Thanks for the talk! ML in CFD is fascinating! Turbulence certainly follows patterns that are not intuitive to unterstand, but ML models can figure it out and extrapolate a better solution from small resolution. With proper validation that is a powerful tool when memory is limited.
@@theastuteangler for example horseshoe vortices for turbulent flow over a surface, and turbulence follows the Kolmogorov scale. The flow field is correlated on small length scales but not on large, this makes it predictable at least for small time and length scales. It's not entirely chaotic.
@@ProjectPhysX isnt any system predictable at a small enough scale? That is to say, for example, with enough inputs any output can be determined? This sounds like a grossly idealized approach, the most spherical of all spherical cows. If we look at anything with a strong enough microscope, we can assume what may happen next right down to quarks and leptons - but once we pull back and begin to include influences continuously, does the deterministic nature of PDEs and the like not break down? I imagine a deck of cards. Taking the whole deck, you cannot determine what card is next. But identifying the beginning state, the precise shuffling and outcome of that, and which cards have come before (being eliminated from the deck), the next card can be determined. We have scoped in to a high degree. Scope out back to the realistic state: a deck of cards whose suits and values are unknown and cannot be known, we cannot determine the next card. I hope I am making my point clear. I am a physics enthusiast, having not been trained at the highest levels. My mathematics is strong however, though I favour intuition and creativity over rote and rigor.
@@theastuteangler not quite; quantum mechanics is fundamentally unpredictable. The thing with turbulence is, it's averages are well predictable over large length scales and large time periods. Its fluctuations are somewhat predictable on small scales. And there are repeating patterns on medium scales. You can't predict it for indefinitely long though, as eventually the tinyest fluctuations from QM will surface as macroscopic chaos. BTW, a cow is quite different from a sphere aerodynamically, I just simulated it in CFD :'D ua-cam.com/video/VyxMZ2vS3dI/v-deo.html
It's quite an interdisciplinary field :) I'm coming from physics, but I also know many mathematicians, computer scientists, engineers and hydrologists who do fluid dynamics in some way. I wouldn't say that one background is better than the other, either way there is a lot to learn and to gain :)
Those are good for applied fluid mechanics - for instance running CFD for an airplane. They are also decent paths into this type of fluid mechanics (less applied, more theoretical). I would suggest applied math, maybe with a few engineering classes, if that's you want to do more theoretical fluid mechanics.
Watched your interview with Jousef on the same topic and you were very positive on ability to reconstruct the flows in near wall flows (boundary layers). This is probably the biggest challenge we face with our AI PIV software. Are there any papers we can read on the proper reconstruction in near wall, you could suggest?
CFD? More like “This has gotta be”…one of the most interesting videos I’ve seen in awhile!
¡Excelente video Ricardo! Me estaba preguntando cuando aparecerías en uno de estos.
@@rvinuesa y me imagino que muy productiva también 😉
Nice lecture, could you explain what is latent space?
Thanks for the talk! ML in CFD is fascinating! Turbulence certainly follows patterns that are not intuitive to unterstand, but ML models can figure it out and extrapolate a better solution from small resolution. With proper validation that is a powerful tool when memory is limited.
Turbulence is chaotic and by definition chaos is that without pattern, so exactly which patterns do turbulence follow?
@@theastuteangler for example horseshoe vortices for turbulent flow over a surface, and turbulence follows the Kolmogorov scale. The flow field is correlated on small length scales but not on large, this makes it predictable at least for small time and length scales. It's not entirely chaotic.
@@ProjectPhysX isnt any system predictable at a small enough scale? That is to say, for example, with enough inputs any output can be determined? This sounds like a grossly idealized approach, the most spherical of all spherical cows. If we look at anything with a strong enough microscope, we can assume what may happen next right down to quarks and leptons - but once we pull back and begin to include influences continuously, does the deterministic nature of PDEs and the like not break down? I imagine a deck of cards. Taking the whole deck, you cannot determine what card is next. But identifying the beginning state, the precise shuffling and outcome of that, and which cards have come before (being eliminated from the deck), the next card can be determined. We have scoped in to a high degree. Scope out back to the realistic state: a deck of cards whose suits and values are unknown and cannot be known, we cannot determine the next card. I hope I am making my point clear. I am a physics enthusiast, having not been trained at the highest levels. My mathematics is strong however, though I favour intuition and creativity over rote and rigor.
@@theastuteangler not quite; quantum mechanics is fundamentally unpredictable. The thing with turbulence is, it's averages are well predictable over large length scales and large time periods. Its fluctuations are somewhat predictable on small scales. And there are repeating patterns on medium scales. You can't predict it for indefinitely long though, as eventually the tinyest fluctuations from QM will surface as macroscopic chaos.
BTW, a cow is quite different from a sphere aerodynamically, I just simulated it in CFD :'D ua-cam.com/video/VyxMZ2vS3dI/v-deo.html
@@ProjectPhysX Good to see you here Moritz ;-)
Is mechanical engineering the best branch for fluid mechanics or aerospace engineering or something else?
It's quite an interdisciplinary field :)
I'm coming from physics, but I also know many mathematicians, computer scientists, engineers and hydrologists who do fluid dynamics in some way. I wouldn't say that one background is better than the other, either way there is a lot to learn and to gain :)
Application of fluids exist everywhere in the engineering and science sectors.. you just have to pick your poison!!
I would say naval architecture or ocean engineering if you want to study hydrodynamics
Those are good for applied fluid mechanics - for instance running CFD for an airplane. They are also decent paths into this type of fluid mechanics (less applied, more theoretical). I would suggest applied math, maybe with a few engineering classes, if that's you want to do more theoretical fluid mechanics.
is the simulation done in OpenFoam?
@@rvinuesa ohh, got it,Thanks for sharing.
How would you prove the universality of the autoencoder? How is it different from guessing the results or simply remembering everything?
yes it has been proved
good technique
Hey, I know that guy! :D
Recorded a podcast with Ricardo here: ua-cam.com/video/TOfwf4ffPnU/v-deo.html
Watched your interview with Jousef on the same topic and you were very positive on ability to reconstruct the flows in near wall flows (boundary layers). This is probably the biggest challenge we face with our AI PIV software. Are there any papers we can read on the proper reconstruction in near wall, you could suggest?
@@rvinuesa Thanks a lot!
hi