CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   CFX (https://www.cfd-online.com/Forums/cfx/)
-   -   Time Scale in steady state simulations (https://www.cfd-online.com/Forums/cfx/94445-time-scale-steady-state-simulations.html)

saisanthoshm88 November 16, 2011 08:22

Time Scale in steady state simulations
 
1 Attachment(s)
Could someone please explain the purpose of a pseudo/false time step in a steady state CFD analysis.

From the CFX help manual I was able to understand how the time step is calculated with a representative length scale and a representative velocity scale but I didnít know its actual purpose.

Also could someone please tell me why it is more likely to achieve convergence with a smaller time step.

It would be much helpful if an explanation is given in connection with the image attached

D.B November 16, 2011 10:56

I am not 100% sure on this, but my understanding is that this time step is use to evolve the flow from your initial conditions to final steady state. That is it marches in time by the given timestep ( in seconds ) for each outer loop of iteration. Although thi is just a representative time scale and is mostly relate to the time scale of your flow like 1/w in rotating flows ( w = rot speed in rad/s ).

Anyone please let me know if I am wrong, I would like to clear my misconception.

saisanthoshm88 November 16, 2011 15:05

Yeah.May be but i'd prefer an elaborated explanation and also I am interested in knowing how a smaller time scale would likely help convergence

ghorrocks November 16, 2011 19:45

In short you have to stabilise the numerics somehow. SIMPLE used under relaxation factors, and the CFX approach uses a pseudo trnasient approach. It is only pseudo transient as the more complex transient terms have been ommitted (ie the ones which do not affect the final steady solution) and there is the capacity for different equations to advance at different time steps so the flow is not time accurate.

So now that CFX has chosen a pseudo-transient approach, what time step do you use? The time scale of the flow is a good starting point, but nothing more than a starting point. If that does not converge you go smaller, and as you approach convergence you probably want to make it larger.

ttu004 April 6, 2014 02:13

Steady State convergence issue by reducing time-scale factor
 
Quote:

Originally Posted by ghorrocks (Post 332428)
In short you have to stabilise the numerics somehow. SIMPLE used under relaxation factors, and the CFX approach uses a pseudo trnasient approach. It is only pseudo transient as the more complex transient terms have been ommitted (ie the ones which do not affect the final steady solution) and there is the capacity for different equations to advance at different time steps so the flow is not time accurate.

So now that CFX has chosen a pseudo-transient approach, what time step do you use? The time scale of the flow is a good starting point, but nothing more than a starting point. If that does not converge you go smaller, and as you approach convergence you probably want to make it larger.

Hi Glenn

Sorry to bring this old thread up again.
I think I have a similar question about using smaller time-scale (factor) in steady state.

So according to your statement, if a smaller automatic time-scale factor (say reduce by a factor of 4 = x0.25) is being used and the solution converged towards desired tolerance (10^-7), I should always increase the time-scale size (factor) at the final few iterations before actually reaching desired tolerance? (a similar approach to one using Local Time Scale Factor-http://www.cfd-online.com/Wiki/Ansys_FAQ)

Please correct me if I am wrong about this
Thanks

Regards
Tom

ghorrocks April 6, 2014 08:24

Physical time stepping does not require large time steps int he final run to convergence, but with any luck if you simulation is stable enough it will allow it. Large time steps will speed convergence. But your simulation is not stable enough to allow it then you just have to put up with small time steps and slower convergence.

ttu004 April 6, 2014 18:43

Quote:

Originally Posted by ghorrocks (Post 484142)
Physical time stepping does not require large time steps int he final run to convergence, but with any luck if you simulation is stable enough it will allow it. Large time steps will speed convergence. But your simulation is not stable enough to allow it then you just have to put up with small time steps and slower convergence.

Ok, so only Local time scale factor requires large time step at the end of the run.

thanks Glenn


All times are GMT -4. The time now is 10:45.