 On the ground, presented by theCUBE. Here's your host, Jeff Frick. Hey everybody, Jeff Frick here with theCUBE. We're on the ground in Santa Monk, California at the headquarters for GE Digital. It's their West Coast Silicon Valley presence. Up to like 1,500 people growing like a weed, really bringing in a lot of software engineers, software expertise to help on this transformation that GE's going through within the industrial internet. And we want to come out and get the story like we love to do. And we're excited to have Jeremiah Stone on. We're going to get GM Asset Performance Management for GE Digital. Welcome. That's right, thank you. Happy to be here. Absolutely. So we just came off the Predix show down in Vegas last week. One of you can just share your impressions. 1,700 people, a lot of good buzz, a lot of good energy. And it was the first ever, if I'm right. That's right. Kind of with our broader journey, we've done a lot of stuff internally and now we're unveiling things that we're really excited about to the rest of the world. And so bringing in now more and more external folks, customers, just general people from the ecosystem, people that are excited about this possibility of the industrial internet, and tremendously exciting. And for our business, for Asset Performance Management is even more exciting because we actually helped to kick off the entire conference with a 24-hour hackathon using Asset Performance Management with both GE and non-GE teams to develop new applications. And it was tremendously successful. We had, I think, about 200 people, 24 hours in a room, and they came out with some just tremendous applications ranging from tracking people inside of cars on hot days to alert the car automatically, roll down the windows, these kinds of intelligent activity at the actual place where you need it, to jet engines talking to each other in mid-air or deep analytics on turbines, et cetera. So it was really exciting for us. So you're right in the middle of this journey because Asset Performance is probably kind of the top line expected behavior and benefit that people are gonna get out of Internet of Things and the industrial internet is how do I get more out of those assets? And you're out in the field talking to customers. So there's the vision, right? And then there's the reality when you show up and things are dirty and you gotta wear a hard hat. So I wonder if you can share some insight. What are some of the real significant challenges that either you didn't expect that people would have or that you're finding once you're in the field and people are trying to adopt some of this technology? Absolutely, well, as you pointed out, Asset Performance Management is really, for many, many of our customers, really the starting point on their journey in the industrial internet. And that's for many reasons. One of them is very easy to describe, no unplanned downtime. In other words, things should not break when you don't expect them to. Industrial operators hate surprises and so being surprised by any kind of equipment or asset failure is not a good thing. And unfortunately, even despite all the trillions of dollars we've poured into our infrastructure globally, things still happen that we don't anticipate, things happen that we don't like to have happen. You have production down situations that is bad from a business point of view. It's bad from a safety point of view. People can get hurt. It's bad from an environmental point of view. You can have spills, et cetera. And then any of these more exciting use cases that we talked about in terms of optimizing your business, optimizing your operations are really not possible if things are just breaking out of the sink out of tomb. And also what we've learned is that the data and information you need in order to provide maximization of reliability and availability for your equipment is the core data for doing all of those other interesting use cases. So it's both easy to describe and understand the business case for and foundational for nearly everything else we do. So we're kind of at the center of the storm right now and loving it. Yeah, absolutely. And then kind of the marriage of this kind of IT and OT and obviously G's been involved in OT and operations of this equipment for a long time and now really bringing the cloud to bear. And G a number of months ago announced predicts and industrial internet focused cloud. But as we've discovered, it's not just about the cloud, right? It's really this continuum from the machine and the edge, if you will, as you guys are talking about all the way back into the cloud, you can't just, as everybody has to say, right? Speed of light is just too slow for these applications. It's a good point. And that's one of the things that is just, I guess for us, it was just assumed because we have over a lot of our businesses after a sales service. So our relationship with our customer doesn't end when we have a transaction around a large piece of equipment. And in fact, it usually begins and it's a decade long, multi decade long relationship because of the lifetimes of these assets. And so we have over 60 monitoring and diagnostic centers globally. We monitor over a trillion dollars worth of equipment deployed globally, if you think about it from a replacement value point of view. And we've been doing so for decades. And so we have decades of experience connecting to machinery at the edge, providing capabilities on the machine itself, advanced control systems, distributed control capabilities, and then bringing that data back for advanced analytics. And so we really look at it as a continuum. We talk about it as edge to cloud, control system or sensors all the way back. And it's really, when we say industrial internet, we're very intentional about that because it's a network. It's not a hub and spoke. It's really a distributed network of data and intelligence. And you have to be very thoughtful about where you place compute, where you place storage, how you use that. And we've experienced that as an equipment provider, our customers, our challenges are just a drop in the bucket for what an operator faces. Because of course, we're looking at specific asset classes of equipment. We understand it very well. We built the things. We can maintain them over time. If you're an operator, if you're upstream oil and gas, you're drilling, you're producing oil, you're a transportation provider with locomotives, an airline, healthcare, et cetera, you're dealing with a blended environment with lots and lots of different equipment from lots and lots of different people. And you're trying to get it function as a system. And when we talk about no unplanned data, it isn't just one thing. It's the entire system. Because of course, if you're looking to provide a service or a good to your customer, you're only as strong as your weakest link and you have to then look at systemic health around that. And of course, it's all about the assets. And so if you only think about the cloud or you only think about the analytics software, then you're gonna miss the point entirely, which is your really connected system of data analytics and work processes that take us to a new level of productivity. Right, so let's kind of break that down by the kind of the three core computing anchors, if you will, pillars. So the first one I wanna talk about is the data itself and the storage of the data and the massive amount of data that these machines are kicking off. And as was talked about in Vegas, you guys have a really significant needle in the haystack problem because you're producing a ton of hay that's coming out of these machines and they're generally pretty reliable. So your needle is even smaller relative to the size of the haystack. When you're thinking about Edge and this continuum back to the cloud, what are some of the processes customers need to think about where they store the data, what data do they store, how much of it do they analyze locally versus how much are they sending back up to the cloud, if you will? It's a great question and fundamentally for us, we really start with the end. In other words, what is the outcome you're looking to achieve? Because as you point out, if you try to bring all of the data you have back, you're talking about exabytes of data and there really aren't data centers that can handle all of that data. And furthermore, as you point out, the majority of the data is not necessarily applicable to the specific things you're looking to achieve. So when we work with customers, that's where we really try to start is, what is the specific thing you're trying to accomplish? Then we move backward from there and we actually look at the different types of analyses you wanna provide. So are we looking at an outcome that requires a physics-based analysis? Do you need to understand the metallurgy? Do you need to understand the machine itself? Or are we looking at statistical or machine learning methods? Are there things that you're looking to identify or understand that you can't really get to with a first principles approach or made both together? And then we derive the data that we need to get there in the first place. It's not necessarily about big data per se, it's about correct data and many times, correct data is small data and many times it's missing data. And so you cannot assume that you have the data you need to accomplish a given output. Work with a upstream oil and gas company here in the US, a large independent oil and gas company, Anadarko. And working with Anadarko, one of the things that we identified was that for what they were looking to do to minimize their production losses, maximize their machinery health and life is that we needed to model all of their assets and model their business architecture associated with that and then start to understand how to stage that data whether it was out in the Rhone Plateau in Colorado, whether it was back in Houston, whether it was in a different data center and then understand the purposes for which they're looking. Do you need the data to help a service person in the field? Do you need a tight loop on a control system to correct performance in a given environment or do you need that for your analysts back at wherever they may be, right? And so designing that information architecture is based really on what you're trying to accomplish and as you pointed out earlier, sometimes the speed of light matters. We do a lot of work in mineral processing for example. So in mineral processing, if you're looking at platinum for example, if you're looking to mill platinum out, the ore body quality is heavily dependent on how you configure the entire mineral processing. So you have different dimensions there, where you've got floaters, you've got grinders, you've got all sorts of things involved in the milling of these things and you actually need to change the set points for how fast the mill is spinning if you have a ball mill breaking this stuff up or a floater, et cetera. And if you rely on human beings to understand that, you're just too slow and you're wasting a lot of product and you're putting stuff out into trash that could be platinum. As an example there. And so even doing sub second control point changes to optimize the system, you need to have data, you need to have any grade data while you need to understand what you're output is. If you hold all that back to headquarters and ask for a human to look at it, it just won't work. It's fundamentally impossible. Just too slow. So then that takes us really, which you've already kind of started to talk about, which is where this is the compute, right? Where does the compute? And the thing that cracks me about data centers, right, is the very temperature controlled, everything's controlled, everything's locked down. That's very different than if you're out in the field and as you said, the Roan Plateau or the bottom of the sea or a lot of these crazy places where your guys machinery operates. So when you're thinking about the compute piece, how do we allocate, where's the compute? How much compute? Because obviously that needs to be a little bit more protected. Those are harsh climate environments, very, very different. So how are people approaching the compute? Where do they put which compute where? It's a fascinating problem. And I think this is one that's gonna take us a while to work through quite frankly because as we start to get a better grip on understanding the outcome we want to achieve, understanding the analytic, understanding the data, then you start to ask, okay, well, where should this sit? Is it at the actual asset? Is it at the family or that the, let's say economically meaningful collection at the plant level? Is it across multiple geographies? Are we looking to optimize a business unit? Once we start to understand that, now you start to talk about distribution of compute and what you've quickly run into is that the compute is part of the physical infrastructure that is subject to regulatory concerns, cyber concerns. Also just distance issues. We do a lot of work in solar. One of the interesting things about solar is that you tend to put them far away from just about everything else. So sending a service person, a service worker out to a solar plant in the field is an expensive exercise and there's not many people that are really good at this stuff, so you've got opportunity costs associated with that as well. And so if you have to send somebody out they better know what you need to do and you have to be able to equip them to make a change without needing to repermit the entire facility, et cetera. So we're doing some work right now with some providers that allow for elastic compute at the actual asset, even looking at architecting new types of hardware to allow expandable computing without having to disrupt the existing control system. Take a combined cycle power plant. You have a large gas turbine and we're looking at analytics that are in a tight loop right at the analytic analyzing combustion can flame out. Now that has to happen right at the asset because it's so fast. However, the original control system, the compute and storage associated with it was not designed to do large scale analytic right there but if you pull it back, you can't prevent the flame, right? And so you look at that and again, what we're seeing come out of this is really as you say, there's different layers where the compute patterns seem to be evolving. We are seeing a big push towards much faster, tight compute around the actual asset itself. And that's hard because now we're dealing with the network topology and the fact that in most of these environments you have an air gap requirement where say NERC-CIP for example in North America, the regulatory body that controls the cybersecurity, we actually have requirements to not have external network access to these environments because we're talking about the power system for North America, right? On the other hand, we wanna be able to do tight control changes there. And so then you start to see a huge increase in cyber capabilities where we work for example with our cyber system here to whitelist control traffic at a bite level to go back in and make updates and changes closely to the asset. Then you have at the plant level, you have at the enterprise level and we even start to talk about cross enterprise level because some of our customers operate in the utility, many different fuel sources. You have solar, you have wind, you have fossil, et cetera and they wanna aggregate information across those. And so we're starting to see at least a four layer network evolving here and now challenges around how do you store the data, how do you pull it out again at each layer, what are the outcomes you need and then how do you structure it. So it's fascinating and this notion of a continuum, of a connected network and graph really just becomes stronger day by day as we discover new needles, needle and haystack. Not only are we producing a lot of hay and we're looking for small needles, the needles we're looking for we don't actually know about yet. And so we're discovering the needles as we go. So designing a system and architecture for continuous learning and evolution is a big part of the challenge. Right, I just wanna close the loop on the big three and then we'll go down to the contextual angle but you brought up the air gap in the networking that you have no prescribed breaks in the network itself when you were still trying to now put in control and pull data upstream and then send control data back downstream. So, and those are regulatory environments. So again, a really challenge to put in a connected cloud infrastructure down to the edge if you've got regulated points that you need to break the thing. No, that's true. And again, here you have to know what you're trying to accomplish. The air gapping requirement does not seek to accomplish air gapping. The air gapping requirement seeks to accomplish security. So first you have to understand the threats that you have associated with that and then the control plan that you have with that and we have a marvelous team led by Russ Dietz and this team works in across all these different regulatory environments helps us from the product group understand what our risks can be upfront and design controls to go after that. And I think specific to this requirement I think people are, there's a lot of fear and uncertainty and doubt in connecting back into these environments and people have taken, I think the very logical step of saying, look, nothing touches our control network. We're gonna pull data out initially. We'll do northbound data. You can use interesting things like fiber optic switches where you disconnect the back channel so you can only push it out and that accomplishes air gapping so you can get data out continuously. And then many customers are at that stage on the journey. The first stage is let's securely get the data out. Show me that there's value outside of the context of the plant or the asset. And now that I understand there's value to show me what the next step is. How do we make a progression, a journey from starting with some level of optimization or improvement asset performance management is a common starting place. And then, okay, now talk to me. How can we get a so-called a southbound connection where we would actually send control signals back? And our wind group, the renewables group has started to do exactly that with our digital wind farm. And as we start to look at that, what they're really doing is not only do we optimize within the wind farm itself, but we now allow industrial data scientists and engineers to start to improve the algorithm within the farm by pushing that back into the farm in a very secure and structured and governed way so that you don't have to roll a truck and have somebody plug in their laptop to put a new control algorithm in. We can actually do that from a distance. And that increases your productivity and improves your ability to test and iterate and really leaps us forward. But it's a tremendous challenge. So the phrase you just brought up, which I think is great, is an economical meaningful collection. Which I think is pretty interesting because the context in which you're making these decisions changes depending on what you're trying to achieve, whether it's optimizing for an individual component, an individual unit, a whole farm, or a whole ecosystem. So it's fascinating to me how the controls and how you might tweak that individual little thing way down the line at the edge is completely dependent on a bunch of factors well beyond just optimizing for that individual unit. That's absolutely correct. And this changes per industry, per industry sub vertical, per region. So when we go and speak to an operator in South America or Asia or North America, they all have different regulatory environments. Every customer is on their own journey on where they're going and where they need help. And so we need to provide a system that's very adaptable around that. But again, what it comes back to is really understanding the business case that we were trying to achieve in the first place. Our head of sales likes to talk about what is the total addressable problem? And then what are the pieces that we can get into that because we can very quickly lose sight of the larger goal and then have technology seeking a problem situation where we get really excited about what we can do with an individual asset when in fact, if you pull back, you see that that's actually not the highest value thing to be working on there, or there are things to work on. There's a lot of assets that you wanna run to failure, quite frankly. No failure really is, how do you define failure? Failure, at the broadest sense, is defined as your plant or your productive capacity not meeting its specification. So in many cases, there are sub-assets there that's perfectly fine to get a couple of spares on the shelf and just replace them if they fail. There are other assets that that's absolutely not okay. So you really have to understand very deeply the business outcome you're looking at, the context that's in, before you start to throw technology at these problems or else you can waste a lot of time and energy and have very little outcome for it. Yeah, it's a great analogy here that often in the big data center world where Google and those kinda guys, they're not swapping out hard drives, right? They're swapping out God knows. Well beyond racks, I'm sure. When it's a failure, right? It's again, exactly how are you defining it? But you brought up another interesting point before we turn on the cameras and we talked about this IT and OT coming together and as you said, a lot of the data that comes off these machines was very specifically built by an engineer that designed that data to feed a very specific data set and now we're trying to pull it into these broader contextual systems and apply knowledge back to it. It's a very different challenge than as you said, if you've got basic relational database, basic IT infrastructure in these machines that we're now trying to get to there. So how are people addressing that problem? I got the old machine, works great, it's got a sensors, give me data outside the context of its app, doesn't really mean much. How are people addressing that issue and bringing it into a bigger context? It's a great question. It's a challenge for us as an operator. We've worked very hard to bring our information technology experts and leaders together with our operational technology leaders as an operator ourselves, it's a challenge. Every single industrial company out there is facing this challenge. I think it really goes back to the roots of what you're trying to accomplish with these pre-existing systems. Operational technology is all about things, right? You're talking about equipment, you're talking about equipment working, run by, organizations are run by engineers that have designed equipment, that have designed machinery and the digital element of that is sort of an unintended consequence of the rise of IT. So we have all this compute and IT is very ephemeral. It's about information, it's about data, it's about exactly not things. And so to achieve the outcome we're looking for, it's really a fusion of the best that the IT world can bring in terms of analysis, in terms of looking at large scale systems and optimizing a large scale. And then operational technology in terms of really understanding, yes, this transformer is at the end of its 70 year life. It's out in the middle of nowhere and right now, given that we've got thousands of them, we need to understand the health and we need to get another 10 years out of that transformer and you can only do that if you take a data-driven approach, which means you have to sort of become ambidextrous almost in terms of being able to operate the machinery, operate your equipment, and then also operate your data and manage your data in a very adroit fashion and that's very, very difficult. And so that's one of the things that we struggle with every day, we have victories every day, we have failures every day and we don't do everything perfectly, but we're learning and we're learning very quickly. And I think the successes we have, ranging from our long-standing relationship with Delta Airlines, for example, and helping them to decrease really every year and they're just tremendous company and we're excited to be part of their team to decrease, technically, cause flight delays year-over-year to other industries that work in such as Power, of course, working there as well and just being part of the team, sharing our journey, talking about, and that's one of the reasons why we're organized the way we are, that our head of IT and our chief digital officer are in the same organization together and we're continuously improving our ability to deal with this and it is a journey, it's going to take us quite some time to get really good at this, but every day is a lab inside G. That's great. And just final kind of concept I want to reflect on with you is an IT used to be, we used to call it the digital exhaust, right? You just had all this data coming off of stuff and it was expensive to store, expensive to manage, nobody knew what to do with it and that's really changed now. Now there is value to that data, some subset of it within a particular context or a particular point of time. Are you seeing that kind of transformation inside the industrial side where we don't want all the hay, but there are now better ways to deal with some of that hay to get more value out of it than before, we just burned it off, I was saying the plants were burning stuff off. Why are they burning stuff off? Flaring the data. But just so you see now where it's less of exhaust and more of value as these machines are generating this amazing amount of data. Every day, every day we learn of a new way to use data that we didn't understand before. We're applying a lot of techniques in artificial intelligence, machine learning, we're kind of an all of the above shop here in terms of how we look at stuff, we don't really have a lot of pretension around one technique over another and I think what's really exciting to us is now fusing machine learning statistical methods with first principles methods and bringing those two worlds together and providing out of the box content to our customers and then really going forward on that and I think what is exciting for us is when we discover a low cost feasible fit for purpose way to achieve something somebody couldn't do before at all and we're able to roll it out quickly at scale across organizations and I think the exciting thing there is how you get to that is a combination of people that know the equipment very deeply and people that don't know anything about the equipment and we're gonna explore the data and work in a very research centric fashion to then work with it. We had one of our data scientist leads she was assigned to look at some of our supply chain logistics problems and working with our aviation organization I'll tell you, if there's anybody who's optimized their supply chain it's our aviation people and working on specific routing problems et cetera and as it turns out she just was able to change the paradigm that they looked at their data she was able to shift the way that they thought about their operations and achieve another couple of percentage points of productivity and when you start to roll a couple of set percentage points of productivity up at scale it's very, very meaningful and really it's what it takes you sort of have this community of people with different skills and all joined in a common cause to radically improve productivity in the industrial world and every day is filled with accidental awesomeness where people find great things and you just take it to a new place and we surprise each other it's a lot of fun. I love it, accidental awesomeness and really a support of diversity and not meaning sex or gender or anything else but when you bring a different set of opinions to a problem with a different lens people find things that the people that have been in the room the whole time often miss. You know what's been wonderful to us is to see just the enormous amount of talent that our customers have and being able to help unleash that talent. You know work with one of my customers is he says look I've got thousands of people in this company there are maybe 20 of them who can use data to change our business processes change how we operate but I know that outside of my organization there are thousands of people who can help me so how do we create a community of people working on the hardest problems of our time free the data help people get into it and start to work with that from the edge from the control system to the cloud understanding those different layers of possibility and help each other go faster and I think that's what it's all about for us is how do we help each other how do we take a multi-disciplinary approach and how do we help accelerate productivity across the board and that's why it's so important for us to start to create now just content it's all about contents it's about the pre-built asset models and analytics something we refer to as the digital twin how do we now have a library of twins for the world to work on so that we basically bootstrap people into a position where then they just take off it's exciting exciting times well Jeremy thanks for stopping by we look forward to hanging out and watching the ride and I'm sure we'll talk again later well thank you very much I appreciate it Jeremy Stone here Jeff Frick with theCUBE we're GE Digital in San Ramon thanks for watching