Thermal management of electronic chips using nano-encapsulated phase change materials in a rotating-cylinder H-shaped cavity: A hybrid numerical and machine learning study

Thermal management of high-power electronic chips is decisive for the reliability and lifespan of modern computing devices, driving demand for advanced cooling architectures. Passive nanofluid-based cooling and active convective control have each shown promise, but their combined exploitation within compact chip-scale geometries remains poorly quantified, particularly when latent heat storage, magnetic damping, and mechanical agitation act simultaneously. The role of these coupled mechanisms in governing heat and mass transfer, phase-change dynamics, and thermodynamic irreversibility inside a chip-representative enclosure has not been systematically resolved. Here we show, through a hybrid numerical and machine learning framework, that integrating nano-encapsulated phase change materials, a rotating cylinder, and a magnetic field within an H-shaped cavity enables tunable and predictable thermal performance for chip cooling. The coupled continuity, momentum, energy, and species equations are solved using the Galerkin finite element method across Reynolds number (10–100), Richardson number (0.1–10), Hartmann number (0–80), Lewis number (0.1–10), buoyancy ratio (1–5), Stefan number (0.1–0.9), fusion temperature (0.1–0.9), and nanoparticle volume fraction (0.01–0.035). Raising the Reynolds number from 10 to 100 enhances the mean Nusselt and Sherwood numbers by 179% and 166%, respectively, while increasing the Hartmann number from 5 to 80 suppresses them by 41% and 35% and reduces total entropy generation by 58%. Nanoparticle addition raises the Nusselt number by 34% with negligible impact on species transport. Three surrogate models — XGBoost, Random Forest, and Multilayer Perceptron — trained on the dataset achieve coefficients of determination exceeding 0.997, providing a rapid interpolative design tool for chip-scale cooling modules within the sampled parameter ranges.