[arjun3020]

Hi all,

I did the analysis of Cyclone separators and I found monitor of mass flow rate at outlet is fluctuating.
please see the attachment.
Is the solution converged?

[LuckyTran]

Originally Posted by arjun3020

Hi all,

I did the analysis of Cyclone separators and I found monitor of mass flow rate at outlet is fluctuating.
please see the attachment.
Is the solution converged?

Sorry but, please define solution convergence and that will answer your question. What are you considering a converged solution? What are your convergence criteria? No one else but you should be telling you if your solution is converged, since it is up to you to determine the convergence criteria.

[jpo]

There are fluctuations, but they seem very small within this range

I would look for the following in a converged simulation:

1) mass is conserved (e.g. same mass flux coming in and exiting the domain)

2) generally speaking, all residuals decrease smoothly below some low value

3) if no energy is added to the flow, total temperature on all exits should add up to the total inlet temperature, meaning (m_dot_in * T_in = m_dot_1 * T_1 + m_dot_2 * T_2 + ... )

4) mass/temperature/velocity etc outlet monitors should occupy a stable numerical value when plotted over a sizeable range of iterations

Hope this helps

[jpo]

On the image m_dot has been plotted over the last 90 cycles. It is best for one to plot a monitor over a large range of cycles (in your case, say, 3000 - 4000, as an example) in order to see how this monitor has been behaving throughout a sizeable part of the simulation.

[arjun3020]

Thanks for your help,

My residuals are fluctuating in cyclic manner,
from one research paper i found that the solution must be changed to transient one.
i dint know how to decide the time step and if i run transient run then how should i know that my solution is converged, and how to analyse the final results of transient run.


Thank You...

[LuckyTran]

Originally Posted by arjun3020

Thanks for your help,

My residuals are fluctuating in cyclic manner,
from one research paper i found that the solution must be changed to transient one.
i dint know how to decide the time step and if i run transient run then how should i know that my solution is converged, and how to analyse the final results of transient run.


Thank You...

arjun, do not take it lightly, but my greatest suggestion to you is to first decide how you are going to judge convergence before anything. You need to begin the problem with an end in mind or else you will never be able to decide when to stop. Again, you are the one that ultimately decides convergence! If you can never decide then you will never finish. You do not want to blindly follow others' advice on this, you need to convince yourself!

For transient run, your time-step should be small enough to resolve the main flow structures (the unsteady features). If the unsteady feature has a time-scale of say 1s, then some fraction of 1s is needed or you will not resolve the unsteady features. Less than 1/3 is recommended, 1/10 or more if you want to be safe. Again, you need to know beforehand what the time-scale of these features are, do not do it blindly or you will only get lost! If you don't know these time-scales then you probably do not need to be doing a transient simulation yet.

Your solution must be converged at the current time-step before you can move to the next one. Hence, it is even more important that you define convergence well beforehand. If you find that it is taking too many iterations to converge at the current time-step, you can reduce your time-step size. You will end up computing more time-steps but it will make it easier to converge each time-step. If your solution converges easily then consider increasing the time-step size (as long as you can still resolve the unsteady features using the 1/3 or 1/10 guideline).

Additionally, typically the time-step is also sometimes chosen to satisfy stability criteria (check-out CFL number).