July 28, 2021

The Automated Microfactory


Augmented reveals the stories behind the new era of industrial operations, where technology will restore the agility of frontline workers.  

In episode 29 of the podcast, the topic is: The Automated Microfactory. Our guest is Brian Mathews, CTO, Bright Machines.

In this conversation, we talk about increasing the speed, scalability, and flexibility of manufacturing using an intelligent, software-driven approach. Can discrete manufacturing, that is, the production of distinct items such as electronics, automobiles, furniture, toys, smartphones, and airplanes, now achieve the same efficiencies that we have seen in the software world? What does the next decade look like?

Augmented is a podcast for leaders, hosted by futurist Trond Arne Undheim (@trondau), presented by Tulip.co, the frontline operations platform, and associated with MFG.works, the manufacturing upskilling community launched at the World Economic Forum. Each episode dives deep into a contemporary topic of concern across the industry and airs at 9 am US Eastern Time every Wednesday. Augmented--the industry 4.0 podcast.

After listening to this episode, check out Bright Machines  as well as Brian Mathews' social media profile: 

My takeaway is that as fully software-enabled platforms take hold, the automation of discrete manufacturing will change exponentially in the years ahead. This has been long in the coming, and the impact is almost impossible to fathom. Factory-level automation is one thing. However, the onset of relatively mobile microfactories and the ability to remotely update, tweak, and even radically improve physical things that already left the initial production facility will not only change timelines, but might alter the very notion of what a product is. Given that the sci-fi that Brian and I both love is coming nearer reality--good luck to sci-fi writers trying to write about the next century's innovations. That is going to take some extra creativity.

Thanks for listening. If you liked the show, subscribe at Augmentedpodcast.co or in your preferred podcast player, and rate us with five stars. If you liked this episode, you might also like episode 42, Business Beyond Buzzwords, episode 21, The Future of Digital in Manufacturing, or episode 27, Industry 4.0 Tools

Augmented--industrial conversations that matter.

Transcript

#29 The Automated Microfactory_Brian Mathews 

[00:00:00] Trond Arne Undheim, host: [00:00:00] Augmented reveals to stories behind a new era of industrial operations, where technology will restore the agility, your frontline workers. In episode 29 of the podcast topic is the Automated Microfactory our guest is Brian Matthews, CTO of bright machines. In this conversation, we talk about increasing the speed, scalability and flexibility of manufacturing using an intelligent software driven approach. Can discreet manufacturing that is the production of distinct items, such as electronics, automobiles, furniture, toys smartphones and airplanes now achieve the same efficiencies that we have seen in the software world? What does the next decade look like?

[00:00:50] Augmented is a podcast for leaders posted by futurist Trond Arne Undheim presented by Tulip.co  that frontline operations platform [00:01:00] and associated with mfg.works. The manufacturing upskilling community launched at the world economic forum. Each episode dives, deep into a contemporary topic of concern across the industry and airs at 9:00 AM us Eastern time, every Wednesday, Augmented the industry 4.0 podcast.

[00:01:24] Brian, how are you?

[00:01:25]Brian Mathews: [00:01:25] Doing great. Trond how are you? 

[00:01:28] Trond Arne Undheim, host: [00:01:28] I'm doing great too. I'm excited to chat with you a lot of things happening since we since we got to know each other and you gave me homework and I've done my homework. So for the people who actually get to see this, I've read the DevOps handbook and you have it too.

[00:01:42] So that was one part of the homework that I think I was supposed to watch a movie as well. And read the book behind the movie so there was a lot of prep for this, Brian. 

[00:01:51] Brian Mathews: [00:01:51] I didn't.  Think you'd actually do it. That's amazing. 

[00:01:54]Trond Arne Undheim, host: [00:01:54] You went to Cornell, I respect people who go to Cornell. You guys are serious people.

[00:01:58]You've been swimming in that little pond [00:02:00] there and there's something magical. 

[00:02:01] Brian Mathews: [00:02:01] I enjoyed it. A lot of good universities, 

[00:02:06] Trond Arne Undheim, host: [00:02:06] There are a lot of good universities. I wanted to just ask you straight off the bat, you've been in a bunch of businesses. You're deeply steeped in technology but now, very much working on manufacturing.

[00:02:20] What brought you into the space? What makes you excited about this space and you obviously have a tech background, but not every bright person in tech saw the light about the, the integration of the physical and the, and the virtual as early, what is it that fascinates you  with this space?

[00:02:39] Brian Mathews: [00:02:39] It's a lot of different technologies, it's been my career in a lot of different industries. So I was trained as an electrical engineer and a lot, the manufacturing work that I do at my current company is around electronics, manufacturing, electronics, prod products. And so on. But my real career was built around software.

[00:02:58]I was in a computer graphics [00:03:00] company and then a CAD company for many years and doing design computerated design in many industries, manufacturing, architecture water infrastructure, you name it. And then I got into a scanning technologies. So things like laser scanning, photogrammetry, things like that, which are measurement and computer vision kind of came in. When bright machines started out I had a friend who was at the company and told me about it and it really combined a lot of different things. Cause you've got robotics, you've got Manufacturing. But  I had seen about this industry is there really aren't many people who bring the thread through all of these things, because what drives manufacturing is CAD, and that turns into simulation.

[00:03:41] How do you simulate your manufacturing? And then that turns into computer graphics. How do you visualize it? And then you've got the actual robotics and so on. But finally the missing link, I think in all of this is computer vision. Robots historically have been blind, numb and dumb, and all three of those things can change [00:04:00] now with new technology, you've got computer vision, you've got sensors that can feel.

[00:04:04] And, they're probably the coolest thing on the block. Now it's all the machine learning stuff going on. And so there's a lot of companies out there that have combined a plus B, you take a simulation and robotics, or you take a machine learning and IOT or this plus that, and what hadn't really seen as anybody who sees how you really have to string all these things together.

[00:04:24] So you really need people to have a background in CAD and how you take design intent out. And how do you marry that up to simulation and computer graphics and material science and robotics, and really bring it all together into a system.. And that's really what excited me about this company and this industry is that we're finally with industry 4.0 and so we're finally at the point where we can do some of these things.. 

[00:04:47] Yes. 

[00:04:48] Trond Arne Undheim, host: [00:04:48] And that's where I want to get in a second. But before we get there, I have to bring this out because not only did I do my homework on the dev ops, but your favorite robot is tars and [00:05:00] there's a story behind this because the movie interstellar obviously has tars.

[00:05:05] You're not just a little bit fascinated with robotics and these kind of infrastructures you really live them. What was it about tars that brought your, that got your attention? 

[00:05:18]It's not incremental thinking. So if you think of every robot that you've seen in every Hollywood film, they all follow the same paradigm of looking techie and so on.

[00:05:27] Tars was a complete rethink of what a robot is and how it works. And it was really functional for a movie to come up with that kind of a design. This is a a robot that was made to be a military robot, tough as nails and the way that it articulate lates, the way that it moves very interesting.

[00:05:47]I'm not going to spoil it, but it's not like any other robot that you've seen and there's some really good ideas in how it actually works now. Physically, it wouldn't work the way it works in the movie. It is a movie at the end of the day but there are some interesting ideas in how it's [00:06:00] put together.

[00:06:00]And I wanted to bring that out just because you said something, you said, we're finally now able to do industry 4.0 What is industry 4.0 mean to you? It means a lot to a lot of people. It's actually not a term that is used very much into US. Smart manufacturing has at least been the industry's term for it but of course in Europe and with the world economic forum and where the term originated.  It's a bigger term you explain it to me. How, what does it mean to you? You said we're finally able to do industry 4.0, how do you explain that? 

[00:06:33] Brian Mathews: [00:06:33] Industry 4.0, there's a lot of actual academic papers that define it and I don't use those definitions. I, use the lay person. So to the lay person, industry 4.0 is just really the digitization of the whole process. And when I say we can finally do this, I'm really talking about a bigger picture of take the concepts of industry 4.0, which is once you've digitized something.

[00:06:57] It really isn't even about the digital data. [00:07:00] That's important. What is your goal at the end of the day? What are you really trying to do here? And what I think we can do now is because of digitization and these different technologies, machine learning, combined with vision combined, with CAD and simulation and so on is that we can automate entire workflow.

[00:07:17] And I think that's what this is really about. If I look at other industries, Salesforce doesn't sell anything, Workday doesn't hire anybody service now doesn't fix anybody's bugs. What are these really successful companies really do. They automate the workflow of doing work and in a way they automate tasks. So they're automating machines, but they're also automating the humans that are participating in the process of working with the machines. So this concept of automating automation, I think is a very interesting concept. And that's actually where my background is that, while I've danced around the manufacturing industry, from a CAD space and cam and simulation and all of these [00:08:00] other parts of manufacturing and work with a lot of manufacturing companies. I'm not a actual manufacturing person, but where I've spent a lot of time is in cloud computing. And that the reason why I like that dev ops handbook that, there's the ad again for these guys. I don't know them personally, but the big aha in that is if you look at what happened to the cloud computing market, it's been amazing. And what did happen in the cloud computing market? That's the big question. What was the change? Cause if you go to United airlines 20 years ago, you had a networking's server, you had a database and you had a file server. And today in Amazon web services in the cloud, you have networking, you have a database and you have a file server, like nothing changed.

[00:08:40]But everything changed. And I think if you take a step back and look at these innovations, the innovation wasn't about actually doing the work. Software engineers don't write more software today than they did 20 years ago, writing Photoshop or AutoCAD or whatever. Where a lot of the work was in the tool set to automate the process of [00:09:00] software tools.

[00:09:00] And so when I look at manufacturing, I think, it's the same thing in software where did the concept of Kanban come from? Where did you know agile come from? Where did just in time come from, it came from Manufacturing, but I feel like Manufacturing stopped and they stopped innovating along those same lines where the software industry was just like manufacturing. A Toyota Camry comes out every four years and so you only do certain types of testing every four years. You only invest so much in automation. And the software industry how often did windows come out, 20 years ago? How often did Photoshop ship? It used to be every four years, but how often does it ship today?

[00:09:38] You get a new copy of, AutoCAD every three months. And so there's something really different about how software is being manufactured versus how products are being manufactured. And I think that trajectory, if you peel away at what has happened in cloud computing, what is there's a huge set of tools around managing the entire process and automating the [00:10:00] process, not automating the actual work..

[00:10:02] And I think in our industry, we focus putting screws in holes and automating that and not automating the workflow. 

[00:10:09] Trond Arne Undheim, host: [00:10:09] Yeah, and that's super interesting. And I guess what starts to happen then? What started with the cloud computing market also is that there were a lot of interesting roles, suddenly specialist roles that were available to non-computer engineers who are using the automated tools where they could actually do automated, efficient things, even not being software engineers. And I guess some of that stuff is starting to happen finally one would say in, manufacturing and just about time, since the industry inspired every other industry.

[00:10:42] And then it stopped, like you said for a while, and now it's very much cool. Again, I wanted to just bring, I guess the interstellar metaphor back one more time, just because  what you told me about it was right. It's so fascinating because it's real. So if you reflect on the industry 4.0, there [00:11:00] are, of course, a bunch of shiny things there. There are robots and the trope of a robot has been around for so long. And like in many other technologies, it is quite disappointing when you actually then look at a real robot, having seen a robot in the movies, many times, but that was one of the things you said was so fascinating with interstellar is that I guess Jonathan Nolan spent a lot of time making sure that this robot wasn't just fake, that everything that was happening in there was no, not only fascinating, but it was physically possible.

[00:11:31] Brian Mathews: [00:11:31] That's true and it wasn't just the robots. It's a, they actually, there's a Nobel Laureate, Kip Thorne that the was a consultant to the movie and the physics of the movie itself are based on possibly. Not profitability, but yeah, if you if you think in this industry that ability to do things with robots is hard today.

[00:11:50] Today, you got to teach pendant, you're working in a real-time operating system with something like a structured text language. If you teach a robot how to move in a certain way to a [00:12:00] certain coordinate usually that that programming is done in a vendor specific language that is not a modern programming language.

[00:12:07] The User interface is quite old. A lot of the hardware companies that make that software, the software in some ways is an afterthought. Once you've made the hardware sale, you really don't care about that user interface and the people who are programming robots are system integrators. And so there isn't a lot of self-service from the end customer the end OEM. And that creates an interesting dynamic there and so I think all of those things are right areas for change because they were the same way in, in the software world. You used to have languages that could only run on a Mac or only run on a PC or only run on a mobile phone.

[00:12:41]You had Blackberry apps and today you have web apps that run the same everywhere. You can write out one app, you have virtualization systems. You have tools that manage, think of GitHub, think of Terraform that configure your system, that deploy your code editors and so on debuggers. And when I look at the [00:13:00] manufacturing market, I see PLCs, I see real-time operating systems I see structured text and ladder diagrams, and I see protocols that are old. And so I, in fact, I did a search a while ago. I was looking for a book on structured ?Text programming. Do you know how many, if you type that into Amazon, guess how many you'll find ? 

[00:13:21] Trond Arne Undheim, host: [00:13:21] I have no idea. 

[00:13:23] Brian Mathews: [00:13:23] Five, five hits, but if you look a little closer, three of them are the same book, just different additions. So now if you type in Python what do you get? And so this is the problem. I think we're not taking advantage of all of the research that has gone into these other industries, where we could apply these same exact concepts to this industry. 

[00:13:47] Trond Arne Undheim, host: [00:13:47] So you call software defined Manufacturing, a new trend.

[00:13:51]When did that start and how can you just line up some examples of the that particular movement? When did that really start? When [00:14:00] software for real started to. Because you're not just talking about having software traditional MES system in, in manufacturing, what I'm understanding from your term, maybe I'm wrong is that you're talking about this cloud aware like low code version of software and when that moved into manufacturing. Can you line up a little bit, when that started to happen and what characterizes, the kind of things that, that changes in the industry when these software layers are becoming more flexible. 

[00:14:27]Brian Mathews: [00:14:27] And I think, again, this is an example of one where, we didn't invent these concepts.

[00:14:33]These are already proven elsewhere in other industries. So where did these concepts come from? There's a movement called SDX software to find anything.. And you can go Google that and read about SDX, but there was a software defined radios is probably one of the first examples of this concept where, you used to have an am radio or an FM radio, and then your cell phone, there was a PCs and CDMA and GSM and every time you wanted to change protocols, like the hardware physically encapsulated [00:15:00] that modulation scheme in, in hardware.

[00:15:03] And, at some point there were these things called software defined radios, where they would digitize the wave form coming off the antenna and they would do the demodulation through mathematics rather than a bunch of capacitors and inductors and resistors, they would find it and do the demodulation in software.

[00:15:23] And then, the cloud computing world has infrastructure as a service or platform as a service. And there's a software defined networking, software defined storage. And what is software defined networking? In the old days, you used to have to, if you wanted to have a secured network, you used to have to physically cable this networks, which to that network switch.

[00:15:41] And then Cisco and others came out with software defined networking, where you can create what's called a virtual LAN. So when I plug my printer into this port of a network and I plug my IP telephone and my whatever, if I don't want those two devices to be able to talk to each other, I define that network in software instead of in the hardware..

[00:15:59] So if we [00:16:00] bring this into manufacturing, this is, the, one of the big ideas is take a conveyor. You can go out to Bosch or YJ link or any of these guys, and they'll have a catalog of a thousand different conveyors. And the traditional way is if I'm trying to make, this mouse here, I'll go get a pallet.

[00:16:17]That's the right size for that and I'll go to my catalog and get the right conveyor for that product. And that'll be the cheapest way to go but when the, when I want to make a different product next. What do I do? I throw the conveyor away I get a different conveyor. A software defined manufacturing approach is to get an, a conveyor that's motorized that has a shaft and coder and knows what its width is set to and you load a recipe and it changes its shape, we've done things like a heat sink installers where, there's 250,000 different types of heat sinks on them. And you can go to any number of companies and find thousands of different grippers for your robot that can pick up heat sinks of different shapes and sizes. But why not have ones that have motors in them that it's under software control, change their shape.

[00:16:57] It's going to cost you more, but you can reuse it [00:17:00] for every future project you're ever going to have. So those are some ideas behind software defined manufacturing. If you look at a Tesla or an iPhone,  what they were compared to their predecessors was a software for of first approach. It's still hardware, but the hardware, isn't the thing that's meant to do the work it's meant to be controlled of sensors and motors and so on.

[00:17:21] It's meant to be controlled from a software first approach. And then what you were saying, there is absolutely part of it, which is how do we layer on that entire workflow? So low-code no-code environments. How do we do what I call declarative programming. Which is you don't say, do step one, step two, step three that's procedural programming. That's the way we program robots today. But if you think ahead, you can be, goal-oriented I want to pick the screw up from the screw presenter, and I want to put it in the heat sink. And why am I telling it the XYZ coordinate when the CAD file knows where the coordinate is and the answer to that as well, because what the CAD file says and what the robot actually has in front of it are always [00:18:00] two different things.

[00:18:00] So that's where computer vision comes in. So if you have all these, so you put them all together. I shouldn't be programming robots in the first place. Things should just be doing the stuff on their own. That's a declarative approach I declare the goal. 

[00:18:11]Trond Arne Undheim, host: [00:18:11] It's fantastic to  hear you explain this because I'm getting an analog a little bit to like the way that we have defined 3d printing, for some people it's just oh, it's actually, in their head, they had this little printer that was, printing ink.

[00:18:27] And then of course the concept broadens and you can start printing other material, but then if the size now starts to change, which it is right also, and not just materials that changing and, desktop metal is now all a project or they're even printing a wood product. But anyway, what you're talking about is essentially automated the notion of a factory. So it's not that you're just necessarily printing anything, but you're modularizing everything about the surroundings. 

[00:18:55]Brian Mathews: [00:18:55] And let's maybe back up from that because it's even the design. 

[00:18:58]Trond Arne Undheim, host: [00:18:58] So it starts with the design [00:19:00] digitally and then you're essentially modularizing the entire.

[00:19:05] I guess the entire workflow, which is all of manufacturing. It's fascinating. So how does that, so what is that micro factory look because right now actually it's micro factory, in the future, I'm assuming the size there will increase as well. But this notion of automating the factory and creating kind of a software defined factory, what does it look like today in the marketplace?

[00:19:27]You have a product in it, but what are some of the other concepts out there that are starting to come on the market where you have essentially said there is a new factory available and, it has some of those software defined features. What does it 

[00:19:41] look like?

[00:19:42]Brian Mathews: [00:19:42] It depends on which industry you're in. Let's just take electronics, manufacturing, where you're making mice or, servers or carp car parts, or, electronics and. And if you're in that, the front of the line where you're making your printed circuit board assemblies with all the components on all that's been highly automated for a long time, because [00:20:00] there's standardization there, you have standard board, you have standard SMT components that go on top and there's standardized machines that can make all that.

[00:20:08] And so that's  fairly software defined way of manufacturing that part, but it all falls apart at final assembly and inspections where you've got to get your plastic case and your battery and your speaker and your, and you gotta put it all together. So that's really where the microfactory concept comes in is the part where there's a lot of human labor now.

[00:20:27]So if you look at the implementation, you go into a factory where you have things like micro factories, you're going to see cells and robotic cells and the different. It, at first glance it may look the same, but the difference is that things are very modular. Traditional system integrators and traditional automation professionals there's a lot of customization for each project.

[00:20:46] So you may choose which robot, which conveyors, which safety systems, which PLC, which everything. And, our approach is to standardize a lot of that stuff. If you look across all these different projects, do I really need, 18 different [00:21:00] safety systems? Do I really need that many different kinds of conveyor belts and robots and so on?

[00:21:04] And so why not make this more like Lego bricks? And standardize these things, and then you can have them in inventory. You can have them, at lower cost at higher volumes reusable on another project. So don't just think of building these cells for this project and depreciating that asset for just one project.

[00:21:22] Think about the, appreciating it into the future. It changes the whole cost dynamic, the business model, everything changes when you think that. So that's part of it. That's the hardware, but then let's back out and say, okay, how do you program this stuff? Because what is, what's the most manual thing you can do in a factory?

[00:21:38] What's the most expensive and slowest thing. It ain't putting the screws in the thing that often is the most manual is deploying automation. And that's very ironic that's why people don't deploy automation. I can teach you to put a screw in a hole in five seconds. If I want to teach a robot, how to put a screw in a hole that's a lot of work and that's really what we need to change. So yes, you get [00:22:00] modular hardware, but it's really in the authorizing system. How do you show intent to the machine? And this is where the low-code no-code environments come in, where you can give intent in a in a more graphical approach that is more self-service that maybe doesn't need as much of a system integrator.

[00:22:18]Or if you do need a system integrator, maybe. You don't need them to continually come back to the factory for every new product revision. You can just tweak the recipes that you already have. 

[00:22:28]Trond Arne Undheim, host: [00:22:28] But that is just the death of the, it's just the death of the system integrator. Or is it just similar to that this is the robots are not the death of anyone? It's just a transfer of different doing different skills. There's still gonna be services that have to be executed. There is some automation, but there's, so tweaking .

[00:22:47] If you look 

[00:22:47] Brian Mathews: [00:22:47] at any other so I had a friend once who said, robot is what you call something before it becomes useful.

[00:22:53] After it's useful, you call it a dishwasher or a vacuum cleaner or something else. If you look at ATM machines, [00:23:00] for example. How many tellers are there working in banks today compared to before the teller machine. So when you make something cheap through automation, you actually consume more of it.

[00:23:09] So you're absolutely right. I don't think it's the death of system integrators system integrator. There's always going to be the last mile. It's what's on the end of the arm of that robot, that gripper, and that pallet is custom. It's custom to that product and I don't see that. I see it being improved through software approaches and using CAD and so on, but I don't see it, that's not going to change overnight.

[00:23:26] That's something that, by the time I retire, we'll get some big inroads to that someday. But what you want to do is make the quality better allow for innovation to be cheaper and so on. And that means backing out and say, not just having this modular hardware and low code environments to program, but how do you, what is your entire workflow from design?

[00:23:49]We've always had the person who designs the cell phone to the person who's manufacturing the cell phone that linkage existed, but it wasn't very strong. And in some companies [00:24:00] I've even seen, it's throw it over the wall and it's somebody else's problem. And there isn't a, there isn't a good loop.

[00:24:05]If you look in a, if we go back to the, my cloud computing example, the industry where I came out of, what is the equivalent of GitHub in this industry, what is the equivalent of Terraform? What is the equivalent of service now or JIRA? What is the equivalent of visual studio? There, I really don't see it in this industry.

[00:24:20] And so if we think of software defined manufacturing, let's not just talk about, the, on the robot. Let's broadened this thinking all the way from the CAD designer in the design department, who is deciding what kind of screws to put in the holes. And shouldn't we be using data from the factory to instruct the design itself, designed for manufacture.

[00:24:42] Trond Arne Undheim, host: [00:24:42] One of the things you said earlier was about the standardization. And I guess you stopped short of talking about interoperability, but certainly for robotics, that's been a big issue. I know there, there are some steps forward now. Some reason announcements of on, among several firms to do work on robot [00:25:00] interoperability, but sure it's inefficient. And it has been inefficient when everyone's creating, everybody loves to be a platform these days, but essentially, in the old days in manufacturing, like everything was not just the platform, but, there were customized factories that would build, like you said, they were built-ins in stone and metal and they couldn't be changed.

[00:25:19] So not only were you locking in your own project trajectory. And so there was a reason why there was only a Toyota every four years, because they actually have to retool the entire factory to create something else, but you are, of course not able to communicate across with even your suppliers or anything like that.

[00:25:37] How is all of that now slowly changing with this new paradigm? Because there's surely is still the incentive to try to lock in some aspect of the process in order to reap some sort of profit in the, at the end of the line. 

[00:25:50]Brian Mathews: [00:25:50] You're right and if you go back to the computing world, I'm sure if you were Blackberry or apple or Android,  there is an attempt by companies to try and lock [00:26:00] people in.

[00:26:00]But at the same time, if you look at what the web has done in various protocols you can still have very proprietary environments that inter-operate with others. And this is that in web servers all of the time, whether you're on Amazon web services or Google cloud or Azure or whatever, what have you, the protocols of how services, how machines talk to machines has been standardized in that world. And you see a lot of efforts in our world in manufacturing with, OPC, UA and Hermes, and all of the different protocols, there's movements to go in that direction and that can help with some of this. But I think this idea that everyone has to talk the official standard protocols and that you're going to get standards.

[00:26:39]The beautiful thing about standards is there's so many of them and no one's going to agree on them and every company wants the standard to be their own. So instead of trying to fix human nature, and try and get to this utopian world where there's this universal standard. How did the software world deal with this?

[00:26:57] And they did it through drivers, which is [00:27:00] encapsulation. So let's come back to this mouse again, this is a logic tech mouse. It's working with Microsoft windows and it works with a Mac, same mouse. Now the protocol of how the mouse movements and the button presses and all that stuff. The protocol of how it gets from here through the USB and into the machine and the language that the logic tech mouse speaks, which by the way, it has non-standard buttons it has its own protocol. How does that get into Microsoft windows or into Mac and work? And it is still proprietary it isn't a standard communications protocol, but it's been encapsulated into a device driver and Microsoft word or PowerPoint doesn't really care which mouse it is, it just needs certain types of events and communication and it works.

[00:27:47] And why can't we use the same kind of analogy in the manufacturing world? I don't need to have a hundred percent standardization of everything. What I need is encapsulation of differences. So again, if I come back to a [00:28:00] goal centered language, if I can define a language that says, I want to pick the screw up from here, and my target is to put it in that hole over there.

[00:28:07] Do I really care in that recipe that I just said that procedural declarative goal oriented language, do I care if it's a four axis robot or a six axis robot? Because if I tell the robot what I want it to do in terms of a goal, instead of move this joint, w to this number of degrees, and instead I give the goal, then device drivers in lower-level encapsulation can do the translation..

[00:28:32] Now what I get is reuse because if I've defined a recipe at a high level that can do this operation of putting screws in for this product. When I come out with a similar product, I should be able to reuse a lot of this even if I can't reuse that encapsulated difference. It just minimizes the work.

[00:28:52] Trond Arne Undheim, host: [00:28:52] Let's take it to people back to people again for a while. Cause we talked about dev ops and dev ops is essentially developers and [00:29:00] operations people. Talking together and building a joint process in a different way and that's what happened in the cloud industry. There is a similar issue happening it seems in manufacturing between it and ops.

[00:29:12]Because they historically, even in manufacturing where there was it, if there was it was the it department dictates and then ops just deals with it and has all their concerns and is waiting for whatever system they have to work with. That seems to be one, one dynamic.

[00:29:26] But if you think of it from like a plant manager, point of view,  what does a plant manager have to do to either stay ahead of these trends or start implementing the kind of industry 4.0 that, that you are have been describing here? Who are the actors that can actually make a difference in terms of how an individual factory makes use of these tools or indeed how the entire sort of industry shifts, what, who are we waiting for to make this happen faster? 

[00:29:57] Brian Mathews: [00:29:57] To make it faster? I would say we're [00:30:00] not waiting for anybody. I think we get analysis paralysis and we look at industry 4.0 because it has big implications against every  task, everything from, design for manufacturing to how you program your robots, to how you maintain them. And we haven't even talked about IOT and all the rich data that you get out of manufacturing and how we're going to use that. So you look at all that workflow, there's so much to change.

[00:30:24] There's so much opportunity. It's very easy to just throw your hands up and go back to the old way. So I think the first thing is is like any kind of innovation don't invent innovate, and what is innovation it's taking other people's inventions and figuring out how to get business value out of them.

[00:30:42] And it's not an all or nothing thing. You don't need to take an entire line and automate every aspect of that line. You can start small and run an experiment that's cheap. This is there's another book that's a favorite book of mine from the software industry. It's called the lean startup and the whole premise of the lean startup [00:31:00] is that when you're trying to innovate, the most important thing to do is to run as many experiments, many inexpensive, cheap, safe experiments that you can, and, effectively find a way to take risks, cheaply so that if an experiment fails, it hasn't really been a failure. It's given you something which is education.

[00:31:20]And so you can do that in a factory like why not take part of your line and audit the labeling and the screwing phases of it and keep everything else the way you have it today and learn how to bring in industry 4.0 techniques into that. Then maybe if you want to use some IOT and see what your first piece RFP, why is your , your various types of metrics, just start with some basics and then grow from there.

[00:31:44] But I would caution again incrementalism and I think this is another problem I see in the manufacturing industry especially in contract manufacturing companies, is that the product that you're offering, this ability to put things together is the same as [00:32:00] everybody else. And you're using the same kind of equipment and you have the same kind of humans with the same kind of training.

[00:32:05] And so I always ask people, what is your strategic difference from your competitor?How are you going to be different? Because if you're not different, if you're a commodity, it's a race to the bottom on price, everyone assumes quality or they won't use you. So what are you really doing different?

[00:32:20] So I think you want to do these experiments and learn how to innovate cheaply, but at the same time this is what I like about space X is they had a big freaking huge audacious goal. And not everyone knows what space X is real goal is, but it's to colonize and so it's a it's to colonize Mars.

[00:32:41] If you go into their, reception, there's a picture of Mars there, that's what everyone in that company understands what they're really trying to do. And we in the press, we read the newspaper and we see, oh, they're going to get me nice wifi in a remote area. And they're going to put my communication satellite up and they're going to do this, that and the other.

[00:32:56] But what they're really trying to do is colonize Mars. They want the, they want to [00:33:00] be the, have a backup for humanity. So they started with a big audacious goal and they work backwards. So if we start with this big industry, 4.0 goal of designing for flexibility having code reuse, not design everything for just one project, but for all future projects, having our hardware be reusable so we can depreciate it, not just on this project, but all projects. And you get into this virtual virtuous feedback cycle where when you have a version control system that's keeping track of all recipes you've done for putting screws in or labels on or glue or soldering or whatever the process steps are and you start building these libraries that you can reuse. Then it becomes cheaper to do the next project, which means you can reinvest in more of these process steps, more of this automation of automation you have version control and how do we do a sign-offs and design reviews and so on. It's everything that we do in the software industry.

[00:33:53] None of these are new concepts. We just have to implement them here and that's what I would do start with a big audacious goal in [00:34:00] mind, and then find the cheapest way to test your assumption that this is the correct path. 

[00:34:06] Trond Arne Undheim, host: [00:34:06] And then Brian if we do that, what can be achieved because, it's a truism these days and it has been for awhile that, manufacturing is difficult it moves slowly. Yes, we are trying to innovate, but it's gonna always be slower than these easy guys in software who sit in a garage and invent software companies. And they have an easy time because yeah, sure. You're doing a brainiac kind of a challenge and you're solving something then no, one's diminishing that.

[00:34:34] But so you'd say hardware is different because there are all these physical limitations. Are you saying that by automating automation, if we're stepping into more like a futurist reference what is already being accomplished, like even just medium term, like next 2, 3, 4 years.

[00:34:50] And then if we look at a longer term, like seven to 10, or even longer than that, when you play the book, the playbook forward on [00:35:00] automating automation, where do you end up do, you do you end up colonizing Mars? Are there other interesting things that we're not even looking at? That's going to be possible now through this.

[00:35:09]Is it possible to think of manufacturing 10 years from now as an industry that innovates at the speed of what we see software innovating now? Is that crazy to think? 

[00:35:20] Brian Mathews: [00:35:20] I don't think it's crazy to think at all. And I would go back and I will call BS on this thing that manufacturing is different or harder.

[00:35:28]When I go into a factory, I see exactly what I saw in the software industry 20 years ago, 10 years ago, lots of keyboard screens in mice, lots of equipment from different manufacturers none of them made to be under programmatic control. All of them with dials that humans are supposed to push. If you went into a data center 20 years ago, you would see tons of keyboard screens in mice.

[00:35:50] You would see a network engineer plugging into their Cisco router, setting up the route tape. Okay, manually logging in. If you go into data centers today [00:36:00] there are no keyboard screens in mice in them. And the other thing is let's talk about just the number of manufacturers in a data center you've got water pumps, you have power management systems, humidity control systems, CPU's GPU's FPGA networking, gear, all of these from different companies.

[00:36:16] None of them had API APIs or protocols or were meant to be automated no different from Manufacturing. I don't see any difference between a data center, which has manufacturing bits and a physical manufacturing center, making Adams. I see them as the same. The difference is, and that's why 20 years ago, AutoCAD and Photoshop shipped every four years, just like the Toyota Camry, because it was the same problem.

[00:36:38] What changed is and let's ask the question. Do you know how often Facebook makes a production level change to their product? That affects users, production level, not a test that they're doing in some little test thing. Do you have any idea how many changes they make? 

[00:36:53] I don't know. Every week

[00:36:55]Trond Arne Undheim, host: [00:36:55] they make 1,500 changes every single day.

[00:36:59] Brian Mathews: [00:36:59] How often does [00:37:00] their site go down? Can anybody think of all the services from there? Instant messaging to video conferencing, to think of all of the services that they have. And you've never seen that thing go down. Maybe once a year, have a bug on one section of the site. And they're making 1,500.

[00:37:17] If you change the Toyota Camry, 1,500 design changes a day and Friday to get that into manufacturing, what would your quality be now? What costs did they do it at the reason why auto CAD and Photoshop, and these things used to ship every, four years back then was because the testing processes were manual..

[00:37:34] And because the deployment processes were manual and any change you made was so expensive and so risky. And this is what I see in factories today we make millions of identical things. We can only afford to use automation where we're going to make a million. And what, if you want to make a small number of something that our systems aren't flexible enough to do that.

[00:37:53] And so what you need to do, and that's what companies like ours are doing. We're trying to overspend in this R and [00:38:00] D of making systems that make it easy to make changes because you only have to do that once.. It does take a lot of capital. It does take a lot of engineering to simplify these things and automate these things.

[00:38:13] Automation is hard and it's expensive, but the thing is you get to reap the rewards for all future. 

[00:38:18]Trond Arne Undheim, host: [00:38:18] I wanted to challenge that a little bit, and I think you have challenged it in a, it also in previous discussions with me when it comes to it used to be that only big companies could produce value in many industries.

[00:38:31] And now you're going back on it a little bit in the sense that this seems to be an industry where there's both going on, like in, there are some major investments that do need to be made. Because there's a certain inflexibility that has been built up, and now I'm making up the story here.

[00:38:48] You can disagree with it, over time we have created all these inefficiencies and in orders that are breakthrough, that you sound have to invest with some scale to it. But on the other hand, I guess the software industry has [00:39:00] also proved that certainly for a decade or two, now a bunch of startups have come out of nowhere, although, no one really does, but they have come out of universities to be honest, right.

[00:39:12] With clever things and have really made a big difference. How do you see that in manufacturing? Is it possible over time to come out of nowhere? Whatever that really means, to come with some clever things. That really changes everything or is it even though it's fast change, it's still have to be, of a certain scale.

[00:39:35] Brian Mathews: [00:39:35] I think you can. I think it's highly analogous. Software did take 20 years to get from, the windows every four years to, to 1500 changes a day at Facebook, it took it a long time and it wasn't one invention that made that happen. You had source code control systems and languages and environments and debuggers.

[00:39:55] And so I think that's what we're talking about here. It's not a [00:40:00] Zero or one it's not an off on it. There's a whole level of gray in between there and I think that's the interesting part of industry 4.0 to look at is start with that end in mind and how do we work backwards? So what could we do in manufacturing to make this better?

[00:40:15]Let's take a computer vision. People spend a lot of time trying to get the right coordinates to get those screws to go right and there are companies that have done digital twins and simulation, but they generally don't work because the real screwdrivers, a millimeter longer than it was in the simulation and you have to do everything over twice. So the auto industry uses a lot of simulation and other industries tend not to, they can't afford to But when you bring computer vision and it's getting better, there's a lot of vision stuff that's been around for decades. It's been 2d, you've got all the self-driving car, you've got machine learning, you've got computer vision, which is going through a Renaissance.

[00:40:49] There's more happened in the last three years than in the last 30 years. And so these things are coming to bear. So now all of a sudden a problem that was two X, you could have solved it, but it was too expensive and too slow. [00:41:00] Now it's not, that's not true anymore. So I think people have been in this industry for 20 years.

[00:41:04] I haven't been, but what I see is things that have been hard problems for a long time. If you just borrow some of these techniques from some other industries and say, some of these problems have actually been solved now they weren't three years ago, but they are now machine learning is pretty amazing.

[00:41:20] And one of the things I'll hear with machine learning, by the way, Is that machine learning is great. If you give a machine learning system, a lot of data, a lot of examples that this is a good part, and this is a bad part and I label what's good versus what's bad. I could have a machine do inspection for me, for example, and inspection in a lot of plants is about 20% of the labor is inspecting the work of the other labor. And so if I can, use computer vision to do that would be a big deal, 20% savings would be pretty big. The problem with machine learning is it needs a million examples of something before it really learns how to do it.

[00:41:53] The human can learn in a few minutes, right? And so in the last day of production of this mouse, it's ready to take over. [00:42:00] And how do you make it take over on the first day of production? And where there's no data to train the machine learning system with and this is where again, I think if you look at other industries, Look at computer graphics, we can do what is called synthetic training data, where you take CAD files of this map.

[00:42:16]And if you think of all the special effects you've seen in the movies, when you see avatar and iron man and, spaceships they all look real. They all have shadows and Glint and glare, and, you can have computer graphics technology apply to CAD files, create synthetic images for synthetic robots and do synthetic machine learning training to do inspect.

[00:42:35] So you can create synthetic scratches and train these things. So these are the kinds of concepts, where again, if you look at other industries and you bring it here, things that seem impossible or really are possible now. 

[00:42:48] Trond Arne Undheim, host: [00:42:48] I wanted to ask you one more question, which is more to this point of a lot of these things seem fairly complicated yet we are dealing with specifically in [00:43:00] manufacturing, a very different workforce, at least then the typical soft in the software industry and certainly with than other industries, in a sense that it's a big, traditionally, a very big workforce, right? It's a big part of the economy. And it's been universally recognized by many people that re-skilling is going to be a massive challenge going forward.

[00:43:24] What is your answer to what kinds of skills are needed in this new world of automating automation that we've just been speaking about for the last a bit of time, how are we going to do that? And if you are an individual who is either in this industry already, or young talents or contempt contemplating going into this new manufacturing environment, which by the way, I think both of us would recommend it's actually gonna, it's heating up.

[00:43:52] It's an interesting place to be. It's not just a joke. It's not, we're not making this up. There's like advanced technologies that you can experience, but [00:44:00] how do you, how can you as an individual gain new skills and how can we as a society facilitate, I don't know what it's going to take, but essentially.

[00:44:09] We need a billion people who are in this sector to essentially continuously be up to date. How's that going to happen? 

[00:44:18] Brian Mathews: [00:44:18] I think it happens at three levels. Let's take that ATM machine it's a robot . All right. And there's three levels. There's the person who wrote all the cloud software and security software and all the operating system in the ATM.

[00:44:32] There's the person that comes in services, the ATM machine, and looks at the debugging code and finds out that there's a motor that's jammed and it needs lubricant and that's why it's not working today. And then there's the person who uses the ATM machine and they're at three levels and you need all three and all three are growth opportunities.

[00:44:50]Quite frankly, not everyone is going to do the most sophisticated writing the device drivers that encapsulates the complex, the complexity doesn't go away. If anything, we have more [00:45:00] complexity in the world, not less. So how do we encapsulate the complexity in a way that yes, there need to be these experts that are even more sophisticated than the most sophisticated people we have today.

[00:45:12] And we have to train those people up and they need to go to university and they need to learn this. Then you have this middle level, which are the people that, that need to not author these systems, but configure them, integrate them and those system integrators, aren't going to go away. I think the analogy there back, if you think of, in my old world, there used to be network engineers who I told you would program a switch today when you have a million networks, which is it, a Google data center, you can't do it that way anymore.

[00:45:42] So they write a script and so all of my network engineers had to learn. Python. They had to learn a programming language in order to do their job. And, a bunch of them were really excited about that some were begrudging and some didn't want to, and there was about a third and a third.

[00:45:58] So a third were a little bit left [00:46:00] behind and that's a real concern, but that is a re-skilling problem and not everyone wants to be re-skilled. But then there's that third level, which is the user of the ATM machine. And I think that's where bringing in really good user design and designing this thing, like an iPhone rather than for a manufacturing industry, this is what we've seen in the software industry, the amount of software that you have to do today to, if you have a new I don't know you've got a guitar in the background.

[00:46:26] Let's say you wanted to make some software to teach people how to play the guitar. In the old days, you would have to learn how to do your own backups. You would have to learn how to manage a database and do sharding and load balancing and all of this technology today in this public cloud infrastructure, as a service or cloud as a service you just write the actual logic. You care about the business rules of how, what your app does and all of that other stuff is done by somebody else. And that's what platforms do. So I see the same thing in that system integrator level, there's [00:47:00] a whole continuum there of sophistication and encapsulation of complexity.

[00:47:05] And then finally, when you get to the end user, if we have this great user interface, which is declarative it allows people who weren't even part of the process, the end customer, the end OEM, who used to have to outsource it, these integrators all the time, they now have something they can tweak that they can turn, they can change their product a little bit.

[00:47:25] I get a new supplier and the dimensions are a little different and I can just tweak the recipe. I don't need to understand all the device protocols and stuff, but I can adapt to this new part, and that is what is going to make Manufacturing cheaper. It's also the interesting thing about robotics is it allows we've the last major it innovation in manufacturing was globalization, offshoring all this stuff. But we've run out of, countries to off shore to, we're not making any new countries and the standard of living has increased. I did some research a while ago, found out that there's about 400,000 people that joined the middle-class every day, because [00:48:00] globalization has brought so much money to these new areas that the middle class is rising and then they're consuming more, they want their dishwashers and stuff.

[00:48:08] And so it's this virtuous cycle and those people don't want to go into manufacturing anymore. So we have this crisis of tariffs and COVID and broken supply chains. And you've got container ships that are all bottle-necked and all of this offshoring stuff seemed like a great idea then, but now we're seeing customers that want to build things closer to home.

[00:48:26] And if you have a system that is more self-service, that has a user interface, that's easy to use. You can afford to do that and it doesn't cost any more to run a robot in Germany than it does in China. It's the same cost to run the robot and that is an interesting insight as we move forward. 

[00:48:46] Trond Arne Undheim, host: [00:48:46] I agree indeed. There, hasn't just been one interesting insight has been a string of them from you, Brian. I thank you so much. This has been a very insightful, thank you. 

[00:48:58] Brian Mathews: [00:48:58] Sure, sure thing. Good talking to you [00:49:00] Trond.

[00:49:02] Listened to episode 29 of the Augmented podcast with hosts Trond Arne Undheim. The topic was the automated micro factory our guest was Brian Matthews, CTO of Bright Machines. In this conversation, we talked about increasing the speed, scalability and flexibility of manufacturing using an intelligent software-driven approach can discrete manufacturing. That is a production of distinct items, such as electronics, automobiles, furniture, toys, smartphones, and airplanes now achieved the same efficiencies that we have seen in the software world.

[00:49:39] What does this decade look like? My takeaway is that as fully software enabled platforms take hold, the automation of this  manufacturing will change exponentially in the years that this has been a long year and the coming and impact [00:50:00] is almost impossible to found. Factory level automation is one thing. However, the onset of relatively mobile micro factories and the ability to remotely update tweak and even radically improved physical things that already left the initial production facility will not only change timelines that might alter the very notion of what a product is given that decipher that Brian and I both love is coming near a reality.

[00:50:29] Good luck to say five writers, trying to write about the next sanctuaries innovations that is going to take some extra creative. Thanks for listening. If you like the show, subscribe, Augmented podcast.co or in your preferred podcast player and rate us with five stars. If you like this episode, you might also like episode 42 business beyond buzzwords episode 21, the future of digital and manufacturing or episode 27 industry 4.0 [00:51:00] tools, Augmented industrial conversations that matter.

 

Brian Mathews

CTO, Bright Machines

Brian Mathews is Bright Machines' Chief Technology Officer. A technology futurist who has spent three decades architecting cloud and desktop software for designers and engineers, Brian most recently led a 500-person platform group at Autodesk that was responsible for software development, cloud operations, product security and compliance. Previously, he led software teams developing 3D reality capture, computer graphics, augmented reality, data compression and other products. Brian holds a bachelor’s and a master’s degree in electrical engineering, both from Cornell University.