Can anyone explain this? The chart bellow show results for a meshing sensitivity study where each design points represent a different meshing level using the automated meshing option. (design point 1 = Meshing level 3 , ......, design point 6 = Meshing level 8).
I can not explain why I see a significant drop in the temperature for the last design point.
Thanks,
Thanks for you answer Siavash..
What would you recommend doing?
If I had stopped at 'design point 5' it would have definitely look like the solution was grid independent.
Hi Sidi,
I would have quit at the third temp that was the same but possibilities include......
Is a steady state analysis? I have seen divergence with descretization in other analysis, mostly on drag numbers, temperatures are pretty well behaved.....can you show an image? Could be due to the emergence of a small flow feature with continuing refinement. What does the goal plot look like?
Hi Bill,
I set a parametric study (only changing the initial mesh density) and let it run.. That is why I caugh the divergence issue.
Yes it is a steady state analysis with natural convection.
The model is very simple. It is an electronics enclosure with a PCB and several heat loads.
Here is the goal plot.
I also see a drop in temperature with the other components but not as significant.
Any suggestion will be helpful.
Thanks.
what is the volume/area that the goal is being measured on? Where is it? Is it a reasonable thing that if could eb affected by a small flow feature that is being resolved at th new scale (seems a bit of a stretch but maybe). Another possibility is that if the cell count in the channels changed so that the empirical corrections are dropped for low cell counts if might make such a difference. You would have to make a test case to sort it out. I think it changes at about 10 cell across the channels.
Did you have 'advanced channel refinement' ticked? Set to minimum of 5 cells across channel if you are in manual mode..
From what I can tell mesh levels 1 to 3 are not useful for final results.
I would use local mesh around your heatsink. I assume you are using gap pad or some other TIM so include this interface resistance.
For natural convection you need a large computational domain (possibly 4x the heatsink height below and above the heatsink).
Refine the domain at level 4 and study local-mesh levels with channel refinement. Again I would not bother with level 1 to 3 except for debugging. Local mesh will more than likely need to be #6 or over. Please post results on your new finding.
Assuming everything is set properly, your solution is not grid independent yet.
PS. It is not a good practice to increase mesh density globally, although it is very convenient.