“Things must keep flowing, or bad things will happen,” says John Carrier, a man obsessed with control. With a PhD in chemical engineering from MIT earned with a thesis on control systems, Carrier’s words neatly sum up his unique perspective on problem solving.
They’re also a pretty good tagline for the MIT course Carrier has been teaching since 2017, Implementing Industry 4.0: Leading Change in Manufacturing and Operations. In it, Carrier urges his students—which run the gamut from factory floor employees to C-suite executives—to stop thinking of Industry 4.0 as a technology problem, but as (what else?) a control problem.
“Industry 4.0 is making the cost to implement these control systems actually affordable,” says Carrier, now a senior lecturer on system dynamics at the MIT Sloan School of Management.
Engineering.com spoke with Carrier to learn more about how manufacturers can take control of Industry 4.0. He shared his thoughts on the changing nature of technology, the concept of “hidden factories” and how to find them, what Industry 4.0 has in common with the human nervous system, and why an old Air Force acronym should guide your technology investments.
The following interview has been edited for clarity and brevity.
Engineering.com: How do you conceptualize Industry 4.0 and what’s your approach to teaching it?
John Carrier: Think about machines, equipment, chemical plants, vessels, engines—we have such huge systems. But what’s actually weak is the information flow. I like to joke it’s easier to move protons and neutrons through a plant than electrons.
It used to be that sensors were expensive. To get one sensor hooked up, it would cost thousands of dollars. And even worse, you’d have to wait five or six weeks to get an instrumentation person to hook it up, and then you’d have to maintain it. The cost of that has fallen substantially. Now you can buy a industrial class sensor for a few hundred dollars that magnetically attaches to a vessel that measures vibration and can handle 200 degree temperatures, and have it up and running in 30 minutes. And that’s just one example.
So that’s the technology side. Now I want to talk about what’s on the other side that the technology is fixing. There’s a concept that came out of the MIT Sloan School [of Management] in the early 60s called “hidden factories.” It means that 30 percent of all effort in any system—the labor, management, attention and things like that—is spent on things that have nothing to do with making profits or serving the customer. So that’s rework, speculation, poor analyses, allowing defects to persist.
That happens because we’re trying to make the system work faster than it can and we can’t see into it. If we had visibility into what’s going on, we could improve it and generate increased profits by 20 to 50 percent. And we can also reduce risks. So that’s a huge win on both sides. But you can only do it by gaining visibility to what’s going on in real time on the factory floor.
So the way I run the course is not simply teaching technology. I want people to leave and say ‘I have that hidden factory.’ So if they go out and implement this system, or these parts of the technology, within three to six weeks they’re going to have it pay itself off. And then you keep repeating that. So instead of a big bang approach to buying all this new technology, you keep adding feedback loops and analysis and shrinking these hidden factories and you see results almost weekly and monthly instead of yearly.
What should manufacturers focus on first?
One of the most grossly undervalued areas is maintenance. Part of the reason why maintenance always struggles is because, traditionally, it’s very difficult to say ‘this is going to break down next week, so we need to shut down.’ Vibration sensors, real-time thermal analysis with cameras, temperature sensors, and measuring timestamps—these technologies are now affordable.
I’d say for the first 18 months, 70 to 80 percent of the technology or data you need is already in your system. You just haven’t gone and analyzed it or no one’s thought how to apply it.
What’s your advice for collecting and analyzing data?
You have to understand your own system. This isn’t just about getting data, you need to know where your system is not performing as expected. Usually the key indicators are you’re late with a product or service, it came out defective, or when you go to start the equipment up it’s not ready or it’s not at operating capacity. Those are warning signals.
What Industry 4.0 can help you do is go down from the 50,000 foot level to a piece of equipment, and talk to the people and collect photographs and thermal images and information and timestamp data, which is really important to figure out what’s going on.
Match the technology to a system, not the system to a technology. Diagnosis is key. A lot of these sensors are really about feeling pain. I often use a human analogy. Have you ever heard the term referred pain? I was talking to a colleague of mine and they had horrible pain in their lower back. To make a long story short, their heart valve was infected. But you don’t have a sensor there, so the pain showed up downstream.
Our nervous system is generally very well developed compared to a typical factory. Industry 4.0 means you can now afford to build a new nervous system, which allows you to run your system faster and make more money at lower risk. But you can’t just build any nervous system, you have to match it to the system you have.
How do you ensure you’re building the right nervous system?
A lot of people believe Industry 4.0 is big and if you don’t invest in it, you’ll go bankrupt. I believe that’s true. But the other part is, if you do invest in it, you might go bankrupt as well. It goes back to these hidden factories. Do you have them? They will tell you where your nervous systems is broken or weak.
Let me give you one quick example. I worked with Heineken. It took them six hours to change over the bottling lines. We got it down to 15 minutes. We basically put spreadsheets online with some simple algorithms. It was the feedback loop that was broken. I think the investment in terms of technology was probably zero, because we already had it.
I advise people to always go look at your current capital budget. Those are problems that you have that are persistent, that people are currently trying to solve traditionally. Heineken wanted to buy two more bottling lines. That would have been twelve, eighteen months and like five or six million bucks. Plus, they would have run at just 70 percent efficiency because we didn’t fix the problem. So look at your capital budget and see if that’s really a hidden factory problem that can be solved with a sensor or real-time analysis of data.
You’ll find that these problems generally cross company functions. One group is collecting the data, another group is storing it, and a third group is actually suffering from the consequences. By tracing ourselves through that information loop, the answer can often appear.
The syllabus for your class mentions the OODA loop. How does that apply to Industry 4.0?
[The OODA loop was initially developed by U.S. Air Force Colonel John Boyd as a means of understanding aerial combat.]
First of all, I’m a control engineer. And the theory of control says I need to know two things: gain and the time constant. The OODA loop brilliantly took that time constant and broke it into four sequential steps: Observe, Orient, Decide and Act. It’s not just for warfare, it’s also for serving customers. If you’re my customer and my time constant to deliver to you is faster than you need it, I’ll get the job.
One of the things Industry 4.0 does is it dramatically reduces the cost to make an observation, and then that speeds you through the rest of the loop. I think of it like I’m buying an observation. If I can measure or observe something, what would it be? And then how would I use that information? That’s orientation, and that’s really like your algorithm. You’ve got to take that information, you’ve got to filter it, and you’ve got to match it against a model. What did we expect to happen and what are we going to do?
So you need algorithms to handle that data. They can be simple algorithms like a control chart, or more complex machine learning, but you need to analyze that data. Otherwise it just sits in a lake. Just doing the observation and storing it in a data lake is insufficient to get value.
Now that we’ve had a better observation, and we know how to interpret that data—which is the orient step—we can make better decisions. And if we make better decisions, we get better actions that cost us less money and create more results. So I’m trying to buy my way through the OODA loop. And relatively speaking, at least for the next five to ten years, your money’s best invested in the earlier stages rather than the later. Robotics will be act. Everyone loves robots because they’re cool, but that’s probably not where you want to start with your investment.
How do you think AI might impact Industry 4.0?
Think of it as a control problem. To control something I have to have a model of it and understand it. Now if I could use artificial intelligence to give me a better model, then that will take me to the next level. So it’s going to be a movement from AI as a technology to understanding it as part of a control problem.
I think that’s a much more useful framework for senior leaders. Their job isn’t to buy technology, their job is to adaptively control their organization. And if AI helps us generate more options quickly and evaluate those, then that will improve their ability to control the organization.
What’s one thing you want everyone to take away from your class?
Go see your system and diagnose a real problem. It typically is going to fall within maintenance failures, delays in communication, disappointed customers, or people under stress at their jobs. Find that chronic problem and then see if you can find the right technology to relieve it.