[fatihyaman661]

I am doing a compressor analysis using NUMECA for a company. The work is confidential so I cannot share much detail. The funny thing is that the very same analysis converges sometimes and diverges at other times. But I don't change anything at all. The mesh, the test setup, the model, the boundary conditions are all the same. I just hit run button twice. And the result changes in two.




I use k-w wilcox model and SST model, and both of them show this behaviour. The problem doesn't persist with SA model. I run the analysis multicore on an HPC system that runs on SLURM.




When it diverges; firstly, the residuals act like the way when they converge. They decrease gradually. However, after a point it gives the density negative error and the analysis stops.




What I don't understand is not that the analysis diverges. It is the fact that it converges sometimes and diverges sometimes. Does anyone have any idea why such an indeterministic behaviour may occur?

[colinda1]

Hi,
Did you already create a support case on support.cadence.com? For such industrial consultancy project I would recommend to do that to make sure to get an expert advice soon.

When you write about "the residuals", do you also look a the residuals for the turbulence quantities? Di you save the solution before divergence and have a look at it to see whether there are any non-physical values for pressure and temperature in the domain?

Best regards,
Colinda

[Hamidzoka]

Originally Posted by fatihyaman661

I am doing a compressor analysis using NUMECA for a company. The work is confidential so I cannot share much detail. The funny thing is that the very same analysis converges sometimes and diverges at other times. But I don't change anything at all. The mesh, the test setup, the model, the boundary conditions are all the same. I just hit run button twice. And the result changes in two.




I use k-w wilcox model and SST model, and both of them show this behaviour. The problem doesn't persist with SA model. I run the analysis multicore on an HPC system that runs on SLURM.




When it diverges; firstly, the residuals act like the way when they converge. They decrease gradually. However, after a point it gives the density negative error and the analysis stops.




What I don't understand is not that the analysis diverges. It is the fact that it converges sometimes and diverges sometimes. Does anyone have any idea why such an indeterministic behaviour may occur?

Hi,
The SA model has the least sensitivity to mesh quality of the domain. However, in case of SST turbilence model, convergence needs really high quality meshes in this software. So, first make sure that you have properly set boundary conditions for your case and then try to maximise the grid quality.
Last but not least, SA gives acceptable results for turbomachinery aerodynamic simulations. So, you can proceed with that in case SST does not converge.

Regards