Friday 07 March 2025
Researchers have long sought to understand the mysterious forces that drive turbulence in fluids, a phenomenon that governs everything from ocean currents to atmospheric circulation patterns. Now, a new study sheds light on this complex process by using machine learning algorithms to solve a mathematical problem that has stumped scientists for decades.
The researchers focused on a specific type of fluid dynamics called shell models, which are simplified versions of the equations that govern turbulence in real-world fluids. These models have been used to study turbulence since the 1970s, but they’ve always relied on simplifying assumptions and approximations to solve them. The new approach uses machine learning algorithms to find solutions directly, without relying on these approximations.
The team’s technique is based on a type of neural network called a physics-informed neural network (PINN). This type of network is designed specifically for solving partial differential equations like those that govern fluid dynamics. By training the network on examples of known solutions, it can learn to find new solutions that meet the equations’ constraints.
The researchers applied this technique to two specific shell models: one that simulates turbulence in a fluid with no viscosity (i.e., it’s extremely thin and can’t resist motion), and another that simulates turbulence in a fluid with finite viscosity. In both cases, they found that the machine learning algorithm was able to find solutions that matched the expected behavior of real-world fluids.
One of the most intriguing findings is that the algorithm uncovered new types of behavior that weren’t predicted by traditional methods. For example, it found that in the low-viscosity fluid model, the turbulence could exhibit a type of self-similarity, where the same pattern repeats at different scales. This kind of behavior has been observed in real-world fluids before, but the researchers were able to use their algorithm to quantify and predict it with unprecedented accuracy.
The study’s findings have implications for our understanding of turbulence in general, as well as specific applications like designing more efficient aircraft wings or predicting ocean currents. By using machine learning algorithms to solve complex mathematical problems, scientists can gain new insights into the behavior of complex systems like turbulence, which could ultimately lead to breakthroughs in fields like engineering and climate science.
The researchers’ approach also highlights the potential for machine learning to augment traditional scientific methods. By combining the power of neural networks with the physical insight of traditional modeling techniques, scientists may be able to tackle problems that have long been considered intractable.
Cite this article: “Machine Learning Unlocks New Insights into Turbulence Dynamics”, The Science Archive, 2025.
Fluid Dynamics, Turbulence, Machine Learning, Shell Models, Neural Networks, Physics-Informed Neural Network, Partial Differential Equations, Fluid Viscosity, Self-Similarity, Complex Systems
