[theflyingfury]

Hello Everybody,

I am kind of new to CFD. I am working on a project where I am doing steady state analysis with K-epsilon realizable model. So since i run many iterations of these kind how will I check whether the solution has converged in Fluent. I set monitors for my output also. But how will I decide that the solution is converged? Please help . Thank you in advance.

[godeny_mecaer]

It depends on how accurate your project needs to be. If you set a value on residuals monitor and, during the iterations, the residuals reach that value, the solution is converged. Note that the smaller the value you set on residuals monitor, more iterations will be calculated.

[theflyingfury]

Originally Posted by godeny_mecaer

It depends on how accurate your project needs to be. If you set a value on residuals monitor and, during the iterations, the residuals reach that value, the solution is converged. Note that the smaller the value you set on residuals monitor, more iterations will be calculated.

hello Fernando, Thank you for replying. Let me clear something. You mean the in Monitors -->residuals--> I have to tick convergence and specify the accuracy for which it should say solution is converged.

If yes, there is one more question I have here. Usually I will put up my case and data files into the server to solve. In this case how can I verify the convergence. Is there any way to write the residual values in a file so that I can verify it later after I retrieve from the server???

Thank you in advance.

[godeny_mecaer]

In this case, I don't know how exactly to generate a file with residuals data since I run my simulations directly on my machine and check the convergence on the residuals plot showed during the calculations (see attached image).

Maybe one way is to check how many iterations you set and how many the solver used to complete the calculations.

Sorry if this seems too obvious but I can't figure now a better answer.

Regards.

[theflyingfury]

Originally Posted by godeny_mecaer

In this case, I don't know how exactly to generate a file with residuals data since I run my simulations directly on my machine and check the convergence on the residuals plot showed during the calculations (see attached image).

Maybe one way is to check how many iterations you set and how many the solver used to complete the calculations.

Sorry if this seems too obvious but I can't figure now a better answer.

Regards.

Thank you @godeny_mecaer for your help. But again can you let me know what is residual concept clearly?? How do you judge by seeing the residuals graph and decide the convergence?

[LuckyTran]

I disagree with the residual concept as a measure of convergence. Residuals are a measure of tolerance and how well the governing equations are satisfied, they don't tell you if the iterative convergence is achieved.

To check iterative convergence, you should monitor raw solution values like pressure, density, temperature, x-velocity, y-velocity, z-velocity and see whether asymptotic convergence is obtained.

However, the convergence criteria is up to the user and each user as their own preferences. I.e. many users run 100 iterations and check if their solution is converged, whereas I have a personal preference for monitoring my solution over 10,000 iterations at a minimum in order to determine the amplitude of the low frequency oscillations in the solution. The finer your grid, the higher the likelihood of low frequency oscillations due to non-linearities. So you run it for some "#" number of iterations and get some amplitude. Then you must simply decide if that amplitude is reasonable or not. I.e. how many significant digits do you want it to be converged to? Ultimately however, for a given grid and settings, there's not much you can do to change the result.

Hence, you don't necessarily need a criteria defined beforehand. You can simply check the asymptotic behavior and take whatever amplitude comes out in the same way that experimenters accept whatever errors are in their measurement; you simply need to report whatever errors there are.

[theflyingfury]

Originally Posted by LuckyTran

I disagree with the residual concept as a measure of convergence. Residuals are a measure of tolerance and how well the governing equations are satisfied, they don't tell you if the iterative convergence is achieved.

To check iterative convergence, you should monitor raw solution values like pressure, density, temperature, x-velocity, y-velocity, z-velocity and see whether asymptotic convergence is obtained.

However, the convergence criteria is up to the user and each user as their own preferences. I.e. many users run 100 iterations and check if their solution is converged, whereas I have a personal preference for monitoring my solution over 10,000 iterations at a minimum in order to determine the amplitude of the low frequency oscillations in the solution. The finer your grid, the higher the likelihood of low frequency oscillations due to non-linearities. So you run it for some "#" number of iterations and get some amplitude. Then you must simply decide if that amplitude is reasonable or not. I.e. how many significant digits do you want it to be converged to? Ultimately however, for a given grid and settings, there's not much you can do to change the result.

Hence, you don't necessarily need a criteria defined beforehand. You can simply check the asymptotic behavior and take whatever amplitude comes out in the same way that experimenters accept whatever errors are in their measurement; you simply need to report whatever errors there are.

Thank you @Lucky tran for your reply. Ok I did as you said. For my simulation I recorded my output monitors and plotted them in a sheet. I have performed 1000 iterations and I observe that the outputs values are going in a form of sine wave (values varying between 47.5 and 48.5). So here as far as I have understood, I should consider the mean value i.e 48 as the solution???

[LuckyTran]

48 +/- 0.5.

[theflyingfury]

Originally Posted by LuckyTran

48 +/- 0.5.

Thank you very much @LuckyTran .