RSM do not converge
 

Dear All

I am currently using RSM (BSL) to do the simulation of the air flow through a simplified gas turbine combustor. But I cannot get the final results converged to the target of 10^-4.

I have tried the local mesh refinment and complete mesh refinement, but they just do not work.

Could anyone help with this problem ? I really need to cathch up with other classmates.

Kind regards and happy new year

Eric

Re: RSM do not converge
 

Hi,

RSM turbulence models are notoriously hard to converge. Only use it if you REALLY have to. If you must use it then make sure your mesh quality is excellent. Even medium quality elements cause RSM models problems.

Glenn Horrocks

Re: RSM do not converge
 

Hi Glenn Thanks for your kind reply As k-epsilon cannot predict the large recirculation areas correctly, my supervisor requires me to use RSM.

By the way, how do you improve the mesh quality ? I tried to reduce the cell size and do the mesh refinement locally, but they don't work.

Re: RSM do not converge
 

Hi Eric,

Most likely, mesh refinement is not your primary concern. Quality of the mesh, and therefore performance of the RSM model, is greatly determined by

- Aspect ratio and shape of the cells. Avoid skewed elements, ie. aim for a high degree of orthogonality in the cells. A hexa grid generally gives better convergence.

- the smoothness of the mesh (smooth growth of cell sizes). Generally you would like to have the growth factor somewhere around 1.25 .

First try to get a proper solution on a coarse good quality mesh, as outlined above. You can try to start with RSM, but if you find that that doesn't work, use the SST turbulence model (which is way more robust). Assuming RSM doesn't converge initially, start with the SST-solution and try to obtain convergence for the RSM model on the same grid (use local time stepping and play around with the local time stepping factor when convergence seems to get "stuck", use factor 1-5 more or less). The last step would be to proceed to a fine good quality grid, starting from the coarse RSM solution. Just run with RSM model straight away. If that doesn't work, step back to SST and try RSM later.

As mentioned earlier, RSM is really sensitive to the grid, so this is where your problem is most likely to be found. Your global convergence might be determined by just a few bad cells. You can get an idea of where the problematic cells are located, by adding the expert parameter: "output eq residuals=on" and checking the location of the maximum values of the RSM-residuals in post.