Hi all,

Is it theoretically correct to alter the turbulece parameters in order to get better resemlence to experemental data and if so what guidelines are there for doing this.

I doent seem to get results close enough to real values.

/Erik

[olwi]

No, it's not correct! Of course, it depends on your application, you may of course stumble on a very special case for which your model is not appropriate. If you are still talking about a simple channel flow, however, then it's a very bad idea to change constants. Standard turbulence ARE tuned to do that very well. It may give you a "better" result, but for the wrong reason. Two wrongs do not make one right. You should try to focus on what is so special with your case, if there are things you didn't think of:
- Compressible?
- Do you have buoyancy effects?
- Is it single phase?
- Chemical reactions/combustion?
- Swirling flows?
- Permeable walls?
- Wall jets?
- Is it really steady-state? If you think it is, do you get the same solution at the end of a transient run?
- Are you using symmetries where you shouldn't?
- Are your measurements really that good? Do they agree with other results in the literature?
- Are your inlet and outlet conditions placed in a good position upstream/downstream? Are they placed well in the experiment?!

If you want to read more about how the model constants are chosen you could check "Turbulence modeling for CFD" by David C Wilcox, or "Turbulent Flows" by Stephen Pope. Wilcox is good for this question, I think.

/Ola


/Ola

Thank you Ola for the thorough answer. It got me thinking of some new things i shall try. I am modelling a air flow with a incompressible because im hopeful to use the model for water flow aswell. will also try a compressible solver

Thanks again
/Erik

[olwi]

Hi again,

Air will indeed be incompressible, as long as you have a Mach number of less than 0.2-0.3. That means it's ok for speeds below 100 m/s, if you're at ambient pressure. Feel free to tell us a bit more about your case, and about the expriment you compare with. Maybe that will give us some ideas...

/Ola

Thanks a lot!

This is the geometry of the case that I run my simulations on:



It is 1/12 of the circumference of a rod. the three other sides are symmetry sides. I have used wallFunctions for the rod boundary, inlet for the inlet and inletOutlet for the outlet. The channel is 11.5 meter long and has a inside air flow velocity of 20.57 m/s. The Re number is 64590.
I count the air flow as incompressible and steady state.


this attached papper explains the experiment that I attempt to reconstruct:



As the papper mentions the next step is to heat the flow through the wall with the following implemented code:
solve
(
fvm::div(phi, T)
- fvm::laplacian(alphaEff, T)
- S*q/(rho*Cp)
- (mu*(gradU + gradU.T()) && gradU)/(rho*Cp)
- turbulence().epsilon()()/Cp
);

where S is the cell wallarea divided by the cell volume. q is the applied heat/area. The last two term hopefully models vilcous dissipation heating.

I hope I havent shoveled too much uninteresting information on you.

Thankful
/Erik

the papper was to big to send. I will also lay up my graph results once iv'e finnished my last turbmodel.

The flow is oriented thou wards the screen in the colored picture above.

here are my graph results:

sampling on the narrow symmetry side to the left

sampling on the wide symmetry side to the right


where the red crosses are from the experement and the other signs are from the models

Should I expect better coherence?

[olwi]

I would expect velocity curves to be a lot more smooth... You sure it's properly converged? I wouldn't be surprised to see a difference between the lines, but their shape should be even and smooth, as in the experiment.

Do you model the whole 11.5 meters of the rod bundle? Do you make the grid using blockMesh? You can make cells very long in the streamwise direction (200-300 mm, maybe even longer), that would save speed and improve convergence. Feel free to send the blockMeshDict to my private mail address, maybe I can have a look: olwix@yahoo.com

/Ola