NISKAYUNA, N.Y. — Every year, General Electric gathers hundreds of its engineers under one large roof at its expansive Global Research Center campus outside Albany for a days-long symposium focused on best practices, recent developments and moving forward with new ideas. The brainpower collected there is, to say the least, impressive.
Part of the week’s activities last month included an hour-long roundtable with four top thinkers and leaders from across the company. What follows is a version of that roundtable, edited for length and clarity, a deep dive into the potential near future of digital manufacturing at GE and across the industry.
GE is rolling out its new Industrial Internet Control System. How does it accelerate GE’s transformation to the digital industry?
Jim Walsh, general manager, automation and controls, GE Automation: I guess the way I would summarize it is, it makes the outcomes real. In other words, when I had a software role, when I thought about data, it was a lot of, ‘How do I get the data together? How do I ultimately run some analytics against it?’ What I didn’t necessarily have was a way to get those insights back to the equipment. So what we’re talking about here is a world where you’re collecting the data, you’re applying a rich domain expertise against it, … and then you’re able to bring those insights back to fundamentally change the way that an asset or process operates. Customers care about outcomes, not necessarily the technology or the theory.
From a perspective of its strategic importance, how is the digital twin important to this effort?
Colin Parris, vice president, GE Software Research: If you look at the digital twin itself, it’s a digital involved in the physical asset that actually uses a constant stream of data to deliver business outcome. So when you combine that with the control’s capability, which gives you real-time capability, what you have is a constantly-adapting system. So I can predict, or I can optimize, but regardless, as the data comes back, I adjust the twin model, predicting and optimizing with the current state. That constant stream allows us to customize it as close as possible to what the customer actually needs.
The twin system is a set of model management techniques. You can build a model, but how do I know when to update the model? How do I know when to retire the model? How do I know if the model is valid? These things have to be done by building a modeling system, which is what we’re doing here at GRC and transitioning to Predix, and we have to use them effectively. It goes beyond analytics. It’s the entire system.
How are you thinking about what these pilot programs teach you and what they do for customers?
Scott Bolick, head of software strategy and product management, GE Power & Water: It’s a tremendous collaboration between the GRC, the business and the customer. Every time we think about the (Industrial Internet Control System) or the digital twin, we have to think about it through our customers’ lenses, through their outcomes. … One plant wants to trade off light and capacity on peak days, and we’re able to give them that 2% to 3%. The starting point is understanding the light, understanding the thermal performance, understanding the transient performance of those machines, and if you have that you can run really complicated analytics. You can make the decision of what your capacity is and how you should optimally operate that plant. … The algorithms we were originally running took over 40 hours to run. Now they can run in seconds, pretty much continually.
GE has a long history of controls. Almost every GE company uses controls to keep their equipment safe and secure, for decades. People are getting excited about the industrial internet. What is the benefit of using GE’s automation controls rather than those from a third party?
Walsh: Well, ours are much better. (Laughs). To be more specific, I would say … when you think about the history of controls and their fundamental purpose, it’s heads down, make sure that that 30-millisecond deterministic loop repeats over and over and over again without fail — and that doesn’t change moving forward. Think about a world where the controller not only does its job as it relates to that deterministic loop, but has the ability to look up, with all the data sources, with all the domain expertise, and figure out how to disseminate that into outcome. That, to me, is the big difference.
There’s been a lot of internal discussion about how to do that well. What can you say about some of the senior leadership conversations in this space?
Kishore Sundararajan, CTO and VP of engineering, GE Oil & Gas: There’s a lot of excitement and anticipation — ‘What can you do? What’s next?’ — and I think we’re in a place where we have to drive faster. That’s the threshold we’re at.
Bringing in the Internet of Things
With the respect to digital, how would you talk about this controls activity dovetailing into the Internet of Things talk that we hear about all the time?
Parris: Initially, IoT was the notion of connectivity, right? Then, with connectivity, it gave you data, and data gave you insights into things. When you combine that with controls, you have the ability to actuate, and I think that is extremely powerful — and it’s not just what you want to get done, it’s the machines themselves interacting. …One example I think is interesting is if you have a wind turbine at a wind farm and a new wind turbine shows up, that new wind turbine now asks a question, say, ‘What’s the prevailing wind direction?’ The older wind turbines can answer that question. That connectivity of machine to machine enables that new machine to know something, and it allows the machines to find ways to connect. That’s different. As that begins to happen at ridiculous speeds, they can react and they can transform themselves and adapt to what you need — optimizing for productivity, or for revenue. The machines are learning from one another at a speed we’ve never seen before.
This system was introduced about a year ago. What has been accomplished and learned since then?
Walsh: We’ve relatively quickly transitioned from concept and PowerPoint to real product. We have to go faster, but I am proud of that. It shows an ability to execute and collaborate.
What’s next?
Bolick: Three things. The first is, we have the initial pilots underway, and we have to make sure we continue to learn with our customers and making sure what we put in place is going to have real value. … The second is just modernizing the controls we have and making sure they’re running on the IICS. That’s critical, not just on the initial install, but on the updates, too. The third is, we have to figure out what those next outcomes are. When I talk with my team, we always say that our mission is to help our customers optimize their profitability through time-sensitive recommendations that push the plant to the outer limits of its operating envelope without impacting down time. That is the mission. How can we make sure that safe, reliable, secure, affordable power is there, and make sure our customers are optimizing their profitability?
What are some of the R&D activities that are going to move this technology forward?
Parris: They fall into a couple buckets. The first is this notion of the models themselves. We have models that are combinations of physical and digital capabilities. … We have to blend the physical and the digital. The second is determining how many models will you have. If you have thousands of models with different characteristics, you will have a huge number running simultaneously, and you want to use those to optimize, some running at the plant, some running in the cloud. How do you distribute that in a way that makes sense? How do you not send data all the way back to the cloud, because that’s a tremendous cost? The last notion is the sensors themselves. What data do you want to select, and how much data do you want to get from these sensors? Sensors are expensive, some are $20 million. How many do you expect to put there? So how do you create virtual sensors that give you access to that data? These are all hard problems that will not be solved in a garage. This is a very different space. These are the problems we’re working on with our partners.
What are some of the important trends that may be affected by industrial controls?
Sundararajan: It’s a broad question, so let’s demonstrate through an example. Let’s take an offshore platform. It’s a harsh environment, it’s harsh for people — safety is always on top of the list — and the architecture system we have allows for unmanned platforms. Unmanned platforms today are no longer a challenge. We can do them using a combination of Predix and IICS. It’s important for two reasons: The cost of the platform, the size of the platform, the safety criteria, all of them are affected by a redesign. And if you look at the demographics of the oil and gas industry, about half of the working population is eligible for retirement in the coming five years. That means we need to be able to bring data to areas where expertise is shared. That’s what this enables. That’s a specific example.
Who else should be thinking about this?
Walsh: I think the beauty of it is there is not an industry I can think of that’s not looking for outcomes. The fundamental concept we’re talking about — in terms of collecting data, being able to apply domain expertise to it, being able to close that loop and change the way an asset or a process is running — is ubiquitous. It translates across all industries. … When I ask my team and they say, ‘Well, Jim, every vertical is applicable,’ I want to jump off a bridge, but I guess that’s the answer I’m giving to you today.
What Took You So Long?
Whenever you get something like this out there, there’s always someone who wants to know why you didn’t have ready yesterday. If someone asked you, ‘What took you so long?’, how would you respond to that?
Parris: One is, obviously, the data. You had to collect a significant amount of data to make this feasible. We collected data before, but more in extreme situations, when there was a problem. The cost of collecting that data was onerous at times, quite high. So I think the data and the digitization that’s happening — not only with us but with everyone — it’s a big factor. The second thing is the computing power. The computing power and the storage has allowed you to do things now with algorithms that are quite different, and because of that computing power and storage, you have the chicken-and-the-egg problem: new analytics showed up that allowed you to compute. You have the data, the computing capability, and you have the actual algorithms, and you put the synergy together. In this case, when you add 1 plus 1 plus 1, it doesn’t give you 3, it gives you 7 or 8. Then the more data you collect, there’s an exponential rise.
It’s not only that the technology’s coming together, it’s the understanding in the marketplaces that if you make this better, you, alone, win. That’s what you see with the Apples, the Amazons, the Googles. Everybody’s rushing to it right now, because whoever gets there first has a shot at being the only one on top.
What are some of the areas to focus on?
Parris: There are new techniques in analytics and AI that show up every two or three years. Unlike the other fields, this is not an emerging field. When I did my graduate studies years ago, there was stability. You had a textbook, and the textbook was the rock for many years. This was the way you learned. Now you look at deep learning, some of the new things coming out every two years — the half-life of your knowledge may be 16 to 18 months. So some of the latest things that are coming out, we can test them for you, we ensure these things meet the requirements you need and give you a view on it. That is vital.
Sundararajan: The way I think about it is, standing on the shoulders of giants. Without this (GE) foundation, I cannot do the application for the oil and gas industry, at scale, at speed.
Bolick: We all believe in the GE story and horizontal. It’s critical to our success. … We have the obligation to be upfront, defining the requirements for each of those components. We’re not on the sideline, getting something back from businesses, and then saying, ‘The digital twin doesn’t quite meet our needs.’ That’s complete failure on our part. We have to be engaged. … We have the obligation to get as many assets connected so we have the richest set of data in the world, and extend the digital twin running analytics to where we can say we have the most comprehensive set of analytics on top of all of that data. And we have to build those applications that deliver outcomes.
What looks different in a year? Or in five years?
Walsh: I’ll pick one year, because in five years, I have absolutely no idea. I would say, my business looks fundamentally different in about every way that it historically has looked. To start thinking about a disruptive value proposition that enables really meaningful outcomes across the entire GE corporation, that’s a heck of a big shift. I still think we have a tendency to have development projects that last a long time. How do we get into a mode where we’re more comfortable sprinting? We can’t cure cancer with every release, but instead we’re trying to make incremental yet impactful improvements.
Parris: I’ll go with the two-to-five-year range. What I’m looking for is this notion of learning, the twin’s beginning to learn. That’s important, because once that learning begins, we get on a bridge nobody can touch. So the operational and environmental data that’s coming in, they modify themselves. … If a machine has experienced something, and another machine is new and asks how it can get that configuration to be more optimal, could a machine tell the other machine what it is doing? That’s another way of learning, machine-to-machine learning. And what about machine-to-human learning?
Sundararajan: I would like to see the organization become faster and more agile, and I would like to see it, one to two years from now, think in sprints and incrementals. The other part is, how can we go from a product mindset to a system and outcomes mindset? That’s a key piece I would look for.
Bolick: We have to start talking about solutions and outcomes, and not get religion around something being in the cloud or on the edge. The reality is there are natural locations for different analytics and actions, and we all have to continue to question ourselves about where we should put the analytics, where we should put the action, and put it where it belongs. We can’t talk about just the software and leave the hardware alone.
