A giant leap forward in solving a classic problem: turbulence

Turbulence, responsible for many bumpy flights and a large part of the energy consumption in vehicles and aeroplanes, remains one of the great unsolved problems in physics. Now, a team from the Universitat Politècnica de València (UPV) and the University of Michigan (UM) has developed a new explainable artificial intelligence model to identify the most influential regions within turbulent flows and advance our understanding of them. The work has been published in Nature Communications,

"A better description of turbulence makes it possible to predict dangerous areas in flight and reduce risks for passengers. It would also help to manipulate it in industrial processes, improve combustion or reduce aerodynamic drag, a goal with a huge economic impact," says Sergio Hoyas, a researcher at the Institute of Pure and Applied Mathematics (IUMPA) at the UPV and co-author of the study.

Solving the turbulence puzzle

For more than a century, turbulence has been a puzzle: overly complex equations, difficult experiments and insufficiently powerful computers have prevented it from being deciphered.

'AI now gives us a new tool with enormous potential to try to solve the puzzle and identify which regions of a turbulent flow are truly the most important in its evolution,' adds Andrés Cremades, also a researcher at IUMPA at the Universitat Politècnica de València.

In their study, the UPV-UM team analyses turbulence using their new AI model. Based on a highly detailed simulation of a turbulent flow, the AI algorithm estimates its importance in the dynamics of turbulence. Unlike other AI-based work, which functions as a 'black box,' this method not only predicts the evolution of the flow but also indicates which specific regions have the most significant influence on its development.

To train the model, the researchers combined high-precision numerical simulations with explainable artificial intelligence techniques, known as SHAP. 'Now we know exactly which regions of the flow we need to modify if we want to reduce drag, improve combustion or decrease pollution,' explains Sergio Hoyas.

More efficient control strategies

The work of the UPV-UM team is of particular interest for designing more efficient turbulence control strategies, which can reduce friction, energy consumption, and wear in industrial systems. Considering that around 15% of the global energy is lost due to turbulence-related effects, accurately identifying key areas of the flow can contribute to the development of more sustainable technologies in sectors such as aeronautics, automotive, and wind energy.

The trillion-dollar problem

According to the team, the technique can be applied to other physical problems where it is necessary to identify which factors are really important.

'Proving the existence and uniqueness of solutions to fluid mechanics equations is known as the million-dollar problem. Solving turbulence in a practical way would be the trillion-dollar problem,' concludes Ricardo Vinuesa.

Reference

Cremades, A., Hoyas, S., & Vinuesa, R. (2025). Classically studied coherent structures only paint a partial picture of wall-bounded turbulence. Nature Communications, 16(1), 10189.

Classically studied coherent structures only paint a partial picture of wall-bounded turbulence | Nature Communications