Hybrid machine learning and soft computing methods for predicting and optimizing natural convection in a C-shaped cavity with double diffusion effects

Ebrahem A. Algehyne, Marwa M. Alzubaidi, Tareq Saeed, Wafa F. Alfwzan, Syed Ibrahim, Muhammad Ibrahim, Shahid Ali Khan, Vakkar Ali,
Hybrid machine learning and soft computing methods for predicting and optimizing natural convection in a C-shaped cavity with double diffusion effects,
International Communications in Heat and Mass Transfer,
Volume 169, Part D,
2025,
109855,
ISSN 0735-1933,
https://doi.org/10.1016/j.icheatmasstransfer.2025.109855.
(https://www.sciencedirect.com/science/article/pii/S0735193325012813)
Abstract: In this study, heat and mass transfer in a C-shaped cavity filled with nanofluid, considering Soret and Dufour effects, was investigated using machine learning models to predict the Nusselt and Sherwood numbers across the design space and to guide parameter optimization. We evaluated Advanced Multilayer Perceptron (MLP), Response Surface Methodology (RSM) via Polynomial Regression, Random Forest, XGBoost, Gaussian Process, and Support Vector Regression (SVR) for predicting Nusselt (Nu) and Sherwood (Sh) numbers across symmetric and asymmetric cavity configurations (CU, CS, CD). The impact of Lewis number (Le) ranging from 1.0 to 7.5 and aspect ratio (AR) between 0.25 and 0.75 on flow patterns and Heat Transfer Rate (HTR) rates was analyzed, with key parameters optimized using the best models. Random Forest for Nu (MSE = 0.0008) and XGBoost for Sh (MSE = 0.0044). Numerical solutions were obtained via the finite element method, while optimization employed machine learning predictions. Results showed that varying Le from 1 to 7.5 and AR from 0.75 to 0.25 increased Sh by up to 148.8 % and altered Nu by 41.2 % in asymmetric cavities. Optimization identified optimal parameters for maximum Nu (5.6517) at Le = 1.0478, AR = 0.7487, Cavity = 1, and for maximum Sh (11.9269) at Le = 7.4807, AR = 0.7235, Cavity = 2, with a combined optimization yielding Nu = 5.1668 and Sh = 11.8960 at Le = 7.4381, AR = 0.7367, Cavity = 2, demonstrating the efficacy of machine learning in optimizing thermal performance.
Keywords: Nanofluid; Soret and Dufour phenomena; Machine learning; Concentration; Double-diffusive convection; Predictions