open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi all!

I have a 3D model in which I would like to model flow below a concrete slab. To model this non free surface situation I try to use a locally increased air pressure.

The geometry is as follows:
-Inflow section is a rectangular opening, the vertical profile has been defined in vel_prof_z
-Downstream of the upstream boundary there is a short portion with free surface before the flow plunges below a concrete slab. The base of the concrete slab is located 1,6 m below free surface and 0,3 below the upstream boundary (the flow will hit the concrete slab before plunging below it).

I did simulate something similar but with a rather loose mesh and it worked rather well. This time I need to model the case into much finer resolution. The mesh I use now corresponds to a simplified geometry with 5 cm spacing in horizontal and 10 cm spacing in vertical (50 planes).

I use the non-hydrostatic version and I tested with constant viscosity and k-epsilon.

The model crashes in subroutine FSGRAD. I tried to tune the Free surface compatibility coefficient (0,0 to 1,0) but the model still fails. I guess I am demanding too much when imposing a gradient of air pressure of 1,6 m between two nodes separated by only 5 cm? Can I try to improve the case by tuning some numerical parameters?

I attach my model files.

For information, I have good results with the same model but without applying the locally increased air pressure (ie. free surface only).

As an alternative I can try to model the concrete slab by applying a drag force instead but I would really like to see if I can manage with the lowering of the water surface.

Thanks in advance for your help!

Best regards
PL

Hello Pierre-Louis,

You have ventured into terra incognita and nothing is guaranteed. However a crash in fsgrad.f is strange. It computes the gradient of the free surface and there is no such thing as a division by 0 in it, even with tidal flats. Unless you send data that is already NaN it cannot be fsgrad.f that crashes. If you can see what kind of crash it is it would be helpful. Otherwise send your case and I'll run it with every checking to know which line it is and the real reason.

With best regards,

Jean-Michel Hervouet

Hi,

Yes, I was guessing that this was maybe "too much".
Anyway, would be the opportunity to test something new!

I don't know the real reason of the crash, I just don't get the "Back from FSGRAD" in debugger mode. I don't have any traceback options in my gfortran set-up.
The model files were actually attached to the first post, can you not see them (in a zip file)?

Thanks in advance!

Best regards
PL

Hello,

I tried your case with 8 processors and the trunk, and it went through the 2000 time steps... Anyway it is clear that the difficult step is the pressure. You could apply a bit less brutally the hydrostatic pressure in meteo, with some intermediary values when approaching the slab, or using subroutine FILTER like in corfon.f to smooth the pressure field.

You can also use GMRES instead of conjugate gradient for finding the dynamic pressure:

SOLVER FOR PPE : 7
OPTION OF SOLVER FOR PPE : 5 (this is the Krylov space dimension, it can be tuned to optimise the number of matrix-vector products, which is:

(number of iterations of the solver) * (OPTION OF SOLVER FOR PPE )

With best regards,

Jean-Michel Hervouet

Hello,

Sorry for late reply.
The only info I have on my crash is the following:

job aborted:
rank: node: exit code[: error message]
...
6: PCSTH13495.sweco.se: 1: process 6 exited without calling finalize
...

Is it gfortran or mpi? If it is related to the compiler then the crash might be caused by some gfortran options?

Anyway, I will try to play a bit with your suggestions (this model was more for fun than something to be delivered so I hope I find some [free] time to continue with it).

I thought of applying a smoother pressure gradient but the aim is to model a horizontal flow meeting a vertical slab so I can't tweak the gradient too much. But maybe it can help if only the upper and the lower part are slightly smoothed.

By the way, I run with 7.1.0, I might try the trunk soon also.

Best regards,
Pierre-Louis