force convergence problems in CFX 6DOF rigid body solver

ajay_ks · March 16, 2013, 22:53

hi
i have a questions regarding convergence of force (default is 1e-02 in CFX). I want to know is whether is it possible to improve the convergence of rigid body solver for force (say upto 1e-04 or 1e-06 . i know that it can be set but would it converge to that value??.)
I am doing a simulation by making a rigid body move inside the fluid by using CFX rigid body solver.
I am attaching the image file of force convergence which i feel is not sufficient for accurately estimating the drag force on the rigid body.

Red in the figure is for the force convergence and it just converges to 1e-02. Is it accurate enough??

ghorrocks · March 17, 2013, 05:47

Rather than just "feeling" whether this is tight enough, how about actually proving whether it is tight enough? Do a simulation with the existing tolerance, another 10x tighter, and another 10x tighter again. Did it change the result of anything you care about? Now you have just shown whether the tolerance is tight enough.

ajay_ks · March 21, 2013, 09:50

Hi Glenn
Thanks for your reply. I think you are correct i should check by trying with convergence of 1e-03 and so on and see whether it makes any difference to the results for the rigid body.
But the problem is that one 3D simulation takes around 20 to 30 days (on i7 system with 8 cores and 8GB ram) and right now i do not have more computational resources. So if you have run some rigid body solver cases then i think you can help & tell whether tight convergence will make any difference or not?????
I also have one more question regarding using remeshing in ICEM for using with 6DOF rigid body solver for a 2D case (i know how to do it in 3d but in 2d i didnot have much success). I tried making thin geometry (i.e one cell thick in z-direction) and it ran successfully but for a turbulent case it is showing results like that of a laminar flow. ( i checked this by running a simple turbulent channel flow case and the velocity profile that i got was that of laminar flow even without using remeshing)
So is there any way of doing 2d simulation in cfx???
Should i take thickness in z direction very small (say 1e-04m)

. i have tried with 0.5 mm but it did not work.

ghorrocks · March 21, 2013, 17:53

Quote:

one 3D simulation takes around 20 to 30 days (on i7 system with 8 cores and 8GB ram) and right now i do not have more computational resources.

OK, so think laterally. If the full model is too big to do sensitivity studies on, can you do a smaller simulation which will have the same (or similar enough) sensitivity? Maybe model one body instead of 50. Or model the first cycle, not all cycles. Or simplify the body, or remove physics not relevant to the sensitivity - there are lots of options.

Quote:

it is showing results like that of a laminar flow.

What do you mean by "it is showing results like that of a laminar flow"? What results is it showing?

The thickness to use in 2D models is described in the FAQ: http://www.cfd-online.com/Wiki/Ansys...tion_in_CFX.3F

ajay_ks · March 22, 2013, 13:26

[QUOTE What do you mean by "it is showing results like that of a laminar flow"? What results is it showing?

ajay_ks · March 22, 2013, 13:27

[QUOTE] [ What do you mean by "it is showing results like that of a laminar flow"? What results is it showing?]
I was just talking about the velocity profile which i got at the end of the simulation. i ran a turbulent 2D channel flow case(Re = 10000) and got a parabolic velocity profile at the exit of channel.
And thanks for your suggestions

ghorrocks · March 23, 2013, 06:06

I see. On the face of it this suggests the turbulence is of a low intensity, not enough to change the velocity profile. Are you sure this flow is actually turbulent?

ajay_ks · March 24, 2013, 21:44

Yes, i think the flow is turbulent as Re is 10000 and moreover i did the same simulation of 2D channel flow in FLUENT with the same boundary conditions and the velocity profile which i got in FLUENT is not parabolic but is just as is expected in turbulent flows.
Has it something to do with the thickness of 1 element in z-direction in CFX or is it something else????

ghorrocks · March 25, 2013, 05:02

No, it should not be affected by the 1 element thickness.

How do you know that Re=10000 is turbulent in your geometry? It would be turbulent in most cases in pipe flow, but if your flow is not in a pipe then this is only a guide.

Your comment that CFX and Fluent are giving quite different results suggests you have made some fundamental mistakes in your simulation setup. There is nothing inherent about CFX or Fluent which makes it incapable for modelling the flows you are talking about (from what you have said so far anyway). Both codes are perfectly capable for this type of flow.

March 22, 2013, 13:26		#5
ajay_ks New Member ajay Join Date: Sep 2011 Location: New Delhi, India Posts: 8 Rep Power: 15	[QUOTE What do you mean by "it is showing results like that of a laminar flow"? What results is it showing? Last edited by ajay_ks; March 22, 2013 at 13:29. Reason: repeat

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
6Dof Solver + Moving Object in fluent	PouyanJ	FLUENT	44	March 24, 2020 04:46
Turbulent solver rigid body mechanics	philippose	OpenFOAM Running, Solving & CFD	53	July 16, 2011 14:35
ATTENTION! Reliability problems in CFX 5.7	Joseph	CFX	14	April 20, 2010 16:45
Problems with SUPG body force term	FEM question	Main CFD Forum	0	January 21, 2006 18:51
CFX 5.5	Roued	CFX	1	October 2, 2001 17:49

March 17, 2013, 05:47		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,871 Rep Power: 144	Rather than just "feeling" whether this is tight enough, how about actually proving whether it is tight enough? Do a simulation with the existing tolerance, another 10x tighter, and another 10x tighter again. Did it change the result of anything you care about? Now you have just shown whether the tolerance is tight enough.

March 21, 2013, 09:50		#3
ajay_ks New Member ajay Join Date: Sep 2011 Location: New Delhi, India Posts: 8 Rep Power: 15	Hi Glenn Thanks for your reply. I think you are correct i should check by trying with convergence of 1e-03 and so on and see whether it makes any difference to the results for the rigid body. But the problem is that one 3D simulation takes around 20 to 30 days (on i7 system with 8 cores and 8GB ram) and right now i do not have more computational resources. So if you have run some rigid body solver cases then i think you can help & tell whether tight convergence will make any difference or not????? I also have one more question regarding using remeshing in ICEM for using with 6DOF rigid body solver for a 2D case (i know how to do it in 3d but in 2d i didnot have much success). I tried making thin geometry (i.e one cell thick in z-direction) and it ran successfully but for a turbulent case it is showing results like that of a laminar flow. ( i checked this by running a simple turbulent channel flow case and the velocity profile that i got was that of laminar flow even without using remeshing) So is there any way of doing 2d simulation in cfx??? Should i take thickness in z direction very small (say 1e-04m) . i have tried with 0.5 mm but it did not work.

March 22, 2013, 13:27		#6
ajay_ks New Member ajay Join Date: Sep 2011 Location: New Delhi, India Posts: 8 Rep Power: 15	[QUOTE] [ What do you mean by "it is showing results like that of a laminar flow"? What results is it showing?] I was just talking about the velocity profile which i got at the end of the simulation. i ran a turbulent 2D channel flow case(Re = 10000) and got a parabolic velocity profile at the exit of channel. And thanks for your suggestions

March 23, 2013, 06:06		#7
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,871 Rep Power: 144	I see. On the face of it this suggests the turbulence is of a low intensity, not enough to change the velocity profile. Are you sure this flow is actually turbulent?

March 24, 2013, 21:44		#8
ajay_ks New Member ajay Join Date: Sep 2011 Location: New Delhi, India Posts: 8 Rep Power: 15	Yes, i think the flow is turbulent as Re is 10000 and moreover i did the same simulation of 2D channel flow in FLUENT with the same boundary conditions and the velocity profile which i got in FLUENT is not parabolic but is just as is expected in turbulent flows. Has it something to do with the thickness of 1 element in z-direction in CFX or is it something else????

March 25, 2013, 05:02		#9
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,871 Rep Power: 144	No, it should not be affected by the 1 element thickness. How do you know that Re=10000 is turbulent in your geometry? It would be turbulent in most cases in pipe flow, but if your flow is not in a pipe then this is only a guide. Your comment that CFX and Fluent are giving quite different results suggests you have made some fundamental mistakes in your simulation setup. There is nothing inherent about CFX or Fluent which makes it incapable for modelling the flows you are talking about (from what you have said so far anyway). Both codes are perfectly capable for this type of flow.