Hi everyone,

I am modelling a flow in a penstock with a reynolds number of 2x10^7. I use k-epsilon model in Fluent.

First, I ran the case with a very fine mesh and had an error like : "turbulent viscosity ratio limited to ... in ... cells". Hence I used another model (rng and k-omega) but it did not solve the problem. Reading the Fluent documentation and messages on this site I went to the idea that the problem may come from the y+ values and the density of the mesh. I checked my y+ values and effectively I saw they were all under 30.

Then I have done a coarser mesh and the problem of turbulent viscosity disapeared. However the y+ values are still below 30. Once again I changed the turbulence model but no better results.

Besides, the velocity contours I obtain are all very different depending if I use the k-epsilon, the RNG or the k-omega model and also when I change the mesh size.

Any comments are welcome.

Regards,

Julie

You want low Y+ values. These will vary on what boundary conditions you are using. The velocities will be different for different turbulence models. Usually the best one depends on your application and also if you have experimental data to compare them to.

Hi Charles,

I thought y+ value should be between 30 and 50 to provide accurate results in the boundary layer.

As boundary conditions, I used a velocity inlet with a 3m/s velocity, 5% turbulent intensity (but this is just an asumption) and a hydraulic diameter of 4574mm. But why do you say boundary conditions can influence the y+ values?

Besides I have experimental data but when I compare the results they are totally different. The range of velocities is the same but the profile is very different.

Any idea?

Yes, Y + should be > 30 to 60. At Y+<11.225 Fluent employs the linear(laminar) law. So using a excessively fine mesh near the wall should be avoided because the wall functions cease to be valid in the viscous sublayer. You said your Y+ was below 30, correct? Fluid properties affect Y+ such as density, viscosity, and friction velocity. This is stated in the Fluent manual. Also, as Reynolds number increases the upper bound of Y+ increases. What wall functions were you using for each turbulence model?

I used the standard wall functions, i.e. the default ones.

be carefull with statements of y+

If you use standard k-eps, y+ must indeed be >30 and somewhat below 50.

But if you use k-omega, y+ values should be very low (around 3) to have good results.

Julie, you have a very high Re. Is your flow swirling? How big is the geometry?

Julie, you state the results depend on mesh size. That looks like an insufficient mesh-resolution...

kind regards,

Laika, still orbiting

Thank you for your help, but what can I do to obtain higher y+ values with the k-epsilon model?

How can I know if swirling. i do not have this info from the experiements.

Concerning the geometry, the pipeline is almost 43m long with a diameter of 3.92m.

regards

julie

[doodek]

Hello,
I have a question. Which should be y+ when i'm using k-omega SST model?
A few posts previously laika said that "around 3", but there:
http://www.cfd-online.com/Wiki/Near-...k-omega_models
is an information, that y+ should be around 30 (if i don't use transitional flows option). What is truth?

Best regards,
doodek

[Philipov]

Try to compute turbulent intensity... not guess! There is an easy formula for internal and external flows... use it.
Advice: use Wiki.. but read manual and consult providers!

[doodek]

I know only empirical correlation for pipe flows. Which are other formulas? In my case I have flows between a few rows of pipes, so I guess that I should set up value of turbulence intensity.

[Philipov]

7.2 Flow Inlet and Exit Boundary Conditions
Uniform Specification of Turbulence Quantities
Fluent UG.. the formula is described

[thecfduser]

U can use y+>30 without transitional flow or y+=1 with trans flow. U will get a near result if ure boundary layer is not subjected to severe adverse pressure gradients....
However it is better to use a y+=1 since the k-w is essentially a low Re....

With k-epsilon, using a y+>30 and law of the wallis usually better than y+=1 with enhanced treatement