We are surrounded by flowing fluids. The air around us, the blood inside of us, both are flows that require complex equations to understand and predict. UCLA MAE Professor Jeff Eldredge, director of UCLA’s SOFIA Laboratory, explores a wide variety of phenomena that occur in fluid flows in nature and technology. In this interview, Prof. Eldredge explains his research and what he hopes to achieve.

What are the main subjects of your research?

Fluid dynamics is obviously a broad subject. I’ve always been interested in one unifying aspect of fluid dynamics, which is to understand how to take a complex fluid flow—that means air and water and other fluids and how they interact with objects, whether rigid objects or flexible structures—and reduce that complex flow into a simpler description. And, most importantly, control the flow or design a better system as a result.

As an example, we have always relied on wings, where the flow is nicely attached. We know stall is usually a bad thing to go through if you’re in a flying aircraft. If I turn my aircraft up too quickly, then the wing stalls, and then we lose lift and increase the drag on the aircraft. Those are the things that generally make aircraft fail, and so we try to avoid that.

But if you look at an insect, it actually flies in stall, permanently. Its wings flap back and forth, using stall to its advantage, because what an insect can rely on is that stall. When it first starts it actually has much larger lift force than we can ever get from the attached flow from a wing. So a stalled flow is an example where we avoid it partly because it’s unpredictable. It’s hard to design an aircraft that can exploit it. I use the word unpredictable, but what I really mean is that it’s hard to predict it with a simple model. How can I exploit stall if I don’t have a simple model? How can we control the wing’s aerodynamics if stall is such a potentially dangerous event?

We want to fill that missing ingredient. Basically, if we can easily describe something like wing stall and all the nice benefits that we can get from it, then we can design aircraft that can fly in and out of stall and as a result they’re a lot more agile and they’re also more robust as well. For example, in light aircraft, the ones we’re tending to build now, that don’t have people flying in them, they’re uninhabited, they might fly through some disturbance in the atmosphere and since they’re so light that disturbance would adversely affect them, probably it would be dangerous for them to fly in that scenario. If they can they take advantage of new tools for controlling the flow around the wing better, it would not be such a dangerous encounter.

This image is a set of snapshots from a simulation of a process called “dynamic stall” past a flat plate wing. It shows the streamlines over the wing (in gray) and vorticity (the local rotation of the flow) (in color). Jeff Eldredge, SOFIA Laboratory.

What are the challenges of studying wings?

Most of the challenges are in how the wing encounters something that is unpredictable. Generally in aerodynamics it’s called a gust, and we think of turbulence obviously when we fly in aircraft like commercial aircraft, and that’s an example of a gust. Heavier crafter are less prone to problems because the gusts we encounter in a heavy aircraft aren’t that dangerous. They’re uncomfortable but not that dangerous.

When you’re flying personally, you have a sense of how the aircraft is put together, more than most people sitting around you. You’re getting turbulence, but in your own mind, you’re saying this is okay, I know how this plane is designed.

I know the plane is designed to absorb turbulence, basically through its own inertia. It’s heavy, and being heavy means you’re less prone to dangerous dynamics of the aircraft. It bounces around a bit but we also know that unless we were flying through big weather systems,  dangerous tropical storms and that sort of thing, then the most dangerous encounter you’re going to experience is still within the design margin of the aircraft. So I’m confident that as long as the aircraft has been maintained well, that a turbulent ride is just uncomfortable, and nothing more than that.

But for small aircraft, lighter aircraft, they don’t have that inertia, they don’t have this mass to them, and because they’re so light, then the same turbulence, the same atmosphere that maybe a 747 would fly through, would be much more dangerous for a lighter aircraft to fly through. And that’s true even about single engine aircraft that maybe one or two people fly in. It’s much more dangerous for them to fly through.

This image shows snapshots of a full computationally-simulated flow field past a flat plate (top row) and a much faster model prediction of the flow (bottom row). Jeff Eldredge, SOFIA Laboratory.

A large gust of wind can blow a small plane away.

Or flip it over. That’s also true about lighter aircraft encountering the wake of a larger aircraft. That’s the reason why they have to space aircraft out when they land, because the wake of a large aircraft like a Dreamliner or a 777 is much more dangerous for a 737 flying behind it or something smaller than that, so they have to space them out by several miles, and that’s part of the airport’s operations. It’s designed to eliminate those sorts of encounters. But if you can design an aircraft so that it’s better able to deal with that encounter, then you can make your aircraft operations run more smoothly.

Are you doing all of your design work on computers?

Our work is computational, theoretical, so we work with experimental collaborators, then test these ideas out by putting new actuators on wings and putting it into a laboratory. Our work is in developing tools to take governing equations of fluid dynamics, and reduce them to their simplest principles, it’s what we call model reduction. We do that in a combination of ways that are physically intuitive and also mathematically rigorous. We use tools now actually from data science, and machine learning, for example, in combination with fairly classical tools in fluid dynamics. By doing that we can make very rapid models, models that are easy to compute.

The equations of fluid dynamics are extremely complicated. Predicting just the flow over a wing that’s stalled, like I mentioned, what we call a high-fidelity simulation, trying to capture all the physics of that flow, it generally take some days to do that sort of simulation now on a supercomputer. We want to be able to do that in seconds, and even better, in real time. Imagine we would do this calculation, at least a prediction of the flow that’s passing over this wing on board the flight computer, and so by doing that, then you put this into a feedback loop, you say okay I’ve taken a few measurements on the surface of the wing – to the extent possible, maybe a few pressure sensors on the wing. And I can take those measurements, and also use reduced order models like we’re developing. And then from there figure out what the rest of the flow looks like, even though I can’t see it. By then filling in all that information, I’m better able to make a decision, and when I say, “I,” what I mean is a flight computer, which would make a better decision about on how should I control my wing so that is able to negotiate this surface of a gust. It’s all part of this feedback loop where you need as much information as possible about what the flow is doing and then also to use that to control the flow better, control the flight better.

What have you learned since you first started teaching? What’s been your evolution in terms of your research? How has your research changed in terms of your approach?

I’ve always approached research problems from the mindset of a student. When I look at a new problem, I naturally want to look at it from the simplest way possible. In some ways, that’s my end goal, to take something complicated, and look at it, find the simplest way that also has the same features to it. The idea of most of our researches isn’t striving to make the most complicated simulations that are amazingly detailed. Actually, something just the opposite of that, which is reduce fluid dynamics to its simplest principles.

Similar to an E = mc2 simplification.

You might think of it that way, exactly. I look at it as trying to find the skeleton of a flow.  We’re trying to look at the essence of the thing that actually makes the fluid what it is, the flow what it is. Teaching is the same way, I’m trying to take a complicated subject like fluid mechanics, and work with people’s intuition. Most students come into fluid dynamics, even before they’ve seen the subject, with some intuition, so they start off with a little bit of intuition, and I try to build that up. We build that up with simple problems, the problems that I think are from different categories, the simplest example of a certain phenomenon, for example. Teaching and research, I think, have similar principles, at least the way I follow it. I think teaching and research has always been coupled in my life.

This image is a set of snapshots from a simulation of a scalpel resecting a liver. Jeff Eldredge, SOFIA Laboratory.

Do you see a project that’s 10 years down the road that you want to accomplish? Something you’ve been thinking about for a long time?

Well I think that there’s a couple of them – one I’ve already mentioned – but I’ll say more about it. My colleagues are building better and better actuators. So for example, a wing of an aircraft flies with the same control surfaces – that means the ailerons, the flaps and so on – that we’ve used for decades. They’re limited, because they can’t respond as quickly as they might need to. There’s new actuators coming in, and you combine those better actuators that are not based on the pneumatics that you use on flaps and so on. In my lab, with our model reduction principles, is the idea of estimating the flow like I’d mentioned. Combine all these things together, you can foresee an aircraft that flies more agilely than any other aircraft ever built, or something that approaches what a bird or insect can do. For example, it may not flap its wings, but it can exploit certain things that we’ve never been able to do before. So that’s one example.

The other half of my research group is working on something entirely different, which is modeling cardiovascular flows, and flows inside the human body. There, we’re working towards developing tools for surgical simulations. Imagine, you have a new medical student who wants to become a surgeon. Of course the best way they learn is by actually applying surgery to a patient or maybe a cadaver. But then imagine you can build a simulator that would simulate the situation of doing surgery on an organ.

Obviously, that has to be interactive, and also it has to be realistic. It has to capture all the physics, and all things that happen during that process. And that includes bleeding. Developing  a realistic model of a human organ, that behaves the way it should when you virtually apply a scalpel to it, for example, is something that’s never been figured out because, like I mentioned before, fluid dynamics and those equations are really complicated. They take a long time to process. If I apply a scalpel to an organ, then I can’t get the bleeding to look right, because I have to solve the underlying physics of fluid dynamics on that computer to make that happen, and I can’t do that quickly enough.

But now, if you combine machine learning principles that have been used in other areas, along with those equations, we might actually have some hope of doing that. An interactive tool that learns, because you trained it with lots of other simulations of cutting an organ.

The liver is our target problem right now. The liver’s very complicated because it’s got three different circulatory networks inside of it, and it’s got tissue that is highly elastic, hyperelastic in fact. It also has to bleed when you cut the liver, so it’s responding properly. We’ve made a lot of progress on that. But it’s not fine-tuned yet, so that’s what we’re working towards.

Crow in Flight, photo by Pheanix, 2009, CC BY 2.0.

You’ve mentioned the flight of insects. What about birds? From your window, you can see crows. They can be in a tree, drop forward, spin once, and land on the ground perfectly, every time. When you see crows do that, do you also see the fluid dynamics behind it?

What I see, is that birds get really far without having to worry at all about fluid dynamics, or at least they don’t have to do all the complicated work that we feel like we need to do to solve for the flow around a bird. They learn to fly. They achieve that remarkable agility by experience, by learning what happens when they move their wings in subtle ways. So they know if they move their wing a little bit this way, then it might cause a roll of their body, for example, or if they bring their wings out in a spread form, they’ll slow down and be able to land with control. All these things are things they’ve learned and so I think it’s something worth finally realizing now in fluid dynamics, that we don’t have to put all these tools together in maybe the way we thought. It’s more of a learning principle that we need to be applying.

Jean-Marie Le Bris and his flying machine, Albatros II, 1868.

Could a plane be built that could adjust its wings like a bird?

A plane on autopilot translates the command of a pilot into a lot of different things. For example, it deploys the flaps when you’re about to land, in order to make the lift a lot larger for the slower flight speed. That’s just a very simple example of something that if we had better materials and better actuators and so on, we could make wings that were much more dynamic, things that actually could tilt upwards in nice ways, basically be deformable – much more deformable than they are. Obviously a bird’s wing, a bat’s wing, or any other creature’s wing is much more compliant, it’s able to change its shape.

Are deformable wings out there right now?

They are, but they’re all “proof of concept” things, and so they still have various deficiencies that haven’t been addressed yet. People are trying to address them. One example is wings built from smart materials may not always respond as quickly as you want them to, or maybe they require too much energy input to make them deform, so there’s lots of different threads of thought on that. Some people are building membrane wings, for example, that have a framework to them, but they don’t know how to design the exact structure of that wing the most optimal way, so it’s more trial and error, which is not a very good way to do things. But maybe we can learn, and that’s what we’re trying to do. I think by using robust learning principles like from machine learning maybe we have some hope of figuring out how biology sources down.

How do you view the human body philosophically, in terms of your research? Is there a “ghost in the machine?”

Well, I stay agnostic on those questions, but I definitely would say that I look at it more as a system that has fine-tuned itself over many, many, many generations to be able to handle lots of disturbances. The human body is remarkably capable, but we know the principles how it responds to these disturbances too, we figured out for example that there are sensors in the cardiovascular system that will detect if there’s a hemorrhage going on, it senses there’s lower pressure and it changes how the heart responds. There’s a neural system that we’re not even aware of, that is always controlling how our heart responds to the various changes in our environment. It happens even when we sit up from lying in bed, sitting up for the day, we know that our heart starts beating a little faster, for example, as the blood flow changes under the load of gravity, and then it happens throughout the day. Or if I drink coffee for example, that raises my heart rate. So all of these adjustments are going on automatically, without us having to know about them. So it’s not a ghost in the machine. The machine itself is just highly tuned, is how I think. That’s a part of the responsibility of designing or developing any simulator that we might be trying to do. We need to be able to capture that automatic response. Just to give you an example of the liver, if a liver starts bleeding profusely, then we’re losing blood at a certain rate, then I need to also simulate how the whole body responds then. We also include in that model the cardiovascular function, the heart’s function, the sympathetic nervous system, that’s the part of the nervous system that I mentioned, that controls how the heart behaves.

Do you need to build a complete body model simulation, since everything in the body is connected?

It’s all connected, that’s true, but you can imagine that there are strong connections and weak connections. It’s not so important that I know, for example, what the fingers are doing, or how much less blood goes to the fingers. It’s a complicated network that’s able to compartmentalize a lot of things, so if the liver is bleeding a lot, it can still regulate flows to the brain pretty well, at least while we haven’t lost too much blood. But that’s part of the challenge, while bleeding continues, how does the body preserve the essential functions of the body primarily has to keep blood going to the organs and sacrificing extremity. Those are the challenges, capturing all those physics in a bottle.

Article by Alex Duffy

  • 165

 UCLA Samueli Mechanical and Aerospace Engineering