GE Digital User Conference: The Future of Digital, Colin Parris

[Music] [Music] [Music] [Music] everyone wakes up every morning to a world that must keep turning the world can't stop so neither can we because the things we make help make the world go round they make it cleaner healthier and more connected it's what we build that keeps things moving forward so with every turn we'll keep building a world that works ladies and gentlemen please welcome colin paris chief technology officer ge digital all right good morning good afternoon good evening it's my pleasure to be here let me first thank you by for joining our 2020 user conference um i we much appreciate the fact that you are spending your time with us understanding what we can do to help you so thank you again for being here so the topic that i'm going to address is the future of digital but first let me give you a context and i think it's a context you would have seen in all of the exhibits and all of the sessions it's really a context around business itself it's a context about how we make your business smarter flexible and more profitable by allowing you to deliver more and more value to customers and so that's the context we have so we'll talk about digital technologies but that's what you should be looking for how do we become more valuable working with you to building a smarter business a more flexible business more profitable business so let me start by talking to you a little bit about digital twins and i'll put it in a context view it's about digital twins in use i'll show you a little bit about what we're doing now you've seen many great examples and then what i'll do is i'll actually take you on a track to the future again using these technologies to make your business smarter more flexible and more profitable then i'll end by putting it together and show you how you actually do it and how we are using one approach to actually doing that within gpu so let's start with this notion of a digital twin you know we've talked a length about it in the conference i think you've seen us describe it but let me start at a high level and describe what it is and let me also tell you about this context about a so what you do with a digital twin is we're taking the knowledge that you have from design manufacturing services any operation that you do combining it with the actual data that's happening operationally or data firm inspection to give you insights to give you business insights right how do i understand and get an early warning on a problem how do i predict how long i have in terms of remaining life on a part how do i optimize my operation how do i utilize you know by assets in the in the most important way but that's just the algorithmic part this is what where most people stop they see twin is a model it's an algorithm it tells you things after you get towards something the real value is and can you take the right action and drive business value okay if i understand that i have you know an early warning can i take an action and do a repair early so that i keep my availability high if i can predict the remaining useful life of a part then i can actually schedule when i should have other parts there the customer when i should build other parts so i can optimize my inventory if i can optimize my fuel burn then i could actually understand my profit profile for the next month next quarter as i see the value that i'm getting so taking these actions to do something that equally is important and that's the entire idea behind a digital twin in use it is not a single technology about modeling it is not a single technology in which we are focused on simply understanding uh physical construct in in a digital world it is about all of the technologies that allow you to do that it's about the data capture technologies that you use simple ones in which you have somebody inputting something or maybe more complex ones and achieve the computer vision to look and understand spoilation on a blade it's about the data tagging technologies that you use to tag that data coming in it's about the edge technologies you use to capture data on the edge the networking technologies you use to send that across it is about the computing runtime technologies you use to run the analytics and then it's the way that you deliver that outcome either through a ui or through a workflow to get a real action done all of those are incorporated in the digital twin in use and that's what you see us doing and that to me is an exciting technology because it allows you to take these inventions you know whether it be inventions and networking or inventions and analytics and put them together to get that innovation and that innovation when you put these things together it's how you deliver the actual customer value and if you think about our digital twin in use context now when new technologies show up like 5g now you put that in context you say okay 5g gives me more data so with more data maybe i could actually rewrite the model because maybe i can get a more accurate model 5g also gives gives me greater latency well less latency so with that less latency maybe i can use that advantage of my control system to write an analytic that could control and optimize my asset so again you can now take a technology like 5g and put it in the context of the digital twinning use and you see how it benefits your business this is what we do this is how we think about digital twins now if i take that and i understand that these are digital twins in use and there are a variety of digital twins the purpose then is how do i actually get value out of it and in many cases what you what happens is that you actually have to change the business process to be able to get value deliver value to the customer monetize some of that value for yourself as well that's a key part as well because again if i give you a tool and that tool is on the shelf or that tool is not a part of your business process the way you make money then the ability for you to improve your business outcome is quite low so we think about all of these twins in the business process itself we find ways to incorporate it so it delivers value now i have here three types of twins you know we have asset twins twins that are focused really on understanding and optimizing your assets we have network digital twins ones that look rather than a single asset but a broad collection of assets in a network and then we have process twins those that understand the processes that you're running in your plants and your systems and and give you insight on how to take the right actions to optimize that so now i'll talk a little bit about each of those giving you examples and then i'll pull that all together because again we're on this journey of twins in use pulling the technology together that delivers the value that you need for your businesses so let's start with the asset digital twins and these are twins that are very well known in fact when you see a lot of literature about them this is the twin that most people are looking at and it's mainly focused on thinking about the health of your asset and initially if you have a view on the health of your asset you usually go after two things how do i optimize my maintenance schedules the first thing you do with the health of the asset is give you early warning on action because if i can take a repair action early that means i increase the availability of my asset and i don't disturb the maintenance schedules i have when i take that asset down and put it out of commission in order to do significant maintenance also by getting these early warnings and understanding the health i can predict and avoid that unplanned downtime and we've been doing this for a long time in fact right now when you look at the you know the assets we have and the capabilities we have out there with our digital twins we are already reducing operations by 1.6 billion dollars we've seen that over several years and we have a number of blueprints available these blueprints allow you to take pre-package you know in some sense configurations and capabilities of twins bring them into your environment and then use your data to tune it so that you can tune it to the level where it is closely approximating your asset and again as these assets run as more data is collected more and more of that becomes available the asset becomes a much more closer the twin becomes a much more closer approximation of the asset itself so these are the asset digital twins one good example of that is the work that we've done with competitive power ventures cpv in maryland what cpv is doing it's using the asset knowledge of the health of the asset and an understanding of the state of the asset to advantage itself so what they're trying to do is that they're a power plant what they do is they generate electricity and what you're trying to do is at the time when the electricity prices are the highest what you'd like to do at that point is to actually over fire your turbine increase the peak power so that you can produce the maximum electricity when the price is highest and that's where you get the most profit now nothing is for free so obviously if you are over firing the turbine if you're increasing peak power at some point in time previously you would have had to actually reduce your efficiency so you may have run a little bit less efficient in that way you are you're actually running not as hot and so you consider the life because life gets consumed whenever you increase the temperature of these spots during combustion so you preserve the life and at times when you have the maximum electricity and hottest days or coldest days in the year you actually then uh over fire increase that peak power that again requires not only the twin but a beautiful connection the controls remember we talked about the fact that we had insights you know so you use the digital twin to get you inside and then you've got to take an action well this is the digital twin combined to the control capability and so that combination allow them to increase their peak power by 10x 10x so imagine at the time when electricity is highest i can deliver my most profit it also inclu improve the day ahead accuracy so once more you're using these asset twins and the insights into health and how your system is running to give you a significant advantage let's go on now and talk a little bit more about you know some of the network twins so especially when we think about network twins one of the twins we spend a lot of time on as i think you've seen in several of the sessions is the grid the grid is the largest you know man-made machine on the planet and i would believe the most complex and guess what it's getting even more complex because we're adding renewable resources we're adding electric cars we're adding batteries people have them right now in power walls at the same time the dynamics of the world are changing because of certain aspects of the weather you're having severe heat in certain areas you're having fires in california that level of dynamics means you have to have some type of model that can dynamically change to understand the situation to give you the best insights possible and this is why you need a digital twin you know in many cases before we had models and models were fairly static we've always had models you know of the grid you know and but the instance you have now where the dynamics are increasing because of new additions of new resources new distributed you know resources as well as conditions around in terms of the environment you have to manage that in some way and while you have a mandate to deliver stability on the grid and availability you also have to reduce cost and so what our digital twin does with them is understand the grid itself and say what's the ways i can reduce cost while i'm keeping that level of stability how do i in think about all of these new additions require planning how do i do them faster so i can reduce the planning time that's needed to add these new distributed resources or respond to conditions in the weather all of these you know allow a lot of the great capabilities that are out there and some of these utilities to flourish especially the time when they need to reduce costs so that they can actually in turn make new investments given the the massive changes that we're seeing in that grid one example is that of energy queensland so this is a government-owned utility in australia provides power to over 2 million customers and what we're doing here is we're trying to help them understand the relationship with that twin between that is the distribution management systems and the geographic information system the geographic information system contains information about every asset they have out there these are transformers the transformers are operating in this way this is the state of these transformers you know here you have a switching system how is that operating you have to have a view of all of the assets that combine to produce this network this massive network are they working and how are they actually responding to that distributed state that's changing especially in the distribution system so this is what the twin does and their focus is actually reducing networking risk every time you do unnecessary switching every time you actually do operations you know there is a greater likelihood of failure will occur so how do i actually get a view a state of my network such that i can say these are the minimal switches i need to do such that i can deliver this power very stably to you then while i'm doing that how do i actually use the right sources of power in such a way that i am reducing the cost of that power on the network itself and this is the work that we've done with them again another powerful digital twin in use the third one i'll talk about is the process digital trim what we're trying to do here with a process digital twin is take a look at complex processes that are in your manufacturing plants that are in this in any of the systems that you have and get a living learning model about that process which is what a twin is because strength in many cases they constantly update themselves because they are adapting given the data in the system and they can adapt by their own data or they can adapt by information they receive you know from other assets you know that looks similar to them in the fleet or simulators that you use to give them insights so these process digital twins are important things so in the process digital twin what you're trying to do in your plant is you're trying to meet those quality requirements that you have and meet those requirements at the same time you want to increase your oee you want to increase your throughput you want to reduce the scrap in your plant you want to keep the quality level as low as possible all of these things happening simultaneously this now is becoming more important because if you look at the world especially some of the manufacturing examples you've seen right it's changing kovid showed us rapid changes in manufacturing plants because not only do i have to deliver certain types of skills or certain types of goods because those are those are needed essential there are others that are not essential that are not in demand right now how do i quickly shift that you know while i'm shifting that how do i ensure my quality is the highest how to ensure my throughput is the highest how do i do all of this with less staff all of these dynamic environments are especially environments that the twins are good at and i don't think these will be changing in the near future so one good example of that is xaru zaru is a heavy coal and um materials company headquartered in south africa and what they're trying to do is they're trying to understand how could i increase the trooper and so what we've done is worked with them building that process digital twin and they're integrating that process digital twins with that their control system such that they can actually take the output from the work we do in c sense and tie that into the process controls in the plant and increase their gain increase their troopers initially they went after a five percent gain what occurred was a ten percent output great results but we find that often because we find in the journey because again a tuning use means you have had initially to pull that data analyze that data understand your process because you've got to integrate it into your process in doing that a lot of things show up it's not just the game that you get from the things you believe you would have gotten new insights show up along the way as you actually go through that industrial data and extract from that value and sometimes you get these bonuses that make sense and again for them ten percent gain so this is the journey that we've been on in these actual twins in use driving this over the last many years to give value to the customers and we have a number of things we see here you know right now where we we have created over 2 million twins you know we have uh under control under management over 6 000 assets now an asset is not one single thing in some cases an asset may be a turbine in some cases and i said maybe a combined cycle plant so even under this banner of 6485 there could be many other sub components that you see as i mentioned before we've created many of these blueprints the blueprints actually help you by giving you a template or something you can use to accelerate your development of twins and then again you can customize them and we monitor a lot of tags the notion is once you recognize the value of data there's a thing about industrial data we spend a lot of time on not all data is created equally by actually doing the work by putting these twins in use by examining that data and building the twin you quickly find that you have a data value map some of these some of these pieces of data have a big impact on the models and the decisions humans make and so if that's true if this data this sort of data has a great impact on the models great impact on human decision making should you preserve that should you say that's highly valuable and so you would like to capture more of it clean it protect it nurture it and that's what we do with these tags these tags not only give us the ability to actually run the twins they give us a view on the data value maps that we have in a variety of companies and you see the value there 1.6 billion the other thing i notice is the six day catches per week because we have an industrial management system that actually is getting input and looking at things we can find a number of things that are happening at customers we monitor these for customers and we say well if you do things like change the sensor that two thousand dollar sensor change could actually you know result in you know avoiding a 30 million dollar failure in one of your plants you know that that's a significant you know issue that you can quickly avoid now again not many of them are as dramatic as that you know but quite a few of them if left for prolonged periods can cause significant significant disruption so this is the journey that we've been on and so that's a journey i think that's going to continue but i'm here to also say we're at it's infancy imagine you know we've seen transformations for the last two decades gone in the consumer space this has now begun in our space so i'm delighted you're with us on this journey but there's more to come in fact let me tell you about the more to come in the next section what i'm going to talk about here is continuing that theme about digital twins in use and driving business value i'm going to talk about what are the new technologies and ways that we're using these twins to actually increase their value right and again we're back on a specific business focus because the technology for us is all about giving value so the first thing we're going to talk about is how do i increase business adoption so i have these great technologies but let's be real there are cultural factors people say well i don't understand this digital technology not sure what it's doing and because it you know i'm not sure what it's doing i don't think i want to adopt it this fast i want to wait until it's perfect well we have a technique called humble ai and we'll talk a bit about that you know that increases business adoption it gives you confidence to adopt these technologies and use these digital twins the second is okay so now i'm at the point in which i'm using more and more of this technology isn't there business risk am i not putting my business at risk for cyber attacks you know tell me how i should think about that well we'll talk about technologies again that we're working on you know with government agencies to deal with that and one of them is digital ghost and then the third is acceleration all right i bind to the fact that this digital twin is going to give me value right but i have that data is locked and how do i extract out of that data even if i unlock it how do i get more and more value and then put that value to work faster well one novel work that we've been doing for the last three years is what if the machines to talk to us what if they could communicate with each other figure out a couple of things that were wrong and then suggest to us well i think the problem could be a couple things we still have to go to the work we still will go investigate and inspect we still actually make the decision as the humans but now the machines are working with us a lot closer and we can accelerate that business value then finally i'll show you how you all put it together as a precursor to what you're going to see betsy talk about how do i actually implement all of this value in a complex organization like the ones we have so come on this journey with me now so let's talk about a few things let's talk about this notion of humbling so what is humble ai so many of the models that we build you know so people say well i build models i've built models for a long time calling what's different about the digital twin that's true we've built models when we do design we build models only have a serious services problem the difference though is that right now with the digital twins i'm taking that model and rather than after i do it after design i leave the model and the model gets discarded until it's it might be needed later or it might be used in some contexts and services i'm building a living learning model i'm building a model that continually gets data and actually adapts to the environment and it's unique to that asset right and so when i build that model i use data right so what happens as a data scientist myself what happens is that well the best models i built are based upon if i have the most data so let me give you this wind example so this win example that shows wind speed over 24 hour cycle why it's so difficult to really optimize these wind turbines is because the way the wind changes you know over time over distance over seasons here is over time so what you see here is the wind speed you know over a 24-hour cycle now what you should also know is that the wind speed varies by the height now these turbine these these wind turbines can be very tall right they can go up to 150 meters so what you'd see is that the bottom of the wind tower that blue band of wind that's the speed of the wind at the bottom of the tower at the tip of that blade at the bottom what you see the yellow is the wind speed and the tip of the blade at the top so at various heights the wind speed is varying at various times of the day the wind speed varies if you take that same turbine and move it a mile down the road it's entirely different if you have that turbine and you look at this profile in summer in winter in fall it's entirely different if you have an anomaly that occurs so for instance you have a hurricane or some storm that goes through these profiles change for the next four or five days before and after so to have one algorithm that a human can write that says i could actually encompass all of this and get one algorithm that can optimize the best energy coming out of this wind turbine is very hard so we use these digital twins with ai technologies and the ai technologies takes the data and can actually create a twin that can generate greater electricity for you now here's the problem much of my data when i do that much of my data was actually derived at a certain incident so for instance in this case you know for whatever reason we had a lot of data between six and seven meters per second wind speed and so because we had a lot of data we built a great twin that between six to seven meters per second gave you much better performance in the algorithms you created before so that's the region of competency so what humble ai is all about that says because i have built this twin out of data in a certain area i actually think this twin is competent in that area outside of that area don't use the twin use the other algorithms you have and then feed more data back in and get the twin to be competent so this is the ai system that's humble it knows what it knows i know stuff between six and seven meters per second outside of that you use another algorithm give me more data let me allow me to learn and that's the journey we were on and so between six to seven meters per second what you have is that that twin that's the region of competency and that twin performs well now when wind speeds change you know maybe there's a there's a storm coming the wind speeds change most of the wind speed is either above it's a you know eight meters per second or below five meters per second don't use the digital print use that deterministic algorithm that you created yes you'll get lower performance right but still you'll get some performance and it's state and then what you do after that is get more data to improve this digital twin to allow this digital twin to learn now why people like this and we've been doing this in a number of situations right now you know and once you get that larger region the competency the twin gets improved now why this makes a ton of sense for folks is that when i go and i tell a customer right now i don't have to learn everything i'm going to take your data give you a twin in a zona competency that you're comfortable with they feel better i can have that discussion with the chief engineer the chief engineer says okay it makes sense because there's more data i can understand it plus it's a limited window i am not putting the entire business at risk at this window which is that zona competency we're gonna try it if it's successful i feel confident i expanded even more i feel confident i expanded even more so it both allows a business to adopt it faster because it's lower business risk to them and it also allows culturally you know for quite a few people to understand well you know it did do better than the algorithm and so now i gain more confidence in it and again this is all on a quest to drive business value for us one example we have we've been testing these things for about two years we tested it on 69 wind turbines in ge renewable and now it's at a point where you know we believe it's right and so it's an atq you know we have it out there available to quote and so right now we have customers picking it up right now now that same technology we're also looking at in parts of power and that's a technology that's going to come to you soon think about what this can do for your business where once more you're dealing with the cultural parts of the the digital discussion which you can say well within that zone competency i can prove steve it's better and then we keep it in that zone outside that zone you're back to the deterministic algorithms so now that we talked about that let me talk to you about the one that everyone worries about we have all these digital technologies calling what's gonna happen now so maybe we have a an episode in which someone you know takes control of our asset we have a cyber attack so how do we think about that so i'm going to show you a little video here but before i do let me give you the concept of digital goals so digital ghost is using the digital twin because in ge we design an asset we manufacture the asset you know and much of it is done with physics you know as the core of it so for instance in gas turbines there's something known as the braking cycle the braking cycle says when you actually bring in fuel and you explode that fuel it generates this amount of that combustion generates this amount of heat and so i know the amount of heat that gets generated so these sensors i have three sensors sensor one should be 100 degrees center two should be 120 centimeter three should be 140 i'm making up some numbers but because you know i have a cycle that has been you know designed you know using physics in mind and that has been tested over the last hundred years this is how the physics work now if i know that i also know when i put in my control systems and increase the fuel or increase the airflow what happens then is the temperature changes but it's known how it changes so if at any point in time i have sensor 1 saying 100 degrees and sensor 2 saying 120 incentive 3 saying 400 degrees i know by physics that's not possible in this universe so i can use the fact that i have designed this according to certain physical principles to actually do cyber physical capabilities i because i understand how my system is running digital twin in use i now have the ability to use that physics to say i think sensor 3 is either broken or it's been hacked and so now what i do is i see that allowed me to first detect the problem and because i know center one and two is correct and i compare it to what my twin says all right sensor three is the problem i can then isolate the problem the sensor 3 and then what do i do next what i do next is i say sensor 3 has been hacked let's turn it off but the control system still needs input from sensor 3. what do i do i give it the input from the digital twin because the digital twin is still accurate and it can generate sensor three information okay it might be slightly degraded slightly less slightly different and so what the controller knows when i give it that signal is maybe the performance will be slightly less but what i can do is still have a running function in turbine because i've used this cyber physical capability to detect an intruder to isolate it to a particular sensor to replace that sensor using the digital twin and so this system runs so let me show you that in action and talk a little bit more as you see it go through so i'm going to run this video for you here now imagine what you have here is that turbine below and you see the centers on the turbine and it's turning red somebody is deliberately hacking your system to cause problems in fact smoke's coming out and so given that you know right away your controller says there's a problem the digital twin and ghost recognizes this alert we have a center that's wrong we quickly have to figure out and isolate that sensor because we'd like to restore normal operation then we use the twin and said we get the information from the twin we bypass that sensor and that twin feeds into the mark 60 controller and now you have a system that's running once again maybe slightly degraded but you continue the use of your asset now these are some of the capabilities that we're building we've been building this with the doe for the last few years we've partnered with them they've invested about 22 million dollars worth of grants and we have built it and tested it this is what i mean in use so we've gone to the greenville facility part of the mandate that the department of energy had was to put it up on the large test center we have which is an actual operating 500 watts megawatt gas turbine that we have in testing and what we do there is we successfully stimulated it we brought them there there's a bunch of criteria they brought there their officials and we actually they simulated it put a couple of attacks we detected it localized it and neutralized it in this capability think about what that could bring to your business later on as you get the confidence because you understand you know how this works it's a digital twinning use you actually have the control capability to perturb it a couple of times to understand truly how these sensors and the actuators you know um actually work and you use that physics advantage to give you that ability to deliver value and protect value with your customers so the next thing i'd like to talk to um talk to you about and so that is one that we have begun working on a while ago but the reason i put this one up you know it is the furthest reaching out you may not see this for the next four or five years but but it's where we're going and you always have to keep these things in mind because where we're going it you know it gives you a roadmap to the future that we as technologists need to look at to see if it's valuable to us understand the use cases so as the technology develops as i mentioned before digital printing use is about putting the right technology to use in the right place it's innovation to deliver value here's one that's named me that i've been investing in a while it's called twins the talk so let me give you the scenario so with wind turbines what we found is that you know they produce signals from the sensors they have and what we were saying is that could we find a way and we've done this with other systems to teach that wind turbine a language so that those twins can talk so the notion is how do you build a language you build a language by doing simple things if you and i spoke a different language how would we communicate i would hold up a cup and i would say the word cup and you would see the word cup and by and by you'd understand that what i'm trying to do is show you by saying the word cup i'm indicating the thing in my hand then i might test you i might hold up a pencil and say cup and you'll tell me no no no that's a pencil oh i might hold up you know a bottle and say cup and you say no no that's not a that's not a cup and then i hold up a cup and you say cup this is the way this is the way humans have for years for generations you know taught each other languages so the turbulence do the same thing you know if you look at a certain set of their signals you know they have taught themselves that this set of signals that come out of all my sensors means wind increasing this in a signal means when decreasing and maybe it's the pitch of the blade because when it increases you want to actually pitch the blade so that you don't have these things becoming unstable so there's all these signals at the pitch that when you tank together you know become unique so when increasing is a signal when decreasing the signal energy increasing is the signal energy decreasing as a signal right wind turbine blade is a signal damage is the signal snow is a signal and so we've been working on this with with some of the darpa folks and we've actually created a language that we've begun to explore and it's become interesting because we've actually run it on a couple of tournaments and for instance what we'll do here is we'll have one tip and ask a question you know have you seen this problem and then other turbines actually make up statements to explain it and so if you look on so what you see there in the middle is a turban language it's a series of ones and zeros from all the signals and they successfully created it then we went to a small group of turbulence and we actually had them engaged in a dialogue it took a while but after a while they actually had a dialogue and here's the interesting thing after one incident that they explained what you see there in blue below is an actual children dialogue so you see the snow icon then you see the damage icon you see the blade icon and you see the energy decreasing icon what that sentence means is one of the turbines actually said to another in a snow event it was damaged by blades which resulted in a reduction of electricity that's amazing because again what you'd want to do is if you can collect this information from the assets themselves you know you can quickly dispatch people to repair now when they go there they're going to conduct their own operation they're going to conduct their own analysis but by at least having some ideas of what's happening it gets us much better prepared and as this develops you have the ability to get more and more precise to the point in which you know twins are helping and anticipating things helping us be better this again is stuff that we're looking at in the future but it's a future that may not be far away given some of the advances we're seeing so this piece of work again real work you know is work that we teamed up with grc with a couple of large data brands work focus on contextual reasoning and grounded ai language acquisition and these brands allowed us to actually build this over the last three years so we still have a ways to go but i want you to see what the future can become because what what you're realizing right now is that you know the rate of growth of these technologies they're getting fast and fast and you need to be prepared because we have assets that live 20 30 50 years so these are things that are going to be used so think about the three things that i told you before you're going to have to think about business adoption that is humble ai how fast can i adopt it protect it with business risk which is you know digital goals and then you have twins the talk at salary business and now you i'm sure you're saying we're calling that that's interesting but but let's talk about the reality it's a complex thing to implement any of this thing in a real business you know how do we do that because i need you to tell me how i make money on this thing right away okay let's talk a bit about that because this is the piece of work that we've been doing you know i've been working with betsy bingham and you'll see betsy talking right after on some very interesting work here so what we've done is because of paths and larry's influence we begin to understand and tie lean and digital together here's what i mean by that so what does lean do lean focuses on reducing waste that's the major focus of lean how do i reduce waste and how do i actually create business value with data how does it do it it does it by the first thing it says let's lay out the kpis the key performance indicators that are valuable for you let's actually get what is the actual benchmark data how fast is this thing run right now what's the trooper right now then let's go to the value stream map to see where value is blocked and then sure tell me numerically how much value is lost here then we'll begin to create from that you know a bunch of action plans based upon data because that's how we know we're doing better we have attack time is there stack time better we have a trooper is this trooper better well by lean focusing on that data that critical business data that is a dream for data scientists like myself because now that data shows up for the first time and it's usually data that comes across the entire value stream so data from the design all the way through manufacturing services you know to finance because that's what lean is doing it's focused on our actual financial outcome that data comes to me and now i use that data with all of the powerful technologies this is truly putting industrial data through it all the powerful technologies i have at mine and in terms of analytics in terms of the ai operations research and i use that to detect the right problems to focus on then i actually say well given that these are the right problems are there digital content measures that i could put in my action plan right things i could optimize things i could actually data mine and show you where the problems are then based upon that can i actually get embedded in software and put it in standard work put it in the workflow itself so that right now i have the digital capabilities in the workflow and now i have that constant stream of digital capabilities in the workflow constantly getting better and if i tie these two together i get improved margins why do i do why does that happen because i'm focusing lean on a value proposition that ends in dollars somewhere and because of that lean capability lean is changing the business process while the process is being changed i then do digital transformation and put the digital transformation inside the process transformation and that's where i get the improved margins so now i've changed the process i've stuck with digital capabilities in it now the margins improved now i can go after reducing my backlogs taking off cost warranty cost inventory cost i can actually do revenue generation i can do a number of things this is something that has been profound for us and because we've been doing it across many parts of ge we actually have laid it out and formulated a strategy first it begins with discovery right this is discovery and strategy we go with the business units the organizations and we say what's the problem you're looking at you know let's financially later i'll get the right objectives the right kpis then let's figure out the value stream where is how is value created where is it lost and then out of that where is that macro point where is that maybe some stream which you're losing money excess cost let's focus on that once we set the focus area then we actually go on we understand in our case we've been focused a lot on services life cycles now we understand we need to bring people from across the organization this is not just a design problem in engineering it's an engineering problem plus manufacturing plus services it's financial in some cases in some cases we bring in the commercial folks because maybe they are selling capability they shouldn't so we bring them all together and more importantly we bring the data sources in many cases data silo data is always siloed for a good reason what we find is the engineering guys say well i need this data done this way and so they build an engineering action system the services guys say well you know i need my services information this week they built a service society system incredibly enough they know all the stock they don't always have the same underlying asset model they don't always have the same underlying way of interpreting cost and interpreting customers so we need to bring that all together we bring it together in what we call a problem solving report psr and then we go through a value stream map then we do a process analysis and then we do a data analysis this data map is where it gets interesting because there you quickly find out in all these cases guess what you know i have data that's incorrect i have data that's incomplete i have data that's untangling you know it's in someplace else it's not here that's inconsistent same set of data in two different places but they have some different values in them right so quickly you learn about what the problems are so you see your data gaps then you get a data value map you say well i use these models and humans make decisions based upon this data so this data is most important so now i know where to focus i'd have to change all the data i have to change these first get an improvement put that in standard work then i have kaizen once i have that i can continually improve because it's all based on data that's now flowing data that's clean data that all of us agree on so this execution structure is what allows me to actually go in and combine lean and digital i am doing process transformation business process transformation and embedding digital transformation in it so i have the digital capabilities coming inside and the new process showing up that allows me to use that and then i sustain it i have bowlers i have daily management i have operating reviews if there's something wrong i go back in my lean process i actually continue to innovate and this is the journey where i want so i'll stop here right now but i want to leave you with two things the journey of taking you on has shown you this digital twin in use it's a combination of technology it's all put at service to deliver business value all in a context so that when new things show up like 5g or blockchain you know where to fit it in so it gives you value it's not well i found the new technology i got to do something with it just because i need to be cool it puts it in context and we've gone on that journey value has been delivered now in the future there's a lot more coming we have to increase business adoption with humble ai we have to protect you know reduce business risk with digital goals we have to accelerate with this business value by using twins that talk then we finally have to implement business value by using the lean plus digital process this is the journey that we're on in you know ge digital right now using the power of the ge business i'd love to have you join me here we have your beck and call because customers are what we serve and we love to spend more time with you exploring how we make your business better and with that what i'd like to do is maybe move to questions now you know and have ninu help us with that thank you again for your time thank you colin i learned something new every time i hear you talk um and i really appreciate it um so the first one is from i believe jovare i apologize if i'm pronouncing that wrong from gulfstream aerospace how is ge applying digital twin for aviation engines does ge have digital twin implemented for aviation engines is fleet data availability and transmission the main challenge applying digital twin in aviation and how is ge addressing this um so let me let me try to keep that show up and again we can answer these things offline as well g is using a lot of digital twin capabilities in aviation um a lot of it is really around at least at the start our own parts so for instance we have situations where we find the harsh environments if you fly into hot and harsh environment things are going to curl on the blades of the the engines pollution pieces that the thermal coating can come off so what we do there is create a twin so before what would happen is that we have no idea when it would um happen so what you do is you say well every 200 flights i need to do an inspection so that's pulling an asset out of service to do an inspection and it may be that nothing happens because maybe it's not flying in enough harsh environments or maybe the contaminants aren't right so you are wasting you know um time pulling it in because it's okay so you've reduced availability and you have also you know put added useless work inside a shop so we built a trend that can detect for us based upon where the asset flies the humidity the contamination exactly how much damage would result on that you know blade and when you bring it in you know to actually do inspections in this way you don't have to bring it in every 200 flights maybe you know for some flights based upon where you're flying it's you know two thousand flights or it's three dozen flights maybe some it's a hundred and that allowed a lot of um benefit um for ge i talk about a little length and some of the other things i have out there so we're using it there we're also using it to try and predict um what parts we will need because some of the lead time of these spots take nine months to a year and so if you can predict the damage and predict what part you would need for an asset you can actually then you know bring those parts in a lot earlier and you don't have money and free cash flow for inventory so yes there's a lot we're doing so let me leave it at that you know but you know great question [Music] thanks son the next one is from seth from metra twin implies the full one-on-one mapping of a system how do you address the blind spots where data is asynchronous or can't be digitized to avoid biasing your models to only the available data yeah that's a really good question so what we do when we look at the twin is we look we start looking at the business value and we have bounds of confidence for it right so there is some data you're not going to have it's like humble ai there is a zone of confidence in which you have data there the data itself may be biased in that so you get a certain when i give you a to an output i give you a probability around that output and so what i do initially when we do work is we do two things one is we actually take a segment that we want to go after that we can address a problem so if i have a problem like i mentioned one of the blades maybe i don't want to address every blade on the planet maybe i want to look at blades blade damage on one fleet that i know flies into a hot and harsh environment and focus on that given that i have more data i know the data is biased and i'm using the bias to help me because it's all biased towards that type of environment and then i have balance of confidence so that's when i start and then what we do is we start with these bonds with confidence and we say even at the beginning when i say the twin says something what customers usually do is to say well for the first six months i won't believe you i'll pull the asset anyway i'll take a look at it if it's true you know i'll say okay the twin is getting better now if it's not true that's okay because i take that same data that i'm wrong about and i feed that data back into the twin and the twin gets smarter so this this is a journey in which you know i am taking the data that you have given the limitations you have and bounding the problem and to use the data that's appropriate twins should never be used in a situation where you say well it's going to cover everything everywhere that's not what it's about you know uh think about it like growing a baby rather than doing a factory you grow the baby and then you add more and more i'm a parent i start off with a baby and that baby grows more and more as i add more data add more capabilities get more believability get more insights and that's the journey that we're on so we start with these type of things you know bounding getting the right confidence interval following the problem and then go from there and then you could figure out because in some cases you'll say well to collect all the data i need i can't afford it so that business is not financially trackable for you and i love that uh parent metaphor that was that was great um next is messier from corber digital does the digital twins simulate the availability in day and day ahead market more concretely can the living model define the pm scheduling based on the expected market clearing price in the day ahead market no um so what we do right now so again it all depends i shouldn't say no blanketly but we have not done that for someone are there other people who may have they may have again it all depends upon you know your level of business risk so if you're thinking about their head clearance um there's a couple of things you've got to truly understand you know whether it be weather whether it be you know the capabilities of your assets it depends on how much data you gather the fidelity of your twin to be able to do certain things in cpv right we do forecasting for them and the techniques we use actually allow the forecasting to be much much more accurate and you use that forecasting to tie that into exactly what you're gonna do uh deliver and you tie that forecasting with the capabilities of your asset to deliver how sophisticated you get is based upon how much effort you're willing to put into it in terms of money and timing as well as what is the business risk you're willing to take you know by um by you know putting these proposals because the the flip side of this is that if i can't deliver that power that i've committed to i've got to buy it from some place and if i have to buy it at the time when the prices were the highest i will financially lose money so it all depends upon the organizations we deal with and so i've talked about cpp there are others who do more sophisticated things that i can't address right now but i think it's a great question but the to me the essence of the question is will you see these things be tied together more and more the answer is yes the reality is that and here's the reason i take that you know part of my background is i spent 20 years in ibm we used to have the same discussions about the financial markets in which we were saying well we had traders and the traders would then go buy the right stock or who made the trades then it got to a point where some of it went to actually automated trading using you know models and financial systems and then bit by bit more and more of it went there because then the traders rather than pushing the button we had enough algorithms and belief in the algorithm that they actually treated at that speed i think as these things evolve you will find that we integrate more and more the business side of it with the control side of it to deliver things from the plants automatically and i think it'll have to be so especially given the event of more and more electric cars more does things that happen in california so i think we're moving to that world do i think we get there tomorrow no maybe five years yes policy has to change you know and and you have to find a way that you can monetize some of that um uh energy that's stolen batteries but i think we're on that path right now great and then last question um to close with comes from tim from dupont sorry timothy from dupont do you distinguish between digital twins and reliability block diagrams and ram models process models compressor performance models or are these all a continuum of sophistication in modeling tools that all classify as digital twins that is a great question so i'm also um g is a founding member of the digital twin consortia um and this is one of the questions we actually spent a lot of time on the taxonomy you know and the description and terminology and taxonomy about these digital twins and you've seen i mean i've seen people um initially in the early years about five years ago amazon said i have a sensor at the bottom of a pump you know when you go into a bathroom and you you pump the soap up in your hand and that sensor you know can tell me if the soap level has reached below level and that's a digital twin you know and again you know there are a lot of things that can be classified as a digital twin you know and it all depends upon how you look at it for us there are five things we're thinking about right um i'll talk about the first four the first one is that it has a specific business outcome that we're focused on you know second is that it's specific to an asset itself the third is that there's a continuous stream of data that literally you know makes the makes the twin adapt to its environment you know fought is learning right um i can learn from the data that i bring into myself because i can i can say well given the data i'm getting i'm going to forecast you know something an hour from now and i wait an hour and if i'm wrong okay i learned from that so i can learn from my senses i can also learn from simulators there are powerful simulators all around the world we have similar national labs have that could simulate conditions like that i can live in the fleet right i can learn from humans so the other the photo requirement is the learning you know and the adaption the fifth requirement really is it's about that notion of scale to business value right um you know can you scale it easily by saying if i do these for these type of bearings can i adapt it to other bearings in my business right and so this is the other requirement that we have so we have sort of confined our discussion to those high value assets so are those models are those models very useful they are tremendously useful you know but for me um i am not also i'm not religious about what the terminology is i'm religious about the business value we're delivering you know so if somebody says we you know we add some other things to it and it delivers value i'm okay with that so i'm very flexible how we think about these things they're things that i think about and why do i think about the things i think about the two reasons one is that i'm also watching what's happening with digital technologies the thing about digital technologies and i've been in that all my life again it amazes me that i want to take advantage of is really two factors one is the fact that ten years ago if i look for if i go one dollar of investment in processing power how much more do i get now it's roughly six and a seventy x from 10 years to now i get a 70x improvement in processing power roughly a 40x improvement in bandwidth roughly 15 17 to 25x improvement in storage that's what i get right on the digital technology side okay no other technology who does that now let me flip that that's the hardware side let me put that on the software side before what i wrote i wrote a similar language right okay good then they wrote interpreters to write the assemblers but then we would compile us the right interpreters then we had drag and drop with libraries to quickly build code now we have low code no code before using all the rules now ai takes the lead and makes up the rules i have a technology that is building itself at a fantastic rate at the same time it drops in price with massive performance increases so when i take that in in the back of my ear and i say to myself well what parts of that can i use to advance myself the first one that comes to mind is learning right these algorithms so whatever model you build is interesting but the ability to have that processing power ai power and have that learn rapidly if that isn't plugged into your system then be really careful that when you think you have a model that you think this model is going to help you as the world adapts so dynamically right so that's what i'm looking at i'm looking at that learning ability and then i'm looking at that zoom ability as applied to controls not a control system to get faster you look at you know the nvidia processing power and 5g so how am i taking advantage of that right now there is close to i think the last report i saw was between 400 to 500 billion dollars worth of investment in i.t structure like that happening outside and i'm looking at my little it my little r d budget no no i'm going to use that billions of dollars and bring it in to give me value and that's the value i look at when i think about my twins so you find i put learning in there you find i put creating new models that actually utilize that wave that's happening that i can't stop that's compelling that other people are paying for and that's the what the things i'm focusing on in the twins we have so learning the ability to deploy them in certain ways you know how i take advantage of edge all of that in context is what i built into the twins that i look at so i think all of those models are powerful but i'm not trying to compete with those i'm trying to say i need to take advantage of a massive technology how do i take advantage of that at the same time and get benefit from my customers so that's how i would think about what we're doing here with the digital trip and with that i know i've run over so thank you very much for your time nino thank you sorry i ran over here but um it's been a delight for me and we're here to serve you so please do reach out to us i think we have a beautiful future that was in g digital working with you colin thank you so much it's always um a genuine honor when we get to hear from you so really appreciate your time my pleasure no no thank you thank you team have a great conference goodbye

Channel: GE Digital

Views: 2,811

Keywords: GE Digital, Industrial IoT, Industrial Internet, IIOT, digital transformation, digital twin, industrial software, industrial automation, user conference, colin parris

Id: MzPLPcQguQ8

Length: 63min 47sec (3827 seconds)

Published: Fri Oct 16 2020