Numerical and AI-based simulation of entropy generation in MHD Jeffery tetra nanofluid flow over an inclined cylinder for energy engineering applications

The increasing need for high-performance thermal management systems in energy, medicinal, and industrial sectors has prompted researchers to investigate improved nanofluids with superior heat transfer properties. Tetra-hybrid nanofluids, created by distributing four unique nanoparticles in a base fluid, have enhanced thermophysical characteristics relative to conventional, hybrid, and ternary nanofluids. Nonetheless, forecasting their flow and thermal characteristics is difficult because of the nonlinear interdependence of several elements, including nanoparticle concentration, viscoelastic effects, mixed convection, and viscous dissipation. An axisymmetric two-dimensional MHD Jeffery tetra-hybrid nanofluid flow across an inclined stretched cylinder contained in a porous medium is studied in this paper to address this. A hybrid computational methodology is used, integrating numerical solutions derived from the Lobatto IIIA technique using MATLAB’s bvp4c solver with an Intelligent Bayesian Regularization Neural Network (IBR-ANN) for prediction and validation purposes. Critical physical quantities of interest, such as velocity, temperature, skin friction, Nusselt number, and entropy generation, are examined in relation to significant parameters including nanoparticle volume fraction, Deborah number, retardation parameter, mixed convection, Eckert number, curvature parameter, and thermal radiation. Based on the findings, it is shown that larger concentrations of nanoparticles increase skin friction and decrease the Nusselt number. On the other hand, viscoelastic effects cause increased retardation, and Deborah numbers substantially change the velocity and temperature fields. The examination of entropy creation emphasizes the roles of viscous dissipation, radiation, and fluid memory in system irreversibilities. The IBR-ANN model has excellent prediction accuracy, as shown by error histograms, regression plots, function-fit curves, performance plots, and training state charts. There are very few mistakes, and the predicted and actual data are quite similar. A comparison analysis demonstrates strong concordance between the results of this investigation and the existing literature, affirming the dependability and precision of the findings.