InteractiveComputerGraphics
disappear issue of fluid particles
Hi,
Just found a weired issue. In DFSPH, rigid-fluid coupling (2012 method) simulation, sometimes the fluid particle will disappear with Nan value.
This issue is more serious in a large scene multiple rigids. I attached my recording .html for your convenience.
Do you know how to fix it?
https://drive.google.com/file/d/1sQPzb6GW77_RgXVEVqfSaSgA-ObSONX1/view?usp=drive_link attached .html. It record 32 rendered results with the searched DFSPH parameters.
Thank you for your reply!
I double test the attached scene, it will lose particles after a few time steps using dfsph + 2012 rigid-fluid coupling method. But it's stable (no particles disappear) when I integrate a rigid body control method.
Please check the scene and the state. As for state I only load the water state by position and velocity.
I also attached my running code for your convenience. To run my code, it require set in diffFR environment: https://github.com/zhehaoli1999/DiffFR
For me the initial configuration directly explodes since the rigid bodies are stuck in the fluid which causes high pressure forces.
For me the initial configuration directly explodes since the rigid bodies are stuck in the fluid which causes high pressure forces.
Thanks! Now I can confirm it's the high pressure forces issue. So, high pressure forces -> high fluid speed / nan velocity -> nan position.
Yes. And this issue will also happen (~10% Subjective) when I use PPO to control a single rigidbody to swim. And it's (~50% Subjective) when I use multiple (9-11) rigid bodys. And it's (~90%) for the shared scenes.
I also see another issue in your scene. You enforce a constant step size of 5 ms. In this case the body might move too close to the fluid particles or even penetrate and the CFL condition cannot prevent this.
Thank you for your reply and this information!
I hope to set a fixed time step in my project (rigid-fluid coupling), may I know what time step is a good option top in your head?
This depends on your scene. If you have small fast moving particles, you should take a small time step. If the particles are all slow, you can use a larger step size. A particle should not slip through another one. This is essentially what the CFL condition prevents.