Conversations on Applied AI - Stories from Experts in Artificial Intelligence

Paul Denman - Where IoT and Industry 4.0 Is Today

March 29, 2022 Justin Grammens Season 2 Episode 5
Conversations on Applied AI - Stories from Experts in Artificial Intelligence
Paul Denman - Where IoT and Industry 4.0 Is Today
Show Notes Transcript

The conversation this week is with Paul Denman. Paul started his career as an engineer focused on new innovative technologies. He is passionate about technology and its integration to solve business problems and has a number of accomplishments in automation, AI, IoT medical device robotics, machine vision, and hard automation. As a member or advisor to corporate leadership, Paul is viewed as a key contributor and collaborator credited with identifying and securing missing links that have improved corporate growth, profitability, and new product sales. Most recently, Paul has been focused on robotics, automation, and machine learning using edge analytics for Industry 4.0. Paul holds a degree in Biomedical Engineering and a Bachelor of Science in Electrical and Electronics Engineering.

If you are interested in learning about how AI is being applied across multiple industries, be sure to join us at a future AppliedAI Monthly meetup and help support us so we can make future Emerging Technologies North non-profit events!

Resources and Topics Mentioned in this Episode

Enjoy!

Your host,
Justin Grammens


Paul Denman  0:00  

I'm just finishing up an article in design world where industry 4.0 is and what the major hang ups have been or slowness of its acceptance. One of the major issues is been white white hasn't caught on, of course, in the way we thought it would from four or five years ago when he collected this term 4.0 is primarily because of the return on investment. People don't want to spend the money on this lake and directly tie you know, predictive analytics or conviction to failure, digital modeling to you know, what will I really say, is worth the manpower and investment.


AI Announcer  0:41  

Welcome to the Conversations on Applied AI Podcast where Justin Grammens and the team at Emerging Technologies North talk with experts in the fields of artificial intelligence and deep learning. In each episode, we cut through the hype and dive into how these technologies are being applied to real world problems today. We hope that you find this episode educational and applicable to your industry and connect with us to learn more about our organization at applied ai.mn. Enjoy.


Justin Grammens  1:11  

Welcome everyone to the Conversations on Applied AI Podcast. Today we're talking with Paul Denman. Paul started his career as an engineer focused on new innovative technologies. He is passionate about technology and its integration to solve business problems and has a number of accomplishments and love automation, AI, IoT medical device robotics, machine vision, and hard automation. As a member or advisor to corporate leadership, Paul is viewed as a key contributor and collaborator credited with identifying and securing missing links that have improved corporate growth, profitability and new product sales. Most recently, Paul has been focused on robotics, automation and machine learning using edge analytics for industry 4.0. Paul holds a degree in Biomedical Engineering and a Bachelors of Science in Electrical and Electronics Engineering. Thanks, Paul, for being on the program today. 


Paul Denman  1:56  

Thank you. 


Justin Grammens  1:58  

Awesome, you know, I gave a short background, I guess kind of what drives you a lot about innovation and sort of applying those to business problems. But if you can give us a little bit more detail, and sort of how you got to where you are today. 


Paul Denman  2:09  

I'm from Colorado, went to Colorado University. My real passion was in life science. And so I worked in cardiology research for about five years at CU Medical Center, and got heavily steeped into microprocessor design, hardware design, that kind of thing. Later, Honeywell hired me to move to the nice, warm Minnesota. So I think it's, it's 21 below zero today. It's exactly, I got to be a hardy Minnesotan to live here. I know, right now we're all looking at where could we live, I got into more commercial design, electronic software development embedded software in and then moved into sort of a technology support area with a company called Western Digital at Irvine. And that was the turnaround situation. They got involved with some just wonderful champions, as far as business development goes. And we grew that company from about four and a half million to 640 million in five years. So wish they were all that good is that is that is that the same company that makes hard drives Western Digital, is we actually acquired tandem drives when we were there. But they're the original, if you've heard of the UART, or serial port on computers, they were the inventor of the first serial port chip, and good old Rs 432. And then they then they designed and manufactured a chip in a foundry for disk drive controllers. And that soon became the industry standard was in actually the IBM PC products and all those so part of the whole PC wave. And we're going to book a return for IBM's first PC release and a lot of fun and all over the world and for sure, yeah. But more recently, in the early 2000s, I got involved with the Human Genome Project and all the automation involved with that, and commercializing that technology. And that got me steeped into back into life science, but more or less medical device and more into sequencing, cycling genomic research, targeted vector genomic, those kind of things, and all the automation that goes with that. And in the last five years, I've focused primarily on the space of laboratory automation in in general, automation and everything that involves also edge analytics and industrial. I'm assuming that's maybe what words of conversation can go today. 


Justin Grammens  4:47  

Sure, I mean, it's all about data at the end of the day. And so how can we apply some of these new things like deep learning machine learning? AI technology is a broad term, I guess, to all this information that we're now getting from these sensors and all this all this crazy stuff. And it sounds like you know, you've you started at sort of the hardware level, right? You're talking about microprocessor design. Those are some of the concepts and stuff and then Industry 4.0 feels to me like it's just a really, really broad term. It's all around. Like you're talking about edge analytics, automation, improving efficiency, all those types of things, which is where you're focused on today.


Paul Denman  5:20  

Yeah, I'm just finishing up an article in design world, on where industry 4.0 is and what the major hang ups of banners slowness of its acceptance, one of the major issues has been, why it hasn't caught on, of course, in the way we thought it would from four or five years ago, when he collected this term. 4.0 is primarily because of the return on investment. People don't want to spend the money in this lake and directly tie, you know, predictive analytics or prediction to failure, digital modeling to what will it really say? Is it worth the manpower and investment? I think the major niche that this has found recently is word is a pay to be able to predict a failure ahead of time to prevent it. Okay. It's down to like seconds. And you know, I talked to the VP of manufacturing for Ford last year, and he had mentioned that there had this huge conversion on assembly plants to go to electric actuators, rather than pneumatic or hydraulic. And the reason for that is one actuator failing on a line for every minute, you know, a cars are averaging somewhere around 50 to 55 seconds, a new car comes off the line. What does the car cost today? You know, 40 50,000?


Justin Grammens  6:46  

Yeah, sure. 


Paul Denman  6:47  

You're losing about $50,000 Every minute of an hour downtime is gonna cost you a lot of money, right? Yeah, for sure. So you know, the payback there is obvious. So there is a big trend toward that. And in staying in that category of predictive edge kind of analytics. The other big niche right now is moved into large motor applications and the 20 to 200 horsepower size. So these would be like an area of pumps, city waterworks, things that when it's down, it has a major impact on society or products. So it's not moved into the smaller motors and devices that yeah, they're assembling a pacemaker, or they're manufacturing a catheter, or liquid dispensing of my inner microplate. But because the payback just isn't there, you know, the impact isn't big enough. Right now, you will get there, the technology is there already. But the emphasis is on the bigger stuff right now.


Justin Grammens  7:56  

Sure. Now, that makes sense on sort of the bang for the buck. I mean, I've been in sort of this IoT realm for at least a decade now, here. And I feel the same way with regards to it has really slowed, right, there were a lot of things I was talking about, you know, 567 years ago, you know, the whole world is gonna be connected, all this data is gonna come off all these things, everything, you know, from your house, to, like you said, Smart City, smart factories, all that type of stuff. And it really feels like it's some people that the bullet, and I think did the early product development and are seeing they're, they're reaping the rewards of it. But I saw for many, many years, a lot of companies just saying, well, we'll work on that next year, work on that next year. Right? What do you think has changed here? I mean, I have they finally done the math to really like realize what's going on? Or? Yeah, I mean, in your article, I'm curious, we'll definitely post links at the as sort of like the liner notes of this podcast, but I'm just kind of curious to know, like, you feel like we're at the turning point, cuz I felt like there'd be a turning point, we'd get the turning point many, many years ago, and companies are still reluctant.


Paul Denman  8:55  

Well, you know, everyone's talking about trillions of edge devices on the internet, right? Your home thermostat, etc, right? And IoT is going to be massively big. I should say that, you know, my focus has been more on the industrial manufacturing commercial, you know, not home automation, which is another whole market in itself. The model is completely different there. But you know, I have a ring doorbell that, you know, tells Amazon and tells my phone, you know, the UPS guys at the door, right? More of a convenience. It is you're right, that's a good point. It's it's really convenience based. It has to be more money based when you're talking to Ford, General Motors, Siemens, ABB, whatever. Right? Absolutely. But just to sort of narrow the focus a little bit. There's a show in in Nuremberg called SPS. It is the premier world show for motor controls in motion control, and automation, but not automation on the fact of just simple robotics. It's focused on the control of that. So the turnout typically is about 30 to 45,000 people every year, this was pre COVID. And it was cancelled this year, with a one week notice. So we we had equipment on its way out there.


Justin Grammens  10:15  

Oh, no, that stinks. Turn it around, bring it back,


Paul Denman  10:19  

People bought tickets and everything you know. But that show is probably to put a plug for that, that is the number one show to see 4.0 Because there's a high focus. And actually, part of the convention centers are just on that subject alone, you see two major players right now, and others moving into it. And I think the ones that have gotten a little ahead of the curve so far have been ABB, and Siemens. And their solutions have been coming along with a cloud based platform also. So it's a matter of collecting edge data and getting it to the cloud and then analyzing it goes Siemens product is called Mindsphere. And ABB is called Ability. So they have names for them, they usually run on AWS platforms. And in the case of dealing with, you know, the China market, they run on Alibaba, rather than AWS, China's insistent on that, but what they are is primarily data collectors. And then it's up to you to trade to collect and analyze that data in the cloud, you can imagine that it's not real time, right? It's pretty, it's pretty fast, I mean, up links are getting with fiber pretty fast these days. But if he has something that's got very high speed cycle rates, you probably want to do more of the thinking, where the motion is not in the cloud, right? Just to make a comment about edge to cloud here is that. Also, the game has to be completely different when you're dealing with semiconductor foundries. So you know, dealing with Intel, Taiwan, semiconductor, micron, whatever, their processes are so proprietary, the quote unquote, air gap, the factory, right, no connection to the outside world. So what do you do with that cloud? So they're on premise cloud systems, and that the catch term for that has been fog. It's not a cloud, it's a fog. But they have on prem servers that are running fundamental software, like Wonder where Microsoft Azure, whatever it is on prem, they'll collect that data, look for anomalies and give them real time information.


Justin Grammens  12:32  

Sure, yeah. Basically more like a private cloud instance for them. Yeah, I mean, the other thing is, is even if they aren't allowed to connect to the internet, like you said, sometimes downtime happens, right? And so you need these edge devices fail to react in a real time fashion, and not have to continually be connected to the mothership. Because sometimes it's just not reliable. Right?


Paul Denman  12:50  

Exactly. Yep. Microsoft, because of how far and how long they've been out there, still has a pretty good head start on what's actually down on the factory floor. So he's other companies that have the edge to cloud solution, or, you know, they're offering the total package. So they're going to offer, for instance, ABB in Germany, two years ago, pre COVID, they introduced a little clip on module, that's Bluetooth 5.0. You clip to the motor, and usually a large one. So they're 20 horse or bigger, that Bluetooth on prem will communicate to a hub via Bluetooth. And then that will in turn, go up to whatever the servers or wherever they are. So they're trying to provide the whole solution. Right. Siemens has the same approach. They have a clip on unit that measures the vibration temperature inductance variation, anomaly detection of the motor. And it's transmitting that data. I asked everyone, I was a little surprised. Why Bluetooth? Do you know why?


Justin Grammens  13:59  

Well, I mean, you could potentially use a phone I guess for it. If you as your edge collection device.


Paul Denman  14:05  

It turns out the short range in the security of Bluetooth is harder, point to point to hack than Wi Fi. So they figure it they consider it more secure. in it. I think the short range and although 5.0 is supposed to have very long range, the new standard, but there's a very low power transmission. So then you're sort of down to Okay, I'm going to detect something where is it better to detect it in the cloud or on the edge? Well, the big trend, or the big logical path for this is to push it to the edge as much intelligence as you can. So now there's a trend toward what are called intelligent actuators. They're basically an integrated solution and say it's a ball screw, lead screw, linear magnetic shaft motor and He has the drive and controller in the data collection algorithm right on the actuator. The uniqueness isn't really the motor, it's the driving controller electronics, right? There's also been a new trend toward, forget the actuator, just a motor. And on the back of the motor is a drive control electronics module. And the motor is smart. Now it's a motor you just talked to it could be even talked to wirelessly, you can drive and run it wirelessly. So you can connect those motors to other motion devices. And there you have your fully integrated, you know, motion control device. Now we're down to what do you get to detect how you get to detect it, and how fast can you detect it? And maybe fast isn't important. Maybe it is right, right? Sure. And how much data do you collect? 


Justin Grammens  15:54  

Yeah, that's a frequency.


Paul Denman  15:55  

There are some people right now that are there's a team within Amazon, the only thing they do is work with exabyte sized data. You can imagine exabyte, right? That's a lot of data. Amazon's collecting a lot of data. We're not quite sure if Amazon or the government has the most right now. Amazon's rate is moving their collection rates moving very fast,


Justin Grammens  16:18  

Like jet engines collect terabytes of information just in one flight. Right?


Paul Denman  16:22  

That's right. 40 terabytes per every two hours. That's right. And that data is collected in the black box. And that box on a Saturday or weekend is dumped to, for instance, is GE his jump is dumped to the GE Data Center in California. They analyze that data for outliers.


Justin Grammens  16:42  

Sure, I did do they still have their predicts platform at all? Is that still a thing?


Paul Denman  16:46  

They do predicts is a package that does not really have internal high speed proprietary algorithm. But it's more of an approach like Siemens and ABB would have this data collection, simple analytics. The function of the algorithms, since we're talking about algorithms is in most of these cases, where we're at here, we're not talking about Google TensorFlow, we're we're looking at is there a cat or a dog sitting on the back of the pickup, identifying that way, that's more supervised machine learning, we're talking unsupervised in anomaly detection. So that's what the world is that for failure. Failure detection is anomalies. The most common algorithm out there is something called K Means you can Google it, big K with dash and then mea ns. And it's a very simple clustering algorithm that looks for basically Multiple Dimension Data is the been the industry standard for gosh, I don't know, 4050 years has been used, it is still the fundamental thing that's used for clustering data and looking for patterns. And looking for anomalies. There is a huge breakthrough that's been made in that in the area of something that would could replace it, at orders of magnitude faster. This Boon logics out of Minneapolis has come up with a almost three orders of magnitude faster algorithm that replaces that. So that particular one that they have is a streaming algorithm. So it can learn while streaming, everything else in the world is batch mode, you collect it, then you move it, and then you analyze it. This can do a real time. Now, so it can do learning in anomaly detection, real time, I think has got some real traction, as far as you know, getting out there and having a big impact. The question too, is lean and low power. You know, that's the other big advantage to this new algorithm. They call it nano appropriately, because it's a very lean algorithm, they can sit on a very small processor. And in the majority of machine learning when you're looking at, for instance, machine vision, surface wide detection, anomaly detection, you're dealing with high volume, high velocity all the VDS. Right, lots of data moving very fast. Can you keep up? So everyone has thrown a lot of processor, high wattage power to this thing, like the NVIDIA boards that have you know, they're drying, Watson Watson burning lots of heat and for the TensorFlow chips that Google you can fry an egg on them, right. So this has some really big potential in that space. I think the other the other trend that's having a big impact, and in this space is just basically the small little ARM processors. They're low power, and they can be embedded on to small devices. I don't know if you're familiar with FPGAs Field Programmable Logic Arrays. They've been the upcoming thing as far as if you really want speed is the fabric of, for instance, relaying out a circuit board by a software. So you can embed a algorithm in a chip and design the chip in Flash, that's called quote unquote fabric that would enable you to run very fast in the same way rather than using a standard CPU. But the cache, the ARM processors, st is guide, I think, version seven, their st seven or something, they're totally completely pained, microprocessor, you know, cache memory, everything, communications all on a chip. And because of the ARM architecture, they're low power, and they're getting as faster equally, possibly, sometimes even faster than FPGAs. And a lot less money, or FPGAs are still pretty expensive. So in talking to these lot, a lot of companies that are building, you know, Turk banner, companies, they're building sensors on factory floors, they're now embedding these they were, they were working on FPGAs, they've all shifted to ARM processors.


Justin Grammens  21:12  

Seems like a lot of the world's moving to arm, you know, Apple's new laptop here. So ARM based, I've been hearing lots of great things. I still have an old, an older Mac here, but everyone who's gotten one says it's amazing, you know, no heat, first of all, you know, it's like, it's, it's, you don't need a fan. It's fanless. And then also just yeah, they're like, I'm getting days of battery power out of this thing. So it seems like, that's where everyone's moving to these days. Yep. You've covered a lot here. You know, it's just, there's just so many areas to start to dive into a little bit. I guess, going back to the original thing you're talking about, you know, somebody is manufacturing a motor that already has all of this components, it has the algorithm has the technology has a connectivity, it has all of it built into it, right? Is that what you're saying? And basically, manufacturers have stepped up. And they've sort of built this all in one solution.


Paul Denman  21:56  

Smart motors been around for the term quite a while, because it's descriptive, that can't be copyrighted. And there was an attempt to copyright that term smart motor, but basically, now it's sort of become what you might call, like you said, smart actuators. So things have become, you know, the trend is always smaller, more compact, higher performance, right? In in a way we actually little bit, you're actually getting better cost savings. Also, you know, Tesla will have on the line just on the pipe for the PLC, they'll have over 50,000 sensors, all running on a controller, you know, you've got to monitor 50,000 sensors, and what's the response of that? And how fast can they all communicate detect something? It's going to be all about? How do you connect all these Smart Edge analytics devices? Which is another whole project in itself? You know, what do you use to have them all talk, get to have to have hubs that will talk to eight to maybe 16 edge devices that will collect some of the data and then move it up farther to the on prem or, or cloud server, right? If you get a shuttle line down to replace it, then you have to plan it, ideally, right. So you want to have everybody ready there. So the downtimes at a minimum, but, you know, knowing that something could fail, you know, Boone is be able than doing tests that he can predict three months out on their new algorithm, if a motor will fail. So I think the you know, the potential for early enough warning is is there now to do it.


Justin Grammens  23:35  

Yeah. I mean, have you seen AI go through a couple of these what they call AI winters, I guess where there's lots of promises. I mean, I'm I'm thinking about the Gartner Hype Cycle, you know, and, and maybe maybe where you think industry 4.0 is in that are we in sort of the trough of disillusionment? Are we getting out to the plateau of productivity? Tracking the Gartner in 2022? And see where they put it? But are you feeling like we're on our way to sort of wide industry adoption with a lot of these things?


Paul Denman  24:01  

Yeah, I think people have gotten real now, right? The fact that you're collecting data doesn't mess it necessarily mean, you have the right algorithm to detect the failure. Okay, fine. I'm throwing all this up on the cloud. I'm charting envelope of, you know, this is the power and the motor the current than the motor, not in a normal curve. You develop an envelope on that current curve. And you're saying, Okay, well, it's going outside that pattern. Now, it's going more like multiple patterns. We're looking at current position, velocity, temperature, vibration, all creating a pattern. And that was very fast, obviously, in its non threshold base. Companies like Arcus technology and people that have developed a motion controller devices that are actually pattern based. And also Boone logic has a pattern based anomaly detection system for motors. So that's going to be the new trend, pattern versus threshold. That's the big jump.


Justin Grammens  25:06  

So I mean, I feel it does a failure have to happen for then to you to say, Aha failure happened. And now let's look back at all the data using k means you're saying, you know, basically this clustering here's because when you're doing unsupervised learning, and maybe we'll step back for a second, maybe for the listeners, you could define maybe unsupervised versus supervised learning.


Paul Denman  25:22  

Unsupervised Learning is where you don't know what you're collecting, you're just collecting the data, right? It could be all over the board, you don't know what it is you're collecting this data, you're looking for patterns. Supervised is where you're supervising something, telling something you already know what you're looking for, you're supervising an action toward a pattern that you're looking for. So recognition would be, you know, supervised learning, right? Because if you're looking for a particular person, you're supervising that processor to go find that pattern.


Justin Grammens  25:59  

It takes human interaction, or take some sort of tagging or labeling of data, right? You know, you say, Oh, this is what the this what this cat looks like, like you're saying, but in these cases, I don't think you don't even know what you're looking for. You just know what failure happened. And then you kind of need to go back and and let the system figure out. Okay, here's a case where I, where I led to this failure is that is that sort of


Paul Denman  26:21  

Exactly what you're doing is you're primarily, you know, what's normal. So in a way, you supervised the learning of that process. But while it's running, it's an unsupervised mode, because you're looking for an anomaly, something that's not normal. So it's sort of a pattern fit or envelope fit of, if you will have, you know, what's normal and not normal?


Justin Grammens  26:45  

So then, is there something to be said that you have to wait for a failure to happen to then see all the all the things that lead up to it or not? Can you just let a motor run, even though even if it doesn't fail for a year, it's still humming along? Can you still build that envelope around it and say, Hey, here's normal operating conditions, when it comes to vibration, and voltage and temperature and all that sort of stuff? Can you still build an out algorithm without actually having any failures?


Paul Denman  27:07  

That's a very good question. Actually, in supervised learning, if you're looking for a failure, you need to teach it that failure. So that has to occur, you could not possibly in a factory floor, any automation platform, conceive of all the failure modes and induced those in order to train your model. It's just virtually impossible. It can be millions of combinations. For instance, if, you know, IBM has got some platforms, that they're just testing fairly big ASICs on their Lawrence Livermore Labs, and, you know, they're, they just got, I think, the upgrade to go from 300 to 500,000 cores on their supercomputer now, and IBM System is using that in atomic structure analysis. But they've got to teach those cores, every possible combination in order to be able to do that. So that's supervised. So you're right. I mean, you can't teach it every possibility. So what you want to do is teach it long enough, what's considered normal. So you can usually, if you have an actuary that let's just say, moves, dowel pin and a factory floor, on a car in and out, you know, it's as simple motion, right? You could cycle that 20 times and you'd have it, you don't have to run it for weeks, right? You know, what you can train the patterns fairly quickly, in some of the platforms that are out there, now you run them for five or 10 minutes teaching, and then you're out of the learn mode, and you're back into the detection mode, looking for an outlier.


Justin Grammens  28:38  

Gotcha. And bring it back to CPU cycles and stuff like that, from what I understand, it's the teaching in the building of the algorithm that actually is fairly expensive, right? Once the model has been built, you can actually run it pretty low power, right, just to do the checking against it validation,


Paul Denman  28:54  

the same device is used this teaching is looking for anomaly is the same processor. So it's, it's a constant. Teaching versus running is not really usually a big power difference. 


Justin Grammens  29:07  

Yeah, okay, yeah, I just know, like, when you go to build a neural network, or something like that, in those cases, like, that's, that's where, you know, you can actually and you have a lot of data to run through the neural net, to try and figure out how many layers you know, and how this how, you know, like, what is the optimal path through it, you know, all the math that goes on inside there, all the weights and all that sort of stuff, is actually to teach it is actually the the more CPU intensive part. That's why sometimes people was what I've heard is, is they they ship that data, they train the model on the cloud, and then they push it back down. And in fact, there's a number of different companies that offer basically solutions like that, you know, you can kick it up to the cloud, and they offer a cloud based solution and they can push it down to a little just even like, like an Arduino. You know, you can just have that thing sort of like running and doing a lot of the predictions with a super low power. 


Paul Denman  29:58  

No, no you're you're you're talking in the world of what's quote, unquote, deep learning, right? Not everything's deep learning not everything is neural nets, the world has come to accept AI as Oh, it's a, it's a neural net, the term deep learning the deep has to do with layers deep of the net, right? So I believe, you know, Google's platforms, the higher performance platforms go 50 deep on a net neural net structure. So those are serious power hungry applications, you have a lot of CUDA cores is is Nvidia would call it running on, on these very big chips. Even with a smaller gate sizes, they're still run really hot, deep learning is different. If you have another way, you know, like the Buddha logic one or just a standard processor that's doing simple Google Analytics for anomaly detection, just standard prot, you know, so slow sample rates on a desti three arm can suffice, you know, very low power.


Justin Grammens  31:02  

No, that's great, you know, by my bring in a huge arsenal, and a lot of other complexity and power hungry


Paul Denman  31:09  

What I've seen even the models on for instance, mindsphere. And things they can do it, they can spin up maybe 10,000 cores on Amazon to do deep learning model. Google can do it on their TensorFlow platform. But it's not done that much. Actually, you don't need that. Because, once again, you're doing your anomaly detection, right? And the models been made simple for you. So you can do that right on the edge. You don't have to do that on the cloud.


Justin Grammens  31:39  

Yeah, for sure. It feels like there are a lot of different technologies, you know, like you said, that, you know, somebody could just clip something onto an existing motor, you know, boom logic has a unique proprietary algorithm. Siemens has a cloud, like, it feels like at least you know, for the for a number of years here, everyone was sort of solving a little piece of it, right? And is it does it feel like now that that now this has become more commoditized, like, you know, so if I own a factory right now, today, can I just, you know, go out for low cost and kind of automated and just by getting standard components off the shelf?


Paul Denman  32:11  

You can, interestingly enough, Siemens has gone open market, or open architecture on their platform. Now, ABB has not done it, as I know, I might be incorrect on that news might be old, but they were talking about it by now, Siemens is open source so you can develop your own, you know, pattern recognition AI platform, plug it into their mindsphere product. And I noticed they had purchased several small AI startups one in England, that was at the show several years ago, once it gets out into the mass market, and now starting to come out on these low cost motor controllers, some of the older established vigor, Motion Control rackmount platforms that are out there are just now getting into it seriously and offering some version of anomaly detection.


Justin Grammens  33:03  

Sure, you know, we talked a little bit about the maybe the technology, right, right, whether it be k means clustering, or deep learning. Now, there's all neural networks, there's all these sort of like different terms. But if you step it up a bit, I like to ask people like, how do you define artificial intelligence?


Paul Denman  33:19  

I like to stay with the term machine learning. Sure, sure. might even go on there. If you ask 10 people in the room, they all think it was something different.


Justin Grammens  33:30  

That's true. That's sort of like the thought behind the question, I guess is, you know, what's your perspective on it?


Paul Denman  33:35  

I've got grandkids, that they would all have a really interesting definition of what AI is, right? The term machine learning is a good term, because yet, is it intelligent? Well, I don't know. You can give it decision making capability. And it's it's making a decision on data. I guess if that's AI, yes, that's AI. But yeah, I think the focus is going to be in the next couple years here, once things cooled down with the COVID issue is getting very practical, right? You know, where's the return on investment? Where do they really need to detect data? And what do they need to collect? And I think in a lot of cases, we over collect data. We don't have to, you know, once you get down to the edge, you don't have to store all the data. It's more like a light switch, right is detecting it real time on that device. There is another big trend in motion that's coming about that's having an effect on how accurately to monitor the data. That's something in the actuator world that are called linear magnetic servo motors. And these are magnetic shafts Nippon Pulse designs and manufacturers that are the leading manufacturer of these, they have a non cogging motor that has very high force that is a good swap out for a lot of these companies and automotive and commercial. Go To these because they're, they're super compact. But as far as IoT, they're really easy to collect data. The reason is the motions can be very accurate. So you can have something pushing a 300 pound bag of grain into a baking oven, or you can have something moving away for the same motor that's used for the 300 pound green bag is used on wafer motion control. At Intel, that moves moves at four atoms wide, that's how accurate there but related IoT, you have very close, tight control, closed loop control of magnetic shaft motors better than you would have a mechanical, lead screw or bearing device. And there's a lot of companies moving into that field right now, I believe, pulses already sold over a million of them as of last December. So the markets growing fast. not taught in schools is sort of interesting. A lot of the younger engineers already are even familiar with them. 


Justin Grammens  36:10  

Yeah, it feels like in a lot of ways I met my background is in is in software. And you know, I hire a fair amount of engineers every year. I feel that way too. Just when it comes to software development, people coming out of school, I almost call them kids, I guess I would date myself a little bit. But these but you know, these students coming out of school, these graduates, you know, they they're learning the basics, but they're, I feel like the first six to like, you know, six to 12 months or so, I'm really making them more street smart than booksmart. And I think in just in general, I think for the United States to compete, you need to have a lot more internships, I guess, right. So that's one of the things that we'd like to do at our business is have more students coming in along the way, so they get a chance to sort of see what the real world is like. But then also, you know, teach them more current things, right? It's like, you know, yes, C C++ is certainly interesting and used in some capacities. But more, there's a lot more other Java and, and C Sharp engineers, you know, out there, and most students are gonna end up using those technologies more than some of these other ones. So, definitely getting more involved with more modern languages. I think in some ways, it's a good thing.


Paul Denman  37:15  

Exactly. So you, you didn't mention Fortran, there's not much out there anymore.


Justin Grammens  37:21  

Not that I know, I haven't written a line ever in my career, actually. But you know, you hear the stories. 


Paul Denman  37:27  

Fortran was a big deal.


Justin Grammens  37:29  

Well, you know, what, I think you can still find a Fortran job is what I heard, and they're kind of in high demand. But I think like everything new technology comes along, and a lot of it goes away. I mean, I this kind of leads into something I do like to ask other people too, is like, you know, what, what would you advise people that are coming out of school? Like, are there classes people should be taking? Or there you mentioned a trade shows and stuff like that? That sounds awesome. You know, people want to get into this area. Looking back at what you did you what you've done over the past five years, you know, what, what are some suggestions you'd have?


Paul Denman  37:58  

Absolutely. Number one is to get in a car or get on a plane and go to some of the major trade shows. A lot of the companies love talking to students, they love going about, you know, we got to train the next generation, seeing the new stuff. And in a lot of cases, a lot of the trade shows, you're showing it in action, right. So I think the big show right now, coming up, this becomes sort of the new standard for the US and automation is automate, I believe it was in Chicago this year it's going to be in Detroit is probably I think that would be a really good show for students to go and sort of see the new automation, blotter robotics platforms, but a lot of IoT, a lot of edge analytic, you know, machine learning things there, too. So


Justin Grammens  38:47  

Awesome. I'll get some of those links here of things you mentioned, and again, pop up on the site. Obviously, just networking with people, you know, and listening to these podcasts is really what I also like to encourage people to do just search around on the internet and just try and be like a sponge absorb as much as you can.


Paul Denman  39:02  

You can think geography. I don't know if a lot of people are aware of this, but when I work in the space of under the umbrella of life science, you know, there there are zones within the US that have, you know, sort of coagulated into learning centers and a lot of students and a lot of universities focusing in that space and is sort of shaken out in the space of genomics. Right now. Boston is sort of the core, you know, Cambridge down in Boston, that area is become the new pharma vector genomic. So, of course, you know, I've worked with Maderna on a lot of their lab, work cells, those kinds of companies, Merck AstraZeneca, that's become the pharma compound, you know, market drug delivery, right. Northern California is now the highest focus area for IBD in vitro diagnostics, so, you know, anything you do with bloodwork, COVID testing, whatever the majority of the companies in that space are going to be in Northern California. And now, yeah, the last remaining fellow, you know, if you've heard of Craig Venter sequences on genome, he's the left east coast, he was the last company to move in this space. And in DNA sequencing, he is now down in Southern California, along with sequel gnome Alpha metrics, and Illumina, the leader in the space. So that's become the DNA sequencing area of the world. And then fourth would be actually where I am, which is, quote, unquote, medical owlie, which are primarily medical device related. So pacemakers, heart valves, catheters, a lot of the manufacturing in that space is still Minnesota.


Justin Grammens  40:55  

Sure were huge in that space, for sure. You know, as you were talking about automation, I just, I was looking here online, because there was a group called ISA, the International Society of Automation, but I'm not sure if you're familiar with them at all. But they have a Twin Cities chapter. I haven't connected with them in years. But they did have a local show, I remember going to that was really around robotics, right and training an arm to pick up a thing and keep doing it over and over again, right, a lot of this automation stuff. 


Paul Denman  41:21  

There's a group called LRIG, which is laboratory robotics interest group, there's six, I believe, or seven different locations for that group. And there is actually although it's a catchphrase Medical Alley, there is a medical alley.org organization here that usually, I believe the XC o of one of the big medical device companies was running that year in Minnesota. And wow, there's bio is another huge organization that's nationally known and many companies are involved in that.


Justin Grammens  41:55  

Sure. Well, just one last question here, I think probably kind of like wind down, you know, how do you think this is going to change the future of work? Right. So as people enter the job market with all this new technology going on, and we can focus on just, you know, I guess, industrial IoT? You know, how do you think entering the workforce now is going to be different than it was, say, 10-15-20 years ago?


Paul Denman  42:15  

Well, I don't think the trends going to change, there's going to be more emphasis on in the software world, things are becoming more integrated, more intelligent, is surprisingly, I mean, when I'm talking to mechanical engineers, they've, in a lot of cases are writing code, machine level, even on a motion control device, that's part of the training, it's part of the package now. So learning, you know, whatever field you pick, even in most automation, if you're a mechanical engineering, student learning software is going to be, you know, probably a key in some respects. So yeah, the trend, I would say, is going to continue toward the higher ROI in machine learning, Edge analytics, pushing from the cloud down to the edge of Smarter devices on the edge in some of these newer inter factory networks.


Justin Grammens  43:13  

Yeah, no, that's, that's, that's interesting to know, it's like no longer are mechanical engineers sitting in their silo, and just kind of working in SolidWorks. And drawing pictures. I, you know, grossly simplified that but you know, they're working in sort of like the physical world, they're ultimately going to need to understand and crossover into this software land, probably everybody will. Yep, that's, that's a great thing to point out.


Paul Denman  43:34  

I know, Nippon pulse is come out with a new product that's called commander, which is really an aid in assist in that space. Because what you find out there is you have younger students, you can have a company like thermo scientific BioRad, all these big companies, and they maybe have 10 to 15,000 employees, they might have five mechanical engineers in that entire base. And they usually are burdened with everything, the software, the motion that control the mechanical engineering of the device, loading a diagnostic tray into an analytics machine. So how do you make their life simpler, right? Do they have to reinvent the wheel, they have to learn what a closed loop servo controller is, and, you know, they just don't want to, why would they do that? You know, right? Sure. Well, the vent a lot of companies for instance, Nippon is Commander product can run they have a small module, easy develop software that's almost basic like that, you know, a 10 year old can set this thing up. It can coordinate the motion of four to four axes, all automatically for you. A lot of companies getting down to a chip level or modular level to make their life easier, right? So that's the OEM market. It's not the, you know, factory floor car manufacturing line, but it's the OEM market. We have To build an instrument to do motion control, you got to build a circuit board, you've got to put chips on it. You know, how complicated does that have to be? Right?


Justin Grammens  45:10  

Sure, simpler is always better for sure. Well, how can people reach out to you if they want to want to connect with you, Paul?


Paul Denman  45:17  

Well, you can reach out to me either on LinkedIn or direct in that way. Probably 


Justin Grammens  45:21  

Gotcha. Yeah. Okay. Oh, good. Yeah, LinkedIn just fine. I'll be sure to put a link to your LinkedIn profile. Yeah. Is there anything else that maybe you want to chat about? Before we wind this down? Are we covered? Pretty much?


Paul Denman  45:31  

Yeah, I just was wondering what your opinion of is, with IoT, where's the best you think source for or shows that would be based around IoT? Yeah,


Justin Grammens  45:43  

I mean, there, there are a number of ones that happen out in California, you know, there's, there's the like IoT world, you know, a lot of these ones have. And, of course, we used to have IoT fuse here in the Twin Cities, windows, if that's going to be fired back up again. And of course, you know, it's so funny, because it's, it's such a big, it's such a big thing, you could consider going to CES, actually, right, as an IoT show, because there's just a lot of automation going on. And IoT, in my opinion, really, is, I guess, the sensors and the network, not so much around the machine learning and AI piece of it. But there's just so many shows that and, and when you say sensors, a network has served me that's like the nervous system of this entire thing that we have going on here, which is, you know, the making, making the world smart. And that touches everything from consumers to embed, you know, embedded devices, you're talking about medical devices to industrial. So yeah, I think, you know, frankly, I always kind of thought the term IoT was just going to go away. I think it I always sort of thought it had a shelf life of it. Because at the end of the day, in the next couple years, or next, you know, five to 10 years, everything's going to be smart. And the term IoT is going to kind of feel dated, it's going to feel like machine to machine what people used to say,


Paul Denman  46:52  

I agree with you. And you know, I think the hype with IoT, AI, whatever, like you said, on the productivity section of this right now, it's trying to, I think it's settled down. And a lot of money has been lost with all these startups, you know, claiming AI capability or applications of AI. But, you know, an interesting fact, just maybe related this. As far as machine learning I, I have a couple of friends that just graduated a couple years ago, PhDs, University of Minnesota, I went to a doctorate defense with them. There are eight PhDs graduating, and I asked where the other six going. I'm just curious. They said they're going dark. And I said, Well, what do you mean by that? He said, they're all going to Wall Street. Yep, they're all Wall Street gobbles up the majority of the high grade high performance students that with, you know, postdocs that are that are in machine learning right now. 


Justin Grammens  47:55  

Okay, interesting. So a lot of micro trades, I guess. Right? Is that pretty much what you're saying? No. And he's trading.


Paul Denman  48:02  

It's all about machine learning and training to get down to high frequency training. It's not there yet, but it's getting close.


Justin Grammens  48:10  

I wouldn't doubt it at all. If you're talking about wanting to make money, I guess, you know, you can stop a company from shutting down their car production here and save them. You know, like you said, $50,000. But on Wall Street, you could probably make even more. You do the right trade at the right time? For sure. Oh, awesome, Paul. Well, I appreciate the time today, I thought we've had an awesome conversation. And yeah, you know, we haven't really delved into industry for it at all, in any of the previous podcast that we did. So I appreciate you coming on the program here. And, you know, giving just a really good not only a high level overview in terms of like, what it is, but also, you know, getting down into the details of really the true impact that businesses can see, you know, and using it. So I think there's gonna be a great program,


Paul Denman  48:53  

I think it's going to be exciting to see what happens is next year, once we get back to normal life, it's given a lot of people a lot of time to think and figure out where they really want to efficiently spend their time on product development also. And in a way, it's been good in that respect. People trying to get more, you know, better focus, if you will.


Justin Grammens  49:13  

I feel like it's at least in the like computer vision area, it feels like it's matured a little bit more, it feels like, you know, you don't need to train a lot of these things. Or like you said, it's very easy to train them where, you know, years ago, it was it was it, there was a huge technology curve, you needed to learn to do this. And now feels like a lot of these things are sort of coming either pre trained or very easy to train. Exactly. So that's good for all of us that don't have a PhD and neural networks, or data science. Exactly. Well, good. All right. Paul, will you take care of thanks again for the for the time today, and I know we'll be in touch in the future.


Paul Denman  49:47  

Thank you.


AI Announcer  49:49  

You listened to another episode of the Conversations on Applied AI podcast. We hope you're eager to learn more about applying artificial intelligence and deep learning within your organization. nation, you can visit us at applied ai.mn To keep up to date on our events and connect with our amazing community. Please don't hesitate to reach out to Justin at applied ai.mn If you're interested in participating in a future episode. Thank you for listening