Hello,
I'm thinking about how to choose the optimal time increment for my simulations. I set up a very simple testcase in order to study the behaviour of contact interaction between particles and STL-geometries, defined as wall.
Video: http://www.youtube.com/watch?v=IBf3vUwHh-w
First, I calculated the timestep as provided by many posts in this forum with the formula dt = 0.2*dt_rayleigh:
r = 0.005
rho = 8000
Y = 5e6
nu = 0.3
G = Y/(2*(1+nu)) = 1923076.92
dt = 0.2*PI*r*sqrt(rho/G)/(0.1631*nu+0.8766)
dt = 0.00021893
So, I choose dt = 0.0002 in my simulation. If I plot the z-Coordinate of the bouncing particle I can see an instability (see Fig.1).
For this reason I halve the time increment and use dt = 0.0001. Now it seems to run stable (see Fig.2).
Are there any explanations for this behaviour and any more hints how to choose an appropriate time increment when using STL-geometries in the simulation?
Thank you!
Best regards,
Jan
cstoltz | Tue, 01/28/2014 - 12:57
What version of LIGGGHTS are
What version of LIGGGHTS are you using? Try setting 'neigh modify delay 0'. This seems to significantly improve stability for some simulations, though will cost you a bit in terms of performance.
Regards,
Chris
aaigner | Tue, 01/28/2014 - 14:07
20% of the Rayleigh time step ...
Hi!
20% of the Rayleigh time step is somewhat high. The simulation may fail.. in your case it failed ;-). Apparently, the bouncing particle moved too far in one timestep and the particle-wall contact was not resolved correctly.
I would stick to a maximum of 10%. In general you can use the fix check/timestep/gran command to check your time step. It also checks the Hertz time.
Regards,
Andi