DOE’s Aurora supercomputer enables next-gen airflow simulations for aircraft efficiency

Questions and answers:

• What is the Aurora supercomputer used for at Argonne National Laboratory? Aurora is used to perform exascale-scale simulations and artificial intelligence research, including modeling airflow around commercial aircraft to improve design efficiency.
• How does the Aurora project differ from traditional aircraft design methods? Aurora’s approach combines high-fidelity fluid dynamics simulations with machine learning to create improved turbulence models, reducing the need for costly wind tunnel and flight tests.
• Who are the key partners involved in the BARD mission and Aurora research?
  York Space Systems in Denver, NASA’s Space Communications and Navigation Program, Johns Hopkins Applied Physics Laboratory, the University of Colorado Boulder, and Argonne National Laboratory are the primary entities involved.

LEMONT, Ill. - The Aurora supercomputer at the U.S. Department of Energy’s (DOE) Argonne National Laboratory in Lemont, Ill., is enabling researchers to explore new methods for designing more efficient airplanes using advanced simulation and artificial intelligence.

Aurora, one of the world’s first exascale supercomputers capable of performing over a quintillion calculations per second, is housed at the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science user facility. The system also ranks among the top worldwide for artificial intelligence performance.

A research team led by the University of Colorado Boulder is leveraging Aurora’s computing power and machine learning techniques to study airflow around commercial aircraft. The project aims to generate insights that will inform the design of next-generation airplanes.

XTI Aerospace taps Oak Ridge supercomputer in development of its TriFan 600 VTOL

Aero optimization

Currently, aircraft vertical tails are sized to handle worst-case conditions such as crosswind takeoffs with one engine out, often resulting in larger tails than necessary during typical flights. "The vertical tail on any standard plane is as large as it is precisely because it needs to be able to work effectively in such a situation," said Riccardo Balin, assistant computational scientist at ALCF. "The rest of the time, however, that vertical tail is larger than would be necessary, thus adding unnecessary drag and fuel consumption."

The team uses Aurora to run large-scale fluid dynamics simulations with HONEE, an open-source solver designed to model turbulent airflow behavior. These high-fidelity simulations provide training data for machine-learning-driven subgrid stress models, which improve turbulence models used in lower-resolution simulations.

Improved subgrid stress models can reduce simulation costs while maintaining accuracy, potentially decreasing reliance on expensive wind tunnel and flight testing.

Unlike traditional turbulence models that depend on stored datasets and offline analysis, the team’s approach employs "online" machine learning during simulations, saving time and reducing data storage needs.