5 ms·
For the people confused with how AI is used for fluid dynamics, you solve PDEs to know how your fluid behaves, and PINNs ( physics-informed neural network) can
by TimVulcanus 4y ago
For the people confused with how AI is used for fluid dynamics, you solve PDEs to know how your fluid behaves, and PINNs ( physics-informed neural network) can be used to solve these PDEs. They have limitations in their uses and have been only recently re-launched, but they have improvements compared to the usual solving approach.
- bhouston 4y agoIs this using PINNs? https://en.wikipedia.org/wiki/Physics-informed_neural_networks https://en.wikipedia.org/wiki/Physics-informed_neural_networ... It is Unity so I guess one could just decompile the asset. BTW on this page, it doesn't say AI used for the main NS, but rather just for object interaction: https://zibra.ai/zibra-liquids-pro/#tariffs https://zibra.ai/zibra-liquids-pro/#tariffs. E.g. "neural collider"
- TimVulcanus 4y agoI read a bit more about the unity plug-in used here and they in fact aren't using PINNs for solving. Here they more or less explain how they do it : https://zibra.ai/blog/what-can-be-done-with-zibra-liquids/ https://zibra.ai/blog/what-can-be-done-with-zibra-liquids/
- amelius 4y agoIs this like using a preconditioner in Krylov based methods?
- dauertewigkeit 4y agoNo, PINNs are precisely what you do NOT want to use for this kind of thing, because you are trying to take shortcuts that mimic the physics without actually caring about it being physically accurate (as long as it looks nice). Reading their approach, they are not actually using machine learning methods for solving fluid dynamics, but are using it for more efficient processing of boundary interactions.
- elcritch 4y agoThats a very interesting take. Setting up boundary conditions is always a pain and often not physically accurate anyway.
- adgjlsfhk1 4y agomost of the literature on pinns is garbage because it uses comparisons to really bad implementations of the "usual approach"