Hello,
I am using for the first time Code saturne for modelling a cyclonic filter. For the first run I disabled the particles tracking to see if a simple calculation converge, and the aim is to implement particles once everything is correctly set-up. I already made a first calculation, but it seems to me that it did not converge as the velocity and pressure scalar were still varying when postprocessing in Salome. Then i decided to increase the number of iteration by using the restart option. I think I did not correctly set this up as, as soon as it starts it diverges. I should go deeper in this restart option because i might have missed something. Nevertheless I launched a new calculation with a higher number of iteration and regularly followed it by opening the listing file in the "tmp" folder". it started nicely and from what I saw it was quite stable, but at one moment for any reasons it started diverging at step 2240, at a far cell, and then it spread everywhere leading to a "Signal SIGFPE (exception en virgule flottante) intercepté !" certainly due to high values of the figures, or something else. I enclosed link to the mesh, the listing file, and the chr.med file. I tried to change the mesh to have a more regular one (the file enclosed is the last one) but it did not change, maybe it is not yet regular enough ...
Can someone be of any help on this ? thanks a lot.
listing is here :
http://dl.free.fr/n8z63M9y7
mesh is here :
http://dl.free.fr/tHK5lN766
chr.med file is here :
http://dl.free.fr/qb8M4nlhR
Fred.
Cyclonic filter
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
Yvan Fournier
Re: Cyclonic filter
Hello,
I did not look at your data yet, but you may want to try several options which may lead to a more stable calculation:
use gradient reconstruction with extended neighborhood use relaxation of pressure increase (0.7 instead of the default of 1)
If this is not enough, try adding a bit (10% for example) of upwinding
Also, depending on the variable which is the initial cause, you may try to add some clipping (define min/max values), though this is quite a barbarous approach to the problem, and may only delay it (though a temporary instability may be convected and evacuated this way if you are lucky).
All of the above options are available using the GUI.
A good solution based on recent experience is to increase the number of sweeps (I believe you need the usini1.f90 user subroutine to do this). Increasing the default from 2 to 5 or 10 may help, but in certain cases, you need to increase it to more than 100 for the first 10 or 50 iterations, and may lower this after (an automated test should be available in version 2.2 when the ivelco = 1 option is activated, but automating it otherwise is tricky), so you need to experiment.
Best regards,
Yvan
I did not look at your data yet, but you may want to try several options which may lead to a more stable calculation:
use gradient reconstruction with extended neighborhood use relaxation of pressure increase (0.7 instead of the default of 1)
If this is not enough, try adding a bit (10% for example) of upwinding
Also, depending on the variable which is the initial cause, you may try to add some clipping (define min/max values), though this is quite a barbarous approach to the problem, and may only delay it (though a temporary instability may be convected and evacuated this way if you are lucky).
All of the above options are available using the GUI.
A good solution based on recent experience is to increase the number of sweeps (I believe you need the usini1.f90 user subroutine to do this). Increasing the default from 2 to 5 or 10 may help, but in certain cases, you need to increase it to more than 100 for the first 10 or 50 iterations, and may lower this after (an automated test should be available in version 2.2 when the ivelco = 1 option is activated, but automating it otherwise is tricky), so you need to experiment.
Best regards,
Yvan
-
Frédéric PAGEOT
Re: Cyclonic filter
Many thanks for the answer. I will try and let you know the results.
-
Frédéric PAGEOT
Re: Cyclonic filter
Hello,
Thanks for your help. The calcultation converged with a 20% upwind as both advises (relaxation and gradient) were already implemented in the configuration. I tried with the same settings to launch a particle tracking calculation, but is failed "Jacobi: error (divergence) solving for VitesseX". I will dig further in detail to understand what's going on. If you have any advises for particles tracking your are more than welcome.
Thanks for your help. The calcultation converged with a 20% upwind as both advises (relaxation and gradient) were already implemented in the configuration. I tried with the same settings to launch a particle tracking calculation, but is failed "Jacobi: error (divergence) solving for VitesseX". I will dig further in detail to understand what's going on. If you have any advises for particles tracking your are more than welcome.
-
Frédéric PAGEOT
Re: Cyclonic filter
Hello, I made some changes to my mesh and calculation option and followed Yvan advises. Now everything is working fine even with turbulence model included which was not the case at the begining. But now everything is converging fine and when visualizing in SALOME, this is stable. But as soon as I restart the calculation with particle tracking it is diverging after few iterations (about 15) if I do not choose the frozen field calculation option. I certainly missed something. I will not upload the file unless you required it, but I would like to know if some of you already experienced something similar, or if you have any thing obvious you can think of that I certainly missed. Thanks.