CFX Ansys convergence(RMS or Max Residual)

farianka · November 19, 2010, 09:18

Hi
Can anyone explain the convergence criteria for CFX solver Out file.
I have achieved convergence for RMS Mass and Momentum of bellow 1^10-4 (loose convergence), but in 'Judging convergence' section of CFX Solver manual is mentioned that it shoud be bellow 1^10-5, and the previous value should apply to Max residuals.
That would mean, that my simulation has not converged. However, when I browse internet, many people seem to be happy with the RMS falling bellow 1^10-4, not the Max residuals.
Which is correct?

ghorrocks · November 19, 2010, 17:54

Neither. The comments you have read are just generalisations.

You have to do a sensitivity study and see how tight you need for your particular case.

farianka · November 20, 2010, 06:24

Thank you for your reply, Glen.
I have searched the subject of sensitivity analysis, and I understand it means I have to e.g refine mesh and solve again to see if there is any change in convergence.
Meanwhile I have solved with K-epsilon model and achieved convergence with RMS bellow 1^10-5, but not sure if I can trust the result.
I still do not know the convergence criteria for my case, even if I perform sensitivity analysis with refined mesh and achieve let's say "the same convergence".
My analysis case is the analysis of aerodynamic flow around the car's body - same case as in tutorial of Blunt body (Ahmed model), with the aim to calculate drag and lift at high speeds and compare the results to the wind tunnel results. SST or BSL models are the ones to use, but difficult to converge.
Can I trust the K-e model and what criteria for convergence would you recommend?
Thanks a lot

ghorrocks · November 20, 2010, 07:13

SST vs BSL vs K-E - it depends what you are trying to model. They all have strengths and weaknesses. There are many types of flow none of them can handle.

The concept of a sensitivity analysis is simple enough - vary a single parameter and assess its affect on the outputs and select a setting for the parameter based on the accuracy you wish in the output. But in reality it is difficult because you are doing sensitivity on many variables at once, and the variable might not converge nicely.

So you start by setting everything to your best guess, and vary one parameter at a time, and probably come back and do a few iterations until everything is good.

farianka · November 20, 2010, 07:33

Thanks, I will try that. I am currently increasing the elements by 500000, so hopefully that will be the change of parameter.
If I am wrong, please correct me.
Thanks a lot

ghorrocks · November 21, 2010, 06:25

Quote:

I am currently increasing the elements by 500000, so hopefully that will be the change of parameter.
If I am wrong, please correct me.
Thanks a lot

I have no idea what you mean by that, but you seem to misunderstand me. You need to establish convergence sensitivity on ALL parameters - mesh, time step size, convergence, imbalances, any other physical models. If any one model is not converged then your results can be rubbish (even if all the others are fine).

November 19, 2010, 09:18	CFX Ansys convergence(RMS or Max Residual)	#1
farianka New Member Join Date: Oct 2010 Posts: 9 Rep Power: 16	Hi Can anyone explain the convergence criteria for CFX solver Out file. I have achieved convergence for RMS Mass and Momentum of bellow 1^10-4 (loose convergence), but in 'Judging convergence' section of CFX Solver manual is mentioned that it shoud be bellow 1^10-5, and the previous value should apply to Max residuals. That would mean, that my simulation has not converged. However, when I browse internet, many people seem to be happy with the RMS falling bellow 1^10-4, not the Max residuals. Which is correct?

November 20, 2010, 06:24	Sensitivity analysis and convergence	#3
farianka New Member Join Date: Oct 2010 Posts: 9 Rep Power: 16	Thank you for your reply, Glen. I have searched the subject of sensitivity analysis, and I understand it means I have to e.g refine mesh and solve again to see if there is any change in convergence. Meanwhile I have solved with K-epsilon model and achieved convergence with RMS bellow 1^10-5, but not sure if I can trust the result. I still do not know the convergence criteria for my case, even if I perform sensitivity analysis with refined mesh and achieve let's say "the same convergence". My analysis case is the analysis of aerodynamic flow around the car's body - same case as in tutorial of Blunt body (Ahmed model), with the aim to calculate drag and lift at high speeds and compare the results to the wind tunnel results. SST or BSL models are the ones to use, but difficult to converge. Can I trust the K-e model and what criteria for convergence would you recommend? Thanks a lot

November 20, 2010, 07:33	Parameter change	#5
farianka New Member Join Date: Oct 2010 Posts: 9 Rep Power: 16	Thanks, I will try that. I am currently increasing the elements by 500000, so hopefully that will be the change of parameter. If I am wrong, please correct me. Thanks a lot

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
TwoPhaseEulerFoam and Boundary conditions	raagh77	OpenFOAM Running, Solving & CFD	99	February 6, 2018 19:31
SimpleFoam k and epsilon bounded	nedved	OpenFOAM Running, Solving & CFD	16	March 4, 2017 09:30
lift and drag on ship superstructures	vaina74	OpenFOAM Running, Solving & CFD	3	June 8, 2010 13:30
reactingFoam: adiabatic flame temperature	rFkhemek	OpenFOAM	2	July 31, 2009 07:42
TurbFoam problemlarge Co number	sunnysun	OpenFOAM Running, Solving & CFD	6	March 10, 2009 09:05

November 19, 2010, 17:54		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,854 Rep Power: 144	Neither. The comments you have read are just generalisations. You have to do a sensitivity study and see how tight you need for your particular case.

November 20, 2010, 07:13		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,854 Rep Power: 144	SST vs BSL vs K-E - it depends what you are trying to model. They all have strengths and weaknesses. There are many types of flow none of them can handle. The concept of a sensitivity analysis is simple enough - vary a single parameter and assess its affect on the outputs and select a setting for the parameter based on the accuracy you wish in the output. But in reality it is difficult because you are doing sensitivity on many variables at once, and the variable might not converge nicely. So you start by setting everything to your best guess, and vary one parameter at a time, and probably come back and do a few iterations until everything is good.