Digital twins (history, examples, models)BIGDATA
Exploring the Future of Digital Twins:https://sirius-labs.no/wp-content/uploads/2020/09/UC18EU-sirius-kharlamov-evgeny-exploring-the-future-of-digital-twins.pdf style="color: windowtext
Del denne Casen
En LØRN CASE er en kort og praktisk, lett og morsom, innovasjonshistorie. Den er fortalt på 30 minutter, er samtalebasert, og virker like bra som podkast, video eller tekst. Lytt og lær der det passer deg best! Vi dekker 15 tematiske områder om teknologi, innovasjon og ledelse, og 10 perspektiver som gründer, forsker etc. På denne siden kan du lytte, se eller lese gratis, men vi anbefaler deg å registrere deg, slik at vi kan lage personaliserte læringsstier for nettopp deg.
Vi vil gjerne hjelpe deg komme i gang og fortsette å drive med livslang læring.
En LØRN CASE er en kort og praktisk, lett og morsom, innovasjonshistorie. Den er fortalt på 30 minutter, er samtalebasert, og virker like bra som podkast, video eller tekst. Lytt og lær der det passer deg best! Vi dekker 15 tematiske områder om teknologi, innovasjon og ledelse, og 10 perspektiver som gründer, forsker etc. På denne siden kan du lytte, se eller lese gratis, men vi anbefaler deg å registrere deg, slik at vi kan lage personaliserte læringsstier for nettopp deg. Vi vil gjerne hjelpe deg komme i gang og fortsette å drive med livslang læring.
Velkommen til Lørn.Tech. En læringsdugnad om teknologi og samfunn med Silvija Seres og venner.
Silvija Seres: Hello and welcome to Lørn podcast series with the University of Oslo on the topics of going from data to insight. This series is going to contain eight mini lectures in digital format and is supposed to work as a warmup to the same course by the University of Oslo, being piloted in December. The University of Oslo course will be to study credit giving and can also be used as a part of a greater degree. The topic I have today is digital twins and my guest is David Cameron. Welcome David.
David Cameron: You're welcome, thank you, Silvija.
Silvija: We're going to talk for about 20 minutes about what a digital twin is. What are the necessary things to think of when you're creating one? Also, how can digital twins be used in science? David, we usually start by asking the guests to introduce themselves very briefly. So, would you please tell us who you are and why do you care about digital twins?
David: Ok, my name is David Cameron and I have been working for the last six or seven years at University of Oslo, and I'm Australian who managed to find himself in Norway about 28 years ago. I have been working in the process industries since then with programmes which were basically digital twins, but no one thought they were before the thought of digital twins came up a couple of years ago. So, my whole career has been in digital twins without knowing it.
Silvija: And you're staying in Norway because of the nice weather or..?
David: No, I've got myself a wife and two kids on the way.
Silvija: Then you're kind of stuck similarly with me, except I believe that it's also probably the best place in the world to be, if you're going to be a lady and try to combine work and life.
David: Yes, I could observe that.
Silvija: So, David, what is this thing called a digital twin?
David: Well, it's an interesting concept, and the Gartner group, which is a group of very prominent IT consultancies who tell CEOs around the world what they should know, has put it into the very top of what they call the hype curve in 2018 in 2019.
I think the idea of this is that we keep trying to collect data about things and if you collect enough data, you see that you can create some sort of digital things. It's also being managed quite a lot by the gaming community and science fiction where you have films like «Matrix» which portray an universe of something that represents you. Also I think last week people at Facebook launched something which is very digital twin equal. So we've got a situation where people make them, not out of being able to make computer replicas of other things for certain purposes and what we've seen in the industry is that these sort of things are being used for useful things for many years. Recent changes in technology makes it more feasible to do more ambitious digital twinning applications.
Silvija: So that's what I wanted to ask you about. Your phrase computer replicas is just another set of synonyms for digital twins. Many people would say, what's new is that we've been doing digital models of things for a long time and yet there is something that seems to be different more recently, in terms of scale.
David: It's certainly something about scale. When you come back to applications is often the scale is very shallow and that you're able to do something that's very, very game like, a visualisation of something, but for visual tools to have value they've got to be able to enable some sort of valuable decision. At the moment there's a lot of yes, we can get lots and lots of data, and we can possibly do simulations. Or often there is a situation when companies have been doing something related to digital twins like simulation, like visualisation, like acquisition of data. Take what they've been doing and say this is all you need for a digital twin. What is very important now is not to go for the very, very ambitious, marketable digital twin, which is like the European Union's digital twin of the planet sort of thing. Because what we really need to do is to build on the facilities we have now, the computational resources we have now, to create computer applications that enable us to make good decisions about things that we encounter in our life about our experiments, about the industrial processes we need to control.
Silvija: You are talking about a three legged stool model related to digital twins. What is that?
David: When people talk about digital twins a lot of the conversation produces some really complicated architecture showing five layers of instrumentation, 140 different applications and seven layers of users. A simple model of digital twins, which is also something that is closely related to what the US military defines visual twins as being, it's about a model where we need three things to describe a certain object of its behaviour.
David: The first thing you need is a description of how the objects are built up in an industry, that is called asset data. If you were a person, it would be your genotype phenotype. Then you need some sort of simulation of what is the behaviour of the thing that you're trying to twin. Now you can either use physical models or you can use machine learning, statistical models. Or the most fruitful way which we believe is to try to combine those two into some sort of hybrid modelling, where you use machine learning to account for things that you're uncertain about and you use the physics to constrain the machine learning. And then the third thing we need is measurement about the observations the thing is doing, because we know when our models are wrong and we know descriptions of how things are out of date and inaccurate. So you need to create this interplay between measurements, simulations, and the structural information about the theme. To so provide the top of the stool, the chair which is the decision we want to support the process with using digital twins and presenting it to the user in a way that makes it useful for the user.
Silvija: Let me try to translate what you told us into sort of a Lego language. I'm going to try to use the human body as the thing that we're trying to make a digital twin of. I believe that this are becoming probably the most hyped topic, but also for good reason. I think all medicine is going to be using them quite heavily going forward, you've talked about four things in this tool. The three legs and the top. The three legs were the asset data, so that would typically be my genotype, for example.
David: Yeah, but it will also be how you connected together, you know the anatomical knowledge about what makes a human.
Silvija: So, measurements of my skeleton, chemical measurements?
David: At the line of the fact that we know the basics of the structure of Silvija.
Silvija: Then the behavioural data would be describing the blood system, the nervous system and it would be modelling and describing some physical, medical or biology chemical rules for how the system works.
David: Yep, exactly. Modelling with potential, modelling about flow, modelling the fact that you've got a pump going backwards and forwards creating more pressure waves that keep the blood flowing around and moving the oxygen tank up and modelling metabolism. Basically, encapsulating knowledge, you have to be able to predict where you are now and predict what can happen in the future.
Silvija: That entails we have enough medical understanding of how the human body works so we can do this. And then the third leg.
David: And we got models.
Silvija: Right, and then the third would be sensor data. Let's say I build a little chip in my tooth or in my lens and it is measuring my blood pressure, my hormonal levels, my metabolism, my blood sugar etc.
David: Yes, exactly. So that's sort of things you would need to do, because we know the models need to be constrained by the measurements and so you get this neutral checking and building up between the measurements and models. So the models are used to fill in where you don’t have measurement to give you predictive power and the measurements are used to make sure that the models keep tracking the actual real you.
Silvija: The really interesting stuff is when things go off the model, when something unusual starts happening. Is that the top of this tool? How do you model your decisions into this model of Silvija?
David: If we move from a Silvija to an oil platform where the things you are already interested in is not the boring reality of me sitting here producing morally away, it is always the situation where something is wherever I want to do something. What are the implications of that? I'm moving metaphor to an exactly successful practical example from the North Sea, which is about now 20 years old, which was the question that we're going to produce oil, gas and water in huge quantities offshore. Do we want to separate out the water and pump it back down into the earth, but then we've got this mixture of gas and oil. Do we build two pipelines? One for gas, one for oil, or do we put this mixture of gas and oil into a single pipeline, and that single pipeline is 40 inches in diameter and 100 kilometres long and send it to land? The problem with that is if you've got 100 kilometres of pipeline and you're putting in a multiphase mixture, two phases or mixture of gas and oil, you've got the danger that you're going to get really big accumulations of oil in that pipeline, which will build up until you've got enough back pressure to push it through, and then you've got this huge slug of oil coming at your poor reception facility. If you go to one of these reception facilities, you'll see that there's a huge structure called a slug catcher to take account of that.
David: The other challenge you have is that you've got very few measurements on that pipeline, you've got a measurement of what's going in, and you've got a measurement of what's going out and then you've got a couple of 100 kilometres of pipeline between those two measurements. So, to be able to build those sorts of pipelines you're really absolutely dependent on the ability to predict what is actually happening in that pipeline and use it in real time to: A detect when things are going to go wrong when you're building up in the pipeline and B as an operation say if I want to increase production a bit, what are the implications of that? How quickly can I do it? If I do it too quickly, about two days later I'm going to have real problems with large amounts of oil turning up in the facility. And so, the digital twin was designed to support exactly those decisions and that technology is the technology which, more than any other technology enabled most of the deep-sea oil field developments of the last 20 years. And it's fundamentally a digital twin where you took a simulation, again developed in Norway project Cordova, connected it to measurements and the description of the pipeline to build a digital twin which supported two decisions. Am I going to have problems with my pipeline operation, and the other decision is what I'm going to do with the operation safe and correct either how this is going to behave over the next three days?
Silvija: I think what you just described is extremely useful for us to have these pictures. I'm thinking of any complex and large-scale system, and it may be a human body like we started, or it may be an oil platform with hundreds of kilometres of pipe subsea attached to them. We need a very advanced computational model. We use it to then make sure that the operation is as efficient and as safe as possible, we need to decide on what kind of decisions need to be made on the top of this tool in order to basically decide how we're going to use the model. I guess for a human body it could easily be something along the lines of managing your health, that companies like Apple and Amazon are beginning to sell to people. So you are not just waiting for symptoms to arrive, but perhaps you can be a little bit more predictive and even more. What should I say? Eat right, exercise right, manage your health for maximum effect.
David: Yes, build up social credit…
Silvija: Not necessarily, just social health, I guess. Tell us a little bit about the history. Is this a long story?
David: It's a long story, if you look back to as soon as people started building mathematical models, they wanted to start matching the whole test in the mathematical models that don't describe your observed reality. You could always go back to the Greek astronomers, Ptolemy with his model with these epicycles and circles going round matching the behaviour of the heavens and the model was very inelegant and so eventually they got it sorted out when Kepler managed to provide these ellipses for the models, but as soon as computer models came along in the 1960s, people were starting to ask questions about what can I do by connecting those models to the actual thing. I'm trying to model or observe, and so the first automatic computer control came along in the 1960s. In the 1980s, people were getting to the point where their models were getting very good, the process industry had developed a good set of models that were robust and useful enough to be connected to actual plants and so what was called model-based control in effect was a very simple digital twin which was used to revolutionise how you controlled oil refineries and chemical plants. At about the same time the air nautical mechanical engineering community, especially NASA, claimed the term digital twin. A man called Michael Greaves is generally credited for us having published a paper about that and I was interested in the behaviour of a large, complex process system, they were interested in building a replica of their product. The hitter spacecraft drawing F35 or a car, which enabled them to simulate the behaviour of that before it was built and then also use it as a tool to follow up the operation of that particular product or piece of equipment. You still have a very slightly different view between the process industries and medicine and the manufacturing industries where the digital twin is very tightly linked to something they're going to make to turn into some sort of product.
Silvija: I'm just fascinated because I'm wondering if Norway might have an extra good reason to be interested in these industrial applications of both digital twins, but also of artificial intelligence related to them.
David: Yeah, I think Norway was being the leader in the industrial applications and digital twins from the oil and gas industry and also potentially through the predictive medicine activities in the country. I have the impression by going to these meetings in the European Union, that what we've done in the Norwegian oil and gas sector is something that is substantially more advanced than, for example, steal or pulp and paper or chemicals. That knowledge and confidence is something we should be able to trade on once we're getting into this transformation of heavy engineering from oil and gas production into large scale production of renewable energy and hopefully still process industry in the future.
Silvija: I know you're against science fiction and the too grandiose visions, but could Norway somehow translate what it knows from oil and gas to other resources from the sea and management of sea?
David: It was definitely the ambition to in fact, if you're talking about scientific applications or digital twins, this is going to be a big topic in the new design centric University of Oslo. Because what we see is that meteorologists have been building digital twins for the last 100 years as well. They have challenged me as a gas person as it's very hard to build models, because you have to build a model of the whole world based on the fact that you only send up 50-60 balloons at various points. So, you have that mathematical problem, how do I make my model fit an incredibly small amount of measurements? There is a great deal to be gained by bringing together all the observations were going to be making with marine environment, to give a lot of models into a framework where it's easier for researchers to access the data and it's easier for people to build applications which should enable you to do new smarter things around the marine environment. We hope that we are able to create a digital twin platform which will enable biologists and meteorologists and climate scientists to collaborate around a prorated set of data and validated models and an agreed description of what the marine environment looks like, to be able to support better monitoring of the Oslofjord, but potentially repairing the Oslofjord trying to get the call back, for example.
Silvija: You had a wonderful set of examples now for making us realise that digital twins are probably the best tool we have for making computer aided decisions in industrial and social settings. What's the connection to science?
David: Digital twin is a model for trying to combine modelling, reasoning, structural information in a way that links it to concrete problems and also delivers it as a concrete piece of software and hardware. Now what we see is that in most experimental science what you have is a situation where you have an experiment with observations with a process where those observations are then dealt with a theory, that theory is a model and then that information is then stored in a way that we can hopefully replicate it. Often your experimental or clinical researcher is building some sort of digital twin framework on their own PC anyway, you've got observations, you've got a model and you're trying to make some decision. Now what we can see is that in a lot of areas you can actually increase the speed or access to data, you can increase the accuracy of data, you can provide support to interdisciplinary research by investing some money into platforms which enable you to get easy access to the observational data that you need, share that observation or data researchers so that you can actually let them go in and try to model your observational data, and create ecosystems around specific problems, like the one I was talking about observation of the Oslofjord where you again bringing in lots of different disciplines when trying to consolidate without into something that is able to, it reach by the fact that you bring in lots of models and providing a platform for people to make different sort of analysis and decisions based on that data.
Silvija: Very interesting. For the people coming to your course, have we kind of summarised the most important points here, or do we need to add something more?
David: Not really, what we are going to do is that we will be presenting a concept of what the digital twin is and letting you discuss what your digital twin is. We'll also be dealing with the background theory. We will be dealing with the IT nuts and bolts of how you screw the digital twin together and we will also be dealing with some of the intellectual challenges which are necessary rather than digital twins, which are related to how you make sure that your models will work for digital twins. How do you make sure models are robust? How do you deal with uncertainty in data? How do you deal with the problem that nobody ever has the same naming and understanding of objects. So you actually have to build an understanding or common way of talking about things to link that data together and get it together with different databases.
David: Then on the third day we will go through and look at some practical examples around digital twin designs, digital twins in medicine, society and twins industry. Then the assignment you get to do, is to go away and write a little feasibility study for your own favourite digital twin.
Silvija: Sounds very interesting and approachable, so thank you so much David for making the topic of digital twins, which we all have been hearing about a little bit closer and very, very inspiring.
David: Thank you for your time and conversation Silvija.
Du har nå lyttet til en podkast fra Lørn.Tech. En læringsdugnad om teknologi og samfunn. Nå kan du også få et lærings serfikat for å lytte til denne podcasten på vårt online universitet Lørn.universitet.
Du må være Medlem for å dokumentere din læring med å ta quiz
Allerede Medlem? Logg inn her:
Du må være Medlem for å kunne skrive svar på refleksjonsspørsmål
Allerede Medlem? Logg inn her: