[otaolafr]

Hello,
I am seeking help in a strange behavior that I have no clue where to go.

I have a case with a complex geometry, and cyclicAMI and symmetry boundaries (it is a quarter of cylinder where the InletAMI and OutletAMI are connected by cyclicAMI and sym_1 and sym_2 are symmetry planes).
I create my volume mesh using snappyHexMesh and after create the symmetry and cyclicAMI patches with createPatch up to here everything okey.

my issue comes here: if I run simpleFoam it runs correctly until it converges to 1-5 residuals for p and U. but if I run pimpleFoam the solver explodes with the typical error divided by zero where there is not much indication about the problem. it crashes after 3 or 4 iteration whatever the deltaT is (it is not a Co number problem as it crashes even with values down to maxCo 0.02)

and I dont know where to go from here as the fvSolution and fvSchemes seems okey, even if I use the results from a converged simpleFoam and from there I run pimple, it will crash after 3 or 4 iterations (in time)

I am attaching the fvSolutions and fvSchemes and also the link for the case as it can be difficult to have an idea with my description of the problem (but I not sure what can I said more than these... as the error does not give any clear information.)
small note, in the case i put a mesh that is quiet coarse, but the problem is still there even at finer meshes.
link for complete case: https://drive.google.com/file/d/1EYO...ew?usp=sharing



fvSchemes_1.txt

fvSolution.txt
any help or point in direction will be appreciated

[HPE]

Please remove the 'bounded' keywords for any transient simulation.

Hope this helps.

[otaolafr]

Originally Posted by HPE

Please remove the 'bounded' keywords for any transient simulation.

Hope this helps.

Hi man, nice to cross you again (as always thanks for the help!)

even when it is running in steadyState mode? see fvSchemes

I am going to try of course, only asking to learn a little bit more
best regards

[HPE]

Hi

No no, please keep the "bounded" keyword for steady-state simulations. But remove it when you run transient simulations. This keyword is useful for steady cases, but destabilises transient cases (to my experience).

[otaolafr]

Originally Posted by HPE

Hi

No no, please keep the "bounded" keyword for steady-state simulations. But remove it when you run transient simulations. This keyword is useful for steady cases, but destabilises transient cases (to my experience).

still exploding like crazy :/
log.pimpleFoam.txt

[HPE]

I had been writing items - and everything was vanished... This forum is really low quality...

Anyways, some ideas:

fvSchemes:

- I assume you don't use steadyState for the ddtSchemes. If so, please change it to Euler.
- please use a limiter (e.g. cellMDLimited) in gradSchemes - and preferably - don't use it a default value there
- if things keep blowing up - use the most diffusive schemes in fvSchemes - you can search them, e.g. upwind for divergence
- check AMI settings - there was a limiter that can be applied to AMI to avoid zero-face issues - please search it (I must sleep - otherwise will post tomorrow) AMIWeights FO

fvSolution:

- relaxationfactor for p from 0.5 to 0.3
- relTol for pFinal from 0.01 to 0
- use the same smoother for the p solver
- nOuterCorrectors from 1 to 3


EDIT: could you please also add fieldMinMax FO to monitor min max p, U as well?

[pferro]

Hello,


I agree with HPE.


you have no limiter on your schemes. This is very unlikely to obtain a non oscillating or stable solution with Gauss linear for div(phi,U). Instead : linearUpwindV grad(U) and add a limiter on grad(U) or limitedLinearV 1 or vanLeerV. Go to upwind for turbulence.
Eventually add a stronger limitation on non ortho correction : limited 0.7 instead of corrected.
For relaxation factor because you are want to use pimpleFoam I assume you want to get a transient result. The relaxation factors for the last PIMPLE iteration must be 1.


Good luck


Paulin

[otaolafr]

Hello guys,
wow so much to learn, it is difficult to even follow you! eheheh, I should learn more about the schemes for the different terms.

Originally Posted by HPE

I had been writing items - and everything was vanished... This forum is really low quality...

Anyways, some ideas:

fvSchemes:

- I assume you don't use steadyState for the ddtSchemes. If so, please change it to Euler.
- please use a limiter (e.g. cellMDLimited) in gradSchemes - and preferably - don't use it a default value there
- if things keep blowing up - use the most diffusive schemes in fvSchemes - you can search them, e.g. upwind for divergence
- check AMI settings - there was a limiter that can be applied to AMI to avoid zero-face issues - please search it (I must sleep - otherwise will post tomorrow) AMIWeights FO

fvSolution:

- relaxationfactor for p from 0.5 to 0.3
- relTol for pFinal from 0.01 to 0
- use the same smoother for the p solver
- nOuterCorrectors from 1 to 3


EDIT: could you please also add fieldMinMax FO to monitor min max p, U as well?

Originally Posted by pferro

Hello,


I agree with HPE.


you have no limiter on your schemes. This is very unlikely to obtain a non oscillating or stable solution with Gauss linear for div(phi,U). Instead : linearUpwindV grad(U) and add a limiter on grad(U) or limitedLinearV 1 or vanLeerV. Go to upwind for turbulence.
Eventually add a stronger limitation on non ortho correction : limited 0.7 instead of corrected.
For relaxation factor because you are want to use pimpleFoam I assume you want to get a transient result. The relaxation factors for the last PIMPLE iteration must be 1.


Good luck


Paulin

- I assume you don't use steadyState for the ddtSchemes. If so, please change it to Euler.
well I was trying to use the pimple algorithm in steady state mode. thats why I was using the
- check AMI settings - there was a limiter that can be applied to AMI to avoid zero-face issues - please search it (I must sleep - otherwise will post tomorrow) AMIWeights FO
I did not understood your comment about these, and I am quiet interesting in this issue, as when I go in higher refimments (lower cell sizes) I am finishing with 0 weighed AMIs and then the simulation explodes.... I understand that this is comming from a face that does nothing that overlaps in the other patch. but what I dont get is as the refiment gets better the cyclicAMI should too (as the geometry is better represented)
the function that you sendme weightsAMI if I understood correctly it only help to visualize the weights in the patches but nothing else. do you know any particular way of getting better weights? the geometry I can not changed....
for the rest of the things I am going to integrate them to the fvSchemes and fvSolutions when I go back to work as I do not have acces right now, but I promis I will have a look. so much to learn thanks guys!

[HPE]

well I was trying to use the pimple algorithm in steady state mode.

I don't think this can work out.