- **FLUENT**
(*https://www.cfd-online.com/Forums/fluent/*)

- - **Compressor Endface Leakage - Convergence Problems**
(*https://www.cfd-online.com/Forums/fluent/221290-compressor-endface-leakage-convergence-problems.html*)

Compressor Endface Leakage - Convergence Problems5 Attachment(s)
Hi all,
I am trying to simulate end face leakage for a swing vane compressor, and I am facing problems with 2nd order convergence. The problem is a compressible turbulent flow driven by a strong pressure gradient. I have started the simulation with first order upwind for all variables and I am able to obtain deeply converged results (residuals of 1e-6), following which I switched the solution methods to second order. The residuals then spike up and oscillate at the same value with no signs of decreasing. I then tried to change from 1st to 2nd order for the variables one by one, and found that I am able to obtain 2nd order convergence for all variables except for momentum! The screenshot of the residuals are attached below. I am using SST-k omega model with the COUPLED solver and NIST real gas model. The geometry of my problem and the structured mesh that I have created is attached as well. The complexity of the problem is largely due to the geometry -- the end face clearance is extremely small at 30 microns, however there is a depression in rotor end face due to the bearings which results in a 1.9mm thick depression groove. This causes problems for my mesh as I am using the student license -- to keep the cell count under 512k, the maximum aspect ratio is 260 along the thin end face clearance layer. Minimum orthogonal quality is 0.98. I have also attempted to manually create inflation layers near the walls by using edge biasing (I was unable to use the inflation function and keep my structured mesh as well). Would it be likely that the convergence problem be due to my mesh and the poor aspect ratio? Also, checking my wall y+ values I see that it ranges from 0.1 all the way to 320. From what I know this is not acceptable and would impact the solution accuracy, but would it cause convergence problems? I have tried using other models (like k-epsilon) and tried reducing under-relaxation factors, all to no avail. If anyone has any ideas or inputs please enlighten me! Thank you very much in advance. |

All times are GMT -4. The time now is 18:38. |