code_saturne User's Forum

Hello,

thank you to EDF and all the developers for providing CS free of charge! It has been fun to learn and use on some simple problems. However, now I am hitting a mountain I can't seem to overcome.

I am trying to determine the air distribution in a 2,300 seat theater with seating at the orchestra, 1st and 2nd balcony levels. The geometry is, say, 38 m long, 18 m wide and 18 m tall, with a symmetry plane on one side. All I want to do is model the audience members as surfaces with constant heat flux, have various air supplies with certain volume flow rates and temperatures, and watch how the air then makes it way to the air returns and what the temperature in the theater will be over time. Though humid air would be great, I'd be happy to get this working with just dry air, where I am currently using Boussinesq for its mass density.

I have spent many, many, many hours in the past three months trying to get a transient calculation to converge, but it invariably blows up. Part of the problem is not knowing exactly what mesh is best for this, and how to generate it in Salome (I have only ever succeeded in generating tetrahedrons), the other is not knowing exactly what settings to use in CS, e.g. turbulence, upwind/centered, multigrid/Jacobi, etc. I can provide the Salome .HDF and CS .XML files, but am hoping someone must have solved this and can provide direct insight?

Many thanks,

Matt Koch

Hello,

It is hard to provide good recomendations with n more details on the mesh. In general, options which reduce results precision or make it hard to capture some physical flow features also make the computation more stable (using upwind, k-epsilon turbulence model, ...). So I would recommend trying to stabilize the computation with those options, then progressively switch options for better precision.

In any case, the mesh and time step are important, so how and when (i.e. after how many time steps) the computation blows up is important. Also, if you have no forced ventilation but only natural convection, you can expect to have pretty high sensitivity to heat exchange aspects, and convergence may be slow (but using a local time step can usually help).

Please check the top post of this forum for detailed recommendations on what info to provide for more efficient help, and a few views of your mesh may also help (if your mesh is small, the mesh itself may do, but given your geometry description, I assume it is sufficiently complex so that your mesh is not too small.

Regards,

Yvan

Dear Yvan,

thank you kindly for your quick response. I am glad you mentioned k-epsilon and upwind. These are the ones I used - so I guess I am not too far out to lunch. Actually, I had even gone so far as using SOLU rather than upwind, without a real difference in outcome. As rusty as I am on this, I even started out using a constant time step, but then changed it to a variable time step, without success. Perhaps the local variable time step you mentioned is the answer - I will kick that off before I go home today.

I would imagine this problem to be initially largely a natural convection one, but then as the air from the supplies in the ceiling reaches the audience below, it will be more forced convection, or perhaps a combination of both.

Lastly, I checked the forum reference you mentioned. I am using Code_Saturne 4.0. I do not have a "compile.log" file or any user functions, as I am setting everything up through the GUI. Other various files mentioned in that forum reference are attached ( I think), though for the XML file, I had temporarily taken out turbulence, just to see.

Anyway, a lot of info, and a lot to wade through, if you chose to do so. Any nuggets (like your local time step hint) would be much appreciated.

Best regards,

Matt

Actually, "listing" and "error" did not seem to make it through, so here they are (I think).

Hello,

The last 2 files do not seem to have been uploaded.

Actually, a limitation of the file upload filtering of this forum prevents you from uploading error and listing files with renaming, but adding a ".log" extension should do the trick.

Regards,

Yvan

Thanks Yvan, I was beginning to think that there was some sort of file limitation.

So, I restarted my case yesterday with the local variable variable time step you mentioned (presumably "Time step limitation with the local thermal time step" under Numerical parameters -> Time step?) - to no avail. Attached are the XML, listing and error files from that run. It appears as though perhaps the Courant-Friedrichs-Lewy number is still way too high? With CFL = u dt/dx, for a fixed u, all I can do is decrease dt or increase dx, and I feel I have done that to the point where anything beyond that would give me useless results.

Let me ask this: I am specifying an unconnected set of supplies as the inlet - basically a set of small disks. When I specify the inlet volume flow rate for that inlet, that is meant to be the volume flow rate for all disks together, right, not just for one disk? If it were for just one disk, then that might result in high u, which would make CFL large, I suppose. Unlikely, but I am asking just in case.

Sorry to be so clueless about this! I must be missing something very obvious. The old forest for the trees perhaps ...

Hello,

I was thinking of yet another time option, which is the default steady solver (idtvar = 2 in in the listing file).

In any case, variable time step (in time and/or space) starts with a reference time step ,and changes it gradually, so your initial time step should not be too high (or your allowed multiplier very high, but that might not be very safe).

AS you immediately have very high CFL numbers, I recommend dividing your reference time step by 100 (you may even try 1000). I would start without radiative transfers, and possibly add those later.

If that is not enough, setting a "relaxation of pressure increase" in global numerical options or adding "sweeps" in equation parameters might also help.

Where you have outlets, it is always recommended to ensure the mesh is orthogonal at those places (so extruding one or two layers to prolong the outlet there may be useful). Version 4.3 has an automatic extrusion of selected boundary faces option, earlier versions do not...

Regards,

Yvan

OK, so shorter time steps or even steady state, some variation on MultiGrid/BiCG and upwind/SOLU, pressure/velocity coupling or even ramping up the flow from the supplies linearly with time does not yield a solution either. I am beginning to think that either the mesh is just completely inappropriate or that Code Saturne just cannot solve this problem, though I think the latter rather improbable. I am at wits' end and will attempt to transfer this problem to FDS/SMV now. Disappointing after all the work I put into Code Saturne. Oh well.

Hello,

Do you see any difference at how fast the code crashes depending on options ? Did you try relaxation of pressire increase ?

If possible, could you send the mesh (or a link to the mes) and data setup using a large file exchange service ? If is is somewhat confidential, you can use the private message option of this forum.

This would allow us to check if the mesh is usable, if we can get the computation running with the right options, or if some options under development can solve your issues.

Regards,

Yvan

Hello Yvan,

When I decreased the time step, it then never stopped the runcase.bat file, like it did before, but the listing file nevertheless quickly showed very large and implausible numbers for all kinds of variables, at which point I terminated the run. I used some factors other than 1.0 for some of the computational settings, too. No luck.

I uploaded some files to my dropbox, but how do I share them with the Code Saturne forum or you? (My e-mail is mattk@clearyzimmermann.com)

Thanks,

Matt