Physics-informed deep learning of radiative MHD bioconvection in non-Newtonian hexa-hybrid nanofluids
Year: 2026
Type: article
Abstract: Purpose This examination explores the bioconvection flow of Casson-Williamson hexa-hybrid nanofluid (HEHNF) (Al₂O₃–CuO–SiO₂–ZnO–Fe₃O₄–TiO₂/H₂O) via an exponential stretchable, permeable vertical surface in a porous material. The model includes the effects of a magnetic field (MF), thermal buoyancy, thermal radiation, Brownian, thermophoretic diffusion, heat generation, viscous dissipation, Joule heating, chemical reaction, and gyrotactic motile microorganisms. Design/methodology/approach Similarity conversions reduce the managing equations to coupled non-linear ordinary differential equations (ODEs), which are solved with the MATLAB bvp4c routine and validated against published results for the limiting Williamson case. An artificial neural network trained with the Levenberg–Marquardt algorithm (ANN-LMA) and physics-informed neural networks (PINNs), which combines bvp4c data with equation and boundary residuals evaluated by automatic differentiation, are developed to predict the wall quantities. Findings The velocity declines with the magnetic, porosity, Casson, and Williamson parameters and increases with the Grashof number, while radiation, Brownian motion, thermophoresis, heat generation, and the Eckert number raise the temperature. Radiation weakens the wall temperature gradient, and the rise in the Nusselt number is due to the additional radiative heat flux. The concentration diminishes with the Schmidt number and the reaction parameter, and the microorganism density declines with the bioconvection Lewis and Peclet numbers, while the motile density number raises. The bvp4c results agree with published data to four decimal places, and in the limiting Williamson case the ANN-LMA and PINN skin friction coefficient (SFC) predictions differ from bvp4c by less than 1 0 ⁻ ³ . Originality/value The study presents a Casson–Williamson HEHNF bioconvective model with a combined numerical and machine-learning framework, in which the trained models act as fast substitutes for the solver in thermal and bioconvective transport applications.
Source: Next Nanotechnology
Institutions: Islamic University of Madinah, Mehran University of Engineering & Technology Shaheed Zulfiqar Ali Bhutto Campus, Khairpur Mir's, Bahauddin Zakariya University, National University of Sciences and Technology, The Women University Multan, Ankara Bilim Üniversitesi, Biruni University, Umm al-Qura University, Benazir Bhutto Shaheed University Lyari, Karadeniz Technical University, Korea University, VIZJA University, Near East University
Language: English
Subfield: Fluid Flow and Transfer Processes
Field: Chemical Engineering
Domain: Physical Sciences
Keywords: bioconvection, magnetohydrodynamics, porous media, thermal radiation, gyrotactic microorganisms, heat generation, Joule heating, chemical reactions, skin friction coefficient, Nusselt number, mass transfer, artificial neural networks, physics-informed neural networks, Levenberg-Marquardt algorithm
Open Access status: gold
Estimated APC paid: $1,280
OpenAlex ID: W7236767125
