code_saturne User's Forum

Hello,

I am trying run a thermal simulation of an H7 halogen bulb with conduction, convection and radiation. I am using SYRTHES coupled with Code_Saturne. I already managed to run this simulation with a simplified geometry. I used a simple cylinder for the coil geometry. 

After that I changed the mesh and used the real geometry of the coil. But with the new mesh the generation of the form factors for radiation – which needs more than two hours on my computer – always fails with this error:

 *** CFDF3D : calcul des facteurs de forme
              10.00 % du calcul effectue
              20.00 % du calcul effectue
              30.00 % du calcul effectue
              40.00 % du calcul effectue
              50.00 % du calcul effectue
              60.00 % du calcul effectue
              70.00 % du calcul effectue
              80.00 % du calcul effectue
              90.00 % du calcul effectue
At line 565 of file /opt/syrthes/syrthes3.4.2/src/s/inisol.F (unit = 65, file = './fdf')
Fortran runtime error: Cannot allocate memory

At first I thought something is wrong with the mesh and I remeshed everything carefully, but the error still occures.

Do you have an idea what may cause this error?

Kind regards,

Martin Isack

Hello,

How big is your radiation surface mesh ? In SYRTHES 3.4, the computation of form factors requires O(n 2 ) time and memory, so the radiative mesh should generally be significantly coarser than the solid mesh, at least where you do not expect too large gradients along the surface.

You may check memory usage or the size of the fdf file before the crash, but I suspect they are very large.

I believe there are optimizations regarding this O(n 2 )aspect in the future SYRTHES 4, but I am not sure (in a convex domain, every face can see every other face, so the problem is inherently O(n 2 ) (though a multi-resolution mesh might be a workaround).

Best regards,

  Yvan

Hello Yvan,

thank you for your reply. The radiation mesh consists of 10 000 elements. I have attached two screenshots of it. I will try ot make the mesh more coarse, even it it is quite difficult, because of the geometry.

Do you mean the radiation mesh should always be coarser? Or only if the solid mesh is very fine?  At the moment the solid and radiation meshes are nearly the same.

Kind regards,

Martin

Hello,

The recommendation that the radiation mesh be coarser is quite general (at least last time I talked with the SYRTHES developers about this specific subject). With 10 000 elements, the form factor matrix has 100 million entries, which would amount to slightly under 1 Gb in double precision. I do not know if there are multiple copies (i.e. matrix and its inverse) in SYRTHES, but in any case, solving this will be slow.

If you can manage to divide the mesh size by 2 or 3 (elongating triangles along the coil, as you seem close to the minimum in the other directions), things should be much smoother.

Best regards,

  Yvan

Hello Yvan,

thank you for your help again. I changed the radiation mesh to 2500 elements and were able to run the simulation without errors.

Regards,

 Martin

Hello,

Good to here that it works.

Actually, it seems that the visibility matrix size is not quite n*n, as it is symmetric, but in any case, its size increases in a quadratic manner with respect to mesh size, so using a coarser mesh for radiation is recommended.

Best regards,

  Yvan