so now i'd like to introduce our keynote speaker who's an industry leader with expertise and artificial intelligence that extends to the energy aviation and healthcare sectors dr collin paris is senior vice president and chief technology officer from ge digital we are delighted dr paris agreed to join us today to give an industry perspective on i'm sure what i'm sure for him has been the digitization of everything over the course of his career uh and applying artificial intelligence to the energy sector and beyond dr parris okay thank you very much dr murray hopefully you can hear me properly i can certainly excellent excellent good so so let me know i see the charts are up which is always a good thing so let me tell you a bit about what i'll be talking about um today the talk is entitled digital role in the future of energy but if i listen to what um that mario was saying as well as evan i mean we're going to be tied quite nicely into our view of what's happening with energy analytics but let's talk about the agenda here so what i'd like to talk to tell you about on the agenda you'll see there are four things right the first i want to frame the forces that are uh are impacting the market and driving forward right now to give you a view how ge thinks about it much of my talk here will be in the context of ge because while we're dealing in the the energy we have two prime things in mind right in this industry one is to deliver value for customers and the other is actually to do so at a cost that actually makes it profitable for us as well so we'll talk about those forces and those drivers the second because we're talking so much about you know the digital nature of things i'll talk to you about what's been happening overall in different parts of the industry as we talk about the digital phenomenas occurring in the consumer sector and we'll use those because a lot of those deal with the fact that you can get technology now at volume and great insights and we'll show you how you use those in the industrial sector so that you can deliver value to those the third thing is i'll talk about something specific to that which is digital twins themselves the digital twin itself is the thing the the entity we use the combination of physics and ai and analytics that allows us to do many of the data analytics things we're going to do and then finally what i'll do is give you some examples of of you know how it delivers value and then talk about one or two technologies that we will be using advanced technologies to drive us into the future okay so so this chart will talk a bit about the you know the forces shaping the energy market here so as you can see there's about four forces right i'll talk to decarbonization efficiency you know decentralization and digitization so in terms of decarbonization i think uh given the audience here i think you you everyone here is fully up to speed on most aspects of that let me give you the part that we think about one of the things that is happening as a result of decarbonization is the rise of renewables we all know that but what the rise of renewables also does at the same time it has a profound effect on the actual power generation assets that are right now you know in in play we have a number of natural gas assets a number of coal assets so when you bring these things to bear how do you actually tie these things together and get them to know get into work in a way that makes sense right now renewables the rise of renewables itself whether it be the wind or whether it be the solar the it's not enough right now to give you the amount of energy you need on the planet so you're combining those you know with the assets that you have that are fossil based the other aspect of it is that even when you do that the renewable assets are intermittent so whenever the sun is not shining as you expected or the wind is not blowing as you expected you have to bring these systems up to bear you know when you bring these systems up to bear you know quickly they no longer run at base law they're cycling they're going up and down based upon the amount of energy you're getting from renewable sources that itself causes inefficiencies which which results in you know more cost but also it produces more carbon so there's a tie in between the systems you have now and what you're going for in the future as you look at these decarbonization aspects of things this is all part of the dynamics if you think about efficiency we talk at length about you know we're going to electrification and what we're seeing there is that we have a number you know of smart buildings smart factories come into play we're bringing some of these capabilities into play well those capabilities are great because in many cases what you find is they have systems that monitor them manage them in certain ways and so you can get a much more better understanding about the characteristics of these things and these characteristics that allow you to understand that demand better so we see the rise you know in terms of the desire for efficiency but at the same time without desire for efficiency we are also capturing more deer which can help us going forward that's a dynamic you want to exploit then there's one that deals with decentralization there is a prosumers as well as again because of the drive to do electrification you have many decentralized sources showing up and those decentralized sources show themselves up in a variety of ways some of them could be you know solar rooftops that you see in many of the residential areas but also you have a number of things like electric vehicles that also show up and the grid has a responsibility to understand what are all these new distributed resources that are coming to bear what is the needs that they have the requirements they have how do i understand them and how do i orchestrate them in a way that makes sense oh can i possibly even utilize them in the event that i don't have the ability to actually generate the power that i need as occurred in california that's another powerful dynamic you see and then the dynamic we're all here to talk about is that one of digitization i'll spend uh um a lot of time talking about that one but let me give you two key points that always resonates with me the first for me is that and i've lived through this you know in my career in ibm and my korean ge right now the it's one of the few technologies in which the growth in performance is sort of balanced against the drop in price both occurring at the same time not many technologies can do that if i look back 10 years ago and i look at what one dollar 10 years ago would buy me now in terms of processing power it's 40x i get 40x improvement over the last 10 years same dollar gives me 40x with more processing power bandwidth is in the 30s storage is in the 20s high 20s right so you have a capability in terms of massive performance while in fact the cost drops you can use that game both ways i can either buy the performance so i can get mid-range performance at a much lower cost that's powerful as a technology the other aspect of this technology that makes it so interesting is that everything that supplies data to it whether it be some of the meters we have the sensors we have those things are significantly dropping in cost if i look at the last five years we have seen a 60 drop in cost in some of the key sensors that we use in our systems but that again it's profound because now we can put more sensors we can use sensors in certain ways and we see that becoming even better in the next five years so that capability becomes very profound for us as things get digitized and then we have the ability to use that data and compute and store it in a manner in which it gives us insights so that's the dynamics that you see now while those dynamics are happening our customers are asking us for things the first thing they ask for is let's make sure it's safe we take that for granted but then the next thing after that is how do i actually get more profit how do i increase my productivity how do i increase my revenue at the same time how do i manage all these new government regulations that are showing up you know and deliver what needs to be delivered then how do i adapt to my markets in many cases customers are saying to us i want more renewable energy tell me how much renewable energy i'm getting delivered to my house and my business how do i manage to fit that in how do i talk about the fact that many of these systems i have to gather data are 20 30 40 years old this internal obsolescence you've got to do all of these at the same time so imagine managing all the dynamics and at the same time managing all of those customer requirements so how do we figure that one out if you could jump to the next slide for me what we do is we begin to take a page in our industrial world out of the consumer world right if you'll see in this slide here we unabashedly borrow from what we see occurring with some of the key market leaders in the consumer world because they've found a way to utilize technology to use that data to gather insights and the insights allow them to actually make good decisions decisions that are optimal and those same capabilities those same insights allow them then to actually go drive work or take an action that also is optimal all of it running on a digital fabric what if we could do that in our industrial space so let me walk you through what that looks like and so if i think about these market leaders you know there are two things that come to mind one is the notion and you've seen some of the names right um in terms of the rapid growth and as a business person i always think about that um so for instance you look at amazon in 2018 2018 they will know 2019 2019 i think they did 278 billion dollars worth of revenue not not market cap revenue right you know if you look and and this is a company amazon that's 24 25 years old if you look at um google another great example you know what you see there's about 146 billion dollars of revenue and that's a company that's again in its twenties now i've been at ibm ibm to get to a hundred billion dollars it took roughly 94 years in ge it took 101 years you have companies that are in the 200 billion dollar mark on the 24 25 year old companies that rate of growth was based upon certain key things and a lot of everything you look back is the ability to use that data to actually create a connection to give them the ability to make good decisions and then act on those decisions we just we see the same thing happening with a number of leaders that i have listed here all using that data to do that now how do they use the data if you look towards the right what you see here is the two aspects of it one is this notion of a business model how do i create value can i use that data to get enough insights so for instance in amazon they use the data i am a member of amazon prime they look at what i buy they look at what i look at in terms of the ads they look at what i've bought you know recently what it's given to me or what i've done historically and they can create insights on what collins should want to buy and that's what they do they create those insights using the data now based upon that data they also do things they also then generate a workflow if i select something what you instantly see happen is that that workflow is generated where this goes off into one of their warehouses in many cases if it's a single item it quickly goes and this it gets dispatched to a kiva robot the key robot goes out it grabs it drops it in a box the box goes down a conveyor line the box folds in it's taped automatically by a robot and that thing comes out humans don't touch it the entire workflow was a digital workflow now in terms of by two three things you know and sometimes it has to be packaged the right way they've gotta add bubble wrap and everything else yes the humans may get involved or if it's at a different smaller site not one of the much larger warehouses humans get involved but the idea is a workflow and that workflow allows them to operate at a certain speed and that's why last year they had 319 million you know buyers people who buy from them frequently that idea is a profound idea for us in fact we took it and we began understanding this notion of a digital operating model so traditionally you have a bunch of human processes with digital support what you have now is a digital process and you insert the human whatever you need before you had human insights using data now you have digital insights then you add occasionally you know maybe a human expertise because it's something new that you've seen and before you had a limited scale now it's an unlimited scale so if you take that idea what can you do with it so if you jump to the next page i'll tell you a bit about what we've done with it we've taken the same idea for the last five six years we've been working on this and we have created this thing that we call the digital industrial transformation using something called a digital twin so we take all of the knowledge we have about the design manufacturing the operations and the inspection part of it and what we do then is we add to that the data that we're gathering the data that happens operationally you know whenever the asset is performing the data it happens when we do inspection the data comes out of the sensors that we have in the monitoring and diagnostic centers we combine that together to give us a digital twin so what is a twin a twin is a living learning model of an asset we've always used models we've done it usually when we go to design something we desire we start with a model and then we use that model to simulate a few things before we design it but then we usually stop because after you build it goes out into the field in services we might if we have a big problem we may design another model audio update the model we have to look at the problem and services and then we put that model away what this is though is that once we create that model and it's focused usually on a specific thing we're looking at a specific situation you don't have to have a full-blown model of every aspect of the asset but specifically what you want to look at and we build a living learning model which means the model gets updated every time data comes in whenever our jet engines land we take a small snap snapshot of data and we update the model every day with our steam turbines or gas turbines we take the data in from them we update the model it's a living model and it's a learning model we look at other parts of the fleet in which they have very similar assets this asset is direct same configuration operating in a similar environment being operated the same way and we learn from them maybe they will see a thought or failure before maybe they'll see a cyber attack before we bring that learning into the model we have so it's a living learning model what does it do it drives insights that's the first thing you see it will tell you things like when the do you think something might fail give you an early warning it can predict these the level of feeling predict the amount of damage you'll have by the time it comes in for repair so you'll know you can replace spots you'll know you need as many parts you will know that well i better bring it in earlier because this thing is actually you know the parts are being damaged at a rate that i didn't expect because i'm using it differently and you can also use it to optimize things so you get the insights now once you get the insights now you have to translate that into work because that's how money is made if i have an insight and i do nothing on it there's no way to monetize that or no way to get things better for a customer so i actually use workflows digital and human workflows to actually do things like can you i can use a workflow to change you know uh in a control center in a grid i can do things there or i can actually go through a control system the mark 6e that runs in our ge turbines and i can actually automate that control system or i can actually send a technician out when the technician can go out and see if there's a problem or i can have you know on the usual repairs i can send information to a robot and have that robot take specific pictures of specific areas of the turbine so i can get more insights because i think their problems there so now i do the work process and that work process gives me a result and i use that as feedback to this twin and that feedback gives me more insights and can deliver more value for customers occasionally we find things out by doing using these twins for a long time like in aviation we found out we believed that the greatest stress on the shroud the area around the turbine was when the actual aircraft was taking off we found out by using the twins and understanding the data that rarely it occurs during climbing when these aircrafts are at significant height and then climb even more again that allows us to do design this is the digital twin you know both giving us the insights and they're actually generating a lot of software and automated workflows with some human interaction to deliver value now if you go to the next page i'll tell you a bit about that value so this is all in our space delivering significant value from us using these analytics now here's some examples especially the energy space so the three things we look at that are value to the customer can i give you early warning about an asset for feeling if i give you early warning and you can take an action you can do a minor repair you can you know change the operating parameters and that keeps the asset running that increases the assets availability key for us so one good example here is what we've done with one of the condensers here which is collecting enough data from the condensers and understanding you know how they were designed itself and how they operated we can actually predict fouling right we can predict the level of following the condenser we can go and have that clean before and if we have that cleaned at the right time these things don't fail or they take a minor amount of time and it increases the availability the key though is to predict that enough time in advance that you can take an action before it impacts your production of energy right and these condensers are vital you know especially these combined cycle plants where we actually are taking the heat off of the turbine and using the heat off of the gas turbine and using that to run the steam turbine right and so what you see here is the ability to do that and get predictions early enough so that you can keep the availability as high as possible second thing we look at is continuous prediction so what we do here in many cases for these wind turbines you sign up for something called a ppe right which is a purchase price agreement it's over multiple years you agree that i'm going to deliver so much energy over so much time and what you do then is you have to keep to that commitment now the interesting thing is to keep that commitment there's variations there's variations in the wind you know there's variations in you know it's in certain cases you have certain types of failures you've put the turbine in the certain place and in that place you know you are five miles down the road for a chemical plant you know and a lot of these you know pollutants get in the air they gum up the bearings they cause other factors there are a number of things that could happen over the lifetime of these turbines disturbance last 20 years so what you've got to do because you've got to actually predict failure you've got to predict not only the failures but you have to predict when the aggregation begins to affect the performance such that you can't meet the commitment and in that case what you've got to do is you've got to say well there will be times that this happens how do i actually find a way whereby i can optimize the the the um the pitch of these turbines so i can generate a little more you know electricity at this time given this win input because i know you know in the next you know week to the next day in the next two hours i'm gonna have that dip so that you keep that average at a certain level and so we do a lot of work on that and that delivers significant value we've seen in many of the wind farms you know given the fact especially those that are larger than 200 turbines we can generate millions of dollars of savings because why we say savings is that if you don't make that commitment you pay a penalty right and with that penalty that's been happening we now can add analytics rdi capabilities in digital twins save you from paying that penalty the third thing you look at is this thing called dynamic optimization here's a here's an example you know of a combined cycle gas plant what happens is that we know that there are times of year when the price of electricity ramps very very high when it's very cold when it's very hot the demand increases and the price jumps what you'd love to do is you'd love to over fire your turbine or have that turbine you know be ready for those times when the price jumps and then actually run it as hard as you can to make the most profit then most people don't do that the biggest challenge they have is that this turbine has to be maintained in certain intervals and every time you over fire them or you actually make them run as hard as they can you have the ability by doing that to actually cause more damage and so when you assume the maintenance and development is going to be a year from now what if that means maintenance interval turns out to be eight months or nine months from now that's a bigger problem because it's all planned because you plan exactly when one turbine comes off when one turbine comes on and this plan structure is tied to the amount of money you're going to produce and the commitments you've made so you can't afford to break that you know structure in which you've made a commitment that this turbine will only be serviced a year from now now what you can do is what you can do is there are times you can run inefficiently run it so it's not as hot internally what happens then is that you actually save the life of the parts and then at the right time when the prices are high you over fire it you run it as hard as you want so you've saved life and then you burn up that life but you do it in such a way that you meet your commitments one year from now that is a digital twin inside it's understanding the remaining life on the part that understands how that remaining life is based upon how hot the turbine runs and gathers or banks that life theon allows you to use it at the right time that's dynamic optimization and we've found techniques with customers to save them millions of dollars because again when that price peaks you can make a significant amount of money and that's what they do they run it inefficiently you know and when the prices are normal and then they save it for that time when it peaks and now it peaks quite a lot because now that you have renewables in the mix there are many times in which people believe the renewables would give you something you know you have all this solar and then all of a sudden something like california occurs in which you have a fire it covers you don't get you get that coverage of some from smoke you don't have the delivery that you think you have coming from the solar fields and again you know the price of electricity jumps because it's based upon the demand and you can stay and make quite a bit of money so these are three good examples all focus on the fact that can i give you early morning can i predict when things will fail or whether you need parts and can i dynamically optimize to give you the most money and we've been running these for years we have seen for customers multiple billions of dollars worth of value being delivered let's jump to the next slide what's at the heart of a lot of this you know we all see and understand the data and the analytics that's involved but let me just draw you to three things one is this notion of ai enabled analytics because we've had a lot of models before right what's different is we're taking data from multiple sources that allow us to do something unique the other thing you know that we've got more insights on is this notion of life what's key for us is we can understand what's happening inside the turbine but what's the effect on the life of the part in many cases you'll find it's hard to believe but when um we do inspections we rely on people you know and there's a ranking so somebody would go in during a maintenance and development they will say well is this part damage is it not damaged you know because the digital twin predicted it would be damaged so then you'd look at the path and the human would say well it's damaged to a level eight there's a scale of one to ten that the human actually you know it registers damage on now how that itself you know in some cases doesn't give you the information you need because one inspector will say that's damage level eight one inspector's enhanced damage level six how do you tune your models when this is so subjective so now what we have is computer vision computer vision techniques that allow us to actually look at the parts use an algorithm to determine to detect a failure could it be is its pollution is it corrosion and then literally measure it so now i know you know a year ago it was pollution or corrosion damage it was this large right and i see that back in the model a year later it's that much larger we have exact actual numbers that allow us to get these models a lot better so a lot of that is the ai enabled analysis a real fingerprint of the source of truth by which you tune the twin second one is software defined controls whenever i have an insight on something i have an insight on you know what's happening with wind or solar energy and then i want to tune the systems there's a control system that sits inside and so what i'd like to do is combine that with the control system and have that control system run as quick as it can to make changes when needed you know to the pitch or the yo of the turbine itself so we spend a lot of time understanding these software defined controls and they're becoming more and more valuable now because you know you can use them in renewables but also whenever the wind or the solar you know fails you it's intermittent then you have to actually rapidly start your gas turbines again it's a control system that allows you to actually bring that turbine you know from a very low level to a much higher level so that it could actually generate the power that's needed again software-defined controls so we spent a lot of time understanding how those software defined controllers can work with our digital twins so you tie information and work together the third thing is the autonomous inspection repair systems like we mentioned before there's a lot of inspection repair work we get done now part of it is done when you bring it in but some of it is done when you go to a customer you go to a site and what you want to do when you go to the site when you actually have it been done you know by some of the owners of the equipment is if you automate it and you give them the tools to do it you collect much better data the data is uniform you know the data allows you to do the fleet analysis i mentioned before so we spent a lot of time on how do we actually do these things using robots especially designed robots and special cameras designed to capture that data how do you put it on the edge and then how do you use that uh to create better digital twins so these are all of the type of technologies that actually help us build the twins now what i'd like to do because i've spent the time before showing you you know the players that we've been running in the industry and i can show you a play for aviation i play for renewables i play for power all focused on the fact that i give you early warning i give you continuous prediction dynamic optimization all based upon these powerful digital technologies let me show you too that we're looking at you know that we've been experimenting with for the last two three years right some of it from funding from the department of energy some of it from darpa funding and uh to give you insights about where we think this is going so if i jump to the next slide i'll begin by talking about the first one which actually is called humble ai so let me give you a concept of it and then i'll then i'll walk through the slides here so we do things like digital twin because they deliver business value now after you've shown that business value and you begin to go to other customers the hardest thing is to get that adoption everywhere quickly why because there's a belief because you're dealing with humans and every time you deal with humans who have deep expertise the notion is well i've seen that before i can do that i mean this is just my job and so many cases the notion is you know that ai is a black box i don't know what it's doing um i don't know why i did it and so i don't think i want to hand control over to multiple million multiple in a few cases billions of dollars of assets to turn out to an algorithm that i don't fully understand so we built an approach called humble and what humble ai does is this essentially based upon the data that i have you know i build an algorithm right i can i build some of the ai techniques we have now the more data i have the better you know this algorithm is so that data gives me a zone of competency where in that the in that zone i know what i know the algorithm works well outside of that zone if i don't have enough samples or enough data my algorithm isn't as well so i know my zone of competency once i get out of that zone what i say is go back and use your deterministic algorithms you've always had and then allow me to learn from that moderator you're collecting so i build my zone of competency so the humble ai says i know what i know in my zonal competency outside we go back to what happens before and then i learn and i get better to increase my zone competency that works we've done this with enough clients now where clients feel a lot better the chief engineers say okay within this zone i've only i only have this risk to my business i could look at it i can see if i like it i can understand it and if you say it's going to be better than what i have now now i can prove it within that one zone and once you begin to do this and you test it for a while with them then it gets expanded more and more and so they believe the ai is working with them you know it makes more sense of that so here's a great example one we've done with wind i have a chart up there and what you see in the charts is wind conditions over 24 hours now wind is is is really unique as a fuel it's a free fuel but with that you know low-cost free version it it has certain dynamics so the wind varies by every hour in the day the wind varies by height so when you look at a wind turbine wind turbine could be as tall as 300 feet you know what you see here in the different bands if you could look at that small diagram we have here is that the wind speed varies and the height you know if you go at the lower levels the wind speed may be running at you know five meters per second at you know 300 feet it may be running at nine meters per second so it varies you know with time of day it varies with height it varies with distance between two wind turbines that are quarter mile half a mile apart it's very different it varies by season it varies by weather event it is very much varying so what we found in many of our cases we have zones like that blue band is a zone in which we collected a lot of data you know for whatever reason maybe you know um that's when we collected the data maybe the data systems we have uh in that part of the world that's where the usually the right operating profile is it operates in that point of view between for instance i think here it's between was it five and six and seven meters per second so once you have that you build an algorithm that works well there so if the wind is at anywhere between six and seven meters per second the wind speed i run my algorithm if you go to the next shot when the wind gets above that say the wind is running at nine or four millimeters per second in terms of the wind speed i go back to my deterministic algorithm right i say you know what i'm humble i know i'm not good in this individual go back to deterministic and then what i do is i learn i bring data in and i learn more and more i may take me multiple events to learn to get enough data that i can learn and it may also be that i do model management i need certain confidence thresholds before i think the learning is there so i know what my zone is and outside of my zone i use you know the thermistic sessions and then if you go to the next chart i expand that knowledge so in this next chart what you see here is now i have more information between five and eight meters per second wind speed and now the twin itself because i've run the algorithm the algorithm works within the i've adapted that model now i can run it and so we've been running these trials about two years we we we've done them in several places the last was across 69 wind turbines now we've built that into part of a product and that product is being you know sold right now again we've applied that to renewables we also applied that to a good friend in power we there's also parts of this that we're bringing onto the grid i'll give you one last example here before i stop and take questions if you go to the next chart and let me take a second and set this up it's something called digital ghost so the notion is this our infrastructure you know we have seen several over the last nine years several um cyber attacks coming at infrastructure in the us and around the world and it's increasing in frequency so in many cases since we designed the asset you know and we manufacture the asset we understand the physics of the asset inside a gas turbine is a break-in cycle that cycle talks about you know when you have heat generated by fuel that you ignite and we know the fuel we know the um the materials we use that actually allows us to understand that explosion should register this temperature on these sensors so i have three sensors i know the first one is going to be a hundred i'm making up a number the second one is going to be 120 degrees the third one should be 140. if that middle one instead of saying 120c is 400 degrees i can look at that and i can say something's wrong either the sense is damaged or the sense has been hacked right because i know in the in that cycle of physics that shouldn't matter you know because it should be 120. i can then because i own the control system i can then do a slight perturbation and see what it turns out to be if that's like probation it's still wrong now i can begin to understand a few things more if it's a damage sensor or more if it's an actual hacking attack so we call this digital ghost we have a digital twin that's running so i have a model that tells me what each sensor should see you know given the output i'm seeing given the different temperatures given the different pressures and that model has to jive with what the actual sensors are saying because i know the physics i created the model based upon the physics and the ai if it does not i know something has happened and then i have other techniques i can use to determine if it's a failed sensor or if it's a hacking itself right so that is first detecting a problem then i can isolate it by saying it's this specific sensor and then at the right time i can say you know i know that sensor has been hacked rather than use that sensor i'm going to use a sensor that's on the digital twin so that's that value that's on the digital twin should be accurate it might be less so so there may be slight degradation and performance but that allows me to continually run my asset in slightly degraded mode and isolate the attack so if you go to the next chart and we hit the play button i want you to watch this video now that you have that background and so the notion here is that's a digital twin that we've created whenever this is an action where something overheats we're looking at the digital twin and we're saying is that action being caused by a sensor a problem or is it because someone is trying to hack the system and so i can use that twin to go in and determine if there has been a hack and if there has been a hack what i do is i just take the digital twin and i use that model as the input to my control system instead and again i can continue running my turbine to make sure i can still produce electricity while isolating the problems that i have and this is deeply invaluable you know to many systems out there whether it be something that you see as a result of a problem that's occurring because of affiliate sensors or failure of some control pieces or deliberate hacking now we've been working with the department of energy on this for several years in fact we then took it to our greenville test site we have a huge gas turbine that we run there we tested it there and as a result of it it does work so we had a number of the government officials come in you know together with with some of our friends in the military tested against certain capabilities we see it works so these are the kind of new capabilities that we're building that we're bringing to bear so let me go to the last chart here um so let me let me summarize with just a few key points so so i think there's some things you know that i would put to this um august body to consider you know as we watch this unfold you have to really be aware of all these forces because while while there's an altruistic bin at some point we've got to deal with the financial aspects of it and tie both of those together that allows this to happen so these market forces are very powerful these digital forces that are happening that are constantly lowering the amount that you pay and improving the performance of things in the consumer space can we bring these across can we bring these to our space and use those as great sources of information and capability to actually figure out insights as well as drive workflows and then the last thing is can you understand what other technologies we should focus on i love the discussion we had before about blockchain blockchain is great blockchain allows me to get traceability so i can understand which individual which parts are actually in what configurations because again i need to know the exact parts in each asset so that i can actually use the right digital twin because if i have a digital twin assuming this is a nine f o three configuration versus a nine f o four configuration the parts are different than the o4 configuration the parts have different tolerances i'm using the wrong model so i need that traceability in things like blockchain to ensure i actually know what configuration what's in the asset when i run the models so there are many new technologies new techniques coming in 5g is another great one can i get more data more data can i get lower latency i can do more things with more data in terms of better insights with low latency i can activate my control system even faster so i put these things together and then what i need to do then is understand the business transformation process you know how do i show value how do i increase adoption how do i remove the cyber risk because the other thing you deal with is every time you talk to customers and they say well if i put in all this digital technology wouldn't that make me even bigger target you know for cyber attacks and and again risk the reputation you know risk first of all you know my delivery to customers and risk my reputation my reputation but in the end we all know that this is what the industry is going to be digital aspects of this industry are going to profoundly transform it in with the next energy transformation so i think you're in a beautiful place and it's my delight to spend the time with you and show you what little we have done so far with that i will turn it back to dr murray and open it for questions i think we have a few more minutes of questions i hope so please thank you so much colin that was such a fascinating presentation i'm afraid of all the questions i wrote down it would take us another 45 minutes just to kind of go through them and so i'm going to defer to you to the audience here who put a number of great questions here in the in the q a and so one one i'm just going to start with sort of a general it's almost like career advice we have a lot of early career data scientists in the audience right now and um so you know question to you um you know you're a few years into your career you've seen a lot happen um after two decades at ibm and i think roughly a decade so what advice would you give to soon-to-be graduates or someone early in their career as they're preparing to start a career applying data science to real-world energy applications um i'll give two pieces of advice um so one you know and all of the advice comes because of my failures you know i have failed royally that allows me to actually give good advice first thing is uh you're skilled in what you're skill in in terms of data science spend time learning about the processes spend time learning about the environment in which you're going to apply that in right so when we talk about the energy transition there's there's several things going on so for instance one is climate okay we want to create new renewable capabilities but the second thing is defense so for instance right now we know that um there are multiple different actors out there these actors are all saying we want to wage total war they'll attack your infrastructure system the financial systems so there's a significant amount of money being spent by the us government to protect these systems can you understand the way the us government thinks can you understand the economics can you understand what motivates them spend some time getting that then if you get if you understand what's happening in the business world you understand what's happening in the government world or the policy world it sharpens your delivery of your technology because the game always is taking a business or a government problem and convert it into the data problem a data problems or data solution and then you've got to get that data solution back to a world solution so you have to understand the context by which you're solving the problem to be most effective so spend the time on that that's the first thing i would say uh second thing i would say is um for me two things become important well three things the first thing is awareness the more awareness you get about what's going on around you the better off you are so attend these conferences spend time at duke with the energy initiative make friends because that gives you a great awareness after that think about resiliency you are going to feel along the way in what you try to do take that blow and get back up and the third thing is learning right i mean you have to learn continuously i made the mistake of believing when i finished my phd at berkeley that will you know that good enough i mean i think i've done it i've written the stuff i've written that was wrong every two years i had to continuously learn and now i realize the learning the reader growth is happening so fast like i look at my phd at berkeley and you you look at at my bibliography and you see things that are two three years five years old and he said that was okay now when you look at some of the phds out you know stuff in the big lot of is at most two years old so you could only learn so much so you've got a lot not only learn yourself but you've got to have friends who are learning other things and reach out to them so your learning is not you your learning is your network so think about awareness get the world in a broad context so that you can see many things that will help you understand where to go think about resiliency you're going to feel along the way it doesn't say anything about you it's nature of the world and think about learning you will have to learn and you will have to get into networks in which your friends are learning and use them as well because you can't learn at all so that that's the advice i'd give brian that's that's that's great too ask the right questions be a lifetime learner and be resilient and i think you know i i don't know if there's going to be a key word for 2020 um but resilience i think is going to be at the top of the list for me um so a couple of other questions uh with ai enabled analytics you talked about computer vision to make inspections more objective uh given this idea does this sort of system have to be rigid um uh and i guess by that i mean you know it's like you're talking about like humble ai or something like that where there's a little bit more interaction with it with the humans i don't know if that's with the question yeah well yeah let me address that part of it though that is quite right you know a lot of the work we do in computer vision we took the great thing about ge is g as a ge healthcare system so we took a lot of what we learned from how radiologists view things so what radiologists do is what we do is our systems indicate the areas the radiologists should look at i'm not making the decision for the radiologists i indicate the areas the radiologist then looks at that decision looks at those areas and then they focus on those areas if they find something else that i didn't cover they make a circle themselves and so the ai system learns i didn't know that and we captured our information as well so we do the same thing right now in in analysis for aircraft systems we initially have the ai system figure out where the biggest problems and it puts circles around those right and the human will take off the sickles and then the human will draw a blue box around things that missed and so the ei system gets to learn along the way then by and by we call that assisted and then augmented is when we get to a point in which the human says you know it's discovering most of these things don't show me all just show me the areas where you're unsure so now what the system does it it says i know enough about capturing the basic things i'm just going to show you only the high level things the things i'm not sure about so now you actually only use the inspector where the system is not sure so it always is going to be a combination of human ai for one simple thing i need to detect a problem as early as possible i may not have enough data to actually train an ai system so if the failure now is now occurring i need the humans to figure that one out first and so we always bring them together there's no system that we have in which it's just the ei by itself and the human uh without human supervision it works together given the dynamic nature change we see yeah i mean one of the questions i had coming in was just thinking about what are some of the lessons that you've learned from applications in other sectors to energy and as i heard you talk about you know digital twins and digital ghosts and humble ais like they've got to have come from health they've got to have come from aviation and manufacturing exactly because you like literally you're trying to avoid building the plane while you're flying and and those are the types of problems that actually apply in energy and across other systems as well exactly we have a question up here that is self-serving for a lot of people in the audience which is a question about this does ge have data science research partnerships with universities and what kind of research do you do in partnership with the university versus in-house yeah we do we do we we do i mean and again we have many i deal with a few i spent time with the university of maryland baltimore county i spend time um as well with some of the work that we've done in stanford um they usually fall into two categories uh ge has something called the global research center grc and this is where we spend a lot of time working with universities on cutting edge pieces and we also have our gene businesses to work with the universities the best way i find to do that um usually what we do is we get together we spend time we share views and then we try to combine our ideas and go after darpa department of energy i opera e proposal and that pulls us together because what it does it allows us to find ways to work together in a construct that makes sense and usually when you do these right the iep gets taken care of you understand how how it works it's a multiple year contract and that allows us to actually build a research moving forward then as that progresses i have the ability to go from the work i'm doing in the global research center and tie it into one of the ge businesses because the real power is when you can take that research and put it in a product get the product out there and we can turn around sell that product right into the the us military in terms of jet engines we sell you know gas turbines across the world so we have a way of teaming working together on you know especially with government products projects and then from there go into products and i think that flow makes a ton of sense no not everything gets into a product because again not everything is at the right time but we do do these this piece of research and i think the best way is to do things tied to some of the government initiatives we have spectrum energy okay so this is going to be the last question if i get actually even though it's a big question i just ask you if we could just maybe have like a one minute answer and then we're already a little bit over and i think we have to move on but but it has to relate to the kind of the crisis of our time so how is the ai system and tech adapting to shocks that don't have a historical reference or data to learn from like the kobe crisis ah boy yeah that's a big question in a one-line sentence what would i tell you i i don't know how to understand it in a one-line sentence we have seen that affect our supply chains because all history changed when this occurred what we have found though is that they are ai uh for instance the recovery incovid in our airplane business we had models that no longer valid we've been using we've been looking at things that are happening in china china is recovering first from the pandemic and then we can use the ai techniques there to show us how to actually augment the current models in order to get them better so they are ai techniques that can be used right with with sparse data we have deliberately built some like that with sparse data it could be used and then what we do is we have feedback loops because again i don't depend on a model with sparse data but i have a rapid feedback loop to tell me what was wrong what was right that i updated so there's certain types of learning that we use to do that but but we have ways of doing it yes that oh that that's great and thank you that was a terrific one-minute answer to a uh two-day question well it might be like to be here brian um it's my life and honor thank you for having me uh and then hopefully that added to your to your seminar oh very much so and and the honor was ours so thank you so much in whatever way you can do and wherever you are please join me in the in an applause for uh dr collin paris for really stimulating thank you thank you have a great conference
2021-03-03