Visions Podcast: Cognex Talks AI on the Factory Floor, Part 2

In this second and final installment of a two part podcast, Visions: A Machine Vision and Automation Solutions Podcast, Sharon Spielman, VSD's head of content interviews Cognex CEO Matt Moschner about a new survey and the real-world impact of AI-driven machine vision. They discuss transparency and auditability for regulated industries, lowering the barrier to adoption through easier-to-deploy AI tools, and advances in embedded hardware that enable higher throughput and better defect detection.

Related: Cognex Talks AI on the Factory Floor

Transcript

Well, hello and welcome to Visions, A Machine Vision and Automation Solutions Podcast. I'm your host, Jim Tatum, senior editor of Vision Systems design and Visions is an Endeavor Business Media production from your friends at Vision Systems Design. Here you'll find the latest on everything from end user machine vision solutions to trends, developments, and perspectives on all things machine vision and imaging. Whether you've been working in the industry for a while or you're just starting to take a closer look at it, this podcast is designed to grow your knowledge and bring greater focus to your understanding of the imaging and machine vision industry. And now on to our show. 

AI driven machine vision isn't a concept of tomorrow. It's changing quality and efficiency on factory floors today. Cognex Corporation recently surveyed more than five hundred manufacturers about how artificial intelligence is transforming machine vision systems in industrial production. The report, which was just released today, reveals shifting priorities, rapid adoption and evolving expectations across industries. I'm Sharon Spielman with Vision Systems Design and joining me to discuss the real-world impacts, challenges, and future directions for AI driven machine vision is Matt Moessner, the CEO of Cognex. Welcome, Matt. 

Hi, Sharon. Thanks for having me. 

So for those sectors that have the stringent safety or the quality standards, so how do these AI systems help to meet or exceed those requirements specifically? 

Yeah. I mean, I think on one hand, you have to make sure that the AI tool and technology is transparent in terms of what it's doing, right? One of the early critiques years ago of the technology was it was too much like a black box, right? Moving from technology that was really about setting a variety of parameters that you knew exactly the parameters that you set to one, that you were trained by example. And then it was kind of creating a model in the background. I think that that made some customers uncomfortable. I think we've gotten over that. But how? It's really transparency, right? Where we have ways to display what the model is doing, how it's doing it and what assumptions it's making that I think better inform our customers and get them more comfortable that it's doing its job correctly. And that's there and it's auditable, right? So that you can audit that those assumptions stay consistent over time. Those are the sorts of things that, you know, an industry that has regulation like food, beverage, tobacco products, like really care about transparency in the system operation and then being able to audit those things over time. And as we've matured the technology and its product development, those are the things that we've brought to bear that have made those customers more comfortable using it. 

So beyond improved defect detection, how do you see the AI driven machine vision evolving on production lines, say, in the next three to five years. We'll just keep it short because we see how quickly things move. 

Yeah, I know. I know. We try not to plan much further beyond that. Yeah. I mean, I really think it's going to democratize the technology. What does that mean? I think, you know, when I started at Cognex, it really did require quite a bit of expertise, not just technical expertise, but machine vision specific knowledge to deploy Cognex. Our competitors really require time and expertise. And I think the best thing that we can do is, is use technology, specifically AI, to bring down that learning curve and make it much more consumable by a broader set of our customers. Right. Our customers, many of them don't have the budgets or the staff to dedicate to deploying machine vision, and I don't blame them for that. And so a lot of what we're trying to do is, you know, use, use AI and improve our product in the way that it can be easier to test, easier to pilot, easier to deploy. You know, we came to market with a new slogan initially internally, but now publicly advanced machine vision made easy. And it's a nice thing to say, but it's really important for us because on one hand, you know, we see ourselves as driving the boundary of what's possible technically, but we don't want to ask our customers to deal with the complexity that often comes with advanced technology. And at the same time, making sure that when we launch those tools or techniques, that they can be consumable by someone that has a few hours of experience with the Cognex vision system. And so I think that's, that's really, that's really what excites me more as we, as we lower that barrier to entry and barrier to trial, I think we'll get more customers interested and willing to test to deploy. And, you know, we'll be able to, to, to not just grow our business, but grow the industry and solve problems that, frankly, are really, really urgently needed to be solved. So can you talk about what emerging technologies complement AI in advancing machine vision capabilities? Yeah, it's such a good question. I think, um, you know, today we and our customers tell us that they prefer to use Cognex technology and deploy machine vision through embedded systems. What is an embedded system? It's really a smart vision system that has all of the software, all of its tools preloaded, and it can do its entire job with the computing that it has on board. And that's, um, a product strategy we've had for many years. But you know, what's happened in the market is because AI tools have accelerated in their sophistication much faster than I would say the embedded computing capabilities have. Uh, I'd say the biggest limiter in the past has really been where to run those tools and the performance of the embedded vision system to keep up with the computing demands of those latest generation models. That is what we're seeing really improve quite dramatically where we're seeing, um, you know, legacy chipset manufacturers, new, new entrants, uh, creating some very interesting, very powerful embedded computing chipsets specifically designed to run frontier AI tools. And those things coming together is very, very exciting where, you know, if the technology is ready, we now have a place to actually run it in a package that customers can, can consume and scale very, very quickly. And so, um, many of the new products that will be announcing this year are really of that vein where we're, we're pairing very powerful embedded hardware with already very powerful software and tools to deliver advanced technology, but also simplicity and scalability to our customers. 

Um, so what advice would you offer manufacturing leaders about balancing innovation and risk management when investing in AI machine vision? 

Yeah, I would say go back to some of the problems that you had over the last few years that for whatever reason, you thought AI technology wasn't appropriate for. Right. And this is the thing I tell all of our customers, which is revisit those problems, right? Because the technology is changing so quickly and maturing, quite frankly, maturing in terms of its performance, but also its usability where, you know, even six months later, some of the things that weren't possible are becoming possible. And so I would say dust off a lot of those, a lot of those applications that you maybe had locked in the drawer, I suspect that they can be achieved now. And don't expect or tolerate the complexity of years ago. Right. Demand that you know, the provider of your machine vision or automation technology is delivering you the best, but also the easiest. And I see Cognex as uniquely positioned to be able to do both. Heading into the future, continue to be the best and most advanced provider of technology, but also the easiest to consume and scale. All right. Let's talk about measuring impact. Sure. So besides quantitative metrics, do you find qualitative measures like operator confidence or decision making speed important? And why? And then how often should the companies review and update their KPIs in relation to this AI vision performance? Yeah, it's really important. The confidence of the operator, I mean, our customer is, is that line operator. Right. I really do think of it that way. Or often the head of quality, right? Where how does your line operate and how well is it operating? And these are the folks that have teams of people that their job is to, is to make sure that the line keeps running at the rates, at the yields, and with the safety and quality that that's required. And so to me, the best measure are the traditional operational measures, often under the umbrella of OE operational efficiency. And so we really align the value our products deliver to those traditional OE measures of throughput and yield. And I think what we're seeing now is AI tools are the best way to drive those key measures, right. And yield. Right. That's obviously detection accuracy and also the performance of that tool, meaning how, how flexible it can be and how many problems it can solve. Um, but also throughput, as I mentioned, as the tools are getting more efficient as the hardware is getting more performant, we're able to run them faster and keep up with the highest rates. And so no, I think the ROI for a vision system has been and will continue to be through putting yield and with AI. I wouldn't say that's necessarily changing. I would say we're just we're able to deliver better results than we ever have before. 

Can you share any lessons learned where maybe those initial success metrics didn't fully capture AI's impact long term. 

Yeah, often. Often our customers look at those two and then they, they, they solve for the value created by the system and the cost of that system. Increasingly, we're encouraging our customers to take a total cost of ownership view. And why? Because, you know, particularly as we are driving not just performance, which is throughput and yield, but ease of use, what you're seeing is other costs are also coming down, right? So the upfront engineering of testing and deploying a vision system has been cut in half or more. The cost to maintain that vision system is being reduced dramatically as its performance is improving. And so I would say even if the performance of the device stays the same, which it's not, it's getting better. The other surrounding costs that maybe were hidden or maybe not as well appreciated to deploy a vision application are also coming down dramatically. And so what we're working with customers on is really to, to see those costs and those complexities account for them when they're writing the business case to their management in terms of when and how to invest in this sort of automation. So, that's what I would say is moving from the cost of the product to the total cost of the solution in accounting for all of those things as part of the business case. 

Well, that's a wrap for this episode of visions produced by Endeavor Business Media, a division of endeavor B2B. Thanks very much for tuning in. If you enjoyed today's show, be sure to subscribe to the podcast and share this episode with a colleague who would find it helpful. Until our next episode, you can find us at vision dash systems dot com or on LinkedIn, Facebook, or for more insights, updates, and breaking news to keep you in the know. Thanks for tuning in. Until next time, stay focused on your visions.