Jimmy Carroll: [00:00:06] Hi everybody. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing. I’m here at SPIE Photonics West 2024. I have the pleasure of being joined by Doug Erlemann of Baumer and my colleague John Lewis. Doug, thank you so much for taking the time today. I really appreciate it.

Doug Erlemann: [00:00:21] I appreciate you asking me to speak today.

Jimmy Carroll: [00:00:24] Absolutely. So Doug, for those who don’t know, could you tell us a little bit about Baumer and what you guys are excited about today?

Doug Erlemann: [00:00:31] Well, Baumer is a half a billion Euro a year company, is a Swiss-based company that manufactures sensors and coders, a lot of very leading-edge sensing things. IO-Link. And I work specifically for the division, which is based in Germany, and they manufacture industrial cameras that are used in all kinds of applications. I’ve been with Baumer 16 years, and they brought me here on board then to get sales started with cameras. And we’ve been pretty successful so far. And I love selling the product because I always tell people we make good stuff that doesn’t break. For me, I think in 16 years I think I’ve had three failures of cameras.

Jimmy Carroll: [00:01:29] You don’t hear it a lot from people out of their mouths directly. But like, having products that don’t break is obviously a very nice thing to have.

John Lewis: [00:01:36] When you’re manufacturing, downtime is really costly.

Doug Erlemann: [00:01:38] Well, it’s not only costly but on the front end, on the sales side, you’re just doing nothing but damage control, which really is not a way to run a business. So I really appreciate that. And Baumer itself has kept up. We keep coming out with the newer . . . we were one of the first camera companies to come out with gigabit Ethernet, which has been the standard since 2005. And now we’ve got cameras that are 10 gig, 5 gig interfaces, fiber. And so wherever there’s good machine vision applications, our cameras will fit in.

John Lewis: [00:02:21] That’s great. And yeah, there’s a lot of hype, hot topics in the industrial automation and manufacturing space. And one of them is AI. It seems like everyone who’s into machine vision and imaging and even robotics these days are really touting the AI aspect of their products. And I’m wondering what’s Baumer’s approach to embracing the AI trend?

Doug Erlemann: [00:02:50] Well, we’ve just come out with a fairly new product about a year ago, and see we make hardware, and AI is really on the software side of things, but we came out with a camera that is an NVIDIA-based camera. So you get a 5 megapixel image sensor feeding into an Xavier NVIDIA module. And you can load your software to do the AI. One of the applications that I’ve seen, there’s a lot of them, but I have a customer that is one of our best customers. He sells into the wood business and into lumber. And how they make the paper that they get from lumber is they grind a lot of the excess  wood down to chips. And it’s on a conveyor about this wide. But the conveyor is like this, but it’s filled to the top, and they’re using AI to try and figure out: are the chips too big in here? Because otherwise it’ll make the slurry a big problem. You’ll get failures when you get these big wide paper rolls, about as wide as this table, going across at I don’t know what speed — very fast. And if they have a tear because they didn’t read the slurry correctly, then the plant is shut down for four hours. And plant managers, they get real excited about that. Not in a good way. So that’s just one of the applications that I’ve seen. It’s fascinating to see how they use this camera to figure out how to do this.

Jimmy Carroll: [00:04:43] Yeah. That’s an interesting one. There’s a few things that you brought up there that I want to ask about. But one of them is some of these applications. And that’s an interesting one, right, because for so long, Doug, I’m sure you were at, or at least your team but you probably were there too. I think it was Stuttgart, maybe around 2016.

Doug Erlemann: [00:05:01] Yeah, yeah, I was there.

Jimmy Carroll: [00:05:03] You know, that’s when the hype cycle really began for specifically the subset of deep learning. And people are just coming out and saying deep learning is going to solve all your problems. You’re kind of even seeing that today with some companies just throwing the word AI out there or robotics. Incorporate AI. Well what does that mean? But now these types of applications, you’re seeing that these AI-based, because AI on itself doesn’t really exist, but AI-based software exists and these tools exist that can augment existing machine vision tools and work together instead of replacing, because it’s not going to replace traditional machine vision, but it’s going to help it in different ways. And that’s a really cool one, because some of the more common ones you hear are maybe OCR or things like that or in welding inspection. But to that end, what are some other examples of applications where you’ve seen your customers or people in the industry that you know deploying AI-based tools or smart cameras or whatever?

Doug Erlemann: [00:06:01] Well, it really comes down to that a lot of customers want to move towards what they call edge computing. They want to do the calculations, failure, barcode reading, or whatever right at the location they’re making the inspection. And this helps them out in so many different ways. And if you look at the overall cost of a system, let’s say you have 32 cameras on a line and it’s doing all these different inspections. And you have to run the software in the main computer, which is about 80 meters away. To do that kind of computing power for 32 cameras, you need three big Dell servers at least. And these servers they’re 15 grand a pop. So that’s 45 grand right there. So when you do the calculations of doing the edge computing and the only thing that’s sent back is “We got a failure,” then you don’t need all those three big computers. You just need a smaller single computer. And all the computing is done out at the inspection site, and in overall cost they’re saving probably about one-third, 33% is a good guess.

John Lewis: [00:07:29] Yeah. So taking a distributed approach to processing that central server requirement. And deploying the deep learning right at the edge just means that you only have to transfer the result right back to the process controller or whatever.

Doug Erlemann: [00:07:44] So that’s right. They see an error and then they send the pictures of the error and everybody gets excited. The line shuts down. Oh but, you know, that’s manufacturing. We keep getting better and better at this manufacturing. And I believe that this AI and distributed computing is going to allow for a lot more complex applications to be solved. We’re just at the beginning of this.

John Lewis: [00:08:15] Yeah, I know there’s a big trend towards trying to make – five, ten years ago, if you were going to deploy deep learning and machine vision, you may need a data scientist or someone to deploy it. And now a lot of the machine vision software companies are trying to facilitate deploying deep learning by providing ready-to-go tools that are a little easier to use and more accessible to people on the factory floor.

Doug Erlemann: [00:08:43] That’s exactly right.

John Lewis: [00:08:44] I’ve heard it’s still very important, regardless of what type of AI you’re using, deep learning in your machine vision application, to train your staff, or you have to be aware of what it is and how it works. What they’re saying is, instead of if there’s an anomaly or something that the system isn’t catching, it’s easy for a human to go in and retag the image as flawed. So you update the model to get better results.

Doug Erlemann: [00:09:11] Yeah. It helps when you have good operators and you train them properly. It just makes the entire manufacturing process work much, much better.

Jimmy Carroll: [00:09:26] Yeah. And it’s kind of another one of those things that dispels . . . I mean robots were kind of the real thing, the real focus of it, but it kind of helps dispel that notion of “Automation is going to take my job.” No, AI is only as good as the model it is trained on and keeping a human in the loop.

Doug Erlemann: [00:09:44] That’s exactly right. We’ve had our cameras used on a lot of  rogue robots, for example, for positioning. And so they sell the camera and the software, and they’ve told me, that particular customer has told me that they really, really appreciate a good robotic operator because the last thing you want happening, which has happened, is when the camera sends the wrong data back to the software, and the software says, “Okay, go here.” And you’ve just wiped out the entire door from a Mercedes. That’s when everybody gets real excited. So I mean, you get a call Saturday night going, do you have an extra one of these cameras? Because it’s not working anymore.

Jimmy Carroll: [00:10:36] Doug, one of the things that you mentioned earlier that I wanted to follow up on is, and not to make this a overly promotional for NVIDIA or anything, but I think about processing power and some of the advancements that have been made in tools like these GPUs. Some of these larger GPUs now are massive. They’re huge. And you’re going to have to have a custom case for it and cooling and all this. But then there’s these Jetson line of products that are small and relatively low power that are adequate for most or many, anyway, deep learning/AI type applications. And it’s cool to see, because for your customers to be able to solve certain problems, it’s required the advancement of multiple technologies, not just camera or sensor or lighting or robotics. It’s a lot of different things coming together, the computing aspect of it. How have advancements from NVIDEA and others in the industrial computing world helped?

Doug Erlemann: [00:11:41] Well, it certainly helped our company in the fact that it gives us a market, a bigger market to go after and more complex applications. And we’re not just stopping on the Xavier model. We’re going to go to the next platform. You know, we have a 3.2 megapixel camera and a 5 megapixel that is hooked in the Jetson models. But we’re coming out with a 16 meg shortly because I’ve noticed in the last 20 years I’ve been in this industry, it’s always, “Well, you know what? I want more data. I want more data because I want to make sure that my manufacturing process and all the details are spot-on.” And in order to do that, I’ve got to get more pixels onto the object I’m looking at so that I can get more data. And there’s a lot more processing needed to process a 16 megapixel signal coming back than a 5 megapixel. And it’s not just three times, 3x. It’s more than that. So we’re pushing to come out with newer models and especially with the Jetson platform because you’re right, relatively speaking it’s somewhat low power. And in our case, you don’t have to cool it. That’s one of the side things that people will sometimes complain about, that this camera sounds pretty big. Well, we don’t cool it, so you need something big in order to wick the heat away. And we do a lot of thermal analysis when we’re designing the cameras.

Jimmy Carroll: [00:13:38] Yeah, it’s big, but it won’t melt.

Doug Erlemann: [00:13:42] That’s true. And you can grab it and it’s not going to burn your hand.

John Lewis: [00:13:48] They really don’t like fans on the manufacturing floor.

Doug Erlemann: [00:13:51] They hate them. They hate them.

Doug Erlemann: [00:13:52] I’m looking forward to the the next 10 years with all this new information and the new products coming out and the new software especially. So it’s going to solve a lot of manufacturing problems.

John Lewis: [00:14:13] Yeah that’s what I’ve heard. A lot of people that are deploying edge learning, they’re finding just new applications that they couldn’t scale before. Like you said, with the with the paper and the lumber, natural products are really tricky for traditional machine vision tools, because it could be fine, but they don’t all look the same. It has natural variations, so it’s hard to quantify a defect. And that’s where the deep learning really comes in.

Doug Erlemann: [00:14:36] Oh yeah. You know, you look at the at those rolls of paper, I mean they’re flying by and they’re somewhat discolored. So is that bad? Does it have too much water in it? Yeah, it’s beyond me. That’s why I let my customer figure those out.

Jimmy Carroll: [00:14:55] It’s in your hands.

Doug Erlemann: [00:14:56] Yeah. I give them the tools to solve the problem, and I do learn a lot from my customers. They’ll tell me this is how it’s working. I say, well, okay, cool. I mean I do have a degree in engineering, but this is a little bit over my head what you’re telling me.

Jimmy Carroll: [00:15:16] One other thing I wanted to ask, Doug. I don’t want to take too much of your time here at the show, obviously, but maybe it’s an obvious one, but it comes up a lot in these conversations and you see it out there in people’s websites and so on is the idea of ease of use. It seems to be, and it’s probably mostly on the software side, but it’s the idea of making a lot of these tools more accessible to somebody who is not an engineer, who works on a plant floor, that can quickly run a program, troubleshoot a program, things like that. How are you seeing ease of use drive, let’s say, increased automation adoption?

Doug Erlemann: [00:15:53] Well, it is driving it because you want the line workers to be more autonomous. You want them to solve the problem. They’re dealing with it every day, and the last thing you want is, okay, shut it down and then let’s wait for the engineer who’s up in the home office to come down and fix it. You don’t want that. You want the line worker to solve the problem. I have a customer that,  using a couple of our cameras, made a replay unit. It’s a pretty high-speed . . . putting sleeves on this product. And sometimes it gets jammed, and you can’t keep calling the engineer out to fix it. The line workers do it. And so he did the replay, and the line worker looks at it in slow-mo and goes, “Got it.” And he goes out and fixes it. And that’s what you want. It’s more efficient for the manufacturing line.

Jimmy Carroll: [00:16:57] Ultimately you’re saving money. You know, minimize downtime. John, anything else to ask, or Doug, anything that we haven’t asked about that you want to bring up that you’re particularly excited about in the industry or the general market?

Doug Erlemann: [00:17:13] No just the general market. We keep going after more and more complex problems, and in itself is going to make manufacturing easier, faster, and no matter what you’re making. And that’s what’s pretty exciting.

Jimmy Carroll: [00:17:32] Very cool. Well, Doug, thank you so much for taking the time.

Doug Erlemann: [00:17:36] Again, thanks for having me here. All right.

Jimmy Carroll: [00:17:37] Of course. Yeah. It’s been our pleasure. So it’s www.manufacturing-matters.com. If you have questions or comments or want to reach out about joining a future podcast. To learn more about Baumer, you can go to baumer.com or find them on LinkedIn, obviously. If you have questions for Doug, feel free to reach out to us. We’ll pass them along to Doug. And thank you so much for watching. All right.

Doug Erlemann: [00:17:56] Thank you guys.

John Lewis: [00:17:57] Thanks Doug.