The Manufacturing Executive
The Manufacturing Executive

Episode 125 · 1 month ago

The Future of Automated Vision Inspection

ABOUT THIS EPISODE

Technology has brought us to a place where mobile phones have better cameras than the most advanced manufacturing lines had ten years ago. Vision inspection technology has come so far that the barriers to entry in terms of investment and technological feasibility have made vision automation accessible to all manufacturers.  

Austin Appel is the Co-Founder of Overview, an AI deep-learning vision platform for manufacturing. At Overview, Austin is responsible for leading product development and operations. Previously, Austin spent four years at Tesla in battery manufacturing and research and development roles. In this episode, Austin talks about the advancements in vision inspection technology.  

Join us as we discuss:

  • The underlying problems within manufacturing facilities
  • How is technology helping to alleviate labor problems
  • What’s different now vs. 20 years ago in vision technology for manufacturing
  • How Overview fits into the overall picture of vision technology

They just want to know when they get a call from there, their t ones supply, when they get a call from the O E M. Saying how you sent us a bad batch, they want to know if he if that's even true? Is it a bad batch? Or did they get screwed up in shipping or did you screw it up and be how what's the extent? Is it a whole bad batch? You have to shift the whole thing back for a sort or can we go through our pictures and see, oh, you actually only have three bad parts in there. Here's your serial numbers. Welcome to the Manufacturing Executive Podcast, where we explore the strategies and experiences that are driving mid size manufacturers forward. Here you'll discover new insights from passionate manufacturing leaders who have compelling stories to share about their successes and struggles. And you'll learn from B two B sales and marketing experts about how to apply actionable business development strategies inside your business. Let's get into the show. Welcome to another episode of the Manufacturing Executive Podcast. I'm Joe Sullivan, your host and the co founder of the industrial marketing agency Guerilla Sety six, where we help B two B manufacturers grow through revenue focused marketing programs. Just about all of you listening right now could reach into your pocket at this exact moment and pull out a camera that's far more advanced both from a hardware and software perspective, then the cameras that the most advanced manufacturing lines had running on them just a decade or two ago. For QA inspections, the advancements and vision inspection technology, from camera quality to image recognition to machine learning, have come so far that we're now in a place where the barriers to entry from the standpoints of both investment and technological feasibility have made vision automation so accessible. My guest today is at the forefront of this movement, and he's here to get you up to speed. So let me enter do some Austin Appel is the head of product and a co founder at Overview, where he is responsible for leading product development and operations. With over eight years of experience in manufacturing and automation engineering, Austin works with customers to ensure that manufacturing operations can quickly adopt and use artificial intelligence based computer vision systems. Before Overview, Austin spent four years at Tesla in various capacities, including battery manufacturing and research and development. He led the design for Manufacturability DFM and automation efforts for the Model three battery pack production at Tesla's first Giga factory. Earlier at Tesla, he expanded the production of Models and X battery packs through equipment design and implementation. Austin holds a dual Bachelor of Science degree in manufacturing engineering mechanical engineering from Northwestern University. Austin, Welcome to the show. Thanks glad to be here. Well, awesome. I think you told me that of factory jobs in the US fall into the inspection category. Obviously, product quality is a big driver of cost, so let's just dive right into it. And have you talked a little bit about this. Yeah, that number is obviously pretty industry dependent, and the job can take a lot of different forms, but at the end of the day, one of the last things to be you know, kind of mechanized in in production in the US is inspecting things. And I think there's a lot of reasons for that. The biggest one is that you can have a robot or some type of automated system that does a repetitive task, but a lot of times quality comes in forms that you may have never seen before. It kind of has that human touch where it's like, oh, this is bizarre looking. It's not necessarily a defect on my list of three defects, but like, I should probably call someone over here to look at this weird thing coming out of my line. And I think it's also kind of a sypnomous...

...other stuff that you see in the US manufacturing where some folks have been slow to adopt automation, where things like outsourcing and and kind of near shoring and moving stuff to Mexico and China has kind of lead us to to not really think about manufacturing as an automation versus labor perspective and more as I should I do it so are cheaper labor versus do it here because it's expensive or as military customers or whatever reason. Um So, yeah, a lot of people who work in plants, and everyone who's gonna be listening to this knows this, are still doing inspection jobs by hand, and it definitely is a barrier to lower and cost and making manufacturing here more effective. Awesome, And we're gonna get kind of into a little later on what you guys are doing at overview and kind of how you fit into this conversation. But it's a good kind of primer to this this we're what we're gonna be talking about here. You've told me that sometimes you'll walk into a factory and you'll see equipment that's that the sixty years old and maybe on the surf us the economics makes sense to keep it going, but under like they're underlying problems I think that you tend to see in these same facilities as wonder if you could speak to that a little bit. Yeah, there definitely is in certain parts of of US manufacturing in particular, a tendency to basically run equipment and processes forever until they truly are unsupportable or they break like they're they're just that thing is worn out to the point where it's no longer serviceable. And this makes sense if you own the business or you bought the business and you're kind of using it to pay down dead or using it to to try to have just steady cash flow for the family owners, whatever it is. But it hasn't left us in an interesting spot where we're not necessarily automating as much as we can and not doing those capital investments in a country that has some of the highest labor costs in the world. So yeah, a lot of what we see with our customers, at least is people trying to squeeze as much as they can out of things that they've already paid for or their previous owners have paid for that are essentially would be very expensive to replace. But that does lead to higher opics at the end of the day, you know, keeping these things going is just pushing that investment for the downline, squeezing out something, but also is one of the reasons that things made here are so expensive. I think that you know, the costs that go in are a lot of labor, they are a lot of maintenance, There are a lot of skill because they don't want to do that capital investment, and so that that's influenced how we operate. But also I think add some color to a lot of what gets produced overseas versus what gets produced here. I think that a lot of things, we don't have a lot of low cost goods coming out of the States unless it's for a very specific reason. You know, paper and packaging is here because it's so bulky that mostly here. Shipping it is tough, right, and that's a reason to do it here. And you'll see an industry's like that. There actually is a decent amount of capital investment, maybe not in like paper conversion, but in some of the downstream stuff in the printing and the stuff that makes them more competitive, but in things like injection molding and you know, metal fabrication. You'll see we generally produced really expensive things here, and I think a lot of that comes from people trying to run these machines forever and not make those new capital investments. And I think the kind of the other side of that, at least that I see. You know, a lot of people I talked to around this labor issue we're obviously having in so many parts of our country, especially in the manufacturing sector, is you know, the perception of manufacturing is being outdated in our in our country, right you and you walk into the facilities and you see these this old machinery and you're a you know, eighteen year old kid who potentially be a you know, the future of this workforce. Like it's kind of a...

...turn off, I imagine to not see technology being invested in and infused into these businesses and it's just sort of proliferates this perception of like dirty, dark, dangerous manufacturing in the US. And so, you know, I'm curious to just kind of hear your take on that where technology fits into it, because I know you recently told me I think one of when your own customers was paying sixty an hour or something for a QA inspector but still having turned problems. So you know, how how does all this fit together? Where's technology fit into the mix in terms of alleviating this labor problem. It's a really good question, and I think that also really good observations. We you do see a kind of a whole spectrum of plants. I think something that we did really well Tesla was was investing, not invest a lot, but not allowed this particular thing, and to trying to change the perception a little bit. We had, you know, all our factories were really nicely painted. We had a POxy flooring, not as easy to maintain as polished concrete always, but it looked great. And you know, you walked in there, there were plants, there was colors, that was really nice lighting and these weren't huge investments. In fact, that led lighting was probably ultimately a cost savings in the long run. But even as me coming straight out of college as an engineering degree could have gone a lot of different places, could have gone until you know, banking, but going into that plant and knowing that we were intending to make that plant a pretty friendly place to work was was definitely important and I think it paid off many times. That isn't a good example of a company that is doing automotive A lot of it is made here, but it's doing cost competitive battery manufacturing, even getting into some of the upstream stuff that historically has has really come from China. Like there, they're doing a really good job competing, and I think it is part of it is bringing in talent, bringing folks who are willing to go into what is historically in a nasty environment, and you know, investing in making that environment a little bit less nasty. That stuff goes a long way. I think. You know, you'll see a lot of manufacturers get the most results and biggest prompt set of things like safety upgrades. I think that environmental upgrades fall in the same category where people are just you get people who want to be there, you get people enjoying it more. You can attract talent who maybe wouldn't have considered manufacturing, but you know, it doesn't you know. And and it was kind of going back to the test the factory was. It was one thing when the factory was here in near San Francisco and Fremont is on an hour, I could live in a big city. But we were also effective at getting people to go to sparks. I think that, you know, people don't think about sparks as a as a destination for wanting to go be an engineer fresh out of college. But when there's enough you know, time energy spent making something really cool, making it appealing, you can get people to do it. And and the second point about you know, the churn, I think that that comes automation is super super key to that. I think that that some jobs, no matter how you know, even if you're sitting in a massage chair, are terrible. I think quality inspection and high volume is one of those. And those jobs, you know, people always have the automation versus note automation. You know, are we killing jobs? I at this point, having been around this stuff for a while and completely on the automation creates opportunity train, I think it's it's not it's no longer even a controversial opinion. It's just clear. I think that, like, you know, the example of we have a customer who pays over an hour for their quality inspectors in the Pacific Northwest. But it's a miserable job. It's a high volume kind of small thing. It's being inspect continuous production, and they still see training because it's just there's there's obviously some amount of money, but within reason, there's nothing that you could do to make someone enjoy that job. And I really can't imagine why that job needs to exist when they're you know, you could have so much more productivity out of the same team if you did automate. And that's part of the reasoning of our customers, right is they see that too. Just historically there hasn't always been tech to do it. But yeah, there's there's there's an issue.

There's a kind of a flywheel problem, and I think, you know, there's people are doing a better job lately. Automotives always been pretty solid about it. The other industries are kind of picking up the value of you know, if you automate and you bring in smart people, you can actually do some of this stuff costs effectively here, and so it's cool to see. I think there certainly are other factors now, I mean, I was reading an article this morning about how natural gas prices are pushing a bunch of manufacturing from Europe to the US, not even from Asia. But folks who you know, use a lot of gas in their processes or use a lot of hydrogen are investing really heavily in like Texas instead of instead of Europe. And so there's always external factors. But there is a way to do it here cost effectively, and it includes you know, creating good environment and automating. Okay, let's take a quick break here. I want to let a couple of our strategists at Grill the seventies six tell you about something pretty cool that we're doing right now for marketing folks in the manufacturing sector. Peyton and Mary take it away. Yes, So, I'm Peyton Warrant and I'm Mary Kio. Twice a month we host a live event called Industrial Marketing Live. Right now, we have a group of fifty plus industrial marketers from a variety of manufacturing organizations. We meet up digitally to learn, ask questions, network and get smarter. Every session has a designated topic, and one of our team members at Guerrilla seventy six opens up by teaching for the first half hour or so. Topics have included how to get better at a manufacturing webinar, getting started with paid social on LinkedIn, how to optimize your website for conversions, creating amazing video content, and so much more. After we break it down, we open it up to Q and A so we can help you apply all of this in your own businesses. This is pure value, no cost, no strings attached, no product or service pitches, just so unadulterated learning experience. Oh and on top of these live sessions, we've also opened up a Slack channel where our attendees bounce ideas off each other and learned together all week long between sessions. We're building a true community of manufacturing marketing professionals here. So if you or someone at your company has the word marketing in his or her job title, please consider telling them about it. They can visit Industrial Marketing live dot com to register. We love to see you there. I think what you're hitting on there, especially as it relates to QA inspectors, it's a good lead into talking a little bit about vision technology and manufacturing, which is, I know your world. You were telling me recently that a lot of the earliest vision technology for manufacturing production lines was created in the early two thousand's, twenty years ago already, and we were carrying flip phones around in our pockets, right, And now the average person has vision technology in our own pockets far more advanced than than that was. Probably, So what talk about what's different now versus twenty years ago for maybe you know, midsize manufacturing leaders who are listening and maybe don't really understand the landscape that well, yeah, and it's actually even earlier. I mean there were there were there were early inspection and code reading and go no go stuff happening with cameras in the eighties like this, I don't even know, maybe an other than that, but this it's not a new concept to automate visual inspection with cameras. What is new is there's been a spectacular amount of investment on there's two sides. There's a software side on the hardware side and then you know services is always going to be services, but spectacular about investment on both. On the On the hardware side, camera tech has gone. You know, not too long ago, people were carrying around DSLRs with multi thousand dollars two small, incredibly high performance, incredibly low cost sensors that you know, it's not always exactly the same tech that goes into the phone as it goes into the factory camera, the industrial camera, but a lot of it's carried over,...

...a lot of the R and D trickles over, and so you have these sensors now that are phenomenal and they're incredibly cheap. You know, these these you can have something that can do you know, insane resolution, insane speeds, great sensitivity for in the hundreds of dollars. And so when you're talking about capital investment, even if you have a thousand production lines, if you're in the hundreds of dollars, it's not it's it's it's affordable. Now computers have come in the same trajectory, you know, GPUs for a variety of reasons, which are really good for vision tasks have come down. A lot of computer gaming, cryptocurrency and a I have all driven a lot of R and D money into GPUs, which are a cool way to do some of these these vision calculations. And so those two together, like to have a system that could process your images, get good images, process them in high speed. Used to cost a hundred thousand dollars in just hardware, and now it's like it's like that. So that's that's really cool. I've been really awesome for folks who have looked at automation in the past decided it's not for me. I can't make a quarter million dollar investment in the camera. The software side is where we kind of live, and that is even more exciting. Artificial intelligence has you know, everyone is kind of talking about it, but one of the things that it's really really applied for and people have gone super deep on and put a lot behind this image recognition. It's one of the most basic but also one of the most well studied fields in artificial intelligence, and it overlaps really well with quality inspection. Those systems from twenty years ago worked on very different principles and systems that we work on where you basically treat you take a picture of one of those old systems, you treat the grid of pixels as as an array with the numbers of values. It would always almost almost always be like monochrome pictures, and then you do math. You look for edges by looking for slopes in those numbers, you look for features, you look for blobs, and you try to write these like really complex rules based off all that math to try to you know, tell a computer, Okay, if you see a bunch of of white pixels here next to a bunch of black pixels next to uh, this gradient, that's probably a defect. And that takes a lot of really specialized engineering, super finicky, requires like incredibly precise lighting, because those are just values that can move a lot based off of intensity alone, and just drove a lot of like complexity and difficulty into the system. Machine learning is a totally different way of handling that data, that image, where you basically train a computer and a specialized algorithm off of that can hold you know, thousands or millions of parameters. You know that that gradient, that blob whatever. You don't even know what it's doing, but you just feeded examples. You feed a ton of examples and you train it in a way where it learns to recognize things from that image that maybe you wouldn't be able to code, or even if you could code, would take thousands of lines, and it's based off data, so it's much better at a someone who doesn't necessarily know intuitially how to make those rules can still say, yeah, that one's defective, it's got a scratch. Now that one's fine. And you can do that enough times to feed the data into the algorithm and get a good system out of it. And on the flip side, it can handle variability a lot better. So you used to have to put your inspection equipment in like this perfect box and have special LED lighting that had a sensor on the LED lighting so you'd know if it was domming over time to correct for it, like insane stuff like that. Whereas you know, as you know, if you have an iPhone you pointed at someone's face, doesn't matter if you're in a basement, you're outside, you're hiking, whatever, it's gonna do. That little face recognition box right, because it's been fed so many faces, it can handle all that environmental variability. No one sat there and coded up like what an eye looks like. They just fed it enough. We're generalized, and so for things that have a lot of variability in the product, things like food, organic matter a lot of ability to process. If you have skylights, if you have an open factory, um you can build a system that is able to handle that without having to go crazy on the hardware and crazy on the engineering. It's just about getting a little bit more data. So it's a totally new paradigm and it has unlocked things that a are not even possible...

...before because they were too complex, like grading meat or looking at metallic services that are super shiny for there's reasons why that's really hard traditionally, But on the flip side, it's also made it a lot cheaper to do simple applications because you can go in. You don't have to have someone get paid two hundred an hour to code up this thing and then have to call them again because one of your life's burnt out. So it's it's it's exciting what's happened in the last ten years. A lot of the investment has been driven by other industries, but it's really trickled into inspection in a way that's that's amazing awesome. Talk about overview a little bit. Tell us where you and what you guys are doing fits into this picture. Yeah, so what we do it overviews. We provide turn key AI solutions, so our customers come to us with a problem. Hey, you know, we either spending a ton on this one inspection, or we're not catching these defects, or customers are complaining because this defect keeps escaping, and we kind of soup to nuts build them a system using that that lower cost new hardware, our own in house algorithms, and some services that we provide around training and support, and so we basically are like like a shop where you call us and and we'll automate your inspection for you using all this new tech. This came from you know, when I was doing lines building production lines at my last company. We spent a fortunate vision systems and they were this traditional type and the amount of effort and that went into trying to get them programmed, coupled with the level of re liability we got afterwards, it was atrocious. And we realized that a the tech that was in these things was was was old, and it was crazy because they're sitting next to the self driving car team and they had the most you know, they had cameras, these little one megafixal sensors and they were driving a vehicle with this thing. We had these ten thousand dollar cameras that couldn't tell us if a screw was installed correctly because we hadn't programming quite right. So that was an obvious gap. And then, you know, I think what we realized is this is not the kind of thing that needs to be You don't necessar sarily want to have to do this if you're running a factory, it's like the last thing on your mind. And so we decided that someone should have a company that use the latest tech and also provided this stuff ready to use and didn't push the programming and the support and then onto the end customer. So yeah, we started the company in late been working on it for four years now and it's going super well. We have customers across aerospace, textile, medical, automotive, food packaging. It's kind of it's kind of taken off all kinds of directions, but yeah, it really is like we think of ourselves as a one stop shop where if you have a problem, you want to automate your vision task, you come to us. Whether it's an old system that doesn't work, or a place where you have no inspection or a place where you currently have manual inspection. We kind of work with all that very cool. What do you see Austin as you look out five years as it relates, you know, technology advances and you think about sort of what QC and inspection technology is going to look like. Yeah, I see a really interesting opportunity in terms of lowering the barrier entry. And we've already started to do this. You know, systems used to cost hundreds of thousands now with us where you know you're intens but I would love to be in a world in five years where it's like the hardware to get started is like nothing, like let's you know, five bucks a thousand bucks, and it's possibly it's heading in that direction. You're seeing you know, AI computing power get exponentially cheaper. You're seeing camera sensors get you know, exponential exponential but a lot better um and you know, you get two year old smartphone tech for nothing now, and so I can I would love to be in a place where the customer getting started is like so easy and so cheap. And then obviously you know they're still work around getting it running and training it and we can work on that. But I think that's that's that's exciting stuff, and we're seeing that happen like already in the market. We don't we don't make any very work with partners, and we're...

...watching our partners performance just go nuts. Where it's like getting so cheap and so good. That's super exciting. I just love to see it everywhere, right like there's right now you have to invest and and we you have to have a specific high value problem where it's worth the time, money and effort to automate it. But cameras should be ubiquitous in factories. You should be able to, you know, as easily as you can, you know, go through logs on on a piece of equipment. You should be able to go through visual records of what happened in your plan. You know, this part came out all screwed up. I want to go and I want to see what it looked like at the four steps previous. Right right now, no one's really offering that. We were kind of one of the first people to offer that, and we're trying to really make that accessible as well. Where you know, you kind of have cameras, is this ubiquitous thing around your plant and you have all that image data and you can go back and you can see for debugging purposes, for customer complete purposes. You know, some of our customers don't have any AI running at all, or they do but they don't use it. They just want to know, you know, when they get a call from there, their t ones supply, when they get a call from the O E M. Saying hey, you said it's a bad batch, they want to know if a if that's even true. Is it a bad batch or did they get screwed up in shipping or did you screw it up and be how what's the extent? Is it a whole bad batch? You have to shift the whole thing back for a sort or can we go through our pictures and see, oh, you actually only have three bad parts in there. Here's your serial numbers. Sort done right, saved. So that's another future that we're pushing towards. And that's just kind of the like cheap ubiquitous camera and like a lot of image information coming off your plant. It must be a pretty exciting space to be in right now. It's fun, yeah, I mean all all factory automation is fun. I think vision is especially fun. Well, Austin, great conversation today. I really appreciate you taking the time to do this. Can you tell our audience how they can get in touch with you and how they can learn more about what you're doing at Overview. Yeah, you can reach out to me directly, Austin at Overview dot AI be happy to dig in. You know, we work with all different kinds of customers, big, small, everything in between. We have a lab where we process samples. So you know, the typical way that we go about it is that I have this problem. It's cool, all right, send me, send me a bad one. Let's see if we can actually detect this thing. And so we have a team out here in San Francisco that goes through and we'll actually handle you know, testing the parts and validating and all that stuff. And also, you know, more than happy for anyone who has you know, not specific applications, but just general questions about vision automation or factory automation in general. Always happy to lend assistance and and help people who are looking to improve, and always happy to take factory tours. That's like my favorite thing in the world. Beautiful, Well, Austin, thanks again for doing this, man. I really admire what you guys are doing and congratulate you on what you've accomplished so far, it's can be fun to watch where you take things. Thank you, Joe. As for the rest of you, I hope to catch you on the next episode of The Manufacturing Executive. You've been listening to The Manufacturing Executive podcast to ensure that you never missed an episode, Subscribe to the show in your favorite podcast player. If you'd like to learn more about industrial marketing and sales strategy, you'll find an ever expanding collection of articles, videos, guides, and tools specifically for B two B manufacturers at Guerrilla seventy si dot com, slash learn. Thank you so much for listening. Until next time,.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (130)