Hallo everybody!

I'm simulating a closed gas spring (Piston-Cylinder combination, no valves, no inlet/outlet) underlying sinusoidal compression/expansion. Movement of one boundary (Piston surface) is modeled by a simple expression, making the boundary Piston make cyclic displacements (BottomDeadCentre - TopDeadCentre) and, thus, moving and deforming the mesh. The simulation works (convergence of RMS=10^-4 is reached only with extremely small timesteps - ~10^-7s; with moderate timestep of 5e^-4 RMS is ~10^-2), but what I can't resolve is why the timestep has no relation to the frequency that my Piston is moving with. Namely, if I run it with 300 RPM and a timestep of 5e^-4, it runs and I get the results, with a lot of "numerical noise" though, as RMS is as mentioned 10^-2 - 10^-3, but it works. But if I run it at 2 RPM, the lowest timestep still has to be 0.001s or 5e^-4, otherwise the simulation stops after a couple of time steps. Cycle in the first example takes 0.2s, while in the later 30s. Thus the conclusion there is no relation between the Piston (and mesh) displacement in time and the actual time discretization!

Any ideas??

Hi,

CFX is an implicit solver so the following comment is not strictly speaking a constraint, but the principle is certainly a guide.

For explicit solvers in incompressible flow it is well known the Courant number is a constraint on the size timestep which is stable, CN=U*DT/DX where U is the fluid velocity, DT is the timestep and DX is the mesh spacing.

For compressible flow the constraint is the Courant Freidrichs-Lewy criteon (CFL), where CFL=(U+a)*DT/DX where a is the local acoustic velocity.

In your flow the velocities are small compared to the acoustic velocity, so even though you run the simulation physically faster, the CFL number does not change much so the allowable timestep is roughly the same.

Regards, Glenn Horrocks