Nice vid as always. However, this is known as topology optimization and has been around for at least 30 years+. See Bendsøe and Sigmund's book from 2003 that summarises the techniques. It isn't really AI, unless you employ a very broad usage of the term much like conpanies like Autodesk have done to jump on the buzzword bandwagon.
I guess the AI in this case is referring to the thing controlling the roots for those softwares is something like ChatGPT or whatever else to make more unique designs and automate the tedious task of applying the rules that generate the iterations. Unless that's what you're talking about, but I believe it's more to do with LLMs exploring design roots much quicker? Unless I'm missing the point, which I could be, lol
That was my initial thought as well but according to several industry insiders I’ve spoken with, it’s far beyond a rehashing of existing techniques, but rather a move to machine learning being integrated into their implementation- to optimize the process of optimization. In computer graphics, ray tracing is currently experiencing a similar evolution, where decades old techniques are being blended with machine learning to overcome limitations. That being said, I understand your point. It is a TH-cam video and I am limited in how deep I can go on the topic to reveal this nuanced “AI”.
@@NewMind all topology optimization softwares used a form of AI. this is nothing new. It isn't possible WITHOUT some form of machine learning. We've looked at multiple such software programs from companies over the years, and it never pans out for the products we develop. It's a neat tool, but not quite as practical as advertised in many industries. And mind you, I work in the AI industry, on the mechanical engineering side though, but working alongside the software and electrical engineers. And we can't use this software in our designs as it can't optimize much compared to our existing designs, and falls flat at production volumes and prices.
@@NewMind Thanks for the reply! I think I was surprised the term wasn't mentioned - that's maybe all it was on reflection, but indeed there are probably new methods that integrate closer with ML, particularly with regards speeding up FE analysis time and fitness assessment. There was a strand of Topology optimization called ESO and BESO which used evolutionary algorithms back in the day. Ultimately all these terms have kind of been fused together under the term 'Generative design', maybe in some part due to the popularity of GANs in ML (although of course these are very different!).
This reminds me of my workshop. Every time I get a new hand tool or power tool, I wonder how I ever, did without it. Each new tool, either improves my projects or reduces their fabrication time. AI is going to add many, new and use full tools, to a designers toolbox.
00:00 The Zinger 21C hypercar concept represents a milestone in Divergent 3D's goal to mass-produce vehicles with limited or no direct tooling in a fully digital end-to-end integrated system called the Divergent adaptive production system (dapps). 01:10 Divergent 3D uses AI-driven generative design to rapidly create complex and highly optimized structures for their vehicles. 02:32 Designers still play a critical role in defining design goals, interpreting the generated results, and incorporating their creativity and domain expertise into the final design. 03:41 Generative design allows for the production of optimized part designs more rapidly than traditional design processes. 05:16 Generative design can be implemented using different computational approaches, including cellular automata, genetic algorithms, shape grammar, L-systems, and agent-based models. 09:30 Genetic algorithms are effective for optimizing part designs by evaluating fitness based on weight reduction, structural integrity, and manufacturability. 12:04 Shape grammar is well-suited for aesthetic design exploration and repeated structure designs. 13:12 L-systems are used to model the growth and development of complex structures and can be applied to various design domains. 15:45 Agent-based models simulate the behavior and interactions of autonomous agents within a system, allowing for the generation of unique and creative designs. 17:25 Generative design technology is still in its infancy, with various computational models being explored, including neural network-based models and swarm intelligence-based models. 18:23 AI has the potential to completely revamp how entire industries operate, particularly in the world of design, by bringing about once unimaginable capabilities and redefining the roles of engineers and designers.
The marketing guys are having a field day with all those buzzwords and jargon. It might look and sound impressive, but it's more like a contest to see how esoteric and science fiction they can make it sound. LOL
@@asimhussain8716 Some ppl see it as ugly, to me it looks straight out of Aliens or H.R. Geiger's works. Makes sense since it was designed by an alien intelligence.
@@asimhussain8716 Then you assume only biological things can be intelligent. That's the crux of what you're saying. If you can't tell the difference between the burger made by a 5 star gourmet chef that spent a lifetime perfecting his craft and a machine that learned through iterative learning with out being told which is which then they are functionally the same thing. I'm sure horses felt the same way when they were replaced by cars. The soul is used as a crutch to define some indefineable trait or human-ness. One could also use pi or the lemniscate for the same thing. Humans make tools and ai is just another tool. A tool that makes tools and a took that can think of other tools.
@@asimhussain8716 It's not so much that it is an AI that created the part that makes it "ugly" so much as it was the fact that the AI designing the part was given utilitarianism as its one guiding principle. We have seen what AI can do when designing with aesthetics as a goal. Midjourney and Stable Diffusion can produce truly stunning, even profound works of art.
16:22 Nurburgring suspension tuning at home. If you have a simulator with some form of AI driving like Assetto Corsa and an accurate enough aero/weight transfer model you could build the ultimate suspension geometry, spring rate, shock setup, and alignment. Do the tests with varying seat and trunk loads in various weather conditions. Tire wear, grip, and air pressure could also be randomized to test stability in un optimal conditions.
I think Formula 1 will be one of the first to adapt this approach IRL. They have incredibly accurate simulations and are doing constant correlation work, which is quint essential for this approach. You know the phrase: garbage in - garbage out. Without the correlation work, you will only gain parameters that are optimal for the simulation e.g. your car will be really fast in Assetto Corsa, but highly flawed on the real Nurburgring.
As a designer, working in civil jet engine devellopement, i can tell you this is future. And by future, i mean FAR future. Like a whole generation, may be two. But it's a thing we already look at....
Imho a lot of these optimizations aren’t really new but had been simply impossible to produce without modern 3D printing. And the costs are still prohibitive for mass production. Adding some material and simply forging a part can be way cheaper. And most parts fail either due to material defects or unexpected loads where 3D parts are especially vulnerable for the later problem because they simply lack the additional material to deal with it.
I did a little bit of genetic algorithms as well as cellular automata when I studied CS at uni, I can understand why these has not been widely used until know due to large computing requirements associated with their use.
not just that, but it's generally not optimum for high volume production either. I work in teh AI industry, and you can run this software on a desktop computer. it's more issues with production manufacturing, surface finish, practicality of it, etc.
@@SoloRenegade The other issue is that expected force is not always what a vehicle experiences. If a hub hits curbing or the car accidentally goes off the track then the generative AI models fail at a higher rate than conventional parts. It's a cool design, but I think structures and vehicles probably won't utilize these that often in the future. Repairs are also probably impossible making these one and done parts if alignment issues happen from damage or some damage, rust, etc is identified.
@@skoparweaver7692 excellent points. I've seen people mentioning fatigue issues with such parts, and what you just described makes sense in that regard, the unexpected forces it wasn't designed for. I can see it working great for things like the shock bell cranks in a car, and such, since those forces are Extremely controlled and predictable. But the bell crank can be so easily designed/made other ways and be good enough as to not matter anyways. How much weight is really saved in an application like that by using the organic approach? you bring up a lot of good points I hadn't even considered, as we never got into making parts this way due to other issues that far preceded actual use and repair.
By themselves, metaheuristics like Simulated Annealing, Ant colony Optimization or Genetic Algorithms can be as fast as you need them to be. You will just get a solution further from the optimum. Metaheuristics were actually designed to lower computational costs on complex optimization problems (ex: traveling salesman and the whole P != NP stuff). In the end, you can lower the computation time/power cost by lowering the number of iterations and accepting a less optimal solution. It's just that in this specific instance (part design), these algorithms have to compete against well established engineering methodology, so the bar for "acceptable solution" is pretty high and so is the computational cost.
Generative design is awesome when it can be used correctly. If you look at a close-up die shot of modern computer chips, you'll see branching structures in some areas connecting blocks of logic. These are connections and logic made by generative algorithms. The most obvious example I can think of is the cache structure of Zen4C's tiny cores.
Generative algorithms are not intelligence. They are a trial and error process that simply tries everything possible until it finds the best solution. A true AI would find the best solution without trying everything.
@@weistonaski6924 Yes. The links and layout of Zen4C's L2 and L3 caches look like generative design was given a set of weights that heavily penalize taking up extra space and reward capacity more than raw speed.
By saying correctly, you mean effectively? Quz I'm not intended to watch a vid only quz of thumbnail, which depicts totally useless piston in terms of low-as-much cost production due to complicated geometry
I am an FE analyst for more than 2 decades, I am working with topology etc optimisation in Ansys, Nastran since 2 decades. This is new algorithm may be more efficient with time. But both seems to be computationally expesive. Where as GPU processing might help. Seems interesting.
Videos like this further their, over sell under deliver, narrative. These clowns want everybody to believe this is some new never used technology. Like machine learning has not been around for more than two decades.
AI is the new hotness so all companies started chanting "AI! We use AI!" to bump their stock prices and attract investors. Actually, real-world AI-designing use is very limited and done basically only for marketing claim. In the field, advanced real-world parts are designed with the now-old Topology Optimization (presented at 3:40, wrongly claimed to be a "generative design algorithm") that are then tweaked / re-optimized / re-modeled by humans. The most time/brain-power consuming part is NOT the (initial) design in CAD, so saying AI will revolutionize it is completely missing the mark. The difficult part is determining part constraints and design goals, and setting up reasonably-accurate FEA/FEM simulations that complete in reasonable time. There's a lot of fine balancing between the various goals (cost/weight/strength/fatigue/part reuse/ease of manufacture/assembly/repair) including talking to/negotiating with other teams to improve/work around particularly costly constraints. Which means many edits and long simulations. As for AI.. it's mostly hype. Sure, a "smarter" topology optimization algorithm could in theory result in a 3% lighter part for the same cost but only the humans in charge could rewrite the constraints / re-design to simplify an assembly and remove the part altogether. 100% win, the best part is no part.
I can see the work that went into making this stuff look quasi-simple. Really nice. I work as a computational design architect, and it takes sooooo long to make visual explainations that convey these techniques in an interesting, non technical way that goes juuust deep enough to get the point across. Generative design solutions take a lot of investment for research on the front end, and so having a clear explaination of what you're trying to accomplish with the fancy new computer algorithms is really powerful. Thanks again! keep it up.
I’ve been using this for years. Topology optimization. I think the biggest change is the availability of more complex manufacturing methods like metal 3d printing.
@@DigitalDissident in mass production wehicles price for mass production is defining factor. Less machining operations means smaller price. In high end motorsports and aerospace more complex designs are used
It might be interesting to incorporate the growth patterns of small polyp Stoney corals into the generative models. Acropora and Montipora species can be both plating and branching in the same species especially when building a base foundation level. A basing growth pattern will be preferable in early growth stages to produce enough support before branching is prioritised. It is interesting because basing phases have less overall external surface area limiting the ability to intake mineral and nutrients from the water column when compared to the branches which have many more surfaces for polyps to receive the minerals and nutrients they need. Corals are also interesting as they are fauna that incorporate flora for additional energy and nutrient production. Could it be possible to grow buildings in the future that have both solar and water utilisation that can work to more effectively and efficiently become homes for humans? Even have some kind of computational system built into its own pattern of creation so as the structure grows the computational system grows in conjunction with the habitable space.
@@alexandermoody1946 no, as a sort of lichen and coral bio-engineered combo. The lichen part would attach and grow down through existing asphalt/concrete roads and secure the organism to the environment underneath. The coral would grow on top, perhaps with some photo-synthesizing cells mixed in to let it be self-sustaining, and the whole thing would consume rainwater, as well as regulate for temperature, melting off snow and sweating off or radiating heat
@@TheseThreeZs I am sorry to say that I fail to agree with your idea. Have you ever grown corals yourself? There is only one invertebrate that shares many characteristic traits as corals in freshwater and they are hydra which are more closely related to anemones which do not deposit any form of calcium carbonate skeletal structure. The only semi encrusting varieties of anemones that I know of are in the order of Zoantharia and they do not as far as I am aware deposit carbonate skeletons either. The other problem is the mineral composition of rainwater will tend to be very low as the process of evaporation does not readily carry salts or other minerals into the atmosphere. The other interesting thought is that of the responses observed when corals are removed from the water and in effort to protect themselves from both ultra violet radiation and the general atmosphere results in mucus or slime production that would create a real problem for traction. Moreover if a small layer of saltwater was hypothetically used what would inhibit the salt from destroying the land adjacent of any surface that used some kind of bioengineered solution like this in a primary form. The difficulties of removing salt from the land or ground water would have a devastating effect on the environment both short term and long term. I do not know enough about Lichens but from what I have observed the ones I have had first hand experience with are not resistant to breakage and also typically feed by breaking down solid substrate surfaces. There are forms of algae that live commonly in sea water that lay carbonate structures onto substrate medium but they are in general not really compatible with road surface materials plus you have the salt issue. If anything is bioengineered what would the mechanism be to limit growth out in the open environment where current species are unable to adapt in the short term and perhaps the long term as food chain collapses due to monoculture would be apparent? There may be an electrochemical mechanism for laying surfaces like you suggest but I would be inclined to incorporate other technologies into the structure rather than biological organisms. One could propose a network of data or computational conduits through the road network which may help mitigate the urban heat island effect created by data centres and potentially improve the mechanism for laying of cables that are used currently. You could even conceive to construct such a surface like a neural network of sorts. The problem will always be to encourage the importance of keeping a stable biological environment for the organisms that already exist. Nb corals are also very dependent on temperature stability, sea water due to volume does not have dramatic swings in temperature in the short term.
To add I understand the cost to the taxpayer to maintain road infrastructure is not typically appreciated and perhaps solutions are needed so I hope you do not think I have not put great consideration to what you have suggested. Kind regards.
No offensive but i slept good to this video. I put it on before bed. Im rewatching it now. You should make an 8hr video about life and all of its aspects.
Pretty awesome. I love how AI is creating almost alien looking organic optimized solutions to engineering challenges. Imagine what a computer processor might look like using this process.
This is already being done to optimise the layout of a processor's circuitry. Given that all the changes are done to the internal circuitry, they look exactly the same from the user's perspective.
@@TriggerHappyRC1 I would imagine on the scale of nano meter circuits there are some significant limitation due to fabrication technologies. I would love to see what the circuitry design of a CPU looks like with infinite capabilities of fabrication techniques.
@@bobsnabby2298 I think im more inclined to believe the people literally working in the industry who said they are using actual AI, machine learning and neural networks, not just standard iterative optimization.
That's why the theory of evolution is my favorite one. It's not just about biology and carbon, it is all encompassing. Energy gradients creating complexity over time. Thank you so much for the video!
Except that biology doesn’t have a goal but these designed parts do. A fish doesn’t sit around trying to figure out how to make lungs to breathe air so it can walk around for a while before making a blowhole and climbing back into the water to swim around like a whale. Biology has written code on molecules to build living things by assembling molecules using other assembled molecules as the machinery in a factory made out of molecules. That means that they wouldn’t just be designing car parts but also the factories to make them include the energy source for building the parts… and it would also assemble everything… and then the factory could also build more factories. Life is on a whole other level that’s not even on the same planet in the same universe. Personally I don’t know how people can still think that random processes could have created life. I get how Darwin could think that because he thought that a cell was a blob of jelly. Every year we discover that the cell is infinitely more complex than we’d ever imagined.
Honestly if you take an engineering course you will find this is a God sent. Like sure it won't be easy to manufacture some parts but if you give the AI a limited parameter for the machine work you may be able to make a cast aluminium part stronger and lighter then a steel part. Most will say well the alluminium will crack. Yes that's true if it's compromised but if we have multiple load points it should be able to handle multiple fractures till it fails. So if we tell the AI we need at least 13 load points that cab withstand 400 pounds of deformation along the load point. The AI will be forced to make it with the least amount of material and with the most structural rigidity without any plastic or elastic deformation. We give the AI the engineering designs humans cane up with and we say "improve apon what you are given". This will help in the time and money spent and actually increase how well the part functions. I've been useing my own AI that requires 2 cpus and 4 gpus in a server configuration just to funcfunction. I'm planning on makeing the casts myself useing evaporative 3p printing and vacuumed casts useing steel. It's gonna take me some time but I know I can make the world's first jet bike that can fly useing VTOL modes.
I can imagine some of these highly complex, pared down components would be highly unpredictable in their response to fatigue or minor damage compared to their conventional counterparts.
You have to understand that any extra information that could help determine weather a part would have longevity traits can actually be input into these AIs to analyze possibles before they are actually produced. The part is really less about the AI, and more about the variables that were given to analyze.
Wow, with this parts could be created for optimal performance, durability at the ideal price point! Next step is to teach the AI to incorporate structural weaknesses so the part won't live too long... It's all about planned obsolescence after all.
You do realize that 3d printing parts in metal is not really scalable to mass production? It might be possible to 3d print replacements for parts not available anymore. That will probably be more important than using 3d printing as a main manufacturing method.
This stuff has been around in the engineering field for a while, but nobody thought they looked good, so they just designed it into a similar, more appealing shape.
this video is amazing, what i take away from this is that the human brain can be ignorant and think we can achieve a better design than nature and everything around us and this system figured it all out in a very short amount of time. i love hand built and developed parts but if we stand back and take in everything thats around us instead of being egotistical we could all achieve so much more
The problem with 3d printing is the time it takes to produce the part. It takes more time to produce the part compared to standard production methods. One of the production optimizations in traditional production is to minimize the number of cuts needed because every cutting pass increases the amount of time to work the part and can increase the cost beyond the benefits of using less material.
I'm interested to see how "organic" some of these designs will come to look. I'm thinking of how the bones and cavities of the human inner ear look and how you could imagine suspension components having similar shapes.
That's not a thing to aspire to. Biological system are usually horrible, horrible systems. Take the eye: due to the sequence of evolution, the nerve and blood supply for every receptor cell is in front of the cell itself. Evolution produces a series of "good enough" and not finely engineered and tailored solutions.
@@anivicuno9473- Wow. These old, disproved arguments don’t die off very quickly but they keep reproducing themselves. Usually I explain why the eye is not a problem and site more current research to back up my claims but this one is going on a decade now so I’m just going to say this and let you look it up yourself- Your argument would be correct if you were an octopus and lived under water but since you live in air you can be grateful that your eyes have the reverse design of the octopus so they don’t burn out faster than a joint at a rave party.
@@anivicuno9473 lolwut? Think for a moment about the constraints under which biological systems have had to develop. The best design we can come up with for the most familiar example is the boston dynamics stuff - it can do one or two of the thousands of things that a human can do, about as well as a human can. Until you can get that to find its own fuel from its surroundings, repair itself when damaged, reproduce itself in a continuous improvement cycle, and invent a piano and then learn to play it, you might consider giving the good old bio-system a bit more respect.
Other than having to hold AI's hand from a professional over all view point. This right here is what AI is actually good for. Multiple testing with multiple concepts. A human would be hard pressed to run 1000's test scenarios, AI can do that with ease. It may fail with 70% but 10% will be something you didn't think of ..
This is how we got the Revolutionary Cellular Antennas we use today … they weren’t calling it AI then, iterative design was it. Rational design looks more beautiful than practical… Organic.
Love this video so muchhh! I'll re watch to make notes from this video lecture:) Mind opening, FULLY contextual, well build arguments and reasoning is put in this video without which content like this becomes news! I hope more creators put efforts like him...
that's a fun point of view. This should be the best use for Ais because, at it's core, whatit does is streamline a part of the design process that is only valuable to be done via pure parameters. In this example, Ai does not determine or find new base parameter or understand the design constraints, it's not it's strength. The in-between that has been guesswork for all these years is now a thing you can properly do in a purposeful way. This is where the power of Ais truly lies. What's more, these tools do not hinder creativity or innovation because new techniques become new parameters and these ways can totally be reserved for performance-critical design. Anywhere weight/performance/material economy isn't paramount, this type of work becomes a distraction. It is facinating.
I feel like this is one of those things where we will be able go look back and tell when humans were designing robots and when robots were designing robots.
it's crazy how most engineering concepts are a simulation of how the natural world (our reality ) works , that makes me think that maybe we are in a simulation and that we are as humans just a prototype of a more perfect design that's yet to be made ...
As a design engineer, I am soooo excited to start using this technology. It is already showing up in professional industry too for things like CAD workflow optimization.
It’s wild to get a glimpse into the future, the things this AI could change… will welcome it with open arms and then we look back in 50 years and will be unrecognizable, like the Wild West, not that long ago but the technology of their time and how they interacted with it
Am I the only one that noticed the Thumbnail for this video is about the future and it’s a ICE vehicle combustion part 😅. Electric motors are the future.
Perhaps, but unless a breakthrough happens in the battery technology it's just unsustainable economically speaking, just look at rivians earnings last quarter for example (still are negative after many years), and the major manufacturers are also backing away from full BEV and are headed more towards developing hybrid systems. So ICE is definitely here to stay, at least for now.
The general theme of the video is generative design, which is well explained, loved it. But the example used around minute 4:00 is of topology optimization, which is not the same. Topology Optimization is reduction of volume/material of an existing model, versus the CREATION of a new part by only defining the needs and load type/location in the program (as was explained correctly at the begining of the example around minute 3:00), after which it will use generative design to CREATE the model. GD creates a part, TO optimizes and existing one (as the example continued at minute 4:00). Result usually have the same vibe, reason why it's usually referred to as the same thing.
Using an AI to generate possible solutions to design a part is completely fine. This whole idea has flaws such as the solution given by the parameters set is only applicable to that one possible state out of an infinite number. I would give more specific parameters over a set amount of possible states the part being designed can be in. To put it in simple terms, very specific/strict parameters over a general/wide array of possible states.
There's a plethora of requirements that clash with the typical generative design results which are typically bone structure shapes made with additive manufacturing. Definitely not limited to high volume products. Hygienics, roughness, coatings/ corrosion, material characteristics, precision/ tolerances, even certification is a problem because the material is literally built and not processed from stock. Especially printed metals suffer from warpage and internal stresses which can be heat treated but drives cost up even further. Same goes for post machining. Additive manufacturing has been around for quite a while now as well as the generative design softwares. It has found its nice and that's pretty much that. Still nitwits are claiming it will radically change the future.
@@bigbattenberg exactly. it has its uses and applications, but we still rarely see it being used in production volume items after all these years. Just too many issues with this design and manufacturing method as you Well point out. At work, the engineers, myself included, get excited over stuff like this, but then reality quickly sets in and we can't use it.
Just to add to the point made and to drive it home: this stuff is supplemental to engineers' work and is _not_ replacing them, exceptions aside. The future engineer's work, however, would be more about control, review, and oversight of AI-based productions. Well, at least that's my understanding.
Yeah, what it spits out is very organic. It's like nature had the right designs for millions of years and we have been playing with CNC machines and crude shapes for the last 100 until we figured out how to do it nature's way automatically.
now program the ai for a 2nd difference change for design, use these prototypes and then add the parameter of simplicity as in reduce number of overall struts, code for a new way to efficiently mount parts together, could be temporarily backdored by setting a lower and upper limit on part thickness so that space couldn't be made in between parts. then set up a second ai to automatically estimate price based on a standard, then combine those 2 and set a lower and upper limit for cost effectiveness vs design improvement/manufacture ease or even against a mass production model if u wanna go big
This sounds well in theory but a big part of designing components is time of manufacturing. And the costs of running a machine are a lot higher than material costs. Especially for high production volumes. The more complex a shape is the longer it takes to manufacture. And even if you use additive manufacturing a hole that deeds to be smooth or precise needs to be put on a mill again
I know AI is the new hotness, but c'mon. Last I looked, none of those five technologies were considered to fall under the "artificial intelligence" umbrella.
There is a big push to slap the word "AI" on absolutely everything right now, because it sells. I've been playing with nTop lately and it is incredible in lightweighting and topology optimization, which is what most of this is. There is a very narrow definition in this category that can be considered "AI" and that is "generative design". Still, since I have dabbled in programming computers since the 90's to me is nothing but really clever programming. If I was the head of marketing for a company who produces a product that does this sort of work, though, you bet your damn ass I'm slapping the word "AI" all over it.
@@nicog8354 We know what it does, but what @Ice_Karma is saying, and what I agree with, is that that concept isn't really "AI". It's just really nice programming. It's not fast because it's "AI", it's fast because it was optimized well by the devs.
it quite literally is AI. it is machine learning algorithm used to generate optimized designs. if this was done using traditional programming, it would require unimagineable amounts of computing power.
So cool! I know they have used AI like in this designing a next gen rocket thruster nozzle, and with SMD printing, some really alien-like shapes are the result.
There's one problem that I see with the way AI designed internal structure(I am not an engineer, or have any qualifiers making me an expert at any part of this potential condition). Sludge, carbon, and other build up is generally considered inevitable in a combustion engine running off fossil fuels. Parts like connecting rods, crankshafts, pistons, pump turbines, or other rotating assemblies that come in contact with heat, and fluids, and other types of metals, etc. can have uneven, mostly unpredictable build up over time in "capture areas" where AI is attempting to lose weight, while adding strength, this build up could potentially throw many assemblies out of balance. Normally out of balance can be tolerated, and corrected, sometimes in very crude but effective ways, but that's when created by humans where the tolerances are "good enough", with a built in secondary, unintended, but appreciated function that makes them durable. AI wants to create highly accurate, extremely tight tolerance interactions, normally a good thing that leads to much higher performance. I don't know how bad it could get, but I would imagine, considering the AI capable tolerances, that it wouldn't take much to have an effect. I'm still learning about all of this, so it's just an observation, I also have the confidence that we can adapt as we learn, along with the AI we work with.
I'm still watching, but I had a thought and I know I'll forget it by the end of the video (if I wait and see if it was talked about) The AI designs are meant to be the best and most efficient, and we're seeing that they can change dramatically from what people have thought up in the past. What about adding the parameter of what material would work best as well? One shape might be great for a certain type of material (aluminum), but using a stronger metal might be able to make that shape (or a different one) more efficient. Aluminum might need x amount of material and/or mass, while steel might need y amount of material and mass.
Relatively random evolution isnt a bad system, but there are clear drawbacks. Evolution can go down a wrong branch of a tree and not "realise", which will implement drawbacks without noticing. Keeping the touch of human improvement will still be very useful for a while.
When the o1 full model becomes available, people won't need to provide detailed instructions; the AI will determine the best solution from very basic input.
It’s unclear why a talking point is it won’t displace people same as saying a dish washer or washing machine wouldn’t displace workers. When in reality it increased productivity decreases lead times,designs,reduces costs. Why on earth would people be claiming we won’t be putting people out of work. It’s exactly what we want and need. So those people can do more and think about new ideas and designs.
Well, its quite obvious that someone forgot to the tell this specific AI, these parts need to be manufactured by earthly means. ALMOST all the parts in this "vehicle", especially the suspension parts are just impossible to produce (You may have a chance if u mill this out of a solid block, tho). I know those manufacturing restraints, as i spent over a decade as a 3D designer with CAD software on automotive parts. And if you tell AI to consider earthly means of manufacturing and the cost involved, guess what, AI will produce the same ugly, soulless cars which are filling the streets today. And btw, the best cars have already been produced. I swear by god. No need for more useless " hyper hyper". AI will certainly NOT drive manufacturing of mass produced cars. The main driver is COST.
Just imagine combining this technology with other aspects of a customer base. So for example you use 23andme to find a genetic sampling of a local community through that you find the genetic likelihood of somebody being introverted or extroverted along with designs that people are particularly attractive to associated with a certain personality type. Through those findings the A.I. would average out what designs people are likely to find familiar or designs people would love. The A.I. scans the browsing data of the local population that lives within a 5-mile radius of the future building. Uses the cameras on cellphones to figure out when a pupil is dilating when somebody is at a certain heart rate with the data of an Apple watch in tandem with the browsing data to figure out interest. You use that data to subconsciously manipulate the local population by taking subtle design cues from content they like and incorporated into the design of the building. Before you know it factoring multiple invasive data points and now you have created a design that the local population would love.
If this is not sarcastic then you forget that the price of those cars will be extremely high since they're basically customized products at that point. Not to mention the headache sourcing parts for repairs of each locality will be. Also isn't this already done via questionnares for most high volume products? They already have the data and target the lowest common denominator.
This has to be a bit. I know AI bros *love* to dive headfirst into enforcing dystopic nightmare futures on the populace so that they can find a problem for their shitty little algorithm to solve, but this is so plainly on the nose fucked up that I can't fathom anyone wrote it without being at least slightly aware of how horrific it is.
Thanks, very interesting, watched closely. A suggestion - I don't think that your suspension component example around 19:00 worked well because a fundamental constraint of the suspension components is that they would be joined at pivots from the start, and it would waste computation to have them find those constraints through iteration.
Cool. Parts that are more expensive to make, and will cost more when they fail because you cannot replace a single component, but instead need a whole new aforementioned complex and expensive part.
![ลองสุ่มไอโฟน 990 โกงมั้ย? [ โกงมั้ยครับ ep.101 ] | DOM](http://i.ytimg.com/vi/bfsWe0j492k/mqdefault.jpg)
Nice vid as always. However, this is known as topology optimization and has been around for at least 30 years+. See Bendsøe and Sigmund's book from 2003 that summarises the techniques. It isn't really AI, unless you employ a very broad usage of the term much like conpanies like Autodesk have done to jump on the buzzword bandwagon.
I guess the AI in this case is referring to the thing controlling the roots for those softwares is something like ChatGPT or whatever else to make more unique designs and automate the tedious task of applying the rules that generate the iterations. Unless that's what you're talking about, but I believe it's more to do with LLMs exploring design roots much quicker? Unless I'm missing the point, which I could be, lol
correct. and it has limitations due to manufacturing methods, and can often not compete with more traditional methods at high volume production.
That was my initial thought as well but according to several industry insiders I’ve spoken with, it’s far beyond a rehashing of existing techniques, but rather a move to machine learning being integrated into their implementation- to optimize the process of optimization. In computer graphics, ray tracing is currently experiencing a similar evolution, where decades old techniques are being blended with machine learning to overcome limitations. That being said, I understand your point. It is a TH-cam video and I am limited in how deep I can go on the topic to reveal this nuanced “AI”.
@@NewMind all topology optimization softwares used a form of AI. this is nothing new. It isn't possible WITHOUT some form of machine learning.
We've looked at multiple such software programs from companies over the years, and it never pans out for the products we develop. It's a neat tool, but not quite as practical as advertised in many industries.
And mind you, I work in the AI industry, on the mechanical engineering side though, but working alongside the software and electrical engineers. And we can't use this software in our designs as it can't optimize much compared to our existing designs, and falls flat at production volumes and prices.
@@NewMind Thanks for the reply! I think I was surprised the term wasn't mentioned - that's maybe all it was on reflection, but indeed there are probably new methods that integrate closer with ML, particularly with regards speeding up FE analysis time and fitness assessment. There was a strand of Topology optimization called ESO and BESO which used evolutionary algorithms back in the day. Ultimately all these terms have kind of been fused together under the term 'Generative design', maybe in some part due to the popularity of GANs in ML (although of course these are very different!).
This reminds me of my workshop. Every time I get a new hand tool or power tool, I wonder how I ever, did without it. Each new tool, either improves my projects or reduces their fabrication time. AI is going to add many, new and use full tools, to a designers toolbox.
Nope.... But keep drinking the marketing buzz word bingo... Loser.
lots of bot-generated posts here . . .
in this situation the 3d printing tech is doing way more the the "AI" is.
Dude, the amount of hard work and passion that goes into your work is noticed and HIGHLY appreciated.
I second this! ❤
Not even, he doesn't even know if the software actually uses AI
Thanks fake praise comment from an alt account!
Very highly!!
nah he just used chat gpt if he likes ai so much
00:00 The Zinger 21C hypercar concept represents a milestone in Divergent 3D's goal to mass-produce vehicles with limited or no direct tooling in a fully digital end-to-end integrated system called the Divergent adaptive production system (dapps).
01:10 Divergent 3D uses AI-driven generative design to rapidly create complex and highly optimized structures for their vehicles.
02:32 Designers still play a critical role in defining design goals, interpreting the generated results, and incorporating their creativity and domain expertise into the final design.
03:41 Generative design allows for the production of optimized part designs more rapidly than traditional design processes.
05:16 Generative design can be implemented using different computational approaches, including cellular automata, genetic algorithms, shape grammar, L-systems, and agent-based models.
09:30 Genetic algorithms are effective for optimizing part designs by evaluating fitness based on weight reduction, structural integrity, and manufacturability.
12:04 Shape grammar is well-suited for aesthetic design exploration and repeated structure designs.
13:12 L-systems are used to model the growth and development of complex structures and can be applied to various design domains.
15:45 Agent-based models simulate the behavior and interactions of autonomous agents within a system, allowing for the generation of unique and creative designs.
17:25 Generative design technology is still in its infancy, with various computational models being explored, including neural network-based models and swarm intelligence-based models.
18:23 AI has the potential to completely revamp how entire industries operate, particularly in the world of design, by bringing about once unimaginable capabilities and redefining the roles of engineers and designers.
Bruh
Haha that was a good idea
The marketing guys are having a field day with all those buzzwords and jargon. It might look and sound impressive, but it's more like a contest to see how esoteric and science fiction they can make it sound. LOL
We're about to finally have the weird organo-metallic futuristic stuff we always wanted,
I still like the old boxy 80's supercars or pointy tail fin of the 60's way more then this futuristic style could ever achieve.
@@asimhussain8716 Some ppl see it as ugly, to me it looks straight out of Aliens or H.R. Geiger's works. Makes sense since it was designed by an alien intelligence.
Everything any Corporate does is for money, power, duplication, and/or optimization weather it works for them or not.@@asimhussain8716
@@asimhussain8716 Then you assume only biological things can be intelligent. That's the crux of what you're saying. If you can't tell the difference between the burger made by a 5 star gourmet chef that spent a lifetime perfecting his craft and a machine that learned through iterative learning with out being told which is which then they are functionally the same thing. I'm sure horses felt the same way when they were replaced by cars.
The soul is used as a crutch to define some indefineable trait or human-ness. One could also use pi or the lemniscate for the same thing.
Humans make tools and ai is just another tool. A tool that makes tools and a took that can think of other tools.
@@asimhussain8716 It's not so much that it is an AI that created the part that makes it "ugly" so much as it was the fact that the AI designing the part was given utilitarianism as its one guiding principle. We have seen what AI can do when designing with aesthetics as a goal. Midjourney and Stable Diffusion can produce truly stunning, even profound works of art.
16:22 Nurburgring suspension tuning at home. If you have a simulator with some form of AI driving like Assetto Corsa and an accurate enough aero/weight transfer model you could build the ultimate suspension geometry, spring rate, shock setup, and alignment. Do the tests with varying seat and trunk loads in various weather conditions. Tire wear, grip, and air pressure could also be randomized to test stability in un optimal conditions.
I think Formula 1 will be one of the first to adapt this approach IRL. They have incredibly accurate simulations and are doing constant correlation work, which is quint essential for this approach. You know the phrase: garbage in - garbage out. Without the correlation work, you will only gain parameters that are optimal for the simulation e.g. your car will be really fast in Assetto Corsa, but highly flawed on the real Nurburgring.
As a designer, working in civil jet engine devellopement, i can tell you this is future. And by future, i mean FAR future. Like a whole generation, may be two. But it's a thing we already look at....
This is just the old but gold Topology optimization. AI can just be used to speed up the process, but AI is the hot thing this days, so...
AI is "just" optimization anyway. Even very big models are just incremental optimization.
He put "AI" for "Machine Learning" in the title. AI is more specific than ML and he wasn't really talking about that.
@@Arnaz87other way around. Machine learning is a method by which an AI can be achieved.
@@Arnaz87 yeah, today every single instance of neural network being called AI.
Imho a lot of these optimizations aren’t really new but had been simply impossible to produce without modern 3D printing. And the costs are still prohibitive for mass production. Adding some material and simply forging a part can be way cheaper.
And most parts fail either due to material defects or unexpected loads where 3D parts are especially vulnerable for the later problem because they simply lack the additional material to deal with it.
I did a little bit of genetic algorithms as well as cellular automata when I studied CS at uni, I can understand why these has not been widely used until know due to large computing requirements associated with their use.
not just that, but it's generally not optimum for high volume production either. I work in teh AI industry, and you can run this software on a desktop computer. it's more issues with production manufacturing, surface finish, practicality of it, etc.
@@SoloRenegade The other issue is that expected force is not always what a vehicle experiences. If a hub hits curbing or the car accidentally goes off the track then the generative AI models fail at a higher rate than conventional parts. It's a cool design, but I think structures and vehicles probably won't utilize these that often in the future. Repairs are also probably impossible making these one and done parts if alignment issues happen from damage or some damage, rust, etc is identified.
@@skoparweaver7692 excellent points. I've seen people mentioning fatigue issues with such parts, and what you just described makes sense in that regard, the unexpected forces it wasn't designed for.
I can see it working great for things like the shock bell cranks in a car, and such, since those forces are Extremely controlled and predictable. But the bell crank can be so easily designed/made other ways and be good enough as to not matter anyways. How much weight is really saved in an application like that by using the organic approach?
you bring up a lot of good points I hadn't even considered, as we never got into making parts this way due to other issues that far preceded actual use and repair.
also ... false local maxima of fitness functions, not an easily solved issue in this context.
By themselves, metaheuristics like Simulated Annealing, Ant colony Optimization or Genetic Algorithms can be as fast as you need them to be. You will just get a solution further from the optimum. Metaheuristics were actually designed to lower computational costs on complex optimization problems (ex: traveling salesman and the whole P != NP stuff).
In the end, you can lower the computation time/power cost by lowering the number of iterations and accepting a less optimal solution. It's just that in this specific instance (part design), these algorithms have to compete against well established engineering methodology, so the bar for "acceptable solution" is pretty high and so is the computational cost.
Generative design is awesome when it can be used correctly. If you look at a close-up die shot of modern computer chips, you'll see branching structures in some areas connecting blocks of logic. These are connections and logic made by generative algorithms. The most obvious example I can think of is the cache structure of Zen4C's tiny cores.
Generative algorithms are not intelligence. They are a trial and error process that simply tries everything possible until it finds the best solution. A true AI would find the best solution without trying everything.
You mean the routing inside the chip linking logic elements ?
@@weistonaski6924 Yes. The links and layout of Zen4C's L2 and L3 caches look like generative design was given a set of weights that heavily penalize taking up extra space and reward capacity more than raw speed.
By saying correctly, you mean effectively? Quz I'm not intended to watch a vid only quz of thumbnail, which depicts totally useless piston in terms of low-as-much cost production due to complicated geometry
I am an FE analyst for more than 2 decades, I am working with topology etc optimisation in Ansys, Nastran since 2 decades.
This is new algorithm may be more efficient with time.
But both seems to be computationally expesive. Where as GPU processing might help.
Seems interesting.
Videos like this further their, over sell under deliver, narrative. These clowns want everybody to believe this is some new never used technology. Like machine learning has not been around for more than two decades.
AI is the new hotness so all companies started chanting "AI! We use AI!" to bump their stock prices and attract investors. Actually, real-world AI-designing use is very limited and done basically only for marketing claim.
In the field, advanced real-world parts are designed with the now-old Topology Optimization (presented at 3:40, wrongly claimed to be a "generative design algorithm") that are then tweaked / re-optimized / re-modeled by humans. The most time/brain-power consuming part is NOT the (initial) design in CAD, so saying AI will revolutionize it is completely missing the mark. The difficult part is determining part constraints and design goals, and setting up reasonably-accurate FEA/FEM simulations that complete in reasonable time. There's a lot of fine balancing between the various goals (cost/weight/strength/fatigue/part reuse/ease of manufacture/assembly/repair) including talking to/negotiating with other teams to improve/work around particularly costly constraints. Which means many edits and long simulations.
As for AI.. it's mostly hype. Sure, a "smarter" topology optimization algorithm could in theory result in a 3% lighter part for the same cost but only the humans in charge could rewrite the constraints / re-design to simplify an assembly and remove the part altogether. 100% win, the best part is no part.
I can see the work that went into making this stuff look quasi-simple. Really nice. I work as a computational design architect, and it takes sooooo long to make visual explainations that convey these techniques in an interesting, non technical way that goes juuust deep enough to get the point across. Generative design solutions take a lot of investment for research on the front end, and so having a clear explaination of what you're trying to accomplish with the fancy new computer algorithms is really powerful. Thanks again! keep it up.
Very cool! Apart from anything else, the aesthetics of those organic looking parts is just delicious.
Reminds me of Organics, a weapon for Cloud in FF7.
The most precious gift we can offer anyone is our attention. When mindfulness embraces those we love, they will bloom like flowers.
AI designs it, 3D printing makes it. Match made in heaven.
I’ve been using this for years. Topology optimization. I think the biggest change is the availability of more complex manufacturing methods like metal 3d printing.
If you work in automotives, sdo you know where one can find optimized wheel hubs such as the ones shown at 4:40 in this video?
@@DigitalDissident in mass production wehicles price for mass production is defining factor. Less machining operations means smaller price.
In high end motorsports and aerospace more complex designs are used
Notice how the parts look like muscular structures
It might be interesting to incorporate the growth patterns of small polyp Stoney corals into the generative models. Acropora and Montipora species can be both plating and branching in the same species especially when building a base foundation level. A basing growth pattern will be preferable in early growth stages to produce enough support before branching is prioritised. It is interesting because basing phases have less overall external surface area limiting the ability to intake mineral and nutrients from the water column when compared to the branches which have many more surfaces for polyps to receive the minerals and nutrients they need. Corals are also interesting as they are fauna that incorporate flora for additional energy and nutrient production.
Could it be possible to grow buildings in the future that have both solar and water utilisation that can work to more effectively and efficiently become homes for humans? Even have some kind of computational system built into its own pattern of creation so as the structure grows the computational system grows in conjunction with the habitable space.
Or self -healing carbon offsetting roads?
@@TheseThreeZs as carbonate?
@@alexandermoody1946 no, as a sort of lichen and coral bio-engineered combo. The lichen part would attach and grow down through existing asphalt/concrete roads and secure the organism to the environment underneath. The coral would grow on top, perhaps with some photo-synthesizing cells mixed in to let it be self-sustaining, and the whole thing would consume rainwater, as well as regulate for temperature, melting off snow and sweating off or radiating heat
@@TheseThreeZs I am sorry to say that I fail to agree with your idea. Have you ever grown corals yourself?
There is only one invertebrate that shares many characteristic traits as corals in freshwater and they are hydra which are more closely related to anemones which do not deposit any form of calcium carbonate skeletal structure. The only semi encrusting varieties of anemones that I know of are in the order of Zoantharia and they do not as far as I am aware deposit carbonate skeletons either. The other problem is the mineral composition of rainwater will tend to be very low as the process of evaporation does not readily carry salts or other minerals into the atmosphere. The other interesting thought is that of the responses observed when corals are removed from the water and in effort to protect themselves from both ultra violet radiation and the general atmosphere results in mucus or slime production that would create a real problem for traction. Moreover if a small layer of saltwater was hypothetically used what would inhibit the salt from destroying the land adjacent of any surface that used some kind of bioengineered solution like this in a primary form. The difficulties of removing salt from the land or ground water would have a devastating effect on the environment both short term and long term. I do not know enough about Lichens but from what I have observed the ones I have had first hand experience with are not resistant to breakage and also typically feed by breaking down solid substrate surfaces. There are forms of algae that live commonly in sea water that lay carbonate structures onto substrate medium but they are in general not really compatible with road surface materials plus you have the salt issue. If anything is bioengineered what would the mechanism be to limit growth out in the open environment where current species are unable to adapt in the short term and perhaps the long term as food chain collapses due to monoculture would be apparent? There may be an electrochemical mechanism for laying surfaces like you suggest but I would be inclined to incorporate other technologies into the structure rather than biological organisms. One could propose a network of data or computational conduits through the road network which may help mitigate the urban heat island effect created by data centres and potentially improve the mechanism for laying of cables that are used currently. You could even conceive to construct such a surface like a neural network of sorts. The problem will always be to encourage the importance of keeping a stable biological environment for the organisms that already exist.
Nb corals are also very dependent on temperature stability, sea water due to volume does not have dramatic swings in temperature in the short term.
To add I understand the cost to the taxpayer to maintain road infrastructure is not typically appreciated and perhaps solutions are needed so I hope you do not think I have not put great consideration to what you have suggested. Kind regards.
No offensive but i slept good to this video. I put it on before bed. Im rewatching it now. You should make an 8hr video about life and all of its aspects.
Pretty awesome. I love how AI is creating almost alien looking organic optimized solutions to engineering challenges.
Imagine what a computer processor might look like using this process.
This is already being done to optimise the layout of a processor's circuitry. Given that all the changes are done to the internal circuitry, they look exactly the same from the user's perspective.
This is not AI.
@@TriggerHappyRC1 I would imagine on the scale of nano meter circuits there are some significant limitation due to fabrication technologies. I would love to see what the circuitry design of a CPU looks like with infinite capabilities of fabrication techniques.
Stop using word AI in a standard iteration process and optimization.
@@bobsnabby2298 I think im more inclined to believe the people literally working in the industry who said they are using actual AI, machine learning and neural networks, not just standard iterative optimization.
That's why the theory of evolution is my favorite one. It's not just about biology and carbon, it is all encompassing. Energy gradients creating complexity over time. Thank you so much for the video!
Except that biology doesn’t have a goal but these designed parts do. A fish doesn’t sit around trying to figure out how to make lungs to breathe air so it can walk around for a while before making a blowhole and climbing back into the water to swim around like a whale.
Biology has written code on molecules to build living things by assembling molecules using other assembled molecules as the machinery in a factory made out of molecules.
That means that they wouldn’t just be designing car parts but also the factories to make them include the energy source for building the parts… and it would also assemble everything… and then the factory could also build more factories. Life is on a whole other level that’s not even on the same planet in the same universe. Personally I don’t know how people can still think that random processes could have created life. I get how Darwin could think that because he thought that a cell was a blob of jelly. Every year we discover that the cell is infinitely more complex than we’d ever imagined.
Honestly if you take an engineering course you will find this is a God sent. Like sure it won't be easy to manufacture some parts but if you give the AI a limited parameter for the machine work you may be able to make a cast aluminium part stronger and lighter then a steel part. Most will say well the alluminium will crack. Yes that's true if it's compromised but if we have multiple load points it should be able to handle multiple fractures till it fails. So if we tell the AI we need at least 13 load points that cab withstand 400 pounds of deformation along the load point. The AI will be forced to make it with the least amount of material and with the most structural rigidity without any plastic or elastic deformation.
We give the AI the engineering designs humans cane up with and we say "improve apon what you are given". This will help in the time and money spent and actually increase how well the part functions.
I've been useing my own AI that requires 2 cpus and 4 gpus in a server configuration just to funcfunction. I'm planning on makeing the casts myself useing evaporative 3p printing and vacuumed casts useing steel. It's gonna take me some time but I know I can make the world's first jet bike that can fly useing VTOL modes.
I can imagine some of these highly complex, pared down components would be highly unpredictable in their response to fatigue or minor damage compared to their conventional counterparts.
You have to understand that any extra information that could help determine weather a part would have longevity traits can actually be input into these AIs to analyze possibles before they are actually produced. The part is really less about the AI, and more about the variables that were given to analyze.
Wow, with this parts could be created for optimal performance, durability at the ideal price point!
Next step is to teach the AI to incorporate structural weaknesses so the part won't live too long... It's all about planned obsolescence after all.
You do realize that 3d printing parts in metal is not really scalable to mass production?
It might be possible to 3d print replacements for parts not available anymore. That will probably be more important than using 3d printing as a main manufacturing method.
I'd love to use this in my homemade car build a decade from now. An organic looking car frame would be sick.
This stuff has been around in the engineering field for a while, but nobody thought they looked good, so they just designed it into a similar, more appealing shape.
Unless your 3D printing its also almost impossible and expensive to make such an organic shape.
this video is amazing, what i take away from this is that the human brain can be ignorant and think we can achieve a better design than nature and everything around us and this system figured it all out in a very short amount of time. i love hand built and developed parts but if we stand back and take in everything thats around us instead of being egotistical we could all achieve so much more
The problem with 3d printing is the time it takes to produce the part. It takes more time to produce the part compared to standard production methods. One of the production optimizations in traditional production is to minimize the number of cuts needed because every cutting pass increases the amount of time to work the part and can increase the cost beyond the benefits of using less material.
Hench Why 3d Printing is Perfect for the Individual but for traditional Manufacturing will be for mass Production.
those constraints such as amount of cuts and production time can or will also be factored into these gererative models, next
Really gorgious presentation and well documented. The amount of effort and the choice of subject is a marvell piece..
I'm interested to see how "organic" some of these designs will come to look. I'm thinking of how the bones and cavities of the human inner ear look and how you could imagine suspension components having similar shapes.
That's not a thing to aspire to. Biological system are usually horrible, horrible systems. Take the eye: due to the sequence of evolution, the nerve and blood supply for every receptor cell is in front of the cell itself. Evolution produces a series of "good enough" and not finely engineered and tailored solutions.
@@anivicuno9473 make a better eye than
@@anivicuno9473- Wow. These old, disproved arguments don’t die off very quickly but they keep reproducing themselves. Usually I explain why the eye is not a problem and site more current research to back up my claims but this one is going on a decade now so I’m just going to say this and let you look it up yourself-
Your argument would be correct if you were an octopus and lived under water but since you live in air you can be grateful that your eyes have the reverse design of the octopus so they don’t burn out faster than a joint at a rave party.
@@anivicuno9473 lolwut? Think for a moment about the constraints under which biological systems have had to develop. The best design we can come up with for the most familiar example is the boston dynamics stuff - it can do one or two of the thousands of things that a human can do, about as well as a human can. Until you can get that to find its own fuel from its surroundings, repair itself when damaged, reproduce itself in a continuous improvement cycle, and invent a piano and then learn to play it, you might consider giving the good old bio-system a bit more respect.
@anivicuno9473 the eye is also backwards :3
Other than having to hold AI's hand from a professional over all view point. This right here is what AI is actually good for. Multiple testing with multiple concepts. A human would be hard pressed to run 1000's test scenarios, AI can do that with ease. It may fail with 70% but 10% will be something you didn't think of ..
I always love it when the algorithm finds some thing that I didn’t know I needed to watch this was awesome
Im glad the organic designing of our vehicles has begun
would love for this to end with all of us being super self reliant and not just having major monopolies that overcharge
This is how we got the Revolutionary Cellular Antennas we use today … they weren’t calling it AI then, iterative design was it.
Rational design looks more beautiful than practical… Organic.
Design is easy. Manufacturing is hard.
This is absolute poetry. Well done. These concepts are so exciting as to the potential.
This is one of the most interesting and thought provoking videos I have seen in a long long while.
Agreed! I'm in awe
Love this video so muchhh! I'll re watch to make notes from this video lecture:)
Mind opening, FULLY contextual, well build arguments and reasoning is put in this video without which content like this becomes news! I hope more creators put efforts like him...
This Hass to be one of my favorite videos I've seen in quite some time! So many great topics covered in the visual representations are top-notch.
that's a fun point of view. This should be the best use for Ais because, at it's core, whatit does is streamline a part of the design process that is only valuable to be done via pure parameters. In this example, Ai does not determine or find new base parameter or understand the design constraints, it's not it's strength. The in-between that has been guesswork for all these years is now a thing you can properly do in a purposeful way. This is where the power of Ais truly lies. What's more, these tools do not hinder creativity or innovation because new techniques become new parameters and these ways can totally be reserved for performance-critical design. Anywhere weight/performance/material economy isn't paramount, this type of work becomes a distraction. It is facinating.
I feel like this is one of those things where we will be able go look back and tell when humans were designing robots and when robots were designing robots.
What you described gone above my head but I successfully completed full video.
Very well done. It gave me a more clear sense of how generative functions are designed and what results can occur.
As a mechanical design engineer, I am both excited and scared of this revolutionary technology.
You scared?
AI is amazing, i love how they will improve our design and thus accelerate humanity technological growth
Fantastic technology. What about repairability? Simplicity is the invention of Mother. Complexity is costly and a pain to fix.
It's just the shape of a part. If you bust a connecting rod you buy a new one and replace it. You don't weld them back together.
I thought this was a sensationalist channel, subbed for great video quality and storytelling
Now that's art in motion, well done chicanos customs! I really think she could win in her class
it's crazy how most engineering concepts are a simulation of how the natural world (our reality ) works , that makes me think that maybe we are in a simulation and that we are as humans just a prototype of a more perfect design that's yet to be made ...
As a design engineer, I am soooo excited to start using this technology. It is already showing up in professional industry too for things like CAD workflow optimization.
bro you just took youtube videos to another level
Finally someone made the video i was waiting for this from the day i saw the czinger 21C
I love how organal its structures are
And no cars or anything is created this way. This says everything about this revolutonary concept.
It’s wild to get a glimpse into the future, the things this AI could change… will welcome it with open arms and then we look back in 50 years and will be unrecognizable, like the Wild West, not that long ago but the technology of their time and how they interacted with it
Am I the only one that noticed the Thumbnail for this video is about the future and it’s a ICE vehicle combustion part 😅. Electric motors are the future.
Perhaps, but unless a breakthrough happens in the battery technology it's just unsustainable economically speaking, just look at rivians earnings last quarter for example (still are negative after many years), and the major manufacturers are also backing away from full BEV and are headed more towards developing hybrid systems. So ICE is definitely here to stay, at least for now.
Nature surely masters this art, the wings of Dragonflies are engineering little wonders
The general theme of the video is generative design, which is well explained, loved it. But the example used around minute 4:00 is of topology optimization, which is not the same. Topology Optimization is reduction of volume/material of an existing model, versus the CREATION of a new part by only defining the needs and load type/location in the program (as was explained correctly at the begining of the example around minute 3:00), after which it will use generative design to CREATE the model. GD creates a part, TO optimizes and existing one (as the example continued at minute 4:00). Result usually have the same vibe, reason why it's usually referred to as the same thing.
Using an AI to generate possible solutions to design a part is completely fine. This whole idea has flaws such as the solution given by the parameters set is only applicable to that one possible state out of an infinite number. I would give more specific parameters over a set amount of possible states the part being designed can be in. To put it in simple terms, very specific/strict parameters over a general/wide array of possible states.
generative design is also not viable for all products. high volume production can be cost prohibitive compared to a traditional tooled design.
There's a plethora of requirements that clash with the typical generative design results which are typically bone structure shapes made with additive manufacturing. Definitely not limited to high volume products. Hygienics, roughness, coatings/ corrosion, material characteristics, precision/ tolerances, even certification is a problem because the material is literally built and not processed from stock. Especially printed metals suffer from warpage and internal stresses which can be heat treated but drives cost up even further. Same goes for post machining.
Additive manufacturing has been around for quite a while now as well as the generative design softwares. It has found its nice and that's pretty much that. Still nitwits are claiming it will radically change the future.
@@bigbattenberg exactly. it has its uses and applications, but we still rarely see it being used in production volume items after all these years. Just too many issues with this design and manufacturing method as you Well point out.
At work, the engineers, myself included, get excited over stuff like this, but then reality quickly sets in and we can't use it.
Just to add to the point made and to drive it home: this stuff is supplemental to engineers' work and is _not_ replacing them, exceptions aside.
The future engineer's work, however, would be more about control, review, and oversight of AI-based productions.
Well, at least that's my understanding.
As a mechanic working on anything with a throttle since I could remember, it is truly unsettling to see these familiar parts look so...biological.
Yeah, what it spits out is very organic. It's like nature had the right designs for millions of years and we have been playing with CNC machines and crude shapes for the last 100 until we figured out how to do it nature's way automatically.
now program the ai for a 2nd difference change for design, use these prototypes and then add the parameter of simplicity as in reduce number of overall struts, code for a new way to efficiently mount parts together, could be temporarily backdored by setting a lower and upper limit on part thickness so that space couldn't be made in between parts. then set up a second ai to automatically estimate price based on a standard, then combine those 2 and set a lower and upper limit for cost effectiveness vs design improvement/manufacture ease or even against a mass production model if u wanna go big
This sounds well in theory but a big part of designing components is time of manufacturing. And the costs of running a machine are a lot higher than material costs. Especially for high production volumes. The more complex a shape is the longer it takes to manufacture. And even if you use additive manufacturing a hole that deeds to be smooth or precise needs to be put on a mill again
I know AI is the new hotness, but c'mon. Last I looked, none of those five technologies were considered to fall under the "artificial intelligence" umbrella.
There is a big push to slap the word "AI" on absolutely everything right now, because it sells. I've been playing with nTop lately and it is incredible in lightweighting and topology optimization, which is what most of this is. There is a very narrow definition in this category that can be considered "AI" and that is "generative design". Still, since I have dabbled in programming computers since the 90's to me is nothing but really clever programming. If I was the head of marketing for a company who produces a product that does this sort of work, though, you bet your damn ass I'm slapping the word "AI" all over it.
@@nicog8354 We know what it does, but what @Ice_Karma is saying, and what I agree with, is that that concept isn't really "AI". It's just really nice programming. It's not fast because it's "AI", it's fast because it was optimized well by the devs.
it quite literally is AI. it is machine learning algorithm used to generate optimized designs. if this was done using traditional programming, it would require unimagineable amounts of computing power.
Wow I hadn’t seen a great video like this in a while, kudos!
i absolutely adore this use of A.I.
This is absolutely mind blowing.
Fascinating
So cool! I know they have used AI like in this designing a next gen rocket thruster nozzle, and with SMD printing, some really alien-like shapes are the result.
I am so excited to be able to print off replacement parts for my 2011 911
There's one problem that I see with the way AI designed internal structure(I am not an engineer, or have any qualifiers making me an expert at any part of this potential condition). Sludge, carbon, and other build up is generally considered inevitable in a combustion engine running off fossil fuels. Parts like connecting rods, crankshafts, pistons, pump turbines, or other rotating assemblies that come in contact with heat, and fluids, and other types of metals, etc. can have uneven, mostly unpredictable build up over time in "capture areas" where AI is attempting to lose weight, while adding strength, this build up could potentially throw many assemblies out of balance. Normally out of balance can be tolerated, and corrected, sometimes in very crude but effective ways, but that's when created by humans where the tolerances are "good enough", with a built in secondary, unintended, but appreciated function that makes them durable. AI wants to create highly accurate, extremely tight tolerance interactions, normally a good thing that leads to much higher performance. I don't know how bad it could get, but I would imagine, considering the AI capable tolerances, that it wouldn't take much to have an effect. I'm still learning about all of this, so it's just an observation, I also have the confidence that we can adapt as we learn, along with the AI we work with.
Thank you very much, this Tek video is the best I saw for a very long time, excellent work. 🙏❤️
finaly, this is how engineering goes forward.
I'm still watching, but I had a thought and I know I'll forget it by the end of the video (if I wait and see if it was talked about)
The AI designs are meant to be the best and most efficient, and we're seeing that they can change dramatically from what people have thought up in the past. What about adding the parameter of what material would work best as well? One shape might be great for a certain type of material (aluminum), but using a stronger metal might be able to make that shape (or a different one) more efficient. Aluminum might need x amount of material and/or mass, while steel might need y amount of material and mass.
I hope they get back to Customer oriented design soon.
Relatively random evolution isnt a bad system, but there are clear drawbacks. Evolution can go down a wrong branch of a tree and not "realise", which will implement drawbacks without noticing. Keeping the touch of human improvement will still be very useful for a while.
those structures look sci-fi alien
When the o1 full model becomes available, people won't need to provide detailed instructions; the AI will determine the best solution from very basic input.
It’s unclear why a talking point is it won’t displace people same as saying a dish washer or washing machine wouldn’t displace workers. When in reality it increased productivity decreases lead times,designs,reduces costs. Why on earth would people be claiming we won’t be putting people out of work. It’s exactly what we want and need. So those people can do more and think about new ideas and designs.
That is the cloud or everything in every person thought funnels to come up with the perfect seam design. Awesome 👍👍❤
showinga new piston is like showing a new Horse Shoe.
Generative design and topology optimization are two different things 😊😊
they never encount for fatigue, this is just a smart talk
Damn, really good video man. you covered so many different topics so well!
As always excellent video great research and very interesting topic. Thank you.
Sounds like the software runs experiments for you and its up to you to make something out of the data. A capable tool in the right hands.
you earned a new subscriber man! amazing work
Truth, and goodness, and beauty are but different faces of the same all.
Thank you wholeheartedly.
This is absolutely brilliant.
Well, its quite obvious that someone forgot to the tell this specific AI, these parts need to be manufactured by earthly means. ALMOST all the parts in this "vehicle", especially the suspension parts are just impossible to produce (You may have a chance if u mill this out of a solid block, tho). I know those manufacturing restraints, as i spent over a decade as a 3D designer with CAD software on automotive parts. And if you tell AI to consider earthly means of manufacturing and the cost involved, guess what, AI will produce the same ugly, soulless cars which are filling the streets today. And btw, the best cars have already been produced. I swear by god. No need for more useless " hyper hyper". AI will certainly NOT drive manufacturing of mass produced cars. The main driver is COST.
You're good, surprised you aren't on Nebula with the other "better than professional production studio" guys.
Actually started watching it and said, oh I should check nebula 😅
fantastic quality
Justifying people's jobs at this point really is just, "we still want a faulty human element involved in this process that could be instantaneous."
Just imagine combining this technology with other aspects of a customer base. So for example you use 23andme to find a genetic sampling of a local community through that you find the genetic likelihood of somebody being introverted or extroverted along with designs that people are particularly attractive to associated with a certain personality type. Through those findings the A.I. would average out what designs people are likely to find familiar or designs people would love. The A.I. scans the browsing data of the local population that lives within a 5-mile radius of the future building. Uses the cameras on cellphones to figure out when a pupil is dilating when somebody is at a certain heart rate with the data of an Apple watch in tandem with the browsing data to figure out interest. You use that data to subconsciously manipulate the local population by taking subtle design cues from content they like and incorporated into the design of the building. Before you know it factoring multiple invasive data points and now you have created a design that the local population would love.
If this is not sarcastic then you forget that the price of those cars will be extremely high since they're basically customized products at that point. Not to mention the headache sourcing parts for repairs of each locality will be. Also isn't this already done via questionnares for most high volume products? They already have the data and target the lowest common denominator.
23 and me will be your downfall
This has to be a bit. I know AI bros *love* to dive headfirst into enforcing dystopic nightmare futures on the populace so that they can find a problem for their shitty little algorithm to solve, but this is so plainly on the nose fucked up that I can't fathom anyone wrote it without being at least slightly aware of how horrific it is.
you should apply for a Black Mirror scenarist position
@@maitele yeah it's amazing what connecting the dots can do.
Thanks, very interesting, watched closely. A suggestion - I don't think that your suspension component example around 19:00 worked well because a fundamental constraint of the suspension components is that they would be joined at pivots from the start, and it would waste computation to have them find those constraints through iteration.
In theory if only mounting plane of suspention joints is defined it can make iterations by moving them around
Cool. Parts that are more expensive to make, and will cost more when they fail because you cannot replace a single component, but instead need a whole new aforementioned complex and expensive part.