Listen: Next in Tech | Episode 48: The everything that is Industrial IoT

View Full Transcript

Transcript provided by Kensho

Eric Hanselman

Welcome to Next in Tech, an S&P Global Market Intelligence podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Principal Research Analyst for the 451 Research arm of S&P Global Market Intelligence. And today, we will be discussing the industrial Internet of Things with Ian Hughes, a research analyst who's part of our infrastructure team that deals with IoT in a range of other subjects. Ian, welcome to the podcast.

Ian Hughes

Thank you. Great to be here.

Eric Hanselman

As I was discussing, I have this feeling we talked about enough of different subjects that I thought you've been on before, but clearly not. So my error, especially when we start thinking about something like what is this overarching part of IoT, a buzzy topic but one that is so critical to so many different aspects of a lot of what we're working with.

Ian Hughes

Absolutely. And industrial IoT sits in areas such as manufacturing. And I used to consider IoT, industrial IoT as sort of a subset of IoT. And then I realized that absolutely everything in the world gets manufactured, including IoT components. So industrial IoT is actually a superset of IoT. That's what the extra I is for, I reckon.

Eric Hanselman

Well, it brings together these a bunch of technologies. I mean, we touched on energy applications with [ Johannes Vermeer ], a few episodes ago. Actually, Dr. [ Katie Ring ] was talking about sort of her perspectives about aspects of IoT and services and support for a lot of consulting stuff. But that's just a small part. I mean there really is this breadth of applications that industrial IoT really covers.

Ian Hughes

Absolutely. And one of the things I've said from the day I started covering this stuff is that industrial use cases have been using all kinds of technology for a very long while. I mean that's what the industrial revolution was. It's like hey, it's steam, let's use it. And so elements of applying extra instrumentation to understand what's going on with something like a manufacturing process is quite a natural thing you would expect industrial companies to do.

So you've always had things like, say, the spinning jenny on the steam engine, which is a relief valve that goes off at a certain temperature or thermostats and triggers and things to work and dials to work at how -- what the temperature of something is or the pressure of something.

And so along comes Industrial IoT and says, well, let's use some more connected ways of applying sensors to this manufacturing process at all elements of it, from what's the raw material going in, what's the quality of that that's being fed into whatever the manufacturing process is, what speed is this motor running at, what temperature is this oven at, what's the pressure here.

And taking all those pieces of information that were probably already being captured because they're already on the machines. But rather than walking around to each machine, take each value and work out what's going on, use the tech to get it in near real time in a centralized place. So you can now understand the status of your manufacturing process as accurately as you possibly can in as close to real time as possible.

Eric Hanselman

And do that thing, which digitization starts to bring us to, which is start to correlate across that data and be able to start to understand it. I mean when we had independent monitoring systems, they clearly were doing a reasonable job, or a reasonable enough job managing each of the different aspects of a particular process.

But it was more complicated to tie back variations and say, oh, well, the oven bumped up 50 degrees and show the parts that we produce then were lower quality over the parts we produce then were higher quality without being able to do that correlation.

Ian Hughes

Absolutely. The elements of understanding the now as precise as possible, that's kind of one part of it. But as you say, you've got historical and contextual data around the process. So -- and if you're going to apply something, for instance, like predictive maintenance is a common use case for industrial IoT to say when is this thing going to fail, not it's failed because it's too late then. So determine when a part is close to failure, when something is starting to degrade more than is acceptable and deal with it there and then.

So you're kind of heading off problems as well as gaining efficiencies by making sure everything is running as optimally as it can because you're measuring it and adjusting accordingly and changing temperatures, changing pressure, changing speeds to just make sure you're in that golden battery, that golden recipe for as long as possible. And that's what manufacturing has always been trying to do.

And as has said, as adapted, we've got SCADA systems, Supervisory Control and Data Acquisition as they're called, that's basically a big virtual layer on top of a manufacturing process. It just happens to have been built maybe with pre-century technology originally. But now we can talk about a virtualized machine the same as you would spark up in a cloud environment now sitting there running that SCADA process.

Eric Hanselman

Well, and you're talking about -- you've jumped ahead to second order functions, which is that predictive capability. It's not just the -- looking back retrospectively and said, oh, what worked and what didn't work and hey, let's do this next time. But to then start to build on that data and learn that, hey, the last time the sequence of events happened, this took place.

And when we start to see that sequence of events happen, then take corrective action like predictive maintenance, when we saw this bearing start to vibrate this way before, we realized it was about to fail. And then to be able to say, hey, guess what, we see that at this other bearing over here. And now we can fix it before it actually fails and do all the good things about doing maintenance upfront as opposed to, as you're identifying, waiting until it dies and then having to fix it.

Ian Hughes

That's correct. And the focus of industrial IoT has been on -- and for the past, over many 5, 10 years, been focused on the machinery of the manufacturing process and finding those. Initially, it was instrumenting the things, and now it's working out what data makes sense and filtering out the noise and the things you don't need to get to the point that you can now train AI models to -- and machine learning models to do this predictive maintenance for you.

But there's also the elements now of trying to help enable the workforce to operate and do their job and fix machines in a better way. Because if -- we've seen lots of stuff about predictive maintenance that we just said, that says this is going to break. But if you then send somebody maybe more inexperienced onto the shop floor to fix a bearing on a machine and they've never done it before, relying on them looking at paper manuals and using kind of brute force to fix it, you're not getting the full benefit, the full kind of virtuous circle of industrial IoT.

So we're seeing a lot more things to help bring that IoT data to -- and the machine data to the person that's actually trying to do the job and trying to fix the machine. And that's where we start to see things like augmented reality being used, where you are presenting this IoT data into the view of the person that's doing the work, and giving them extra directions as to which bolt to undo, how tight something needs to be, whether something is live or not, which bit they need to replace.

But trying to help them through that task still using their expertise, but giving them as much either remote support or AI support as needed to get the job done. And so I often say that augmented reality is the user interface for IoT because of that, because you've got this spatial nature of machinery and somebody trying to engage with that, and you've got all this instrumentation, which is also spatial.

It knows which part it's talking about on a machine. So it's therefore relevant to say it's a bit over there, not the bit over there that you need to engage with. And so we're seeing that improvement in how the workforce can do their job.

Eric Hanselman

That's a fascinating idea that AR is that -- the user interface for IoT because you've identified what are those key pieces to being able to bridge that gap. We've got all this data, we know where it came from, we're working on a particular piece of equipment. But yet, how do we tie in the people aspect? And as you're identifying at the outset, this is -- fundamentally, industrial IoT is this transition into environments that have always existed, have always had systems in place to be able to get this work done.

But yet the next stage is not only just keeping track of how it's operating, but then taking that step into the operational realm, and being able to now engage the humans that are part of this in ways that they can be more productive as part of that transition.

Ian Hughes

Absolutely. The term that we often hear, and it's one of the many buzzwords that kind of sits out there, but is the evolution of digital twin and particularly the digital thread of that digital twin, which is its life cycle. So whereas original IoT instrumentation is just sensor data going down the wire and kind of appearing somewhere else on a dashboard. Here, people are now gathering that data into a virtual model that matches the physical device or the physical plant or it can work at any level of abstraction.

But let's just say the entire plant. So you've got a model of your plant that is now a digital model of the plant, but is injected with the live IoT data onto that virtual model. You now have a 3D model that you can kind of spin around, have a look at, engage with, get an overview of what's going on, zoom in on a particular part. But you've also got a live stream of information about the actual state of the plant.

So now you can take that digital twin, which sounds like it's a gimmick initially, doesn't it? That's -- why would you do that? But you've now got this model infused with data that you can basically take a copy of, for instance, and then use in a simulation as a training tool for somebody to say, well, what would happen if we changed these IoT values and this value, what would you do?

You then help them learn to fix something in a virtual environment or in an augmented reality environment. It's there, you're using that digital twin. Or you take that twin and you run it through even more simulations to say what if all these values and this machinery was operating in this particular environment in this particular way, and we changed supplier? So we had 10% less good quality of prime stuff coming in resources that we were using.

But how could we increase the quality of the product, nonetheless? And what's the cost benefit of doing that? So you've got lots of simulation opportunities. And then you've also got, as I just mentioned, the -- then taking elements of that model and delivering it, that twin delivering it to the shop floor when you actually stood in front of the machine and superimposing it over the actual machine that you're trying to fix.

So you've got lots of scope for doing things with that digital twin. And I said, it's about the thread as well that you could, and should, start -- if you're doing a greenfield factory, obviously, you build the factory virtually first, so everything is commissioned digitally. So you know where everything should be, you know how far it's going to be from some way you can dive in and check the ergonomics of it for your workforce, you can train your AGV robots to wander around the place and do the job they need to do because you've got a virtual version of it.

Then when you commission the factory, you end up with a physical factory, but you keep your digital model now instrumented with IoT. But then if you need to make improvements or you need to build a factory next door or you need to interface something to it or you need to improve something, you've then got a digital version of it that you can alter first to see how that works and then commission the change. So you have this thread, as I said, digital thread of what's going on that is the life cycle of it.

And that can also include the things that you're building. So if you're an electronics company, for instance, or -- the products you're building might be IoT-enabled, they might come off the production line with a digital twin of them anyway because you provide it to your customer, but you may need to monitor what's going on to provide an extra service to your customer and try and help them use it, or to try and upgrade it. So you end up with a digital twin and digital thread coming off a factory floor that's already a digital twin and digital thread.

Eric Hanselman

And we go from an environment in which the physical world is represented by blueprints and a lot of dimensional and maybe some performance data, to now actually being able to hand a customer for your product of a digital twin that fully models the behavior. And so that as you're identifying if you really are doing the digital simulation first before the factory build, you're going to reconfigure the floor, you're going to manage new components, new materials that you're feeding into a process. Now you've got the ability to actually see how that will run before you do the expensive part of actually putting the physical world together.

Ian Hughes

Absolutely.

Eric Hanselman

It puts a lot of focus on the building of the models. And actually, it was one of the things that Katie Ring was talking about in the previous episode, which is all of that data management. It puts that focus back on ensuring that the model is -- has sufficiently captured the various aspects of whatever it's representing, and sort of shifts what the engineering focus is. I guess we've talked forever about software that builds in scope and capability to be able to not only get the job done but to test itself and integrate itself well. This is really that next stage.

Ian Hughes

It is. But obviously, there, I was referring to the future, if you like where when you're doing something in greenfield and you're starting now. But the challenge, and this has been the challenge all the way through for all things manufacturing, all pieces of technology, is that we're also dealing with proper legacy manufacturing equipment that doesn't need to change, that has been there for 100 years, that has been pressing metal or stamping on things or cutting things for a very long while. And things don't -- everything on a factory floor doesn't get refreshed at all, really.

Eric Hanselman

Yes. I mean this is equipment that, if nothing else, the cost alone is significant enough that you're not going to swap this out just sort of willy-nilly as your requirements change.

Ian Hughes

And that's why there's always been this tension between the operation of technology world of manufacturing, particularly and the IT world because IT is rip-and-replace, patch, dump some more software on -- if it doesn't work, send a patch out. Whereas engineering is this thing is working, this is my core business, it's assembling this, it's cutting this, it's pressuring this, it's pushing this.

I need it to keep doing that. There's the entropy of the physical world that means it's going to wear out or bits are going to wear out, but it's not going to get ripped out. So that means there's lots of brownfield IoT in applying sensors to something or increasing, like pointing a camera at something.

The camera is a very, very good example of a high-end IoT sensor that can tell all sorts of things about something and the quality of a product on the line, whether there's oil leaking out or something. All those kind of things that cameras can do are now applied to these existing brownfield sites. And also, obviously, things like a camera, it's not intruding, it's not engaging with the machine.

As you know, most things when you measure things by sticking something on them actually affects the machine. So if you drill a hole in it to put a sensor and you've now got a hole in your machine. So you have to be kind of -- the IT side of things has to be cognizant of that and to let the OT people do what they do because they're engineers.

Eric Hanselman

Well, it's very different time scopes between the IT and OT expectations. And a lot of the places where there's friction is that grind between the OT, hey, we've got operational systems. We're not going to mess things up to spin around at the kind of pace that the IT folks expect to be able to do. So there's a certain matching that has to happen between them.

Ian Hughes

There is. And I think -- I mean it's adjusted. I mean lots of the OT companies are now deploying and they're effectively becoming software companies anyway. So they might produce hardware, but they're producing hardware with a lot of software in it because they want to try and provide good service for their customers.

And so they've very quickly adopted the more -- they've just kind of jumped a generation, if you like, into more modern software development in the fact that everything is virtualized and everything containerized and everything can run at the edge or in the cloud. It almost doesn't matter.

I mean it does, from an architecture point of view. You have to know where something is. But it's not the traditional embedded software that we've had in engineering applications where you have a PLC and it's got a particular level of code on it that might be different from the machine next door to it and some engineers to go in and kind of flash the problems to upgrade everything.

Eric Hanselman

Physically touch it to be able to update it.

Ian Hughes

Yes. I mean we've moved -- there's even kind of alarm systems that sit -- and monitoring systems that sit on the shop floor that are able to share resources with one another and act as -- in the same way as you see, kind of Elastic Cloud and things that say, well, I need more resources, let's spark something up. There are things on the shop floor that do that and basically borrow resources from machines next door to them to do a particular task when they need to and then give it back. And so load balancing of a hypercritical manufacturing systems in ways that even 5 years ago, you would say, well, nobody -- they're not going to do that. But now they can because it's robust enough.

Eric Hanselman

It's getting into edge cloud. And we're once again on the podcast back at the edge.

Ian Hughes

Well, manufacturing is edge first, it always has been. So the -- everything about manufacturing and industrial process is what's happening there, isn't it? It's -- this thing is being heated up. This thing is being stuck to this thing. So all the monitoring and all the understanding of what's going on has to be at that process.

So the initial IoT thing of, well, put a sensor on this, send it over to the cloud, was not really ever the right architectural shape or concept. So now we're having edge -- proper edge processing actually on the shop floor or on that piece of machinery, so that piece of machinery can run some machine learning model to say, well, I'm working. I'll adjust. I'll do this. I'll tweak.

I'll keep myself going. Hang on, I've got to have scope, what do I do now? After predictive maintenance let's sort it out. That's the model that works for the OT world. And the thing that might have been trained in the cloud, you might send your history and data up to the cloud to say, well, what -- how is this machine run for the past 20 years to train a model, but now you can drop the model down and run proper edge processing.

And I mentioned cameras. Cameras aren't streaming all their information up to the cloud, so something can figure how they are sitting there running machine learning algorithms to say, I will look at everything until I get the out-of-scope case that I need to say this thing is broken or the quality is not right.

And then I'll send an alert, a very simple quick message, not tons and tons of bandwidth being used. And so we're ending up with a manufacturing as the kind of most interesting architecture, if you like, because it's trying to solve those real-time, close to the edge problems.

Eric Hanselman

There is that real physical world that is the end product. We can't abstract away the physical components of the industrial environment.

Ian Hughes

Absolutely. That's where it meets.

Eric Hanselman

Where the virtual meets the physical. So how should organizations be considering the various implications of industrial Internet of Things and integrating into their plans? I mean, instrumentation has been sort of the big push historically. But there are a lot more complex interactions and considerations when you really start to think about how that takes place.

Ian Hughes

Yes. And I think the -- clearly, the pandemic has woken people up to the fact that there needs to be some degree of remote control or automation or ability for the workforce to do things in a slightly different way because you can't just have crowds of people wandering around and doing the jobs all the time.

So a lot of companies have woken up to the fact that they can engage in some form of digital transformation that doesn't have to be everything -- it doesn't have to be a big bang transformation. But something like, for instance, kind of workforce shift handover or something like that. So digital handover of information where maybe, at the moment, their workforce writes it down on the white board and then rubs it off the next day.

So just the simplicity of the stuff that maybe has been taken for granted in back office and the enterprise of things, but I'm just saying, well, here's the machine, put information in here, don't leave it on the whiteboard. Or do a screen capture or video capture the whiteboard and then do some image recognition to work out what's been written down. Anything to make it simpler, but capture stuff.

Or just the basics of -- and I've seen a lot of companies saying, well, their IoT journey started by putting a sensor either under a machine just to know whether it's still working or not, or which speed it was -- how fast was it churning stuff out. And so that begins the journey.

Eric Hanselman

Those early steps of digitization, which is capture first.

Ian Hughes

Yes.

Eric Hanselman

And then start to figure out, now that you've got it, what can you do with it? If you're grabbing the shift change information from the whiteboard, do you track that over time? Do you see what got noted 3 days in a row and then whatever happened and being able to then start to look back once you've got the information. But you have to have it to begin with.

Ian Hughes

Well, and that's the good -- one of the good things is that a lot of manufacturing processes and bigger manufacturing processes have had these historian databases that have been recording stuff, it's basically just a big log of what's going on. And that would be there, traditionally, to be kind of cracked open if something went wrong so you could do some forensic analysis of what the problem was.

But that data has been sitting there, accumulating for 20-plus years, maybe 30, maybe more. And now that can be used because usually, they try and end up with well-formed data in the historian, that can be used as your trading aid for your next step to say, can we do a machine learning agent on this machine that can help run it with this small device stack on the machine in edge processing rather than a big massive infrastructure, can we have something to help this machine through and keep it at its optimum and let us know when things are going wrong. So very few places are starting from zero.

Eric Hanselman

It seems to be one of those things where the time scale differences between IT and OT actually work strongly in OT's favor because there is that backlog. You've got historian data that goes back so much further from an IT scale, holy cow, why would capture and hang on to things for 5 years. I'm happy to hold them for 90 days, right?

Ian Hughes

Absolutely. And obviously, this sort of sits in the whole area of understanding what your process is to make it more efficient. So it kind of fits into ESG goals of companies now as well. You're saying, well, what's going on? Can it be more efficient? Can I use slightly less energy? Can I make the workplace safer because I know what's going on.

And also, I'm capturing all this information, so this is governance. This is -- I can prove that I'm doing this. I can prove that -- how much energy I'm using. I can prove how much waste producing or saving, how much CO2 is coming out of this. Those -- that all kind of sits into the -- into that big sustainability bucket that everyone needs to get into.

Eric Hanselman

More secondary capabilities. Wow, well, we could do an entire -- another -- well, I'm coming out of this thinking that we've probably got maybe another 3 podcasts. You mentioned robots, which I wanted to get into, but we didn't we have time for, being able to understand a lot of those secondary and tertiary pieces would be great. And digging into the AR aspects of what are these strong motivators, what are the big pieces of this.

Ian Hughes

Absolutely, heading to an industrial metaverse.

Eric Hanselman

Oh, and now you said metaverse. Ian, all right, well merit. So at least 3, maybe 4, we'll see. But we are at time for today. So Ian, thank you very much. Clearly, we've got a lot more to cover in this. So I would love to have you back in the very near future and dig into a whole host of these additional topics. But thank you for being on.

Ian Hughes

Thank you for having me, been a pleasure. Thank you.

Eric Hanselman

And thank you to our audience for hanging in there with it. That is it for this episode of Next in Tech. I hope you will join us for our next episode where we'll be discussing a whole range of different topics because there is always something Next in Tech.