transient simulation

Numerical methods and mathematical models of Elmer
Post Reply
Karuna
Posts: 12
Joined: 29 Jun 2015, 11:51
Antispam: Yes

transient simulation

Post by Karuna »

Hi. I would like to perform a transient simulation. But, I'm slightly confused with the terms timestep interval , timestep sizes and output interval.
I would like to run a simulation for 500s and for the output to be displayed/ recorded for every 20s. how should I set the values?

Thanks n regards,
Karuna
annier
Posts: 1168
Joined: 27 Aug 2013, 13:51
Antispam: Yes

Re: transient simulation

Post by annier »

Karuna,
In the simulation block you will need to write like this:
1. For normal and larger geometry

Code: Select all

Simulation
...
Output Intervals(1) = 20  !outputs after every 20 timestep intervals
Timestep intervals(1)=500 !total time = 500*1=500.0 s
Timestep sizes(1) = 1 !this means 1 s when u stated like this unit
...
End
or

Code: Select all

Simulation
...
Output Intervals(1) = 1 !Outputs after every 1 timestep intervals
Timestep intervals(1)=25 !total time = 25*20 = 500.0s
Timestep sizes(1) = 20 !this means 20 s
...
End
2. For very smaller geometry with coordinate scaling, generally timestep size should be slightly smaller than coordinate scaling and thus numerical convergence can be guaranteed.
This is based upon the requirement of solver convergence and is based upon hit and trial of any magnitudes of timestep sizes less than or equal to the normal coordinate scale length.

Code: Select all

Simulation
...
Coordinate Scaling = 1.0e-2 ! in cm scale
Output Intervals(1) = 4000 !Outputs after every 4000 timestep intervals and output time = 4000*5.0e-3 = 20 .0 s
Timestep intervals(1)=100000 !total time = 100000*5.0e-3 = 500.0s
Timestep sizes(1) = 5.0e-3 !this means 5 ms
...
End
3. For outputing simulation results with several timesteps interval , you can use

Code: Select all

Simulation
...
Coordinate Scaling = 1.0 ! in m scale
Output Intervals(2) = 1 2  
!Outputs after every 1 interval (total 10/1 = 10 outputs) for first timesteps sizes of 4 s (i.e. after every 1*4= 4s) and
 ! after every 2 timestep intervals (total 10/2=5 outputs)  for second timestep sizes of 5 s (i.e. adter every 2*5 = 10 s)
Timestep intervals(2)=10 10 !total time = 10*4 + 10*5 = 40+50=90 s
Timestep sizes(2) = 4 5 !this means 4 and 5 s
...
End

4. For more information on these timestep sizes, timestep intervals, and output intervals, please study: viewtopic.php?f=3&t=4317&sid=72271a5140 ... 2cdf4562e7

Yours Sincerely,
Anil Kunwar
Anil Kunwar
Faculty of Mechanical Engineering, Silesian University of Technology, Gliwice
Karuna
Posts: 12
Joined: 29 Jun 2015, 11:51
Antispam: Yes

Re: transient simulation

Post by Karuna »

Hi Anil. thanks for the reply. However, I still don't really understand how it works. I'm running a small size geometry simulation. I used coordinate scaling = Real 0.001.

maybe u could provide me with some definitions of the terms and relate it to an example.

Thanks and regards,
Karuna
Karuna
Posts: 12
Joined: 29 Jun 2015, 11:51
Antispam: Yes

Re: transient simulation

Post by Karuna »

does this means for my case, it should be?
Coordinate Scaling = 1.0e-3
Output Intervals(1) = 20000 ; for output every 10s
Timestep intervals(1)=1000000 ; for total time = 500.0s
Timestep sizes(1) = 5.0e-4
annier
Posts: 1168
Joined: 27 Aug 2013, 13:51
Antispam: Yes

Re: transient simulation

Post by annier »

Karuna,
Yes your setting is correct for the simulation timestep sizes and intervals distribution.
The selection of timestep size of 0.5 ms (5.0e-4) is totally empirical. As u select it to be 5.0e-4 (<1.0e-3) and so your computer needs to perform calculations for 1000000(10^6) intervals.

1. If computational speed is not your limitation (e.g. using ElmerSolver run HPC< high performance computers> or clusters of computers in parallel), then you can perform computations quickly for such greater timestep intervals. Even, you can perform calculations at lower timestep size.

2. If computational speed is your limitation (using ElmerSolver in PC computer ), then for the coordinate scale of 1.0e-3; you can start the timestep size to even 1.0e-3 s (1.0 ms) which will reduce the calculation to 5.0*10^5 intervals. If the system does not converge, you can lower the timestep sizes to less than 1.0 ms. If the system converges, u can continue at larger timestep sizes till u find that the system does not converge. Also (Moreover), for increasing the speed of numerical solving you can use domain decomposition method to use ElmerSolver_mpi (in parallel) as described in viewtopic.php?f=3&t=4513&sid=d9226fd23b ... 0df2c2f51f.

The summary is, timestep size selection or design is the optimization between computational speed and numerical convergence easiness for solver.

Yours Sincerely,
Anil Kunwar
Anil Kunwar
Faculty of Mechanical Engineering, Silesian University of Technology, Gliwice
Post Reply