High number of time steps

redza010 · August 4, 2012, 10:25

Hi, I am just wondering. I am currently running a transient simulation and currently investigating the effectiveness of synthetic jets in cooling of hight temperature surface in a channel flow.

My total duration is 400/f and the time-step is 0.01/f in which the frequency is 600. Therefore the number of time step is 40000 which is very big. Do you think it would take a very long time to complete the simulation? Do you have any suggestion as I am currently running out of time

Thank you.

evcelica · August 4, 2012, 12:25

The simulation time depends on much much more than just the number of timesteps. Your mesh will make a much larger difference, as will your convergence criteria. 40000 timesteps is not really a lot, I'm running one now that is simulating 40 hours, with timesteps of ~0.1 s, so that is 1.4 million time steps.

Just start the simulation and see how long each timestep takes, and extrapolate about how long its going to take. If its taking too long, reduce your mesh size, or your convergence criteria. I would make sure you have a high quality mesh since that will aid in faster convergence.

redza010 · August 4, 2012, 15:07

the number of element of the mesh is around 8 millions. Is it normal for the residue graph to be fluctuating instead of converging?

flotus1 · August 4, 2012, 15:17

What kind of fluctuations do you observe in the residuals?
It is normal that the residuals go up in the first iteration of a new timestep.
If they fluctuate within one timestep instead of dropping smoothly, then there is something wrong.

Concerning the runtime: Is it really neccessary that one period is resolved with 100 timesteps? Maybe you could use longer timesteps, but this will be difficult to judge without enough time for a sensitivity analysis.

redza010 · August 4, 2012, 15:23

The graph is fluctuating throughout the run with the same pattern and not converging

evcelica · August 4, 2012, 20:42

About the convergence, in a transient run the graphs will look a bit different than in a steady state. In steady state you see the solution converging and the graphs display this. In transient we don't see the actual convergence of the individual timesteps, but just the residual level at the end of each timestep. So you will see a graph that can look like a non-converging bouncy line, or a sawtooth patten. It should meet your convergence criteria, then advance in time, and iterate to meet your convergence criteria again. Are you meeting your convergence criteria? You have to look at the "out file" to see how well each timestep is converging, and in how many iterations.

ghorrocks · August 5, 2012, 08:43

Why do you need to model 400 cycles? What takes this long to develop?

This looks like a simulation with a big range of time scales. The flow time scale for the jet is fast, but the thermal time scale (I presume that is what it is) is far slower. These simulations are always difficult to simulate as you need fine enough time steps to resolve the fast thing, but enough time steps to resolve the slow thing.

A good approach is often to decouple the simulation. You will have to determine what is the best way to decouple it, but one approach could be to work out the cooling effect of the jet at various fixed thermal conditions, then do a simple ODE thermal model using the cooling effect from the simulations. Much easier - as long as the simulation can be decoupled as I describe.

August 4, 2012, 10:25	High number of time steps	#1
redza010 New Member Join Date: Aug 2012 Posts: 4 Rep Power: 13	Hi, I am just wondering. I am currently running a transient simulation and currently investigating the effectiveness of synthetic jets in cooling of hight temperature surface in a channel flow. My total duration is 400/f and the time-step is 0.01/f in which the frequency is 600. Therefore the number of time step is 40000 which is very big. Do you think it would take a very long time to complete the simulation? Do you have any suggestion as I am currently running out of time Thank you.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Orifice Plate with a fully developed flow - Problems with convergence	jonmec	OpenFOAM Running, Solving & CFD	3	July 28, 2011 05:24
Full pipe 3D using icoFoam	cyberbrain	OpenFOAM	4	March 16, 2011 09:20
[blockMesh] BlockMeshmergePatchPairs	hjasak	OpenFOAM Meshing & Mesh Conversion	11	August 15, 2008 07:36
Unaligned accesses on IA64	andre	OpenFOAM	5	June 23, 2008 10:37
DPM: Particle Tracking	Madhukar	FLUENT	1	July 24, 2007 03:51

August 4, 2012, 12:25		#2
evcelica Senior Member Erik Join Date: Feb 2011 Location: Earth (Land portion) Posts: 1,167 Rep Power: 23	The simulation time depends on much much more than just the number of timesteps. Your mesh will make a much larger difference, as will your convergence criteria. 40000 timesteps is not really a lot, I'm running one now that is simulating 40 hours, with timesteps of ~0.1 s, so that is 1.4 million time steps. Just start the simulation and see how long each timestep takes, and extrapolate about how long its going to take. If its taking too long, reduce your mesh size, or your convergence criteria. I would make sure you have a high quality mesh since that will aid in faster convergence.

August 4, 2012, 15:07		#3
redza010 New Member Join Date: Aug 2012 Posts: 4 Rep Power: 13	the number of element of the mesh is around 8 millions. Is it normal for the residue graph to be fluctuating instead of converging?

August 4, 2012, 15:17		#4
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	What kind of fluctuations do you observe in the residuals? It is normal that the residuals go up in the first iteration of a new timestep. If they fluctuate within one timestep instead of dropping smoothly, then there is something wrong. Concerning the runtime: Is it really neccessary that one period is resolved with 100 timesteps? Maybe you could use longer timesteps, but this will be difficult to judge without enough time for a sensitivity analysis.

August 4, 2012, 15:23		#5
redza010 New Member Join Date: Aug 2012 Posts: 4 Rep Power: 13	The graph is fluctuating throughout the run with the same pattern and not converging

August 4, 2012, 20:42		#6
evcelica Senior Member Erik Join Date: Feb 2011 Location: Earth (Land portion) Posts: 1,167 Rep Power: 23	About the convergence, in a transient run the graphs will look a bit different than in a steady state. In steady state you see the solution converging and the graphs display this. In transient we don't see the actual convergence of the individual timesteps, but just the residual level at the end of each timestep. So you will see a graph that can look like a non-converging bouncy line, or a sawtooth patten. It should meet your convergence criteria, then advance in time, and iterate to meet your convergence criteria again. Are you meeting your convergence criteria? You have to look at the "out file" to see how well each timestep is converging, and in how many iterations.

August 5, 2012, 08:43		#7
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,700 Rep Power: 143	Why do you need to model 400 cycles? What takes this long to develop? This looks like a simulation with a big range of time scales. The flow time scale for the jet is fast, but the thermal time scale (I presume that is what it is) is far slower. These simulations are always difficult to simulate as you need fine enough time steps to resolve the fast thing, but enough time steps to resolve the slow thing. A good approach is often to decouple the simulation. You will have to determine what is the best way to decouple it, but one approach could be to work out the cooling effect of the jet at various fixed thermal conditions, then do a simple ODE thermal model using the cooling effect from the simulations. Much easier - as long as the simulation can be decoupled as I describe.