[Hing]

Hey guys,

I have a CFX conjugate heat transfer simulation with a similar geometry to a standard ice box. I have a isothermal block inside the container (at the bottom) which touches the walls of the container and a known amount of trapped air contained above the block inside the container.

It is a buoyancy driven flow with a Rayleigh number of about 10^6 and so I am using the laminar model. The max temperature difference is below 20 degrees and so I am using the Boussinesq model. The mesh is not sensitive and does not affect the surface temperatures by more than half a degree. I am getting convergence to 10^-6 in a few hundred iterations.

The problem I am having is that the temperature of the air inside the container is about 5 degrees higher than what I would expect. The simulation is being done as steady state (rightfully so?). Should I play with the time steps? I am clutching at straws. The boundary conditions seems correct and the material properties have been checked multiple times.

[Hing]

I think have found the root of my problem and I believe it is radiation. Without radiation, I was relying purely on the conduction of the inside air (convection inside is very low) and the natural convection on the outside. So I have tried to get an understanding of the radiation models available and I have turned on the P1 model for the inside and outside air domains. The average temperature inside has dropped to 3.3 degrees, which is much much more realistic IMO.

However I want to understand more about the model. From previous literature in a similar area it seems that people have used a surface-to-surface radiation model in FLUENT. Is this done by using the Monte-Carlo radiation model in the solid container domain? At the moment I do not have a radiation model turned on for the container (solid) because it seems I don't have the stacked memory available to even run the simulation (have tried increasing the allocation).

In the CFX theory guide it only includes a very small section on the P1 model and so I am going to pick up the Siegel and Howell Radiation book from the library today. Does anyone have any experience using these models? Is it valid to use the P1 model for the two fluid domains and nothing in the solid domain? It seems from the CFX guide that the P1 model does have wall interaction but I am unsure of what it is actually doing?

[ghorrocks]

Your original question is an FAQ: http://www.cfd-online.com/Wiki/Ansys..._inaccurate.3F

You need to have a careful look at the CFX documentation on radiation modelling. The P1 model is only applicable to optically reasonably opaque materials. The models applicable to what most people think of as a radiation model (ie through optically almost transparent materials) is the Discrete Transfer and Monte Carlo models. Have a look at the description of both these models in particular - Discrete Transfer is the one most simple radiation applications require.

[Hing]

Hey Glenn, thanks for your reply. It is interesting, running the P1 model, the results were pretty accurate compared to the experimental. I did read through documentation about optical thickness figured it may not be applicable however the annoying thing is the result are well within the bound of error of my experimental results.

I cannot seem to get either the monte carlo or the discrete transfer model to run. I keep getting the following error:


Domain Name : Inside Air

ERROR: Element(s) with open faces.
Details of first element below.
...
..
Fatal error generated in gKElEl
Message :- Some boundary elements may be missing BCs.

gKElEl called by :- SU_RADFLOW_ZONE


Any ideas on this one?

[ghorrocks]

Don't be fooled by an inappropriate model making the parameter you are looking at about right. Everything else will be wrong and the results will be meaningless.

The decision as to what radiation model to use is determined based on the nature of the flow (how optically thick, whether the radiation predominantly causes face to face heat transfer etc).

I am not familiar with the error, but it sounds like you have a problem with your set up. I would do the set up again and make sure you do not miss something. IN this case it sounds like you have missed the radiation setup on a boundary condition.

[Hing]

Yes Glenn, I agree. However I read the following excerpt in the CFX documentation and it made me think that if my results were accurate (ie. the entire model, not one parameter) that it may be valid to use the P1 model.

"The P1 model is valid for an optical thickness greater than 1. For example, the model has proved adequate for the study of pulverized fuel (PF) flames, in regions away from the immediate vicinity of the flame. However, it has been used for lower values with varying success."



In any case, the problem with the Monte Carlo and Discrete methods is down to the inside air cavity but the geometry is simply a cube and all boundaries conditions have been set and there are no errors or warnings in CFX-pre. Oh well, I guess I will keep trying.

[ghorrocks]

What is the fluid the radiation is passing through? How opaque is it?

[Hing]

I feel like that might have been rhetorical question but it is stagnate air trapped between two surfaces, the bottom one at -2 degrees and the top one at an unknown temperature somewhere between 0 and 15 degrees.

[ghorrocks]

No, that was not a rhetorical question. I cannot recall you ever saying what the fluid is, and if I guess I will no doubt get it wrong.

For most cases air can be considered optically transparent, that is optical thickness = 0. This means the P1 approach is not suitable, you will need Discrete Transfer or Monte Carlo.

[Hing]

Yup I figured as much. I will keep trying to get one of those to run. Thanks a lot for your help mate. Much appreciated.

[Hing]

Just in case anyone else gets a similar error in the future, it was some sort of bug in the mesh. I changed the minimum element size in the mesh by a negligible amount and re-meshed. The simulation then solved correctly with no issues.

Glenn, the results using the Discrete Transfer method are looking good. Once again, thanks for your help.