velocity profil in wake doesn't match experiments
 

Hello!

I'm trying to do a simulation of a mixer and it's wake for a validation with experiments. According to the experiment the velocity profile should look like concentric circles. Unfortunately it looks quite similar to the blade as seen below.

http://www.abload.de/img/velprofiletransientjrk01.jpg

Does anybody have an idea how to fix this and match the results to the experiments?

I did a steady state simulation with SST model. Then I tried to make a transient run but the velocity profile didn't change...

Thanks in advance!

aside from needing more info; are you measuring the profile at the same distance from the mixer as the experiment? And, I have done this before, did you make sure the Fluent has the correct fluid?

Thanks for your reply!

It is the same distance in both cases (3 diameters behind the mixer). And yes, it's pure water in both cases.

Anything else you need? Time Step info or setup of experiment?

as much as I hate to suggest this, but I would take the mixer out and run the experiment and get a velocity profile and then do the same in Fluent..see if the mixer is the problem. I know that will be a headache. I have always had pretty good results when comparing experimental fluid flow with fluent's output.. I would be leaning on a set-up/variable definition in fluent somewhere.. Have you done a grid-study - increase the number of cells in Fluent and see what happens?

hmm... I haven't done the grid-study but it is already pretty fine (max y+ < 3)... the velocity profile in the experiment matches the analytical solution - so that's not the problem imo...

the simulation doesn't seem to capture a full rotation of the flow in the wake... would a bigger timestep make a difference?

I still think a grid study would be nice. It is easy to do, and I would be surprised if one of your committee members doesn't ask you about it. The easiest way to do it is cut the number of cells in half and see how that affects your answer.

You can always try a different time step. I am assuming that the solution has converged. Have you tried setting your convergence criteria tighter? That would mean more iterations tho

Not converged?
 

I suspect that the result has not fully converged. To get good convergence for wake regions, the convergence criteria must be tightened up considerably. The grid requirement is also quite tight and you must be careful to avoid too much diffusion in the solution.

OK, I gather your "container" diameter is about .6 meters and the cutting plane is 3 meters so the L/D is 5.

What is your "container?" Is it an axisymmetric jar or pot with the mixing blades at the bottom? Or is it a tube/pipe? If it is a tube, what is the axial velocity of the fluid at the input? Also, how fast (rpm) are your blades spinning?

From what I see, I gather that your flow is in a pipe and that the velocity that you have plotted may be total velocity, sqrt(u*u+v*v+w*w). Therefore the axial velocity is around .2582 m/s. So, what velocity components are you comparing with to experiment?

Please, more detail.

The mixer is 150 mm in diameter and rotating at 465 rpm.
The domain is tube-shaped, 10 diameters long and 4 diameters wide with the mixer placed in center.
The flow is axial along the tube with an inlet velocity of 0.3m/s.
Measured and plotted are the axial velocity components.

Oh, and when I said l/d is 3 before, I was referring to the mixers diameter. So the cutting plane you see is 450 mm behind the mixer.

The max residuals for this calculation where 10e-3. If not, one or orders of magnitude smaller. So it already converged quite a bit...
I can check it again on Monday.

Are you sure your boundary conditions are correct? Your blade is physically moving, correct? In other words you have imposed a angular velocity condition on the blade and you are moving the blade. Or, you could just impose angular velocity on your outer tube and input flow, assuming it is axisymmetric.

Or, are you solving this quasi-steady? Your plot looks like a quasi-steady state result. And, some of your comments have made this sound like a steady state problem. For example, you mentioned steady state earlier and you mention converging residual. The residual will only converge for steady state runs, quasi runs, or fixed time step iterations, such as sub iterations. Maybe a helpful plot would be a 3D iso-surface of vorticity (or any other variable of your choice) to make sure that the blade tip vortexes are corkscrewing as expected. Of course the vortex will eventually burst, but, to get that accurately your grid might need to be much finer. Also, how well you do will depend on the type of solver for the areas away from the blade, Euler, NS, LES, or URANS.