CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   VOF Speed!!!!!!!! (http://www.cfd-online.com/Forums/fluent/41370-vof-speed.html)

soy June 13, 2006 09:26

VOF Speed!!!!!!!!
 
Hi, i am using the VOF (Geo-R, unsteady formulation) with NITA solver and Variable Time Step. The geometry contains arround 1 milion cell.

I need more than 15 hours to simulate just 1 physical second!!!

Can anybody help me to reduce the CPU time?

Thanks in advance

Gernot June 14, 2006 01:37

Re: VOF Speed!!!!!!!!
 
Hard to say without any more information. What about mesh quality, and computerpower (RAM)?

soy June 14, 2006 02:37

Re: VOF Speed!!!!!!!!
 
Hi, thanks for reply

mesh quality: tétra with max skew 0.85 and max aspect ratio 5.

I runs calculation on server IBM HPC. 15h for 1 sec is from calculation on 8 proc.


Michael Bo Hansen June 14, 2006 02:48

Re: VOF Speed!!!!!!!!
 
Convert your mesh to polyhedral (beta feature at the moment)

soy June 14, 2006 03:07

Re: VOF Speed!!!!!!!!
 
is it possible with fluent6.2.16 (and tgrid 4.0.6)?

Michael Bo Hansen June 14, 2006 03:14

Re: VOF Speed!!!!!!!!
 
No. You have to download the beta version 6.3.17, but be carefull.... By the way. The conversion from tet to polyhedral in FLuent is memory demanding limiting the use to apprx 350000 tet cells. Hopefully They will change that when the new official version is released.


soy June 14, 2006 03:27

Re: VOF Speed!!!!!!!!
 
thanks a lot for help, i will try it

if someone know an other way to decrease the CPU time, i am still interested.

thanks.

Andrew Wick June 17, 2006 11:12

Re: VOF Speed!!!!!!!!
 
Well you could...

-Try increasing the under-relaxation factors -Try increasing the NITA correction factors -Try increasing the NITA residual tolerance -Try changing the scheme from geo-reconstruct to Euler-Explicit with the Modified HRIC discretization scheme

Unfortunately, changing any of the first three could and probably will affect your solution stability and/or final results. I would compare any changes you make to the original to make sure you are properly converging. Also, the euler-explicit scheme is much faster than geo-reconstruct, but the interface is not as sharp so you'll want to compare the changes to this as well. Overall the EE scheme isn't too bad though.

Also, since you're using NITA, try the Fractional Step Method. It's a little bit quicker (depending on how much under-relaxation is needed for stability)



All times are GMT -4. The time now is 06:02.