Researchers at USC, the University of Michigan, and Argonne National Laboratory are using artificial intelligence to model turbulent flows more accurately than ever before, potentially transforming how engineers design aircraft, power plants, and fusion reactors. The U.S. Department of Energy has selected the team to lead a multi-institutional project through its Genesis Mission, a national initiative pairing DOE's 17 national laboratories with universities and industry to pursue AI-driven scientific discovery.

Why Is Turbulence So Hard to Simulate?

Turbulence is everywhere. It's the bumpy ride at 30,000 feet, but it's also the invisible fluid motions that dissipate energy in engines, affect combustion efficiency, and influence how fluids move through industrial systems. Engineers rely on computer simulations to predict turbulent behavior, but here's the problem: calculating every motion down to the smallest scales typically exceeds the capabilities of even today's supercomputers.

The current gold standard for turbulence prediction is large eddy simulation (LES), which calculates larger fluid motions while using mathematical models to approximate the effect of smaller motions. The accuracy of these smaller-scale models is often the limiting factor for the entire simulation, since most models work from a single snapshot of the flow's current state to estimate how smaller scales will behave.

How Does AI Improve Turbulence Modeling?

Bermejo-Moreno's research team proposes a fundamentally different approach. Instead of relying on snapshots, they plan to use AI and machine learning to identify coherent structures within turbulent flows, map their geometry, and track how each formation interacts as the flow evolves over time. The result is a model that learns from a history of interactions, not just a single moment in time.

"The bridge here is between artificial intelligence and machine learning methods applied to this analysis of turbulent flow physics. We have methodologies that we've been developing over several years, and now we want to inject AI and ML techniques to accelerate and enhance those approaches," said Iván Bermejo-Moreno, associate professor at USC Viterbi's Department of Aerospace and Mechanical Engineering.

Iván Bermejo-Moreno, Associate Professor, USC Viterbi Department of Aerospace and Mechanical Engineering

The team estimates that this added historical context will generate a model that holds up across different types of flows and different simulation codes, making it more generalizable and robust than current approaches.

What Real-World Problems Could This Solve?

The practical implications are significant. Aircraft, turbomachinery, and wind turbines are typically designed using physical experiments and lower-grade computational methods because the most detailed simulations are too expensive to run repeatedly while comparing designs. The same limitation applies to emerging technologies like inertial confinement fusion, one of the methods the DOE is exploring as an alternative energy source.

Faster, more reliable turbulence models could break through this bottleneck, enabling engineers to use high-fidelity simulation while comparing designs and running enough simulations to test how designs perform under changing conditions, a process called uncertainty quantification. This could accelerate innovation in several critical areas:

• Aerospace Design: More accurate turbulence models would allow engineers to optimize aircraft designs for reduced aerodynamic drag without relying solely on expensive physical wind tunnel testing.
• Energy Systems: Better turbulence prediction could improve combustion efficiency in power plants and help develop more reliable inertial confinement fusion reactors.
• Industrial Fluid Systems: Improved models would help optimize how fluids move through pipes, reactors, and other industrial equipment, reducing energy waste.

"By trying to develop these fine-scale models that are more robust, more capable and faster, we can tackle problems that we couldn't before. That's going to be highly impactful for design optimization and uncertainty quantification in turbomachinery, aerospace and energy systems," explained Bermejo-Moreno.

Iván Bermejo-Moreno, Associate Professor, USC Viterbi Department of Aerospace and Mechanical Engineering

How Will the Research Team Approach This Challenge?

The project brings together complementary expertise from three institutions. USC's group specializes in identifying and analyzing turbulence structures; Ricardo Vinuesa's team at the University of Michigan focuses on AI applications for fluid flows; and computational scientists Ramesh Balakrishnan and Riccardo Balin at Argonne National Laboratory provide high-performance computing expertise and AI infrastructure support.

The research will run on some of the world's most powerful supercomputers, combining Argonne's high-performance computing and AI capabilities with Michigan's work on machine learning for fluid dynamics and USC's decades of experience in turbulence-structure analysis. Bermejo-Moreno emphasized that this combination of skills creates a strong foundation for advancing the state of the art.

Physical experiments will remain essential to validate the models, Bermejo-Moreno noted. However, experiments are expensive and time-consuming to conduct and analyze. By developing numerical simulations that are both accurate and efficient, researchers can reduce the burden on experimental facilities while enabling faster design iteration.

The Genesis Mission represents a broader shift in how the DOE is approaching scientific discovery, pairing national laboratories with academic institutions and industry partners to accelerate AI-driven breakthroughs. This particular project exemplifies how machine learning can enhance traditional physics-based modeling rather than replace it, creating hybrid approaches that leverage the strengths of both computational methods and artificial intelligence.