Hello,
I am currently using Elmer for thermal simulations of bodies in close proximity at high temperatures with open surroundings. The basic mechanism for coupling of the different bodies is radiation. I performed several simulations in 2D and the results are very promising. Unfortunately, the results of 3D simulations are rather disappointing. I traced the problem to the view factor computation.
In 2D, the viewfactor sum is on average 0.3 with a few single factors with a magnitude larger than one. If I extrude the 2D mesh in the third dimension, the view factor computation goes totally south with a view factor average of 4.2 and single viewfactors with a magnitude of 99.
Is their a way to improve the accuracy of the view factor computation?
Regards
carsten
ps: Using heat gap as a substitute is not a practical option for me.
Viewfactor Computation for Elmer
-
- Site Admin
- Posts: 4868
- Joined: 22 Aug 2009, 11:57
- Antispam: Yes
- Location: Espoo, Finland
- Contact:
Re: Viewfactor Computation for Elmer
Hi
It may be that the extrusion from 2D to 3D is not the same problem. The 2D may be closed and then normalization of the view factors improves the accuracy. However, when you extrude the geometry the system is not closed since no periodicity is enforced in the z-direction. As a result, if you try to normalize the view factors to 2PI things go even more wrong.
The 2D and 3D cases are treated by ray tracing whereas axsiymmetric case uses determenistic integration. This probably means that you'll have more variation in the cartesian cases. It is also natural that the 3D case would require significantly more CPU time for comparable results.
The following keywords (with defaults) control the accuracy of the ray tracing. Perhaps you can play with these:
-Peter
It may be that the extrusion from 2D to 3D is not the same problem. The 2D may be closed and then normalization of the view factors improves the accuracy. However, when you extrude the geometry the system is not closed since no periodicity is enforced in the z-direction. As a result, if you try to normalize the view factors to 2PI things go even more wrong.
The 2D and 3D cases are treated by ray tracing whereas axsiymmetric case uses determenistic integration. This probably means that you'll have more variation in the cartesian cases. It is also natural that the 3D case would require significantly more CPU time for comparable results.
The following keywords (with defaults) control the accuracy of the ray tracing. Perhaps you can play with these:
Code: Select all
Viewfactor Area Tolerance = 0.1
Viewfactor Factor Tolerance = 0.01
Viewfactor Raytrace Tolerace = 1.0e-5
Viewfactor Number of Rays = 1