Manufacturers rapidly address new challenges with AI-enabled vision sensors and simplified calibration for robotic applications. 

In this episode of Manufacturing Matters, SensoPart’s head of applications engineering, Alex Rudolph, joins TECH B2B Marketing’s Winn Hardin and John Lewis to discuss the latest in machine vision technology. ​They discuss SensoPart’s unique offerings in the industrial automation sector, particularly focusing on the integration of AI into vision sensors, its applications in quality control, and the evolving landscape of robotics.​ The conversation highlights SensoPart’s commitment to simplifying machine vision processes and enhancing user experience through innovative calibration methods and adaptable technology. 

Winn Hardin: [00:00:07] Hello, Everybody, and welcome to “Manufacturing Matters,” where we talk about the technology and trends that are reshaping the global manufacturing industry. Thanks for joining us today. I’m with my co-host John Lewis from Tech B2B. How are you doing, John?

John Lewis: [00:00:17] Doing great, man. Wonderful.

Winn Hardin: [00:00:19] Pleasure. Pleasure.

John Lewis: [00:00:19] How are you?

Winn Hardin: [00:00:20] I am pretty good. It’s been a good day. I’m enjoying myself. And today we’re lucky enough to have Alex Rudolph with us. Alex is from SensoPart. Alex, how’s your day?

Alex Rudolph: [00:00:30] So far, so good. I would say.

Winn Hardin: [00:00:31] Beautiful. Beautiful. So, I know SensoPart to be a machine vision expert. You guys play in that field. Can you tell us a little bit about what makes SensoPart special?

Alex Rudolph: [00:00:39] Okay. What makes SensoPart special? I would say we are maybe a hidden champion in the market. We are not as big as others, but I would say we do a pretty good job [given] our size. And sometimes the small one beats the big ones. And I would say we have outstanding products, but I’m sure I can talk a little later about what are specialties are.

John Lewis: [00:01:02] Wonderful.

Winn Hardin: [00:01:02] Yeah, we’ve got some questions all set up for that.

John Lewis: [00:01:04] Yeah. Can you tell us a little bit about what you do there at SensoPart?

Alex Rudolph: [00:01:07] So, I’m the head of applications engineering worldwide. So, my team is taking care . . . Like a consultancy for our customers. So, I like the way we do. We have very well educated technical sales people, and they work together in a tech team with our application engineers. So, if the customer has a question, we are there to help them try to find a way, the right way to solve these applications. And that’s a very good way that works out very well. And that’s how we compete in the market against big [companies].

John Lewis: [00:01:39] Yeah, there’s a lot of trends in industrial automation, machine vision. And one thing I really want to talk about is AI. I know when I was at VISION Stuttgart in 2016, everyone was kind of talking about deep learning and doing OCR applications with that. A few companies were coming out with tools for that. But now in these real-world applications . . . And I noticed on your website you have some involvement in AI. And I’m wondering, could you talk a little bit about what SensoPart’s approach to AI is?

Alex Rudolph: [00:02:07] So, our approach was to bring AI into a vision sensor. So, you know, the vision sensor hasn’t the highest computing power ever, like a PC-based system or whatever. And in the very beginning, I was involved when we had the first conversations with R&D. They said, oh, you need to have cloud computing and everything behind, and you need to have, I don’t know, octa core and GPU and whatever, and you need to teach in like 1000 images or whatever of the different classes. But we ended up with a very, very good and easy-to-use solution, I would say. So, you open up the conflict mode from the visor. You open up the detector, which we called classification AI, and then you can train up the pictures you would like to have. And it’s . . . After like eight pictures of one, of one sample, let’s say, you click on train and it takes a minute. So, it’s not time at all to take a coffee. And then you get a pretty good result. I would say. It’s crazy.

Winn Hardin: [00:03:07] Alex, are you working with some pretrained models to be able to expedite that whole process?

Alex Rudolph: [00:03:12] Yes, we have a pretrained model, and we rely on a, on a, let’s say pictures we got from . . . or training models we got. And then we teach our additional pictures in the applications. And that’s the way he is finding always the right model and also the right specifications of the samples.

Winn Hardin: [00:03:35] Okay. Beautiful.

John Lewis: [00:03:36] What kind of applications are you seeing customers deploying AI-enabled products into?

Alex Rudolph: [00:03:41] As I said, the detector is called classification AI. And in the beginning we considered having a detector which is able to distinguish between parts. But now that we have this, like nearly one-and-a-half years in the market, we see that the people are more using it to distinguish good and bad from each other. So, I see there are advantages, mainly if the part is not coming all the way, the same position, and you have different reflectances in the image and whatever. And then, it [turns] out that here AI is really a game changer because in the, in the past, you’d have to spend a lot of money for illumination or in, in the handling of the part to get it in a better way. And now we are more flexible just by adding those images. And that’s really cool.

Winn Hardin: [00:04:30] That’s nice when you can take some of the stress off of the lighting system design because, for so many years, that was a key part of what people would refer to as the black art of machine vision, but we’re making it a much more reliable solution. As you said, a vision sensor, which traditionally has been kind of an entry point into the machine vision market, it’s amazing that they’ve brought AI and deep learning capabilities to sensors.

John Lewis: [00:04:54] How do you see traditional machine vision tools and deep learning complementing each other? You know, single applications.

Alex Rudolph: [00:05:02] So, I would say they fill up. So we do have tools like caliber tools where we do geometrical measurements. We also have now in the latest release a detector implemented which we call the contour check. So, it’s like a shape tracing tool. And I would say, with traditional methods, it’s still obvious that you need to have it because you may have applications where you want to get, I don’t know, the radius of something or the diameter. So, here we need to stick with the traditional methods. But if you have something like a texture . . . So, at our booth, you may have seen we, we differentiate different tire types. And you know it’s, it’s just a different way to approach an application now with AI because in the past you needed to consider, okay, what’s the difference? Is it the contrast? Is it the brightness? Is it the contour? Is it the geometry or whatever? So, with AI you don’t need to think about something like that. You just click in. So, that’s my entire time. A or B or C or D. And then you click on training and then you’re done. But I see and that’s the reason why we have our object sensor, the VISOR Object, additionally equipped now with the AI tools so they can take use of both worlds and choose what is best for my application. Our customers.

Winn Hardin: [00:06:21] Earlier you mentioned that traditionally AI has been a very processor-intensive application at least function set. How . . . Is it, is it processor development capabilities that have progressed that have enabled this? Is it how you’re approaching your software design, or is it a merger of the two?

Alex Rudolph: [00:06:38] I think it’s more or less the software design. And for sure, we also will develop more processor speed into the camera because we see now there is, you know, you give something out to the market and then there is a big pull out of the market. So, currently, I would say our AI inspection rate is very fast. So, you can classify or check a lot of samples per second, I would say. But now there is demand coming. They would like to get more and more use of this and now, we need to keep up with this as well.

Winn Hardin: [00:07:15] Right. So, you know customers just have it go faster and cheaper right. So, yeah, always two easy problems to solve all the time.

John Lewis: [00:07:23] Have you noticed any recent trends in customer requests. And can you discuss how SensorPart is addressing them.

Winn Hardin: [00:07:32] Other than power and cheaper. Because that’s the gift.

Alex Rudolph: [00:07:36] Yeah, I don’t want to talk about cheaper. You know, I’m here to . . . Yeah, let’s say, the way we support customers is really being a partner of our customers. So, you know, I’m from Germany, and we have the big automotive brands in Germany. They’re, they’re really partners. I don’t want to call them customers. And we learned a lot from them and then we are able to spread this out to all our other customers. And so I would say, for instance, you have seen at our booth, we are also very present in robot guidance and not just AI. So, AI for me is object inspection and classification, robot guidance, where it is about positioning parts. There we see a big demand in the, in the market. It’s amazing how many robots you see here in the trade show. And I saw companies growing up like mushrooms, coming out of nowhere. And they built nice robots, I would say, but they also need to have good eyes on the robot. So, again, it’s the handling. Either you get your handling in control, or you need to have somebody tell you, Okay, now the object is, I don’t know, a little bit to the left, a little bit up, and a little bit rotated. And I would say we have implemented beside AI also very powerful VISOR Robotic we call it . . . a vision sensor, again. So, you know, there is a term, vision-guided robotics. I better like to call it vision sensor-guided robotics because, you know, vision sensor is more ease of use. More people can [put] their hands on. And we find a pretty nice way for calibrating . . . this little sensor to the robot. And we learned this from our customer, Mercedes. So, they told us, hey, look, if you do it like this and that, it’s much better for us because then I don’t need to send a whole maintenance team into the robot cell. And then that’s what we learned, and that’s what we implemented. And that’s what we now can give to all our customers.

Winn Hardin: [00:09:33] Simplified software, simplified interfaces, simplified hardware. That’s, that’s the progression that drives all of this, I think. I know you offer a number of different cameras in your camera family. Are some of them . . . Are they AI cameras going on robotics? And how does the application drive that? Or do you have other more, either traditional systems or is AI across the whole product line?

Alex Rudolph: [00:09:55] No. So, you know, we have a visor. We call it The Family. So, it’s all related on the same hardware. It’s also the same software we have behind [it]. But the choice of the detectors you have inside these different models, they vary because not every robot customer needs to have an AI for inspection. Not every AI inspection application needs to have code reading or whatever, so you can choose which is the best for you. So, that’s one of our possibilities to keep the price low. And on the other side, we also have, it’s called the Allround device, where we put everything in. So, that’s pretty good for the laboratory, but it’s also good if you have mixed applications in your application at all. And then we can [make] use of applications in robotics and AI, in code reading, and whatever.

John Lewis: [00:10:49] Yeah, you mentioned calibration. And I know in the past . . . robot calibration vision is kind of complicated. There’s intrinsic calibration and extrinsic calibration and . . .

Winn Hardin: [00:11:01] Global coordinate calibration.

John Lewis: [00:11:03] Yeah. Is there, is there . . . How do you make it easy for customers?

Alex Rudolph: [00:11:07] That’s that’s a . . . I like it a lot. So, when we started with robotics, we had a calibration that was based on point pairs. So, so that’s very common in the market. So, you place parts in your field of view, then you take your robot pointing to them. You look in the camera, what are the coordinates, and then you start filling up the list. But that’s, I would say not so nice. And there is a worker in the need to go in and, imagine we are talking about automotive. And in automotive, you are just not able to make a point pair list out of it. And what we did this, I called it the marriage dance in German — the Hochzeitstanz. So . . . you know, we mount typically our camera end of arm because she is really robust. She is very compact and we can really implement it into the robot. We have industrial cables. Everything is shockproof and we have integrated illumination, integrated lens. And this makes it really suitable for end-of-arm mounting. And now there comes the calibration into it. So, you can place our calibration plate somewhere you can reach in, in the whole robot. So, it could be under the roof, it could be on the floor, it could be on the side wall. As long as you can go there with your camera. And then you do this marriage dance. So, you go into the measurement position and then you move the robot in line moves around. So, like I was learning the Wiener Walzer. And then afterwards, you get a calibrated system because the robot is sending the coordinates to the camera. The camera is taking them. She is filling up the list of the points and then she’s calculating back to the robot coordinate space. And from this moment on, you are calibrated. No need to know the exact TCP. No need to work or get into the groove.

Winn Hardin: [00:12:57] That’s . . .

Alex Rudolph: [00:12:58] Brilliant. The customers love it.

Winn Hardin: [00:13:00] Absolutely.

John Lewis: [00:13:01] There’s nine robot moves?

Alex Rudolph: [00:13:02] Yeah. And nine points. Yeah.

Winn Hardin: [00:13:06] And I assume that’s pretty much an automated routine, right? In terms of teaching. That’s just, you know, set, go, calibrate, comes back. We’re good to go.

Alex Rudolph: [00:13:12] It’s easy.

Winn Hardin: [00:13:13] Beautiful.

Alex Rudolph: [00:13:14] And we did this . . . Now, we started with ABB and Kuka implementing implementing this. And nowadays I would say we, we are able to handle . . . Now I see some new robot brands. I would say we can handle like 80 or 90% of the robot brands on the market. So, we are completely independent from the robot’s manufacturer. And if there is a new one, it’s just a little bit of maybe copying from what we did for UR or copying what we did for ABB or . . . So, there’s also, for instance, a good partner here in the U.S. And I watched them code, and they said, oh, we copied a little bit here and there, and after like 20 minutes, they had a running library.

Winn Hardin: [00:13:54] That’s it. So, your solutions are definitely going on traditional and cobot on a regular basis for machine vision. Is that a fair statement or no? Is it more still a traditional big industrial bots that are using VGR versus cobots?

Alex Rudolph: [00:14:05] So, I would say, you know, cobots are really trendy because they are more easy to install at customer sites. Every application looks a little different. Sometimes it’s loading a test station, sometimes it’s taking a part out of a machine. I would say the classic robots, you know, the really big ones like you have in machine handling, in automotive, or bigger parts, or if you produce a washing machine or whatever. I would say . . . For us it’s . . . We are not related to one or the other.

Winn Hardin: [00:14:35] So, still traditional robotics is leveraging . . . And you see . . . I think that’s pretty common. I mean, the cobot folks are expanding their ecosystems to include machine vision, but in many of the applications . . . And I think maybe another part of that is that they’re targeting really small and medium enterprises, relatively, who might have tighter budgets, have less complex applications, need that repeatability but really aren’t as focused on maybe throughput, for example. And of course positional accuracy with cobot is traditionally lower than industrial. So, that’s another component.

Alex Rudolph: [00:15:05] As you said. So, there is definitely a trend for automation which goes to robotics. And also now the smaller and the midsized companies are investing in robots, and the integration time gets lower. Also with our tools, with all technologies, they can reduce their the complete integration time. And also if I look back at the production at SensorPart, as you know, we produce it also on our own. So, we develop and produce it. Now, also, the first robots came into . . . Traditionally, we are 30 years old. We have an anniversary this year.

[00:15:37] No kidding. Congratulations.

[00:15:37] All was handmade. But now we started also to implement robots into production because the return on invest is being reached very soon.

Winn Hardin: [00:15:48] Are there any topics we haven’t touched on, Alex, today, especially in major trends maybe related to what’s next? I know you can’t share with us your future roadmap, of course, but if we look at machine vision, we look at VGR, we look at all these applications and the additional AI, what’s something to keep an eye on, on the horizon?

Alex Rudolph: [00:16:05] So, I think the way you teach your sensors will change. Nowadays, you know, if I look back in the, in the past, you need to be able to handle a programming language like C# or Java or whatever. Then companies like SensorPart, we made, made it parameterizable, let’s say. So, we configure what we want to do. And I think, you know, everybody now knows ChatGPT. So, where you go into an . . . I call it like an interview with you. And I could imagine that one of the next steps will be, you say, look, here’s my image. I’m interested in getting the, I don’t know, position of the three holes. I would like to know the diameter, and I would like to know the distance from the center to the corner or whatever. And then he’s doing like boop-boop-boop-boop and then you get your results. So, I could imagine that this will be getting into . . .

Winn Hardin: [00:17:00] I think you’re right. I mean, we’re already seeing some hints of that to, a little bit, you know, being able to take maybe part of a spec from in particular part tolerances, put those right in and kind of develop a routine for inspecting those key tolerances. John, do you have any other thoughts?

John Lewis: [00:17:14] I don’t know. Is there anything that we haven’t talked about that you think is important for our audience?

Alex Rudolph: [00:17:19] I don’t know. So, I could talk another eight hours about SensoPart, even if we are quite a small company.

Winn Hardin: [00:17:26] Anything special for the 30th anniversary?

Alex Rudolph: [00:17:30] Yes, absolutely. We do have, our ISMs. So, there is my colleague, Karen, sitting. So, we do have a big party coming soon in four weeks, and we will have a nice trip on our home River Rhine in Germany. And I’m really looking forward to this. And everybody is coming from the whole world.

Winn Hardin: [00:17:52] Wonderful.

Alex Rudolph: [00:17:52] So, I’m traveling a lot. I’m traveling to, as you see, to U.S. I was in China four weeks ago, Korea. And now everybody is coming to Germany. And I’m just looking forward to meeting all these people and having a nice time.

Winn Hardin: [00:18:04] You’re looking forward to taking a nap on that boat after all that kind of world travel. That’s crazy. Well, Alex, thank you. We can’t thank you enough for taking some time out of your day to come share with us about SensoPart and talk about AI and traditional machine vision and vision-guided robotics. Always exciting topics. And you know, it’s nice . . .

Alex Rudolph: [00:18:20] Don’t call it vision-guided robotics. Call it vision sensor or guided robotics.

Winn Hardin: [00:18:24] Gotcha. Got it. You’re absolutely right. And.

John Lewis: [00:18:27] The new acronym.

[00:18:28] That’s just what we need. Now, John and I will be looking forward to our invitations to the party next month, so that’d be great.

Alex Rudolph: [00:18:37] We’ll see if we have two spots left.

Winn Hardin: [00:18:40] That would be beautiful. Need a bigger boat. So, if you guys have got any questions for Alex at SensoPart or want to learn a little bit more, go to SensoPart.com, and you can check out the educational materials. Look at their product information, including what’s hot and being shown recently at the Automate conference. If you’ve got any other questions, send them to us. We’ll be happy to pass them along to Alex. He’ll get back to you as soon as possible. In the meantime, thanks for watching “Manufacturing Matters.” You can catch our old episodes at Manufacturing-Matters.com. If you’d like to be on a future episode, ping us down below and make sure you check out our podcast on all your major podcast platforms. Until next time, thank you so much for your time, and we look forward to seeing you again.