Pressure Drop Not Converging as Mesh Fineness Increases

Hi! I have been running into an issue with my simulation runs. I am trying to see how the fineness of a mesh affects the pressure drop of air flowing through a cylinder at 60 LPM (0.001 m^3/s). It should converge on the true pressure drop as the mesh increases from very coarse to very fine. I am using Hex-Dominant mesh and increasing the finenesses (very coarse → very fine). When I have run simulations keeping everything the same besides the mesh fineness, the pressure drop is not converging on a value and is instead fluctuating (sometimes pretty significantly). I checked to make sure that the residuals for each run converged and they all did. The values I got when I ran the simulations with the automatic boundary layer on were:

Very Coarse: 27.481 Pa
Coarse: 24.207 Pa
Moderate: 20.749 Pa
Fine: 21.635 Pa
Very Fine: 24.874 Pa

I tried doing runs with and without the automatic boundary layer turned on, but that did not converge on a pressure drop either. I have included screenshots of the average pressure at the inlet with the automatic boundary layer turned on (the outlet pressure is 0 Pa) and a link to the project below. Any advice or help would be super helpful, thanks!

Very Coarse:




Very Fine:

Project Link:

Hi, thanks for posting on forums!

Before any comments about meshes, I noticed that you are running a simple internal pipe flow. Have you already done a hand calc for the pressure drop? (e.g. using the Darcy-Weisbach formulation).

If you have a value to compare your results with, then everything is easier (e.g. see this validation case).

As far as meshes go, I noticed that the coarser meshes are very (very) coarse, with the first one starting at less than 10k cells. It’s likely that the mesh in this region is far away from the asymptotic convergence region.

For sensitivity studies, it’s a good idea to refine the domain evenly, which is much easier to achieve with a hex-dominant (parametric) mesh. The hex dominant tool is likely concentrating new cells around the edges/walls of the domain, which is not optimal.

I’ll link a doc that talks about the GCI, which is one of the grid convergence methodologies used. It should provide you more insights on this topic.


1 Like