Integrating machine learning with numerical simulations of magneto-convective flow and irreversibility dynamics in a partially porous cavity with hot blocks

This article explores the integration of machine learning (ML) with computational fluid dynamics (CFD) to enhance thermal efficiency in a structurally intricate domain subjected to magnetic force. Unlike previous works that primarily address thermal dissipation in uniformly structured cavities under different constraints, this research investigates the flow and irreversibility dynamics within a geometrically complex cavity containing ternary hybrid nanofluid, partially porous layer, and heated obstacles of varying sizes and positions − a configuration highly relevant to battery cooling applications. Numerical simulations revealed that increasing the porous layer thickness from 0.45 to 0.65 impedes fluid motion, resulting in 1.05% reduction in thermal dissipation. While this challenge can be mitigated by tuning other physical parameters, such optimization is computationally intensive when relying solely on traditional simulation technique. To address this, we developed a Gradient Boosting Regressor-based ML model achieving the predictive accuracy with R 2 of 0.998 and 0.992 for average Nusselt number and total entropy generation, respectively. This model enables rapid estimation of thermal dissipation and entropy generation across a wide parametric space and identified an optimal parameter set ( H a = 1 . 5 , D a = 1 0 − 1 . 201 , δ = 0 . 1 , ɛ = 0 . 54 , n = 3 ) that improves thermal dissipation under the Top-Left,Top-Right (TL-TR) hot block configuration. Additionally, optimal parameter sets were determined for other hot block configurations. Sensitivity analysis revealed that hot block position greatly influences the average Nusselt number, while nanoparticle shape significantly affects total entropy generation. Overall, this work demonstrates that coupling ML with CFD offers a powerful and efficient framework for optimizing thermal behavior in complex porous geometries, paving the way for scalable solutions in next generation battery thermal management.