Machine Learning Prediction of Skin Friction and Heat Transfer in Walter-B Ternary Nanofluid Flow with Cross-Diffusion and Entropy Generation

This study investigates the steady magnetohydrodynamic flow of the Walter-B ternary nanofluid (composed of water-ethylene glycol (WEG) base fluid with graphene, alumina, and titanium dioxide nanoparticles) over a nonlinear stretching sheet, incorporating the effects of cross-diffusion, couple stress, and viscous dissipation. Using similarity transformations, the governing equations are converted to ordinary differential equations and solved numerically with MATLAB's bvp4c solver. A Bayesian-regularized artificial neural network (BRANN) is developed to predict skin friction, Nusselt, and Sherwood numbers with R² > 0.99 accuracy. Results reveal that fluid velocity decreases with increasing couple stress but enhances with the Deborah number and Darcy parameter, while temperature rises with the Eckert and Dufour numbers. Concentration profiles decline with chemical reaction but grow with the Soret number. Entropy generation intensifies with Brinkman and Biot numbers, whereas the Bejan number shows opposite behavior. Empirical correlations for skin friction, Nusselt, and Sherwood numbers are developed, showing a 6.3% rise in skin friction with the Forchheimer number and a 13.14% improvement in heat transfer with thermal radiation. This work provides critical insights for thermal management systems, leveraging machine learning to optimize ternary nanofluid flows in porous media under cross-diffusion effects.