Learning Forum

[JamesWright]

I'm running an SBES simulation of a swirling flow in a pipe. Per the suggestion of Dr. Menter in "Best Practice: Scale-Resolving Simulations in ANSYS CFD, Ver. 2.00", I tried out the Non-Iterative Time Advancement (NITA) scheme on my simulation since it should be significantly more CPU efficient than Iterative Time Advancement (ITA).

The results were a bit of a mixed bag, as the NITA (using either PISO or Fractional Step) was significantly slower per time step than ITA. NITA would take ~40s per timestep where as ITA would take ~15 seconds. However, NITA would converge in one time step and would have significantly lower residual values than ITA (around 2 orders of magnitude lower).

First question, are these expected/normal results?

Second question, how might I go about speeding up the NITA? Like how to reduce the residual criteria for it (since it's well lower than what I need)?

[JamesWright]

Alright, I found the help section for NITA (37.3.3 in the Fluent User's Manual). Looks like my choices for tuning NITA are "Max Corrections", "Correction Tolerance", and "Residual Tolerance".

My initial instinct is to reduce the residual tolerance, but according to the help, the only adjustment I should make is to lower the tolerance. I'm running CFL values below 1, so it says I should be lowering it significantly.

The manual entry mentions "Accelerated Time Marching" for LES solutions, but I don't see that option anywhere. While I'm technically running a hybrid RANS/LES model, I still don't see an options for "Accelerated Time Marching" when I switch to pure LES. Should I manually implement the settings described for the "Accelerated Time Marching" setting (changing "Max Corrections", P-V scheme, and MultiGrid settings)?

After that, I'm kinda lost as to what the other two settings are referring to. From what I can tell, I could increase the "Correction Tolerance" or decrease the "Max Corrections" level to reduce the number of sub-iterations, but I'm not sure what to do exactly.

[DrAmine]

Accelerated Time marching is beta feature. It makes just some setting adjustments.  The general rule of thumb if ITA does converge for the time step in 5 to 10 iterations then NITA will not perform better. NITA residuals might be misleading as they are related to the ratio of first  AMG cycle at the current sub iteration to the intial sub iteration / last AMG cycle current sub-iteration/initial sub iteration.

[JamesWright]

Gotcha. ITA was converging within 4-5 timesteps, so that makes sense. I'll keep that in mind if I run into more convergence timesteps though.

 

Thanks for the answer!

[DrAmine]

Please mark this question as solved. Thank you.