Multi-layer neural network on 3D transition metal hydrides-based ternary nanofluid flow with entropy: Application to spinning disk reactor

Abstract

Artificial Neural Networks provide effective solutions to simulate and optimize complicated fluid systems, especially thermal management systems. This research examines the flow and heat transfer behavior of a ternary hybrid nanofluid made up of Titanium hydride $\left(\mathrm{Ti}{H}_2\right)$, Lanthanum hydride $\left(\mathrm{La}{H}_3\right)$, and Zirconium hydride $\left(\mathrm{Zr}{H}_2\right)$ nanoparticles dispersed in dihydrogen oxide in a spinning disk reactor. The governing equations, formulated in terms of conservation laws, are transformed to dimensionless ordinary differential equations by similarity transformations and solved numerically using MATLAB’s bvp4c solver. Artificial Neural Networks models are then used to capture nonlinear interactions and enhance prediction accuracy for important parameters such as skin friction and Nusselt number. The study shows that radial velocity decreases with higher Casson parameter, porosity and 2nd order radial slip, while azimuthal velocity increases for Casson parameter and 1st order radial slip but decreases for porosity. Temperature rises with thermal radiation and Brinkmann number but drops with Reynold number. Casson parameter and stretching parameter enhance skin friction but reduce the Nusselt number. Entropy increases with Brinkmann number but decreases with Casson Parameter. ANN models for first order tangential slip, porosity, and Casson parameter achieved $R\approx 1$, with minimal errors, ensuring reliability. First order tangential slip and porosity stabilized after more epochs, while Casson Parameter converged faster. MSE analysis confirmed accurate predictions, validating the model’s effectiveness.