unable to obtain residual and lift coefficient convergence in SU2
I am trying to simulate a generic fuselage configuration using SU2 (v2.0.8) at M = 0.3. I ran the simulations for around 5000 iterations (with no multigrid, as the use of multigrid was giving a segmentation fault and the solution was blowing off). However, I'm unable to attain satisfactory convergence for residuals (hardly 1-2 orders of magnitude wrt initial values) and the lift does not converge at all but oscillates a lot throughout. I tried to run the simulation with the settings of the SU2 test cases which converged but that didn't work either. For reference I have attached the input file. I have the following questions :
1. What grid quality does the SU2 solver expect? I generated the grid using ICEM-CFD with the min quality of volume elements around 0.11 (and 0.3 and above for surface elements). Is SU2 able to handle very low quality elements ?
2. Is convergence slow in general in SU2?
3. While using multigrid I'm getting a segmentation fault and the solution starts blowing off. Is anyone else facing the same issue? If yes, how could this be rectified.
4. Also, in the log file at various times I get a msg saying n number of non-physical points? What does this non-physical point refer to and how does it arise ?
Would appreciate if someone could have a look at the uploaded files and give any suggestions on these questions. Thanks for your time.
I have no solution for your problems but only ideas to try.
I had some problems of convergence on the NACA 0012 at low speed, doing the computation from scratch. The options that changed the convergence were:
- some multigrid parameters (but it does not concern you since you don't use multigrid)
- reduction of LIMITER_COEFF to 0.05
- use of SLOPE_LIMITER_TURB= VENKATAKRISHNAN
You can find in attachment 3 files: 2 correspond to divergence cases, and one with a real convergence. They were all applied to a partially converged solution (after 15000 iterations)
For this case (NACA 0012) the convergence was slow, but my strategy should probably be improved to get quicker convergence.
Please tell me if you manage to solve your problem, since I am also interested in doing computations like you (aircraft mesh with tetras+prisms cells, which could get quality of roughly 0.1)
Thanks Laurent. I will try it from my end and let you know.
Dear Laurent, I tried your settings for the fuselage configuration but it didn't work. The solution starts blowing off after several iterations. :(
I am exploring other options. Will let you know once I have some success.
|All times are GMT -4. The time now is 16:39.|