What is a Digital Twin with Colin Parris CTO GE Digital - CXOTALK 685

Show video

What is a digital twin? How can we use it?  What are the benefits? How do we implement   it? What are the challenges? Colin Parris is  the chief technology officer of GE Digital.  GE Digital is a company inside of GE, and our  focus is on putting industrial data to work.   We operate in roughly about four industries.  We operate in grid software. We operate in gas   generation in oil and gas, power generation  in oil and gas. We operate in aviation, and we  

also operate in the manufacturing space, so  we have four segments we go after. We produce   industrial software that's designed to help our  customers deliver value from the industrial data.  I am the CTO. I am the CTO of GE Digital,  but I have two roles. One role is  

developing technology that's deployed in  the GE Digital products. But the second role   is helping digital transformation  inside of the GE businesses.   Those are the two roles that I share right now  and it provides for an interesting time for me. 

Colin, you're chief technology officer and  digital transformation has a technology aspect   as well as (very often we talk about) a cultural  dimension. Shed a little more light on the   technology dimensions that you're involved with. When you think about technology dimensions, I   think about three potential dimensions that I get  involved in. One is identifying new technologies,  

technologies that we believe could be  used to give our customers an advantage.   That is more me sitting in the role  of advanced development or research.  The second is deploying current technologies  where I'm thinking about, well, this technology,   I think it's hardened. It's mature. It can be used  in our products to advance those, and it should be   part of the roadmap, so I spend time there. The third is an interesting combination  

of business transformation tied to the technology  itself because, in many cases, customers will come   and say, "Well, I'm doing digital transformation,  but I can't see the business value I get from it."   I spend a lot of time finding ways to  integrate business process transformation   with digital transformation. In this way,  I create (or help you create) a digital   tool, a digital capability that goes inside your  business process and gives you business value. 

Those are the three dimensions I tend to  work on from a technology perspective.  Layer the concept of digital twins  on top of that for us, please.  In GE—and a lot of this was started in  the GE businesses—we have these very, very   large assets. The assets have dimensions  of them that make a ton of sense  

but also provide us a math of complexity.  These assets cost multiple millions of dollars.  They deliver things that are at the basis of  the foundation of the world. One-tenth of the   electricity that's generated comes from GE assets.  We have a significant, 60% to 70% of the engines   that fly in the world, so we have to  bring people back and take them safely.  We also do a lot of the healthcare. We have  a lot of the large healthcare machines that   take care of the health of the people. When we're looking at those assets,  

you're thinking to yourself, "How do I run  them with increased amount of availability?   I want it to fail the least, and I want to  do it in the most cost-effective manner."   This is when we sit back and we think, "How do  we find a way to do that using the data we have?"  What I have now is created a digital  twin. A digital twin is a model,   a special type of model. It's a living-learning  model. If I have the model of that asset  

and that model changes the corresponding  state of that asset, then I can predict.  Can I get an early warning on a failure?  Can I predict what I should have ready so   that when I bring in that machine for it to be  maintained, I can it very fast? Can I optimize it?   Can I use it with the least amount of fuel? Can I  use it in a way that it delivers the most amount   of value with the least amount of labor? All the twin does, it's a living-learning   model that allows you to deliver business  value by constantly making sure that twin   is the exact replica of the asset and  then getting insights to take an action.  What are the components that go  into creating a digital twin?  When I think about the components that go into  creating a twin, I first start with—most people   will see it's not a technology component  but it's the most important component—what's   the value that you're trying to deal with? Am I  trying to increase the availability? Do I want   that gas turbine and that powerplant to run the  most, so it could generate the most electricity?   Do I want to take out cost in terms of  fuel cost of a jet engine that flies?  You start with what's the value, the  business value you're going after.  Then secondly, you back into, "Can I get the  right data and domain knowledge?" to see if   I can get some insights that allow me to do  that. Can I get the right insights that allow  

me to know when a failure is going to happen? That's domain knowledge. Somebody says, "Well,   usually we see these fails happen and that's  what causes the failure, and here's the data   associated with it." That's the second thing  you get. Then the third thing is the models.  Then I actually put to use physics models,  AI models, or a combination to use that data   to try and see, can I predict when that  failure is going to happen clearly enough   in advance that I take an action? Think in terms of the business value   you're looking for. Think in terms of the data and  domain knowledge, and then the models that we have   to build. Then you get into the complex things. You have to figure out, how do you deploy it in   a way that you can test it so that you're sure  it's not going to damage your equipment. Then we  

figure out when does it actually work as accurate  as possible. When do you need to tune it? Those   are the major things you think about to understand  value: data and domain, insights, and then models.  Why do you call it a living model? In many cases, many people are familiar   with models. If you're a good designer—before you  design a gas turbine or a steam turbine or a wind   turbine—you create a model because the model tells  you. The model lets you. The model lets you use   that to figure out what components will I build,  what materials will I use, how would I design   it. Then we use models a lot in design. Also, if you have a problem in services,  

somebody goes out and builds a model so that  you could sort of find a way to emulate what's   going on or simulate what's going on so that you  can take an action. Usually, after we do that,   we put those models aside. We leave them alone. What you do with a twin is you create that model,   focus on a specific problem, and you  make it a living model in the sense that   you bring data in continuously and you let the  model evolve as the state of something evolves.  Let me give you one good example. Think about a  gas turbine. While this gas turbine is running   (over months, over years) the materials inside  break down because you have a lot of heat. You're  

exploding fuels in order to spin this turbine.  If you have a way of saying, "Whenever I do this,   that level of heat changes the material structure  and wear and tear occurs. Ball bearings get   degraded," the model begins to measure that. The data that you are taking from all of the   sensors begin to say, "Well, I think  these materials are wearing down.   The oil is wearing down." That's why you want it  living. It constantly updates what it is that's  

happening inside that complex machine so that it  gives you an accurate view of what's going on.  You also want it to be living because what  happens then is that your state changes.   In some cases, maybe you're operating something  in winter or you're operating something in summer.   The conditions are going to change. Coming into the digital twin, it's not just the   information from the sensors about the machine  but it's the sensors from the environment,   the sensors from how they operate it.  All of that changes and what you want  

is your model to reflect that change. If you make a decision on how to change   the performance, get it better, or how to get  the maintenance done or when to do maintenance,   it reflects the actual state.  That's why it's a living model.  Colin, you were talking about this living model  that's adapting and that is taking data from the   environment as well, so it sounds like the  construction of a digital twin is quite   complex and quite involved. It can be. It all depends upon the type of  

problem you're trying to solve. In many cases, we  have been doing this for a while. Everyone I know   that actually runs complex machinery (or even  simple machinery), they usually have, at times,   and M&D center, a monitoring and diagnostic  center, so you've been capturing information.  The idea is, can I take the  information I'm capturing?   If it's enough for me to understand the state  of the machine, then I can use that directly. 

Often the case, though, especially when these  machines have a much larger length of time—a   jet engine lasts 40 years, a gas turbine lasts  30 years, a wind turbine lasts 20-30 years—with   those long-lived systems, the environment  changes. The way people operate change.   Sometimes, the maintenance changes,  or people slip up on the maintenance   – for a variety of reasons. With those changes in state,   you often have to not only take the sensor  data, but you've got to reflect the data from   the environment and from the operations.  That's the new data you have to collect.  In many cases, that data is already there. It  may be in a different system. It may be in your  

MES system. It may be on your SAP system.  You may have to fuse the data in order to get   the information you need to give to the twin. It all depends upon the outcome you're going   after, and so I always advise people, "Start  small. Start at what you have. Tune your  

skills using that. Then expand to look at  other things that you can put into place to   get the model more accurate or to reflect  the cost in a much more adequate manner."  Ultimately, what are the  benefits of digital twins?  If I think at a high level, we tend to think  about three of them. Can I have a digital twin  

give me the early warning about a problem that's  about to happen? In aviation, for instance,   can I get a view of when would a #4 bearing –  this is a bearing that is inside the turbine   itself fail, because usually, these things fail. You're at a gate and there's a light that pops up   for the pilot to see, and so you have a problem.  Now, you've got to be playing all the people.  Can I give you that information 30 days in  advance? You need 30 days because you need enough   time to allow the airline to get a new aircraft in  that slot, to get a new crew in that slot. Crews   have to be certified. To get new support in terms  of the right things in terms of fuel or food.  You want to get an early warning. That's the  first thing. Can I get enough of an early   warning so I make a small tweak in my business  rather than a large mishap or catastrophe?  The second is, can I do continuous prediction? Can  I predict when something would fail? Can I predict   the type of wind I have? With  wind turbines, if I can predict   the wind a day ahead, I can know what I bid in  order to sell my electricity into a utility. 

Can I predict when something will fail?  The lead time for parts for these large   turbines may be six months. If I can predict  way in advance, I can have the suppliers build   that so, when it comes, it's there right away. Then the third thing I do is dynamic optimization.   Can I optimize how a system runs so that  what I have is the maximum electricity   with the lowest fuel cost or the maximum  electricity with the lowest carbon emission?  Those are the three things: early warning on  a problem happening, continuous prediction   so I can better position myself, and dynamic  optimization, optimizing the way it runs so I   do it at the least cost or most performance.  Those are the things we tend to focus on.  Colin, how is this different from historically  building up models in spreadsheets, for example,   or with software, to do these kinds of analyses  and predictions that you were just describing?  There are two major things we look at here.  Everyone asks that question. We've always   built these models. What's different right  now? What you find out is that the models   we have built have been fairly complex models. First of all, what I'd like to do is a simplified  

version of that model because I have a model of a  jet engine. We tend to build these before we build   any jet engines. But to run that model, it takes  something that looks like a supercomputer and it   takes a couple of hours to run the entire  model. That's not what I'm looking for. 

I'm looking for a surrogate, a smaller version of  that model that's focused on a particular problem.   I want to run that and that should run fast. In some cases, we need it in a half an   hour. In some cases, it might be a day or two.  But it has to run within the right timeframe.  What I'm looking at, first and foremost,  are models that are specifically focused   on things that cost me money. I'm not  trying to look at the entire performance.   I'm looking at a specific point of what is costing  me money right now: the blade is a problem;   the nozzle is a problem. That's the first thing. The second thing is, I am now looking at complex   systems. Before I can have a model of my  jet engine, I want to know that jet engine,  

in the environment, it's flying in—and it flies in  different places—operated by the power to operate   it in this way, given the cost of the fuel right  now, so I am bringing in multiple sources of data:   financial with the fuel, operational with the  pilots, sensors from the actual monitoring,   and environmental data. I can do things like  wind temperature and dust contamination levels.  When you look at all of that data coming in, that  is a big data problem. It's different. Your model   is now in the context of all these things and  you're trying to optimize it. Now, how do you pull   that data in? That's the second way I look at it. The first one is all about a surrogate model  

focused on the exact thing you're trying to  solve, not an overall high-level model of   something. Then the second is taking in all that  other data that gives you context because that's   how business value is realized in the right  context. Those are the two things I would say   that make a big difference at this point in time. That's really interesting. You said that digital   twins are fundamentally a big data problem. In many cases, it is. But again,   it all depends on what you're looking at. Let me give you an idea here of how it could be,   at times, a small data problem but  equally as valuable. For instance,  

I'm a data scientist, so when I think about  things in my normal role, if I look at the   consumer world, I have lots of examples. 2014, if you look at the amount of packages   Amazon sold, it's like two billion packages.  As a data scientist, I love that. That's two   billion examples of people buying something. In Google in 2014, I think it was 10 million or   12 million ads a day get selected. I  have 10 million or 12 million examples.  Now I go back to GE. GE has a fleet of engines  that are the GE90 model. That GE90 model,   in a year, we do a little over a  million flights, some airlines. 

You think, "How many failures do you have?"  because what I predict are failures. Well,   the answer is in the low 20s, so that's not  big data. That's very, very sparse data.  Now, where it becomes big is that I have  got to take that from multiple fleets,   put it together, and then I want to take that  data in the context of the environment, in the   context of how it's operated by all these pilots,  in the context of the cost of me not having that   airline, that plane travel. Then the other  sources that pile on become big data around it. 

Initially, I'm doing things with sparse  data. That's why I have to use physics   models to aid that sparse data. Then  I put it in the context of much bigger   data to figure out how does that affects me  financially and what action should I take.  These are somewhat complex problems. It's both  big data and sparse data at the right time. 

Would it be more accurate then to say  that the digital twin is a data problem   more than anything else? It is a data problem, but it is also   getting that physics right. I need the big data,  but at times I have to go back on the physics   because I can't wait to get all that data;  I can't wait for thousands of failures. By   that time, we would have put lives at risk. I've got to use the physics models and the   simulators that I've developed. I can use the  simulators to build a huge amount of data,   and then I combine that with the data I  currently have. It is a data problem all around.   How can I get that data that allows me to do it? Some of my data come from empirical data that   I take from the environment and some of  it may come from the models I have and   the simulators I've built on those models.  That combination of physical plus digital  

gives me something unique right now. You've got this set of data together   with the physical calculations. Exactly.  The model of physically how it works.  I'm not expressing this well, so please.  You are. You are. You have the perfect  example because I have all the data and   I have the insights of when I designed  it, I designed it using certain physics.   That physics of how I designed  it, combined with the data,   is what comes together to make this all work. The reason we have to use the physics is that  

I can't get a lot of examples of things at the  extreme end. If you do things like commerce, you   can get examples of things on the extreme end. You  can buy blue jerseys for sale and they never got   sold. You can buy blue jerseys for sale and you  run out in two minutes. There are extreme ends.  When I'm dealing with a jet engine,  a gas turbine, or a wind turbine,   I don't want to go to an extreme end. I can't  afford to have it break. I can't afford to do   anything that would impact safety. In this really interesting way that   the only way I get to the extreme ends  is I use the physics models we know   and the simulators we knew that can go to  the extreme end. I can run that physics model  

on a system, on a supercomputer, and I can go to  the extreme end. Then I can take the data that I   have from normal running and tie it together.  I use these things to put the pieces together.  You were perfectly right, Michael. I'm  using both to give myself a breath of data   through a breath of experiences—some created  real, some created artificially—that allow   that twin to express itself. Correct me if I'm wrong. The   quality of the digital twin must be based  on a combination of the quality of the data   as well as the quality of the physical models. Exactly. That is the perfect way to describe it. 

What we do is, even if you don't have the quality  perfect at the start, we have a technique we call   Humble AI. What that technique does is  that technique says, "Can I figure out,   based upon the quality of the physics I have  and the quality of the data, what is my zone   of competency in which I am very, very competent?" In that zone, use the twin. Outside of the zone,   go back to the usual deterministic model  or go back to the human process by which   we evaluate things. Then give that twin  more data and get it more competent.  We spend a lot of time understanding the zone of  competency. That zone of competency tells us what   is the competence interval around the answers that  this twin is giving you. In that zone, the twin   works well and people like that because what you  have are customers saying, "That's the right zone   for me to be in because I have the least business  risk and I can get the most business value." 

Also, that zone, I do what I normally  do. I understand the risks there. Then   I feed more data so this thing gets better.  That's what we spend a lot of time doing.  How do I enlarge that zone? Can I  build better simulators to enlarge   that zone of competency or can I get more  information like, for instance, from the fleet?  Your jet engine may only fly in one environment.  Your gas turbine may only work in one environment.   But at GE, I have access to these huge fleets. Can I not take that data and compare it and say,  

"Well, this turbine, is the model similar  to yours? Is it operating similar to   you?" Maybe I could transfer some of that  learning, bring some of that data to bear,   and give me more data in a way that I can create a  better model that more accurately reflects what's   going on with our assets, and so I can get better  business value for you. That's exactly what we do.  You called it a living model earlier, the digital  twin, a living model. It's improving as the data   gets better and as the physics gets refined. Exactly. A lovely way to say it because we   also say it's a living-learning model.  It improves because it actually learns.  One way you learn, for instance, you learn  by actual experience. Some of our twins,  

we predict the damage that you will see  inside parts of an engine. Then what we do is,   whenever our engine comes in to repair, we  go to those parts and we use computer vision   to take pictures and detect not to only where  the damage is, but the size of the damage.  Did we predict that the crack would be this  length? Did we predict that the damage would   be this widespread? If we did and if we  were correct, we're good. If we're not,   see take that information and feed it back into  the model. The model learns from real, actual  

data that comes back. That's one way to learn. It's learning from itself. It predicted it   would be here. We actually took it. We got it  inspected. It's a little bit off. It learns from   itself. That's one way of learning. The second way is from the fleet.   It can actually find other things that are  similar to it and we then take that data   and bring it in. It's almost like medicine. In  medicine, what you do is you have a section of  

the population that you have used this medicine  or this drug on and you see how they react.  Then you compare that to the person. This person  is this age. This person has these genetics.   We have given this drug to this section  of these people that are the same age,   the same genetics. It works well. You bring it  in and you say, "Well, let's try this drug."  The same thing we do. We look at engines that  look the same. They operate in the same way,  

configured the same way. Can we bring that  learning in so that you learn from the fleet?  Then you can also learn from humans.  We learn from simulators the humans run   or we have huge test sites. We have a huge test  site in Greenville, one of the large turbines.   We have a 200-megawatt turbine there. What we do is we have that instrumented.  

We can run extreme scenarios there and see  what happens. We can learn from that and,   from that learning, send that to the model. We could use simulators. We combine with Oakridge   National Lab and we have very powerful simulators  there. We take that data and bring it in.  You learn from yourself. You learn from the  fleet. You learn from simulation. You learn   from humans doing lab experiments. All of that  learning makes this model a living-learning model. 

Where is the model located? Do you run  the model on behalf of your customers?   Do they connect to your system? Do they have  it in their data center? How does that work?  It all depends upon the model you're running.  For instance, if you take a gas turbine example,   some of the models actually run at the  customer on the turbine because in some cases,   if you're doing a digital twin for performance,  the latency – I've got to make a decision and   I've got to talk to the control system within  milliseconds – that can't be at an M&E center   remotely. That is actually on the physical  machine itself that have models running in   the control system and it's adapting right away. Now, if you're thinking about the life of a part,   and we do twins that measure the life of a part,  that's much longer-term horizons. These parts   live four, five years. You can have that. I can  send that data all the way back into the cloud.  

Those are much more complex calculations in many  cases, and so you'd want the computing power.  It all depends upon what you're trying to  evaluate. In many cases, I have it on the   site running with the asset. Then in some  cases, it's on the cloud. It's all based   upon the value you're trying to pull up. Colin, we have a question from Twitter   from Arsalan Khan. Arsalan is a regular  listener to CXOTalk. He asks great questions.  

Thank you, Arsalan. We appreciate that. Arsalan is making the point that you're chief   technology officer, yet you're speaking about both  technology but very much also about the business   problems that are being solved. It then leads me  to the value that's being provided to the custom.   What's the value for customers and improving  the customer experience of the digital twin?  Let me give you two examples. Let's  start with the first one in wind.  When a customer buys a wind turbine, usually  they're buying multiple wind turbines and they   put it in a wind farm. Usually, the contract is  based upon you have set me up with a machine that,   given the wind is going to be blowing this way on  average because we've spent a year looking at the   way the wind blows in that location, I am going  to produce this amount of electricity for you.  That's how they make money. This amount of  electricity is produced. They actually bid it  

in or they have these things called purchase  power agreements where somebody has said,   "For the next ten years, I am going to buy this  electricity at this amount for you," so they buy   these machines based upon the fact that we can  deliver this amount of electricity for them.  But then things occur. The wind changes slightly.  Maintenance problems. What you have is a digital   twin that's constantly monitoring the state of  this asset, making sure that we are providing the   level of performance we said we would provide. We know some days the wind is going to blow   harder. Can I generate more electricity and  maybe save some in a battery? You know some  

days it's going to blow less. But on average for  the year, we have a contract that says this is the   amount that's going to be produced. The twin is monitoring that.   The twin is then tweaking things and tweaking the  low and the pitch and understanding the damage   so that I can predict what's the best way to  align the turbine so I get the most performance.  It is also figuring out when is the best  time to maintain it because we only have   so many days scheduled for maintenance. If we  do more days for maintenance, it means you have  

less days where you're generating power,  and so that affects the contracts we have.  I am monitoring to the point where I want to  catch the repairs as early as possible so I do   a one-day repair rather than a five-day repair.  It's okay if I do three one-day repairs. It's   better than waiting a while and then having to do  a five- or six-day repair because I lose money.  

The twin is the thing that's managing  the performance as well as managing the   life and the maintenance. That's one example. In jet engines, you see the same thing. People   only make money when they fly, so many of the  contracts are written such that I want that engine   to be highly available, so if we can predict when  there would be an early warning on a failure.  A failure of a jet engine at an airport  is recognized as something that's bad,   so we're only allowed so many. Once more, I'm  trying to predict what's the failure rate.   How do I determine an early problem? Once I could  get to that early problem and do a minor repair,   I could increase the availability. People make money, customers make money on having   the jet engines more available. People make money  on having their electricity meet the contract  

requirements. That's what the twin is used for. Ultimately, then, it's a matter of delivering   on the promise of the product or  the service that's being modeled.  Yes, it is. But here's where it gets  hard, Michael. The world changes,   so wind profiles change. Sometimes, there are  weather problems. There are climate shifts.   There are burns. So, those profiles change but  the contract didn't change, so what do you do?  Sometimes, things occur. You are flying a  jet engine in a harsh environment. In many  

of these emerging nations, they are doing lots of  building. They're building skyscrapers. There are   different dust and contaminants in the air  we never suspected. These get on the blades   of the engine. These wear the blades in  a different way, erode the blades in a   different way, so you didn't anticipate that. Now, you've got to have these twins give you  

enough early warning on these problems and enough  ways to mitigate because when that builds up   on the blades of an engine, you find quickly that  that disturbs the airflow and the performance is   not the same, so you increase the amount of fuel  you use. I've got to find ways to balance that.  While it's meeting the commitments, it's meeting  the commitments in an ever-changing world with   every changing operators. That gets hard. The benefit accrues both to the   operator as well as to the customer. Exactly. Exactly. 

We have a question from Sal Rasa on Twitter who  asks about the cultural changes that are necessary   when you implement a digital twin. Maybe we  can start there and you can walk us through   how do we build a digital twin and get it running.  The twin, for us, as I mentioned before, starts  always with where is the business value because,   again, these are assets that are operating in a  certain environment. Unless you can talk about the   business value, very few people are interested. We figure out where is the value. Do I want   to get more yield out of the factory, more  availability of the asset, more performance?  Now that I have that; I go backward into  that. All right. Given that I have this,  

what's the data I need to collect? What is the  insight I want to get? That tells me the model.   Do I have the right insights at the right time? Like I mentioned before, when I am predicting   a failure on a jet engine, I need to do it 200  days in advance for a bearing failure, if I can,   because it takes the airline 200 days  to get a spare engine on that aircraft,   200 days to get a new pilot, a new crew, because  all of these things are tied up or planned.  I need that domain knowledge. Then, based  upon that, I have to build a model. Then   I have to put it in the process. Now is when you do business process   transformation. It's not enough for me to tell  you to do something. I have to know what's your  

normal process by which you switch engines. What's  your normal process by which you get information?  In many cases, we find people just wait until it  breaks. Now I've got a slip this in your business   process. That's business process transformation. All of these pieces need to come together because   only when you do the business process  transformation do you see the value.   In fact, having the tool and you never used it,  you don't see the value. Think first the value   story, the data and the domain story, the models  you build, and then slipping in the process. 

Now, let's come to culture.  One of the hardest things   is to actually think through the cultural changes. Again, in my history, I spent the last 6 years   with GE, 20 years in IBM. IBM, and other companies  that work in that evolution and that revolution,   have a data culture. What you find  in some of the industrial companies,   the data culture is not so pronounced. It is really a product culture. What   I build is I build the most aerodynamic  engine. It runs fast with the best fuel.  

Yeah, the data is a way towards an end. I don't  have to keep the data once I've gotten what I   have. I may keep some of it for FAA requirements,  but the rest of it is not really there.  You first come with understanding the data  itself is a golden asset. Right now, at GE,   that helps because we have a huge  services backlog, over $300 billion   of services. Having the data helps me predict  what I can do for customers better with services.   The culture has to start with a data culture. The second aspect of it is that you've got to  

make believers of the chief engineers that are  there, the financial leaders that are there,   the CEOs, and general managers that the  data can make a difference. You have to go   in and do specific pilots, usually on their  hardest problems that can't be figured out,   using just the knowledge they have and using that  as a way to say, "Look at what has happened and   look at what the money is we've saved." In many of the industrial companies,   you have two things: safety and money. Once you  get the safety right and you get the money right,  

people take notes. People now say,  "Wow, this data could really help me."  Now you've got to build a financial equation that  says, "All right. Now, if I do this with the data,   I can save so much money," or "I can gain so much  money." Ah! Now it becomes truly interesting.  I think it is all starting off with  understanding and getting that data   culture going by hitting a few key problems  in which you can show the value of it. Then   building that business case in which you could  clearly see it on a problem they're focused on.   Then coming back into the data  culture that actually says, "Well,   how do you respect your data? Do you store  it? How do you clean it? Where is it kept? How   long is it kept?" Then reminding people. The people who use data respectably—Amazon and  

Google—they spend $4 billion or $5 billion a year  (or have spent $4 billion or $5 billion a year)   for five years to get this right. They don't do  it on $100 million or $200 million or $5 million.   It's not something that you flip a switch on. That data culture is really gotten because you   have won some battles where they didn't think  you could win and you've built the right heroes   and they see how we can help them. Then you  go into the cultural war of changing things. 

Arsalan Khan follows up saying, "It sounds like   you think like an enterprise architect." With pride. Thank you. [Laughter] I mean   I don't think I'm that good, but I have  to because the challenge we have when   you're dealing with these larger assets (or even  smaller assets) is that you're thinking about the   design phase where the engineers are. You're  thinking about the manufacturing phase where   it's manufacturing. You think about the services  phase in light of the money you make then   and the response to the customers. This is an enterprise-wide play.  

If you don't do the right things, if you  don't understand what occurred in design   and manufacturing, chances are you're not  delivering the right value and services.  If in services you're trying to make  changes and you don't deal with the actual   life as designed capabilities or problems  inherited in the materials and manufacturing,   it doesn't work. This is a systems problem so,  every time you look at a systems problem, you have   to come at it from that enterprise-wide view. Yeah, I would love to call myself that. That  

would be an honor for me, but I am just here  dealing with, how do I actually make sure that   I deliver value for the customer, then value  for the company, and it does fall across the   entire enterprise itself – make no mistake.  Also, recognize that quickly and work at the   enterprise will get the fastest results  and build the greatest sustainability.  Colin, you mentioned earlier that the cultural,  the human dimension is the hardest part. Why is  

that given the obvious complexity and size, scope,  scale of both the data and the physical models   that you're creating when you build digital twins? I think there are two things here when you think   about the human dimension. One is,  in many cases, especially when you   meet some of the people who have been working  there for many years, there's a notion that,   "I've seen that and I have a gut instinct." In many cases, they do. I wouldn't decry that   at all. There is an instinct built  upon many years of doing something. 

The challenge we have in many of these  environments, though, is that the environment is   not the same. You may have built that gut instinct  in an environment that wasn't that dynamic   a few years ago, but now it's very dynamic. I'm in the energy space. The rate of change of   renewables coming in—large-scale solar panels,  large-scale wind turbines, or even small-scale   on people's houses—that industry is so dynamic and  they are changing. Regulations are changing. There   is a variety of electric vehicles showing up. The gut feel that you have right now, in the  

era it made sense is now being reshaped because  there are so many new, dynamic things. You are not   going to be able to understand the relationship  between all these things and the impact that   those things could have on you in a way that makes  any sense, especially when it's rapidly changing.  I think that's a hard thing for humans to get  in mind because it says two things to them. One,   it says, "Am I less valued?" That's not true  at all. You are valued in directing what the   AI and the data do, but that's a feeling we have  and it's a personal feeling we all think about. 

Then the second thing that tends to happen  is that there are these new solutions that   come in that say things like, "Well, I can  replace talent by this AI solution." Again,   maybe in some jobs where it's now due to  interest, but that is not true in many of   the environments I'm in because what you find is  that there is a lot of data you have not captured.  In many cases, when you think  about data that's been captured,   people seem to capture a lot of data. But we  generally capture data to solve a problem.  

That's why databases have schemas. I had  a problem. I use a database to capture it.  If the problem is different, you may not have  captured all the data you needed. The notion that,   "Oh, I have all the data I ever need to  solve every problem I have," is not true. 

You may need different data or you may  have captured data, but you may have   captured the wrong quality of data. It may not be  synchronized in the right way, so you still need   the human there saying, "Here's the way you  should capture this data. Here's the extra   data we need. Here's the value of that data."  Those judgment calls still need to be made.  Everyone worries. Well, I shouldn't say  everyone, but quite a few people worry about  

the fact that this replaces me. It does not. Those are the things that you think about.   That gut feel that you worry that you've lost  that edge. No, it's not true. The thing is just   more complex. Or that I will be replaced because  this thing already knows more than I know.   Again, not true because we may not have  the data that reflects what we need to do. 

Then the last thing is the business process  itself. A lot of my business process is still   human. You still need to call somebody and do  something. You still need to get the engineers   to get something done, to  field folks to get something   done. Then you still need to explain it to them. There is still that part of it that we've got to   work through in which it's the humans and the AI  coming together that makes sense for productivity.   It is not one or the other. That is part  of the cultural revolution we need to have.  On this top, we have another question from  Twitter. "To create a data-driven culture,  

do you think about specific incentives that  you can provide to individual contributors   or, alternatively, how do you drive and  create that kind of data-driven culture,   as you were describing?" In that data-driven culture,   I think about first of all high-level purpose and  motivation. If you don't have the purpose right   in what you're trying to do and you haven't laid  that out there, people don't get on the bandwagon.   That's the first thing you get right. Then the second thing you get right is,  

can I get the right standard work or the right  process? If the process doesn't include the data,   and if decisions aren't made on the data,  then almost nobody wants to collect it   because it's the process itself. You've got to  change that business process where the business   process is reflecting the fact that you are making  a decision based upon that data and everyone will   be rewarded (customers as well as your employees)  based upon the right decisions to that process.  Then the third is motivating and incenting  people. Now, again, to collect the data,   sometimes it's not just the employees. I have to  incent my suppliers to give me that data. I have  

to incent my partners to share the data. I have to  incent the customers to give me the data on when   they're doing things to make the twin more  accurate. There's an incentive on that dimension.  In terms of people, yes, there is an incentive.  There are a variety of ways that we look at doing   that. One is, we have these innovation metrics  that talk about how many key—I should say—hard  

problems that you solved using the  data. How much was your solution reused,   you know, your modeling solution used? Again, those are incentivizing people   to build these models. Then there's another set  of incentives in which we talk about how much   reuse did you have? One way I can get a lot of my  data scientists and my engineers more productive   is to have them reuse things that are done. In many cases, you talk to them and they say,   "Well, no. I'm brilliant. I'm the only  one who can build this." Well, you know,   you have other brilliant friends and  colleagues. Can you not use what they use? 

In some cases, we've begun incenting  people to reuse things because,   if you reuse things, you actually  do more. You get more productive.   You can maybe tackle four problems instead  of two, and so you make more money because   you've tackled four of the harder problems. Yes, we have done a variety of very—I would   say—targeted experiments to help us do  that. But we are still in the early stages   and we need to emphasize that more. That  is a big fact in our innovation metrics:  

all our own reusability, the model generation,  and also the collection of gold data.  It's amazing. There are people who collect gold  data. Gold data is data that's reused and of high   value. If you collect that, everybody shows  up to run models against your data sets.  They say, "Well, Colin has collected this  great data set. I want to run my model  

against it. People keep giving him good data,  he cleans the data, and he has it ready for us."   That's value. We would reward that. Yeah, we're doing it now, but it is   very early stages and we do not have  it right. Again, it's an evolution.  Colin, as we finish up, what advice do you  have for organizations who are listening,   folks listening, and saying, "You know this sounds  pretty good"? How should they start? What are the   types of problems that are most amenable and  make most sense to begin with when it comes   to thinking about creating a digital twin? The first thing I'd do as a technologist is   don't think about the technology. Think about the  business problem. The business problem is the one  

thing that will galvanize everyone to give you  the data you need to do the work you need to do.  I would think, "What is the biggest business  problem you have that I think that I can solve?   Can I express that in a way that your  finance people, your sales folks,   your engineering folks, the services folks,  and manufacturing folks would understand it?"  That's the first thing I would suggest. Look  at the actual business problem because you will   get advocates once that happens.  They'll understand your purpose. 

Then the second thing I would suggest—and again,  I've made the mistake so many times versus me   telling you how to do things right because  of all my failures—I go after cost problems.   The problem I have is, every time I go after  revenue, if I say I can grow the revenue base,   everybody is not sure. Was it what you  built or was it the way we sold it?  Cost is really easy to find. Why? Because I can  show up inside my manufacturing plants, inside my  

engineering organizations, and I can say, "Which  costs do you need to remove?" More than that,   after I use the data to remove  the cost, you can measure it.   You can remove enough things so that I can go  after removing that cost. Then what I say is,   "Well, if I can save you so much money, can you  not invest that money in me to save you more?"  The great way about looking at costs is that by  doing that, I create a pool of funding for myself,   which is good because then I can  say, "Now we'll save some more costs,   but then we'll put the rest of it in going after  new revenue." That's the other thing I would   say is find a well-known cost pool where it's  defined so that you could show the value of it.  The third is, make sure you have the data. Okay,  I'm a data scientist. Whenever people tell me,   "Oh, I have captured all that data," if you have  never used the data, make sure you have the data. 

You go into their databases and you realize,  "Well, oh, a lot of the sensors at that time   weren't working, so I have data that's corrupt.  Oh, I had the time stamps wrong, so, oh, man,   this thing doesn't jive at all. I have  inconsistencies. I took the same things   in four databases and they all look different." Make sure you have the data. Eighty-percent of   the task is getting that data right once you know  the problem. Spend time and really ask them that. 

Then the last thing I would say is that after you  build that model, make sure you go back into the   business process and transform it. If you can't  show how that model you created with the data   delivers money, everybody is going to say,  "Well, I'm not sure it changed anything."  This is the way you look at it. Really  think about it from looking at the money,  

understanding, going after things  like cost and then growing from there,   making sure the data is there, and then looping  it back in so that you surely deliver the money.   That's been tried and proven. I've made many  mistakes not following those things. That's the   advice I would give, Michael, at this time. Wow. That was awesome. Thank you so much.  We've been speaking with Colin Parris. He is the  chief technology officer of GE Digital. Colin,  

thank you so much for the education today. Oh, Michael, it's been my pleasure and honor to   be here. I'm delighted to have this conversation  with you. I would love to come back in any way and   help whenever I can. Thank you again. Everybody, thank you for watching,   especially the folks who contributed  questions. We have great shows coming up.   Check out CXOTalk.com. Subscribe to our YouTube  channel and hit the subscribe button at the top  

of our website, so you can subscribe to our  newsletter. Do it now and tell a friend.  Thanks so much, everybody. We will see  you again soon. Have a great day. Bye-bye.

2020-12-26

Show video