Coefficient loops vs physical timescale

Turbomachine · February 12, 2011, 16:20

Dear All,

I am running transient simulations with mesh deformation of a full 20 blade fan assembly (7 mill. nodes). I am using coefficient loops as the option in convergence control with a maximum number of 100.

Previously I have successfully run simulations with only 2 blades only using the same convergence control. That time adequate convergence was achieved after max. 20 coeff. loops during one mesh deformation step. However this time during many (about 60+%) of the timesteps the coefficient loops pass up to 100 before the simulation proceeds with the next mesh deformation step.

This is unfortunately very inefficient since the computational cost of the simulation becomes very large (20 days of using 8 cores for each run).

I am looking to modify the run in progress either by decreasing the number of coefficient loops. Or by changing to other methods such as physical timescale.

Can anyone advise me which of these is the better way to go? I must also note that the simulations need very high accuracy, however the time to complete them is also finite.

Thank you very much in advance!

ghorrocks · February 13, 2011, 05:54

Why are you using moving mesh to model a fan? Why not use a rotating frames of reference? It will be an order of magnitude faster to run.

Turbomachine · February 13, 2011, 12:22

I am doing static aeroelasticity analysis on the fan hence I need to model mesh deformation (yes, I am using rotating ref. frame too). I did not enclose the details about this because I thought is was irrelevant. My question was regarding the timestep control though.

I read that it the coefficient loops reach the limit then the timestep will be reduced. Does that mean that by choosing a lower number of coefficient loops (say 50) the change for having too large a time step (and bad convergence) is less likely to occur due to the more frequent modification in timestep? Is there a reason why I should allow more coefficient loops during a timestep? Can someone recommend a number for it?

ghorrocks · February 13, 2011, 19:36

I do not have any experience in FSI so cannot comment on that directly. But in general in CFX you want to do 3-5 coeff loops per timestep. If you are doing 20 or more it sounds like your timestep is too big.

February 12, 2011, 16:20	Coefficient loops vs physical timescale	#1
Turbomachine Member Join Date: Aug 2010 Posts: 31 Rep Power: 15	Dear All, I am running transient simulations with mesh deformation of a full 20 blade fan assembly (7 mill. nodes). I am using coefficient loops as the option in convergence control with a maximum number of 100. Previously I have successfully run simulations with only 2 blades only using the same convergence control. That time adequate convergence was achieved after max. 20 coeff. loops during one mesh deformation step. However this time during many (about 60+%) of the timesteps the coefficient loops pass up to 100 before the simulation proceeds with the next mesh deformation step. This is unfortunately very inefficient since the computational cost of the simulation becomes very large (20 days of using 8 cores for each run). I am looking to modify the run in progress either by decreasing the number of coefficient loops. Or by changing to other methods such as physical timescale. Can anyone advise me which of these is the better way to go? I must also note that the simulations need very high accuracy, however the time to complete them is also finite. Thank you very much in advance!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Superlinear speedup in OpenFOAM 13	msrinath80	OpenFOAM Running, Solving & CFD	18	March 3, 2015 05:36
Viscous resistance coefficient in physical v-model	AL	FLUENT	0	December 15, 2008 14:52
Automotive test case	vinz	OpenFOAM Running, Solving & CFD	98	October 27, 2008 08:43
Advice on Physical Timescale	Oli	CFX	0	January 16, 2007 10:41
SST in 2D?	Jesper Sørensen	CFX	16	December 23, 2006 08:40

February 13, 2011, 05:54		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	Why are you using moving mesh to model a fan? Why not use a rotating frames of reference? It will be an order of magnitude faster to run.

February 13, 2011, 12:22		#3
Turbomachine Member Join Date: Aug 2010 Posts: 31 Rep Power: 15	I am doing static aeroelasticity analysis on the fan hence I need to model mesh deformation (yes, I am using rotating ref. frame too). I did not enclose the details about this because I thought is was irrelevant. My question was regarding the timestep control though. I read that it the coefficient loops reach the limit then the timestep will be reduced. Does that mean that by choosing a lower number of coefficient loops (say 50) the change for having too large a time step (and bad convergence) is less likely to occur due to the more frequent modification in timestep? Is there a reason why I should allow more coefficient loops during a timestep? Can someone recommend a number for it?

February 13, 2011, 19:36		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,703 Rep Power: 143	I do not have any experience in FSI so cannot comment on that directly. But in general in CFX you want to do 3-5 coeff loops per timestep. If you are doing 20 or more it sounds like your timestep is too big.