Alright, well, let's start with an easy question. How many of you are wearing a Fitbit or an Apple Watch or some other kind of health tracking device? And how many of you have got a smartphone with you here today? Maybe I should say how many of you have not?
The fact that so many of us have these technological marvels in our pockets or on our body is a sure sign of the revolution that's taking place in computing over the last decade. And I want you to think with me for a second about the elements of that revolution.
So first off, are the data. These devices are collecting data about our health, our movements, our habits and more. And what's really important is that those data are not generic population data, but they're data that are personalized to us, each as an individual.
Second, and just as important, are the models. Inside these devices are very powerful mathematical and statistical models. Some of these models are learned entirely from data, perhaps a machine-learning model that has learned to classify whether I'm running or walking or biking or sleeping. Some of these models are based in physics, such as a physiological model that describes the equations that represent cardiac function or circadian rhythm.
And now where things get really interesting is when we start to put the data and the models together. Mathematically, this is known as data assimilation. So we have data and we have models. With data assimilation, we start updating the models as new data are collected from the system. And we don't do this update just once, but we do it continually. So as the system changes, as I get older and my circadian rhythm or as my cardiac function is not what it once was, the new data is collected and the models are evolving and following along with me.
Now that data assimilation is really important because it's what personalizes the models to me and that then gets us to the fourth element, which is the element of prediction. Now that I have these personalized models, it's so powerful because I can now get predictions or recommendations that are tailored to me as an individual and that are tailored to my dynamically evolving state over my life.
So ... what I’m describing, this working together of data and models, is likely very familiar to all of you because it's been driving your personal choices in retail and entertainment and wellness for many years. But what you might not know is that a similar revolution has been taking place in engineering systems.
And in engineering systems, the story is much the same. We have data and we have increasing amounts of data as sensors have become smaller, lighter, cheaper and more powerful. In engineering, we also have models. Our models are usually grounded in physics. These models represent the governing laws of nature. They're powerful models that let us predict how an engineering system will respond.
What you see up here on the slide is a picture of the unmanned aircraft that I have in my research group that we use for a great deal of our research. And for this aircraft, we have powerful finite element models that let us predict how the aircraft structure will respond under different conditions. So these models let us answer questions like, will the structure of the aircraft hold together on takeoff if I design it in this way? Or, what happens if the aircraft wing gets damaged and I continue to fly it aggressively? Will the aircraft hold together?
And again, just like the Fitbit and the smartphone example, we can put the data and the models together to build a personalized model of the engineering system, a personalized model of the aircraft. And we call this personalized model a digital twin.
So what is a digital twin? It is a personalized, dynamically evolving model of a physical system. And I want you to think about the digital twin of my aircraft. So as I create that digital twin, I'm going to be collecting data from the sensors on board the aircraft. I'm going to be collecting data from inspections I might make of the aircraft, and I'm going to be assimilating that data into the models. And what's really important is that I’m not building a generic model of just any old Telemaster aircraft. I am building a personalized model of the very aircraft that is right now sitting in my garage down the road in South Austin.
And so that digital twin will capture the differences, the variability from my aircraft to say, my neighbor's aircraft. And what's more, that digital twin will not be static. It's going to change as my aircraft ages and degrades and gets damaged and gets repaired. We will be assimilating data all the time and the digital twin will follow the aircraft through its life.
So this is incredibly powerful. I want you to imagine now that you're an airline or maybe in a few years’ time, you’re an operator of a fleet of unmanned cargo delivery drones, and imagine that you would have a digital twin like this for every vehicle in your fleet. And think about what that would mean for your decision making. You could make decisions about when to maintain any one aircraft, depending on the particular evolving state of that aircraft. You could make decisions about how to optimally fly an aircraft on any given day, given the health of the aircraft, given the mission needs, given the environmental conditions. It would really let you optimally manage that fleet of aircraft.
So this idea of a digital twin is pretty neat. The term “digital twin” was coined in 2010 in a NASA report. But the idea, this idea of a personalized model combining models and data, is much older. And many people point to the Apollo program as being one of the places where digital twins were first put into practice.
So in the Apollo program, back in the '60s and the '70s, NASA would launch Apollo spacecraft up into space, and they would also deploy a simulator, a virtual model on the ground in Houston, to follow along on the mission. And now this became very important and it became very useful in the Apollo 13 mission. And again, perhaps you all know the story because we've seen the movie. In the Apollo 13 mission, the spacecraft suffered a malfunction. It was very badly damaged. It became stranded up in space. And so the story goes that NASA were able to take the data from the real aircraft, the physical twin stuck up in space, feed it into the simulator and to the virtual models on the ground in Houston, do the data assimilation, dynamically evolve the simulator so now that it represented the conditions of the damaged spacecraft and then use that simulator to run predictions and ultimately guide the decisions that brought the astronauts back home safely.
So more than 50 years later, this idea now has a really great name, the name of digital twins. And what's really exciting is that it's moving well beyond just aerospace engineering. So in our engineered world, we're starting to see digital twins of bridges and other civil infrastructure for structural health monitoring and predictive maintenance. We're starting to see digital twins of buildings for energy efficiency, digital twins of wind farms to increase efficiency and to reduce downtime. In the natural world, there's a lot of interest in creating digital twins of forests, farms, ice sheets, coastal regions, oil reservoirs and even talk of trying to create a digital twin of planet Earth. And in the medical world, there's a great deal of interest in creating digital twins to help guide medical assessment, diagnosis, personalized treatment and in silico drug testing. So, many, many exciting potential applications of digital twins.
But now, I would not like you to leave my talk today thinking that all of this is a reality, that we can create digital twins today of all those complex systems. It's still beyond reach to create a digital twin of an entire aircraft. It's still beyond reach to create a digital twin of a cancer patient or of planet Earth. Creating digital twins of these very, very complex systems is very, very challenging. And let's think for a minute why it's so challenging.
So one reason it's very difficult is because of the scales that these systems cross. If you think about my aircraft, damage at the microscopic level on the material on the wing of the aircraft translates across scale to impact the way the vehicle flies at the vehicle level. In medicine, we all know that, again, changes at the very fine level, at the molecular or the cellular level in our bodies translate across scales to have impacts on us at the system level, at the human level. And computational models that resolve all of these scales, from the microscale all the way up to the system level, are computationally intractable. We can't solve them even with today's supercomputing power.
But then you might say, "OK, well what about the data? You said we had a lot of data. Can we not just learn digital twins from data?" So yes, we live in an era of big data and we have a lot of data often for our systems. But when it comes to these very challenging, complex systems in engineering, in science and in medicine, the data by themselves are almost never enough. The data are almost always very sparse in both space and in time. The data are almost always noisy and they're indirect. As an engineer, I can almost never measure what it is I want to know. If I want to know about the health on the structure inside my aircraft wing, I can't just break it open and take a look. I am limited to those few sensors that are on the surface of the wing, taking those measurements and then trying to guess. More than guess, trying to infer what's happening inside the wing. The same is true in medicine. A medical practitioner can't open somebody up to take a look at an organ. Again, we are limited to sparse, noisy and indirect observations taken from the outside to try to infer what's going on.
So then you might say, "Well, we just have to wait a few years because sensing technology will get better and better and better." And that's true. Maybe, maybe then we'll have enough data to really be able to characterize what is going on inside these very complex systems. But even that's not enough, because all that would tell us is what's happening now. And remember, we have to do more than that. We have to be able to predict what might happen in the future if we take different actions. So we're always going to need the models.
So this sounds like a huge challenge, and indeed it is. But the good news is that we have a lot of hope for addressing this challenge. And a big part of this hope rests on this notion of predictive physics-based models. These are the models that encode the governing laws of nature that let us make predictions -- predict how a cancer tumor might grow or how a cancer tumor might respond to radiotherapy treatment, or predict how an Antarctic ice sheet might flow under different future temperature scenarios. And bringing these predictive physics-based models together with powerful machine learning, with scalable methods and data simulation and optimization and decision making, and with high performance computing, that's the realm of the interdisciplinary field of computational science, and that's the focus of the Oden Institute here at UT Austin, where we bring together faculty from 24 different departments across campus to tackle these kinds of challenging problems.
So I’m going to close by provoking your imagination and I hope you’re excited, like I am, about the idea of a digital twin. And maybe as you go home, you can look around and think, "Oh, what if we had a digital twin of that?" But let's look at some examples of some of the really exciting areas where digital twins could make a difference in tackling some of the biggest problems facing society. And as I go through this, you'll also see some of the really exciting research that we have going on here at UT Austin.
So the first area is space systems. You probably all know, we are at the dawn of a new space era. It is so exciting and it's so exciting for our students. And what's even more exciting is that central Texas is right in the midst of that new era. So digital twins clearly have a role to play in managing the health and the operations of space systems, of launch vehicles, of satellites. You can see here, this is some of the work that I'm doing together with my colleagues from the Cockrell School, Renato Zanetti and Srinivas Bettadpur. Digital twins also have a big role to play in tracking and managing space objects and space debris. And here at UT Austin, we have one of the world's leading experts in this area, that's Moriba Jah. Moriba is building digital twins for space domain awareness.
If we think about the environment and geosciences, again, digital twins could play such a role here. This picture you see, Omar Ghattas's, high-resolution, physics-based model of the Antarctic ice sheet, which is put together with observational data of all different kinds to understand what might be going on, to help guide decisions about where to drill ice cores, where to take observations, and ultimately to inform the decision-making around our future climate. We see also here the work of Clint Dawson in building a digital twin of a coastal area, here, the Gulf Coast, again combining powerful physics models with all the different kinds of data and here, focused on making storm surge modeling for hurricanes even more accurate, again, in support of critical decision-making.
And then in medicine, I think it's pretty clear that digital twins have such a role to play in realizing the promise of personalized medicine. Here we see some of the work of Michael Sacks from our Oden Institute Willerson Center, in moving towards patient-specific, personalized heart care, and the work of Tom Yankeelov and David Hellmuth, also in the Oden Institute, also working with Dell Medical School and part of biomedical engineering, in building digital twins for cancer patients.
So I hope that helps to, as I said, provoke your imaginations to think about what might be possible. I personally could not be more excited about a future world where digital twins are enabling safer, more efficient engineering systems. They're enabling a better understanding of the natural world around us and they're enabling better medical outcomes for all of us as an individual.
Thank you.
(Applause)
