Upcoming Physics & Astronomy Mathematics & Statistics
Physics-Informed Neural Networks for Boiling Flow Reconstruction: Resolving Microlayer Dynamics
Summary
Original abstract (not yet simplified)The relentless miniaturisation of transistors and the shift towards 3D integrated circuits have created an unprecedented thermal management crisis. Heat fluxes in next-generation AI hardware and data centres exceed 1kW/cm², with over 55% of electronics failures attributed to overheating. Two-phase immersion cooling, where components are submerged in dielectric fluids to exploit boiling heat transfer, emerges as the most promising solution,...
View original technical description
The relentless miniaturisation of transistors and the shift towards 3D integrated circuits have created an unprecedented thermal management crisis. Heat fluxes in next-generation AI hardware and data centres exceed 1kW/cm², with over 55% of electronics failures attributed to overheating. Two-phase immersion cooling, where components are submerged in dielectric fluids to exploit boiling heat transfer, emerges as the most promising solution, achieving heat transfer coefficients up to 100 kW/m²·K, three orders of magnitude higher than air cooling. This proposal aims to fundamentally investigate the intricate transport phenomena governing boiling, with a particular focus on microlayer evaporation beneath the bubbles, a local mechanism that can dominate interfacial heat flux but whose quantitative role remains uncertain because it occurs at sub-micrometre length scales and evolves on short timescales. Advanced experimental techniques have enabled simultaneous measurement of temperature, velocity, and heat flux distributions during bubble nucleation. However, the spatial resolution of current systems is insufficient to capture steep temperature gradients at the bubble interface, and are fundamentally limited to single-bubble or low bubble density studies due to optical path obstruction by multiple growing bubbles. The pressure field governing interfacial forces cannot be measured directly from the experiments. This leaves critical physical quantities “hidden” to experiment. The MICRO-PINN project will develop a Physics-Informed Neural Network framework that reconstructs complete 4-dimensional velocity, pressure, and temperature fields from sparse experimental data. The goal is to generate new physical insights and engineering guidelines for the design and optimisation of next-generation immersion cooling systems. This project will contribute to SDG-7 (Affordable and Clean Energy), SDG-9 (Industry, Innovation, and Infrastructure), and SDG-13 (Climate Action).
Related Research
Grants with similar aims, by meaning.
Uncovering the secrets of the liquid microlayer: towards next-generation thermal management of extreme power electronics
Boiling in Microchannels: integrated design of closed-loop cooling system for devices operating at high heat fluxes
A Machine Learning Approach to Unsteady Fluid Flow Characteristics in Boiling Approaching Critical Heat Flux
Physics informed deep learning for fusion thermal hydraulics
NANOmaterial-enhanced two-phase COOLing for breakthrough thermal management systems
Original classification
HORIZONPlain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research. Is something wrong? Let us know