Aircraft simulations scale new heights with frontier supercomputer

High-accuracy aerodynamic simulations are essential to designing commercial aircraft that are safer, faster, and more fuel efficient. But the physics of airflow is so complex that the aerospace industry needs more computing power to capture turbulence in its most chaotic states.

That’s why Boeing researchers turned to Frontier, the world’s most powerful supercomputer for open science. Located at the Department of Energy’s (DOE’s) Oak Ridge National Laboratory (ORNL), Frontier enabled Boeing researchers to enhance their simulation capabilities to better predict how the accumulation of ice and microturbulent airflow affect flight performance, especially during scenarios involving complex landing configurations.

The Frontier simulations are expected to help guide aerodynamic design improvements, mitigate issues found during flight testing, and support efforts to develop more fuel-efficient aircraft.

“We want to fly in the laminar regime as long as possible because that’s the most fuel-efficient way of flying, where airflow is smooth and steady,” said Adam Clark, an aerodynamics engineer at Boeing. “Reducing fuel consumption helps lower the operational costs for the airline, which could also translate to less expensive tickets for consumers.”

Boeing’s research shows that reducing the amount of drag by 1 percent while flying in cruising conditions can lower fuel consumption by about 0.75 percent on a flight, which means small changes can add up to big savings.

“Improving aerodynamic designs means we need to advance our ability to simulate airflow on grids with incredibly dense meshes,” said co-investigator and fellow Boeing aerodynamics engineer Konrad Goc. “But that’s something we really can’t do as effectively without more powerful computers.”

The molecular mechanics of airflow

Turbulence is a chaotic, nonlinear system that’s constantly moving. That’s why solving turbulence, or predicting how it will behave, is such a difficult problem mathematically. In fact, turbulence is one of the few problems that traditional physics can’t fully solve.

Accurately predicting airflow requires an almost microscopic understanding of how air molecules move, but the more closely researchers look, the more complex the problems become. Their focus is on the boundary layer — the thin layer of air closest to the wing — where interactions with the surface create drag, which slows the aircraft down and increases fuel consumption.

“It’s not enough to study just the turbulence we experience when we’re on an airplane and the captain turns on the seat belt sign,” Clark said. “We need to understand it at a fundamental level.”

“Imagine a bunch of marbles on a table. If you push them slowly with two hands, they will kind of stay together, but if you move them impulsively, they will interact in a complex and chaotic way,” Goc explained. “That’s a very basic way to describe the difference between air molecules in laminar and turbulent flows, except we’re dealing with millions and billions of marbles.”

The engineering approach to calculating wind

There are three primary approaches to evaluating aircraft aerodynamics: wind tunnel experiments, flight tests, and computational fluid dynamics (CFD) simulations powered by high-performance computing. Each method provides complementary insights, iteratively validating and refining designs until they converge on a final solution.

Wind tunnels are large facilities used to replicate a wide range of aerodynamic conditions in a controlled environment, whereas flight tests provide the ultimate assessment of an airplane’s performance. Both are essential but can be resource intensive. CFD simulations offer a faster, lower-cost, and lower-risk alternative to some physical tests.

“Over the last 20 to 30 years, computer modeling has allowed us to go from having to perform 20-plus wind tunnel experiments on different wing variations to now just a couple,” Clark said. “The goal isn’t to replace wind tunnels, but oftentimes, in physical testing, we see things that we can’t explain — things that we literally have no data for, and that’s what CFD simulations can shed a light on.”

The science behind CFD

The team’s computational efforts center on high-lift aerodynamics, which includes the low-speed takeoff and landing phases of flight.

“That’s when the parts of the wings move forward and backward as the plane gets closer to the airport,” Goc explained. “We’re looking at how the aerodynamics of the wings change as the angle of attack, or the pitch of the airplane, increases. So we’ve really been trying to hone our engineering practices around getting better accuracy in our predictions.”

CFD simulations model airflow by overlaying a grid, or mesh, onto an aircraft or its individual components. The mesh can contain hundreds of millions of cells, or more, down to millimeters in size. The simulation calculates properties such as velocity, pressure, and temperature for each cell. These results are combined to visualize airflow over components and then integrated to provide a complete picture of the aircraft’s overall aerodynamic performance.

Turbulence simulations vary widely in complexity. Engineers have long relied on Reynolds-averaged Navier-Stokes equations, which involve modeling turbulence statistically — fast and practical for full aircraft but limited in capturing complex, unsteady flows.

Higher-fidelity methods such as wall-modeled large eddy simulation resolve larger turbulent structures and finer scales near the boundary layer but at much greater computational cost. At the highest level, direct numerical simulation resolves all turbulence without modeling, thereby delivering maximum accuracy. However, the enormous computing power it requires makes it impractical for full-scale aircraft simulations.

“If we were to run a full-scale wall-modeled simulation on our in-house system, it could take several months just to get a single solution,” Clark said. “But we’re not looking for a single answer. We need ensembles of answers to understand how airflow changes throughout a process — for example, not just when an airplane is taking off but also as it gains altitude and airflow begins to stabilize. That’s what we need to be confident in the performance of our turbulence modeling code.”

The power of exascale computing

Frontier is the world’s first exascale supercomputer, with a peak performance of 2 exaflops — or 2 quintillion calculations per second. Boeing gained access to Frontier through DOE’s INCITE Leadership Computing program, which provides industry, academia, and government researchers with access to the nation’s most powerful supercomputers.

To simulate turbulence, Boeing collaborated with computing experts at ORNL’s Oak Ridge Leadership Computing Facility (OLCF), a DOE Office of Science user facility, to optimize their tests and increase the speed and resolution of their simulations.

With higher-fidelity simulations, the team focused on two key challenges: investigating how ice accumulation affects wing performance and examining how wings respond to airflow when turbulence is modeled at full complexity. This work helped shed light on previously unexplained wind tunnel results for some high-lift wing designs.

“All of our airplanes are certified to operate safely in those conditions, and with Frontier, we were able to simulate flight through cold, icy clouds to understand how different ice shapes, sizes, and textures affect performance,” Clark said.

Because the temperature of the wings influences how air moves over the surface, the GPU-accelerated simulations also gave the team a powerful new way to evaluate their anti-icing technology — both in preventing ice buildup and in removing it once it forms.

“I would consider laminar-to-turbulent transition to be one of the central challenges in fluid mechanics, including the impact wall heating has on a turbulent boundary layer. It’s an invisible problem, something we can’t see in wind tunnels or by looking out the window during flight tests,” Goc said. “Frontier was really instrumental to that end because direct numerical simulations can be very computationally and cost intensive, limiting their use for airplane product development.”

The researchers said the accuracy of the simulations allowed them to predict the coefficient of lift — the upward force on an aircraft — to within 1 percent, which gives them the confidence they need to advance certain prototypes into production much faster than before.

“Not only did Frontier allow us to run simulations that we couldn’t do before, but we were also able to get results in a matter of days, which would have taken months using a typical industrial-scale computing system,” Goc said.

The broader impact for the aerospace industry

The team says it isn’t just Boeing that will benefit from the Frontier simulations. Because the team used a widely studied high-lift model, the findings will be easier for the broader aerospace community to build on, thereby helping advance aerodynamic designs across the industry for safer, more economical aircraft.

“As an aerospace company, we live at the intersection of aerodynamic theory, scientific computing, and physical testing,” Goc said. “This work helps translate fundamental scientific advancements into real-world engineering applications.”

Related Publication: Goc, K. A., and J. D. Crouch. “Surface heating effects on attachment-line transition and relaminarisation.” Journal of Fluid Mechanics (2026). DOI: https://doi.org/10.1017/jfm.2026.11949

This research is supported by DOE’s Advanced Scientific Computing Research program.

UT-Battelle LLC manages ORNL for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit https://energy.gov/science.