Hello, we are trying to model heat transfer appearing at heat sinks for electronic equipment (both, natural and forced convection). As reported in the literature, for correct heat transfer calculation the boundary layer has to be modeled with a certain number of nodes within the layer. This leads to huge grids if many sinks participate in the calculation. How many grid points are enough for sufficient heat transfer calculation (e.g. depending on the Reynolds or Rayleigh number)? Is there any other way to keep the grid small while having good heat transfer results? Our calculations were carried out in Fluent and CFX but this question seems to be more general since the convection models do not differ that much for different CFD codes.

Thanks in advance.

Hi, Jan; If you already have an initial simulation, you already have the answer. You just have a geometry problem. All you need to do is remesh the geometry where you need a large number of nodes and elements. I know TGrid will do this. I'm not sure about Gambit. Or you can use another meshing program and just export it to Gambit. -Edward

Hello Edward, thanks for your answer. Yes, the answer ,in principle, is written down in all textbooks. But correct boundary layer resolution means many grid points. This is not applicable to complex geometries. So the question should be updated as follows: Does anyone have any experience in the resulting errors comming from insufficient resolutions of the boundary layer? I dont know exactly, but any flow inside enclosures containing electronic components should have this problem. How do they solve it? Are there any models (like turbulent wall treatment) to get correct or nearly correct heat transfer?

Thanks. Jan

The courser grid spacing then the thermal boundary layer thickness affects the results significantly, specially at high Prandtl/Peclet numbers where the thickness goes to very low. I think there is no relation available for the thermal boundary layer thickness for complex geometries. A simple assumption of the heat transfer from a flat plate can give a rough estimate of the thermal boundary layer thickness and refine the result. The correlation relating the thermal BL thickness and Prandtl number can be found in any std. textbooks. I think 10-15 grid points inside the thermal BL will do a good job with not so high grid elements.

Hi, Jan; The geometry refinement is the only thing I can suggest to you to resolve the boundary layer(quickly solve and yet tedious). I think it can be done on your geometry. As for the residuals of your simulation, that is just a check of how good your model is. You could try another CFD model(LES?) in your case which is like opening a new can of worms. -Edward