CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   convergence unsteady state (http://www.cfd-online.com/Forums/fluent/98861-convergence-unsteady-state.html)

cfdhydraulic March 20, 2012 20:01

convergence unsteady state
 
Hi everyone!
I am studying a transient state study (unsteady).
The first iteration is correct and it achieves a perfect convergence but, the second one... it doesn't converge at all
it's that a problem?? is there anything I should change?
I decreased the time step from 1E-4 to 1E-05 but it didn't converge whatsoever

any idea??

thank you :)

cfdhydraulic March 20, 2012 20:03

convergence/divergence unsteady state
 
1 Attachment(s)
I post a picture so you can figure it out
Thanks in advance


Hi everyone!
I am studying a transient state study (unsteady).
The first iteration is correct and it achieves a perfect convergence but, the second one... it doesn't converge at all
it's that a problem?? is there anything I should change?
I decreased the time step from 1E-4 to 1E-05 but it didn't converge whatsoever

any idea??

thank you :)

LuckyTran March 21, 2012 14:54

Quote:

Originally Posted by cfdhydraulic (Post 350528)
I post a picture so you can figure it out
Thanks in advance


Hi everyone!
I am studying a transient state study (unsteady).
The first iteration is correct and it achieves a perfect convergence but, the second one... it doesn't converge at all
it's that a problem?? is there anything I should change?
I decreased the time step from 1E-4 to 1E-05 but it didn't converge whatsoever

any idea??

thank you :)

It's working right. Your definition of "convergence" is the problem.

From 30-40 you can clearly see the solution convergence.

Zigainer March 21, 2012 15:01

Quote:

Originally Posted by cfdhydraulic (Post 350528)
I post a picture so you can figure it out
Thanks in advance


Hi everyone!
I am studying a transient state study (unsteady).
The first iteration is correct and it achieves a perfect convergence but, the second one... it doesn't converge at all
it's that a problem?? is there anything I should change?
I decreased the time step from 1E-4 to 1E-05 but it didn't converge whatsoever

any idea??

thank you :)

Your residuals (for the second step) do not fall below 1E-3, but that does not mean, that your solution is not converged!

cfdhydraulic March 21, 2012 21:31

thanks you all!
another question though...

if my time step must to be 0.00001 to achieve convergence, I have to iterate 10000 to simulate 1 second!!! and it takes hours!! like... forever!

what can I do if want to simulate 5 minutes simulation? what about 1h simulation?

To simulate 1 second takes 10 hours or even more... what about 3600seconds????


itīs impossible... is there anything I can do to acelerate this ?
THANKS!!

LuckyTran March 21, 2012 21:39

Quote:

Originally Posted by cfdhydraulic (Post 350780)
thanks you all!
another question though...

if my time step must to be 0.00001 to achieve convergence, I have to iterate 10000 to simulate 1 second!!! and it takes hours!! like... forever!

what can I do if want to simulate 5 minutes simulation? what about 1h simulation?

To simulate 1 second takes 10 hours or even more... what about 3600seconds????


itīs impossible... is there anything I can do to acelerate this ?
THANKS!!

it's impossible, unless you redefine space-time. If I drive 120km/h, can I go 120km in less than an hour?

10hrs is not very long. It's not even one day, let it run on a weekend if need be.

You should figure out why you need such a small time step and/or why you need to simulate 1 second. You can use an implicit time advancement scheme, stability is guaranteed, and just use more iterations per time-step. It all depends on what you are really after, and WHY. Why did I need to accomplish both 120km and in less than an hour?

cfdhydraulic March 21, 2012 21:46

I need to simulate what's happening in my process after 5 minutes. The time step is 0.0001 to small because I have a good mesh. Good mesh, smaller cells, so small time step.


"It all depends on what you are really after, and WHY." well I need to simulate a discharging process of a tank, and it will take around 5 minutes. however, 1 second is more than 10 hours... 300 seconds will be...3000hours... which means that its not possible...
any suggestion?

LuckyTran March 21, 2012 21:57

Quote:

Originally Posted by cfdhydraulic (Post 350782)
I need to simulate what's happening in my process after 5 minutes. The time step is 0.0001 to small because I have a good mesh. Good mesh, smaller cells, so small time step.


"It all depends on what you are really after, and WHY." well I need to simulate a discharging process of a tank, and it will take around 5 minutes. however, 1 second is more than 10 hours... 300 seconds will be...3000hours... which means that its not possible...
any suggestion?

with an implicit time advancement, time step does not matter. and use a coarser mesh

cfdhydraulic March 22, 2012 01:39

But NITA is only operative in pressure-based cases and I am using density-based...

on the other hand, my experience says that the tank that I am simulating will discharge in about 5 minutes but fluent calculations says more than 1 hour... which is impossible... something wrong...

LuckyTran March 22, 2012 01:42

Quote:

Originally Posted by cfdhydraulic (Post 350805)
But NITA is only operative in pressure-based cases and I am using density-based...

on the other hand, my experience says that the tank that I am simulating will discharge in about 5 minutes but fluent calculations says more than 1 hour... which is impossible... something wrong...

I am not talking about non-iterative time advancement.

cfdhydraulic March 22, 2012 01:44

mmm.... where is this option located in the interface program? i am going to google it
:)

I appreciate your help, thanks

LuckyTran March 22, 2012 01:53

Quote:

Originally Posted by cfdhydraulic (Post 350808)
mmm.... where is this option located in the interface program? i am going to google it
:)

I appreciate your help, thanks

The implicit scheme is the default.

for the pressure based solver, only the implicit formulation is allowed.
for density based, you can choose implicit or explicit formulation.

Again, implicit is default.

cfdhydraulic March 22, 2012 17:04

1 Attachment(s)
I am using implicit, as you said, it is default. But time seems to matter...I am going to try with a coarser mesh by changing advanced control parameters "max coarser"
I will let you know

please see photo, for advise
thanks

cfdhydraulic March 22, 2012 17:42

here is the problem
 
1 Attachment(s)
Hi! this is the problem...
I need to perform unsteady state study to get info about how a system behaves during 60 seconds.
If I use a time step of 0.00001 it converges perfect and the results seem to be right. But it will take weeks of running calculations... so it doesn't work

If I use another time step... it diverges.
If I use a coarsen mesh, it diverges because of the poor mesh.

I am working with density-based , implicit time advancement , courant number 0.1

any suggestion???

LuckyTran March 22, 2012 18:22

Quote:

Originally Posted by cfdhydraulic (Post 351000)
I am using implicit, as you said, it is default. But time seems to matter...I am going to try with a coarser mesh by changing advanced control parameters "max coarser"
I will let you know

please see photo, for advise
thanks

I recommend you do not touch any of the multigrid parameters unless you know exactly what's going on and after you have already debugged everything that is already wrong. The max coarser levels parameter is not the same as running with a coarser mesh. That control is to control the maximum coarse levels used in the multigrid algorithm. The coarsen by parameter controls the new resolution of the coarse grid by a factor of the original grid. In the multigrid algorithm, all solutions eventually are solved in the finest grid (the actual grid).

By use a coarser grid, I mean to go back and generate a coarser mesh from the start.

Also, for larger time step, you will need more iterations per time step to achieve convergence. If you increased the time step size by a factor of 100, you can guess that it will take 100x iterations to converge (it will be slightly less). It will definitely take more than 1x iterations. You increased the time step by 5 orders of magnitude, of course you should wait more than 40 iterations.

banty March 23, 2012 14:07

In tank depletion problem, Two time scale (Long time scale(to deplete the tank) and small time scale ( to track back wave ). it is vary important to set proper time step to capture( even in implicit solver) the phenomena that u want. U can not capture the phenomena if u are overstepping it.

to know the tank depletion time, u can use density base solver with large time step and large courant no, but for tracking wave, u must use fine mesh and smaller time step (wave should not cross one cell length in single time step.dt = cell length/wave speed)

Quote:

Originally Posted by cfdhydraulic (Post 351000)
I am using implicit, as you said, it is default. But time seems to matter...I am going to try with a coarser mesh by changing advanced control parameters "max coarser"
I will let you know

please see photo, for advise
thanks


cfdhydraulic March 23, 2012 14:22

actually I would liket to know how long is the discharging time. How long does the tank take to discharge form 300bar to 75 bar
but my experience tell me that it will take less than a minute, but Fluent... well.... tell me 4000 seconds or something like that... which is impossible.
I believe itīs a problem of setting a proper time step but, I donīt know how to do it as Velocity is unknown as itīs a discharging process.


How can I track (calculate) the back wave?

:)

cfdhydraulic March 23, 2012 14:25

by the way what is "large" time steps or "small" time steps....

large = 1 second? 5 seconds? 30 seconds???

small = 0.001? 0.00001?

banty March 24, 2012 04:48

its all about relative scaling. u know time to deplete the tank as about 60 sec approx. so 100 or 500 time step(60/100 or 60/500) with implicit methodology should be sufficient otherwise u have problem with solver setting.

cfdhydraulic March 24, 2012 08:55

why 100 or 500? is it 100 or 500 iterations with 0.6 time step or 0.12 time step? this is it?

60/100 = 06
60/500 = 0.12

but why this number ?

I am going to perform the calculations and let you know but I donīt think my Velocity will go down after 60 sec or around...


All times are GMT -4. The time now is 21:12.