Strange Result using Diffuse Radiation Boundary Condition

[carstenp]

Hello everyone,

currently I am playing a little with ElmerSolver's HeatSolve capabilities and radiative heat transfer. However, in some simulations I observed a very strange phenomenon. I attached a simple example to illustrate the problem. I placed two layered cylinders in close proximity. The inner part of each cylinder contains a volume heat source while the outer is simply a cylindric shell with bad thermal conductivity. When I place the layered cylinders in close proximity to each other I would expect a temperature increase at the area were both cylinders get closest. Instead, this is the coldest part of the model. So instead of heating the opposite cylinder, the cylinders seem to cool each other.

To see the strange results, simply unzip the file and open/solve it with ElmerSolver

Cheers,

carsten

RadiationBug.zip

(1.53 MiB) Downloaded 395 times

[mzenker]

Hi Carsten,

I had a look at your case. I also find the results strange - if you look more closely, you see that the parts of the cylindres directly facing each other are not the coldest parts. Instead the areas directly next to those are coldest. I don't see an obvious physical explanation for this behaviour.
Due to lack of experience with viewfactor calculation in Elmer I cannot tell what is wrong with your simulation, but would suspect the viewfactor calculation. Maybe it is a problem that the two inner and outer parts have the same body number, and that the external boundaries also have the same number? You could try to reprocess your mesh with Elmergrid -autoclean and see if that changes things.

Matthias

[raback]

Hi

It seems that you should finish off with Newton, or even start with it. It may be that partially open cases were verified insufficiently. So try with

Code: Select all

  Nonlinear System Newton After Iterations = 0

-Peter

[carstenp]

Hi peter,

thank you for your advice but this only works in a few cases. In many cases I modeled so far the Newton procedure produced a NAN as soon as it was used. So it is not really useable for the general cases. Do you have any other suggestions?

carsten

[raback]

Hi

Well, I could see that the ILU preconditioner could have problems. What if you used "none" as preconditioner (diagonal), BiCGStabl as the iterative method, and additionally set "BiCGStabl Polynomial Degree = 4". I.e. improving with iterative method while using more simple preconditioner.

-Peter

[carstenp]

Hi peter,

I tried your suggestions, unfortunately it didn't work. I think the problem has nothing to do with iteration but something goes wrong during the matrix assembly:

1.) I also tried umfpack as direkt solver => Same result: NaN
2.) The solution is not slowly diverging. Instead from the first steady state iteration and the first nonlinear Iteration the result is NaN.

So I guess something goes wrong with the matrix assembly or a related step.

carsten

[cemg]

Hi all,

it seems to me that the colder parts are those elements with, at least, a nonzero viewfactor to some element on the other cylinder, so part of the radiation flux goes to the other cylinder (even a zero net flux if the temperatures are equal) and the rest goes (ideally) to the surroundings. The rest of the elements (those that have all zero viewfactors) are not radiating because they doesn't have a viewfactor defined (look into the viewfactors.dat file, there aren't zero values, the values are missed), so the complementary isn't defined neither and there's no ideal radiation to the surroundings.

Regards,

Cesar

[raback]

Hi Cesar

Just the non-zero vf's are saved. However, the determination that a boundary element is radiating is done independently in the code. Hence the code should return correctly the open fraction. There seems to be some issue, but it is probably not this one...

-Peter

[carstenp]

Hi Peter,

I did some further experiments yesterday:

I had a model that produced a NaN in the first iteration. Then I converted the mesh in salome to second order and it worked well. If however I simply increase the order of the basis functions in Elmer it won't work. So my guess is that somehow the node ordering is related to the problem.

carsten

[raback]

Hi Carsten

I would think that issues related to node ordering would already have surfaced by now. Perhaps you could share the case with NaNs.

-Peter