Connecting data and unlocking new opportunities

In this episode of Control Intelligence, editor in chief Mike Bacidore talks to Daymon Thompson, senior software product manager at Beckhoff USA. They talk about how machine builders can get off on the right foot in the IoT journey, and how analytics can establish compelling competitive advantages to optimize machine efficiency, and even to create new revenue streams.

Transcript

Mike Bacidore: Hello, welcome to Control Intelligence, the Control Design podcast that goes deep inside the automation and technology that machine builders, system integrators and end users rely on to keep production humming efficiently.

I am Mike Bacidore, editor in chief of Control Design, and in this episode, I'm joined by Daymon Thompson, senior software product manager at Beckhoff USA. Daymon leads the Beckhoff software product management team in the United States, focusing on advanced automation and control software TwinCat 3. Additionally, he advocates for innovation in manufacturing using technology enabled by the convergence of industrial automation technology with IT and computer science technologies.

A seasoned presenter on new automation technologies, Daymon has also led numerous technological initiatives at Beckhoff USA, including the support of innovative concepts such as Industry 4.0, industrial Internet of Things, edge computing and cloud connected manufacturing systems.

He's achieved more than two decades of success in automation and industrial controls applications. This this experience extends internationally while living and working in Switzerland, during which time he was part of a new technologies group for a multinational medical device manufacturer. This group researched and built machines deployed for advanced manufacturing of various medical devices.

Daymon's a graduate of Regis University in Denver, and he holds two Bachelor of Science degrees, one in computer information systems and one in computer science.

IoT and analytics have been hot topics in the engineering community for many years, and we've moved well beyond theoretical discussions and into applied technologies in the field, especially for large end users. However, there are still many machine builders who may not know where to start on the path to connected machines or about easy to use tools to analyze machine data and find insights.

In this episode, I'll ask Daymon to talk about how machine builders can get off on the right foot in this journey and about analytics as a way to establish compelling competitive advantages to optimize machine efficiency, and even to create new revenue streams.

Daymon Thompson: Yeah. Thanks, Mike, for having me. That's great. Great intro. Thanks for that great introduction. I’m excited to be talking about this topic with you.

Bacidore: Awesome. Well, let's kick right into it then. The Internet of Things or Industrial Internet of Things and analytics, like I said, that's been a very hot topic in automation for some years now. So, what are you seeing in terms of the current adoption and implementation of the Internet of Things?

Thompson: Oh yeah, so you're right. It has been a hot topic for seven or so years, maybe a little longer, and it was the topic there for a while, right?

Bacidore: Right.

Thompson: And you know, like everybody that's listening to this podcast, I'm sure, has been through the beginnings of this was just figuring out what do we do? This is hot. Everybody's talking about it. What do we do? And now it's really boiling out. I mean, there's real projects, and we've been for some years having some real projects with real successes. And we see this break into kind of three different groups. There are the end users, and especially the large end users that are really kind of pushing it and they see the value, they have the resources and the dollars to put behind the big projects. And they're really, they're doing it. The machine builders are sort of trying to figure out how to support that, so when they supply a machine to that end user that it supports the infrastructure. And they're looking for their own goals of how can they use this to help the end users and create some business models and revenue streams for them. The integrators have found some success there with integrating and helping the end users implement new hardware or the infrastructure to make it happen. But they're also looking for ways to kind of turn this into a revenue stream. You know, how can they do that and not just sort of job-based stuff. So, there's there are things going on.

Bacidore: Right, right. So, with that advancement, I mean, there must be barriers to entry for a lot of these companies. What would those be for starting, say, an IT or an analytics project? I mean, is it mostly the connectivity aspect or is there something else?

Thompson: Yeah. So in the beginning days, that was a big conversation. You know, the getting started first step was, “well, I gotta connect my machine.” We don't see that too much anymore. From Beckhoff’s side, we looked at this years ago and said, “alright, hey, we deal with a lot of integrators and OEMs, and we want to help them be successful at their end customer and they're end customer may have completely different infrastructure, right? They may have internal systems, they have cloud systems, they may use a bunch of different things.” So, we said, “well, let's kind of build out our offering so that we can have our machine builder customers and integrator customers really support any infrastructure the end user is using directly from the controller.”

So, that's kind of what we went after. And then the other one we said is, “we offer all these field buses and connectivity to DeviceNet or Profinet or CANOpen, you know, anything. So, we can really bridge those to any of these IoT infrastructures.” So, we built that out so that we could use our controllers as a gateway for existing stuff too, sort of brownfield. So hey, I want to pull something from device net and post it to Twitter, no problem. All that's built into our controller, actually. It's kind of cool. So, connectivity is kind of taken off where I think that's really not the barrier to entry anymore. I think it more is defining the business use cases behind it.

Bacidore: Right. OK, so so the does that connectivity, does that include the remote access to the machines then?

Thompson: Yeah, it's a good question. And that's one of the really common questions is “hey, with all this IoT stuff, can I get on the machine and access problems and fix things faster and push updates?” and it really depends on the infrastructure. So, if an end user says, “hey, there's no connectivity to the Internet, to the cloud for my machine,” then yeah, of course there's connectivity within the network. If they allow that, that, that connectivity to the Internet, then absolutely the same connectivity, the same MQTT protocol or transport path that you're using to send data, from our standpoint at Beckhoff, we can absolutely have the OEM jump on with the development tools, do debugging, download new code, really help without having to put somebody on an airplane, and all that infrastructure is already in place and you really just setup some secure communications and away you go. So yeah, once there's that establishment to the cloud, then yeah, remote access, remote support is definitely, definitely an option and built in.

Bacidore: Well it sounds like you thought two or three steps ahead for what some of these barriers might have been in the first place, right?

Thompson: Yeah, yeah, yeah. We really, really sat and evaluated, right? Like, you know, “hey, this is going go to different infrastructures, how do we really support this?”

Bacidore: Right.

Thompson: So yeah.

Bacidore: So, some end users, like you said, they don’t allow connectivity, especially to the cloud. With every end user having different kinds of requirements, I mean you've touched on this a little bit, but the machine builders are kind of the go-between there. How can they still take advantage of the data analytics?

Thompson: Yeah, another very common question we get. And, you know, like the IoT connectivity side where we said, “OK, hey, our machine builder customers need to be able to support all these different things.” We did the same when we looked at our analytics offerings. We said, “you know, it's not just a multinational, huge end user that's a producer, and they need some really massive IoT solution that lives in the server room and there's teams of people working on this. We think that analyzing data really happens even at the machine builder. And analyzing data happens with the controls engineer while they're building and developing that machine.”

So, for decades we've had the ability to be able to visualize the data coming out of a controller. And so, the next step to that was, we want to be able to answer questions, right? So, not just what is the current vibration, but what was the maximum and minimum? Where was I when the vibration happened? What step of the PLC code it was in? So, what we did was build out kind of a suite of analytics products that start with running on a laptop, on a controls engineer's laptop. They can look at data, live off the machine and run it through about 90-plus algorithms to figure out what the frequency is, what's going on in the machine, to do a few things. One is to find issues so that they can find bugs in their code and get that machine out the door quicker. And the second one is optimizing. So, doing things like timing analysis. So, I want to make that my machine is faster than my competitor’s machine. I need a good set of tools in a toolbox to do some analytics to figure out where in the program can I squeeze out a few seconds or, the example I give a lot is I have this machine with a loading sequence and the loading sequence takes between 1 and say 6 seconds and you go, well why?

So, you need a good set of tools that help you figure out, how do I make that more consistent? Because my customer really wants this to happen at one second, right? Not one to six and sort of load parts. So, we sort of enabled it that point, which takes no connectivity. Then the next part of that is we thought, “all right, that's great sitting in the machine, optimizing it, what happens then when I'm the OEM and I ship the machine to the end user and the end user says, hey, I got this weird behavior.” OK, great. And I used to build machines before I worked at Beckhoff and you'd get this call the first thing in the morning, and they say, “hey, by the way, your machine at 2 in the morning last night, it did something weird. And you’re like “cool, send me the log file.” And there’s a bunch of error messages and things, and you’re looking at it and thinking, “I don’t know. It missed the sensor 10 times, I have no idea why it did that.

Bacidore: Right, right.

Thompson: And so, at that point you're thinking, “do I send somebody on site.” like what do we do here? So, the next thing we said was, “alright, no connectivity, no hooking up to anything we enabled in the controllers.” We can turn on what we call a logger, but we call it the machine flight recorder, and really you're logging almost every, or you could, every PLC cycles worth of data motion control, Fieldbus for something like 24 hours, you know, 48 hours. And then, when there's a problem you can tell the end user, “hey send me that file,” and then I can use that same analytics toolbox or the work bench to go through and figure out what happened. Or same scenario where they say “your machine is not really consistent cycle here, it's not producing parts consistently.” Usually, what do you do? You go, I don't know and send a guy on site take a look. But with the flight recorder file you can really get back at your engineers desk at the OEM site and say “OK let's just take a look at what's going on. Maybe I can make a code change and just push the code change to them remotely.” So, there's a lot of options there for those guys that is not just a cloud solution or an enterprise end user solution but there's a lot of looking and analyzing data from an OEM standpoint.

Bacidore: Yeah, that there's a lot to unpack in that answer that you just gave this the well. I mean just on the on the analytics side and the troubleshooting side, I mean that is almost tailor-made for machine builders who are interested in offering, say subscription-based production or production-as-a-service or equipment as whatever they want to call it. You know, just to be able to implement the machinery and then maintain the machinery remotely, and then the end user just pays a monthly, yearly, semi-annual fee based on how much they use, and then you can optimize the machinery.

Thompson: Yeah.

Bacidore: Yeah. And and back to the first part of your answer, the optimization is really where I think a lot of the advancements or a lot of the potential for a lot of this remote analytics exists because yeah, we can do predictive maintenance remotely.

Some companies are doing that already, but that's great. We can save money, reduce downtime, blah blah, blah. But it's really that fine tuning for the for the operations and for making the machine operate at a more efficient level and reduce energy costs.

Thompson: Yeah, exactly. And on the on that same thread, you're totally right.

So, the way we architected that was you know from the OEM, kind of you know carrying the story forward here, you know now they're using that same set of toolbox, workbench, and like you mentioned, some of the value add for the OEM might be, “hey, I want to have a dashboard that says how are your machines running, OEE, that kind of stuff.” That same toolbox from us, really once you build that up with this drag-and-drop and even browsing variables in the controllers and dragging them into algorithms and dragging results of the algorithms and other ones, you can kind of hit a button that says deploy. And what it does is it generates out what code, actually PLC code, that can run a server, run on controller and a dashboard, which is completely auto generated and built in our HMI software so that you can do 24/7 monitoring and OEE and all that predictive maintenance and all of that running on constant live data and go back and say “hey, what was the OEE yesterday between 6:00 AM and 12:00? What was the vibration on this motor between noon and 1 p.m. yesterday?” We generate all those things in PLC and HMI because that's what the controls engineers know, right?

That’s kind of the base of who's using this stuff.

Bacidore: Right.

Yeah, for the most part, yeah, but I, and I'm sure you know, too, that is changing over time, and that's actually a great segue into the next question that I have. Because especially with computer languages becoming much more common, and a lot of these organizations are even looking for low-code or no-code programming examples for sure. So, what kind of skill set or programming knowledge does someone need in order to implement some of the scenarios that you're talking about?

Thompson: Yeah. Yeah, it's a good question. So, as we know, the whole industry is, there's kind of a skilled labor shortage, right? You know, if you talk to any of our integrators, they have a hard time finding guys to fulfill the needs in the market at the moment. And you're totally right. So, we designed this the analytics product, our workbench or the service tool we call it, with drag-and-drop, basically no-code configuration, and your browsing and dropping and then you can even take things like optimizing machine where you get a timestamp, you can drag it down into a nice graphical view of all the data and say “OK, I can see exactly where this is happening on the graphs” and then instead of having to sit down and code some of this stuff you just say, “hey, I want to generate,” and it generates out all of the code that you need, that you can just run untouched. We like the idea that you give the flexibility of the drag-and-drop environment plus the code generated in the background because then you can add a lot of things to it that says, “hey, not only do I want to see this on a dashboard, but I also want to send it somewhere else to a database. I also would like to format it in a different way. I would also like to do some extra things.” That's code that a PLC programmer knows, like structured text, that they can add that functionality to it. So, it's kind of nice, yeah.

Bacidore: Sure.

Absolutely. So, on the other side though, are OEM's, are integrators going to need someone with a new skill-set outside of what current controls engineers or programmers have in order to accomplish any of these?

Thompson: Yes and no. So, really the easy answer. Next question. No, I’m just kidding.

So, we really built all that analytics toolbox into the same environment, even the same project actually is like the PLC programming environment, motion control, Fieldbus setup. So, anybody that's kind of familiar with controls or especially the Beckhoff TwinCat software, it's in that same thing. And because it's drag-and-drop and generates out standardized isix 1131 code and HMI code.

It's not really like it's a completely different world and completely different environment, which is great because, I mean, I know at the moment every OEM and integrator’s got plenty of stuff to do, the last thing they need is to try to go figure out somebody with a new skill set, integrating them into the company, grab a whole new platform. So, I think this really, as I think Beckhoff usually does over the last decades, we sort of take it a little bit different look at the technology in the market than what everybody else is doing, a little more innovation there. And I think we've done the same thing with our analytics suite.

Bacidore: Right. Absolutely, totally agree.

So, what about these larger companies that were talking about before? What happens in in cases where, or even say an end user, they already have their own large or mature analytic solution like an Amazon Web Services or Azure or SAP?

Thompson: Yeah. And it's another common question we get because they say, “ohh, you know, we standardize on, you know whatever might be, Azure and that's what we use for our corporate, how are our factory is running kind of dashboard.” And you say, “that's great, that's a great solution. Never ever would we want to try to displace that, but we think that there is a place to sort of augment that.” And yeah, I guess I'll explain that a little bit.

You know, there could be many solutions out there to look at OEE, how’s the machine running, how much downtime. But I wouldn't call it really high-fidelity or high-speed data collection. A lot of things, even a lot of the machine-learning algorithms in the cloud provided by say AWS work very well, but they're looking for about 1 second resolution in the data, and really, they want you to collect for 90 days to a year, and then they'll run through a kind of machine-learning algorithm and go through that.

Bacidore: OK.

Thompson: And that's great, that's great. But what happens sometimes is if the OEM's trying to figure out some weird behavior in the machine or really optimize to really tweak out as many products as they possibly can, that's not at all 1-second level. Generally, that's looking at things at a much lower level. So, for example, if your PLC is running at one millisecond, that's 1000 times it's run through that logic, and possibly 1000 times it's updated the I/O, per every 1 second. So, what we've had some customers do really successfully is implement when they say, “OK let's send that data at very high resolutions and very compressed into kind of an analytic server running on-site or even in the same machine. And then, we can store off that high resolution data for, say, 24 hours, 48 hours, but in the meantime, be calculating like maybe a Boolean that toggles several times a second.” You know, I don't think the higher-level systems care when exactly it toggled, what they care about is hey, in the last second or the last 5 seconds this thing's gone this many times, and probably they don't care about that. They care about “in the last 5 seconds, 10 seconds, 30 seconds, this is how many products went by my photo eye,” for example. So, that it's kind of an edge-compute use case where you're saying, “alright, I want to store the high-fidelities, the high-speed data for optimization, figuring out problems, and at the same time I can absolutely still pass that data along to the existing infrastructure at whatever update rates and whatever format that they're looking for.”

So, there's kind of a good balance there, and I don't see a lot of people in the market thinking down at that level of “hey, we're already connecting the machine, we're already collecting data. What if we can also use it a little higher speed to optimize, find problems and give you all the insights and OEE and dashboard that you're looking for.” It seems to be the missing thing there that I think we're fulfilling.

Bacidore: Right, yes.

So, I mean, I think that kind of addresses the next thing I wanted to ask you about, which were opportunities that OEM's might have to offer kinds of analytics as a value-add. Are there any others?

Thompson: Yeah. So, some of the value-add things that we've seen and helped implement a couple different strategies. So, one is that customer that says “well, my machines definitely not getting a direct connection to the machine or to the to the cloud, not even through a gateway, I just don't want to take the security risk.” Great. Sometimes the OEMs ask us, “hey, I got shut down, I can't connect” and you think “well, no. Actually, you couldn't put into the facility.” Like we have one customer that they say “if you buy one machine for me in your facility and you add the analytics option, I’ll embed a PC in the machine cabinet, if you'd like, that then looks at all the analytics for that machine, gives you a dashboard, and then if you buy two machines from me, 10 machines, 50 machines from me, you already have the analytics runtime server or PC, and we'll just connect those up and have them as more data points or more end points for the machine to be collected, and they automatically kind of show up on the dashboard as well.” So that's offering it kind of built into the end user's facility and adding some value there.

The OEM can come in and update those occasionally if they need new things. The added value to that is what we just talked about. It's on that same analytics server looking at the algorithms that can be storing that high-speed data. So, if there's ever a problem on any of their machines, we can say “OK, go in, cut out that hour’s-worth of weird behavior, send it over to my engineer. We'll get you addressed pretty quickly. We'll figure out what's going on.” So, there's some value there. Then, obviously, if they have cloud connectivity, we can take that and just scale it up so that each one of their end users can have a portal, they log in and look at how their machines are running in their facility.

And there's quite a few different infrastructures that can be can be laid out, but there's definitely some potential for the OEMs, whether there's connectivity to the cloud or not.

Bacidore: Right, OK. So, that kind of made me start to think about these algorithms being used to yes, as machine-learning tools and potentially developing some sort of artificial intelligence, even that can make those decisions for you rather than even bring it back to the machine builder level, but that brings me to my last question more so than anything, which is what, what are some of the new or future technologies that are going to come into play in these areas?

Thompson: Yeah. So, you're completely right.

Yeah, back up a little bit. So, in our analytics toolbox with the drag-and-drop, you know, like I mentioned, there's 90-something algorithms at the moment, and we built in a way that you can also add whatever. So, if you come up with, hey, I know my specific piece of equipment or my specific line.” Great, you can implement something in C++ add it to the drag-and-drop everybody in the company can use it. Cool.

Then the next step to that is, like you said, it's kind of machine learning at the moment for a couple years now. We've had the ability to train a machine-learning model and really put it in the controller and run that machine-learning model in real time. And the idea is you know, hey, maybe there's a complex motion or there's some decision that ends up taking 10 to 15 things into consideration, and to write that by hand would be crazy to hand write that algorithm. So, yeah, we can embed that into the PLC. So, what we developed is using that same technology to say “if you train an algorithm, we can drop that into that analytics toolbox. And then from there maybe your guys that aren't machine-learning experts can just drag and drop that and use that in their machine-learning algorithm and in their analytics toolbox, using that as either a machine optimization. So, either at the OEM using machine learning to help them figure out how to optimize, or at the end user site looking at machine learning for maybe predictive maintenance, for example, looking at vibration to figure out maybe when a rotating motor is going to fail things like that.

So, you are right, the machine learning will get more and more common in industry and in in two ways, right? One running on the analytics side, which we hear a lot about today, and the other one is that we've been doing for a while is running it actually in the controller and calling the machine-learning algorithms from PLC ladder logic. Even so, machine learning really is the future on in a lot of aspects of industrial controls and IoT analytics.

Bacidore: Yeah. I mean, that's mind blowing. It's ieven beyond that old Home Depot, remember the Easy Button, where you just hit the Easy Button and it would do it. Like, you don't even have to hit the button. It already hits the button for you.

Thompson: Yeah. Yeah, after somebody trains the model. Yep.

Bacidore: Right, right after somebody trains the button to hit itself. Yeah. Then it does it for you. Wow. Absolutely Amazing.

Alright, well, I know you've got other things to do, so I don’t want to take up all your time, but thank you so much, Daymon, for taking the time to talk this afternoon. That's certainly some very sound advice on connecting data and unlocking new opportunities, and some just wild progressive steps forward in terms of what's possible for machine builders and the end users in order to unlock a lot of that data.

Thompson: Yeah, it was fun. We'll do it again someday, Mike.

Bacidore: Yeah, absolutely. Next time with a beer.

Thompson: Yeah, sounds good.

Bacidore: Yes, alright. Well thanks again, Daymon. And thanks to our listeners for joining us today on Control Intelligence.

For more, tune into Control Intelligence: A Podcast from Control Design.