In this episode of Manufacturing Matters, Manny Romero of Teledyne discusses the pivotal role of automation technologies in today’s evolving manufacturing landscape. The conversion dives into topics including artificial intelligence and deep learning, including Teledyne’s Astrocyte software, which end users can deploy without the need for AI expertise or coding, delivering an automated means for anomaly detection, classification, object detection, and segmentation.

The discussion also covers Teledyne’s extensive product lineup, which includes area scan and line cameras, 3D sensors, smart cameras, image sensors, frame grabbers, and additional software solutions, all tailored to meet the diverse needs of automation systems. Throughout the episode, Manny sheds light on various interface standards like GigE, USB3 Vision, Camera Link, and CoaXPress, elaborating on their individual benefits and use cases.

The episode concludes with a look toward the future of automation, both on and beyond the factory floor. Manny envisions the industry’s growth being propelled by growth in areas including autonomous vehicles, agriculture, and electrification and environmental consciousness.

Winn Hardin: [00:00:07] Welcome to a new episode of Manufacturing Matters, where we talk about the automation technologies that help manufacturing to continue to matter here in the U.S. and across the world. And today I’ve got something special for everyone. I’m sitting here with Manny Romero from Teledyne.

Manny Romero: [00:00:22] Thank you.

Winn Hardin: [00:00:22] Manny, so we probably spoke the first time 20-plus years ago I want to say?

Manny Romero: [00:00:28] Yeah, absolutely.

Winn Hardin: [00:00:29] You know, and I knew you from all the standards work that you’ve done with A3 over the years. So really, thanks for taking some time out today.

Manny Romero: [00:00:36] I really appreciate it.

Winn Hardin: [00:00:38] Manny, let’s start with the background. Well, first of all, give me your title and what you do at Teledyne, and then we’ll dive into the company you represent.

Manny Romero: [00:00:48] So currently I’m product manager for the area scan product that we have within Teledyne. I cover the Teledyne Lumenara brand, the Teledyne DALSA brand, and some of the Teledyne FLIR IS portfolio. So I’ve been doing it for, as you highlighted, a long time. In product management 16 years and nine years prior to that, still with the organization but in a different role.

Winn Hardin: [00:01:19] Awesome. Awesome. So you just touched on a couple of product lines at Teledyne, but let’s talk a little bit more about that because I don’t think people always understand the breadth and the fact that Teledyne can be a one-stop shop for so many of our automation needs.

Manny Romero: [00:01:35] Absolutely. From the different type of sensors that go into our camera. So 1D cameras, 2D cameras or the matrix-type camera, 3D cameras. We’re also very well-known for our frame grabber, and we have software capability. We have two software packages. It’s been now 20-plus years. Sapera. We had also Sherlock. That was for end users. So people that absolutely don’t know anything about coding and they want to get up and running to add image processing and we even added lately the capability for deep learning within that Sherlock software. So somebody that does not know programming whatsoever can get up and running. Part of that group includes also our smart camera that we also have with that group.

Winn Hardin: [00:02:36] I want to delve a lot deeper into that because it’s amazing you can combine object-based programming in Sherlock with the capabilities of deep learning. But let’s not totally get ahead of ourselves real quick. So we were just talking about the portfolio about 1, 2, 3D. Is there more there before I start asking, because I want to talk about, you mentioned earlier the Lumenara acquisition and how some of those sensors are starting to really make a difference in what you guys are producing.

Manny Romero: [00:03:04] Yes. Well, from the Lumenara, from that acquisition, it was a portion that was more microscopy oriented. Which was not a risk originally what let’s say DALSA was doing. So that was something that was added to the portfolio. Same thing with the acquisition of Photometrics and Princeton Instruments. So they went more into the scientific. That technology got added to the family. The sensor technology that we had with the acquisition of e2v bring that whole concept into what we can offer to customers. So when you said one-stop shop, you’re absolutely right. We’re covering anything that anybody could need in that industry.

Winn Hardin: [00:03:46] So Lumenara, they came from the scientific sensor space, right? And Princeton brings a little bit of the infrared spectrum then?

Manny Romero: [00:03:53] Yes, and obviously photometrics was really those high-end scientific cameras. Actually you were talking about the different type of cameras. I mean, even if we’ve been doing line scan, so 1D for a long time, 2D and 3D has been more recent, but the 1D we have not stopped innovating into that space. Some of our latest product that we just announced was the AxCIS, which is the image sensor contact. The beauty of that is the concept of line scan, but with the integrated light and lens. So it’s relatively easy to use, to install. So we brought that experience into that space. Additionally, another product that we announced a few weeks ago was the Linea2 multispectral. Super interesting. We’re going elsewhere with different wavelength.

Winn Hardin: [00:05:00] Now is that preselected, the wave bands that you’re interested in or are you able to select?

Manny Romero: [00:05:05] No they’re predefined. So very useful when you’re talking about food inspection.

Winn Hardin: [00:05:13] Yes, especially it would be a large area if we’re doing vegetable inspection.

Manny Romero: [00:05:17] Yeah. Free-fall type, and you want to identify the good – can be coffee beans, rice, you name it – the good from the bad. And then in the 2D realm again, we never stop innovating, which is why it’s so cool. And the 2D realm, with our Teledyne FLIR IS group, which is in Richmond, British Columbia, we introduced the new 5GigE camera, the Forge. You know, we’re seeing as how 1GigE was a dominating interface in the industry, the 5GigE is really the next big thing, if people just want to go fast. And we pushed the logic even further by making sure that the camera itself has an even faster internal acquisition. So we can do very interesting things and that either the mix of the sensor and the link speed is not the limiting factor. So, for example, you want to do a wide dynamic range imaging, we want to see better. You make a double capture very rapidly. You could sequence to a series of images at different wavelengths to actually see more on whatever you’re trying to inspect. Again, in the realm of the 2D, we came out with the, I would say right now the smallest 67 million pixel 10GigE camera you can find in the industry, super tiny, 49 by 49 – sorry, 59 by 59 millimeter. It’s very optimal for such a large resolution. And then again, in terms of innovation, using our sister company e2v, we added some of the fastest sensor that there is out there. So we have, again, even faster version than the 67 that I talked about that run at 90 frames per second. Very interesting for multiple – you know battery inspection. So it’s very, very, very cool. And then the absolutely latest we came out with, it’s a 2.8 million pixels. So you might say, “2.8 it’s not that much,” but it’s running at 1,200 frames per second. So you’re like way out there in terms of throughput.

Winn Hardin: [00:07:45] You know when you mentioned earlier about 5GigE cameras coming on, it made me think about handling the thermal load, thermal dissipation, and running that kind of resolution speed at that kind of frame rate. Obviously, the technology has come a very long way from 10 and 20 years, when I was first starting to get my head around it. I’m just curious, how do you manage that?

Manny Romero: [00:08:07] ***Q I would say that you don’t need to go 20 years, even in the last few years. You know, you were PJs, the Fi much more efficient in terms of power dissipation and obviously a good design and making sure that we thermally can dissipate the heat properly, especially when you push it that far. So that’s, again, a product that we’re very excited to see coming out into play.

Winn Hardin: [00:08:36] What was the name of that one?

Manny Romero: [00:08:40] The Falcon4, which is using the Camera Link HS standard, which is fiber optic, so distance is not an issue. And then you can go up to, I would compare the standard versus a 1GigE link speed as a reference point. You know, our fastest one is basically 70 times faster, with like 70 times the 1GigE camera.

Winn Hardin: [00:09:10] So one of the things you wanted to talk about was standards in the interface. I mean, there’s so much confusion between Camera Link, between CXP, USB3 Vision, GigE Vision. There’s probably no one better to ask this question to. Can you simplify for us, what is the sweet spot of each one of these? And as a follow-up question, is there any one of the standards that Teledyne doesn’t support right now?

Manny Romero: [00:09:34] Uh, actually no. We actually have it all. And actually your question is very on the dot. Some people don’t actually realize the pros and cons of each of them. So if you’re looking at a longer distance, like you don’t have a 2 meter limit, 2 meters from your processing engine or your computer or an embedded for your camera, and you need to go, I don’t know, 20 meters, USB3 Vision is not the ideal solution. But that’s a great interface when you got a nice embedded system, not that far, and you want a VMA directly from the camera to the computer, that’s a great interface that includes the power with it. Super for that. Now if you want to go far . . . 

Winn Hardin: [00:10:38] Security applications or even just large plants.

Manny Romero: [00:10:45] Actually, I can go as far as – I can give you a very concrete example. As far as, in a stadium, multi-camera around sports event.

Winn Hardin: [00:10:54] Which is what it’s all about these days.

Manny Romero: [00:10:55] And it happens a lot. Obviously people want to, well, it is what it is. People are betting sometimes, right? So they want to know about, okay, which team is going to win? How do they do that? Well, themselves, they want to feel secure that they’re looking at all the data to guarantee that they win. Winning is something else. But they will be looking at the event.

Winn Hardin: [00:11:19] At least they’ll be informed when they lose the money they’re going to lose.

Manny Romero: [00:11:21] They’re going to look at the event’s statistics. Well, that is coming from the main players that are buying our cameras to actually go into the stadium to capture those data, synchronize it, all the cameras. So that’s actually a scenario that would mix the duty of GigE for the distance and for perfect synchronization with PTP, so multi-camera synchronization, very precise. You can go down easily to 8 nanoseconds accuracy from camera to camera, very far. It does include power. So if you have a very good system, easy to cable. And it’s a robotic application, right? You got some nice cable that we can find on the floor there, super flex, so that they can go on those systems. It’s perfect. You need to go much higher bandwidth than you’re going to start looking. Camera Link, it was great when it was created in early 2000s. And I remember the comedian talking and the nightmare of cabling back in those days was the main reason why it was creating that standard. Ease of use for customers, interoperability between the cameras and the multiple frame grabber. It was great. Time went by. Then that’s where we saw CXP rising, again, same principle, added power, like the Camera Link did, higher bandwidth, because the first version was CXP 6. So 6 equivalent, I would say equivalent to 5GigE link speed by wire. And you know, there’s four of them or can go up to four. Then eventually they moved to the CXP 12, which is a 12 gig range, but because of the technology of how it encodes on the wire, I would say it’s a fair statement to say that a 10GigE Ethernet is equivalent to a one-lane 12GigE CXP. That’s something that people have not realized. But in terms of absolute bandwidth, they’re equal because of the way the data get encoded on the wire. And then if you want to go even faster, that’s where the Camera Link HS standard was created, almost at the same time as the CXP, and then obviously our we’re way up there, and it’s available with fiber optics. So distance is not an issue. So it kept moving forward. I’m expecting, from my of view, I expect that within the next few years we’ll be hearing more about GigE Vision. You know, right now, as we say, 1GigE, 5GigE, 10GigE. I’m expecting in a few years we’ll hear way more about the 25GigE, 50GigE, 100GigE. I was joking to some of our sales guys in the office. I was asking them, do know actually where the broadcast industry is right now? And I was saying, I don’t know. So where is it? But right now they were talking about the 800GigE that they’re using. And that basically is the platform of what the whole GigE Vision is based on. So it basically laid the table for a lot of available bandwidth in the future.

Winn Hardin: [00:14:54] So how does broadcast do that? With multiplexing? Essentially, they’d be able to achieve the 800?

Manny Romero: [00:14:58] No, they’re using multiple lanes of fiber optic. And instead of being four wires, they’re eight wires. That’s why those modules they’re called, I think it’s OSFP. You might have, I’m sure you heard about the QSFP, right? Or SFP module. So SFP module is a single connection. QSFP, the Q is for quad, and the O is for octo. So it’s basically eight lanes of 100GigE, each of them.

Winn Hardin: [00:15:33] Is there a latency difference between the CXP and the industrial GigE or are they all pretty much deterministic safety rated?

Manny Romero: [00:15:46] No, I would say that they’re . . . I’m a fan of GigE for the simple reason that the nature of the interface is designed that you got error correction. It would allow a packet resend, so a good system that is designed will never lose data. Now having said that, there’s a portion to it that I think has been minimized over the year, which is the quality of some implementation on the software side, on the computer side. Because of our background in line scan and our customer base at Teledyne still asking us to push that bandwidth further and further, it’s much tougher to transfer on line scan than area scan because the amount of line you’re getting on the wires is high. Same thing with one of the products that we didn’t even cover. But in term of capability, it’s following the same line as our 3D camera, right? Sending that profile on the wire. So we have a our Z-Trak2, which is 5GigE, and it transfers profile, pretty much the same concept of line scan getting a line, being able to sustain that high throughput when you got a really good product. But if you don’t have a good software at the end of it, that basically sustains it, and the definition of sustain mean that when our customers start to process on top of the acquisition, then they really want to take advantage of their computer. It’s not to say I’m going to buy an 8-core CPU to only use half of them, right? They want to make sure that they utilize that as much as possible. When our customers start using the processing with the acquisition, and they’re getting close to the edge of what the computer can do, that’s how you know if you’ve got a good driver or not a good driver. I know that I’m very impressed with what our guys were able to do over the years, and the last, I guess since 2005 I think, so you’re talking about basically a 17-year-old experience on all that matter . . .

Winn Hardin: [00:18:20] Without having to use the frame grabber.

Manny Romero: [00:18:21] No, no, no. Yeah, that’s exactly, that’s the quality of implementation.

Winn Hardin: [00:18:26] And it’s funny because just a few years ago, when everyone would quote bandwidth for a specific standard, it was always, that’s the burst, that’s what it can achieve for short periods of time. And everyone was really thrilled and happy that they didn’t have to sustain it across, like you do in line scan, infinite imaging, tenuous imaging systems. But now we’re pushing right up against it. I mean, that’s incredibly cool.

Manny Romero: [00:18:51] But I would say, I do not minimize the frame grabber to the equation because as we got more and more applications pushing that bandwidth further and further, I see space for having, let’s say, dedicated let’s call it GigE frame grabber, that when you really, really want to push, imagine this, 50 or 100 camera GigE. You’re talking about 100GigE of bandwidth. Now at that point, you could have a very good software, but you’re pushing the boundary too far. So there is still a place for those

Winn Hardin: [00:19:32] Yeah, that’s not going to be a PC. That’s going to be something, without having to go to some kind of rack-mount monster.

Manny Romero: [00:19:37] Yeah. And you know, you might want to ease that data acquisition by using a GigE frame grabber, for example.

Winn Hardin: [00:19:44] Especially if you’re going to do any kind of preprocessing to kind of cut it down, the multi-core. Okay, so we talked about the benefits of GigE. We talked about bandwidth and speed. We know that we’ve got strong latency, has a little bit more on the hardware cost us because a need a frame grab with the CXP. We talked about USB3 Vision being great for embedded systems for short runs. And I suppose there’s actually a thermal benefit there too, right? Because USB3 is so low power.

Manny Romero: [00:20:13] Yes, there’s definitely an advantage for the power consumption.

Winn Hardin: [00:20:17] All right. All right. So let’s pivot to what everyone is talking about these days. But we’ve been talking about it for 10 years. And I’m thinking of artificial intelligence of course and deep learning, right? So but it’s been a lot of the smoke and mirrors until I would say the last couple of years for me, when I’m actually seeing automation companies eat their own cooking, so to speak. Right? I mean, we’re not just selling AI or deep learning solutions out there. We’re actually using it ourselves to optimize our systems in a certain way. So let’s talk a little bit about Teledyne’s approach to DL.

Manny Romero: [00:20:49] Well, for AI, we obviously have solutions from a software perspective.

Winn Hardin: [00:20:55] The Astrocyte.

Manny Romero: [00:20:56] That’s the Astrocyte. Astrocyte is our tool within the processing or Sherlock. So Astrocyte allow to the customer to make his model in a very easy manner. So we understood how there was a complexity in tweaking the parameter to get good results. So we made sure that we would find a way that can automate, generate the model very rapidly and efficiently.

Winn Hardin: [00:21:25] With high accuracy.

Manny Romero: [00:21:26] With high accuracy. And so that’s one portion. Now, as you highlighted, AI is not everything. It coexists with the traditional function.

Winn Hardin: [00:21:42] And honestly have you ever heard of an application that was ever solved solely from using artificial intelligence or deep learning software, that didn’t need some kind of fiducial fine edge, fine something?

Manny Romero: [00:21:55] There is, but they’re very basic, they’re very basic. But if you want to do something a little bit more fancy or smarter, then that’s where your AI will be your co-process to whatever you end up doing.

Winn Hardin: [00:22:15] A hybrid.

Manny Romero: [00:22:16] A hybrid solution. I know that from a Teledyne perspective, there was an initiative, and you might have seen it on the booth at the show today, which was our, we have actually a camera that basically it’s designed to enable customers to offer a solution. But it’s really more, I would say a . . .

Winn Hardin: [00:22:42] Is this a smart camera platform?

Manny Romero: [00:22:43] Well, actually, we have the smart platform, which is called the BOA, and that is more oriented for I would say a lower-performance type of application the way it can process. The IVS product that we have, it’s targeted to work with customers to tweak and customize the solution for the customer. So it does include whatever sensor that you need, whatever AI person need.

Manny Romero: [00:23:18] Better platform, better product.

Manny Romero: [00:23:19] Yes, absolutely. So that I do expect to see more and more with the year to come, obviously.

Winn Hardin: [00:23:27] Okay. So we mentioned the Astrocyte making things simple, right? And let’s go deeper into that, right? Does it help us to identify outlying data that’s undermining our model efficiency?

Manny Romero: [00:23:40] It does.

Winn Hardin: [00:23:41] And I think you said something about its ability to grow.

Manny Romero: [00:23:44] Well, yes. It’s not the Astrocyte that does that part. The Astrocyte is really to create the model. But within our platform,  we added capability that when you add actually an online system and you find a new defect, that we’re not already trained, instead of going back to the drawing board and retrain everything, we added capability that we can update that model so that on the inline system, you’re not going down, you’re just updating it and it keeps going. That’s important. The other part about AI that we’ve been hearing a lot lately, like we talk about those buzzwords, going in the cloud, right? So many of our customers, they’re not ready to share data and losing control over it. So they want to make sure that it’s at the edge or it’s within their control. That’s a portion that I don’t know at which moment or if ever will have customers really ready to really upload.

Winn Hardin: [00:24:55] Yeah. You know, and I have the same questions and I need to talk with Ericsson and others about 5G. I mean, just in terms of security. Although, it was interesting, I was speaking with Omron earlier today about how they’re building cybersecurity systems into their Sysmac platform. Right? So I mean the industrial world is waking up to the dangers that are out there. And so all we’re talking about is liability risk, right, and companies trying to protect themselves and their customers, make sure they’re building quality products. If I could follow back up on the AI, when you were saying it learns new defects, how much operator involvement is required?

Manny Romero: [00:25:33] Very little. The second that they notice that something that was absolutely unforeseen is seen, they’re able to revalidate the model with Astrocyte, and that generates I would say an update file that can be used.

Winn Hardin: [00:25:54] Across the network. Would have to be individual inspection nodes. That is so cool. That is so cool. Okay. So I’m out of questions, but I want to just throw the big open question, right? So Manny, if you look out here, we’re at Automate 2023 right now, and everything that’s cool and nifty is on that show floor right now. But what excites you most? I know it’s not a fair question.

Manny Romero: [00:26:15] That is not a fair question.

Winn Hardin: [00:26:17] And you’re not allowed to say like the jazz band down at the network.

Manny Romero: [00:26:23] You know, one of the things why I’ve been in that industry for the last 25 years is exactly about that, that evolution of the technology. It keeps moving. It’s super amazing what kind of new problems rise from customers. And in every field you can think of, from recycling, feed sorting, PC inspection, you name it, people are pushing the boundary on every angle, and to work with a bunch of amazing guys that keep pushing that boundary and you know sometime they’re a partner, sometime they’re a competitor. It’s a big industry and it’s a small one at the same time, right, so from the decision making, new idea that’s coming out or a new sensor technology or new type of wavelength. Every little piece, all of those products that got together over the last 25 years is just been seriously amazing. But you know passion is basically what is driving us. And same thing with the many people that I work with. We’ve been working for many years together. We have new bloods that come in, come up with new ideas. But that passion is still there. And that’s really what fundamentally is excitable about that.

Winn Hardin: [00:27:56] I swear the last two years seem like a tornado or something has hit our industry in terms of automation technology, the amount of change, the fulfilled promises that are coming true. When you think about AMR and you think about all the autonomous things, they’re no longer just in the R&D and in the conference tracks, but they’re right over there on the show floor. And I’m wondering this: Are we truly seeing an exponential growth and acceleration in technology or are we just coming off a two-year COVID hangover? And it’s like all your favorite bands who wrote two or three albums during COVID and then suddenly they release them all at one time? Is this a Taylor Swift moment or is this nope, the world’s just changed. We flipped a switch.

Manny Romero: [00:28:45] Yeah, definitely. I mean, people are more, you know, everybody care more about environment today.

Winn Hardin: [00:28:55] Don’t have a lot of choices on that.

Manny Romero: [00:28:56] So the growth of that electric battery. Be green, recycling, making sure we’re not wasting. Food sorting, right? Making sure that there is no waste. Grade one, grade two product or recycled to make juice. So we’re not wasting anything. So I guess there’s more consciousness about all those elements being important. And that’s why we’re right in the middle of it.

Winn Hardin: [00:29:32] I think you might have identified something really important there. Like in the past, automation has been driven by, we want to make people’s work lives better. We want to help our customers to build products cheaper, faster, safer. And right now, those traditional drivers have been joined by a real push, because the world is so fast changing. When you talk about going to electrification of vehicles, I mean, this is a sea change. And this is something we’ve all talked about for 40, 50 years. Then throw in autonomous vehicles. It’s not going to be too long before we’ve got that figured too out across the vast majority of our roadways, which brings in so many different imaging modalities, both on the safety side, on the navigation side. I mean, it seems like applications are just exploding, and they’re outside of the factory. One coolest things for us. It’s almost like a payoff later in our careers. We’ve been stuck in industrial environments for so many years, and now our technology is everywhere.

Manny Romero: [00:30:38] Yes. Agriculture.

Winn Hardin: [00:30:40] Which is something I’m super excited about, actually, autonomous harvesting and the ability, much more optimization for pesticide placement and, again, resource optimization.

Manny Romero: [00:30:50] I even seen a going back to a more mechanical method. Not using more pesticides but going back to mechanical chopping. So you need AI to start looking at those plants to identify which one is the right one and which one is not the right one. Make sure you cut the right plant. Because obviously if you’re going with mechanical, let’s call them scissors, it’s again, prevent more pesticide back into the land.

Winn Hardin: [00:31:26] Which would be better for all of us, that’s for sure, that’s fantastic. Manny, I want to thank you so much for taking a little bit of your time out of one of the busiest days of Automate this year to talk. I mean, this is one of the more insightful episodes that I’ve been in, and I’ve really enjoyed it to the max. So thank you so much for coming here today.

Manny Romero: [00:31:42] It was great.

Winn Hardin: [00:31:44] Thanks, Manny. So this brings to an end our latest episode of Manufacturing Matters. I hope you liked it. Feel free to watch it on your favorite podcast platform. If you want to join us someday, shoot us a quick note through social media or at Manufacturing-Matters.com. And until next time, just remember that wherever you are around the world, manufacturing still matters. Till next time.