[falsc233]

I am trying to run a simple 2D square cavity natural convection simulation. The top and bottom edges are insulated (adiabatic), while the left and right edges have specified temperature.

What I'm trying to figure out is how does CFX handle the boundary conditions at the corner where two different boundary conditions meet. Here I quote the CFX modelling guide:

"For Fixed Temperature walls, the wall temperature is the specified value Tw. For all other heat transfer boundary conditions, the wall temperature is backed out from turbulent wall functions when running a turbulent flow model. For laminar flow modelling the wall temperature is just the local fluid temperature at the vertex adjacent to the wall."

So, at the top right corner node where the adiabatic and fixed temperature boundary conditions meet, would CFX force this node to have the fixed wall temperature or some other value that satisfy the adiabatic condition? From my results, the corner node temperature is not the fixed wall temperature I gave it, and it is not "the local fluid temperature at the vertex adjacent to the wall" as explained in the guide either. I hope someone could clarify for me what exactly CFX is doing in this case. Thanks in advance!

[ghorrocks]

CFX uses control volumes centred around nodes inside the domain, so the boundary conditions are the boundary faces of this control volume. Each face of the control volume can only be one type of boundary condition - and there are no nodes located on the boundary faces - so there is no conflict as you describe.

This is also why you have hybrid and conservative values in the post processor. If you are not aware of this you should read about it in the documentation.

[falsc233]

Originally Posted by ghorrocks

CFX uses control volumes centred around nodes inside the domain, so the boundary conditions are the boundary faces of this control volume. Each face of the control volume can only be one type of boundary condition - and there are no nodes located on the boundary faces - so there is no conflict as you describe.

This is also why you have hybrid and conservative values in the post processor. If you are not aware of this you should read about it in the documentation.

Thank you for your reply! Would you please clarify what you mean by "there are no nodes located on the boundary faces"? I'm a bit confused because during meshing I can see the nodes (vertices of the elements) on the boundary faces as well as at the corner, so is the picture I attached not a correct representation of the mesh discretization? If so, how does the CFX discretization look like for a 2D domain with square elements? Thank you!

[Opaque]

Your drawing is spot-on for the nodes. However, it is a matter of interpretation and vocabulary.

In your 2D drawing, faces are lines, correct?. Then, the nodes are NOT on any of the faces, they are at the end of the edges. ON the faces means anywhere along the line.

My advice is to read the documentation and have a clear understanding of the fundamental steps in the control volume discretization approach. Differentiate what happens in a volume integral, and how BC comes into play when discretizing and interpreting what those BC represent.

Hint: nodes are just placeholders to store data, and do NOT represent a specific location in space for that data.

[falsc233]

Originally Posted by Opaque

Your drawing is spot-on for the nodes. However, it is a matter of interpretation and vocabulary.

In your 2D drawing, faces are lines, correct?. Then, the nodes are NOT on any of the faces, they are at the end of the edges. ON the faces means anywhere along the line.

My advice is to read the documentation and have a clear understanding of the fundamental steps in the control volume discretization approach. Differentiate what happens in a volume integral, and how BC comes into play when discretizing and interpreting what those BC represent.

Hint: nodes are just placeholders to store data, and do NOT represent a specific location in space for that data.

Thank you very much for your reply! I think now I have a better idea of what my question really is. What I'm seeing could be a combination of multiple issues:
1. Due to low grid resolution and the high temperature gradient, the adiabatic wall temperature at the corner CV is a bit different from the fixed wall temperature I specified.
2. CFD-Post uses hybrid values (user-specified boundary values) to show the boundary temperature at the corner CV/node, which I think is either the fixed wall temperature or the calculated adiabatic wall temperature?

So, my understanding is that under low Rayleigh number when the temperature gradient isn't too high at the corner, the two boundary values are close enough and either one is basically the same as the fixed temperature I specified. However, at high Ra, CFD-post might be using the adiabatic wall temperature which is slightly higher than the fixed wall temperature I gave it, causing all the confusion.

So, I think now my question becomes how does CFD-post select which boundary value to use for the hybrid value at the corner CV/node? I've found that there might be a hierarchical order in choosing the boundary values: https://ansyskm.ansys.com/forums/top...aring-an-edge/. But for my case when the two boundaries are both "walls", what would the order be then?

[Opaque]

The arithmetic average for boundaries of the same physical type.

You can verify that by setting a problem with 300 [K] on wall 1, and 500 [K] on wall 2, the corner node should be @400 [K]

[falsc233]

Originally Posted by Opaque

The arithmetic average for boundaries of the same physical type.

You can verify that by setting a problem with 300 [K] on wall 1, and 500 [K] on wall 2, the corner node should be @400 [K]

Thank you, Opaque!