[Dónal]

Hello,

I'm running a 3D compressible (M=2.5) RANS simulation in SU2 version 7.0.0 and the results I am getting are not smooth. The smoothness is clearly linked to the mesh, and it appears that the issue may be with the calculation between cell edges and cell centres. Slices of the simulation with the x component of vorticity are attached for clarity, along with my config file

I have tried different convective schemes and got similar results on each. A similar effect is apparent in both unstructured and voxel type meshes, however it is more apparent in voxel meshes due to their regularity. I have also tried different slope limiters and turning off the slope limiter and seen no change in this effect. I should note that the best convergence I have managed to get with this solution is residuals almost at 10^-3 and less than 1% change in results over 1000s of iterations.

I am also having trouble running the same cases in version 7.2.0, but this is a separate issue and I will put this in a separate post.

Any ideas on what may be causing this or other things to try?

MeshEffectVoxel.jpg

MeshEffectVoxelwithEdges.jpg

MeshEffectDelaunay.jpg

closeup.jpg

Medium.txt

[pcg]

With only 3 orders of magnitude convergence it's hard to say whether the problem is from convergence or from the numerical scheme not liking the flow conditions.
With AUSMPLUSUP(2) and SLAU(2) you try using option USE_ACCURATE_FLUX_JACOBIANS=YES to improve convergence.

[Dónal]

Thanks for your reply Pedro. I get a similar results from a first order simulation with residuals of 10^-4 and accurate flux jacobians on

First order medium.PNG

residuals are still decreasing on this one so I can continue running further

[Dónal]

I am still getting a similar effect for first order results with better convergence statistics

I have tried refining the mesh around the geometry features that are giving poor mesh quality, running in first order and using Lax-Friedrich convective scheme as it's more dissipative, just to try and get some initial results. While I can get to 10^-5 and beyond with, these the mesh effect is still visible.

The next things I am trying are:
- URANS. A steady state solution seemed possible from the experimental results this work is based off, but this may not be the case

- Increasing the domain length. Currently using a Riemann boundary condition at the outlet, but previously had more trouble using the supersonic outlet condition which I believe were related to the vortices and shockwaves interacting with the outlet condition. The longer domain will hopefully remove/reduce this

Any other suggestions would be greatly appreciated, I am running out of things to try