Sound Policy podcast: Inside the next industrial revolution

Roland Schaefer: I'm Roland Schafer. I was hired by FM to head up a research area in essentially industry 4.0 technologies. 

Brian Amaral: What is Industry 4.0? Or the fourth industrial revolution in other words?

Roland Schaefer: So, you worded it quite well. I always talk about industry 4.0 being the marketing term behind the fourth industrial revolution.

In the 1700s there was the first industrial revolution, which was steam engine. About a hundred years later came a whole new technology called electricity, electric motors, and everything that totally revolutionized the industry again for a second time. Then about another hundred years later in the 1960s came electronics, microelectronics, having reprogrammable computing systems to control your industrial facilities.

And about 10 years ago, 10, 15 years ago, they started the talk about all of these new technologies that were coming to maturity that would spur a new industrial revolution, the fourth industrial revolution. So some of these technologies are just communication, getting the machine to talk to some other machine.

Having both wireless communication or internet communication. The whole concept of cloud computing and putting all of your data into the cloud became a lot more mature, became a lot more accepted. Once you've got all of this data in the cloud, you've got large volumes of data, data mining, data analytics, machine learning, you know, that was hitting maturity levels that were making it very useful.

The whole idea of autonomous vehicles, navigation, GPS, you know, these various technologies and some other technologies were part of this industrial revolution are things like additive manufacturing, virtual reality, those sorts of topics as well, which are not quite as relevant for us. But all of these technologies sort of coming together is essentially going to revolutionize the way factories or the way industry works.

Brian Amaral: And why is that important for FM to really fully understand?

Roland Schaefer: This is the whole reason why I'm here, is you know, to a large extent a lot of our clients are going to be implementing, are going to be passing through this digitalization, are going to be living this industrial revolution. And a lot of these technologies they'll be bringing in potential risks that we don't know about yet. Trying to understand what are these technologies, how are our clients using them? What does this mean for them? What kind of risks are they introducing themselves to by implementing some of these technologies? That is essentially what I'm trying to figure out.

My goal is to provide enough information and enough knowledge that FM can make, you know, informed decisions about where do we go from here?

Narration: One of the key pieces of technology that Roland will be looking into is something called a digital twin. These twins aren’t some sci-fi, dystopian, robot. They're a system to make industries work better.

Brian Amaral: How do they work and what's the benefit of a digital twin?

Roland Schaefer: Okay, so there are various, you know, stages of all of this. So you, if you've got a real-world item, and at some point, you know, you've got a drawing of that on paper. At that point you've got a digital blueprint. So you, you've just got some sort of digital drawing of this and you can make a 3D model out of all of this and all in the digital space. And there you've got a digital model of all of this.

So if you take this model and you now start augmenting this model with data, real time data that comes from the real world. So, you know, a digital model in itself will not change unless you physically go in and change it. If something in your real world changes, then you would have to physically go in and into your digital model and actually change the model. With a digital shadow, the real time data coming from the real world will now in real time change your model. So it's shadowing what's happening in the real world. And this is what most people, when they talk about the digital twin, that's, you know, mostly what they're talking about. They've got some sort of digital representation that is updated on a regular real time basis based on the real world situation.

Now, once you've started collecting all of that data you can start doing analytics on it in the digital space. You've got historical data. You can add more data to this. You can add weather data, you can add maintenance data. You can add all sorts of other data to this. Do analytics. And then when you start, based on those analytics, start automatically feeding back and changing your real world. So changes done in your model get reflected in the real world. That is the ultimate sort of building out of a digital twin. And that is where anything that happens in the real world is mirrored in your digital model or anything that happens in your model is mirrored in the real world. 

Brian Amaral: And that helps with diagnostics with certain things like, you know, to, to see if your, if your equipment is going to break down? And maybe predict when things are going to go wrong?

Roland Schaefer: So there are multiple different applications for this. You're describing the most common and the most obvious is predictive analytics. Trying to understand, you know, the equipment before, before it breaks down. So a lot of people use this to simulate and increase operational efficiencies. 

So I've got a chemical mixer, I've got two materials going in with temperatures, pressures, whatever else. I can model how should I change my temperatures, my pressures, my speed of, you know, adding the various chemicals to each other. I can model all of this. I can do simulations, and based on optimizations done in the digital space, I can say, okay, now I'm going to change the real world. How I actually mix these things together and how all of this works.

Narration: Roland says some new factories in Europe are being built with the help of digital twins.

Roland Schaefer:  So they did all of the, all of the programming and all of the parameters for a whole section of their production facility in digital space, they built the models, they did all of their simulations. They ran all of the, you know, conveyor belts and matching the conveyor belts and making sure the robots don't collide, all of these things. They did all of this in the virtual space, and then downloaded all of this once the building was actually built and everything was put in there.

And then they, you know, reduced their commissioning time they said from something like six months down to six weeks. A huge improvement in time in terms of getting everything up and running and once it's built to actually get it, functioning.

Training is another one. If you're training new operators, you don't necessarily want to train them on the real thing. If you've got a really good digital model, you can do all of this, all of this training, also with real time data. You know, have them react to things as they are happening in a digital space without actually affecting the real space. So these are just some of the examples. 

Brian Amaral: You mentioned there's some risks that we don't know about yet. Are there risks that we do know that this is going to bring and what are they?

Roland Schaefer: Well, cybersecurity is clearly, you know, the most obvious and the, and the easiest one to understand. When you inter interconnect all of these systems in a digital transformation, cybersecurity is, is quite high on, you know, the potentials here and someone having some sort of cyber attack that causes a turbine to overspin and, and, get destroyed. You know, that's clearly quite a simple one. 

Narration: But, Roland says, there are other real-world examples of digital risks, like companies using technology on their truck fleets—collecting data through a cloud application and running AI bots to identify when trucks have to go in for maintenance. 

Roland Schaefer: The question is, you know, what happens if your model's wrong? What happens if you know some sub-sub-sub-supplier of your models made a mistake? You know, how do you test that? How do you check for that?

Brian Amaral: And this is the sort of thing that's happening right now.

Roland Schaefer: There are some companies that are doing this. You know, a lot of these companies are at the beginning of their digital transformation.

So think about a control system. Most machines are controlled by PLCs or some sort of, distributed control system. It's a computer that's perhaps sitting right next to the machine that it's controlling. At some point that is going to be virtualized. So you're not going to have this computer sitting next to your machine anymore, it's going to be up in the cloud. And all of the signals from all of the sensors and everything are going to go to that cloud virtual control system. The decisions are going to be made there and perhaps at some point in the future, they're going to be made by an AI model as opposed to, just a deterministic program that sort of runs from top to bottom.

The challenge with those is if you get the same inputs to machine learning models or to AI models, you won't necessarily always get the same outputs. And if it's probabilistic and not deterministic, what does that mean? How does that concern us?

Brian Amaral: And by that you mean if somebody's using a generative AI model and asks for a pasta recipe, they might get two different pasta recipes.

Roland Schaefer: For the same inputs

Brian Amaral: For the same inputs. But when you're running a factory, that might have more implications.

Roland Schaefer: Yes. And what kind of implications does that have? How do you test for this if it goes wrong? How do you determine what went wrong?

Brian Amaral: I think people can imagine right now what a factory floor looks like. What's it going to look like in 10 years and how's that going to be different?

Roland Schaefer: That's a good question. If I knew that I'd probably earn my money on the stock market or somewhere else, if I could, predict that. There will be a lot more automation and a lot less human interaction. And I think that is certainly where it's going. There are some facilities where there are five people walking around the facility and everything else is automated. We did have a debate not too long ago about whether a factory will even, you know, have doors and those sorts of things. 

Brian Amaral: Or lights on right?

Roland Schaefer: Well, yeah, lights on is the easy one, but the humanoid robots, they look like humanoid robots. Potentially one of the reasons why they look like that is because they're mimicking the spaces that humans walk in, sort of up and down stairs, through doorways, those sorts of things.

But if I can design my factory around robotics as opposed to designing it around people. maybe, you know, the whole walking in and out of a building and those sorts of things is going to be a lot different. And so you know if the lights are on or not, you know, if there's even people in there. There are some warehouses where you walk into at the moment. They are highly automated. There's three people in the whole place. and you know, that already exists at the moment. Some of our clients, if you walk into their facilities, they already look like that.

Brian Amaral: What excites you about this? We’re talking about the risk, but there must be some big opportunities with this for our clients and for businesses more generally.

Roland Schaefer: Essentially, you know, we're looking at a whole bunch of different technologies. So, you know, we're at the cusp of all of these different technologies and, you know, we get to look at all of these and to try to understand, you know, what does all of this mean?

And it's a lot of exciting things to look at, and try to figure out. My life has been very much dominated by doing new things. Coming here to FM was me coming way out of my comfort zone in terms of different country, different language, a different industry and everything. And, you know I get to do this every day.

Brian Amaral: What does that research look like specifically? You, we were talking about some of the technological changes that we're trying to understand, but how are we doing that

Roland Schaefer: We're doing that mostly by learning, by doing. We don't want to just talk the talk and just read look at reports and go to conferences and everything. We are actually gonna be implementing these technologies ourself. So we want to walk the walk as well. To show our clients, you know, we know what we're talking about. Because we went through all of this already ourselves.

So part of what we're going to be doing is creating a digital twin of the Research Campus, of parts of the Research Campus, that are about as close as an industrial facility as we have at FM. And looking at what would our clients be going through as they try to digitalize a water treatment facility. So we're going to be doing those activities. We're going to be setting up a digital platform. We're going to be modeling all of the equipment. We're going to be looking at all the data, ingesting all of the data into the system, doing analytics on this, and looking at how can we– what do we learn from all of this?

You know, what sort of challenges did we have along the way? What sort of challenges do our clients have along the way? And these might be technological challenges, they might be cyber challenges, they may be algorithmic challenges. They may just be cost challenges of, you know, what does this cost to do all of this?

In Luxembourg, we're in the process of acquiring a number of these humanoid and dog and sort of, driving robots that are in automated factories and automated logistics systems, and looking at what do they mean? You know, nowadays you can get humanoid robots, they cost about $6,000. These things are going to be hitting the market. And you know, when they lose power and they lose energy, they fall over just like humans. They faint. What does that mean? They're going to be moving themselves autonomously through facilities. What does this mean? If there's something in the way and they have to make a detour, what sort of decision making goes into all of that? What does that mean if they walk into high hazardous areas? 
So, you know, there's a lot of trying and a lot of doing. And the idea behind all of this is fail fast. You know, we're going to try some things, we're going to look at them, then we're going to, you know, on a regular basis, look at our results and say, do we need to continue? Do we need to, are we on the right path? Do we need to change our path? You know, do we need to stop? There is no endpoint already defined. We're going to be defining the endpoint as we go along.

Brian Amaral:  Can you talk a little bit about the facility that we are building in Luxembourg? 

Roland Schaefer: So there is a new building. We did the groundbreaking in the spring this year, due to move in the summer, late summer of 2027. So it's going to be a large research facility, Approvals, the headquarters of Europe, operations, and, you know, training is going to be there and a lot of client interaction.

So we have a very large high bay facility there, in terms of the research, so we're going to probably going to be building in a small, version of an automated storage and retrieval system, just to have a look at, you know, what kind of risks and challenges do we have there. We'll have the robots running around probably, you know, maybe even bringing people coffee when they come in the main door. We've got a whole cyber lab that's going in there. And we'll be creating a digital twin of that facility so that you can walk through, virtually, walk through this space as well, and see what in real time, what is happening there. In various areas, obviously there's going to be security and confidentiality issues. But, those are, you know, the things that we're going to be looking at.

Brian Amaral: You mentioned procuring some of these humanoid robots for Luxembourg. What are they going to be doing?

Roland Schaefer: Well, to start with, they're just going to be running around and we're trying to see. What happens when you take three different robots from three different suppliers and you have them running around the same space? Do they walk into each other? 

There is this concept of fleet management, software that manages a fleet of these mobile robots, you know, autonomous, robotic vehicles. and you know, they typically get run by one fleet management software, but you know what happens if you bring in a foreign device into all of this? So that's just one example. We'll have them walking up and down the stairs and going outside and, you know, differing floor surfaces. And just trying to understand what, what can they do? 

Narration: Roland says they plan on getting a robotic dog, to test what it can do.

Roland Schaefer: How can we use this to do things like, fire pump reviews and have it sort of walk around in enclosed spaces. And can we use it to do, you know, checking whether you know, all of the valves and all of the, you know, gauges and everything are correct or not.

You know, it's just an example. We've also talked about the research campus, about what kind of applications might they have with this and to learn what if this works and what if it doesn't? How much of this is the marketing of these robotics companies and how much of it is, you know, truly usable and viable as a solution?

Brian Amaral: I think people might be familiar with the grocery store robots that are checking stock and shelves. But there's robots that could actually check on fire equipment to make sure it's in good working order?

Roland Schaefer: We are about to find out.