Electrohydrodynamic-induced micropolar gravitational convection of dielectric nano-scaled phase change materials with heat source: Numerical and machine learning approaches

Advanced thermal systems in usages like electronic cooling, solar energy harvesting, and latent heat storage demand enhanced control over heat transfer processes. This study explores the electrohydrodynamic (EHD)-induced micropolar natural convection of a dielectric suspension containing nano-encapsulated phase change materials (NEPCMs) within a two-dimensional enclosure. The enclosure features curved heaters at the corners, functioning as homogeneous charge-emitting electrodes, and three internal cold channels, which act as collector electrodes. A uniform volumetric heat source is applied throughout the domain. The governing equations, incorporating micropolar fluid theory, electric body forces, and phase change heat transfer, are solved numerically using the finite element method (FEM). Key dimensionless parameters, including the Rayleigh number, Eckert number, Stefan number, electric field number, charge diffusivity, vortex viscosity, NEPCM concentration, and fusion temperature, are systematically varied to assess their influence on fluid flow, thermal behavior, electric charge transport, and entropy generation. Quantitatively, micropolar effects suppress convection, reducing the average Nusselt number by approximately 23% as the vortex viscosity parameter increases, whereas buoyancy remains dominant, with the Rayleigh number rising from 105 to 106 enhancing heat transfer by about 56%. Viscous dissipation further strengthens convection, producing nearly a 32% increase in heat transfer at the highest Eckert number, while the electric field parameter yields ∼5.8% improvement. Increasing the NEPCM concentration to 4% enhances heat transfer by ∼7%, and lower Stefan numbers associated with higher latent heat provide the most effective thermal regulation. Additionally, a supervised machine learning model based on an artificial neural network (ANN) with a multi-layer perceptron (MLP) architecture is trained using numerical data to predict the electric charge density, temperature, microrotation, and stream function along the cavity's vertical midline. The ANN demonstrates excellent predictive performance, with regression coefficients exceeding 0.997 and a minimal mean squared error (MSE), confirming its reliability as a surrogate model.