[RodriguezFatz]

Dear all,

I am now and then faced with the following heat transfer problem:
Some kind of "pencilish" device is heated on one side. All other sides are connected to ambient air. The temperature on the surface and inside the device is needed.
Since the heated side is plugged to some pipe / fluid or has some electronics with constant temperature, I don't simulate that heat source, but just set the wall temperature to some fixed value. Let's say 350K.

I see two options here:
1) Just set the other walls to heat sinks, with some more or less arbitrary heat transfer coefficient, maybe about 5. I think that is basically what all the FEA guys do !?!
2) Simulate the device + ambient air, with compressible air and buoyancy.

If I do the second, I usually create a box around the heated device, top of the box is a pressure outlet and all other sides are simple walls. The box must be large enough, of course.
excellent_painting.png
I made this (superb ) painting of the setup.
My question is: Does that make sense? Should I set some boundary in a different way? How do you do that kind of stuff?

[Hamidzoka]

Hi,
As far as I know, the second strategy has some issues to consider:
- surrounding space around your device is small relative to the device dimensions
- side walls is not physical. since you force the natural convection just in Z direction while the realistic convection with the surrounding air can be expected in other directions as well. in order to get better results the dimensions of the cube must be quite larger.
- Note that natural convection and buoyancy phenomena will be a difficult task from view point of convergence.

instead, maybe u can use existing NU correlations for a vertical cylinder in an iterative process to find the temperature distribution. In fact, the first approach makes sense to me.

[RodriguezFatz]

Hi,
I don't understand your concerns.
The device is about one order of magnitude smaller than the surrounding space. So it's exactly the opposite of what you state in your first point.
I also see point two as a problem, but if the walls are "far enough" from the device the exact choice of boundary condition is not important.

[FMDenaro]

if the device is so small compared to your computational domain lenght, you could also consider to use periodic boundary conditions instead of walls

[RodriguezFatz]

Ok, do you see any advantage? As I understand it, if the domain is "large enough" the boundary conditions at the box wall don't matter, as long as they cool down the ambient air. Up to now, I just made the domain quite large so I was confident it was large enough.

[FMDenaro]

Originally Posted by RodriguezFatz

Ok, do you see any advantage? As I understand it, if the domain is "large enough" the boundary conditions at the box wall don't matter, as long as they cool down the ambient air. Up to now, I just made the domain quite large so I was confident it was large enough.

It is the same think to consider the wind tunnel walls or the real flow in open air ... it is your physical understanding of the problem to address the quality of your computational setup... Have you any idea of size of the largest flow structures the buoyancy will be driven?