Automating and Simplifying Digital Product Creation | Clothing Tech | PI Apparel

So let me introduce myself. I have spent 40 years in other industries, only recently entered and started working in the fashion industry. My background is mostly looking, solving efficiency and manufacturing problems in automotive and aerospace and electronics. About 5 years ago, Hexagon, which is our partner company, we got interested in what could we do with the technology that we have that we use for a number of other industries. How could we use that same technology and solve problems in the fashion industry? We created a startup called Clothing Tech, and we're based in Austin, Texas, in the southern part of the United States. I want to talk today a little bit about automation, and we'll get to digital product creation. But first, I just want to talk generally about automation and the current state of automation in the fashion industry, again, from an outsider's perspective. I have learned a lot in the last 4 years about garment construction and the processes, but I can't say that I'm an expert or I'm nearly as knowledgeable about as many of you. There's a great deal of automation actually in the industry, but it tends to be in the early stages of production. When you talk about fabrics, fibers, converting raw materials into fibers, spinning fibers into threads, weaving threads into fabric, etc. Lots of automation, lots of big machines that do it all. There's not so much manual labor, some, it depends on the fabric and the and what it's coming from, but mostly it's automatic knitting, weaving, thread production, trims, buttons, all kinds of things, elastics, zippers, all produced in large quantities with a lot of automation, very little human interaction and touch. So far, that's a good beginning for the industry, but then things slow down. A bit once we get past the basic production of the bill of materials that would go into a garment or a shoe. When it comes to apparel, so the design, the technical design, the pattern-making, manufacturing remain largely manual. There are some automated steps in there, but overall, I think it takes a lot of people with experience and expertise to convert the intent of the product into the information that is needed to actually produce it and create that product and bring it into market. This is where things get a little bit challenging in terms of the automation. Partly it's because of the structure of the data. If you look at the data that we're using here, Designs often begin as a sketch. Maybe they're in a computer, in Illustrator or something like this, but there's still, it's a manual process. It also requires a person or beginning, there's more options now with AI, but still it's beginning, it's in its infancy. So the design process is still largely sort of human experience, human expertise. And the information out of that process is not digital information. The designer is not giving information about the product that a robot could understand. The designer is giving visual cues about a product that a person typically would need to be understood to be able to say, OK, how do I create that product, which is the next step, which is typically technical design. So a tech pack, generally manually produced by an expert, interpreted at a factory also by an expert, to figure out how to produce that garment. Of course, the sewing and the processing, much of that is also manual, which is why there's always a constant search for lower-cost labor. What we've noticed coming again from an outsider's perspective, we noticed coming into the industry is that there are two areas where we see investment and focus. One is in modernizing production, a little bit more automation, more cycle machines like buttonholer, button setters, collar, cuff creators, pocket placement, and sewing, and so forth. You see a little bit more automation as over time in the factories and production. And you see a lot of investment and effort to create digital products. So digital product creation is a relatively new thing within the last few years. And the main objective right now is to reduce the amount of physical samples by replacing them with virtual samples. Instead of giving the pattern to a tailor or a seamstress to create a sample, we give it to a 3D designer, a 3D expert who uses a tool, one of several tools that are available on the market to create a digital sample. The problem is that that's a manual process. We've just introduced one more manual step into the design process. We have simplified, replaced another manual process, which is sample making, which is a good thing. But it's not still— there's really still very little automation in this process. So what are the obstacles for automation? If we talk about automation in general, automation requires digital information about the product in order for a robot to understand it. Automation requires information that is— you can't give a tech pack to a robot. They don't understand it. They can't interpret that. Even with the best AI systems, you're going to get at least 20% of that wrong. Who wants a garment that has 20% error rate in it? You want the garment that's the way it's supposed to be designed. So you can't have 20% error. You have to have 100% understanding. You can't get that with the data that you have today. So product data is currently scattered between all of these different formats, sketches, tech packs, DXF files, PLM systems have bill of materials and sometimes points of measure and all these different things are spread out and there's a lot of domain knowledge. For us, that was a little challenging because we're mathematicians, computer programmers. We have a different skill set and so trying to understand all the domain knowledge was a big challenge for our team. As we started 4 years ago. Most of the product data of all of those In another industry, you would call that a subassembly, and you would make those in a different factory potentially and ship them together and, and assemble them in a final assembly factory where you— like the engine The car is made of many components preassembled, brought into the assembly factory in the assembly line, and inserted into the car as it's rolling down the assembly line. And this is very normal for most industries. So digital information, digital definition is really a key part of automation. And I think that you can't understand, you can't really ever automate, the industry can never automate unless they solve the problem of data, how the data is structured. Clean data, in my experience, is a result of standards. So I want to just spend a few minutes talking about standards because to me this is a super important topic. So automating garment design and engineering and production. I know you don't call it engineering. It's sort of technical design. But it's really the engineering phase of the product development process, which works across all products. We would like to propose 3 standards and we have sample standards here, but the way standards work, I don't know if you've ever been involved in the process of creating a standard. It's typically the users of the information, which is the brands generally, that need to decide what the requirements of the standard are. They have to come together and say, okay, this is what we want this standard to accomplish. Then somebody, usually a research institute, a school, or a group of interested parties get together and work out the details of how those requirements are to be met. Then it's iterated and eventually there's a standards governing body like potentially ASTM or there's a dozen different standards bodies in the world that create standards and manage. One important part about standards is not just the definition of the standard but the compliance to the standard. You have to have both. Otherwise, if you don't have compliance to the standard, people say, oh yeah, I support the AAMA, DXF files, but do they really? Maybe, maybe not. You need compliance as well. So there's 3 areas that seem to be missing right now. Digital garments. If you have a garment that you create in CLO, can you export it and bring it into Browzwear? No. Can you take the Browzwear file and bring it into Style3D? Nope. They're all proprietary formats. All of that information, somebody has said how to create that garment. How to fold it, stitch it, layer it, place it around the avatar. If you export, what do you get? You give a 3D visualization of that garment, but you don't get any of those details. Nothing is— all of the details are lost. This is what we talk about, digital garments. All of the information that's needed to create a digital garment, which includes the pattern. The pattern is part of a digital garment. You can't have a digital garment or a physical garment I'm talking about cut and sew, not knit. You can't have a digital or physical garment without a pattern. You can't have— without digital fabrics, you need to have digital trims as well. And so those are the other two parts of the standard is that right now we have proprietary formats kind of scattered about. For each vendor has their own format for how they want to define fabrics. You can move between one to the other, but it's a little bit awkward. It's not easy. Why not have a standard? It should be one way of defining digital fabrics. It's all about the physics of the fabric. It doesn't matter about simulation. It's just about how much does it stretch under what amount of force? How much does it bend under what amount of force? How much friction does it have when it slides across skin? How much weight? These are physical measurements that are not ambiguous. They're just physical measurements. Why can't we create a standard format that is applied to all vendors? And you just, as the brands— I have an observation to make. The brands in this industry, if you go to automotive industry, the vendors of technology are multi-billion-dollar companies, sometimes with 50,000 employees. This is the vendor of technology, not the user of the technology. The brands in this industry, in this case, are multibillion-dollar companies, the vendors of technology are 100 people, 200 people, very tiny little companies. And as a result, I think you forget that you're a lot bigger. As brands, you're a lot bigger. You just have to say, support the standard or I won't buy your product. And guess what? People support the standard. But we have to organize and decide what that means. What is the standard? What are the requirements of the standard? So digital fabrics, wouldn't it be simpler if every producer of fabrics produced also their digital version of that fabric and made it available? So you don't have to have your own group of people or outsource or pay somebody to scan your fabrics. Why should you have to do that? It should be done by whoever you're buying fabric from. They should provide you the digital version of that fabric. Same with trims. If you're buying YKK, I know there's YKK people, don't be offended. I like you a lot. But if you're buying YKK zippers, you should get a YKK digital version of the zipper. You shouldn't have to figure out how to hire a 3D artist to create your own version of the zipper. You want the digital trim makers, elastics, all of the different digital trims. There should be a standard. There isn't a standard today. We use OBJ, FBX, GLB, all kinds of things. That works great for shape. But a lot of trims have flexibility, have physical properties that have to be simulated. A lot of trims also have to be assembled, like hook and eye. How do they get hooked together? This information, where is the point where they get attached to the garment? Where is the attach point? This should be standard. Right now, an OBJ file could be anywhere. How do you know? A person has to look at that and say, ah, that's the attach point, and has to deal with that. You can't automate that. A standard would allow us to say, OK, I take that hook and eye, I put it on the garment, automatically sew it in the right way. I don't have to have the person to do that for me. So standards, again, are important. Let's go through some of the potential requirements. And I'm going to go through these quickly. These will be available to you later, and you can comment on them. But I'm not going to go through every little comment. But the basic idea, there's— why standards? Because they enable automation, which we've been talking about. You can't have automation, whether it's design automation and creating the designs faster and easier, or it's in a physical factory where you're creating the products, you need standards to enable automation. Interoperability, we talked about this, going from one vendor to the next, you have to cobble together a set of tools that work together and you have to work with the vendors to make them fit. And there should be standards interfaces between them, which is easy, much easier for everybody. And the single source of truth. Right now you have tech packs, you have digital— I guarantee if you go to any of your digital products, your digital version of the product, your pattern, the production pattern that's being produced at the factory, and your tech pack that is used to send to the factory are not going to be the same. There's going to be differences between the tech pack. Because if you change the tech pack and update it, it doesn't update the 3D garment. If you change the 3D garment, it doesn't update the tech pack. If you change the pattern, it doesn't— maybe if you're doing inside of the 3D software, but that's typically not your production pattern. That's typically the pattern used for visualization. It's a little bit different. So again, single source of truth is important. So digital garments, digital fabrics, again, interoperability, standard testing equipment. Why have to have a special kit? It doesn't make any sense. There's lots of material tester vendors around the world, and they have very precise, very accurate products. You can just buy their products, stretch your fabric, fold it. You don't even— shouldn't even have to do this. But the providers of the— the makers of fabrics, the looms and whatnot, they can use standard testing equipment. They don't have to have a CLO version and a Browser version and a whatever version. They can have just, here's the physics of the fabric, period. This is what the standard entails. Then also, if we go on, we have— sometimes this doesn't quite— digital trims. Digital trims, same thing. You have testing equipment with physics. There's a lot of physics in a lot of trims. Think of a shoulder pad. Shoulder pads actually are pretty complex in terms of the physics of how they shape and mold around the shoulder. You need standard testing equipment. You can't put one of those into a CLO test kit and say, OK, how do I— how does this shoulder pad work? It doesn't work. You have to use standard testing equipment. So standards become very helpful here. The textures, the trims, the physics, the attach points, the shapes, all of these— yeah, why not use OBJ? Still fine. But there's more information than just the shape that is needed for effectively creating, especially when we're talking about automation. Automation. Now, why are we so passionate about this topic? Because we— and really, it's automation and clean data. The first is because we want to make things more efficient, and that allows you to do more and also to do it in a much shorter time span. And I'll show you what that means in a little bit here when it comes to digital product creation and product development. So we've been basically spending the last 4 years trying to clean the data. We've worked with some key early brands that have worked with us from the very beginning, and they send us tech packs and patterns. And guess what? All the patterns are a little bit different. They're all standard. I think these were mostly ASTM standard format with graded nests in them, but they're all a little bit different. They're not quite the same. So you have to clean all of that data to make it so you can consistently import that information and extract the actual information that you want, which is the pattern shapes. The second thing is then about the tech pack expert systems. Again, we've trained expert systems to try to read the language of tech packs to be able to extract the information from them so we can automate the DPC process. And that is really what we started our first goal was to just to try to take the information that you have, which is patterns and tech packs generally, and convert them into 3D assets automatically. That was the first goal that we started with. So we've moved on a little bit towards that, is that we said, well, we want to be able to not just convert because— and the process of design right now typically starts with sketch. Then it goes to a technical designer who creates a tech pack. Then a patternmaker generally is involved. And then 3D sample maker. All of these things take time and cost and slow down the process. We said, why can't we design in 3D, directly in 3D? Well, what are the obstacles for that and so forth? We started looking through that. Let me just to show you our approach to automating garment design. So this is DPC, fully automatic DPC. So we'll just show you a few garments we've loaded out of many that we can convert. Just load the tech— let's see. I think that— there it goes. There. So we're just going to bring out an avatar. Again, these are standard OBJ type FBX avatars because there is no standard for avatars. But we load a DXF file. This is— we do have to tell our system what type of garment it is. The Expert System works better if we know it's a pair of trousers or if we know it's a shirt, blouse, or top. In this case, it's a pair of running shorts, but we've added automatically the waistband with the gathered elastic in there and sewn the shorts around it. You'll see a few others that we've done. How is this done? Well, again, it's a lot of effort of cleaning data to try to get from the information you have, which is patterns and tech packs, to get the information needed to generate the digital product. Is it 100% accurate? No, but it's about 85% accurate. There is sometimes a manual step where you have to tell it, oh no, this is a front top and not a back. Maybe it recognizes it incorrectly, and so we have to sometimes help it out a little bit with human intelligence. But again, it's still just a matter of telling us what it is and then it automatically does the rest of the job from there. You can see different types of garments. This is a singlet, rowing uniform for a company that makes uniforms for crew teams, and so forth. This process requires— so one of the advantages of automation, and you think, OK, well, this is much faster than what you're doing today because today you typically have to place pieces around. And yes, some of the vendors talk about automation, but this is a level beyond what you'll see at any of the current 3D CAD solutions. You have to place the pieces, you have to sew them and say attach them, you have to create the finishes and all of that. For us, all of this is done automatically. I'm going to go ahead and continue on while we're— while it's— we'll go to the next one. I think I can skip out of this one. Yeah. So then let's say now what? We have a design. We want to create a design from an existing design. So our contention is that most of the time when you're creating a new garment, you're creating a change. You're changing a style that already exists. It's not really a brand new style. It's a derivative style. It's an update or refresh. All industries do this. The iPhone you buy today is not that different from the one you bought 2 years ago or 4 years ago. But it has been updated and it has been improved and so forth. This is what the process is. There's usually 3 ways that a garment changes when it gets updated. Fabrics, prints, and colors get changed typically in a new updated garment. In addition to the— I'm going to go ahead and start this. Sorry, I forgot to have to— there it goes. In addition, shape changes. Maybe it's a different silhouette, maybe it's a longer length garment now, shorter, or maybe the finishes change is the third area. A lot of times finishes will be different from one to the next. So in this case, we're taking a garment that already existed from a few years ago, and it's just a pair of 5-pocket jeans that have been cut up with shorts. So it's a typical thing, and we say, well, I want to make something new out of this. So we've provided tools in 3D to change the shape, So I want to change the circumference, the silhouette of a garment, I can change that using Bézier curves. If I want to change the length, the proportions of the length, I can change that with, again, simple tools, drag and drop. If I want to change the finishes, I just, again, drag and drop finishes onto the garment. Like here we're adding side slits on the legs. Because of the automation, we solve another problem, which is the complexity of the process. Most— many designers are not comfortable using today's 3D software because they perceive it as too complicated. It's too dealing with patterns and pattern making. So we took the pattern making and automated it in the background. So the pattern is getting adjusted. When we create this garment, we have to change the garment to the pattern shape to create the slit. We have to change the pattern shape to create the different lengths. If they change, if she wanted to make it baggier, the designer's doing that or change the pocket bag shape. All of those things are possible to do with simple Bézier curves, similar to working in Illustrator. In this case, you know, she's adding some embroidery onto it, some illustration onto it, and, you know, trying to, again, trying to create a new style. Prints, fabric colors, and so forth are being updated and adjusted. Not much skill in terms of design skill, yes, but in terms of technical skill is necessary any longer with this approach because again, with automation, it becomes super, super simple. In the end, you end up with a process that allows you to use a tool in a totally different way. There's a few more examples here and I'm just going to go through one more before we talk through this one. This is taking a shirt and adding a collar. Normally if you were in a 3D software today, this becomes quite a technical challenging problem. To do that, you have to create the collar, bring it in from another garment, you have to take the collar and fold it, add the second ply, put the collar stand on. All of those things have to be done manually. But again, with automation, it's drag and drop. All of the instructions for creating the placket, cutting that placket down the front of the shirt, extending it so that we have enough material to fold to create the placket, or adding the placket pieces on top of the edges, either way, all has to be done by It can be done by automation. So we've been able to automate that process. And the result is then I can not only do that, but I can change the fabrics as well and update the style and have different versions of it, different colorways of the same garment. One other thing that's not talked about here, because what we do for pattern making is essentially the concept of parametric CAD. And if you're not familiar with parametric CAD, don't worry about it. But it's heavily used in most industries. Normally when you change the design of something, you do it by parameters. You don't actually go in and change the individual points of a pattern. You change the parameters of it. So the waist, I want the waist to be 1 centimeter more, or I want the waist to be 1 centimeter less, or I want the length to be adjusted in a certain way. These are the parameters of the garment. The points of measure, you can think of the points of measure and editing those, but not just editing them, but actually having those edits it's changed the pattern to accomplish that. So the pattern making is done automatically. It's called parametric CAD. So we've taken that same concept and applied it here to garments. And as a result, you can create these different styles and lay them out and visualize them separately. And she'll bring out some more avatars here in a minute. But in any event, the whole idea between— The ability to adjust and update the garment. In this case, this took, I don't know, 3 minutes, 4 minutes to create 3 new styles, 3 colorways of a new style. If I wanted to make another style, maybe more of a polo shirt or something with maybe a a V-neck or a scoop neck crew neck instead, and I wanted to change the adjusted, I wanted to add longer sleeves. Again, you can create so many styles because the automation allows you to do that. The automation which comes from nice clean data. I'm just going to continue on. Let's talk about the design, today's process. And I know there's probably some questions about how we do this and why we do this, but— and we'll come back. I'll try to leave enough time for questions at the end. The 2D sketches is usually where a new garment starts out today. And usually the technical designer, the pattern maker, the digital product, which is the new part. The digital product creation used to be a physical sample maker. Now it's a digital product creation, which adds more value and speeds things up, and you can use the digital product for your connection to your customers. You can visualize it on the e-commerce and social media and so forth. So there's a lot of advantages to have a digital product. But the same process is really what's— it hasn't changed the real— the process is basically the same as it has always been. It's just that we've replaced one technology with a better one. But just one part of this step is is really different. There's still a lot of delays because if a designer wants to make a change, the designer most likely is not going to be able to change it directly in the software. So they're going to tell the technical designer. The technical designer is going to tell the pattern maker. The pattern maker is going to create a new modified pattern. And the new modified pattern is then going to be visualized in 3D. And all of those steps might take a week or two. But with automation and standardized data, this can be done in seconds. The change the designer wants to see, they can see in seconds. It's really that simple. So what we think is a more efficient process, using, again, with current technologies, is to take a tool that designers can actually use and are comfortable using. They don't have to deal with patterns. They don't have to interact with, oh, I have to be a pattern maker now. They don't have to do any of that. They just take the garment that they have, which we can convert automatically into a 3D garment for them, visualize and say, okay, what do I want to change to make this new style? And make those changes. They drag and drop the slits on there. She dragged and dropped the, the embroidery on the edge of the pocket. She changed the color of the fabric. She shortened it. You could do many other changes to it as well. There's unlimited number of things you can change, again, through this very simple approach. What do you get from that though is you get the pattern. The pattern is made for you. We create the pattern as part of the process. The pattern is an output. Not an input into the 3D process. The pattern is the output. In addition, you get the tech pack. The garment is essentially the digital tech pack. Everything about the garment is, in fact, the definition of the product. And we can convert that automatically into a physical file— not a physical file, a digital file, but a human-readable file that can be sent to a factory so employees and people working in the factory know how to produce the garment and have instructions to do so. From that, you get the potentially product validation. But the difference here is that in one day, a designer can create 10 different new styles. It's really that easy. One day, they can create 10 new styles. Each of those styles comes with its pattern. With its tech pack and with all of the information necessary to automate actual manufacturing. Because this is completely digital, the definition of the product is completely digital. It's not— the tech pack is just a human-readable version of that information. It's not the information. The information is digital. We convert it into human-readable. All right, let's just stop for questions now. I don't know if we can go on to the Slido here. Industry standards, this is a great point. So what we're doing now is trying to get interesting brands that are interested brands that are interested, because right now it starts with the brands. It has to start with the brands. It can't start with the vendors. The vendors will get into a catfight and argue about what's this and that and the other. Don't trust the vendors to create a standard that you like. The brands need to be involved in the process. If you want a good standard, the brands have to be involved, 'cause you're gonna live with it. You're the ones who are gonna live with it. So what we're doing now is gathering information about who's interested in participating in the process. And I've done this same kind of explanation in Milan. Had quite a few large brands and some small ones that were very interested in participating. If you're interested in participating, I think, Jane, I can't quite see you, the lights are bright. She's standing in the back of the room. Just have her scan your badge so we know to contact you, keep you informed. You don't have to commit to it. It's just if you're interested in potentially participating in the process, just come to our booth or talk to Jane and we'll get you scanned and we can include you in the process as we connect the brands together and try to create an organization that will allow the brands themselves to— and there are some brands that are potentially hosting this and so forth. So let's go on to the next question. Let's see. Standardization is desired from a product point of view, but is there a thin line between— oh, differentiation. So I don't actually think so, because part of the standardization is essentially the process of how you create the product. So let's take a side slit on a shirt. We've actually purchased a bunch of shirts to take them apart, figure out how all the different ways you can do it. And there's at least 10 different ways to create a side slit. How you fold it, how far you fold it, where the stitching is done. Is there a top stitch on the top to kind of close the slit so it doesn't come apart? All of those things could be different from one brand to the next. Once you create your own standard, you're creating— the standard is not the fact that there's a slit on the side seam of the top. The standard is the instructions that create the slit. It's the set of steps that you have to go through because you need those steps not only to produce the virtual garment, but the physical garment as well if you're going to automate. So the standards there are the instructions. We have discovered that there's about 100 different instructions needed to create all of the different types of finishes that are used on the myriad of garments you go into any store or high street or mall, et cetera. All right, so I don't think you'll see that there's a lack of differentiation. The designers handling the elements such as curved cut lines, flares, and the system, yes, this is all done by Bezier curve tools. So you have a silhouette curve that you can choose what you want to flare it. You can change the curve of it. You have a control over how far the flare goes. This is all done using very simple UI cues, whether it's lengthening or changing the silhouette, or whether the silhouette is on the pant leg or the trouser leg or the arm sleeve, etc. All of those things can be adjusted by, by Essentially by opening shapes, you do all of this with Bezier curves tools and you have complete control over that. The designers adding complex detail construction like concealed placket or vent. Yes, the question is, again, we don't create the finishes. You create the finishes with our toolset, which is all of these commands. You say, okay, take this piece of fabric, do this, cut a hole in it, put the pocket bag in, sew it this way, and all of those steps are done manually one time because it's reusable. And then the instructions are general. So yes, they can be— it'll generate the correct pattern because it's— you're saying how to do it. You're telling us how to do it, essentially.