time step and # of coefficient loops

hamed.majeed · September 22, 2012, 03:58

Hi,

Guys I have some questions regarding value of time step needed and the # of coefficient loops specified. I have two questions:
1. How to perform time step dependence test.
2. What is the appropriate value of # of coefficient loops needed.

hamed.majeed · September 22, 2012, 04:00

According to ansys cfx help I found the following..
Time step:
Time step dependence is done at a single time step,. doubling the time, and halving the time step.
ref: ANSYS Help // Modeling Guide // 15. Advice on Flow Modeling // 15.4. Timestep Selection // 15.4.2. Transient Timestep Control

Now, I have plotted curves for cd at a wall for (i) time step = 0.01 s
(ii) time step = 0.005 s, and (iii) time step = 0.02 s
I have another question
3. How to find the appropriate range of max and min %age error for time steps, and then finally select one?

ghorrocks · September 24, 2012, 23:06

1. Time step dependance check - run a simulation, then do it again with half the time step size. If the two simulations are the same within a toelrance you are happy with then you are finished. If not the half the time step size and try again.
2. 3-5 coeff loops per time step for most applications, 5-10 for trickier (eg multiphase) simulations. There is no guide for steady state runs, only the tightness of convergence matters.
3. I do not understand your question.

hamed.majeed · September 25, 2012, 23:07

Hi,
In time dependence you said that the results must be withing "tolerance". How do we specify this tolerance value?

ghorrocks · September 26, 2012, 09:08

The tolerance is how accurate you are happy to have that result. Are you doing a quick design check - maybe 20%. Want a good answer - maybe 5%. Need high accuracy - maybe 1%. You have to choose for your application.

Far · September 27, 2012, 02:54

I have seen an official Ansys CFX presentation on circular cylinder where he took the one coefficient loop per time step. Whereas I think there must be some coefficient loops per time step as you have already mentioned. Do you think they (CFX team) have done it for purpose?

I have shown some results on this thread http://www.cfd-online.com/Forums/flu...lat-plate.html as well, if any one needs complete presenation, I am more than happy to provide one.

ghorrocks · September 27, 2012, 08:47

Doing even less coeff loops per time step is good for temporal resolution (ie fine time steps) but in most cases not optimal. Also be aware that I understand the second order components of the temporal terms only get enforced on the second coeff loop, so you need at least 2 coeff loops per iteration to get second order to work.

hamed.majeed · September 28, 2012, 07:53

Hi,
I remember reading in ANSYS Help that the solution resolution must be increased by increasing the time steps rather increasing the coefficient loops

ghorrocks · September 28, 2012, 08:33

To clarify - increasing the number of time steps (ie smaller time steps) with 3-5 coeff loops per iteration is usually a better approach then larger time steps with more coeff loops. I think that is what you said.

cfdesf1990 · August 16, 2017, 02:22

Hello
I am simulation a moving mesh case. It is a turbulent, free surface, moving mesh, and transient problem. I set 1 to 4 for COEFFICIENT LOOP ITERATION and 4 for COUPLING/STAGGER ITERATION which a parameter related to moving mesh. So Every time step is about 16 (4*4) iteration which seems a bit weird. Do you have any idea? Since it takes a lot to solve such time-consuming problem, I think I am doing something wrong?
Do you have any other idea to increase the solution speed? Actually, I have to choose 0.001 or 0.002 for time step because for larger ones it gives me a negative volume.

ghorrocks · August 16, 2017, 07:24

Coupling/stagger iterations are used in FSI simulations. Are you intending to do a FSI simulation? If you just need moving mesh you do not need FSI and the simulation becomes much easier.

cfdesf1990 · August 16, 2017, 08:58

Thanks Glenn. Actually, I want to have some degree of freedom to see the rotational and horizontal movement of an object in the presence of surface waves. Based on buoy example in CFX tutorial I thought it should be FSI. Is it true? The deformation of the solid object is not important for me. I am just going to investigate movement of a given object.

ghorrocks · August 16, 2017, 09:00

If the body is rigid, and its motion is just translation and rotation, then this is not FSI and should be modelled as a rigid body motion. You only need FSI if the body deforms due to the forces applied by the fluid.

cfdesf1990 · August 16, 2017, 13:22

Yes you are right. Thank you

September 22, 2012, 03:58	time step and # of coefficient loops	#1
hamed.majeed Senior Member Hamed Abdul Majeed Join Date: May 2012 Location: New Orleans, LA, US Posts: 147 Rep Power: 14	Hi, Guys I have some questions regarding value of time step needed and the # of coefficient loops specified. I have two questions: 1. How to perform time step dependence test. 2. What is the appropriate value of # of coefficient loops needed.

September 24, 2012, 23:06		#3
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	1. Time step dependance check - run a simulation, then do it again with half the time step size. If the two simulations are the same within a toelrance you are happy with then you are finished. If not the half the time step size and try again. 2. 3-5 coeff loops per time step for most applications, 5-10 for trickier (eg multiphase) simulations. There is no guide for steady state runs, only the tightness of convergence matters. 3. I do not understand your question. hamed.majeed and mmcpr like this.

September 26, 2012, 09:08		#5
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	The tolerance is how accurate you are happy to have that result. Are you doing a quick design check - maybe 20%. Want a good answer - maybe 5%. Need high accuracy - maybe 1%. You have to choose for your application. hamed.majeed likes this.

September 28, 2012, 08:33		#9
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	To clarify - increasing the number of time steps (ie smaller time steps) with 3-5 coeff loops per iteration is usually a better approach then larger time steps with more coeff loops. I think that is what you said. aero_head likes this.

August 16, 2017, 09:00		#13
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	If the body is rigid, and its motion is just translation and rotation, then this is not FSI and should be modelled as a rigid body motion. You only need FSI if the body deforms due to the forces applied by the fluid. cfdesf1990 likes this.

September 22, 2012, 04:00		#2
hamed.majeed Senior Member Hamed Abdul Majeed Join Date: May 2012 Location: New Orleans, LA, US Posts: 147 Rep Power: 14	According to ansys cfx help I found the following.. Time step: Time step dependence is done at a single time step,. doubling the time, and halving the time step. ref: ANSYS Help // Modeling Guide // 15. Advice on Flow Modeling // 15.4. Timestep Selection // 15.4.2. Transient Timestep Control Now, I have plotted curves for cd at a wall for (i) time step = 0.01 s (ii) time step = 0.005 s, and (iii) time step = 0.02 s I have another question 3. How to find the appropriate range of max and min %age error for time steps, and then finally select one?

September 25, 2012, 23:07		#4
hamed.majeed Senior Member Hamed Abdul Majeed Join Date: May 2012 Location: New Orleans, LA, US Posts: 147 Rep Power: 14	Hi, In time dependence you said that the results must be withing "tolerance". How do we specify this tolerance value?

September 27, 2012, 02:54		#6
Far Senior Member Sijal Join Date: Mar 2009 Location: Islamabad Posts: 4,558 Blog Entries: 6 Rep Power: 54	I have seen an official Ansys CFX presentation on circular cylinder where he took the one coefficient loop per time step. Whereas I think there must be some coefficient loops per time step as you have already mentioned. Do you think they (CFX team) have done it for purpose? I have shown some results on this thread http://www.cfd-online.com/Forums/flu...lat-plate.html as well, if any one needs complete presenation, I am more than happy to provide one.

September 27, 2012, 08:47		#7
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	Doing even less coeff loops per time step is good for temporal resolution (ie fine time steps) but in most cases not optimal. Also be aware that I understand the second order components of the temporal terms only get enforced on the second coeff loop, so you need at least 2 coeff loops per iteration to get second order to work.

September 28, 2012, 07:53		#8
hamed.majeed Senior Member Hamed Abdul Majeed Join Date: May 2012 Location: New Orleans, LA, US Posts: 147 Rep Power: 14	Hi, I remember reading in ANSYS Help that the solution resolution must be increased by increasing the time steps rather increasing the coefficient loops

August 16, 2017, 02:22		#10
cfdesf1990 New Member Fluidflow Join Date: Oct 2015 Posts: 24 Rep Power: 11	Hello I am simulation a moving mesh case. It is a turbulent, free surface, moving mesh, and transient problem. I set 1 to 4 for COEFFICIENT LOOP ITERATION and 4 for COUPLING/STAGGER ITERATION which a parameter related to moving mesh. So Every time step is about 16 (4*4) iteration which seems a bit weird. Do you have any idea? Since it takes a lot to solve such time-consuming problem, I think I am doing something wrong? Do you have any other idea to increase the solution speed? Actually, I have to choose 0.001 or 0.002 for time step because for larger ones it gives me a negative volume.

August 16, 2017, 07:24		#11
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,870 Rep Power: 144	Coupling/stagger iterations are used in FSI simulations. Are you intending to do a FSI simulation? If you just need moving mesh you do not need FSI and the simulation becomes much easier.

August 16, 2017, 08:58		#12
cfdesf1990 New Member Fluidflow Join Date: Oct 2015 Posts: 24 Rep Power: 11	Thanks Glenn. Actually, I want to have some degree of freedom to see the rotational and horizontal movement of an object in the presence of surface waves. Based on buoy example in CFX tutorial I thought it should be FSI. Is it true? The deformation of the solid object is not important for me. I am just going to investigate movement of a given object.

August 16, 2017, 13:22		#14
cfdesf1990 New Member Fluidflow Join Date: Oct 2015 Posts: 24 Rep Power: 11	Yes you are right. Thank you