This study aims to develop an efficient Physics-Informed Neural Network (PINN) model for simulating the convective heat transfer of a viscous ternary hybrid nanofluid (THNF) flow over a wavy inclined surface on a porous medium. The suggested method gets around the shortcomings of classic numerical solvers that compute nonlinear heat transfer in complex geometries that are frequently found in heat exchangers, solar collectors and cooling systems.
Continuity, momentum and energy governing equations are obtained under Boussinesq approximation of two THNFs of water-Cu-Al2O3-TiO2 and ethylene glycol-Cu-MWCNT-graphene oxide. It considers nondimensional parameters such as the Rayleigh number, inclination of the surface, wavy amplitude and the volume fractions of the nanoparticles. PINN is intended to use automatic differentiation of its loss, to impose physical law and boundary conditions to provide mesh-free approximation, modeled by a feed-forward and hyperbolic tangent activation neural network. The process of optimization is carried out by minimizing residual losses with the Adam optimizer, and validation of the model shows the variation of less than compared to benchmark data in Nusselt number.
It has been found that when wavy surfaces are large, convective heat transfer is reduced due to the restriction in fluid motions, and when the incline angles are high, heat transmission occurs through buoyancy forces. Increasing the concentration of nanoparticles is significantly effective in increasing the heat transfer. The thermal performance of Tnf1 is always better than that of Tnf2, and it is reasonable to note that it is suitable in high performing thermal management systems.
PINNs is developed and used to compute the solutions of differential equations. Heat transfer enhancement due to two optimal sets of ternary hybrids nanofluid flow over an inclined wavy surface is investigate.
