Wow, A Simulation That Matches Reality! 🤯

❤️ Check out Lambda here and sign up for their GPU Cloud: lambdalabs.com...
📝 The paper "Fast and Versatile Fluid-Solid Coupling for Turbulent Flow Simulation" is available here:
www.geometry.ca...
❤️ Watch these videos in early access on our Patreon page or join us here on KZbin:
- / twominutepapers
- / @twominutepapers
Credits from the paper:
Ansys Inc. 3D meshes were provided by GrabCAD users Aisak (Fig. 1), Mehmet Boztaş (turbine blade in Fig. 20), Vedad Saletovic (turbine tower in Fig. 20), Dhanasekar Vinayagamoorthy (Fig. 22), and CustomWorkx Belgium (Fig. 25), as well as Sketchfab users Cosche (Fig. 4), Opus Poly (Fig. 21), and Lexyc16 (Fig. 27).
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bryan Learn, Christian Ahlin, Eric Martel, Gordon Child, Ivo Galic, Jace O'Brien, Javier Bustamante, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Mark Oates, Michael Albrecht, Michael Tedder, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Peter Edwards, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: / twominutepapers
Thumbnail background design: Felícia Zsolnai-Fehér - felicia.hu
Wish to watch these videos in early access? Join us here: / @twominutepapers
Meet and discuss your ideas with other Fellow Scholars on the Two Minute Papers Discord: discordapp.com...
Károly Zsolnai-Fehér's links:
Instagram: / twominutepapers
Twitter: / twominutepapers
Web: cg.tuwien.ac.a...

[@epicthief]

There is so much love for the scientists who focus on efficiency, this paper is drenched in sweat from squeezing

[@FireyDeath4]

Nah I think it's scrunched plot twist: the paper is made of titanium and there are nano bots facilitating the deformation accurately in real time to a degree of 0.999999217711794934

[@TwoMinutePapers]

🤲📜

[@JrIcify]

"This airplane simulation took only minutes. How many minutes? 60." He should be in marketing or politics if he ever gets bored of light transport research

[@pineapplerindm]

@@nagualdesign Rome was built in seconds!

[@sebastianjost]

I think when you have some context of usual simulation times in the field, that is actually an accurate and useful statement. Even the compared techniques shown in the video took much longer and got worse results. So getting similar results probably took many hours before. The technique being able to do that within a few minutes to a couple of hours is still much better than (probably) waiting tens of hours for similar results.

[@njdotson]

@@sebastianjost and some of the simulations took less than an hour. Even if it took 75 minutes for example, it would be more practical to measure in minutes compared to hours

[@u.v.s.5583]

It took seconds. How many seconds? About 3.6k

[@zrmsraggot]

I'm out of small time units to make you laugh. Can I get my thumbs up still ?

[@QuixEnd]

Now THIS is the level of aerodynamic Sim we've wanted for ages. They've got it in professional labs, but its not nearly this easy to use

[@Jaepeaa]

video games are about to peak in the next couple years, It has been truly amazing to see the tech evolve over the last 20 years, truly amazing and the applications driven by video games are endless for science!

[@theresa_sweetheart]

all 14 of them

[@w花b]

@@Jaepeaa I don't think having this realistic of a simulation is useful for most games. You want to just have an approximation and unless it's a simulation like a car or plane simulator, I don't think it's going to be a priority.

[@VoxelMusic]

I love these videos. Tbh, this is one of the only smoke sims ive seen that actually had me second guessing myself. Usually you can tell when smokes not moving naturally, but this is really good. Way too good for the amount of time it takes.

[@10thletter40]

One hour for a simulation that is relatively accurate? That's crazy. Plus, the flow following the object's curve isnt easy to predict all the time

[@Quantum_Beyblados]

2 more papers down the line, we'll be able to make it in real life with a good grapics card. (I hope)

[@_John_P]

1 hour running on what?

[@nathanholyland9493]

@@_John_P that is the question. I presume a pretty high spec graphics card atleast

[@nathanholyland9493]

Mabye even multiple

[@mihailmilev9909]

@@_John_P good question

[@RogerFrigola]

I can't help but feel that this paper is somewhat oversold. It would have been nice to see a numerical comparison of these results vs a traditional Computational Fluid Dynamics simulation run at low resolution. If the claim is that this algorithm can be useful for design/engineering, there needs to be a numerical assessment of the simulation fidelity, not just pretty graphics.

[@jietao8870]

Unfortunately lots of ppl only care about fancy pictures. It it even more dangerous getting an answer looking good but actually wrong.

[@NicholasMati]

Agreed. I'm especially sceptical given the fact that it's a structured grid. It will likely fall apart in problems where global behavior is determined by very small structures outside of the boundary layer that can't be refined and properly captured. Speaking of which, they also didn't mention what turbulence model they were using. I pray it's not a DNS.

[@thomasludwigkaiser2701]

@@NicholasMati I think it is SPH, since he frequently refers to video games.

[@NicholasMati]

​@@thomasludwigkaiser2701: if you look at time stamp 1:04 (kzbin.info/www/bejne/rGnRZX17gq-lqLc), it shows a grid resolution of 400x100x200, so I assumed it was a structured grid, probably just convecting a scalar to do the coloring. I've worked on the finite element CFD code PHASTA and am familiar with the formulation of finite volume and finite difference codes, but have never really looked at smoothed-particle hydrodynamics. From Wikipedia, "it is a meshfree Lagrangian method" which would seem to be inconsistent with having a grid - at least from my very limited understanding.

[@olistiktok]

It surely looks good, but, for industry (automotive, aerospace, energy and so on) applications, the problem is rarely to distinguish qualitative differences in simulated flow, rather very sensitive/precise/barely visible differences in quantities. Pressure, velocity, vorticity fields, coupling frequencies, etc etc. If these new models can't bring something new in that extent and only remain in the visual/qualitative domain, for now they will hardly be useful anywhere else than for new (visualy amazing) creative CG animations.

[@LokiScarletWasHere]

I’ll have to dig into the info about this when I wake up, but those colors you were seeing in the wind tunnel simulation indicated air density/pressure (or velocity magnitute? I dunno, I’m no physicist, but what I do know is the colors indicate important values, which is the point here). It would stand to reason you could pick out quantitative information from any point you wanted. Maybe this is available to play with.

[@h00db01i]

could be useful even if it merely delivers approximations efficiently

[@SSJ3Tim]

This is very true, although it's worth noting the error-vs-resolution convergence plot in the paper (Fig. 16) shows that - for that particular case - it is able to reach higher levels of accuracy than ANSYS Fluent, a commercial CFD solver. Fluent is more accurate on coarser grids, but reaches an accuracy plateau. In general it's important that progress is made on both fronts, model physical accuracy and computational efficiency, because better efficiency means we can scale up to larger problems and/or refine the mesh for more numerical accuracy.

[@chadd_robertson]

@@LokiScarletWasHere You're correct; the colours are representative of pressure, fluid velocity, etc. However, just because these quantities are present does not mean they are at all accurate. To determine their validity, we would need to compare the simulated results to experimental data extensively. The complexity of fluid simulations can not be overstated. Our understanding of the physics is still developing, especially when considering transient and turbulent flow conditions (what I was taught two years ago is officially outdated).

[@chadd_robertson]

@@SSJ3Tim I've not read the paper yet, but knowing that these comparisons have been made has definitely put it on my to-do list.

[@nathanholyland9493]

Wow. So many new papers so fast. I am excited to see this one.

[@HighTher3]

Crazy! I can't wait for 2 papers down the line. Engineers utilizing this are going to do great things

[@sanikku7359]

I hope this paper gets implemented into a easily accessible CFD program soon! Aircraft design is one of my biggest interests and being able to see how they actually behave with the air quickly is a VERY helpfull thing.

[@Xero_Wolf]

Us: OMG only 60 mins? Sooo fast! People 5 years in the future: Yea fast if you running it on a toaster lol

[@oo7squid]

I'm always skeptical when there's no quantitative results presented alongside the pictures. Without a Cd compared with known results, it is just pictures.

[@mvmlego1212]

"CD"?

[@acruzp]

@@mvmlego1212 Try asking in a friendlier way and we might be bothered to explain

[@hfkssadfrew]

Lol I am not surprised for someone does not even know cd to comment here

[@ruandurand3971]

@@mvmlego1212 Coefficient of drag.

[@mvmlego1212]

@@ruandurand3971 -- Thank you.

[@WikiSnapper]

I want to put 3d models through those simulations now so bad!

[@JustinianH]

I used to study Lattice Boltzmann models back for my bachelors degree, it's a so cool topic I'm glad to see it at the forefront of fluid dynamics, and nostalgic I didn't pursue it further ...

[@KuroSamaMG]

I was just starting to investigate and study about exactly this for my own simulation agine! I'm so glad i found this! thank you for the info, as always!

[@adamhenriksson6007]

It would be interesting to add this simulation into a machine learning training loop for generating aerodynamic structures. More funky AI generated car designs please.

[@slitthroat6209]

Yay two minute papers

[@mihailmilev9909]

Yay

[@StrangerHappened]

*Doctor, the prior method from 2020, which does incorrect simulation, is presented as if it was the only one.* Such simulations are done accurately for many years with multiple methods beyond the one wrong cited.

[@skrimper]

It wasn't presented as if it was the only one. "Beyond the one wrong cited"

[@StrangerHappened]

@@skrimper How it was not? The video shows a new method and compares it to an old method with pointing out how it was inaccurate and how now it is done correctly. This is not an accurate picture. I guess Doctor did not notice that in the script, which might happen sometimes as he is focused on the things right in front of him at the moment.

[@antipastamony]

@@StrangerHappened my assumption is that they chose to reference that paper in particular because it produced results in a similar time span, though I agree that the way it was presented should have addressed this. the way it was put in the script was somewhat misleading, though I'd imagine this was unintentional. an algorithm that focuses on accuracy alone without respect for calculation time would of course be more accurate than the algorithm described in this video, but wouldn't be a particularly meaningful comparison. this paper is about balancing computation time and accuracy with a slight preference towards the latter and as such it makes the most sense to pit it against a paper with similar goals.

[@TriSamples]

Wow. I’d love to see an AI off the back of this work calculate shapes that reduce or eliminate turbulent flow. Now that’s next level.

[@SSJ3Tim]

Look into adjoint-based shape optimization, I've seen such techniques applied to airfoil design! Although it's worth noting that often turbulent flow is desirable, as it reduces drag relative to laminar flow. It's flow separation and wake drag that you really want to reduce.

[@StephenRoseDuo]

Very impressed it can simulate vortex lift!

[@caiocc12]

The miracle is when these vortices on the wings were discovered and the Concorde was designed computers as we know didn't exist yet, everything done by hand or physical models.

[@SBImNotWritingMyNameHere]

This is both ridiculously amazing and beautiful Thank you for making these videos :)

[@mateuszabramek7015]

There is one parameter that could make car simulation more close to reality namely car vs wind speed. It changes the distance between upper part of the wind tunnel and car body.

[@Shooting_Gaming_Channel]

Been working on aerodynamics and CFD simulation. Such a beauty. Wish I could re run my research stuff om this method 😶

[@Roriloty]

Wow two videos in one day you are the best!

[@PeterRichardsandYoureNot]

I just can’t get over the dude’s delivery. It’s like listening to Peter Laurie invite you to a dinner with his master in Dracula’s castle.

[@Ceelvain]

I've been watching this channel for years. And I haven't yet had the heart to tell Károly that I'm not a scalar, despite what he says in the intro.

[@JMPDev]

Fantastic work. I wonder how many more papers that iterate on this before we can see comparable quality in real-time.

[@thomasamassey]

Is this implemented in any software yet, if so where can i get my hands on it?

[@tHEuKER]

I love your videos. Every paper you present is more incredible than the previous one. That said, I suffer from anxiety and something in the way you talk triggers my anxiety, so I'm forced to watch your videos with no sound, using closed captions instead. I know it's not easy to change one's ways, but it would mean the world to me if you didn't make a pause as if there was a comma every three words, as that would surely enable me to watch your videos as intended. Sorry if I sound rude or annoying. Hope you understand. Keep up the good work.

[@julians3danimations]

Featured on Blender Guru's newsletter again! Congrats!

[@Fogmeister]

Ok… I don’t know if this is the same for anyone else but in this video I’ve literally learned for the first time what those little spoilers on road cards actually do. 😂

[@schumzy]

Awesome explanations of use cases!

[@johnnysparkleface3096]

I love your channel. You always have incredible videos.

[@Paulo_Dirac]

4:48 "So we can go from idea to conclusion in a matter of 1h" well from what you've said it is not exactly true... as I understand it, it would allow to go from one idea to another in a matter of an hour, then the "final" idea would need to be tested "old fashion" -> about a week. So from idea to conclusion would need around a week instead of around one or many mounth.

[@ZongHoops]

I am thinking the following question might be replied before: how do you reproduce the simulation from a paper?

[@smorty3573]

I think this would be amazing for simulating planes with realistic air particles and airflow.

[@shriharir6450]

In which software were these simulations made on?

[@hedayetulislam8211]

which software is use for that kind of simulation ?

[@cwk18]

what is the software they are using? Ansys, Altair, Abacus?

[@jacobrogers2214]

This has become my favorite channel.

[@PrinceWesterburg]

Blender Devs: Oh crap, the roadmaps just changed again!

[@JustsomeSteve]

2 Videos in 1 Day? What a time to be alive!

[@Jewelsonn]

The fact that it doesn't account for pressure and temperature of surrounding and object itself makes it look a bit off... if your paper can implement that.. it might give you a good result.

[@kotsariniskonstantinos7440]

Are there any quantitive comparisons from this solver and experimental data? For example drag coefficients? It would be very interesting to see such a comparison!

[@MatrixPermutation]

this is really amazing considering that simulating chaotic systems normally equals extreme amount of resources

[@OayxYT]

I wonder if this can help with more accurate weather forecasting. That would be neat

[@bluesonkel]

I could watch these turbulence simulators all day long

[@ruandurand3971]

It would help If you include the flow field mesh, as that plays an incredible critical role in accuracy. Is a mesh dependent study performed? Judging by the thickness of that boundary layer on the hood, it is a very coarse mesh. Is there any numerical validation? As these CFD simulations are not time averaged I don't even want to know the computational resources needed to determine the time steps (This is probably why the coarse mesh). Just because it "looks nice" does not mean it is accurate. If this can be used as a preliminary design tool like any panel code or empirical code, sure that's fine.

[@parthsavyasachi9348]

Finally colorful fluid dynamics is here.

[@edocor8081]

Man, this time is getting better and better, when it comes to being alive

[@zackmackeil3754]

I really want to see what one could do with a Deep GAN and this simulator to create as effective an airplane as possible

[@markoj3512]

Please show us more data about it. It‘s nice to have colorful dynamics. But we need more reliable data for comparison than funny pictures. Real cfd codes use benchmarks to compare their results with experiments.

[@Elrog3]

Yes, and something about algorithmic complexity or computation required rather than just simulation time which varies with the hardware its run on.

[@fluxophile]

*4x Two Minute Papers in one week, feels like Christmas

[@linkrobertsonus9347]

Excellent video. Hold on to your papers!

[@BryanAndKareem]

i wonder what implications this paper will have on the construction of energy harvesting devices involving wind. hopefully we can use it to optimize harvesting efficiency by utilizing multidirectional wind strengths. i imagine a rotating base at the bottom of the windmill. it may cost more at first but later down the line it will save much more.

[@zain786ification]

How does it compare to Ansys fluent? Is it finite volume? Does it mesh the elements or it does not , I guess it uses LBM method and meshless solver , what's the name of the software? Does it solve in cloud ?

[@Infinit3Enigma]

"Hold on to your papers ".. I love that saying

[@hilo7729]

Truly what a time to be alive

[@ianliu88]

Eyeballing two images isn't exactly a measure of accuracy, though.

[@AndreasZachariou]

it's not meant for pin point accuracy, rather testing if the behavior is going in the right direction we want which this does. Of course if they want accurate results they're gonna need to do a proper sim but this already lets you see the essence of the results you can expect

[@ianliu88]

@@AndreasZachariou to what degree? If you want to improve the design of a wind turbine, or an airplane, I bet you already need accurate simulation.

[@getsideways7257]

@@ianliu88 This would make a great flight sim... with some dumbing down and if it could run in real time. For anything serious, on the other hand, it's not extremely helpful - but the doc said it himself that it's more of a "preview" than an actual analysis tool.

[@ma_nu]

the "reality check" wasn't even the same car...

[@mamo987]

amazing!!! your content inspires

[@Bruno-cb5gk]

Is there any way to try this out?

[@mvmlego1212]

I've never even taken a class in fluid dynamics, so I assume that I'm in the wrong here, but thought that 1) the phenomenon seen in the lower-left panel at 2:35 is known as "vortex shedding", and 2) that vortex shedding increases drag, not decreases it.

[@rhushikmatroja5909]

This is mind blowing!

[@arcadealchemist]

we're getting there soon.

[@apeckx5090]

I can't wait for 2 papers down the line when this runs in real time!

[@sidjindal]

Can i use this for free or is it not publicly available?

[@shanewilson3653]

This software can only be truly tested by comparing results with testing done during Major Hardware's Fan showdown.

[@epiczeven6378]

Man... my dream is to have a all-in high fidelity simulation virtual world, to play around :D

[@DogeMultiverse]

laughs in metaverse

[@theairaccumulator7144]

@@DogeMultiverse metaverse is a low poly online game, where is the high fidelity simulation?

[@epiczeven6378]

@@theairaccumulator7144 exactly! I want a real-time simulation at the atomic level. Let´s hope we don´t have to wait until super quantic computers for that :(

[@serta5727]

Awesome 🤩

[@davidberger5745]

Could you please go back to giving a short overview of *how* that stuff works, not only the results?

[@Newenu]

With these videos you can understand the future

[@liamkelly3333]

What software the guy at 4:56 using, really would love to know!

[@lamarrrrr]

I have riddle. What can move without making rustle of the leaf?

[@vilchico]

I love this channel

[@abdelhakkhalil7684]

I wish you could share with us the hardware spec too. 60 min is highly relative to the kind of hardware.

[@michaelleue7594]

It takes seconds of skimming the paper to find what you want. How many seconds? 60. All simulations presented in this paper were conducted with our fluid simulator platform, on a workstation with a 14-core Intel Xeon E5-2690 CPU, 128 GB of memory and up to two NVIDIA RTX 3090 GPUs.

[@mattheModest]

Holy... shit. This really changes the game.

[@WildGamez]

What degree do u need to get into this field?

[@BursaProduction]

Awesome. Addon for blender??

[@김아무개-y2v]

Do you know CFD? Abaqus Ansys? Gemens NX?

[@bimDe2024]

How do they create such simulation? Using NVidia's Physx?

[@Mr35diamonds]

Pretty odd that Boltzmann models, that is, kinetic or particle based modelling across complex geometries take so little time! Rivalling with an FVM solver (Fluent) for 2d Taylor-Couette flow! That’s actually very interesting.

[@russhicksart]

LBM is certainly well suited to low-speed flows but the computational expense quickly becomes prohibitive with increasing Mach number, hence you will still see most aerospace companies continuing to utilize RANS for these applications.

[@tiagotiagot]

Does anything changes if they use it to simulate things at smaller scales, like the airflow inside a computer case?

[@nehalhosalikar29]

Excuse me, but what software is this?

[@Groveish]

What is the software used for the graphics of this simulation? ANSYS Fluent looks nothing like this!

[@0xdeadbabe240]

"Visualization was achieved, depending on the examples, through cross sections showing the color-coded magnitude of the velocity field, or through smoke visualization using smoke particle tracers advected by the resulting vector field on the GPU and rendered by Redshift" ~ from the paper.

[@marc_frank]

does it calculate forces, too?

[@maxborkowski717]

It took 1 hour but how long was the simulation? I wanna know the FPS damnit! Or in this case probably SPF :D

[@theairaccumulator7144]

The raw data is calculated first, frames come later.

[@xE92vD]

Looks like it won't be that long before games get like real simulation of objects and like real graphics.

[@Mr.President]

man, MSFS 2030 is gonna be so sick!

[@serta5727]

I believe this to be one of the best things I saw today 😍

[@AngelLestat2]

Now they can add all these outputs to train an ANN and achieve similar results in seconds.

[@danuvip]

Two Minute Papers Merch: "PAPER"

[@ManjinderSinghHanjra]

You, my brother, are a magician!

[@brainsanitation]

"only minutes, SIXTY minutes" thats just a full hour

[@nightmisterio]

Need a water sim to test boat hulls

[@iTube4U]

will this be released to the public or still in labs and beta ?

[@TundeEszlari]

Nagyon jó lett a videó.

[@WillyLangley]

Nice Video!

[@jaygaikwad8173]

On which softwares were these simulations performed, I would like to know in detail if anyone has an idea.

[@OktoPutsch]

Have a look at Xflow

[@cgbasic808]

im confused. what are these made in? what is rendering it? what makes the output?