CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   y+ problem (http://www.cfd-online.com/Forums/cfx/24363-y-problem.html)

 Gui August 5, 2007 23:43

y+ problem

I've read over and over about y+ including this forum and I still don't know how to interpret this value.

First how do you check your y+ value in post ? Do you need to create some contours or else ? I just clicked on the Default Domain and selectioned the variable y+. The max value is approximetaly 1000. How can I interpret that value? I'm using the k-w SST turbulence model in a steady incompressible and isothermal flow. I created 10 inflation layer with first layer thickness 0.3 mm.

 hammam August 6, 2007 09:11

Re: y+ problem

check that

http://geolab.larc.nasa.gov/APPS/YPlus/

 Gui August 6, 2007 09:54

Re: y+ problem

ok but how do you know what y+ value you should have ?

 hammam August 6, 2007 10:24

Re: y+ problem

hello

eash turbulence model amd wall function has a certain limit for the y+ cfx manual and the posts here have comprehensive information for that matter , just search here with ylus keyword

regards

 Gui August 6, 2007 12:18

Re: y+ problem

ok I've been reading through all the archives but I still have some questions.

I am simulating a incompressible and steady state High Reynolds number flow (10^7) with the k-w SST model with Automatic Wall treatment.

The max y+ value observed is 1000 (though I'm not sure if I checked the value at the good location, I just displayed the variable y+ in the default domain). I read that SST requires y+<1 in order to integrate the sub layer (no wall functions). So if my value are that high (though I don't understand because I created 10 inflation layers with first prism height 0.3mm and expansion factor 1.2 wich is practicaly the finest resolution I can get) then my nodes are not in the sublayer. Then with Automatic treatment,SST reduces to k-epsilon and uses wall functions. So it means I'm not resolving the viscous sub-layer but only the log-layer. Is that correct ?

If I'm not resolving the viscous sub-layer, and if this effect is important for my problem (I'm studying the flow that goes out of a 0.15 m side square tube through windows located on each lateral face of the tube, and what is important is the flow (mass flow rate and direction) out of these windows, wich I need to impose inlet conditions on another geometry), then my results are not that good.

 Glenn Horrocks August 6, 2007 18:29

Re: y+ problem

Hi,

Not resolving sub layer with y+=1000 - Correct.

What y+ should you use? - Problem dependant. Do a mesh sensitivity check. I suspect you will need a finer wall resolution than that for good boundary layer resolution.

Glenn Horrocks

 Gui August 6, 2007 20:05

Re: y+ problem

Thanks for the answer. My question then becomes : for a Reynolds number of 10^7, can the linear sub layer have a significant effect on the flow ? And so if you're not resolving you get wrong solutions ?

More precisely, if y+=1000 then the k-w SST model reduces to k-epsilon and uses wall functions, does it lead to an overprediction of the turbulence ? And that would be the reason why I can't get stable values of my interest quantities after having refined the mesh 5 times (from 200000 to 10000000 elements).

I'm really looking forward to an answer because I'm thinking maybe this is exactly the problem (because I already did a mesh study)

Thank you

 Glenn Horrocks August 7, 2007 18:27

Re: y+ problem

Hi,

Can sub layer have significant effect? - Yes, it is the first layer of a boundary profile. But do you need to model it (ie y+=1 approx) or can you use wall functions to model from y+>11? That is problem dependant. Most flows at Re=10^7 should be OK with the wall function approach.

Wall functions do not over predict turbulence for flow over a flat plate, which is what they have been "tuned" for. They do, however have problems in separating flows, stagnation regions etc etc and that can lead to under/over prediction of turbulence and/or gross flow errors, such as not predicting a separation.

Your mesh refinement study is hopefully showing a trend towards a value. If so you can use Richardson extraplotion to get a "zero sized mesh" solution. If will also give you an error estimate.

What are you modelling?

Glenn Horrocks

 Gui August 7, 2007 20:02

Re: y+ problem

I'm modelling the flow in a control rod guide tube. Basically the flow enters at the base of a square tube wich contains very dense structures, and then exits the tube through windows that are located on each lateral face of the tube. My mesh study doesn't seem to converge to a value. What really interest me are the mass flow through the windows, because I need it in order to impose an inlet on another model. I did 4 uniforms meshes (from 200000 to 800000 elements) and the solution seemed to converge faster and faster as the elements were increased, and the mass flows through the windows seemed to converge to a value. But then I used a finer mesh (of 1 million elements) and then the solution totally changed, the convergence was slower, and the mass flows really different. With even more elements, it's still different, and in addition when I check the value of mass flows at different iterations it's still fluctuating a lot

 Gui August 7, 2007 20:05

Re: y+ problem

but if I'm not resolving the sublayer then my layer is only turbulent and that's why I suggested that turbulence is overpredicted, wich could explain my fluctuating values...

 Glenn Horrocks August 8, 2007 07:00

Re: y+ problem

Hi,

I assume your simulation is steady state. Did the fine mesh simulations have problems converging? It is common for a coarse mesh to successfully converge to steady state and a fine mesh to not converge as the finer mesh can resolve some transient wake shedding and other effects which the coarse mesh did not resolve.

This could be a cause for a variation.

Glenn Horrocks

 Gui August 8, 2007 08:32

Re: y+ problem

yes that's exactly what happens. My simulation is indeed steady state and the finer meshes converge really slower and to a lower value of residual (between 1 and 2 order) than the coarse mash. But why is that exactly ? You say that with the finer meshes, more terms of the transient are resolved?

 Glenn Horrocks August 8, 2007 18:40

Re: y+ problem

Hi,

Coarser meshes have more numerical dissipation than fine meshes, so coarser meshes can damp out flow instabilities. Also if the mesh is not fine enough to physically resolve an instability it generally will occur. Hence fine meshes can go transient when a coarse mesh solution was steady state.

Whether the transient is significant is significant is problem dependant - in your case it sounds like it is. If possible I would run a transient simulation of your fine case over a weekend (or maybe over a week's holiday!) and let it establish itself.

Glenn Horrocks

 Ganguse August 30, 2007 15:08

Re: y+ problem

Hi, I have the same problem as you, in finer mesh occurs instabilities nor present in coarser mash, can this avoided manipulating sw parameters?

for Gui, I think u have y+=1000 because u calculated it in the whole domain, it has to be calculated only on wall boundaries

 All times are GMT -4. The time now is 05:01.