I'm encountering problems with the RSM SSG turbulence model with a 2d-axisymmetric swirling pipe flow simulation. My setup is:

- 5-degree cell in circumferential direction, with rotationally periodic sides. -Good quality hexa grid. -Total Pressure inlets/Static Pressure outlets.

This setup works for similar geometries. For this particular geometry RSM-results (although not fully converging/instationary) have been obtained on a grid of poorer quality (tet mesh with inflation layers running as far to the centerline as possible).

The simulations run well with SST and BSL-EARSM, but RSM-restarts with converged solutions crash almost instantaneously with an overflow error.

I had a look at the residuals to figure out what/where the problem emerges. The flow field quantities don't show singular points ie. diverge massively, however the RSM-residuals diverge immediately to very large values. The strange thing is that the locations where this happens appear to be somewhat random, ie. not at the wall, centerline, refined areas or other critical regions such as around a shock.

Neither grid refinement nor coarsening, larger/smaller physical/local timesteps alleviate the problem. I know RSM is sensitive, but this immediate divergence (crashes in 1 or timesteps) from a fairly good initial guess (starting from scratch doesn't work either), at appearingly random node-locations, puzzles me.

Any ideas would be greatly appreciated. Cheers.

Hi,

RSM is very sensitive to mesh quality. If you are doing a 2D axisymmetric simulation this is a challenge as in CFX you do this by modelling a small wedge (maybe 5°) - but this means you have poor quality elements with only a 5° face angle.

I would: a) Model maybe a 45° wedge as 3D with symmetry/periodic boundaries. It is a much bigger simulation but to get the good quality mesh RSM needs I think you have to.

b) Contact your CFX support rep and request a real 2D model be developed in CFX. It is a major failing of CFX that it does not have a true 2D model, just about every other CFD code does.

Glenn Horrocks

Hi Glenn, thanks for your response.

That's true, but it's not only a quality problem. Refining the mesh near the centerline, such that the elements are of good quality, removes the biggest issue (the instantaneous crash), but leads to non-physical oscillations around the centerline. There are some fundamental problems with RSM in combination with rotationally periodic boundaries, which is acknowledged by CFX.

whether we can do heatpipe analysis in cfx with wick structure and multiphase flow?