WEBVTT - The Future of Technology and Manufacturing 0:00:04.320 --> 0:00:07.160 Take a second to think about every single item in 0:00:07.200 --> 0:00:12.800 your home. Your television, your refrigerator, your desk, lamp, your laptop, 0:00:13.400 --> 0:00:15.520 even the smartphone you might be using to hear my 0:00:15.640 --> 0:00:19.360 voice right now. All of these things, and so many 0:00:19.400 --> 0:00:23.919 more items in our lives, began in a factory. There 0:00:23.960 --> 0:00:27.360 are more than six hundred and twenty thousand manufacturing businesses 0:00:27.640 --> 0:00:31.080 in the United States right now, responsible for nearly twelve 0:00:31.120 --> 0:00:35.040 percent of the total US economic output. The numbers are 0:00:35.080 --> 0:00:38.160 even more staggering in China, which makes up nearly twenty 0:00:38.240 --> 0:00:43.000 nine percent of the total global output. For manufacturing. Factories 0:00:43.000 --> 0:00:46.240 have been around since the late eighteenth century, and today 0:00:46.520 --> 0:00:50.199 they're used everywhere from South Korea to southern California to 0:00:50.280 --> 0:00:55.960 make cars, airplanes, textiles, and even space vehicles, and each 0:00:56.040 --> 0:01:00.920 one depends on a carefully choreographed system of steps, each 0:01:00.960 --> 0:01:04.360 one as essential as the next before the final product 0:01:04.640 --> 0:01:09.360 rolls off the production line. Mistakes, however, are also an 0:01:09.400 --> 0:01:14.279 unavoidable part of this process. Manufacturers simply can't check every 0:01:14.360 --> 0:01:18.120 piece of every product, and it's nearly impossible to achieve 0:01:18.120 --> 0:01:22.479 perfection when some manufacturing plants produce thousands of items a day. 0:01:23.720 --> 0:01:26.640 So how can technology help an industry so crucial to 0:01:26.680 --> 0:01:30.319 our daily lives, how can factories use AI to reduce 0:01:30.680 --> 0:01:38.119 and even prevent defective products? Welcome to Technically Speaking, an 0:01:38.120 --> 0:01:43.119 Intel podcast produced by iHeartMedia's Ruby Studio in partnership with Intel. 0:01:43.760 --> 0:01:47.480 In every episode, we explore how AI innovations are changing 0:01:47.480 --> 0:01:51.880 the world and revolutionizing the way we live. Hey there, 0:01:52.080 --> 0:01:55.280 I'm gram class, and today we're headed into the world 0:01:55.320 --> 0:01:59.480 of manufacturing, an expansive and essential industry that drives the 0:01:59.520 --> 0:02:03.120 global economy and both the history dating back nearly two 0:02:03.200 --> 0:02:06.840 hundred and fifty years, we've seen manufacturing create a revolution, 0:02:07.400 --> 0:02:11.520 resurrect nation's economies, connect people around the globe, and even 0:02:11.560 --> 0:02:16.200 send mankind into space. But what's next at the intersection 0:02:16.280 --> 0:02:20.840 of manufacturing and technology. In this episode, we'll be focusing 0:02:20.880 --> 0:02:24.680 on how AI technology can help optimize manufacturing and improve 0:02:24.760 --> 0:02:27.600 quality thanks to no small part to the minds at 0:02:27.600 --> 0:02:31.360 Intel and at Eigen Innovations, a company committed to helping 0:02:31.440 --> 0:02:35.919 organizations unlock the power of machine vision to automate quality inspections. 0:02:36.800 --> 0:02:41.680 Before we go any further, let's welcome our guest joining 0:02:41.720 --> 0:02:44.520 us today is John Weiss, the chief revenue officer at 0:02:44.520 --> 0:02:48.799 Eigen innovations. John oversees all revenue generation activities at Eigen, 0:02:49.080 --> 0:02:53.440 including driving sales in Eigen's machine vision software and engineering services. 0:02:53.840 --> 0:02:56.079 Welcome to the show, John, thanks for having me. Graham's 0:02:56.120 --> 0:02:56.800 great to be here. 0:03:00.760 --> 0:03:03.160 Let's start with a bit of background on manufacturing and 0:03:03.400 --> 0:03:05.799 the role it plays in our society. I mean it's 0:03:05.800 --> 0:03:08.240 fair to say that I phone, our car, laptop, even 0:03:08.280 --> 0:03:11.840 the food we eat involves some sort of manufacturing process. 0:03:12.480 --> 0:03:14.960 I'd like to get your thoughts on just the importance 0:03:14.960 --> 0:03:17.400 and scale of manufacturing plants around the world. 0:03:17.760 --> 0:03:20.720 Yeah, sure, Well, like you said, just about everything in 0:03:20.760 --> 0:03:25.480 our daily lives comes from factories or plants. But sure, 0:03:25.600 --> 0:03:29.280 depending on if you commute on a train or in 0:03:29.320 --> 0:03:32.400 a car, lots of those components are coming from factories. 0:03:32.600 --> 0:03:35.440 Very little these days are really kind of hand crafted 0:03:35.520 --> 0:03:40.160 and handmade and smile batch, especially large scale consumer items. 0:03:40.160 --> 0:03:44.120 And there's many different types of processes and many different 0:03:44.120 --> 0:03:45.839 types of ways things are made. 0:03:46.640 --> 0:03:50.480 And look, I know there's a multitude of ways and 0:03:50.800 --> 0:03:55.120 types of manufacturing processes. Like a Volkswagen built in Germany 0:03:55.160 --> 0:03:56.880 is going to be very different from an iPhone built 0:03:56.880 --> 0:04:01.680 in China. Do you find common things, reads or similarities 0:04:02.160 --> 0:04:04.200 across manufacturing industries. 0:04:05.000 --> 0:04:07.560 Yeah, definitely, so, I guess maybe as a kind of 0:04:07.560 --> 0:04:10.040 a starting point, it's important to understand there are two 0:04:10.200 --> 0:04:14.920 types of manufacturing processes or approaches. One is called process manufacturing. 0:04:15.320 --> 0:04:19.360 This is things like chemicals, plastics, things that can't really 0:04:19.440 --> 0:04:24.960 be broken down or deconstructed easily. Or you have discrete manufacturing, 0:04:25.000 --> 0:04:27.560 which is much more of the process of putting stuff together. 0:04:27.640 --> 0:04:30.720 Think about a watch or a car. Now, both of 0:04:30.760 --> 0:04:35.160 those processes, discrete and process manufacturing, they're quite different, but 0:04:35.720 --> 0:04:38.920 there are certainly similarities. And between the two methods, you 0:04:39.040 --> 0:04:41.560 basically have all of the things that we use every day, 0:04:41.640 --> 0:04:44.640 right then oftentimes actually they kind of bleed into one another. 0:04:44.720 --> 0:04:47.159 Most things have a little bit of process manufacturing involved 0:04:47.200 --> 0:04:50.640 and then a little bit of discrete manufacturing as well. However, 0:04:51.120 --> 0:04:54.279 I would say the commonalities across both are really heavily 0:04:54.320 --> 0:04:57.719 reliant on technology. We see a very large push for 0:04:58.000 --> 0:05:03.120 data driven decision making. We see large patterns or trends 0:05:03.160 --> 0:05:07.760 in both realms of manufacturing around empowering the workforce, trying 0:05:07.760 --> 0:05:11.479 to opt skill workers via technology to get them to 0:05:11.520 --> 0:05:15.799 be focused on more mission critical tasks or higher value 0:05:15.839 --> 0:05:18.920 activities while letting some of the technology do more of 0:05:18.960 --> 0:05:20.480 the mundane tasks. 0:05:21.120 --> 0:05:23.240 Yeah. In a previous job that I had, we were 0:05:23.279 --> 0:05:27.960 doing consumer electronics, and we struggled quite a bit with 0:05:28.080 --> 0:05:31.120 the quality side of things and being able to ensure 0:05:31.560 --> 0:05:34.839 a good product can you manufactured in our China plant. 0:05:35.480 --> 0:05:39.000 And one thing that struck me was that there was 0:05:39.000 --> 0:05:41.960 a very manual process in terms of the quality inspection, 0:05:42.360 --> 0:05:44.680 and it will take samples in one out of every 0:05:44.720 --> 0:05:47.039 ten and the test that and then if that worked, 0:05:47.040 --> 0:05:49.599 then okay, and then we assume the rest kind of work. 0:05:50.080 --> 0:05:53.480 I'm wondering if you could share any stories or examples 0:05:53.520 --> 0:05:57.560 of I guess problems with quality or defective products that 0:05:57.600 --> 0:05:58.480 stick in your mind. 0:05:59.520 --> 0:06:01.920 Yeahsolutely. I mean that's all we do at Eigen, right. 0:06:01.960 --> 0:06:05.839 All we do is industrial machine vision for inline quality inspection. 0:06:06.120 --> 0:06:08.720 So a couple that stick into my mind. Actually very 0:06:08.760 --> 0:06:11.800 relevant to what you said, sample testing. Lots of manufacturers 0:06:11.839 --> 0:06:14.440 do this, right if they're a high volume shop or 0:06:14.440 --> 0:06:17.719 a high volume process. For example, we have some customers 0:06:17.760 --> 0:06:20.599 that use our technology to inspect upwards of forty thousand 0:06:20.680 --> 0:06:23.320 units a week per facility. The challenge is if you 0:06:23.360 --> 0:06:25.680 do find a problem, now you're kind of scratching your 0:06:25.680 --> 0:06:28.600 head wondering how many in between the last one hundred 0:06:28.680 --> 0:06:32.920 or last fifty also had a problem right, And unfortunately 0:06:33.000 --> 0:06:35.240 you tend to find out the hard way when you 0:06:35.279 --> 0:06:38.800 get returns or warranty claims that maybe something wasn't right 0:06:38.880 --> 0:06:42.080 in that process. And so technology is a great way, 0:06:42.160 --> 0:06:45.640 especially the arena that we operate in computer vision, we're 0:06:45.720 --> 0:06:48.920 helping customers actually get away from that. A great example 0:06:49.120 --> 0:06:52.240 is one of our manufacturing customers who makes fuel tanks 0:06:52.320 --> 0:06:56.360 for a variety of different vehicles, and they do what's 0:06:56.400 --> 0:06:59.640 called destructive testing. They don't just test, they actually break 0:06:59.680 --> 0:07:02.240 the fuel tank, that cut it up, and they look 0:07:02.240 --> 0:07:04.760 at all of the plastic components inside and they see 0:07:04.960 --> 0:07:08.360 was it molded correctly, was it welded correctly? And if 0:07:08.400 --> 0:07:10.560 they have a problem, well, now they have to reverse 0:07:10.560 --> 0:07:12.720 engineer a whole bunch of stuff and try to figure out, 0:07:12.800 --> 0:07:14.920 holy cow, what went wrong right? And how do we 0:07:15.000 --> 0:07:17.559 ensure that no bad fuel tank gets on a truck. 0:07:18.080 --> 0:07:20.200 They started the journey with us about three years ago, 0:07:20.240 --> 0:07:22.800 and fast forward today, we're builds back on every new 0:07:22.880 --> 0:07:26.000 machine that gets put into those plants for fuel tank inspection. 0:07:26.600 --> 0:07:30.440 So they know unequivocally, every single product that they ship 0:07:30.480 --> 0:07:32.800 out the door is of the highest quality standard and 0:07:32.840 --> 0:07:36.040 if it's not. If something happens, now they have complete 0:07:36.040 --> 0:07:38.920 traceability on everything they've made, so they can figure out 0:07:38.960 --> 0:07:41.200 exactly what went wrong in the process. 0:07:42.960 --> 0:07:45.400 What John is talking about here is the output of 0:07:45.440 --> 0:07:49.040 the manufacturing process. How can we ensure every fuel tank 0:07:49.120 --> 0:07:53.000 that leaves the plant will work as designed? Just as importantly, 0:07:53.360 --> 0:07:56.000 we need to consider the quality of the input components. 0:07:56.400 --> 0:07:59.520 Everything from the greater steel to the precision of the 0:07:59.560 --> 0:08:03.120 fuel gate, These need to be expected to ensure that 0:08:03.160 --> 0:08:07.000 these are up to the manufacturer standard. I asked John 0:08:07.280 --> 0:08:08.280 for his thoughts about this. 0:08:11.720 --> 0:08:15.200 We don't often look at raw material although it's possible 0:08:15.240 --> 0:08:18.680 in some cases, but more often than not our inputs 0:08:18.720 --> 0:08:22.200 that we're looking at it's actually process inputs or parameters. 0:08:22.360 --> 0:08:25.960 So we're looking at feed rates of raw materials, temperatures 0:08:26.000 --> 0:08:29.080 of raw materials, things like this. In the process that 0:08:29.320 --> 0:08:32.200 become more of a scientific look of what's happening on 0:08:32.240 --> 0:08:36.480 the assembly line and ensuring that everything is inspect We 0:08:36.520 --> 0:08:39.040 don't just look at the output of you know, did 0:08:39.080 --> 0:08:41.120 you make a good or bad product? But we'll actually 0:08:41.120 --> 0:08:43.600 show you all of the process data that went into 0:08:43.640 --> 0:08:47.000 making that product. The other side is on the discrete 0:08:47.040 --> 0:08:50.880 world where you're actually assembling things. In this instance, what 0:08:50.920 --> 0:08:54.160 we do is we'll actually monitor the assembly, so we'll 0:08:54.200 --> 0:08:58.240 look at how people are placing door panels into a 0:08:58.320 --> 0:09:02.160 doorframe for example, on to motif asset, or look at 0:09:02.200 --> 0:09:05.559 tail lamps for lighting purposes right the way that they're 0:09:05.679 --> 0:09:08.679 assembled and put together. And what we can do in 0:09:08.720 --> 0:09:11.400 real time is tell folks, hey, what you're putting together 0:09:11.559 --> 0:09:14.840 is misconfigured, or it's missing components, or it has too 0:09:14.880 --> 0:09:18.240 many components. Those are defect types that are pretty common 0:09:18.240 --> 0:09:19.360 in the assembly world. 0:09:20.280 --> 0:09:23.160 And what are some of the technology that is used 0:09:23.160 --> 0:09:25.880 for that? Is it vision? Is it sensors? Is it combination? 0:09:26.800 --> 0:09:30.360 Everything we do is vision based, so we don't make cameras. 0:09:30.400 --> 0:09:32.560 By the way, we are a software provider, we also 0:09:32.679 --> 0:09:34.720 act as a system integrator, so a large part of 0:09:34.720 --> 0:09:39.319 our business is actually delivering turnkey solutions, not just the software. 0:09:39.679 --> 0:09:42.680 But we don't make hardware, which is actually really cool 0:09:42.720 --> 0:09:45.040 for us because that means we get to use tons 0:09:45.040 --> 0:09:47.800 of different types of options that are available for our 0:09:47.840 --> 0:09:52.160 customers and it helps us really find the perfect design 0:09:52.360 --> 0:09:56.320 and configuration that is definitely going to solve problems, and 0:09:56.360 --> 0:09:58.679 so having the flexibility is really nice, and of course 0:09:58.760 --> 0:10:02.520 that's a large and why we partner with Intel. We're 0:10:02.559 --> 0:10:04.880 built on the open Veno tech stack, and that means 0:10:04.960 --> 0:10:08.719 we can run our software really on any device that 0:10:09.000 --> 0:10:12.439 leverages an Intel chip, which gives us tons of options 0:10:12.600 --> 0:10:15.319 for deployments. What's really cool about this though, from a 0:10:15.400 --> 0:10:18.120 quality perspective, is that it means you now have one 0:10:18.240 --> 0:10:21.679 vision system that can integrate with different types of sensors. 0:10:22.120 --> 0:10:25.600 So if you want to do, say an optical inspection 0:10:25.880 --> 0:10:29.240 for surface defects like scratches and dents, but you also 0:10:29.360 --> 0:10:32.599 want to look at perhaps inside that product in a 0:10:32.679 --> 0:10:35.840 thermal application, if it's a molded part or something like that, 0:10:36.160 --> 0:10:38.480 Well you can look at all those different types of 0:10:38.520 --> 0:10:41.960 sensor in one easier to use screen, right, So it 0:10:42.000 --> 0:10:44.559 removes the headache of having to have five six different 0:10:44.640 --> 0:10:47.000 vision systems to do a variety of inspections. 0:10:47.520 --> 0:10:51.280 And I'm also interested in the deployment of these sorts 0:10:51.280 --> 0:10:54.080 of new technologies. I'd like to get your thoughts and 0:10:54.160 --> 0:10:57.200 experiences around what's some of the tips and tricks for 0:10:57.240 --> 0:11:01.080 people out there trying to deploy not just for manufacturing quality, 0:11:01.080 --> 0:11:04.880 but technology and AI in general into a workforce that 0:11:05.000 --> 0:11:07.120 maybe is a little bit hesitant. 0:11:07.520 --> 0:11:10.320 Yeah, humans don't like change, that's for sure. I know 0:11:10.360 --> 0:11:13.680 I don't. I'm guilty of that. And it's certainly like 0:11:13.760 --> 0:11:15.960 that when you go into a factory and you've got 0:11:16.000 --> 0:11:19.200 folks that have been on the same line or in 0:11:19.280 --> 0:11:24.200 the same steel plant for twenty five thirty years, and 0:11:24.520 --> 0:11:26.640 you show up and you've got this bright, new shiny 0:11:26.679 --> 0:11:29.280 software and you say, hey, don't worry, data is going 0:11:29.320 --> 0:11:33.320 to solve everything. Naturally, people can be quite apprehensive. We 0:11:33.360 --> 0:11:37.480 don't often run into technology challenges anymore now, it's really 0:11:37.640 --> 0:11:41.680 we run into people challenges and organizational challenges. So first 0:11:41.720 --> 0:11:43.560 and foremost, I'll give the advice that I give on 0:11:43.760 --> 0:11:45.600 most of the times I'm asked this question. But it's 0:11:45.600 --> 0:11:49.960 so true. Is you don't ever start adopting technology just 0:11:50.080 --> 0:11:52.960 for the sake of adopting it, just because competitors are 0:11:53.040 --> 0:11:55.800 using something, or just because somebody way up the chain says, hey, 0:11:55.800 --> 0:11:58.240 we need an AI strategy. Go invest in AI boom, 0:11:58.679 --> 0:12:01.680 spend some time and really think about the problems that 0:12:01.720 --> 0:12:04.720 you're trying to tackle in my world, in the quality world, 0:12:04.800 --> 0:12:07.560 in manufacturing, it's looking at things you can do to 0:12:07.679 --> 0:12:12.880 increase yields, increase your throughput, reduce your waste, reduce your rework, 0:12:13.160 --> 0:12:17.160 and ultimately lower what's called the cost of quality. Start 0:12:17.200 --> 0:12:20.359 with that, find a way that you can or process 0:12:20.440 --> 0:12:24.199 that you can optimize by using some of this newer technology, 0:12:24.200 --> 0:12:26.720 and then of course do a cost assessment or a 0:12:26.760 --> 0:12:29.920 return on your investment analysis, and ensure that the business 0:12:30.040 --> 0:12:33.400 justification is there. My experience, that's where a lot of 0:12:33.440 --> 0:12:36.079 these projects fall short, and where folks get stuck in 0:12:36.120 --> 0:12:39.520 these pilots and pocs is because they get really excited 0:12:39.559 --> 0:12:43.600 to try something, but there is no proven business value 0:12:43.679 --> 0:12:47.440 or business justification behind it, and naturally then you don't 0:12:47.440 --> 0:12:50.240 get the executive sponsorship you need, your budget falls through, 0:12:50.320 --> 0:12:51.680 and the project goes nowhere. 0:12:52.240 --> 0:12:56.280 And in your experience, what industries do you find actually 0:12:56.800 --> 0:12:59.240 a little bit more advanced in terms of adopting these 0:12:59.320 --> 0:13:02.520 new technology both on a technical level but also at 0:13:02.520 --> 0:13:05.600 an organizational level that it seems like the teams are 0:13:05.600 --> 0:13:09.839 actually involved and successfully deploying these sorts of techniques. 0:13:10.480 --> 0:13:13.920 Yeah, that's a great question. We see pretty advanced deployments 0:13:13.920 --> 0:13:17.160 in the automotive world as far as discrete manufacturing goes, 0:13:17.200 --> 0:13:19.760 they tend to be far ahead of the curve compared 0:13:19.800 --> 0:13:24.400 to say, steel manufacturers or something like that, or concrete manufacturers. 0:13:24.880 --> 0:13:27.640 There's a lot of very advanced technology and those automotive 0:13:27.640 --> 0:13:31.560 facilities that make sure what you buy is actually perfect. Similarly, 0:13:31.600 --> 0:13:34.920 in the process world, pharmaceuticals tends to be on the 0:13:34.960 --> 0:13:38.040 continuous process side that tends to be pretty advanced. They 0:13:38.080 --> 0:13:40.440 have a lot of vision systems in place looking at 0:13:40.920 --> 0:13:44.680 the vaccine vials to ensure the integrity of vile caps 0:13:44.760 --> 0:13:48.160 and seals and things like that. Some of the laggers 0:13:48.160 --> 0:13:52.800 would be metals, some of the plastics organizations. But there's 0:13:52.840 --> 0:13:55.320 also a kind of a bigger dynamic in manufacturing that 0:13:55.360 --> 0:13:58.760 I think folks don't really understand that also contributes to 0:13:58.840 --> 0:14:03.000 who's advanced and who's which is the sheer size of 0:14:03.040 --> 0:14:07.240 these organizations, right, Manufacturers are not all large. Folks tend 0:14:07.320 --> 0:14:10.320 to think about John Deere and three M and you know, 0:14:10.400 --> 0:14:13.560 the largest players in the world, and the reality is 0:14:13.600 --> 0:14:18.680 that makes up such a small fraction of the manufacturing pool, 0:14:18.920 --> 0:14:22.800 especially in America, most manufacturing facilities have you know, twenty 0:14:22.840 --> 0:14:27.160 people or less. Small to medium manufacturers anywhere from say 0:14:27.200 --> 0:14:29.840 like the twenty to two hundred range of employees. That's 0:14:29.840 --> 0:14:33.160 who makes up the vast majority of our products. Even 0:14:33.200 --> 0:14:35.160 when you buy something really big, you know, whether it's 0:14:35.160 --> 0:14:38.160 a whirlpool dishwasher or a hot tub or whatever it 0:14:38.240 --> 0:14:41.760 might be, all those little components that make up that 0:14:41.920 --> 0:14:45.400 consumer good, Well, it came from probably many different suppliers, 0:14:45.440 --> 0:14:46.560 and most of those are small. 0:14:47.200 --> 0:14:49.520 It's nice that you mentioned that because my father has 0:14:49.560 --> 0:14:54.000 a small manufacturing facility here. And just to talk a 0:14:54.040 --> 0:14:57.120 little bit more of the technology stack that you're using 0:14:57.120 --> 0:15:00.840 with open Vino and Intel's edge devices. I'm really interested 0:15:00.840 --> 0:15:03.920 to see how some of the smaller guys can actually 0:15:04.040 --> 0:15:07.080 use this sort of technology so that it can actually 0:15:07.120 --> 0:15:08.160 be more competitive. 0:15:08.880 --> 0:15:12.080 Sure, well, leveraging open Veno helps us have a real 0:15:12.160 --> 0:15:14.840 wide range of how on the hardware side, how we 0:15:14.880 --> 0:15:20.080 can install our software. What that means for smaller manufacturers 0:15:20.160 --> 0:15:23.280 is that we can be quite flexible in the design 0:15:23.320 --> 0:15:26.880 of a system and can accommodate just about any budget, 0:15:27.080 --> 0:15:30.560 which that alone is pretty significant to understand. I still 0:15:30.600 --> 0:15:33.920 think there's a misconception that it's too expensive or too 0:15:34.000 --> 0:15:36.840 cumbersome for the little guys, so to speak to really 0:15:36.920 --> 0:15:40.320 innovate in their plants, and it's simply not true. You know, 0:15:40.360 --> 0:15:43.480 we have customers that make as little as twenty parts 0:15:43.480 --> 0:15:47.840 of shift, and even for them, having the flexibility of 0:15:47.880 --> 0:15:50.440 how we design and configure these systems, it ensures that 0:15:50.600 --> 0:15:54.000 even they can embrace newer technology and provide the highest 0:15:54.000 --> 0:15:55.720 amounts of quality to their customers. 0:15:57.840 --> 0:16:00.000 Part of the reason I can can design and configure 0:16:00.080 --> 0:16:03.520 of those systems is because the company uses Intel's central 0:16:03.520 --> 0:16:08.520 processing units or CPUs, as opposed to GPUs or graphics 0:16:08.640 --> 0:16:14.040 processing units. GPUs are specialized processes are originedly designed to 0:16:14.120 --> 0:16:18.320 accelerate graphics rendering. The key difference in the manufacturing world 0:16:18.600 --> 0:16:22.120 is that CPUs, like the ones Intel provides for Ogen, 0:16:22.600 --> 0:16:26.360 are able to perform under harsher or hotter conditions like 0:16:26.400 --> 0:16:31.320 the ones you might find in a factory or manufacturing plant. GPUs, meanwhile, 0:16:31.320 --> 0:16:33.720 are prone to overhitting without the use of a fan 0:16:33.840 --> 0:16:36.880 to cool it down, and most factories won't use fans 0:16:37.280 --> 0:16:40.920 so they can avoid spreading dust and debris. There's always 0:16:40.960 --> 0:16:44.280 a trade off between designing software optimized for CPUs or 0:16:44.360 --> 0:16:47.960 GPUs and a manufacturing plant. I asked John about this, 0:16:48.360 --> 0:16:50.880 and I found his answer to be quite illuminating. 0:16:53.160 --> 0:16:55.720 It's always an interesting discussion when people ask, why don't 0:16:55.760 --> 0:16:58.720 you just go on GPUs and what's the real difference? 0:16:58.840 --> 0:17:04.159 And from a manufacturing perspective, just logically thinking about what 0:17:04.280 --> 0:17:07.720 happens in a plant. If you remember, like late nineties, 0:17:07.760 --> 0:17:10.200 you remember you had your COMPAC or your Gateway PC, 0:17:10.480 --> 0:17:13.440 this big old white box on the floor, and every 0:17:13.480 --> 0:17:15.480 so often you'd take the front panel off and it 0:17:15.520 --> 0:17:18.800 would just be totally caked in dust. Right, you'd hear 0:17:18.880 --> 0:17:22.480 the fans humming, and well, this is what happens to 0:17:22.560 --> 0:17:25.520 GPUs and factories. This is why we don't use fans, 0:17:25.720 --> 0:17:30.879 because factories are dirty. There's dust everywhere. And what we 0:17:31.000 --> 0:17:34.320 found is that when we explored using various types of 0:17:34.320 --> 0:17:36.720 mediums to do our processing, what we found is that 0:17:36.800 --> 0:17:42.240 fanless intel boxes were not only just as performant and 0:17:42.280 --> 0:17:44.960 in some instances probably even more beneficial to use, but 0:17:45.840 --> 0:17:48.919 on the maintenance side of it, we didn't have to 0:17:49.000 --> 0:17:52.919 worry about dirt and debris, which exists in every single 0:17:52.920 --> 0:17:55.600 plant that we deploy these in. We also didn't have 0:17:55.640 --> 0:17:59.879 to worry about heat. GPUs generate tons of heat. Had 0:18:00.080 --> 0:18:03.399 this discussion with somebody who did deploy GPUs in a 0:18:03.480 --> 0:18:06.320 manufacturing environment and they were looking at in tens of 0:18:06.359 --> 0:18:09.560 millions of dollars in HVAC improvements just to keep the 0:18:09.640 --> 0:18:13.800 factories cool enough to operate effectively. Right. And then the flexibility, 0:18:13.840 --> 0:18:17.639 like I mentioned, being able to very easily scale the 0:18:17.720 --> 0:18:20.880 hardware for more advanced use cases, if we need two 0:18:20.960 --> 0:18:23.439 or three different edge boxes, it's really easy to do, 0:18:24.240 --> 0:18:26.080 and also be able to scale down for the smaller 0:18:26.119 --> 0:18:28.160 applications where we want to make it a bit more 0:18:28.280 --> 0:18:30.840 cost effective for the smaller manufacturers as well. 0:18:33.280 --> 0:18:36.840 Coming up next on Technically Speaking and Intel Podcast. 0:18:37.000 --> 0:18:40.520 Computer vision specifically for quality is becoming more and more common. 0:18:40.560 --> 0:18:43.920 I think this will become completely commonplace over the next 0:18:44.000 --> 0:18:44.800 twelve years. 0:18:45.240 --> 0:18:47.679 We'll be right back after a brief message from our partner. 0:18:47.680 --> 0:18:48.240 Is that Intel? 0:18:56.680 --> 0:19:00.400 Welcome back to Technically Speaking an Intel podcast. I'm here 0:19:00.440 --> 0:19:06.439 now with John Weiss. I'd actually like to get you 0:19:06.480 --> 0:19:09.000 to talk a little bit about Eigen Innovations, if you 0:19:09.000 --> 0:19:11.640 could tell us a little bit about the company and 0:19:12.160 --> 0:19:12.720 its mission. 0:19:13.440 --> 0:19:17.040 Sure so, Iigen Innovations has been around for twelve years. 0:19:17.760 --> 0:19:21.120 We started in academia out of the University of New Brunswick. 0:19:21.160 --> 0:19:24.399 It was founded by a PhD student and a professor. 0:19:25.240 --> 0:19:27.760 We started as a system integrator, so we were going 0:19:27.800 --> 0:19:31.840 into factories actually installing vision systems, and over the course 0:19:31.840 --> 0:19:35.080 of about a decade, we developed our own software to 0:19:35.119 --> 0:19:38.600 make our job as a system integrator easier. And about 0:19:38.840 --> 0:19:40.439 I don't know, two and a half years ago or so, 0:19:40.520 --> 0:19:43.679 we realized there's actually a ton of value in IP 0:19:43.880 --> 0:19:47.480 and the software we created, and so we reinvented the 0:19:47.520 --> 0:19:50.480 company and moved away from leading as a system integrator 0:19:50.560 --> 0:19:54.200 to actually leading as a software SaaS based company. We 0:19:54.240 --> 0:19:57.440 really only do one thing. We do inline quality inspection 0:19:57.640 --> 0:20:01.720 and actually, to be more specific, our specialty thermal applications 0:20:01.760 --> 0:20:05.960 that leverage AI. So when you think of like injection molding, 0:20:06.040 --> 0:20:11.760 blow molding, metal welding, plastic welding, void detection, and plastic goods, 0:20:11.960 --> 0:20:15.000 anything that has a heated process that the human eye 0:20:15.040 --> 0:20:19.000 can't easily see defects, we do really really well there. 0:20:19.680 --> 0:20:22.720 And we talked a little bit about AI, and I 0:20:22.720 --> 0:20:26.920 think we've also talked about the software that utilizes machine vision. 0:20:27.800 --> 0:20:30.640 Where do you see AI models and the CPU based 0:20:30.800 --> 0:20:35.520 technology being able to compete with machine vision use cases? 0:20:36.160 --> 0:20:37.959 Yeah, it's a good question. Look, I think there are 0:20:38.000 --> 0:20:41.840 pros and cons of both approaches. We actually have not 0:20:42.119 --> 0:20:46.640 yet come across a project that we had any kind 0:20:46.640 --> 0:20:51.240 of processing limitation on being CPU based. We have applications 0:20:51.280 --> 0:20:56.560 in production running yet thirty inferences a second across cameras. Right, 0:20:57.040 --> 0:21:01.800 that's quite quite fast. There are definitely higher demand applications. 0:21:01.840 --> 0:21:04.560 But in our world of process in discrete manufacturing and 0:21:04.600 --> 0:21:08.960 the types of projects we typically focus on, speed has 0:21:09.000 --> 0:21:12.080 actually not been a problem for us with CPUs, even 0:21:12.119 --> 0:21:15.880 at quite aggressive speed. I see the tools getting easier 0:21:15.880 --> 0:21:19.280 and easier to use, more and more self service, if 0:21:19.280 --> 0:21:23.120 you will. Years ago, we had this phrase of democratizing 0:21:23.200 --> 0:21:25.440 data if you remember that, around the days of big data, 0:21:25.560 --> 0:21:28.239 kind of empowering everybody to be a data scientist, and 0:21:28.720 --> 0:21:32.439 I see the same movement happening in the AI world. 0:21:32.640 --> 0:21:34.879 In fact, actually we're a good example of that. You 0:21:34.920 --> 0:21:38.120 can use our tool to build deploy train models across 0:21:38.160 --> 0:21:40.959 factories and you don't have to touch a line of code. 0:21:41.200 --> 0:21:43.240 So I think that's the future. I think the tools 0:21:43.280 --> 0:21:46.240 get easier and easier to use, so that my good 0:21:46.240 --> 0:21:48.679 friend Jimmy, who's down in Texas at one of our 0:21:48.680 --> 0:21:51.400 customer plants, who's been in that same plant for over 0:21:51.480 --> 0:21:55.320 thirty years, that he can blow me away with how 0:21:55.320 --> 0:21:57.919 he can build a model that does thermal inspection on 0:21:57.960 --> 0:22:01.399 metal welding. And years ago, oh, somebody that didn't have 0:22:01.480 --> 0:22:04.399 that kind of training from a data science perspective or 0:22:04.400 --> 0:22:06.920 a programming perspective, they would never be able to do that. 0:22:07.040 --> 0:22:11.040 And today they're building dashboards and building models that are 0:22:11.520 --> 0:22:15.000 literally redefining the way these manufacturers operate. It's amazing. 0:22:16.800 --> 0:22:19.240 You heard John say earlier that eigen has been around 0:22:19.240 --> 0:22:21.760 for more than a decade and this technology has been 0:22:21.760 --> 0:22:26.040 implemented across a variety of manufacturing spaces to thermally inspect 0:22:26.080 --> 0:22:32.080 items like metal paper, cardboard, box adhesive, automotive windshields, and 0:22:32.160 --> 0:22:36.600 high glass plastics. With such a lengthy track record of achievements, 0:22:37.080 --> 0:22:40.520 John spoke about one specific company success story that stuck 0:22:40.560 --> 0:22:43.800 out for him. 0:22:44.000 --> 0:22:46.840 A couple that come to mind. I mentioned we inference 0:22:46.880 --> 0:22:49.840 about thirty images per second in this one process. This 0:22:49.960 --> 0:22:53.679 is a paper process, so it's continuous, very high speed, 0:22:54.160 --> 0:22:57.679 and it's for a high glass specialty paper. And what 0:22:57.800 --> 0:23:00.719 happens is this high glass coding goes on paper very 0:23:00.800 --> 0:23:04.440 rapidly as it's going down the line, and unfortunately there's 0:23:04.480 --> 0:23:07.040 a problem where this coding can build up and if 0:23:07.080 --> 0:23:09.439 it's not caught in about eight seconds, it will do 0:23:09.600 --> 0:23:11.960 roughly one hundred and twenty thousand dollars worth of damage 0:23:12.040 --> 0:23:15.000 to the equipment. This can happen multiple times as shift. 0:23:15.440 --> 0:23:18.159 This is a very expensive problem if it's not caught. 0:23:18.200 --> 0:23:21.520 And so this one's a great example of a thermal application. 0:23:21.560 --> 0:23:24.200