Hybrid finite element and machine learning modeling of natural convection, thermal energy enhancement, and entropy analysis in a ternary nanofluid-filled octagonal enclosure

The octagonal enclosure geometry represents practical engineering configurations in enclosed heat transfer devices, electronic equipment, solar thermal collectors, battery thermal management systems, nuclear reactor cooling, and compact heat exchangers. This work explores naturally convective ternary nanoliquid flow and irreversible processes through an octagonal cavity with a centrally heated circular obstacle. The walls of the cavity on the horizontal and vertical sides were cold, whereas the inclined walls are adiabatic. Formulation of problem is provided via dimensional differential setup in (PDEs) with the implementation of Boussinesq approximations, which are then converted into a non-dimensional form via a new set of variables. Numerical computations are performed via the COMSOL Multiphysics package based on finite element simulations. A machine learning approach consisted of feedforward backpropagation and the Levenberg-Marquardt technique is also employed to predict the outcomes of the FEM-based simulations for engineering quantities. The results revealed that the convection phenomenon is dominated by elevation (Ra), leading to higher kinetic energy and heat flux coefficient. However, intensification of the Hartmann number suppresses the fluid motion due to the generation of the Lorentz force, which causes convection regime to depreciate. ANN predictions showed high accuracy with minimal deviation, particularly in convection-dominated regimes and the nanofluid case, assuring the robustness of the hybrid FEM-ANN framework for nonlinear thermofluidic prediction.