Numeric instability switching from 2.0.1 to 3.3

[andrea-dyn]

Dear all,
since last year I've used CS 2.0.1 to analyse the pressure and velocity field in complex geometries (oleodynamic valves) without problems.
Recently I've switched to CS 3.3 and the same geometries (same meshes) that worked perfectly with CS 2.0.1 have started to do convergence problem with CS 3.3.
It seems very strange, but with all the meshes I've tried to use the calculation becomes unstable near 100 steps.
I think there is something I wrong in the calculation parameters setting (I use only the GUI), but after a lot of test I still not understand where is the problem...

In attachment you can find a simplified mesh that reproduces the problem, together with the xml file produced by the GUI.

Thanks in advance for the help....

Andrea

[andrea-dyn]

sorry, the mesh was not attached...

[andrea-dyn]

maybe the listing and preprocessor files can also help to find the solution...

[Yvan Fournier]

Hello,

2.0 was a fully validated version, while 3.3 is an intermediate version (the next full validated version being 4.0).

Did you test your case with version 3.0, which is also a stable version ?

Otherwise, we fixed quite a few bugs between 3.3.0 and 3.3.1, and a few more fixes have been pushed to SVN recently. We'll probably release 3.3.2 within 1 or 2 weeks, so I'll try to check your case on it, to see if we reproduce the issue with the current version.

Regards,

Yvan

[andrea-dyn]

Hello Yvan,
thanks for your quick reply...

I've already tested also with the 3.0 (in attachment you can find the XML file produced with the GUI, the listing file and the preprocessor log file), with the same result.

I've also done some test changing the gradient method to "Least squares method over extended cell neighborhood" (as indicated in the BPG), but the result seems not to be changed...

The calculation become unstable after more or less 100 steps.

Regards,
Andrea

[Jacques Fontaine]

Hello,

Did you compare 2.0 and 3.0 listings?
Could you try with IDTVAR = 0 (unsteady algorithm, with constant time step) and also with another turbulence model (eg: k-eps)?

Regards,

[andrea-dyn]

Hello Jacques,
thanks for the reply.

What do you mean when you suggest to compare the listing files? They contain a lot of data, can you please indicate what are the most significant parameters to compare?

In any case, I've tested the calculation with unsteady algorithm and constant time step (I've left 0.1s as per default) and the stability seems improved, but near to 200 steps the calculation starts to diverge.
In attachment you can find the xml generated by the GUI and the listing file.

Today my time on calculation machine is ended, so I'll test the case with the k-eps model tomorrow.

Thanks in advance for the help.

Andrea

[andrea-dyn]

Hi,
I've tested also the k-eps turbulence model (both k-eps and k-eps-PL), but the result is similar to the mixing_length one; the stability seems improved, but near to 200 steps the calculation starts to diverge.

In attachment you can find the xml generated by the GUI and the listing file.

Thanks in advance for any help...

Andrea

[Yvan Fournier]

Hello,

I just checked your last listing, and the CFL and Fourier numbers are is very high right from the start (iteration 1 for Fourier, 2 for CFL).

Are you sure you used the same time stepping scheme for versions 2.0 and 3.x ?

In you last listing, you have a constant time step of 0.1 s, and the CFL numbers are very high.

Did you try reducing the reference time step (even if you switch to a variable time step or pseudo-time step, this will determine the initial time step, and help the calculation initialize better).

Regards,

Yvan

[andrea-dyn]

Hello Yvan,
thanks for the tips.

I'll try again now to redo the calculation on the basis of advice that you gave me ... I'll let you know the results ...

Regards,
Andrea