Video: Closing the Loop: How Generative Design and Digital Manufacturing Are Accelerating Product Development​ | Duration: 3296s | Summary: Closing the Loop: How Generative Design and Digital Manufacturing Are Accelerating Product Development​ | Chapters: Introduction & Welcome (18.645s), Steering Knuckle Components (102.05s), Engineering Process (186.625s), Generative Design (261.78s), Generative vs Traditional Design (356.185s), Traditional CAD Challenges (462.655s), Generative Design Process (590.78s), Design for Manufacturing (793.315s), Performance Benefits (952.18s), Testing and Metrology (1077.995s), Generative Design Applications (1220.105s), Thermal Camera Retrofit (1444.185s), Thermal Imaging Applications (1799.145s), Conversational Design Future (2004.535s), Questions and Closing (2191.125s), Pricing and Production (2420.625s), Design Software Tools (2619.855s), Generative Design Software (2718.11s), Post-Processing Techniques (2931.135s), Q&A and Wrap-up (3090.175s), Closing & Contact (3242.98s)
Transcript for "Closing the Loop: How Generative Design and Digital Manufacturing Are Accelerating Product Development​": Hello, everyone. My name is Matthew Verwick. I'm the founder and principal engineer of Pathfinder Labs. Today, we're going to be having a discussion called Closing the Loop, How Generative Design and Digital Manufacturing are Accelerating the Product Development Process. We're going to go over some real life examples of things that I've built and things that other companies are currently using generative design for. What you see right now on the screen is a project that I did while I was at NASA Goddard, it was called BOOP. It was the balloon office opportunity payload, and essentially we, needed to take a camera and put it on a balloon up to about a 115,000 feet. And so, that camera, and and whole box are designed and the, arms were the arms that you see that have that cool like striation pattern were actually manufactured by Proto Labs and designed by us in about two-three hours. Generative design really speeds up the design process and how quickly you can bring things to market and how quickly you can iterate. Oftentimes, a simple change has a lot of trials and tribulations and back and forth, things like that. Here what we actually find is we can capture the inputs and then the outputs because they're computationally solved, we can recreate those as we go. Can I ask the group, what are these? If anybody has any idea, what these are as you're sitting there and thinking, these actually they're apart for a car. These, that you see right now are steering knuckles. What that means is you have the wheel of your car and you have your suspension that comes in, and so this piece is actually really complicated. This is where the bearings for your wheel sit, the brake attaches to this, so your brake caliper has to attach to this. It's really a lot going on there. But as you can see in each of these parts, there's a lot of planar geometry. A lot of flat edges, a lot of someone clearly made a shape to make a little bit of... Save a little bit of weight, put some fillets on it, kind of rounded it out. But these things, you know, there... There's a lot of assumptions when you go into these, designs, they're kind of baked in. The engineer will look at it and go, well, the last time we did it, maybe this piece was an inch thick and we got good performance, so let's just start with an inch and just go from there. That drives your design, your mask, things like that, and it becomes almost like a guess and check. Let's talk about the actual design process. What we tell ourselves is in the engineering process, we need to identify a problem, we're going to ideate on the solution, we're going to come up with the design, analyze it, manufacture it, test it, and then if there's any problems, we'll go back around. Well, what we tell ourselves is that this is a nice circle. It's a nice, neat circle and everything takes the same amount of time. What we actually find is that we don't spend a whole lot of time identifying the problem, really focusing on the requirements. We jump right to ideating on the solution. I could do this, I could do this. Oh, I did this other project that looked very similar to this. What do we want to do? We really want to get to the design. We love playing CAD, we love doing stuff like that. We pass it off to someone who does analysis and then, maybe we have a partner like Proto Labs where we can send it off to get manufactured, we get it back in three days and we go, sweet, let's test this. We don't focus a lot on what we spend a lot of time about what we think the solution should be and not really what's the requirement, what really has to be done. Let's go back to that steering knuckle. This is actually a generated design of a steering knuckle for an ATV. Well, that's small, that's child's play, things like that. Well, what if you're a Ferrari? Ferrari has started putting generative design. You notice this steering knuckle out to the right. Part of their suspension system now is a optimized component. What is generative design? Generative design is giving us these cool organic looks. But what it is, it's maximizing our stiffness while minimizing our weight through a computational means. We typically have inputs like here you can see the big round areas where bolts or fasteners are going to go through, and we can identify, hey, we don't want material here, but we need like a fastener here, we need an adhesion point here or something like that, and then we define our loads in all of our different load conditions. What's really great about that is that we can put in all of our anticipated load conditions. If you're an F1 car, you're a sports car, I know I'm going to turn at X number of Gs, I'm going to get this much load through the pins. You're actually already setting up the analysis as you begin to design these. It really starts to crush and shrink your design and your analysis time into giving you an optimized solution right off the bat. I'm going to hold up something that I designed and Proto Labs was nice enough to manufacture for us. We're going to actually explore the difference between what would be a traditionally designed component and what would be a generatively designed component. I think I'm going to point out here on the left, you'll notice a lot of straight edges, a lot of planar geometry, things like that. On the right and the generative part, you're going to see a lot of what we call organics. A lot of weird twisty geometry, especially if you look down where where there's a basket. This is a UAV holder for a UAV to sit and hang on. I happen to have one right here that we'll get into a little bit later. The idea is that this UAV can sit here in between these arms and then it will hang and its body will rest right here on this little basket. A traditionally designed part, we'll notice here as I'm showing as the camera focuses. So it's right here down along the bottom where we would expect a more traditional design to make it easier to, we think easier to machine. Down here, this is the generative design part. Part. It's raised up off of the bed. It's actually optimized for our load condition. What does traditional CAD look like to solve this problem? What I did was I three d scanned the drone, I brought it in and I made this. I took some slices, which you can see in the right image in purple, and I lofted between the two. I essentially told the CAD program, fill in the geometry between these two. Well, in in the time to make these, these traditionally designed arms, right, one thing that you you you might notice in the geometry immediately is that one side is lower than the other. So this surface and this surface are at different heights. So you end up, as you can see on the picture on the right, this really kind of weird, bracket and you have it in two different planes, you have to loft between the two of them. Well, you would think that a CAD program could just loft here, here, cool, good to go, walk away, you have your structure. But sometimes the geometry gets confused especially when you have all of the interpolated lines there in purple. You end up with these weird volumes and voids and everything. I actually fought with lofting this. This is the reason why I'm focusing on it so much. I fought with this for a number of hours. After I finally got this design, I was able to perform the FEA on it so I could, it would be about 14 grams per arm. It's first mode, so how it vibrates naturally was about six Hertz, so not really all that stiff. The idea of one of the applications is that this blue object here, the anodized aluminum could actually be mounted inside of like a police vehicle or a deployment vehicle, so that, users could could see their drones that they're ready to go and ready to launch. And so, you know, you you wanted to have a high first mode so that it's not bouncing the UAV around, it's not inducing a lot of loads onto the UAV. And it took me about four hours to get this one bracket done. I uploaded it to Proto Labs and got a cost of about $150 to make it with AVS via the SLA process. What does it look like to design this with generative design? In the picture on the right, what I'm going to point out is there's geometry in yellow, that's the scanned version of the drone, and then in blue, what I have is geometry that I want to keep or geometry that I need. That is the defining geometry of the structure. In this particular bracket, it's this U shape right here and then closer down here to the base, it's this little pad where the fastener goes through and the pad up here where a fastener goes through as well. You'll notice that there's a red cylinder coming out of those two bolt locations. What I'm specifying there is do not put geometry there. Whatever your solution whatever the solution the computer is going to work on, I'm removing that volume from the design space. That cylinder is actually the insertion volume of a fastener. I put a fastener in a washer over and in and drop it in and be able to fasten it to the structure? Special thanks to a company by the name of InfiniteForm, they run a Cloud based solver that is able to work on these geometries. Right here in the middle, what you can see is this is the solution that was ultimately printed. However, here you see a yellowish volume around it. That is our design space, that is me telling the computer there's nothing in this volume that, will stop you or inhibit you from building. Some things you might notice is there's some geometry right here, so I don't want it to build anything inside the cradle for the arm. You'll notice there's a flat back right here and so I don't want it to design or put any material that would extend into this aluminum plate. You'll also notice that there's a cylinder down here, so that's actually removed from the design space. Then over here, you'll notice that there's volume that's removed, but it also opens up to, I'll call it the outside. What can happen sometimes is it'll leave that volume, but it'll put something wrap it around that empty spot. This extended cylinder allows me to ensure that I can get a finger in, get an insertion path for the bolt, and then I essentially said, Hey, here's the loads. I need you to go from this huge volume, optimize it based on the loads that I have. I had the mass of the drone and I had, you know, in in a car at certain static g loads, I had different vibrational loads, and you can see I'm talking to an AI agent, over here on the right hand side. And so it's actually giving me feedback, on the optimized result that I'm receiving through the infinite form engine. One of these solves once you have put it in, takes anywhere from about ten-fifteen minutes. What's really fantastic about generative design and why it works really well with Proto Labs is you put in and you've built in DFM, which is called design for manufacturer. Design for manufacturer allows you to specify the process by which something is made. If you are three d printing it, if you're machining it, some engines even go as far as saying casting now as well. But there's a lot of limitations with the process that you choose. For instance, if you're machining something like this, you need to be able to get the cutter and the cutter head down in to make that pocket. Not only is there a tool associated with it, but there's the actual head of the CNC machine that goes in. And what's really fantastic, about working with Foto Labs is they actually have a CNC and a three d printing design guide. And so it tells you all of their their end mills, what they'll use for different materials, and you can build those into, these design for manufacturer, programs. And so you know that the design that you're going to get out can be made by the machine or by the process that you expect. And so with this, I was able to actually put in a, I made both a three d printed m a c n c machine variant. There's a lot of confusion, I've I've seen online and over the last couple of years it say, you know, generative, you know, only works with additive manufacturing and that's just not not true. When I worked at NASA, we had approximately 65 different, applications and every one of them were CNC machined. The reason for that is NASA has NASA standard sixty thirty, which prescribes additional testing for additively manufactured parts for aerospace applications. What we found was most of the time or a lot of times, the additional testing that you would have to do would almost overcome all of the time saved from using a technique like this. We really focused a lot, my experiences heavily based and influenced on CNC machining and not so much three d printing. In your application, be it aerospace, meaning planes, cars, something underwater, or like a static holder, you may not have that rigid of requirements. If you're just three d printing something to hold on to a wall, all of your loads are static, then you're good. You can three d print it, you can make out of plastic, you can do whatever you want. What does the performance give us? I told you about traditionally designing the part over here and how long it took. Now let's talk about the generative part. This part was about 1.7 grams less. That doesn't seem like a whole lot. Yes, this is a really small plastic part. But in the aerospace application, NASA estimates that a launched kilogram into space is about a million dollars and so we've had designs that have saved three, five, six kilograms. That's millions of dollars in launch costs and testing and things like that that we didn't have to do because we use this application. You'll notice that our first mode is 16.2 hertz, so that's 10 hertz higher. That's incredible. We've seen a lot of times where these type of designs will actually go from something like 10 to 15 hertz and bump it up to a 100, or a 120 hertz. I'll go into another application for a product that we're launching today on this this webinar, where we actually went way above that almost, four times the area that we needed using this type of technology. It took me only thirty minutes because I literally placed exactly where I needed them to be, had the little, trunnion to to hold the arm of the UAV and, you know, shifted out the... Out the door. It was $20 cheaper to make it, via proto labs. But let's really think about, you know, if you're somebody making, I don't know, a $100,000 a year, as a senior engineer or $150,000 that three and a half hours in your design time is a huge savings. Being able to pass off a design like this to an analysis person and say, hey, I've essentially done the analysis for you. Do you agree with this analysis? You're also saving their time as well. So not only do you see a lot of savings in manufacturing, and and in design, but just in in engineer time. So I hear a lot about, like, how are you going to test these? How are you going to do metrology on this? How do you ensure, you know, that the the part that you have is what you need. Right? Traditionally designed parts, you know, I'll go back to this blue anodized part. You know, you would do something like, you know, have a part probe that would go down and touch, random surfaces to ensure planarity of the machine part. You do all those steps. What we've actually found is that there's a lot of metrology tools now that have come out, especially with the advent of CNC machining and and how advanced that has gotten, metrology has also gotten that advanced. We can actually use like blue line laser scanners and throw a blanket, I don't know, 10 thousandths tolerance over the whole picture or the whole part, and we can get that information back immediately. If we need to do something like very, very, very advanced metrology for down to less than thousandths of an inch for whole placement or faster locations or things like that, we can still use our existing metrology. In terms of testing for performance, what we need to do is actually use our existing processes. One of the biggest challenges with using generative design and using artificial intelligence and automation is that everyone wants to jump right into it. Well, one of the things you really need to do first is take a step back and start discretizing your process. Go from, once this person designs it, it goes to this group for analysis, it goes to this group for testing, it's tested in this standard way, then take each one of those parts and swap it out and then validate that step in your process. Here, we're not actually changing anything about the design process. I'm sorry, the testing process, we are looking at the design process. Let's design it a new way, but put it through the exact same testing process that we had to ensure that, it still meets all the performance criteria that we're interested in and that we're looking for. With this, where do we go? Where does our development time actually go? What we're finding is that the CAD got really fast, the design did, the analysis got faster, but we're still spending the same amount of time in getting parts out the door. Our bottleneck isn't our design software in a traditional sense, it's how many learning cycles we can iterate through. I bring this up because what's fantastic about generative design is that it does all of these iterations and it can quote and machine and you can synthesize all of this and you can get 30, 35 different iterations in the course of a couple of minutes. And you can do it for all the different manufacturing processes. So instead of saying, you know, I'm assigning one person to design this part and they're gonna go off, it's gonna take them a week or two to design this part and they're gonna analyze it, they're gonna do this and that. With generative design, what you find is because you're able to automate kind of from the beginning, to the end, then send it off to a partner like Proto Labs that's able to machine it, three d print it or make it out of, you know, all the different materials they have in their capability set, then you can actually get those those parts super, super, super fast. I'm going to talk about one commercial product that I launched at the commercial UAV Expo. Right now I'm holding up our sensor backpack. This is actually a really cool tool. What it does is you can attach this via our attachment points here onto any drone. We have different adapters for a number of different commercially available drones and they all have a common interface that goes to the square. The square is able to do, air quality monitoring. The process that we used is called multi jet fusion. It is a great process that HP has been working on really perfecting. Proto Labs is a wonderful partner of theirs. What you can do with this is you can actually use these materials, gray or I believe they have a dyed black, and you can get them super, super thin. There's a new design paradigm using, MJF where you can actually get walls that you can almost see through. I want to say it's about 50 micron walls and you build kind of a structure and it's just the the skin. Because sometimes in an aerospace application, you know, what what do you really need? You just need the surface. You don't need that surface to be an eighth of an inch or a sixteenth of an inch, but we end up building that in because our three d printer or, machining process can't make anything reliably that's thinner. And so MJF has really unlocked a bunch of these these capabilities. With this one, we're able to put it onto existing drones, fly around, get air quality measurements, check for exhaust from buildings, some hazardous air condition, and its distance is really only limited by the flight endurance and signal endurance of the drone. But what we've done very recently is we've actually added a thermal imaging backpack. You'll notice that it uses the exact same attachment mechanism to the drone and then uses those same four holes, but I was able to generatively design a bracket that would hold a thermal sensor and its controller to the drone. Now, I want to just go ahead and describe the timeline here. I was very fortunate to go to Las Vegas with Proto Labs and went to the commercial UAV Expo. This was, I want to say it was like two months ago, And that company there, RPX Technologies, had a very small swab thermal sensor. And I was like, great, you know, how are you integrating these? How do you retrofit these onto existing platforms? And they said, well, we we really just make the sensor. We don't do the the retrofitting onto platforms and things like that. I said, I I need to do that. Right? Right now we have a lot of things coming out from the the US federal government about buying drones and and using drones and UAV technology that were all made domestically. And so, you know, how do we we retrofit the drones that we have existing to, have these capabilities and not go out and spend 15 or $20,000 on another drone that has a thermal camera, in the front of it. So, my goal was to make something that was, 200 grams or less so that, small scale UAVs, it could attach to small scale UAVs. I needed to endure, about three g's of load, figuring, handling, acceleration, deceleration and and, you know, as it moved. The, 150 to 200 hertz first mode comes from the propeller wash, so it's standard, kind of hover noise if you will, has a structural born vibration within that range and wanted to be able to have someone connect to it over WiFi and using aluminum 6,061. Just so everyone is aware, this is a real part. This is a DJI Air 3S. Here is our thermal camera that is attached to the side of the drone and it goes up and over and then there's actually a controller over here on this side that we're able to use to wirelessly get and process the images. I used the exact same process that I did for the earlier demonstration that you heard of, that you saw. I used our lightweighted sensor backpack interface and you'll notice that there's four blue squares there. Those are threaded holes for, hey, when this is made, I want you to attach to the outside of the structure, but I don't want you to attach to the front face because those are going to fit inside the bracket. You'll notice back here, in the back that there are two, kind of footholds, if you will, that's to hold the controller and a battery. So this is completely independent of the drone. And then, over here on the... What would be the left hand side of the drone, this blue piece right here is actually the interface plate to the back of the camera. You attach some thermal paste to that and that provides, heat removal. So we actually had to have, the the back of it made from a metal material that could wick heat away. You'll notice that there was, four of the red cylinders here, that's for the insertion of the fasteners, and then a larger one which is actually the the insert location for the USB C cable. I also had a volume down here, which is the field of view of the camera, so I wanted to make sure that nothing else was going to be in the field of view. I went, Alright, cool. Go for it, make me a complete design. As you can imagine, this had a lot more moving parts than the other example that I had provided. This all in all took about an hour and a half or so and I got a design back. You'll notice that I did find an element analysis on this bracket. Despite hanging a weight from each side, you notice that the right side was held very solid because of this thick vertical geometries that go by. The largest vibrating part is this little piece down here in the corner, but my first mode is almost five ninety Hertz. So what's fantastic is, you can take this, you can put it on and you, onto a drone, you can fly around and you won't get any sort of vibration, won't blur your image, anything like that. Everyone will notice in the upper left hand corner, this is actually the quoting tool from, proto labs. I love using this tool. I'm able to upload a lot of parts and just see rough order of magnitude. How much is this part going to cost? How soon can I get it? Things like that. This part, I want to say, I was able to get it in three days, machined from ProLabs, and the only issues, that I had with manufacturing was that this part was going to be thin and down here where the USB C connector was was going to be a little thin piece. But was able to actually remove and machine, the entire part and so, you know, completely completely clean. I can, you know, once again hold this up. You can see all the the bracketry. You can see, you know, all of the the arms kind of coming out here. And so we as as engineers oftentimes go, oh, well, I I couldn't machine that on if I was standing in a mill, you know, turning wheels and and things like that. But, you know, CNC has just come so far and if I needed to do a test fit before I went to a metal part, I could send it off and get it either three d printed, I could get it made out of MJF, I could get it from DMLS, Direct Metal Laser Sintering, and sent back just for testing and for fitment. Then this video is actually a demo. I took this outside on the drone. It was actually a very windy day. What you see a lot of the bouncing of is actually the drone stabilizing in the wind. There's a few shots here in this video. You can actually see the trees moving a little bit from the wind. But it's a fall day here. I live in Maryland and so you can imagine a person walking with kind of muted colors underneath some trees, you wouldn't be able to see them. And so this is, you know, really powerful now because if you're doing, for instance, a search and rescue mission or if you're doing something with, you know, trying to find maybe animals in the forest or something like that, you can actually kind of see through trees. It's it's almost like cheating a little bit. You can see this this person having to be walking down the, down down the sidewalk. You can see also there's a car up here in the top, we'll see another angle, this setup here in a second. This is another person walking behind this car. You can see the car is bright, so it is warmer. Things that are colored black or colored dark blue are cooler. This makes sense to us. The sun came up, it hit the car, it warmed up the car, and then under the car didn't get any sunlight. And so it's actually a lot colder under the car. And then this person's walking by and they are hotter than the ground around them. They're hotter than, kind of the... Excuse me, the road behind them or or this tree. Now, the same situation applies. You can see them walking around leaves and branches and things like that that you would think maybe you'd have a hard time visually seeing. Then I had another target of opportunity happened to just fly by or drive by at the end of the street, this is just a car driving down the road. This is why I say it's cheating because you can see here the engine is running. The engine is a lot hotter than anything else going on around the car. You can actually see things like, if I was at a low enough angle, you can see things like if the car had been sitting for a long time, you get on the tires, you get strips of color, and so you can tell like it is either cold or or hot. You can tell if the engine's running, if the front is really warm, things like that. And, you know, this is... For for me, you know, I... I'm now making this product. I'm I'm bringing it to market. You can now, go to pathfinderlabs.ai and fill out an information form. But I'm able to do this and able to bring it to market because Fertilabs has such like awesome capabilities in rapidly manufacturing and rapidly prototyping things. You can also use it for an on demand, machining service or, you know, or injection molding. Hey, I need, you know, 10 of these. Now you're not holding on to stock, so you're not, kind of, you you don't have all of that upfront cost. You can kind of make them as as you need to go. And so this car, you know, drives away, and then you can just kinda see it heading down, down the street. So very cool technology, to be able to, play with. Thank you very much to RPX Technologies for the camera and, Protolabs for helping me in in making the part, attaching it to the drone. So going back to generative design and digital manufacturing. So What's next? Where does this go? Well, what we really see is conversational design. Much like AI has really helped people code or design, you know, flyers or write emails or things like that. We've gotten kind of good at prompting. Right? I mean, email in a response to the following message, make it sound like this, etcetera. What we can and should expect is more conversational design within a CAD program. Hey, I need to interface these two parts. Here's my load conditions, here's my load characteristics, and then expect the CAD program to come up with the, I'll call it the foot pads, those rectangular things that you see right there, and then do some level of generative design or some automated design system in order to make those parts for us without needing specialized CAD software, without needing someone sitting and clicking a 100 times to make a part. We're also looking at like standards aware workflow automation. Instead of, someone coming in and going, oh, do I want this type of fastener, this type of fastener, do I want to make a tapped hole or clearance hole? Do I need, to put in, an insert for for, threading or a Tapcon or something like that? We we expect that pretty soon AI is going to start knowing, hey, if I'm attaching these two things in an aerospace application, maybe they're going to be under our rocket launch condition, they're going to be using like stainless steel hardware of a certain grade. This one maybe is an electrical box in a house. Well, when that design comes up, we expect electrical code needs insulated materials, so they're not going to pick aluminum for an electrical box or something like that. We also expect to see a lot of things about manufacturing and quoting access inside design tools. So, you know, Fusion, SolidWorks, Onshape, you know, can you have a a proto labs connector that allows you to, you know, click a button and automatically get a quote inside your CAD tool? Can you get geometry back from, you know, a machine shop that says, hey, I I can't make this one, cut or this one kind of, thing that you've designed, or it will take me additional time, or can you have the digital machine shop where you know this shop has 10 machines, they can get you in, right now and get a part where you go, ProLabs is going to be better for this one because I need to make a bunch of them on demand versus a local shop or something like that. So three takeaways as I wrap up and we open up for questions. So speed lives in in every step of our design loop and, you know, the the the direction is consistent. The time between decision making, and making the part keeps shrinking. Right? Really focusing on our requirements upfront and making sure we have those right will really help us and speed up the decision process and will make it easier to make the correct part and get that tested in a timely manner. The best time, number two, for design for manufacturer is before the geometry exists. If you know that there's limitations like we can't use DMLS, we can't use a three d printed part, it has to be CNC machine or CNC machining is going to be too complicated. We just need it three d printed, make it an SLA. The best time to do that is before geometry exists and you are defining requirements, but honestly the second best time is now. Start working on and designing for the method by which your part will be manufactured and you will be shocked to see how much quickly, how much faster your parts are able to be made. Then again, trust is earned by test. Don't just expect to walk in and say, hey, we have this new AI tool and it's going to make things and it's just going to be awesome and make all sorts of parts. Everyone will be immediately skeptical as they should be because even though AI gives you an inference, the engineer is still responsible ultimately for the outcome. We need to really get fast, accurate, and affordable hardware that we can test and validate and then really prove out like this design methodology, generative design, or having another company make our part or making it out of a different material or a different process, you know, I mentioned MJF earlier, you know, passes the muster and, you know, that's definitely something we can do. You can learn a lot by testing and not experiencing analysis paralysis. Some questions for you to think of as we end, where in your development process, does your loop grow or maybe even open and not close? Where is your critical point that you really need to fix and what would be possible if you could not only close it, but then tighten it and make it faster. Right? It's not a one and done. We are definitely at the... In the age where replacing, you know, just buying software and saying, oh, I'm I'm switching to a different, you know, CAD platform and, you know, we're we're gonna use Windchill or something that's gonna solve all of my problems. We are now at the point where, you know, we need to find gains at every single step. And for design, that might be using generative design. For manufacturing, it might be turning to a company like ProLabs to make something really fast, and it may come for using like a testing house or doing something in house and becoming vertically integrated because you need to test so many of these things so quickly. So with that, I'm gonna open it up to questions. Thank you very much for your time and, look forward to hearing from everyone. I don't see any questions from the audience. One just popped up. Thank you very much. Pricing on these parts. We actually don't see any real increase in pricing. Now, as I mentioned in this one demonstration, the generative part saved, 20 to $30, on that web plastic part. If you are, making something that is like a a sheet good that's punched out and bit, Right? That's probably about as fast as you're gonna go. You can optimize something like that. Generative doesn't work very well for like large sheets or for cylinders, but if you're looking at a bracket, right, traditionally what you would find is, if you needed to have a bracket that was gonna say hold something, you're gonna put a base on it, put a gusset here, gusset there, you're you're looking at, okay, I'm gonna take this sheet of aluminum or the sheet of steel. I'm gonna drill some holes. I'm gonna come back. I'm gonna weld it. I'm gonna do whatever. You know, making one of those parts, the machining of it might only take, I don't know, maybe a water jet or a plasma cutter or something like that, and it only takes you an hour or an hour and a half to machine it. But then the assembly process takes three or four hours. I mean, that's, you know, that... That's kind of your baseline. Well, if you... Again, going back to discretizing your process, go through and say, alright. Now if I can send this to a different shop, for instance, and I can get 20 of them in three days, so I get 20 of them in four days and they show up and then I'm done, does that work for your process? Traditionally, haven't seen them much more expensive than traditionally designed brackets. That said, a lot of my experience with this has been with, I'll say, bespoke aerospace components, where adding a few extra $100 for a lot of performance increase is not an issue. If you are using this for, I need to make 10,000 of them, at that point you're going to have to do the analysis based back and forth of if a little additional machining time is going to be worth what might be a a more, faster machining process. The company I mentioned earlier, in my in my analysis, infinite form, they actually have a prismatic that will come up that doesn't have all of the organics and at least kind of, prism shape, more traditionally looking, shape. It could be machine just really fast and not need all the the contours, and that might be an avenue that you go as well. I see a question says, do you size the red areas or they do not build geometry zones based on predetermined hardware sizing ahead of time? Also, what was the name of the software that you used? Yes, a lot of times you're presizing fasteners based on load and based on edge sheer strength of a fastener. For that, if you can change it back and forth and you have a design envelope, what's really fantastic with a lot of these tools is you can just say, you know, I wanna try these three shapes. You can either have an agent do it or, you know, you start out an analysis and you say, I'm gonna do a quarter 20, analyze, go. And then from there, you get another, you know, while that one is running, you go and you change the size to different one and say go. It will all come back and then you can look at, you know, overall displacement versus weight or or mass. You can see, like, maximum mice stress versus mass. You can actually compare all of them, in a in a big, dot dot plot essentially. Name of the software used, used primarily three, softwares. The first one I mentioned, InfiniteForm for their generative design engine. Another one that I've used is Autodesk Fusion. They have a really good one. Also, there's a workflow program that I use and I'm actually a developer for called Scenera. You can think of Scenera like instead of making the part in CAD, you actually build the timeline and then it generates the parts for you. Really cool tool and I'm actually giving a webinar on Tuesday about that and optimizing for design using, using that one. So, question from Noel. Can you explain if the generative design software infinite form requires a paid subscription? Additionally, do you have to provide base models first and then build upon this for optimizing, for the design or you prompt the software and create a design based on your conditions? So yes, InfiniteForm does require a paid subscription. Most of these, do. Fusion requires the simulation package for you to have access to their generative tools or you can pay as you go per generative solve. Scenera has a token based system where you buy 20,000 tokens or something like that and the base software is 5,000 tokens and the generative design add in is 3,000 tokens or 5,000 or something like that. So, yes, so all of them have an associated oftentimes cloud compute costs. Do you have to provide the base models first and build upon them for optimizing the design? So this is the difference between topology optimization and generative design. So So topology optimization is the traditional design methodology where you make a bracket or you make some structure, you put in loads, and then it will subtract material from your design. The challenge with that is that it's already set down a predetermined path and you can get stuck in local minima and local maxima with those type of designs. What generative design allows you to do is generative actually, wraps your whole design space. So typically, it takes, you know, all the volume around your fasteners and makes kind of a big blob, oftentimes like convex hull or a cube or some sort of shape. And then from there, it does, the most common one is called, man, sorry, it just... It rolls off the tongue. It's level set minimum compliance to policy optimization. What happens is the software defines the surface of that convex hull, does analysis, divides it into Ith layers, goes inside the part, outside of the part, determines stress strain at each of those voxelized points. Then if there is a high stress point, it will add material. If it's a low stress point, it removes it and then it iterates again. And so that's actually how you get these kind of, weird geometric shapes is it makes the volume and then as it goes through, you'll get, you know, those kind of spindly looking organic things. One thing that you can tell, oh, this is a generatively designed part is you get a lot of bifurcations with it. What I mean by that is you might have a leg in topology optimization. Once that part is removed or once that node is removed, it can't go back and so you'll get just one stress path. With generative design or more specifically level set, it can actually go here and go, oh, I don't have a lot of stress and remove it, and then we'll find that there's stress on on these two sides, so we'll add material. And then you start to get two legs that'll come out of it. So that's what I mean when I say, it will bifurcate supports and you'll start to get, you know, really the the organic, look out of it. So that's really the original topology optimization and and generative, topology optimization being what you were referencing. And then, Jeremy Parsons asked, can you delve a little deeper into how you are generating the generative design model? Your view of the process all is excellent, but I'm more curious how to get from your parametric bracket gets to the optimized organic shape by a generative. I believe I just, answered that. Using level set is one of the methods that you can do. You can also use like, simp and things like that, which is just a volume optimization reduction in order to to get the optimized design out. So, a lot of these engines are being designed by, or there are a lot of engines being designed by companies, to solve exactly these problems. They all have a little different flavor of how they do it, how they do reconstruction. A lot of them will, also give you for instance fusion will give you the organic shape and actually allow you to modify it. And so sometimes, with generative, you can get, maybe like a little nub, somewhere that was removed but the stress was still found. It reached the optimized solution but didn't go through the cleanup to remove that little nub that's sticking out. You can actually see that in the generative design bracket. I'm actually gonna flip back through some slides here. Right here, you'll notice that there's a little disconnect, kind of between those two right here. And, you know, so what what happens is there's a number of iterations that it goes through and then at some point it goes, hey, I've I've found the optimized solution. I don't need to keep computing on this and that was kind of an artifact. Well, what you can actually do in, as I mentioned in Fusion specifically, is you can go inside that tool and actually modify the organic and kind of push that whole structure, back in to give it a smoother shape. Fusion also allows you to do things like, cylindrify the supports, so I actually did that in some of these. Instead of maybe being something that has a lot of undulations on it and things like that, you can select it and say, I'm going to cylindrify this and it makes it an easier to follow geometry. You can do those post processing things. I know that infinite form also allows you to go in and do some post processing and because with scenario you've gone in and made the process, you can actually put in your own post processing after that as well. Just kind of simplify those things. I see a few more questions that popped in down below. Does ProLabs actually machine the product or do they provide only DFM analysis for the component? And can we get the DFM analysis of the component as soon as we upload the design into ProLabs? So, yes. ProLabs actually machines the product. So one of the differences, primary differences between them and one of theirs, another service you've probably heard of is Xometry. Xometry sends out to a network of machine shops to have their parts made. Proto Labs has full control of their process. If you order a part through Proto Labs, they make it in their machine shop facility. I've been to their one in Brooklyn Park in Minnesota. It's a absolutely incredible facility. There will be some videos coming out shortly. I know they have some gen... Generally of the facility, but, you know, it's just rows of like Haas v f two CNC machines. It's just... It's really, really, really cool to see and that's how they can scale and and get on design. They also have Proto Lab Network. Proto Lab Network, does, send it out to, other shops, but when those parts come back, Proto Labs actually does all the Q and A on it to ensure that it is the high quality part that you want and that you expect. The TFM analysis of the component, as soon as you upload, if it's a clean part, meaning that, if Proto Labs can just machine it right off the bat and there's no overhangs or no like real challenges, yes, you can get it immediately, sometimes it takes up to an hour. If it's really complex geometry, I have some on the shelf behind me that I've made that are this big and they have really weird overhangs, things like that. That will take human analysis and sometimes they can go, Oh, I see where the automated software and you can get that feedback typically in less than a day. A lot of times I will actually have two or three optimized designs and because for me it's effectively nothing to say, Oh, let's see if I three d printed it, if I machine it, if I did whatever. What would all these cost? How long would it take? I send them to Proto Labs and then I get a a quote back for all the different parts. I know all of their subsequent performances and I get their costs and their delivery dates and then I can make, kind of a decision based on that. Next example is huge opportunities ahead. Thank you. Yep. Thank you, Artur. Appreciate it very much for your comment. And, last call for questions, per Kaylee. If you have anything else, they'll wait for about another minute and then we will go ahead and wrap it up. Alright. Hearing none, thank you so much. It was a pleasure and opportunity to present to you. You noticed there was a QR code previously on my site that takes us to LinkedIn. We're also at protolabs.ai And, LinkedIn, you can also reach, or pathfinderlabs.ai. You can also reach protolabs@protolabs.com, and, looking forward to see what you come up with. Thank you very much. Bye.