Fluidic Pinball dynamic #94
Comments
|
Alright, when running the simulation again with an higher noise Vorticity field which looks good I guess. However I don't get why there is explosion of the dynamics when applying random control forcing when starting with a null vector field after 40 time steps. I could soon provided more details on the controls applied. I am going to:
|
|
Yes, this looks about like what you might expect. Nothing jumps out at me as being obviously wrong here, except that you're having trouble with the random forcing. I think initializing from a more natural condition like steady state or post-transient flow is a good idea though. If you have trouble with the coarse meshes, just keep in mind that these aren't considered "validated" at all and are basically for debugging purposes, so if the numerics are bad I would move to the default mesh ("fine" in this case). I'll leave this open in case you want to report what you find with the other initializations. |
|
Actually, this might be a bug from my side, I am about to disantangle this. Thanks for all your answers:) |
|
Just remembered this - we did also end up finding a bug in the control terms for the pinball, so #132 (still in review at the moment) might help if you're still having any trouble with this. I'm going to close out for now, unless you have other issues after we get that merged. |
Hello:)
As not so much examples are given for the Fluidic Pinball, I would like to have your feedback on those elements:
First, as @jcallaham suggested for the Cavity Flow Fluctuation KE objective for cavity & step flows #51 and since the Fluidic Pinball at$Re=130$ is chaotic I chose to derive the steady-state solution and add Gaussian noise on it $\mathcal{N}(0, 10^{-2})$ (otherwise if initialised with a null vector field the simulation could blow up as Cavity but I am not sure it is that critical in this case).
Second, I chose a$\delta_t^{IPCS} = 10^{-3}$ , I am not sure it is the right one and references often omit to mention the used values. Consequently, the flow is qualitatively constant after $t=10s$ and the KE value somehow converges. What is the best strategy to quickly obtain the chaotic fluctuating flow ? Larger noise, most instable eigenvalue (but eigenvalue part is not trivial...) ?
Vorticity field
Velocity field
I started to investigate this flow because I observed exploding dynamic when starting to apply control forcing directly (sampling random actions to initialise a replay-buffer) from the transient flow initialised with a null vector field (I guess it is null by default with UFL?) through the$Re=130$ with a coarse grid.
stepinterface atPS: sorry this might be moved in discussions ?
The text was updated successfully, but these errors were encountered: