Engineering an Accurate, Affordable Model for Turbulence in Air and Space Travel
Unreliable computer models are holding aerospace back. Penn Engineering’s George Ilhwan Park is trying to break the jam with critical work on super-small currents.
By Gwyneth Shaw
The future of space travel could hinge on the management of tiny swirls of air turbulence. But first, engineers have to figure out just how far they can push the envelope.
George Ilhwan Park, an assistant professor in the Mechanical Engineering and Applied Mechanics Department in Penn’s School of Engineering and Applied Science, focuses on understanding and modeling those turbulent eddies, which happen when the flow of air over a wing is disrupted by even a small amount, producing a complex tangle of currents.
Armed with a grant from NASA and an astonishing amount of supercomputing time, he’s on the forefront of developing models that are affordable but also highly accurate in their predictions. Better models could improve the efficiency of aerial or waterborne vehicles and could also help build wind turbines and wind farms that are more stable and energy-efficient.
One of the major limits in aerospace involves the angle of attack, or the angle at which the wing of an aircraft is inclined to the direction of the flight. A sharper angle of attack creates more lift, a critical part of launching an aircraft and keeping it aloft.
However, at too steep an angle, air pressure opposing the natural flow passage starts to build up on the wing, and that can eventually create a slowly recirculating current. This can grow into a large blockage to the air flow and a sudden loss of the lift. That’s called a “wing stall,” and it can be difficult, or even impossible, for a pilot to recover from it.
For example, the Lion Air flight that crashed in October 2018 may have been caused by a domino effect set in motion by faulty sensors that misread the plane’s angle of attack, prompting part of the plane’s automatic steering to push the plane’s nose down to avoid a stall. Investigators examining the recent Ethiopian Airlines crash are looking at the same sensor issue as a possible cause of the accident. Pilots are taught to avoid a stall, and safety margins are built in to further reduce the odds.
“It’s better for us to operate at high-lift conditions so we can be more energy efficient,” Park says. “But we still want to be safe enough.”
That’s where the field of fluid dynamics, and Park’s work, comes in. His research is applicable to the design of an aircraft, not how it’s flown.
To improve airplane stall characteristics, dozens of wing designs need to be explored at a wide range of flight conditions. This is very expensive and difficult to do through actual physical experiments. Companies like Boeing and Airbus have long aspired to replace such costly wind tunnel tests with the tests performed in numerical wind tunnels: the computer simulations.
The crux of the challenge in simulating turbulent flows over wings is the tiny nature of the eddies comprising the turbulence close to the wing. To understand the relationship of that eddy, which could be smaller than a submillimeter, to the far larger wing, engineers need to create a sort of mesh, filled with grid points where the air velocity and pressure are calculated. That’s a huge number in one dimension, and the calculation must actually take in all three dimensions.
“We have to raise the power to the third power, and that number is just ridiculously large. We are at least a century away” from being able to simulate the problem that way, Park says.
Even the most powerful computers can’t handle it. What aerospace companies have been doing instead, he says, is to do calculations for some of the work and then model the effect of the fluctuations in the turbulence. These approaches are cheap but require artful calibration.
“The worst feature of these approaches is that they often completely miss the stall in situations where it happens in reality,” Park says.
Companies build in large safety margins, probably more than they need to. So even now, aircraft are almost certainly operating less efficiently than they could be. More sophisticated methods should crank up the accuracy while still keeping costs down.
“The basic idea is to compute what is resolvable and model only the features that are challenging to resolve,” Park says.
He aims to model only the smallest, worm-like turbulent eddies residing right above the wall surface, while computing the much-larger eddies in the outer layer directly. This approach is called wall-modeled large-eddy simulation.
This work is never complete unless the models can be applied to real-world problems.
“Many of the works done in this field stop when people demonstrate models in simple geometries. But we need to go beyond the flat plates or two-dimensional geometries to meet the industry need,” Park says.
Park has a separate grant to pay for the many, many computer hours he needs to run his simulations. His current joint grant with researchers at Stanford pays for 240 million core hours at the U.S. Department of Energy’s Argonne National Laboratory in Illinois, home of some of the most powerful machines in the world.
Yes, 240 million core hours.
It won’t take all that long, though, because of the capacity of the Argonne machines. Think about this way, Park says: A core hour is an hour on a single central processing unit, or CPU. An average laptop has 2 to 4; some of the computing clusters at Penn have thousands.
Argonne’s machines have more than 1 million cores. Park’s work won’t have access to all of them at once — these computers are often running multiple projects — but his results will be available in months, not decades.
“We were really excited about getting that award,” says Park. He’s using this resource to produce a one-of-kind calculation of a full aircraft geometry at flight conditions.
“This is a grand-challenge problem in computational fluid dynamics. This will be a breakthrough for a number of applications, from a passenger aircraft to a space shuttle to a fighter jet.”
This research was supported by NASA grants #NNX15AU93A and #80NSSC18M0155 and a U.S. Department of Energy INCITE Leadership Computing Award.