[Diger]

Hello,
When modeling a flowtube with low reynoldnumber (1600) I guess the advise is to use a laminar scheme.
However if I use a turbulent scheme, is the turbulent kinetic energy a good approximation for the representation of turbulence or is it usually overpredicted?
For example I get with a k-omega-SST scheme between 5 and 13 % in terms of tubulent intensity ( sqrt(2/3 k)/|v| ) - 5 in the middle where v is large and 13 when going to the wall of the cylinder where v goes to zero - velocity profile should be sth like 2(1-(r/r0)^2).
To me personally this seems rather large. What do you guys think?

[LuckyTran]

Turbulence intensity increases with decreasing Reynolds numbers.

How is there turbulence in a laminar flow? If you use a turbulence model then obviously you'll overpredict all the turbulence because the real turbulence is zero. k is 0 everywhere in a laminar flow, where k>0 for turbulent flows (k can't be negative).

The problem seems rhetorical.

[Diger]

No the issue I have is the following:
First of all exactly zero turbulence is something rather theoretical which means that also in a "real" laminar flow I'd expect some order magnitude which I could get from the turbulence model. On the other hand if a turbulence model predicts relatively large turbulence how is it ensured that also within real turbulence there is no overprediction?

[LuckyTran]

I still don't get what you are trying to do exactly.

Originally Posted by Diger

On the other hand if a turbulence model predicts relatively large turbulence how is it ensured that also within real turbulence there is no overprediction?

The only hard limiter is turbulent viscosity. There are some flux limiters as well that are dependent on the discretization scheme and vary wildly. The problem is how do you know it's an overprediction? Equivalently, how do you know that pressure,temperature are not too high or too low for your problem?

Originally Posted by Diger

No the issue I have is the following:
First of all exactly zero turbulence is something rather theoretical which means that also in a "real" laminar flow I'd expect some order magnitude which I could get from the turbulence model. On the other hand if a turbulence model predicts relatively large turbulence how is it ensured that also within real turbulence there is no overprediction?

Turbulence is broadband. Laminar flows strongly damp most frequencies so this broadband behavior does not occur. Hence, there's no real turbulence in a laminar flow either. Hence why I argue (in the weak sense) that any prediction of k in a laminar flow is an overprediction, via the Boussinesq hypothesis which is invoked by RANS models (this k inevitably is synonymous with turbulent Reynolds stresses).

Now if you want to study these fluctuation velocities that are permitted in laminar flows (however small they may be), then you can simply run an unsteady laminar model. However, I don't understand what you hope to obtain by imposing a turbulence model onto what should be a laminar flow.

RANS assumes separation of time-scales which severely limits what you can (or should) do with the turbulence model.

[flotus1]

Originally Posted by Diger

No the issue I have is the following:
First of all exactly zero turbulence is something rather theoretical which means that also in a "real" laminar flow I'd expect some order magnitude which I could get from the turbulence model. On the other hand if a turbulence model predicts relatively large turbulence how is it ensured that also within real turbulence there is no overprediction?

Laminar flow is not something hypothetical. It exists in real life.
A turbulence model predicting turbulence quantities >0 in a laminar flow is a modeling/user error, not a simulation result.

[deenriqu]

Is it not option to run fluent's "Laminar" model for you flow. This is essentially pseudo-DNS solution of the NSE. For a fully laminar flow, this should provide a decent solution.

[Diger]

Thanks for your replies. I was just checking from the very beginning that my flow (in this case just the flowtube) is very well modeled.
In this case I found the Reynold-Stress / Stress Omega to be reasonably well to the laminar case (as long as I just use a standard mesh without any wall refinements / prism layers).
Why don't I need wall functions for the stress-omega anyway? In the manual I only found that the pressure strain has not wall reflection term. Or does it mean that instead of an equation for epsilon I have one for omega? Like the switch from k-epsilon to k-omega?!?
Is a sensible wall resolution not necessary? How can I extract the y+ from my simulation? In the manual again it says I have to look for the maximum of the turbulent viscosity, but there is no maximum for me?
In many other models wall functions seem to be independent of mesh refinement at the walls (apart from standard wall functions).

Now if at the end of my flowtube there is a nozzle (as an outlet with specified massflow) and a volume where the overflow goes (the geometry is more complicated since at the side there will be also a pump which is probably modelled as pressure outlet because I only know the mass flow into the system) then I might (does not have to be) expect some turbulence. However I do not want overpredicted turbulence/turbulent kinetic energy from the flowtube where it should be laminar to be transported into the region where I might then also overpredict turbulence. I hope that is more clear.