Adaptive meshing percentages meanings - in absolute and relative terms

[milos28HR3]

I'm going through the docs on adaptive meshing and I'm struggling to get a bit more definitive answer on what the percentages in the results mean.

By this I mean, if I went the manual route I'd run a simulation, refine the mesh by 30%, run the sim again, check if certain parameters in the results are within the 5% of the initial results. Easy enough to understand, time consuming and laborious to do manually.

Now, from the adaptive meshing help files:

"Mesh independence: Pressure: 85% Velocity 98% Temperature 97%

These values indicate how converged your simulation is for each quantity. Higher values indicate less sensitivity to the mesh, which means your solution is approaching mesh independence."

I understand this part and it makes sense, nevertheless it's too vague in context of making decisions about the accuracy of the simulation.

Say I have a simulation where I ran the solver once for 250 steps, then enabled the mesh adaptation (3 cycles, 1000 steps each). The first cycle finishes before the 1000th step and I get the message:

Mesh independence: Pressure: 97.23%, Velocity: 97.07%

I understand that the percentages are high, and that they indicate less sensitivity to the mesh, and that solution is approaching mesh independence. But what do they mean exactly, are my results mesh independent or not? Is the only way to be certain to go to 100%? Am I looking at how far are these percentages from 100%? When it says that pressure results are 97.23% mesh independent, what is in the remaining 2.77%?

After the second cycle I get:

Mesh independence: Pressure: 97.89%, Velocity: 97.17%

I understand that my results are even less sensitive to the mesh, and the solution approaches mesh independence even closer. But again, how much better is this exactly? Is it good enough, is it excessively precise, etc.?

Comparing the first pair of percentages to the second one we see increase of 0.67% for pressure and 0.10% for velocity. Do these two relative differences mean anything or should I just observe the absolute value?

And so on for the remaining cycles.

I guess where I'm coming from is, it's convenient to set up adaptive meshing and let it run for 3* cycles but if anyone asks me if my results are mesh independent all I can answer after reading the docs is something vague: they are more close, less sensitive, etc.

Any help in understanding this would be deeply appreciated.

* Also, perhaps related - why three cycles? Why not 2 or 4? How can I tell if Autodesk CFD has refined the mesh enough after the 3 cycles so that the results are mesh independent?

[KubliJ · Accepted solution]

Hi @milos28HR3 

Welcome to the fun of CFD. It isn't really about know if your results are right, but more so about knowing if they are wrong.

The adaptive mesh convergence percentages for velocity and pressure are global change comparisons from one adaptive mesh to another. So that is why you may see very small changes in the percentages from one mesh iteration to another. The problem here is that overall your model may be converged, but you may have very localized areas experience mesh dependency. And depending on where that is at in your model, may have a dramatic effect on the results you are looking for.

Mesh adaptation is also dependent on if you gotten enough analysis iterations for convergence. Our automatic mesh convergence again only works on global results, so sometimes it has shut down a good meshed model too soon.

Models with big domains such as building modeling, natural convection and external flow are all dramatically hampered by these issues due to the large domain, but small areas of interest.

I've been doing this for a decade and all I can tell you is how to prove your model is wrong. Once you do that, and can't prove it wrong, then you assume it is probably right. And your "I can answer after reading the docs is something vague: they are more close, less sensitive, etc" is 100% correct.

Thanks,

James

[milos28HR3]

James,

Thank you for your reply.

Fair enough.  I appreciate the inputs and the clarifications nevertheless, especially the part discussing which variables are monitored.

My last cycle resulted in both pressure and velocity reaching more than 99% independence each, so in the spirit of this discussion - let anyone prove it is not accurate 🙂

Thanks again and have a nice one.