Over the past week I have found myself dealing with a stubborn natural convection ANSYS Icepak model with convergence plots that would have been more aptly named divergence plots that looked like this:
In this post I’m going to show you the process I went through to find and fix my problem.
First, a few things to know about Icepak:

Many of the problems associated with your Icepak model are very likely mesh related.

If the bad elements are in a solid, you are probably OK, but if they are in the fluid, watch out!!
So, what is the conclusion? I have a mesh problem.
Second, how do you find the problem?
According to the above “convergence” plot, the continuity equation is diverging (or to my frustrated, onadeadline mind, it was GOING CRAZY). Well, a diverging continuity equation indicates that I have a conservation of mass problem. After consulting with one of my more experienced colleagues, Clinton Smith, he suggested that I do the following to work towards pinpointing the problem:

Plot the gravity direction velocity (in my case, this was Uy)

Look for the Minimum and Maximum Uy locations in the model
Plotting Uy along a cut plane produced this:
As Clinton thought, plotting Uy instantly showed me the section of my model that was producing unphysical results. Next, I looked for the maximum and minimum velocity locations because this would further show me problems.
Next, I need to determine why this area of my model is the problem. Like I said above, it is likely a mesh problem. In the Mesh Control panel under the Quality tab checking the Face alignment values often help to locate very bad elements:
Clicking on the pink block above displays the elements in the graphics window and it was instantly obvious that my problem was due to distorted elements in my area of interest:
When I look at these elements with a perspective of my model geometry I see that the elements are obviously in the fluid domain:
I have found my problem.
Third, how do I fix the problem? Well, the location of my bad elements happens to lie on a CAD body in Icepak. This means that I am limited in my ability to control the mesh on the actual body. So, though there are likely multiple ways that this problem could be solved, I had the idea to create an air block in the area above that I could much more easily control from a meshing perspective. Having a real Icepak primitive in that space would force the mesher to conform to the boundary of the CAD body.
Like I thought, the air block worked!
I should note that in order to get the mesh to conform exactly, I had to put the air block into its own meshedseparately assembly. And now my residuals look much better!
Summary:

Diverging continuity residuals indicate a conservation of mass problem

Plot velocities to locate problem region

Plot min/max velocity to further identify problem

If bad elements are in the fluid region, they must be fixed

Consider creating an air block in the region of interest to more finely control the mesh