code_saturne User's Forum

Hi all,

I am working on a simple geometry : a pipe of a 33.7mm diameter and 2m long. My mesh is made of tri/quad elements but with a high number of nodes (1502501 nodes!). The simulation is a simple Poiseuille flow with a velocity inlet (Reynolds number=60000), an outflow and the rest is a wall. In laminar regime everything is OK, but when I try to simulate a turbulent k-epsilon flow it doesn't work : or divergence error solving dissipation or I have a velocity profile which is not a turbulent one. Actually, I have tried different combinations of initialization and boundary conditions for turbulence. But it never works! I thought initializing the volume by values (turbener and dissip) would work. I did this with Fluent at home (I took turbener=0.01 and dissip=0.001) and it worked well. I am importing my meshes in Code Saturne in a neutral format (I do them with Gambit), has this anything to do with my problem? I don't think so, because the checking of the mesh by Saturne is OK.

So my question is, and I need your help :
1-What should I choose in the turbulence model panel k-epsilon advanced options : one scale model, two scale model or scalable wall function?
2-What should I choose in the volume condition initialization panel for turbulence : by values for the selected zones, by reference velocity for all zones, reference velocity and reference length for all zones?
3-For the turbulent boundary conditions for the inlet : calculation by hydraulic diameter or by turbulent intensity (and which value)?

I tried too to deactivate the reconstruction flux box in the equation parameters scheme panel as I had been advised, but it didn't changed anything to the result.

I attach one of my cas.

Thank you for your answer,

Alexis

Hello,

Checking your log file, we can see that the maximum velocity starts increasing after only a few time steps, so the calculation starts diverging progressively.

For 1), I'm not quite sure, so I'll have to check.
For 2) and 3), reference velocity and hydraulic diameter (easy to determine for a pipe) should do.

Did you check your physical coefficients (volume mass, viscosity, ...) ?

If your data is not confidential, could you send a link to your mesh, xml file, and possible user subroutines to a large-file transfer service so that we may reproduce and analyze the cause of the diverging calculation ?

Best regards,

  Yvan

Thank you again...it's not the first time I ask you some advice!

The physical coefficients are OK, I redo 2 calculations and will send you a link for all my data when they will be over.

The thing I don't understand is "reference velocity" to initialize turbulence. Is it the friction velocity (which is lower than the inlet velocity), the velocity on which the Reynold number is based  or the characteristic velocity of the fluctuating flow u' (saying u=umoyen+u')?

I think the gradient calculation method (in the global parameters panel) has an influence on the calculation (it depends on the king of mesh used) because sometimes it diverge after a few time steps, sometimes not but my values of velocity are always wrong.

I ll make a link for my data as soon as possible.

Best regards,
 
Alexis

Hello,

I send you a link to a new calculation. The velocity at the outlet seems OK but I have some excessive values (in the listing look at Umax which should not be higher than 2.09 and dissipation which is too high). Moreover, the calculation seems to diverge at the very beginning but finally it calms down. I don't understand this behavior. I hope you will have all the files needed, here's the link with the password "alexis" :
http://dl.free.fr/vm8fC5o7Y

I attach two pictures with this which are 10 cross sections regularly spaced (the first one from x=0 to x=1 and the second from x=1 to x=2). They represent the contours of U-velocity colored by velocity magnitude. The planes number 2 and 3 are not regular : the number 2 shows a too high velocity in the center of the pipe and the number 3 an offset velocity contour . What do you think of this? I am sure that the method of  gradient calculation has an real impact on the result. Maybe you could advise me the proper one for the mesh I send you (which is not a really good one I know but it's for beginning simulations, a hexa mesh may be better...)

Best regards,

Alexis