Thermal performance prediction and optimization of magneto-thermal radiative heat transfer in ternary-hybrid nanofluids via deep learning and sensitivity analysis

The present work is concerned with an analysis of heat transfer through the magneto-thermal convective transport of ternary hybrid nanofluids filled inside a trapezoidal enclosure with a heated circular cylinder. Ternary-hybrid nanofluids improve the thermal efficiency of heating, ventilation, and air conditioning systems by optimizing the heat transfer mechanism in heat exchangers. The finite element method and the Newton–Pardiso solver in COMSOL Multiphysics 6.2 are used to numerically solve the governing equations. The two main objectives of this research are (i) predictive analysis of the average heat transfer rate utilizing a Levenberg–Marquardt Artificial Neural Network (LMS-ANN) and (ii) performing sensitivity analysis through response surface methodology (RSM) and analysis of variance (ANOVA) to identify the most dominant parameters. A gradient descent optimizer and a tangent sigmoid activation function are used in the LMS-ANN. The results reveal that the RSM model achieved high accuracy performance, with R 2 and predicted R 2 at 99.07% and 97.38%, respectively. Sensitivity analysis shows that the Rayleigh number is the most dominant parameter that controls the average Nusselt number. The LMS-ANN model attained high accuracy with mean squared errors lying between 10 -7 and 10 -13 , and regression plots for all dimensionless parameters are unity, which shows that the predictive model is reliable. The integration of ternary nanofluids with data-driven machine learning models and sensitivity analysis provides a robust platform for improving the thermal performance in engineering systems.