Dr. Colin Parris, Digital Twin presentation at the 2017 Naval Future Force S&T Expo

thank you very much okay hi good morning like to begin by thanking Admiral Han for this unique opportunity so it's my delight and honor to be here and what I'll walk you through is a view on the transformation that's happening at GE right now and specifically about one aspect of a technology called digital twin so you'll get a combination of two things you'll get this notion of what this digital transformation means to a hundred and twenty five year old company and you'll see how we apply it for specific technology that makes a significant difference so let's start first with the motivation what would move a company for 125 years of a certain direction well the two things one is a stick one is a carrot so let's look at this stick the stick is on the left side so if you look at global productivity in the industries and we look at especially those industries around energy transportation and healthcare it significantly declined in the last five years from 2011 to 2015 significantly what does that mean in an environment in which every customer is asking for a decrease in price because there's not a situation you go into in which a customer says please increase the price I'll pay more for it if this significantly impacts your profits that's the stick as your profit gets impacted so does your stock what's the carrot the card is this thing called industrial Internet or the Internet of Things that is growing at a fantastic rate but it's also projected to grow at an incredible rate without spending too much time here just look at the numbers and what you need to notice that everyone begins at a B or T so the number of things that will be connected between 20 to 50 billion things these are not my numbers I just have at the bottom of that slide maybe six or seven different sources it gets in easily into the 20s that many things connected if you look at the revenue that's created from selling selling products to companies Internet of Things products industrial internet products that easily goes into the billions or trillions if you look at the economic value created once those products are so those companies until them sell to their customers easily in the trillions easily in the trillions so that's the carrot so the stick and the carrot are the motivations now what is that carrot actually tell us that carrot tells us that if it's the Internet of Things there's a digital aspect of this that's predominant what is the stick tell us because we're looking at productivity in an industrial space the industries we look at a very prominent it's a combination of this digital and industrial capability that's going to add to this productivity in this growth so let's talk about how we began to face that we began to face that by trying to understand this notion of exponential technologies what does that actually mean because that's what you see occurring with the industry industrial internet or the Internet of Things so in an exponential technology is one in which the performance is growing at a fantastic rate while the cost is coming down and what characterizes these exponential technologies is that think all these technology curves it's not just Moore's law Moore's law takes you up one curve but then on top of that curve we build other things as the memories begin to build new technologies show up that allow you to build arrays of memories that make them faster as the speed of transistors level off the way the group transistors allows you still to have the performance in different ways if you could few years if you go to GPUs there's those nested s curves is another thing that's unique about this technology I grew up I spent two decades in IBM I started off with processors I built a processor and then I used programs I wrote on that processor to build a faster processor then I use programs I built on the faster process that I built a faster processor the technology was building itself a routine software started with assembly language and I wrote compilers to write the assembly language interpreters to write a compiler scripting language to write the interpreters now I can speak and generate software the software building itself then I combined the software and hardware together the technology seems to be building itself this is the fear most people have about AI which are the industry do you see that has that we're in the aircraft industry there's the steel bill itself if you're in the design industry does the concrete build itself no but in this industry does and while it's doing so the cost is dropping so the interesting thing 128 megabytes of memory in 2008 was $99 128 gigabytes of memory this year is $99 a thousand fall increase with no change in price a very different capability so what do we do we look at these exponential technologies and then we try to determine how can they change the business value in our industry because it will and there's a mismatch of speeds in the industry so we have to look at these technologies very differently so a jet engine lasts 40 years of steam turbine last 30 years wind turbine last 20 years a locomotive last 20 years in IBM I was there for 20 years in that time we went through the PC revolution we went to smarter planet who went through analytics who went through cloud for evolutions in 20 years every four or five years exchanges how long do you keep a cell phone for so how do you match these two together it becomes interesting when you do that so we decided on two approaches one was the notion of making sure we blended these technologies but the other one equally is important was we have the cut we looked at the cultural aspect of what it meant because without the two it would never succeed so with that backdrop I'm going to take you into one technology we call digital twin and I'll show you how we use that technology to drive the Industrial changes we needed I'll talk about how we created those systems and how we adapt those systems using AI capabilities using robotics to drive to some of the outcomes we have then I'll end by talking about the lessons we learnt around the way so let's talk about the first thing we did first thing we did was we learned that you have to have an awareness of what was happening convince yourself it was real and it was happening and find the best practices that you could out in the industries that were successful then adapt it to what you need it evolved it to the manner in which you needed to have it happen so we started with the transformations that we saw on the consumer internet we looked carefully at this for a number of years and we saw there were three very interesting players leadership players and these three leadership players actually use the same digital transformation approach to be successful those were Google Amazon and Apple when we say leadership what do we mean well we are obviously a commercial company so we look at financials as one clear definition of leadership now you look at the numbers there you see Apple 216 billion and 41 years Amazon 136 billion in 25 years Google 90 billion in 19 years when you look at that and you think it took GE little over 100 years to get 100 billion I was at before IBM 93 years to get 100 billion dollars look at the rate of growth of these companies Apple itself was - I was eight billion dollars in 2004 it jumped over 200 billion dollars there's something happening here now let's talk about the approaches they use now what I'll do is I'll use Amazon then I will apply to Apple and to Google so what did Amazon do Amazon first started using data in order to try to understand a consumer then use analytics to drive a behavior in which a consumer purchased and then found ways to deploy that in other consumer spaces now it started with this data all in a very demographic focus so I look the groups of customers I'm looking at a group of people and this group of people likes this book and if you should be a member of that group then I would suggest that book to you started with demographic data data on groups it quickly realized they do this right I had to move the data on individuals so it went from demographic data the individual data it looked at what you bought when you bought it what you bought on Mother's Day Father's Day who sent you things than those days what different types of commercials or advertisements you selected where did you send the things to what did you buy when you bought certain things what did you group together it took all of this data and it got a profile on you as an individual in fact it was a psychographic profile a profile of an individual how you buy and in what context you buy it's used that to get a model of one not a model of a group a model of one and it has 268 million consumers so notice model of one now you apply analytics to try to drive the behavior it began with segmentation if you belong in this segmentation and this segment is buying these books I'm selling these books it moved quickly to things like profiling if I know by the ads you're looking at that you've about it on a ski trip I offer you these skis at a discount then if you buy this ski is it moves quickly to prediction if you bought the skis I should tell you a ski helmet at a discount as well then I look exactly the ads we send to you and I see which ads you select the size you don't and so I know don't waste my time sending these ads to you I only send you the ads I think would be most beneficial to you by doing that I get the maximum amount of revenue for you but I reduce the cost so I get a P&L a profit and loss focused on you so I have a model of one and a P&L of one then to move quicker I put it in a platform a platform allows me to deliver something at the lowest possible cost as fast as possible then I take that platform and I go from books to retail to movies and video to web services and that's how I get two hundred thirty six billion dollars then to get more detail upon you I give it to you on a Kindle you keep the Kindles personal and so I get more details so it becomes better I do the same thing if you look at the Apple approach Apple created a model of one by looking at your iTunes list there 800 million people that use iTunes I can quickly a few new songs on the lessons if you buy these sounds if you buy these sounds I've been into a few more sounds like that I build a profile of you that's how I get to a model of one then the P&L of one is where I go after iTunes then on a platform I go from iTunes to video to applications to home to health but then I put it on steroids I give you ipods ipads iphones and i collect an enormous and of information about you I give you watches what is on steroids that's how you grow to two hundred sixteen billion dollars Google does the same thing Google get the profile from you by what you search for in terms of Google search but it's also YouTube videos Google Maps Gmail all to build a profile at model of one then you use analytics to apply that so they can deliver the right ads to you in the Adsense business that's the PNL of one you put it in a platform and they go from ads to applications to hold to health that's the pattern that's been used and it works so how about if we apply that in the transformation in the industrial state so what you see here is a jet engine we normally collect information about the jet engine but in terms of the fleet we collect fleet information usually when things break when we worry about the life of performance of something now instead we collect individual information about each particular engine or each particular turbine we collect information when it's manufactured when it's maintained when it operates the environment it operates in we quickly create a model of one then we apply analytics to understand how we could look at the efficiency and performance of these things before we had fleet analytics we did that primarily because we didn't know enough so we made assumptions so if you have the life of a part what do you do I don't know the environment that the app the engine is running in and I don't know the way it's used so I'll assume things I'll say well I assume in this average environment with this average usage that the path the part would last three years but now I know exactly the environment it's running in exactly the usage because I'm collecting that data so I could predict precisely this part will last five years so I don't need to bring the asset in in three years inspected to find out there's nothing wrong with the park also when I bring the acid in I know exactly what I should find in terms of damage and I know the replacement to have and the right engineer to repair it so reduce the cost and the time to do that how about performance I mean assumptions and performance of my gas turbines because they didn't know the environment they were run in and I didn't know the way they were run so I'd look at it I'd say well I assume this should be the maximum performance because I want to limit my risk so I don't have damage that would void my contracts but now I know the exact environment the exact usage so I have the ability now to see you can perform at a little higher level at this point in time so with the understanding of the life I change the efficiency of the system or availability of the system the understanding of performance I change the performance of the system when I combine those two together the economics changes so now I have a P&L of one so I have a model of one on the asset and a P&L of one on the asset allowing the customer to have the power of taking a specific action at the right time being more efficient much better for four months knowing how they can vary that asset to meet the in need then I put it all in a platform I call a platform predicts I make sure that platform has two components a cloud component and an edge component why do I need the edge component because some of that information has to be delivered in real time when I run control systems that affect power stations and locomotives and engines airplane engines after run it in real time also I may do things that require video capabilities I don't have the bandwidth so I need an edge capability and a cloud capability all this I have on predicts I put it in a platform again focusing on speed and time and I deliver it to aviation to power the transportation to help here it's the same approach that we've used this fall except we've evolved it and innovated it specifically for this environment let me give you some examples because the other aspect of this because with GE it's all about business outcomes it is not the use of technology for technology sakes but it's directed specifically towards an outcome so the digital twin which is what I just showed you has five characteristics it is per asset it is attacking a specific business problem it is a continuous stream of data again the timing of it is based upon the problem or attacking it's adaptive when the environment changes it changes it's scalable if I get it for one device I can quickly scale it the things that look similar to that device it's focused on three things how do I provide sufficient early warning so I can change an unscheduled maintenance or scheduled maintenance giving the customer the maximum availability for the asset continuous prediction how do I predict the remaining useful life so that I know exactly when to move that asset out of service and be prepared when I do move it out of service to have all the things I need to reduce the services time or how can I predict performance capabilities so if I see the benefit of a much higher electricity price I can increase the performance of this gas turbine to take advantage of that how do I do dynamic optimization where I balance the needs operationally in business wise against each other so those are three examples here so the first one is efficient early-warning are just touch briefly on that so the notion here is they're disappearing in the engine and number forbearing and if you can detect early enough that number forbearing giving you problems you can stop a light from showing up on the dashboard that says you've got to have this repaired now before the warning was four to six hours which can leave a plane stranded on a terminal out of service passengers upset now when we create a digital twin and we combined physical and digital capabilities and I'll talk about that in a second we have the warning capability to go from a few hours to a day to 15 to 30 days that's advanced warning when I look at continuous prediction the challenge here is that we were flying these aircrafts and hot and harsh environments highly contaminated very hot maybe sand the environments once you find in the Middle East's or certain parts of of Asia and the blades that rotate have a coating on them and that cotton was getting contaminated been worn away exposing the blade itself the heat was causing problems that again results in a warning light we began to pull these engines every 200 flights to look at them to determine buildup on the blades when you pull an engine it's a way from the customer what we did is recruited a digital twin that began to tell us exactly what the level of damage is the cumulative damage on the blade by understanding the physics but also the environment it was flying in as well as the wheat was being flown what was the trust when it took off how much it'll climb where is the level of which to airport it flew between it's combining again the physical and the digital to give us an outcome in this way I can actually say after 200 flights don't bring this air graph in the damage was not sufficient or I can see at 160 flights bring this one in it's been flying continuously in the wrong environments damage has built up too much third aspect of this is a dynamic optimization whenever you bring in all of these aircrafts what you'd like to do is quickly replace it with a spare engine and put it back out so that a customer has maximized the availability of the aircraft how do I know if I've bought enough spare engines it all depends if I'm keeping the aircraft out there a lot longer what tends to happen is when I bring it in there's a lot more damage I have many more parts to replace then many more parts to replace I'm going to have to keep it in the shop longer if you have to keep it in the shop longer I'm going to need many more spare engines how do i balance the time I keep it on wing with the damage it gets done with the tuner long time in the shop in such a way that I have the minimal number of span engines because these engines cost a lot but if you have to lease an engine it cost a lot so you want the balance between the two that's a dynamic optimization problem with many factors the outcomes you see here all real in terms of the numbers 90 plus million saved in continuous prediction 30 million saved in dynamic optimization so I'm sure you're asking how do we do it what's this notion of physical and digital let me simplify it to some level so the physical aspect of it is the fact that I have certain sensors and from those physical sensors have created certain models physical models those physical models give me data about the asset I'm thinking about but what I need is more data about this asset so I create virtual sensors from the digital data I have those virtual sensors combined with the physical sensors give me an enormous amount of data that allows me to use machine learning techniques and other techniques to determine some of these new capabilities I never had access to before for instance on a jet engine there may be only 20 sensors what is 20 sensors our sensor cost over 20 million dollars to build it's gotta last 40 years on the plane it's got to fly a 10,000 feet it's gotta go through incredible heat it's got to land and take that jarring impact cost the lock we may have sensors at these two points what occurs in the middle between those two sensors to that piece of metal what if I see that as a problem I create a virtual sensor that lies right here that I've created because I have data about where it flew how it flew that virtual sensor gives me additional information that allows me to deliver this value it's the combination of the math and the science the physical and the digital coming together to give us this unique capability so let me talk about learning systems now because I'm doing this digital twins that these digital twins are adapting to the environment so I have a living model because the model every time the plane lands I get snapshot data it's living how about if I can have it learning how about of the model itself can learn from the environments it's in so if I have a living learning model I have the best of all worlds so where do we see learning happening let me use the Tesla example Tesla right now have something called enhanced autopilot so if you look at Tesla cars this is the ability to do you know what we talked about as autonomous driving now I just start we talk a lot about things like Google Google has 2 million miles with their cars well Tesla when you look at the entire fleet as of last year October had over 218 million miles of autonomous driving but it also captures all the data whenever the car drives it's like a mini computer on wheels it has over 1.3 billion miles would update its collected so what to do what is it doing so if you get on a Tesla there's a big panel and this panel has a button that you can push the $5,000 called enhanced autopilot what does the autopilot do it does several things but it does three things that I personally like first thing it does is on freeways it can keep you at a constant speed and it can change lanes to keep you at constant speed how does it do that there are 16 sensors below the car there 48 cameras on the car and there's a reader in the front of the car so the reader actually can go and jump ahead and see cars and understand the movement of cars and the cameras can look at the white lines on the ground and so it can move you between lanes to keep you at that constant speed it can also take you off the freeway and take you on the freeway so when you get on the freeway the most boring times the times in which several dangerous accidents occur you sit in the seat and you relax because this is driving for you the other feature I like a lot is when you pull up you could come out of the car and you can tell it Park itself it goes in it parks itself when you come back out of the venue you open your cell phone hit a button the car comes to you great features how does it do it let's talk a little bit about that so here's the one I like so I'm going on the freeway and I'm coming down I'm using the white lines and the cars driving correctly and then I come off and run a country road and which is only one line is a line in the middle of the country road so on this country road I'm on one line the car drifts I'm in a driver seat I adjusted back after a while the road shifts car drifts I adjusted back what is the car do that point in time it recognizes that the human is teaching it I have moved the car back it records the exact GPS coordinates and the actions I do and it sends it up to the cloud that point in time once it gets to the cloud every other Tesla vehicle knows how to drive on that road that's a human teaching it how to drive so it's learning from humans now when it's on a road and the road begins to waver the sensors are adjusting the sensors are adjusting because it doesn't want to damage the undercarriage of the car it also wants to ensure that the Sharks as well as the wheels last as long as possible so it adjust itself so it's using the sensors and calibrating against the road mapping the road keeping things right that's the living from the sensors but at the same time it takes that information sends it up to the cloud so any other Tesla car that drives on that road knows that information so it looks like magic when you Phil show up on that road that you've never been on because it suddenly balances itself and it knows exactly where to go so it learned from humans and it learned right away from the sensors it's also learning from the fleet the fleet of Tesla cars 70 to 80 thousand strong now are driving on roads that none of us have driven on mapping those roads mapping those conditions so when we show up looks effortless so it learns from the fleet last thing it learns from is simulations so it collects all this data and it puts it in the cloud what does it do when it puts it in a cloud it runs machine learning algorithms what do they look for they look for patterns which patterns are important here the patterns they found the first patterns if a ball goes into a road a child follows it said the camera detects the ball it slows down and continues to slow down because the child is coming behind it second pattern detect if a dog runs into the road a human follows it if it sees the dog and it could recognize the dog it slows down and continue slowing down because a human is coming behind it third pattern is that they are travel in groups so whenever it sees day it slows down it learns in Australia that kangaroos hop over cars so the kangaroo can come directly to hop over the car that's an anomaly but is it lifting to new so it's using simulation it's learning from four different methods it's learning from the sensors it's learning from the simulations it's learning from the humans it's learning from the fleet if you can learn from four different manner modes you can learn at a much faster rate than anything else on the planet including humans we have applied the same capabilities jet engines we have capabilities where we're learning from the sensors themselves and can allow us to tune some of the twins that save us millions of dollars we have some of the capabilities in which we're learning from the simulation that one I showed you an optimization of the jet engines we had to run simulations projecting powered to figure out the patterns and allow us to figure that out we're learning from the fleet when some engines are flying in these hot and harsh environments everyone doesn't need to I can learn from that example and apply it transfer learning is the term we use in machine learning we're also learning from experts in the number forbearing how do we do it we look at a certain profile we constantly monitor that bearing and we know what's normal when we see something abnormal we look in the old rule databases we have to see if it ever occurs and if we can't find something that tells us why we have this anomaly we send it off the chief scientist she or he typed something in we capture that in text mining and so we're learning from the human with these four rates of learning environment changes imagine a situation you're in in which you've got a little at a fantastic rate wouldn't this be useful let me give you another aspect of the use of the twin somebody we call adaptive defense so the notion is we use this to create a digital ghost now when we think about cyber we rely on a lot of IT system to protect us protect us at the boundary so that someone can't come in and pull the sensors or pull the control system you can pull the sensors by giving it fake data and thus manipulate the control system to do an action or you can directly send that signal as a control system signal and have the Machine do something so while we depend upon our IT brothers and sisters to help us with this we also know we have some insight into the machine itself we have one thing that's very hard to spoof which is physics so what you see here are 17 sensors we have 17 sensors around this combined cycle gas plant that has a gas turbine and a steam turbine and the 17 sensors are all related by physics so if I try to fool one of the sensors like the fuel flow and because I have a digital twin I know the relationship between that one and the other 17 so that one sensor looks strange that physics relationship looks quite strange as well and I can respond to it if they fooled two or three sensors I can respond to it five or six I can respond to that as well because I know the physics relationship it's the Brayton cycle of heat if the wave physics works it's not just looking at what the data flow is it's understanding the data flow and the physics itself so when I do detect that four or five of these sensors don't seem to be right what do I do I use a digital twin and I build virtual sensors of what it should be and I use those virtual sensors that continue controlling the system so I can detect the problem because I'm using the anomaly detection techniques of the little twin I can realize when it's not right because I'm using the physics that I know should happen so I can locate which one of these sensors are problem then I can isolate the sensor and I can create virtual ones so I continue maybe in the degraded fashion but I'm not totally put out of play another use of the digital twin in the environment is sin and if I can adapt and I can learn I can evolve fast enough to be better suited to handle these environments because my learning gives me again other capabilities to allow me to adapt faster so now when I think about that what do I have I have an immune system I have an immune system that's learning about the environment and adapting to the environment in a very very quick way now let me talk a little bit about the data that we have to gather talk at length about the fact that we're using physical and digital and combining these two together what another exponential technology that I can use that actually helps me with ensuring that my systems are kept at the highest level but also give me the data that I need robotics so what we've done in G is we spent a significant amount of time understanding how we use these capabilities so while we love the technology itself and we have a fascination for it our real focus is how do we apply this technology to deliver value that I can monetize in the right way here are three examples we have something called the inspection means answer repair approach what we do here is we use these robotics the first two inspection gathereth data we generally want to do this in dull dirty dangerous environments then we focus on how do they do simple maintenance actions for us then how did it do more complex repair actions for us so you see three examples here the first one rooted revolves around this notion of G internal services so the notion is you bring the engine in or you have the engine and you like the engine to be inspected I'm using many aviation examples here but they apply to power they apply to locomotives they apply to wind turbines they apply to healthy equipment what you see below that G internal services are something called the turbine surgeon it's roughly about two inches long it can go inside a complex turbine and it has sensors that allow it to take video and that video is connected to an NG device and on that edge device we have algorithm that can detect certain things corrosion spallation cracks and so that assists a human being in getting the places where we can't normally go it can also give us a speed-up of 8x in determining a problem the middle one you see manufacturing what you see below is one of the Baxter robots this is a robot that's helping the collaborative robot so you can have it in the open with humans doesn't need to be in a cage and it helps build a heavy parts of a chassis it loads the heavy pieces in the best positions and then the human goes in and does the wiring parts so the human doesn't have to lift the heavy pieces and put them in the right places and then go from a physical skill to to much more fine-tuned skill we figured out that if you do that in the right way what we've done is we've reduced the number of injuries to humans as well as allow the human to focus on the thing that's most valuable again reducing the time increasing productivity significantly the other one who sees the service solutions we have many situations in which we're putting people in very dangerous environments so this is a rig on the rig there flare stacks and the other structures in which you've got to go and inspect so you inspect it either by taking a helicopter and flying around that rig and trying to get a camera to look at it in places but even so you don't get all the places or you send climbers up climbers on ropes in which the wind is blowing at 40 50 miles an hour very dangerous environments how about if we use robots in these environments and I'll show you an example of this one in the video I have so the notion here is can we use these BOTS with humans and we think the tree steps the first one is using the robot as a tool the second is using robot as a partner the third is automating the robot to the point in which the bot can do it themselves and what's the bot doing it's learning I'll show you the examples of the bottle learning from a human learning from simulations learning from its own sensors the same learning pattern we talked about before how do we use this to drive value so on the left side our focus here is on aviation so what we have is a situation where I have an engine normally if I have to inspect that engine I would have to use this camera for the borescope it's a little camera on the edge of a tube and I have a person that actually tries to sneak it in and so you sneak it in but they're part you can't see what this robot does is that we take the robot in we attach it to a part of the turbine and then we spin the turbine so now it gets to see every part because they were part of a block from there from the borescope because the turbine was actually in the way now because it's spinning very slowly you can take a full picture of that environment so we can do an analysis on a wing we don't have to actually pull that off wing to do it do it on wing do it very quickly detect the problem early then we can take an action early maybe do a water wash rather than wait until damage occurs that actually reduces the overall visits the number of times we have to take the entire gin in by 20% other example is one of a turbine gas turbine so what we're doing here is that we can take the top off the gas turbine what we normally have to do then is pull the entire to buy enough so that you could look at the bottom side now we take the top off and we send a robot in and that robot walks through that turbine chart in its own path based upon what we've asked it to collect it collects that data again using a camera gives us insights by using that turbine because we can use it effectively it's cut down time cost and the down time cost is how fast it takes to actually do that repair by 40% we use three basic areas when we do this the first is this notion of perception what perception is can I give the robot a view of their real-world environment can I teach it what the environment is so that it understands how to access an engine what is the part in the engine how do I Traverse that engine how do I do this automated inspection because it mean conquer something that's different the blockage in some way it's got to be smart enough to move around it then reasoning how do I teach it how to find a crack how to find a spoliation how to find corrosion how do I actually have it give me the right data to see that digital twin the third is dexterity how do I have it position the camera in the right way so I get the right images zoom in the right way or if it comes to the tool how do I have it use that tool to actually repair a crack so we tend to focus on these three things let me run a video here to show you what we tend to do so this is one in which we're teaching a robot to do an inspection we first start by planning we ourselves as a human look to what we're trying this is a flare stack so we're trying to understand what's the scheduler in fact of normal inspection on the flare stack what's the risk trade-off if I don't inspect the flare stack how much does it cost me then based on that we say okay maybe I do want to do this inspection at this time right now so if I want to do this inspection let me at least get some historical background because if I can understand where ISO cracked before and what occurred before it can give me insights into how I have the robot do what it needs to do so I look at prior inspection reports then I decide let me go out and do a site survey because I've got to build a robots world view so the robot and a human they go out and we fly this around the flare stack that tall object is a flare stack because we've got to understand and teach it that this is a flare stack they're things next to it that are relevant but not exactly flare stacks and so we fly it around then we take that data back and we have a human say these are the areas you look at this is the angle this is the zoom the size of that blue ball is the zoom angle and you map that out then you give it to a computer who determines the flight path that's the optimal one and the angles is the optimal one for the robot to do exactly what it needs to do then when the computer does that we give it back to human and we trace it out for the human so this is the human and the computer working together to learn how to do a task for the first time and the human looks at it and says okay that sort of makes sense we want to make sure it's not next to too many things then we have the robot God autonomously on their own and fly around the stack itself it's smart enough to know when the variation this is a wind gust here's how I angle myself so that I still get the right angle but I complete the job and as it's flying along it's looking for cracks it detects the crack and records a crack it continues its journey regardless of the heat it doesn't get bored it doesn't lose focus doesn't worry about the Jets game it continues on the path it can not only do cracks it can the things like corrosion it can detect it at an early point so that I don't need to replace many things it can put on different sensors it can do Timmel imagery normally I would have humans do this every three or four months I can have the robot do it every week so I can detect something went small and so it costs a lot less to fixed so I can adapt in my environment a lot quicker so when I have all of this information I put it together in a report that both the robot and the human generates so part of this is robots being used as a tool the human helping the robot understand what to do and then over the course of time guess what more robots in the fleet talk to each other and figure out how to use their fleet information so that they become smarter again using this learning paradigm to get it smarter to get it quicker so with that I'm going to end but before I end let me tell you a bit about the journey because I think you've seen parts of it here they fight things that we learnt along this journey first aspect of it is that to do this right you have got to have a new strategic awareness about what's happening around you when you look at the timeframes of 20 30 40 years in GE compared to the timeframe that the technology three four five years something is vastly different there you've got to blend those together what do you do you have to deliver new ideas about how you do things a lot faster so the first notion was okay let's just do it faster maybe I can have people work more hours people work more days doesn't work first notion is this thing known as a string of how about rather than people work faster what about if everyone works on a platform and builds modules as they do work and if I have a lot of these little modules collected on the platform when I want a new feature how about if I just string these modules together four or five of them to create a new feature that's a string of pearls what if I needed something new can i string these four together a new capability Connect string these five a new capability so we have a way of a distributed community or working on the same platform putting together and adapting as fast as possible to a situation not four groups running like hell all in one direction or in two directions and trying to respond the string of rules idea the vision aspect is make it believable to me that's why we use consume examples cuz when the G engineers brilliant arrow mechanical engineers have seen we've done it on a consumer side could you not apply it here they agreed then they said no it's different it's different we do things in which people's lives at stake okay well Tesla's doing that people's lives are at stake in a car hmm what it becomes believable and you challenge them then they say okay you know what maybe we need to look at this in a different way isn't Tesla billing rockets isn't that hard to do as well maybe we should look at this as well then make it repeatable innovation there's no sort of projects was predominant in GE I have 10 projects in my organization I'm good whoo what's the outcome I have 10 projects show me the outcome will here's a chart with 10 projects that's the outcome you have to focus on the outcome so we began in 2014 we did 25,000 twins 2015 I looked at me and said okay Colin how much more can you do my guys came back and said ok we can do 35,000 twins with you he said 250,000 now what occurs when you have that type of demand that type of outcome the projects went away you couldn't do it using an incremental approach with the projects you had then so we have to scrap that idea to get to 250 get to think differently now the project's mattered or figure out which fuel matter and actually deliver what you need to deliver the outcome focus he said it should deliver a hundred million dollars with a value everybody stopped in the end at twenty sixteen when we actually finish we had six hundred thousand-plus twins over 200 million dollars in value because once you get the mind shift to an outcome everything changes this whole notion of human nature we have cultures we've build these cultures all these years you want me to change the culture no Colin you mean it's it's our culture when do people suspend their culture they try new things in two times chaos an opportunity we have a chaotic situation security change a 911 people suspend their culture they say ok let's do something different you have a massive opportunity you go to a Silicon Valley company which person can make ten million dollars Wow we change our culture so we began creating these things the 250,000 twins was a chaos how much money you could get was an opportunity they suspended their culture not asking you to drop your culture suspend it and try a couple new things once those things worked they began to incorporated in their culture what was different in this one dough is that is a winner-take-all mentality when you ask about Google Apple and Amazon in the commerce in the entertainment space or in the advertising space gee normally would are through the top three players if you ask that question there who number two and three the answer is it doesn't matter there's a winner-take-all culture in this world right now and if you don't have that you've lost the last one is you have to make visible manifestations of your organization's gee moved all of their IT into one centralized organization no longer they eight silos where I sit in the GRC the global Resource Center a real Research Centre we package all of these small labs and these projects in the two big areas one physical one digital and then we create breakout projects which we fuse the two together with an outcome the outcome is in a year the outcomes in two years there comes in a few months if it can't be done we fall them back in visible changes the way we invest visible changes the way we reward so I'll stop here but this has been the journey for us we are still on this path it is a journey that's inevitable for everywhere you look in this industry so it's one you have to embrace so with that I will stop thank you very much for your time it's been my pleasure and honor to be here [Applause]

Channel: usnavyresearch

Views: 7,943

Keywords: #navaltechexpo, GE, digital twin, future

Id: XM4DlLjsL0g

Length: 53min 17sec (3197 seconds)

Published: Sat Jul 22 2017