USC Researchers Use AI to Crack Classical Physics' Turbulence Puzzle

Researchers at the USC Viterbi School of Engineering are utilizing artificial intelligence and machine learning to decode the physics of turbulence. The initiative seeks to overcome the massive computational hurdles that have long hindered the accurate modeling of complex fluid flows, which govern everything from aircraft wing aerodynamics to global weather patterns.

Led by Associate Professor Iván Bermejo-Moreno, the research group is moving beyond traditional "snapshot" models. Instead, the team is developing AI models capable of learning from a history of interactions, a shift designed to surpass the limitations of conventional simulations such as Large Eddy Simulation. This work is part of the U.S. Department of Energy's (DOE) Genesis Mission, a strategic collaboration that includes the University of Michigan and Argonne National Laboratory.

The Computational Bottleneck

Turbulence remains one of the last great unsolved problems in classical physics. Traditionally, scientists have relied on the Navier-Stokes equations to model these flows. However, solving these equations for high-Reynolds-number flows—where turbulence is most intense—is computationally expensive, often requiring resources that exceed current hardware capabilities. By shifting the burden to AI, researchers hope to find shortcuts to these complex calculations without sacrificing accuracy.

Industry Implications

By integrating AI to uncover hidden patterns in turbulent flows, the Genesis Mission aims to revolutionize design optimization and uncertainty quantification. The practical applications span several critical sectors: aerospace engineering, the development of more efficient turbomachinery, and advanced energy systems. Specifically, the research could provide breakthroughs in inertial confinement fusion, where precise control of fluid dynamics is essential for sustainable energy production.

The Path Forward

If successful, these AI-driven models could lead to significantly more efficient transportation by reducing drag and improving the precision of climate modeling. The project continues to refine how machine learning can synthesize historical flow data to predict future turbulence. While the theoretical framework is advancing, the full integration of these models into standard engineering workflows remains the next major milestone for the collaboration. This transition would allow engineers to test designs in virtual environments with a level of fidelity previously reserved for the most expensive supercomputers.