Possible Bug in Elmer Heat Solver with Radiation BC

[carstenp]

Hello everyone,

I tried modeling the thermal coupling of several bodys in vacuum in Elmer. I started rather simple with a few cylinder with high thermal conductivity, internal heat source (body force) and diffuse radiation boundary conditions. The simulation works in principle quiet well unless I place the objects very close together, in this case the simulation breaks down at a certain iteration step after applying the boundary condition with "NaN". The problem seems to be related to the closeness of the radiating bodies. Has anybody any experience with the simulation of radiating structures in close proximity with elemer.

I attached a simplified model, the simulation diverges in the fourth iteration.

I hope anyone can help?

Thanks!

[mzenker]

Hi,

I have tried your simplified model and get the same result, i.e. NaN at the 4th iteration.
I haven't used the viewfactors calculation before, so I cannot tell if the problem lies there. But one thing that seems strange to me is that you have the 9 cylinders declared as only one body. What happens if you declare each cylinder as its own body instead?
If that doesn't change the problem, this is a question for the real gurus...

Matthias

[carstenp]

Hi mzenker,

thanks for trying I modeled all cylinders as one body since I only have one material and one boundary condition so I guess this should work.

Cheers,

Carsten

[carstenp]

Ok, I modeled the same setup with every volume body as seperate group and every surface as distinct boundary group an I still get the same strange behavior of NaN in the fourth itteration.

Any other suggestions?

Best!

[cemg]

Hi,

try using only the Picard method for the non-linearity, as Newton method may not converge. That is, in your sif file

Code: Select all

  Nonlinear System Newton After Iterations = 20
  Nonlinear System Newton After Tolerance = 1.0e-7

Regards,

Cesar

[raback]

Hi

I didn't look at the case but I may comment on possible problems: the view factor computation may run into problems when the gap between boundaries is smaller than the elementsize. Then the numerical integration may fail more easily. If the gap is really small it would be better to use the "heat gap" model. Can you comment on what the elementsize vs. gap size is in your system?

A good sign of an accurate computation of radiation factors (in a closed system) is that the row sum equals to 1 (normalized full angle). After that a normalization procedure is applied but it can not always resolve the problems.

Finally, if in your system radition is dominating you might try to set the "Nonlinear System Relaxation Factor" to ~0.5 or so. Otherwise the computation may start to oscillate.

-Peter

[carstenp]

Hi Cesar, Hi Peter,

the advice to omit the newton iteration did the trick, the simulation did converge (but I am a little confused since I thought until now that the newton-rhapson method has better convergence properties than the picard iteration).

The structure with the cylinders was just a test case for a slightly more complicated problem (multiple bodies in close proximity in vaccum - so my system is open). I also observed sometimes viewfactors larger than one. But I think that the ViewFactors code just needs a little reworking. If elements are close together the view factor should asymptotically converge to one. So if the ViewFactor code computes a Viewfactor sum larger than one the ad-hoc solution would be a renormalisation of that row to "1" .

Peter, thank you for your suggestions, I will try them when I run into new convergences issues.

Thanks everyone for the quick help!

Carsten