k-omega SST discrepancy between v516 and v530

Questions and remarks about code_saturne usage
Forum rules
Please read the forum usage recommendations before posting.
Post Reply
fmastrip
Posts: 3
Joined: Tue Apr 06, 2021 3:43 pm

k-omega SST discrepancy between v516 and v530

Post by fmastrip »

Hello,

I did last year some simulations of a jet in cross-flow using the version 5.1.6 :

Image

I try this year to run the same case in 5.3.0 (the switch of version is motivated by the use of the HTLES model developed in the thesis of V. Duffal).

I expected to find the same results, but it was not the case :oops: . Using the bdiff tool, i found a maximum difference about 10 m/s (for a maximum velocity in the domain around 20 m/s) located just at the outlet of the jet (Uj on the picture).

I try to find the origin of the discrepancy, without success. The latest (and simpler) setup I have tried is :
  • k-omega SST model using the Neumann condition for omega (I have modified the clptur.f90 file for the v530 to be sure that this condition was chosen by the code)
  • Spatial schemes : upwind for all variables
  • Linear solvers : Conjugate Gradient with Multigrid preconditionner for Pressure and Symetric Gauss-Seidel (without preconditionner) for the other variables
When I compare the setup.log files, I have the same information except for the pressure solver where we can read "algebric, faces in natural order" in v516 and "defaut" in v530. But I don't believe that the 10m/s difference come from that...

The two setup as well as the mesh used are available here (the coarser mesh size is 18Mo...) : mesh and cases
If you could have a look to this problem, it would be wonderful :D

(Please note that the incoming profile are defined into two .dat files stored in the DATA folder. The interpolation at the boundary of the domain is done using a Python3 script also in the DATA folder. You should change the system command line into cs_user_boundary_conditions.f90 to match with your installation of Python).

Thanks in advance !
Regards

Franck
Yvan Fournier
Posts: 4069
Joined: Mon Feb 20, 2012 3:25 pm

Re: k-omega SST discrepancy between v516 and v530

Post by Yvan Fournier »

Hello Frank,,

both versions 5.1 and 5.3 are "retired" versions, which were never fully validated (compared to v5.0, v6.0, and soon v7.0) and are not maintained anymore, so we only run these versions anymore, except as a last resort to trace a regression, when all other means fail, as this can be quite time consuming, especially with complex setups and large meshes such as you may have. They would probably not even compile anymore on machines such as our latest cluster ,with more recent compilers (gfortran 10 or Intel 19 or above) so patches would need to be backported...

Are your flow structures very stable or unsteady ? If the structure moves a bit from one time step to the next, even a small perturbation can explain a local difference if you look at instant (instead of averaged) values in areas of large velocity gradients. If you postprocess the 2 cases, do you see important flow structure differences ? Since "bdiff" checks only a local maximum, it is probably very sensitive, especially on large meshes where even truncation error differences using the same code version on a different number of processors can lead to high local differences for very similar flows.

Which version seems more stable/realistic ? v6.0 seems a bit more sensitive to some outlet boundary conditions than v5.0 on some cases I have seen, so if 5.1 is more stable than 5.3, that would be and interesting (and possibly) related information.

I would recommend checking the case also with v6.0, or even v7.0-beta (from GitHub), as validation is 2/3 finished, and so the version is probably already more stable than 6.3, 6.2, or 6.1. And if there is an actual bug and not simply a sensitivity/interpretation issue, fixes would also go there.

Depending on what you observe with postprocessing, we can determine how best to proceed.

Best regards,

Yvan
fmastrip
Posts: 3
Joined: Tue Apr 06, 2021 3:43 pm

Re: k-omega SST discrepancy between v516 and v530

Post by fmastrip »

Hi Yvan,

Thanks for your response, and sorry for the delay, I faced some troubles in running the new simulation.

I've tried ton run the case with the 6.0.0 version.
I can show you a slice in the hole of the jet
comparaison_5.1.6-5.3.0-6.0.png
.

The results of the 6.0.0 are close to the 5.1.6 (perhaps the discrepancy is related to time averaging or linear solvers).
The 5.3 gives weird results, with an under estimation of the velocity and the turbulent kinetic energy...

Do you have any idea ?
Thanks
Franck
Yvan Fournier
Posts: 4069
Joined: Mon Feb 20, 2012 3:25 pm

Re: k-omega SST discrepancy between v516 and v530

Post by Yvan Fournier »

Hello,

Looking at the release notes and git commit history, I don't see anything obviously related, but there were many small changes and bug fixes, so short of a full bisection (which may not be a priority for anyone), I can't be sure.

Not sure I understand if results are time-averaged or not.

I can asks colleagues if they remember a specific change.

Best regards,

Yvan
fmastrip
Posts: 3
Joined: Tue Apr 06, 2021 3:43 pm

Re: k-omega SST discrepancy between v516 and v530

Post by fmastrip »

Hi,
Thanks for your reply and your help

The result are time-averaged. It was done over the same time period and the same time step.

Regards
Franck
Yvan Fournier
Posts: 4069
Joined: Mon Feb 20, 2012 3:25 pm

Re: k-omega SST discrepancy between v516 and v530

Post by Yvan Fournier »

Hello,

It may be interesting to use multiple cuts, to see if the structure is simply "shorter" in v5.3. That would still not explain why you have different values, but maybe help evaluate how strong is the difference.

Best regards,

Yvan
Post Reply