Machine learning-enhanced nusselt number prediction for thermal convection around a cylinder in porous media
Year: 2026
Type: article
Abstract: This study develops an integrated numerical and data-driven framework for predicting heat-transfer characteristics of natural convection around a horizontal cylinder embedded in an infinite porous medium under local thermal non-equilibrium conditions. The coupled momentum and fluid-solid energy equations are formulated using the Brinkman-Forchheimer-extended Darcy model and solved using the second-order Keller-box method. After validation against benchmark data, the numerical solver is used to characterize the effects of the Rayleigh number, Biot number, conductivity ratio, porosity, Prandtl number, and cylinder-to-particle diameter ratio on the fluid- and solid-phase Nusselt responses. Deep neural network (DNN), multilayer perceptron (MLP), and Random Forest regression models are then evaluated as computationally efficient surrogates of the numerical solution. Among the evaluated models, the DNN yields the lowest prediction error for both fluid and solid thermal responses. Mutual-information-based feature assessment and particle swarm optimization are additionally used to examine parameter relevance and derive explicit predictive correlations. The contribution of this work lies in the integration of a validated LTNE porous-media solver, dual-phase surrogate prediction, parameter-relevance assessment, and correlation development within a single engineering workflow, rather than in proposing a new machine-learning algorithm. The resulting framework enables rapid estimation of the investigated thermal responses within the considered parameter ranges.
Source: Journal of Engineering Research
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,100
Award: G7300154059
OpenAlex ID: W7203636443