ICEM vs Solver
 

As the solver runs through, it will tell you the three main mesh constraints (aspect ratio, etc.). As I make a mesh in ICEM, I try to keep an eye on these topics to make sure I'm not making a monster that won't converge. Recently, I noticed that ICEM will say my max aspect ratio is around 76 or so. The solver says its around 618. This is kind of confusing to me. Am I missing something blatant? Am I not setting something up right?

-W

Re: ICEM vs Solver
 

Hi Wooster,

The solver one rules. It is based on the way CFX actually does the discretization and is the most relevant to error generation. You can find a description of what the solver is actually calculating in the documentation in ANSYS CFX-Solver Modeling Guide>Advice on Flow Modeling>Mesh Issues>Measure of Mesh Quality.

The difference arise because CFX is calculating quality criteria of it's control volumes. The mesh aspect ratio is actually calculated as the ratio of the maximim to minimum sub-face areas (because this is what actually leads to discretization errors when aspect ratio's are high). If you have a very large expansion from one element to the next, this could lead to high aspect ratio's when calculated in this manner.

-CycLone

Re: ICEM vs Solver
 

Thanks Cyclone.

That's makes sense. I do have a model with high mesh expansions. Still, I'm trying to figure out how to better gage (or gauge?) the mesh properties so I can fix things before I go through the time and trouble to put them into the CFX triad. I'm getting better at getting those things right, but I'm not expert class yet.

-W

Re: ICEM vs Solver
 

A quick way to use the solver to check the mesh without setting up the whole thing is to load the mesh into a new session and run a single iteration. You could easily create a batch script to do this as well.

-CycLone

Re: ICEM vs Solver
 

I need to learn how to make batch scripts. That's something that needs to go on my to-do list.

usually if I have doubts, I use the post processor to look at a mesh, but using the solver seems reasonable as well.

One problem I face, however, is the time it takes to convert my models to .gtm. Sometimes it takes >10min on a Win64 machine. So I'm always looking for quick methods on large meshes (>5mil nodes). Thanks!

-W