Prediction of double diffusive mixed convection in a partially heated and concentrated polygonal enclosure using artificial neural network and hybrid nanofluid

Abstract

The objective of the present study is to perform a numerical analysis of double-diffusive mixed convection within a polygonal cavity subjected to lid-driven motion. The cavity, partially heated and concentrated, is filled with a hybrid nanofluid composed of Al2O3 (50%) and Cu (50%) nanoparticles suspended in water. The lower middle portion of the cavity is heated, while the upper horizontal boundaries are maintained at a low temperature. The upper left wall moves at a constant velocity in the positive direction, whereas the other walls are thermally insulated. The physical problem is mathematically represented by a set of governing equations along with appropriate boundary conditions. Using a class of suitable transformations, the governing equations and boundary conditions are converted into a non-dimensional form, which are then solved using a finite element-based Galerkin weighted residual method. An artificial neural network (ANN) model is also developed using simulation data obtained from the numerical solution to predict different performance parameters within the present framework. The study investigates the effects of dimensionless parameters such as Richardson number (Ri), Reynolds number (Re), Lewis number (Le), Buoyancy ratio (Br), and nanoparticle volume fraction (ϕ), with a constant Prandtl number (Pr = 6.8377). The results are presented in terms of flow patterns, temperature distributions, solute distributions, average Nusselt number, and average Sherwood number. It is observed that the average Nusselt number increases with increasing Ri, Re, Br, and ϕ, but decreases with increasing Le. Additionally, all parameters show an upward trend in the average mass transfer rate. Specifically, the average heat transfer rate increases by 37.42% when Ri increases from 0.01 to 10, and decreases by 11.25% when Le increases from 0 to 5. Using the simulation data, an innovative ANN model is developed for accurate prediction. For training and validation data, the model predicts the average Nusselt number with 99.57% accuracy and the average Sherwood number with 99.63% accuracy. For test data, the model predicts the average Nusselt number with 99.34% accuracy and the average Sherwood number with 99.49% accuracy. Therefore, the proposed ANN model accurately predicts the responses for the current framework.