fun reading and testing!!!

[ali-st]

Hi, In summary it is related to type of your problem and sometimes is kind of experimental choice! In my experience, if I don't know the most appropriate model for a problem I usually start with k-epsilon model. For more information about models see the User Fluent Guide or other sources that are available (There are a lot of books and videos. Just google it!).

[LucasGasparino3110]

Yeah, you're kinda on the right way...
Just keep in mind that all RANS models are quite deficient in some way, and most will be pretty innacurate if you have strong swirling, even the RSMs.
Be extra careful with the k-e model: it's all too common to use it and not perceive it's flaws. What I can suggest you is that you do a comparison study, maybe in 2D if you PC is not a monster... The wall mounted square is well documented, and at Re=25000 exhibits enough turbulence to be interesting. The only general rule is: avoid algebraic models if you can't calibrate the closure coefficients beforehand. They are brutally dependant on those. Finally, Google is fine enough to gather some ideas, but turbulence modelling is a pretty serious subject, so please check some of the literature. Stuff like production limiters, curvature correction and even coefficient values can be pretty important, and require deep understanding of the model to become useful. In the end though, Menters SST model takes care of most normal flows. If the Reynolds number is not too high, use the intermittency model, preferably in a transient sim. There's no correct answer, no universally better model: you have to find the one that fits your constraints of data quality and sim time. And don't forget to check your numerics: your discretisation method for time and space can have a huge influence in the end results, if you even get them that is. I'll take this little space here to complain that neither FLUENT nor CFX offer an RK4 time stepping scheme... I really miss having a 4th order scheme for time, at least for LES...


HAVE FUN!