Multi-objective optimization of cylindrically arranged triply periodic minimal surface (TPMS) porous structures for enhanced thermal and hydraulic performance in microscale thermal management

This study presents a multi-objective optimization framework for cylindrically arranged Triply Periodic Minimal Surface (TPMS) porous structures, aimed at enhancing thermal and hydraulic performance in microscale heat sinks for electronic cooling. Integrating conjugate heat transfer simulations in COMSOL Multiphysics with genetic algorithms (GA) and machine learning (ML) surrogates in MATLAB, the approach optimizes seven geometric parameters including cell periodicity, inner radius, and radial/axial level-set gradients across four TPMS types (gyroid, diamond, Neovius, primitive) to maximize volumetric Nusselt number (𝑁𝑢𝑣) while minimizing friction factor (f) at a constant Reynolds number of 500 and heat flux of 300 W/cm². Direct GA optimization over 30 generations yielded Pareto fronts dominated by gyroid structures, achieving up to 300% 𝑁𝑢𝑣 improvement with controlled f increases, favoring porosities of 0.50–0.60, inner radii of 0.16–0.18 mm, positive axial gradients (0.3–0.8), and high axial/angular cell counts with minimal radial partitioning. Gaussian Process Regression ML models (R² > 0.85) trained on GA data enabled an indirect GA method, reducing computation time from 400 to 30 hours while preserving solution quality. Feature ranking identified porosity, TPMS surface area, and periodicity as key drivers. CFD analyses confirmed superior flow uniformity and heat flux distribution in optimal designs, with gyroids excelling at high porosities via enhanced connectivity and diamonds at lower porosities through inward-concentrated convection. The cylindrical arrangement outperformed traditional cuboidal lattices in flow equalization and hotspot reduction which advanced TPMS-based thermal management.