Heat flux at plane of symmetry

DaveStupple

Hi all, I've got myself rather confused and I am having trouble thinking this one through. I have a large model (many nodes!) for static thermal analysis of a system where 100 W of heat is injected into a sub-mm volume. The model is a quarter of the whole, as there is symmetry in both the X and Y directions. The images have the centre of the whole towards the viewer, appearing as a corner of the quarter model. Elements are quadratic, and where the heat flux is highest they are hex20.

The temperature distribution looks reasonable (see image), as does the flux magnitude (see image). However, nodes in the plane of X symmetry do not have zero heat flux in the X direction (see image), and likewise with the plane of Y symmetry. I expected to see zero heat flux through the plane of symmetry in these nodes. Of course, Mecway does not know that these are planes of symmetry as I have imposed no special constraints here. I can think of two possibilities:

1. These nodes have an elevated temperature like many other nodes in the model, and as there are other nodes in the X direction with a different temperature, the heat flux will have an X component. This would be exactly balanced by a heat flux in the mirror direction, out of the nodes and into the non-existent (assumed) mirrored part of the model, but of course Mecway doesn't know this. I need not worry. Or...

2. I need to constrain the nodes in these faces to have zero flux through the plane of symmetry.

One other small question: I was puzzled to see negative flux magnitude values at the bottom of the colour scale (magnitude, not X). A quick search did not find any individual nodes with a negative magnitude, so should I not worry about this?

Thanks all, I look forward to your ideas.

Cheers, Dave

Victor

What happens is as you suggested in point 1 - there's a temperature difference across the element and the heat flux is calculated from the nodal temperatures so it ends up being non-zero. This is just discretization error and it would approach zero if you refined the surface like how you refined the top surface. If there were mirrored elements on the other side, their error would cancel out this error and reduce it even further - perhaps to exactly zero. So yes, in principle the software could do this cancelling itself but it doesn't.

The reason for negative magnitudes is that all the field variables are interpolated quadratically in quadratic elements to obtain the max/min on the scale, so there can be a point in an element where the value is less than at any node. Technically it shouldn't do this with magnitudes because they aren't really quadratic but as with the non-zero heat fluxes, it reveals something about the size of the discretization error. The same effect can lead to negative absolute temperatures too.