[thatdiverguy]

What turbulence model would you use for the following simulations with justification:
a. Low-speed airflow past an airfoil;
b. Wind flow past a mountain or large bluff-body;
c. Convective heat transfer in magma.
d. gas flow inside an internal combustion chamber;
e. viscous flow inside a mixer
f. wind flow and pollutant dispersion inside an urban street canyon;

[t.teschner]

since you ask about models, i presume direct numerical simulation (DNS) is not an option (which of course would be the trivial answer and would work, in theory, for all cases but probably not for practical applications)

large eddy simulation (LES) would be my next best guess but might not be really what you are looking for (in terms of models).

in general, the spalart-allmaras turbulence model was derived for aerospace applications (airfoils) but is does not perform very well for 3D flows and flows with high shear stresses and separation but is very fast (1 equation model) and is very robust. i have not once discovered a case where i could not get the S-A model to converge, though you may get better result using a more sophisticated model.

in the family of the two equation models you have, most commonly used, the k-epsilon and k-omega model. k-epsilon is said to perform best away from walls (and their modification like the RNG and the realizable version) improve the prediction for flows with a shear, swirl, vortices and mild separation (separation is a very tricky thing to get right)
the k-omega model on the other hand is said to perform best near the wall i.e. flows with boundary layers (even for adverse pressure gradients) however separation is also poorly predicted. (excessive and early)

moving on to more sophisticated models you get for example the SST k-omega which is actually a combination of the k-epsilon and k-omega model and it uses a blending function to determine if the calculation of the flow property should be "more" based on one or the other model (in this way you get the best of both world, i.e the good performance of the k-omega model near the wall and the good behavior of the k-epsilon away from the wall)

the previous models are all based on the eddy viscosity and you use the approach to close the equation (see: closure problem of the RANS equations). you can go a conceptual different way to close the equations using the reynolds stresses. using this approach, the most popular model you'll get is the Reynolds stress model (RSM) which solves one further transport equation for each individual stress. fortunately the stress tensor is symmetric so you "only" have to solve 6 equation to obtain the stresses. It is tricky to get it to converge, the equations are coupled to some degree and is not very robust. if you have to use it, it is usually a common practice to start with a robust turbulence model (say the spalart-allmaras for example) and use the result you obtain as an initial guess for the RSM. you may even use another, intermediate model to further improve on your initial guess for the RSM if you are still having difficulties to get convergence.
Due to its complexity it is said to be good for complex, highly swirling flows but my personal opinion is, that you probably get a better result with just slightly more computing time if you switch to a large eddy simulation.

speaking of which, there are also detached eddy simulations, where you model the near field with RANS and the far field with LES but that has other difficulties, for example to decide where to switch from one model to the other.

so based on that, you hopefully get an idea which model to use, i am not an expert for the topics you mentioned, i have some knowledge in some of the fields but i am not an expert in these and other users may give you a more experienced based answer.

my selection would be:

a. spalart-allmaras (aerospace application, S-A was developed for that sort of simulations)
b. k-epsilon (the flow away from the wall is of interest)
c. - e. either SST k-omega or RSM, for e. you may get away with a k-epsilon model
f. k-epsilon

a note at the end. if you need a time history of the flow, i.e unsteady flows then your best candidate is probably LES. RANS equations have been obtained by time averaging the Navier-Stokes equations and although there are unsteady RANS (URANS) equations, as I have been mentioning before LES does not necessarily need to be a lot more computational expensive and if you can allow the extra CPU time you will probably get away with better results.

[harishg]

http://www.metacomptech.com/index.ph...cfd/turbulence