Understanding turbulence through artificial intelligence

When one hears the word turbulence, it is common to think of the uncomfortable movements we suffer when traveling on an airplane. However, turbulence is much more and is continually present in our lives.

By turbulence we refer to the irregular and chaotic state presented by the movement of fluids, gases and liquids, in most situations. Examples of turbulent flows are the movement of air in our cities or water in seas and rivers, but also that which occurs inside engines or around cars, boats and airplanes.

In fact, turbulence is one of those responsible for the loss of energy in these means of transport, being responsible for up to 15% of the carbon dioxide (CO2) released by humanity annually.

Now, an international team made up of scientists from the Polytechnic University of Valencia and the universities of Edinburgh in Scotland and Melbourne in Australia and led by Ricardo Vinuesa from the Royal Institute of Technology (KTH) in Sweden, has developed a new technique that allows studying the turbulence in a completely different way than that used by the scientific community in the last 100 years. His work has been published in the academic journal Nature Communications.

Artificial Intelligence, fundamental

The main difficulty of fluid mechanics is that “although the equations of fluid mechanics are about 180 years old, the problem is still open. These equations are unsolvable algebraically or numerically for practical cases, even for the largest computers in the world. For a typical commercial airplane, we would need a memory equivalent to a month of internet just to be able to configure the simulation,” says Sergio Hoyas, professor of aerospace engineering at the UPV and researcher at the IUMPA. “We need to understand turbulence so we can improve the simplified models that are used every day. And there is a new tool: artificial intelligence,” adds Ricardo Vinuesa.

Although there are already several works that apply artificial intelligence to fluid mechanics, the great novelty of this study is that it allows for the first time not to simulate or predict, but to understand turbulence.

Andrés Cremades and Sergio Hoyas, from the research team. (Photo: UPV)

Using a database of about 1 terabyte, the team of researchers has trained a neural network that allows predicting the movement of a turbulent flow. Using this network, he has managed to follow the evolution of the flow by eliminating small structures individually, subsequently evaluating the effect of these structures using the SHAP algorithm.

“The most important thing is that the results of this analysis exactly match and expand the knowledge acquired over the past 40 years. Our method has managed to reproduce this knowledge without the neural network knowing anything about physics,” highlights Andrés Cremades, postdoctoral researcher at KTH and first author of the study.

“Experimental validation with data from the University of Melbourne indicates that our method is applicable to realistic flows and opens a totally new path to understanding turbulence,” concludes Vinuesa. (Source: Polytechnic University of Valencia)