
[Sponsors] 
May 29, 2007, 19:23 
Help with large sim

#1 
Guest
Posts: n/a

Hi, I am trying to simulate a 45,000 Ltire being filled with water. I have created a multiphase sim with the inital volume fraction of air at 1. The intlet has a vf of water of 1 and air 0, naturally. The water is being pmped into the tank via a 2" inlet at 3.11 m/s (uniform speed).
When I originally ran the sim, I was using a mesh of 1M elements and a timestep of 15 seconds. Things were bad! I then refined the mesh to 3.4M elements, changed the time step to 0.25 seconds. This was good, the courant number was around 30, a bit high but livable. The problem is I wanted to simulate the tank filling for 2 hours, after 8.5 hours of solving, my quad core CPU with 4 gig ram had done 15 time steps, that was 3.75 seconds of actual 'flow'. Under these conditions it would take 1.86 years to complete. Note that CPU was at 100% and the sim was entirely in RAM, no spool to HDD. My question is, if I am only interested in the water portion and don't care about the air, is there any way I can setup the sim to make it viable in my lifetime. The objective is to examine the effect of different inlet velocities and inlet anlges have on the mixing of the fluids. eg when the tank was to get to 30 % full, a different liquid was going to be injected into the tank, I have not done this yet as the sim would have taken too long to get to 30% full. Can I create say 10 tanks, the first tank being the bottom 10% of the tank, run this sim on this until it gets reasonably full, then use another tank that is the botton 20% and map the previous results into the new domain. That way I am not solving the extra 'air' elements/equations that I don't care about. What are your thoughts about this kind of approach? I dont know how to map results from the 10% domain to the initial conditions of the 20% domain. Any help would be appreciated. Thanks Stu 

May 30, 2007, 06:21 
Re: Help with large sim

#2 
Guest
Posts: n/a

Theoretically, you could calculate the location of the free surface analytically and then use an icemcfd macro to generate a different mesh for each timestep. This would eliminate the air phase but this would only reduce the solving time modestly.
Try using symmetry. A 2D approximation would reduce computing time by an order of magnitude. If you simply have to solve the 3D problem hire computing time on a cluster. Physical scale modeling is another option. 

May 30, 2007, 06:36 
Re: Help with large sim

#3 
Guest
Posts: n/a

I can't take advantage of symmetry but I have had a thought. Part 2 of tutorial 20 uses junction box routines to 'change the mesh' maybe this approach is possible.
I have had another idea, could I use multiple domains that each represent say 10% of the height of the tank? Is there a way to turn off all calculation in a domian until I am ready to process that domain? I could process the bottom doam, wait until it gets to sday 50% full, then allow the domain that is sitting above it, that would be the 10 to 20% Height domain, to start calcualting. Any thoughts? Stu 

May 30, 2007, 07:04 
Re: Help with large sim

#4 
Guest
Posts: n/a

Hi,
Even if you managed to restrict your calculation domain to the liquid phase, your computation time would only be reduced to half: from 1.86 years to 0.93 years I think you'll have to stay quite below the 1M elements mesh. And try to use hexa elementes wherever possible Take a look at this: http://www.ansys.com/industries/cons...oiletvalve.htm It's about discharge instead of filling Here's the paper http://www2.dem.uc.pt/antonio.gameir...SEAS_paper.pdf I think it may give you some ideas about meshes and time steps Rui 

May 30, 2007, 08:15 
Re: Help with large sim

#5 
Guest
Posts: n/a

Why do you do this transient CFDcalculation? Do you really need to know the full history of the filling of the vessel? Isn't it sufficient to create a mesh for the 30% fill and perform the stationary flow study with different inlet angles and inlet velocities?
I don't think the situation at 5% filling ratio will influence the situation at 30%, unless you want to solve complex chemical reactions that are competitive. If you have to solve that......... GertJan 

May 30, 2007, 08:57 
Re: Help with large sim

#6 
Guest
Posts: n/a

Can I use the Step function in CEL to change the timestep each of the 10% full domains runs? eg initially start the second doamin which represents 1020% Height at a long 1 minute time step, but when I want to turn it on, I use the step function to change the time step to 0.25 seconds, which ios the same as the 010% height domain? That was I would not have to calculate what the air is doing in a much large all encompassing domain.
Would this work for a transient analysis? Thanks Stu 

May 30, 2007, 13:38 
Re: Help with large sim

#7 
Guest
Posts: n/a

What you can do is create a flat cilinder which represents the first 10% of your vessel. Using dynamic meshing, you can let this grow with the pace of the amount of water. When it is 20%, regrid your domain, interpolate the previous solution and let it grow further to say 40% ... and so on. You can represent the free surface as a wall with free slip. The pressure on this surface is a rough guess for the height of the disturbances.
This is a possible route but I still have my thoughts if this CFDsimulation is useful...... It won't be my approach. Good luck, GertJan www.bunova.nl 

June 4, 2007, 17:13 
Re: Help with large sim

#8 
Guest
Posts: n/a

Hi Stu,
I offer a low cost processing only service for CFX simulations: http://www.pointzeroconsulting.co.uk...rocessing.html Drop me a mail if interested. Cheers, Lee. 

Thread Tools  
Display Modes  


Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Postprocessing large data sets in parallel  evrikon  OpenFOAM PostProcessing  28  June 28, 2016 03:43 
Suggested unsteady, implicit solver stable with arbitrarily large time steps  djbungee  OpenFOAM Programming & Development  45  March 23, 2015 05:14 
LES on a scaled geomentry (Mesh and Large scales)  comb  Main CFD Forum  0  February 8, 2011 07:21 
large vortex on outlet when gravity accounted for  Alejandra Menchaca  Phoenics  0  March 3, 2009 20:53 
what happens if y+ is too large or too small  Bo Jensen  Siemens  12  February 20, 2003 00:31 