Saving Heat Flow using Savescalars on internal boundary

[aminer2k]

Hello,
Thank you all for the assistance in the past.

I am trying to extract the heat flow (W) from boundaries between domains that are internal to the model. Is there an accurate way to output the heat flow from internal boundaries? The topic of getting heat flow from boundaries has been discussed before, and Peter described two methods to do so in a post on 05 Oct 2009 10:49 titled Generating gradients. I have been try
Method 1: uses savescalars to generate temperature loads, which are summed and saved.
Method 2: uses savescalers to compute the normal flux.

Method 1 outputs only 0.000... for the flux values, no matter what I adjust. Code used for savescalars for this case:

Code: Select all

Solver 3
  Procedure = "SaveData" "SaveScalars" 
  Filename = case.csv
  Operator 1 = boundary sum
  Variable 1 = Temperature Loads
End

Method 2 results in good data only in certain situations. I have made a simple model to illustrate the issue. The model is a 3d rectangular solid with 4 domains, or quadrants (top1, top2, bot1, bot2). When all quadrants are the same material, the heat flow results from save scalars are good. When the top is all Cu, and the bottom is all glass, the results are good. In other situations the outputs are unreasonable. For example, in case 4 in the results in the attachment, 82% mopre heat flows through the center of the object than enters at the hot side. Code and images are attached. Example results are also attached, showing which cased result in energy conservation, and which do not.

case.sif

(3.79 KiB) Downloaded 393 times

[raback]

Hi

1) The flux is continous over internal boundaries and hence the residual of the matrix equation is zero. This method is not really applicable to internal boundaries.

2) Perhaps you could try to increase the element order, or improve the mesh density towards the boundary. FEM does not conserve flux by construction (as does FVM) and therefore you cannot expect perfact match. Of course 87% is rather poor accuracy. How does this behave as a function of mesh density?

You may also determine in the BC from which side the flux is computed. There is a keyword "Flux Integrate Body" in the BC section that may be used to give the correct body id. Otherwise the first parent is used and depending on how your mesh was created it may not always be consistant.

-Peter

[aminer2k]

Hello,
Thank you for your help. I ran several runs to look at your recommendations. I attached a screen shot of the results. Run number 8 is a baseline run.

a. the use of 'Flux Integrate Body = X'
Run 9,10, and 11, as compared to run 8 show the effect of using 'Flux Integrate' in the boundary specifications. These runs take look at specifying different combinations of bodies that are adjacent to the boundaries in question. The heat flow result changes, but does not improve, and does not illuminate the problem for me.

b. mesh convergence.
Runs 12,13, and 14 illustrate the effect of mesh density, with respect to run 11. As the mesh density increases, the flux calculated at the outside surface goes from 60, then 53, then 48 watts. The internal surface heat flows do not however converge to reasonable values that conserve heat flow. Heat flow in collumns mid1 and mid2 must be equal, and thier sum must equal bot1 and bot2.

The model and mesh are made in Salome using netgen mesh (as part of CAE Linux). 'max element size' is used to control mesh density. It was .01, .05 and .025 for the mesh convergence studies here.
Any guidance would be appreciated as far as how to monitor the heat flow across an internal surface.
Thank you,
Andy