Episode Transcript
Speaker 0 00:00:00 Machine vision is largely focused on the wrong problem. We need to think beyond the lens to move manufacturers from defect detection to defect prevention.
Speaker 2 00:00:17 Scott Everett is the co-founder and CEO of icon innovations, Inc. An independent software vendor out of Fredericton, new Brunswick, Canada that delivers enhanced AI enabled machine vision solutions software for industrial manufacturers. Scott is a PhD candidate at the university of new Brunswick where he graduated with a master's degree in mechanical engineering, driven by a passion for bringing technology to the factory for to solve complex and unseen process and quality problems for manufacturers. Scott and his team have designed a machine vision platform that's successfully adopted by tier one automotive suppliers across the globe. And that's scaling rapidly with partners like Intel and AWS for helping kick open the right doors. Scott, welcome to the podcast. I'm glad to have you.
Speaker 0 00:01:03 Yeah, thanks for inviting me on Jim. It's going to be great.
Speaker 2 00:01:06 Scott, how'd you get involved and started with icon innovations?
Speaker 0 00:01:11 Well, the journey began, um, about 10 years ago and it was, it was actually during, um, while I was working on my master's in mechanical engineering and I was studying, um, a lot of boat process control actually, and specifically in the plastic plastic space, plastic injection molding. And so my supervising professor, uh, was engaged with a lot of client projects and we were out in the industry. And so we kept kind of seeing the same problem over and over again, there was a ton of sensors, a lot of data on the factory floor, but none of it really being put to use. And this was long before AI and, you know, data science was a household term. So it was what we started to really recognize was the fact that to really put this data to use, we really needed to help provide better feedback about the quality of the product in real time.
Speaker 0 00:02:10 So a lot of quality control quality assurance, uh, happens offline. It's destructive testing and it's, you know, it's, uh, out of cycle. So the, the feedback can take hours if not days. And so we really started to wrestle with, okay, how do we just reduce that time and get the process engineers the right information. So they can actually optimize how they're running their machines. And that led us to the world of computer vision. And we, you know, went down this path of machine learning really early on to, you know, start to try and process that data. So it was, uh, a journey bit by bit. Eventually we decided to found Eigen and productize a lot of the solutions that we were developing from the research
Speaker 2 00:02:55 And machine. Vision's a hot space right now. Can you define the difference between AI and ML or machine learning for our, the, our audience?
Speaker 0 00:03:04 Yeah, for sure. You know, I think about it in terms of, okay, what is intelligence and that's really all about being able to acquire and apply knowledge to different problems. And so artificial intelligence is this really big topic or field and it's how do you have computers and algorithms have that same level of problem solving capacity? Uh, machine learning really is just a process of teaching computers, how to do tasks without having to be very explicit about, you know, programming every step of that task. So machine learning came about because we needed ways to create algorithms that had a lot more complexity to them without taking years and years of, of programming time. So machine learning is like one of those, one of these tools, that's been a big step in the progress towards artificial intelligence. I don't, you know, we're, we're a long ways off, uh, from true, you know, computers that can solve problems, self guided. Right. Um, so it's still something that's quite a target, but yeah, machine learning really has, uh, become very important when you start thinking about, uh, how do you actually process the volume of data that you would get from images and video and, and all that good stuff.
Speaker 2 00:04:25 Yeah. It's, we're still grappling with a lot of these things, like you say, to some vast amount of data that everyone's collecting and making sure that we do the right thing with it. And we'll talk more about that as we go through. So again, got involved, uh, early with precision molding and what's happened since then.
Speaker 0 00:04:40 A lot of our research, um, in terms of controls was we were using plastic injection molding as the, sort of the research process, because it's pretty complicated, right? Like you're going from raw material through to these very complicated parts, all within a sequence of stages on a single machine and it's happening, happening very rapidly. And in that journey of trying to figure out how to, uh, have better control and basically make better parts, um, at the end of the day, faster and cheaper, we really started to want to capture more information about the product itself, right? And so, like I mentioned that we started to turn to these, uh, different types of sensors, like a thermal imaging, for instance, it's a, it's a type of camera that basically gives you a picture of the heat signature. And so, you know, these new sources of information just gave us a whole new way to actually start to think about, okay, how do we control the process?
Speaker 0 00:05:43 And ideally what we've found as we've been working with our customers is, you know, it's, it's a great to be able to detect the defects that you make. So they don't get out into the wild, but really if you, if, if you don't want to ship a bad part, just don't make one. So it's that paradigm shift of, yes, you need to be able to detect quality issues, uh, in your facilities, but realistically you want to engage in the activities that are helping make you better on a daily basis. Is that true? Continuous improvement. And, you know, the journey for us has been okay, how do we actually start to leverage machine vision to that end? How do we actually start to use this type of technology in the continuous improvement rather than just catching the problem at the end of the line?
Speaker 2 00:06:31 I love that quote, but just don't make bad parts. I mean, it's easy,
Speaker 0 00:06:35 Right? Yeah. Right. Yeah. Exactly.
Speaker 2 00:06:37 And vision has kind of suffered through this robustness issue and, and is getting better and certainly you're out to solve this challenge, right?
Speaker 0 00:06:46 Yeah. I mean, when you look at the challenges with machine vision, what makes it such a powerful, uh, sensor is the fact that you can capture so much information in a non-contact type way. Um, but because it's in an open environment, there's a lot of things that that can change. And so, you know, getting that consistent measurement, because really when you're trying to get to that zero defect state and make things better, what you're really trying to measure are small changes over time and really understanding those trends. So when you look at what goes into all of the stages of being able to have a highly repeatable measurement system using machine vision, you know, it comes down to everything from the installation, you know, how you process and normalize the data. A big thing for us is really relating the images back to the product itself.
Speaker 0 00:07:45 So I think one of the really exciting spaces right now is 3d vision. And what we actually do is to take the flat images coming from a camera and map it back to the actual geometry of the product itself. So when you, when you're trying to understand the quality of your product, you're not just looking at an image and trying to interpret that you actually can start to visualize that in with respect to the actual product itself. So it opens up a lot of really interesting dynamics about having that, that really rich data that's consistent, um, and is something that you can really start to use to drive operational changes.
Speaker 2 00:08:23 And that's wonderful. What, um, I want to ask you a question was one of the product question, what is the open Vino toolkit when I was doing some research on you? Uh, it looks like it's a product from Intel.
Speaker 0 00:08:35 Yeah. So, I mean, in terms of our software suite, we've actually, uh, created probably one of the first true hybrid vision systems. That's a platform, right from the edge all the way up through to the cloud. And, you know, in some cases we're, we're actually capturing and storing video clips of every single part, you know, in the orders of thousands per day. So there's, you know, there's a lot of data and analysis that's that can happen on that. But then we also have to train these models and deploy them back down to devices in factory so that the decision-making is, is happening in real time. So we've been working with Intel and they have a new product called open Vino. And it's really quite interesting because it's helping, uh, create a pipeline between the model that you're training and the actual device that the model is running on. So it optimizes the model and for us, it makes it possible to deploy to any hardware without us having to worry about the performance on that hardware. It speeds up the model and makes that process of deploying models out, into factories, much more consistent and much more optimized. And I think the other interesting thing is that that optimization helps you run more complicated programs on smaller, cheaper devices, right? So keeping that capital expenditure down, uh, when you're trying to deploy a lot of systems within a factory line is, is important.
Speaker 2 00:10:09 So let's talk a little bit about data because I think we're all skirting this right now. Like where is it and, and how does say quality department access it?
Speaker 0 00:10:19 Yeah. You know, that was one of the, uh, interesting kind of evolutions on our journey. Like I said, we were probably one of the first actually create, uh, a platform that is not only in factory on the edge running models, but as actually capturing that data and transferring that data to the cloud. And it really starts to create a whole nother layer in terms of value, because what actually ends up happening is you're creating this digital record for every single part. Quality control is largely been based on sampling. So you pull a couple of parts off the line, you test them, if they're good, then you hope that that represents, uh, all the rest of your parts that you're not testing, but when you're actually starting to capture this digital data on every single part, uh, now you've got this record of the changes and it really opens up interesting opportunities for analysis to truly understand the trends and what's changing over time.
Speaker 0 00:11:21 So while the decision-making has to happen at the edge, so you're reacting to real-time changes in your process. That data set now starts to become a really powerful resource for the organization because you can, uh, you know, you can really, number one, you can reduce the amount of risk in terms of the quality assurance process. We've had multiple occasions where, you know, something came back or there was an issue, and we're able to go search through those images and find exactly what happened and, and the process to say, okay, was this an isolated case or did something change that reduction in risk is just huge value in terms of the overall return on investment for these types of systems,
Speaker 2 00:12:05 It must surprise some of your clients to say, we're going to actually take, take data on every single part that you produce like that that's kind of a game changer.
Speaker 0 00:12:15 Yeah. And honestly, it came with it, it took time, right? Uh, one of the things that we really started realize deploying machine learning early on was, you know, having your, the right data in your, in your data set in the way that your training needs algorithms, the management, the change management becomes really important. And so your traceability and your, your record of, of the data that you're using to adapt these systems over time, it's what enables a scalable, robust, repeatable solution. And it took some time, I think, for our clients to understand the value of that. But what it actually means in the long run is you're able to automate and create this consistency because, you know, you, you have, um, a system that's continually updating and adapting, and then you've got the proof, right. Um, in terms of the digital part record.
Speaker 2 00:13:11 And so I'm kind of curious about who calls you like when, when, uh, when a prospect is interested to say, Hey, we got to get these guys ranging and into our factory. Uh, is it the industry 4.0 person? Is it plant manager? The quality person may be the CEO.
Speaker 0 00:13:25 There's a lot of stakeholders in, in the factory. And we ended up kind of touching a number of them. Process control, quality control are usually early in the conversation when you're looking at new applications. A lot of times what we found is we tackled really challenging machine vision problems that people couldn't solve previously. And like I said, it's not just the simple defect detection, you know, that's, that's a big part of it, but it's also, how do we measure the process? And what we've found over time is, you know, you start out with the machine vision system that's in factory and you're kind of achieving goals of improved quality control. But once the organization starts to see and understand the value of the data and the data set and what that means in terms of being able to prove to their client, that every single part is certified and there's this, you know, valuable resource there, then you really start to have interesting conversations with, um, you know, the advanced engineering groups and throughout the organization, you know, the, on more of the corporate side, right?
Speaker 0 00:14:37 Because what it starts to create is the potential for true repeatability. And I think that's one of the things that machine vision has suffered from. There's a lot of one-off solutions that are disconnected. You know, somebody created one over here and over there. And so where we've seen customers really take full advantage of the platform capability is you start to create a solution that's consistent across all of your factories. And then you can actually learn from the differences in the variations of different machines in those different locations. And so that really starts to create new ways that there's, that you can unlock value.
Speaker 2 00:15:18 I was going to ask you, what are the benefits and savings, but you've kind of gone over overall some of the benefits, but from your client's perspective, what are some of their savings?
Speaker 0 00:15:28 Yeah, I mean, what's really interesting about return on investment is tangible versus intangible, right? And the things that you look to in terms of the immediate return on investment largely is the reduction of, you know, the defects that get out to customers. Some of our customers have had challenges where it's kind of strained the relationship with their customer. Um, and we were able to bring defect rates down like 20%, you know, sometimes higher, if it's a real problem, like there's certain processes that are obviously making have more challenges than others. Um, and so that's, you know, hundreds of thousands of dollars, millions of dollars a year per line in some of these factories. So it's, I mean, the cost of quality is estimated somewhere between 20 and 40%, right? Depending on who you're talking to, but the intangible things are okay, what starts to happen when you're preventing recalls?
Speaker 0 00:16:30 Like it's, it's that risk reduction that's more difficult to calculate, but when you're consistently creating and using the data to its full capacity, you're drastically reducing your risk on a daily basis. And that makes you more competitive. And it's really like, you know, it's like choosing to embrace automation and robotics. What's the exponential value of making a decision to have more repeatable, you know, not firefighting all the time, actually ahead of the issues and, and engaged in continuous improvement. Those things really stack up over time, right? So it's, we often measure the ROI based upon the hard cost of quality. Uh, a big one for us lately has been actually reducing the amount of destructive testing. So because we're capturing digital records of every single part, you know, we're able to say, look, you don't need to actually test these parts. Here are the ones that might've changed a little bit, and those are the ones that you, you wanna, uh, distract because that, that process takes a lot of time and, you know, you're actually scrapping product. So yeah, it's, it's one of those things that I think will continue to evolve. I think organizations, as they start to see the, the benefit of a lower risk profile in terms of value know, they'll, it it'll change the conversation about how you actually look at ROI.
Speaker 2 00:17:53 Well, and I kind of imagine the president of the company that's manufacturing as part or whatever, saying we have the best quality system. We inspect every single part and it just changes the conversation, I think. So it makes it almost easier to, to sell things to their customer or their customer's customer.
Speaker 0 00:18:09 Yeah. And I think that's a big part of how manufacturers are going to continue to be competitive, right. Is, is being able to show that they have that best in class. And, and that's, it's not just the digital twin of your factory. It's actually the digital twin of all of your product. Like how do you explain and certify that every single thing in your product is built a spec, you know, and a lot of that in traditional, uh, manufacturing is you try and create a very capable process. You do your sampling and you hope that everything stays consistent, but we, we know that's just not the case, right? So, uh, you know, that the world of digitizing manufacturing really is all about changing the way that we, we think about quality control. And we think about, you know, these different processes,
Speaker 2 00:18:54 It was, is return on investment is ROI gone.
Speaker 0 00:18:58 I would say, no. I think in terms of the traditional mindset of manufacturing, it's still really important to be able to prove that ROI. But I will say that I think when organizations start to really embrace digital transformation, they start to realize that exponential benefit of investing in the right technology. And the sooner you do that, the more that that opens up. And so it does change the perception of how am I going to measure this ROI, but it is one of those things, like at a certain point, I think the digitization of manufacturing is, is going to be so necessary that you're going to have to kind of suspend the immediate calculation of ROI and really start to factor in those longterm things. Right.
Speaker 2 00:19:45 I totally agree. So how do you sell your product? Do you use integrators or distributors?
Speaker 0 00:19:50 Yeah. So more and more, um, you know, in the early years, a lot of our opportunities were through direct sales. We actually, I think one of the, the interesting channels is the actual hardware providers of the cameras. You know, when you can create really compelling solutions and new solutions using their technology, you know, FLIR cameras is a, uh, you know, they're a partner of ours. We've been working with for a very long time. And, and that opened up the doors to start to investigate different applications. Um, but what we're finding now really is there's a much more focused ecosystem on digital transformation. That's involving a lot of different players from the integrators. You've got the traditional systems integrators, but now you've also got the whole world of it. And so we're seeing players like Intel and Amazon and the cloud providers that are coming to the table now and saying, okay, a solution, a true digital, um, stack of technology is inclusive of in factory in cloud technology. And so there's a lot of channels now that are starting to collaborate and work together to bring much more scalable solutions that are easier to adopt. I think that's always been the challenge with the fragmented landscape of everything that has to come together to create, uh, an AI solution in manufacturing. It really slows things down. So if we can streamline everything involved in, in the involvement of the different players, then it makes it much easier for the industry to adopt and envision how it can scale throughout their organization. Yes.
Speaker 2 00:21:34 It seems like all the planets are aligning, right? We've got our processes are really good and we've got good data hygiene. And do we still, I think haven't nailed security down yet, but it looks like we're kind of all lining up in a lot of the things that we have to do to make your solution work. Well,
Speaker 0 00:21:51 I think, I think the biggest thing that is really going to sort of tip the scales. You gotta put yourself in the mindset of the production teams, right? They're the ones that are responsible of getting, getting product out the door. And so how do you create solutions? The benefit of AI and machine learning is they can be very dynamic, a lot more powerful in terms of interpreting data, but they can also be fussy and you have to really create that trust so that, you know, production teams, if, if a technology started failing too many parts in the run of a day, you know, you just say, okay, I gotta shut that off. Cause I had to get parts out the door. Um, and so the, the management of this technology in that real live environment and just creating that, that trust and, and creating systems that are able to react to the needs of, of production is really important, right? Like I think that's, that's one of the things that we can't lose sight of. We can get really excited about the power and the capability of AI, but you're really got to keep in mind what the end goal is.
Speaker 2 00:23:02 Scott, do you feel that part of your responsibility bringing this technology to your prospects and your customers is that you also have to train them how to use this, or do you feel that there's a, the training is really, is quite adequate right now?
Speaker 0 00:23:15 No, and I think that's another thing that's, that's been a challenge for the industry is it's abstract. You know, if you create a simple rule of one plus one equals two, it's easy to understand, but all of a sudden you put these artificial intelligence algorithms in the mix of your automation. Um, and how do you know how that thing is going to react under all these different conditions? Right. And so understanding the process of training and, you know, one of the things that has been very interesting in terms of observing, um, quality and the way quality assurance happens in factories, when you have human, uh, people that are inspecting parts and they're, it's up to them, a certain level of interpretation about what's good versus bad quality when you start labeling data and asking them, okay, is this a good part, or is this a bad part?
Speaker 0 00:24:10 You're going to get so many different answers. And so there's this whole process of helping the customers, not even just understand AI, but understand their data so that it can be effective in teaching the AI, the right things. And so it is a journey we've seen a lot more awareness. I mean, when I started talking about machine learning eight, 10 years ago, um, a lot of blank stares, but I think we're seeing that, you know, the benefits it's here to stay and it's something that is going to be a long-term competitive advantage for, for manufacturers that learn how to adopt it. Well, one of the things I think is also really important to note, especially with what we've experienced in terms of machine vision is we, we say, think beyond the lens. So a lot of times machine vision is, is very much in isolation of, you know, inspecting part of a process or a product. But one of the things that we've seen become very, very important is integrating it with all of the surrounding data in the factory as well. So that you're contextualizing what the AI and the decisions that it's making based upon, you know, what's the sensor data upstream and, and all that. Right. And so connecting the dots on all of that data is really one of those things that's essential to maximizing the performance and capability of the system. Yeah. Machine, vision system in the factory. So,
Speaker 2 00:25:41 So what do you see as some of the future of your industry? You see, it's more data, more cameras, more integrations.
Speaker 0 00:25:48 Yeah. I mean, I really do believe that the value of the digital twin, if you will, of the product is it's going to become essential. I think we're going to continue to see an expectation that if you don't have the data to support why your product is good, you know, that's going to create a lot of risk and it's in. So, you know, we're, we're seeing it in, in the automotive space, which is really fascinating. You've got E V batteries, right? And now with autonomous vehicles, the headlights and taillights, they're packing so much technology into these components now they're, they're very, very complicated. And if you don't have a good understanding of everything, that's, that's gone into that, that part in product, it can really come back to bite you when there's, um, failures out in the real world. So I think it is more, it's more data, it's more cameras, but it's, it's really about the integration of all those technologies into this continuous improvement cycles so that, you know, you're certifying every single detail about your product. You're not just checking a few at the end and, and hoping it's going to be good enough.
Speaker 2 00:26:59 I see these tier two manufacturers who are making products and saying, and this being a big checkbox, like tell me about your quality system. And all of a sudden they say, well, let me tell you what my quality system do you think that's going, that's going to be continuing to be a thing.
Speaker 0 00:27:14 Yeah. And honestly, we've started to see it in terms of, uh, from the OEM level. Like there no new best practices in terms of quality assurance. I think even the OEMs are actually starting to figure out how to understand the implications of AI and data. And then what that means through the supply chain, right? Like how do you create more consistency, um, and reduce risk throughout their supply chain. So as those independent solutions become standardized, you really start to see that effect. And it just becomes, well, of course we should be doing this, right. It's not a, it's not a question anymore. So that's really been largely in the last year kind of year or 18 months that we've seen that type of progression now.
Speaker 2 00:28:04 No, I, I think that's a really good point. And I think it's one of these things that will differentiate a manufacturer from one to the other is one has data and the other one just does not.
Speaker 0 00:28:13 Yeah, exactly.
Speaker 2 00:28:16 So thank you for very much for coming out of the podcast. I did want to ask you personal cards. What got you into this? Automation saw a business.
Speaker 0 00:28:23 Yeah. A good question. I, I was fortunate to grow up on a eight generation, uh, family potato farm when I was a kid. And my, my father actually, uh, had a little side business where he was building homegrown equipment for manufacturers and or for farmers. So we had a little manufacturing, um, facility. I think I learned to weld when I was like 10 years old or whatnot. Uh, and so it was just always, I was always fascinated with equipment and, you know, a farm, especially a potato farm, it's a mini manufacturing line. You get conveyor belts and all of these different processes to sort and whatnot, right? So that always fascinated me and an engineering just became an obvious calling very early on in my life.
Speaker 2 00:29:15 Great story. Thank you so much. Our sponsor for this episode is Earhart automation systems. Eric builds and commissions turnkey solutions for their worldwide clients. With over 80 years of precision manufacturing, they understand the complex world of robotics, automated manufacturing and project management, delivering world-class custom automation on time and on budget contact. One of their sales engineers to see what Earhart can build for you. And Earhart is spelled E H R H a R D T. I'd like to thank and acknowledge a three, the association for automation for advancing automation. They're the leading trade association in the world for robotics, vision and imaging motion control and motors and artificial industrial intelligence technologies visit automate.org to learn more. And I'd like to thank our partner painted robot painted robot builds and integrates digital solutions. There's they are a web development firm that offers SEO and digital social marketing and setup and connect CRM and other ERP tools to unify marketing sales and operations. And there are painted robot.com. And if you'd like to get in touch with us at the robot industry podcast, you can find me Jim Beretta on LinkedIn. We'll see you next time. Thanks for listening. Be safe out there. Today's podcast was produced by customer attraction, industrial marketing, and I'd like to thank my nephew, Chris gray for the music, Chris Coleman for audio, my partner, Janet, and our partners eighty-three painted robot and our sponsor Earhart automation systems.
