In the screenshot below, we can see that I set max iterations to 20. There will be 4 output for t=0,0.01,0,0,02 and 0.03.
I asume it would take at most 80 iteration. But currently my simulation is at iteration 200+ and the output folder only has a subfolder for timestep 0
What have I done wrong?
I couldn't get the max iteration paramater
Moderator: oliveroxtoby
Forum rules
and Helpful information for the FEM forum
and Helpful information for the FEM forum
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: I couldn't get the max iteration paramater
The max iterations isn't actually used for the solver you are running. We use adaptive timestepping without outer iterations for the segregated solvers, so it is probably doing many small timesteps. If you look at the output on the report view you can see the time and delta t.supahaka wrote: ↑Fri Apr 01, 2022 4:51 am In the screenshot below, we can see that I set max iterations to 20. There will be 4 output for t=0,0.01,0,0,02 and 0.03.
Untitled.png
I asume it would take at most 80 iteration. But currently my simulation is at iteration 200+ and the output folder only has a subfolder for timestep 0
Screenshot 2022-04-01 at 11.50.37.png
What have I done wrong?
Re: I couldn't get the max iteration paramater
Thank you for your message
I am trying to do a proof of concept of how a machine learning method can be a surrogate model for such flow. In such POC, the temporal length of the simulation is much more important than the accuracy.
I am facing big a disadvantage due to limited computational resource. Therefore, I'd like to ask what are ways to make simulations run faster.
My simulation is about the flow of gas through a 50 cm pipeline. It should be a transient simulation. I have thought of options like:
- Increase max characateristic length: Currently 1 mm
- Increase convergence tolerance: Currently 0.01
As I have no experience on CFD (being a machine learning participant), could you suggest me anything?
I am trying to do a proof of concept of how a machine learning method can be a surrogate model for such flow. In such POC, the temporal length of the simulation is much more important than the accuracy.
I am facing big a disadvantage due to limited computational resource. Therefore, I'd like to ask what are ways to make simulations run faster.
My simulation is about the flow of gas through a 50 cm pipeline. It should be a transient simulation. I have thought of options like:
- Increase max characateristic length: Currently 1 mm
- Increase convergence tolerance: Currently 0.01
As I have no experience on CFD (being a machine learning participant), could you suggest me anything?
-
- Veteran
- Posts: 3156
- Joined: Sat May 20, 2017 12:06 pm
- Location: Germany
Re: I couldn't get the max iteration paramater
Hi supahaka,
welcome to the FreeCAD-Forum.
Running a fast simulation:
- use a 2D mesh
- run a steady-state simulation
- avoid turbulence model
See example, bullet shooting through pipe. Meshing/simulation approx 3 min on intel i5.
When you play the produced iterations in Paraview, you see the moving fluid in the wake of the bullet.
Which in fact is a fake. But the optical illusion suggests the "real" flow.
Maybe you can use this approch to train your AI.
Regards Thomas
BTW:
please update your cfdof-wb and cfmesh, there where several changes in the last days.
Re: I couldn't get the max iteration paramater
I am interested in the 2d rectangular domain. What should the "thickness" of the domain be? 0.1mm, 1mm, or what. Is there a criteria / range for it?
- oliveroxtoby
- Posts: 837
- Joined: Fri Dec 23, 2016 9:43 am
- Location: South Africa
Re: I couldn't get the max iteration paramater
It shouldn't make any difference for the simulation itself, but for the quickest meshing it's best if it's roughly the thickness of the base mesh spacing (since the meshing is done in 3D first and then extruded).