What is it about?

Moving heat out of a hot surface is hard when the surrounding fluid is swirling chaotically. We trained an AI to learn when—and where—to briefly switch tiny heater patches on or off at the wall. This pattern makes the fluid carry away more heat than standard approaches: in our main tests it increased the amount of heat removed by about 38.5%, and a distilled, easy-to-implement on–off rule reached 40.0% under even tougher conditions. The simple rule comes directly from what the AI discovered and doesn’t require retraining. Physically, the on–off pattern triggers deeper, stronger “hot plumes” that reach beyond the thin layer near the wall, so more heat leaves the system. Improving how much heat can be removed matters for technologies like heat exchangers and electronics cooling.

Featured Image

Why is it important?

Smarter, low-cost control that pulls more heat from chaotic flows can translate to more efficient heat exchangers, cooling of power electronics, and other thermal systems, cutting energy use and costs. By showing that an interpretable AI rule can outperform and complement classic controls—even under realistic sensing/actuation limits—this work opens a path from simulation to deployable heat-transfer upgrades across industry.

Perspectives

What began as a curiosity—could an AI really help turbulent flows carry more heat—turned into months of lively debates, failed ideas, and, finally, a simple insight that surprised us: a switch-like, on–off pattern was enough to draw out much more heat. Seeing something so minimal emerge from a complex system felt both humbling and thrilling. I’m especially proud that the result is not just a performance number but a clear, interpretable rule that others can try. My hope is that this work nudges engineers and researchers to look for simpler, more robust ways to improve heat removal—whether in heat exchangers, energy systems, or electronics cooling—and that it sparks new conversations between control, AI, and fluid physics.

Xiaojue Zhu
Max-Planck-Gesellschaft zur Forderung der Wissenschaften

Read the Original

This page is a summary of: Deep reinforcement learning control unlocks enhanced heat transfer in turbulent convection, Proceedings of the National Academy of Sciences, September 2025, Proceedings of the National Academy of Sciences,
DOI: 10.1073/pnas.2506351122.
You can read the full text:

Read

Contributors

The following have contributed to this page