Artificial neural network prediction of an electrohydrodynamic thermosolutal buoyancy-driven convection of NEPCMs-dielectric suspension within an oblique enclosure with active blocks

Abstract

Studying double-diffusion natural convection in Nano-Encapsulated Phase Change Materials under Electro-Hydro-Dynamics is crucial for enhancing thermal management across various applications. This technology has the potential to greatly enhance cooling in electronics, electric vehicle batteries, and photovoltaic panels, as well as contribute to energy-efficient building designs, solar heating, and desalination. Nano-Encapsulated Phase Change Materials also hold promise in thermal energy storage systems, where they can absorb, store, and release energy, aiding in grid stability and renewable energy integration. This study investigates the thermosolutal natural convection of Nano-Encapsulated Phase Change Materials suspension within an oblique enclosure with differently heated and salted blocks under the influence of Electro-Hydro-Dynamics. To obtain the solution of the governing equations, the finite element method was utilized. Moreover, an Artificial Neural Network is employed to model and predict some important physical quantities within the system, providing an advanced tool for optimizing performance. The findings reveal key influences of various parameters on heat and mass transfer efficiency. Increasing the Eckert number (Ec) causes about a 13.2 % decrease in mean Nusselt and a 2.0 % decrease in mean Sherwood, while raising the Lorentz force number (S_E) from 0.1 to 5 results in a 4.8 % reduction in mean Nusselt and a 2.2 % reduction in mean Sherwood. The diffusion number (De) has a secondary effect, with an increase from 0.25 to 0.75 producing a 1.5 % rise in mean Nusselt but only a 0.3 % decrease in mean Sherwood. Adding 3 % of Nano-Encapsulated Phase Change Materials concentration (ϕ) at a lower Stefan number (Ste) enhances heat transfer by 4.4 % while reducing mass transfer by 2.3 %. In addition, Artificial Neural Network analyses show plausible training with minutest errors and the best-fit model for the suggested factors. These results underscore the benefits of employing machine learning techniques for both scientific inquiry and engineering applications.