MHD natural convection and entropy generation of a CuO–Al₂O₃–Ag/water ternary hybrid nanofluid in a zigzag-corrugated porous cavity with ANN-based prediction

This is the authors' abstract. We don't add key points for this paper.

The heat and thermodynamic performance of magnetoconvective heat transfer in porous enclosures can be greatly influenced by the interactions between thermal damping by a magnetic field, porous resistance, corrugation of the wall, and loading with nanoparticles, which makes fast prediction difficult. The aim of this research is to investigate steady two-dimensional MHD natural convection and dimensionless entropy generation of CuO–Al₂O₃–Ag/water ternary hybrid nanofluid inside a porous rectangular cavity containing a zigzag heated wall. The coupled governing equations are solved numerically by the Galerkin finite element method (GFEM), and an Artificial Neural Network (ANN) is built using the Levenberg – Marquardt algorithm with the parameters of the averaged Nusselt number (Nu Avg ) and total entropy generation (Si Total ). The effects of (Ra = 10 3 – 10 5 ), (Ha = 0 – 50), (Da = 10 −4 – 10 −2 ), ( ϕ = 0 – 0.05), and (NC = 0 – 16) are examined. Increasing (Ra) from (10 3 ) to (10 5 ) raises (Nu Avg ) by 59.1% for the smooth cavity and 68.0% for the zigzag cavity, while (Si Total ) increases from 2.3152 to 8.6573 and from 2.5152 to 9.2726, respectively. With a smooth configuration, increasing (Ha) from 0 to 50 reduces (Nu Avg ) by 85.1%, and for the zigzag configuration, it reduces by 82.9%. Increasing ( ϕ ) from 0 to 0.05 produces a (Nu Avg ) increase of 9.2% for the smooth cavity and 86.6% for the zigzag cavity, whereas increasing (Da) from (10 −4 ) to (10 −2 ) increases (Nu Avg ) by 6.0% and 7.8%, respectively. The ANN provides a good reproduction of the GFEM database, both the training set and the independent testing set R 2 = 0.99998, and an overall value of R = 0.99998. Results show that the integrated GFEM–ANN system can be used to perform a physical analysis with high fidelity and provide a computationally efficient surrogate for the thermal and thermodynamic responses over the explored parameter range.