Levenberg–Marquardt artificial neural network surrogate modeling of magneto-thermal hybrid nanofluid flow over a vertical plate

Hybrid nanofluids have become promising working fluids for advanced thermal management systems due to their greater heat transfer properties than conventional fluids and mono-nanofluids. In the current study, the magneto-thermal properties of an Al₂O₃–Cu/H₂O hybrid nanofluid flowing over a vertically stretched surface are investigated by considering the effects of mixed convection, thermal radiation, magnetic field, and the effects of suction/injection. The governing boundary-layer equations are converted to a system of nonlinear ordinary differential equations by using similarity transformations, and the system is solved numerically by applying the Lobatto IIIA collocation method. The accurate numerical dataset is then used to develop a Levenberg–Marquardt artificial neural network (LM-ANN) model to predict the velocity and temperature distributions rapidly. Five different physical cases are considered in which the variation of the magnetic parameter, mixed convection parameter, stretching parameter, suction/injection parameter, radiation parameter, and Prandtl number is applied. The developed ANN model shows a great prediction capability and has the corresponding mean square errors of 10⁻¹ ⁰ and the regression coefficients are very close to unity. It is revealed that increasing the strength of the magnetic field restricts the movement of the fluid by applying the Lorentz force whereas the influence of thermal radiation increases the thickness of the thermal boundary layer. Higher Prandtl numbers diminish thermal diffusion thereby decreasing temperature profiles. Besides, hybrid nanofluids perform better thermally than traditional nanofluids due to the combined influence of Al₂O₃ and Cu nanoparticles. The most important innovation of the study is the creation of a powerful LM-ANN surrogate framework for effective magneto-thermal hybrid nanofluid transport prediction working under the action of different interacting physical processes. The proposed method is able to dramatically cut the computing expenses keeping accuracy high enough for implementation in the area of real-time thermal control, intelligent cooling technologies, electronic heat dissipation technologies, and engineering design optimization systems.