 Welcome back everybody, Jeff Frick here with theCUBE. We're in downtown San Francisco. The GE Mines and Machines Conference, about 3,000 people at Fear 48, talking heavy industry, really, where software is eating the world in the form of GE Industrial Internet. It's very exciting times, and excited to have our next guest, Colin Parris, the VP of GE Software Research. Welcome, Colin. – Delighted to be here. – Absolutely, so first off, just kind of what are your impressions of this thing? It's like 50% bigger than last year, it's a great buzz. – It's great, it's great. I've been here for the last two years, and I've seen a rapid rise. I think the big difference for me, before we were actually having a discussion about, is this a relevant topic? Now it's all about how do we take the partnerships we have, and take it to the next level. I think everyone's convinced that there is a digital industrial space. How do we get that as fast as possible, moving in the right direction? – Right, and Jeff clearly said this morning in the keynote, he's all in, in it to win it, I think. – Yes. – Using the great sports analogy. So your area of expertise, we talked a little bit before we got started, is the digital twin. – Yes it is. – And we've heard a lot about the digital twin, so for those that aren't familiar, just give them kind of a quick overview. – Okay, the digital twin is a digital representation of a physical asset, of a specific asset. So what you have is a combination of models, and because you have a continuous stream of data coming at those models, that model now adapts to the environment, and so you get a precise representation of a physical asset that exists. And because of that, you're able to do some interesting business things. Like you're able to predict warnings of problems, you can predict the damage or the opportunity, and you can then optimize to get the best possible results. – Right. – So that's what it is, a digital representation of a physical asset. – So it's interesting, we've been to G and talked about this before, so specifically for say like a jet engine. You know, if jet engine number 0001 serial number goes to an airline that operates out of North Africa, and jet engine 0002 goes to one that operates out of Antarctica, same engine, same airplane, very different lifetime characteristics, and this enables you guys now to, or the operator, to manage that engine within its own specific environment as opposed to just open it up the standard operating manual and change it in the oil every 5,000 miles. – Exactly, exactly, because what we're looking at here is we realize initially when you have a model, where you operate and how you operate changes the model. So we're taking data like the operating mode. How fast do you ramp up when you take off? You know, how fast do you get to a climb? How often do you climb? That's an operating parameter. Then the environmental parameters, what city pairs do you travel in between? What's the temperature, the pressure, what's the dust level? All of those shape the model. And so what we do then is we actually tune and change the model based upon those characteristics. The other thing we do is we learn. So we look at other planes in the fleet that have certain specifications like this one, and if you've discovered a failure in that, you send that technique across, it's called similarity techniques, and it allows you to incorporate that into the model. So before this actual entity, this actual asset has a problem, it can get that learning about the problem. So it's adapting based upon its knowledge and adapting based upon learning from the fleet. And that's the digital twin, that's what it does for you. And who's managing kind of the collection of the data to that twin and then the analysis of what I'm doing with that data in the twin. Is that the operator of the airline? Is that you? Oh, that's G.E. right now. As part of our services packages, that's what we do. We have 550,000 twins right now, and the twins we have are the people that have elected to have an actual service agreement with us. So in many cases, people after they've bought these engines, they decide to have a service agreement. In the service agreement, what we're doing is constantly monitoring the diagnostics from the engine. In fact, every time the plane takes off and lands, we get a snapshot. A snapshot is a piece of data that says, here's the right set of parameters for this situation. We can use those to tune the digital twin every time it takes off and lands. So then, so that's kind of looking backwards based on the climate, the way that's been worked. But you can do forward-looking things too, right? You can run scenario analysis, you can do all kinds of stuff, because it's digital, right? Exactly, so what we also do is, given the data that we have collected, historical data, we can run simulations. Now, the other value of having this digital twin capability is that normally simulations take many, many hours or days to run. We've had a technique called surrogate modeling in which we can extract, say I'm specifically looking at the damage on a blade, I can extract the right parameters and get a much smaller version of exactly the model focused on blades itself. And that smaller version we only run in a few minutes. Now I can go get on a supercomputer or a lot of cores and I can run 10,000 simulations and learn precisely in the future what the damage impact would be. So now we can actually take this and fast forward. And we can forward a few months or we can forward a few years. It helps you plan for a number of things that you have. Right, and then of course predictive maintenance, I'm sure that's a no brainer, right? But you could also plan for fuel changes. You could also plan if I change routes, what would the damage be if I'm flying through these atmospheres? You can decide what routes you run and what planes you have and from that how much profit you get. Very different approach. Yeah, because the other thing I find fascinating by this is again, if you use kind of your standard operating services manual, you optimize to whatever it was written for in the first place, whether it's longevity or meantime between failure or whatever it is that you chose to optimize for. But in the reality, in a dynamic situation, there's a whole bunch of things you might want to optimize for in a particular situation. Exactly. Low fuel, high fuel, expensive routes, all types of things that you can then plug into this model. But you had to initially, right? Because initially what happens, you didn't know. So you made assumptions. I made assumptions on the operating mode, assumptions on the environment. And I would say the assumption would mean the life of this spot is two years. But now that I know the exact operating mode, the exact operating environment, I can say the life of this particular engine is five years. Now I operate very differently. Once I know that, the economic changes. Right, right. And that's the key for us. Also the changes in weather patterns, right? There are things that happen that you don't know about. The jet stream suddenly shifts down. Well, they say there's no global warming, but for whatever reason, the jet stream is shifting. And now you're flying into a different jet stream. You could not have anticipated that. So there are new things that are happening, right? They're opening a bunch of new airports that are in parts of China that could be dusty. You couldn't have anticipated that. So how do you react to it? So you need a twin that changes and adapts to your environment. And then so aviation is an easy one. What are some of the other areas inside the GE ecosystem where you're seeing kind of this adoption of the digital twin and where is it really having the most, I want to say surprising impact. I'm sure the impact is huge all over, but things that you didn't necessarily anticipate or that the clients really had no visibility into. Well, part of what you see is also in combination. So for instance, you have a wind farm and you have a gas turbine. Right now that over the last few years, people are going to more wind power, we find that they're putting up more wind farms. So now you have a digital twin of the wind farm to estimate the damage. Since you're using a wind farm, you use your gas turbine a lot less. So normally you would have had it running at 60%. But because of the wind farm, now it's cycled. Sometimes it runs at 20% at night because sometimes it has less wind, you have it running higher. No one ever intended for these things to cycle as fast. So as they cycle as fast, more damage occurs on the parts. So now you've got to sort of look at a twin and say, can the twin begin to predict for me the life because I'm doing a lot more cycling. Can the twin predict the life of the rotating parts on the wind turbines because that's happening? So now the twin is a combination of uncertainties, right? And that becomes very interesting. So is it a system twin then? Do you create a system twin for the combination? Exactly, you have system twins that can do that quite easily for you. Yeah, it's interesting. You know, one of the concepts of the IT that we've seen from the hyperscale of people like Google and Amazon, right? Is you no longer treat, it's pets versus cows, right? You no longer treat the server as a server or the drive. You know, you move the whole rack in and out. And Jeff mentioned in his keynote, starting to think really more in terms of systems and aggregate as a combination of the whole. And that's really what you're talking about. These are systems that were never designed and specced to necessarily work in conjunction, but you need to analyze the function, the delivery, the optimization at a system level, not necessarily component level. Oh, definitely. I mean, we see that in spades. Now you could also extend the system to be something even larger. There's an ecosystem around that system. So for instance, what we find out is that in many cases in the oil and gas industry, they are groups that operate, that make changes to the actual machine because they have service contracts. Sometimes when they make those changes, they put in different parts. So they put in the wrong parts. Now that ecosystem is affecting you and you as the owner never knew it. Because what you did is you bid on the lowest possible contractor that you can get. Right? For that one particular- For that one particular part. Service part. So now it radiates outwards beyond just the system itself to the people interacting with your system and the suppliers you have. So this never ends. And as far out as you go, it's the more that you can actually get the right amount of utilization and extract as much cost as possible. Law of unintended consequence is one of my favorites. And the other is you said kind of what's the logical economic unit by which you want to manage the system? Because as you said, there may be an instance where you just are burning the crap out of one little piece of component because in that total, that's really still what you want to do. Exactly, exactly. But we have twins that go from parts all the way through. From parts to components, groups of parts, to the actual machine, to systems, machines together, to ecosystems or processes. I can look at a twin of a process, a manufacturing process, and I can say the manufacturing process is supposed to be this way because I've done enough analysis of it. Now if that process changes, so for instance, you're missing a part or you didn't order enough parts, I can recognize that as a warning and say, oh, I have a warning because the twin of this system says that you don't have enough parts, then I can predict the failure that you're going to have in the system because your trooper drops, and I can optimize that. So that we've used the twin actually across processes. So the twin works, the concept works across all of the things. That's why with ServiceMax on new introduction, the twin works in ServiceMax as well. Not just assets, processes as well. So how small of a component can economically support a twin, whether that be by value, usage, how small can you create, economically create a digital twin? It's all based on value. The digital twin process is I look for where there's the biggest problem financially or problem from an operational level and then I go backwards. If the problem happens to be in a single part, for instance, the jet engines or in the steam turbines, what you have are the parts that sit in the hot gas bath, the parts that get the hottest, right? Some of these parts are fairly small in terms of what they do, right? They may be a couple inches wide, but they're vital to the process down there. In the jet engines, you get the bearings. The bearings in the jet engines, the number four bearing and the number three bearing have a significant impact. Oh, it's a significant impact. If the bearing liberates, the blade begins to shake. Tremendous problems with fuel economy as well as damage on the engine. So because the value is where we start from, the business value, and you go backwards, you end up in very different places and it could end up into a number four, a number three, a number one bearing. The other very interesting thing we were talking about briefly before is open. And again, Jeff and Beth will talk about open, open systems, open source software. The fact that you are willing to open up kind of the analytics capabilities, open up API so that it's a classic case. Not all the smartest people are in your four walls to let other people apply analytics to the digital twin. That's pretty interesting. I think it's also a pragmatic point of view, right? So I've had experiences in other industries, as you well know, for instance, in the banking industry. There was a time in which people hoarded data. They thought data was vital, let me keep all my data. But then the tier two banks realized that they put their data together, they could discover things no one else could, like they could discover people who were committing fraud against them. Because how fraud is committed at that level is that I commit fraud against one bank and I'm hoping that bank doesn't tell the others so I can try the same technique. So when you put all the data together, I could quickly discover what fraud was happening. In our world, we see the same thing with failures or things that affect safety. If we put our data together, all of a sudden I'll detect when a failure is happening and if I can tell you a failure before it occurs, I help you. So you're helping each other and especially with safety. You can't do enough to help safety. So there'll be times in which, as you bring that data together, you bring the analytics, we can discover ways in which we reduce failures as a group that helps everyone. We have much more safety with our employees that helps everyone. So coming together, it's not just an idealistic view, it's a pragmatic view that says, in these industries, lives are at stake. You have to think about this very differently. And if you've got a cloud, by rule, you have to be open to get people to participate in that cloud. It's the network effect that brings the value. So if you look at digital twins in terms of the value delivered as a percentage, how much of it's really kind of operationalizing and making sure I'm taking care of that particular engine and doing the right things versus really providing the data for the next generation engines that you guys are really proactively using really to build and design, look forward. 50-50, 20-80, what do you think? Well, let me go on a journey with your question and then give you an answer. That's to show you where I'm going, right? So for instance, as you think about digital twins, what's the next level for us with digital twins? Here's an interesting thing. We know that Google never writes white webpages, but their value is in putting them together. So the same thing for Amazon. They never produce products, but their value is in putting them together. So how do I get machines to contribute into a network that makes it attractive for other machines? For instance, you have a wind turbine operating on a wind farm, and a new turbine shows up. The new turbine says, what's the prevailing wind direction that I should tilt my blades at the right level to get the maximum power? Who does it ask? It asks the old wind turbines. If a new... That is vision of the old man on the mountain taking care of the little guy. If a new, you put in a new gas turbine, a new gas turbine says, I've got this digital signature. I'm not sure if it's a failure or somebody's hacking my security. Who would it ask? An older gas turbine. So now the turbines are talking together and sharing information in such a way that the entire system is raised. When I begin to do that, two things surface. The machines attract more machines. You have to join the network because that's what the knowledge is. The next thing that happens is that these machines get together and find the best practices. They didn't tell those best practices to designers. The designers can now build a next best machine. The machines are helping to design themselves. That's where you want to go with this. And how much of it's done with the machines in the field versus their digital twins back in the data centers? The digital twins are the ones that are talking because the minute you give me the data and you say this is the prevailing wind direction, my machine may be slightly different from yours. I may have a newer version of the blade. So that data goes to the twin and the twin says, ah, they said till 10 degrees but for you it's eight degrees. So the twins are the ones that are talking. The digital counterparts have the discussion. And they're the ones who feed the designers to build the new systems. So in the end it all comes back to these twins working together. And so you grab my next question. It's going to be, where is this going to go next? So we've already answered that. So before we wrap, you know, as you kind of look out into the future, what do you see as next? Is it just more penetration of digital twins further? Is it this integration of the communication? What are you excited about? I see the penetration of the twin and the integration. But I go back and I look at 10 years ago when three billion people on the planet got connected through the Facebooks and the Googles and I saw what occurred. We saw businesses being transformed. We saw elections being transformed. We saw education being transformed. Windows, now it's going to be about five billion people connect with 50 billion machines. What won't be transformed? Every single thing on the planet. Now I know exactly how much electricity I need for this exact neighborhood. Now I understand how I deal with certain health problems and how they arise. Now the interconnection becomes something enormous. Now the planet changes. Sighting times, sighting future. Definitely. All right. We'll call them. Thanks for taking a few minutes out of your day. Really, the digital twin. Pay attention. That's where a lot of exciting things has happened. Come join us. Absolutely. All right. Call in Paris. I'm Jeff Frick. You're watching theCUBE. We are live at the G-Minds of Machine 2016 Pure 48 San Francisco. Thanks for watching.