An investigation published in Discover Nano demonstrates that a Bayesian-regularized artificial neural network can accurately simulate complex transport dynamics governing magnetohydrodynamic radiative flow of a carbon nanotube-water hybrid nanofluid containing oxytactic and gyrotactic microorganisms over a stretching sheet. Conducted by researchers Ansar Abbas, Shouki A. Ebad, Dhouha Choukaier, Abed Saif Ahmed Alghawli, Fuad Alsarari, Farkhod Alisherov, and Gulnar Hamidova Abdulhamid, the computational study evaluates how Joule heating, viscous dissipation, and porous media alter thermal and concentration fields under Soret and Dufour cross-diffusion phenomena. By training the neural network on synthetic data generated through numerical solution of governing transport equations, the authors established that cross-diffusion mechanisms intensify heat and mass transfer while microbial distributions respond sensitively to advective and diffusive transport ratios.
The physical problem models two-dimensional boundary layer flow along a flat sheet at the horizontal boundary, where coordinate x aligns with the surface and y extends into the porous medium. Within this domain, a hybrid nanofluid containing carbon nanotubes in a water base undergoes motion under an applied transverse magnetic field. The fluid system incorporates living microorganisms, distinguishing oxytactic microbes that steer toward oxygen gradients from gyrotactic microbes that orient relative to fluid shear. In commercial and ecological contexts, oxytactic microbes improve wastewater treatment by migrating into oxygen-rich zones, while gyrotactic microorganisms assist mineral and nutrient transport by aligning with streamlines. The mathematical description accounts for simultaneous conservation of fluid mass, linear momentum, thermal energy, nanoparticle species concentration, gyrotactic cell density, and oxytactic microbe concentration across the boundary layer.
To describe these interacting transport phenomena, the research team constructed six coupled nonlinear partial differential equations reflecting mass continuity, momentum conservation with Darcy-Forchheimer drag and Lorentz forces, thermal energy transport with radiative heat flux and viscous dissipation, nanoparticle mass diffusion, gyrotactic cell conservation, and oxytactic cell conservation. The momentum formulation accounts for hybrid nanofluid density, dynamic viscosity, porous medium permeability, and quadratic Forchheimer inertia resistance. Energy conservation incorporates hybrid suspension thermal conductivity, radiative flux via the Rosseland approximation, internal viscous heating from fluid shear, electrical Joule dissipation from magnetic interaction, and Dufour cross-diffusion energy flux induced by concentration gradients. Nanoparticle transport incorporates Brownian mass diffusivity alongside Soret thermal diffusion flux. The biological equations couple microorganism advection and diffusion with cell swimming parameters and local oxygen concentration variations.
Applying similarity transformations based on a stream function formulation, the investigators converted the partial differential system into coupled ordinary differential equations defined with respect to the dimensionless similarity coordinate. The resulting non-dimensional equations capture flow velocity, dimensionless temperature, scaled nanoparticle concentration, gyrotactic cell density, and oxytactic cell concentration across the boundary layer. Boundary conditions prescribed at the surface impose unit values for dimensionless velocity, temperature, species concentration, and microbial populations, while enforcing zero stream function value at the wall. In the distant ambient region far from the sheet surface, boundary constraints require velocity, temperature elevation, excess concentration, and microbial densities to decay asymptotically to zero, reflecting quiescent ambient reservoir conditions. The transformed mathematical equations decouple spatial coordinates while preserving nonlinear coupling among fluid momentum, temperature gradients, solutal concentrations, and microbial swimming behaviors.
The transformed governing equations incorporate dimensionless parameters that govern momentum, thermal, and biological transport rates. These numbers include the bio-convection Schmidt number representing kinematic viscosity over gyrotactic cell diffusivity, the standard Schmidt number measuring momentum diffusivity relative to nanoparticle mass diffusivity, the Prandtl number defining momentum to thermal diffusion ratios, and the magnetic parameter reflecting Lorentz braking strength. Cross-diffusion effects are governed by the Dufour number, which scales concentration-gradient-driven energy transport, and the Soret number, which quantifies temperature-gradient-driven mass flux. Biological transport is governed by the Peclet number, balancing cell swimming speed against diffusion, and the oxytactic microbe Schmidt number relating kinematic viscosity to oxygen diffusivity. The Eckert number accounts for kinetic dissipation, while the Darcy parameter and Forchheimer coefficient scale substrate permeability and drag.
To capture the enhanced thermal conductivity of carbon nanotube suspensions, the authors implemented a modified Hamilton-Crosser model tailored for cylindrical nanoparticle morphologies. For cylindrical particles like single-walled and multi-walled carbon nanotubes, the shape factor is conventionally taken as six; however, the modified framework introduces an augmented shape influence defined as six plus a correction parameter multiplied by the nanotube length-to-diameter ratio. This formulation computes effective thermal conductivity as a function of constituent nanoparticle and base water conductivities and volumetric fractions. The authors note that integrating carbon nanotubes elevates thermal conductivity compared to conventional base liquids, enabling sophisticated bioreactors and diagnostic systems, while suspended microorganisms simultaneously promote fluid mixing and nutrient transport across the flow domain without requiring mechanical agitation.
The mathematical solution framework combined numerical differential equation solvers with machine learning optimization. The constitutive nonlinear ordinary differential equations were initially integrated using the numerical shooting technique, alongside analytical framing via the Homotopy Analysis Method, which systematically controls convergence without requiring small perturbation parameters. The resulting numerical dataset served as synthetic training and testing data for a multi-layer artificial neural network optimized through Bayesian regularization algorithms. Within this architecture, interconnected neurons distribute information across input, hidden, and output layers through weighted connections adjusted iteratively via back-propagation. Bayesian regularization constrains network weights during learning, mitigating overfitting and ensuring robust convergence across diverse parameter ranges. The model aims to reduce error and achieve accurate approximation results across multiple operational scenarios for the hybrid nanofluid system.
The predictive performance and numerical stability of the Bayesian-regularized neural network were evaluated across five operational scenarios using statistical metrics including mean squared error trajectories, error distribution histograms, regression analysis, transition state indices, and solution fitness plots. Error histograms demonstrated that prediction deviations clustered tightly around zero, confirming minimal divergence between numerical targets and model approximations. Regression analyses across both training and testing datasets achieved regression coefficients of R equals one, indicating near-perfect agreement between neural network estimates and numerical benchmark solutions. Furthermore, tracking mean squared error and gradient norms over advancing training epochs revealed steady, monotonic decreases, verifying stable numerical convergence and strong generalization capabilities on unobserved parameter configurations. The neural network architecture provided a precise computational structure for simulating complex heat transport processes.
Parametric evaluations detailed the direct physical influence of nanoparticle volume fraction on velocity distributions and thermal transport. As the volume fraction of carbon nanotubes increased, the velocity profile and overall flow intensity of the hybrid nanofluid decreased, accompanied by an increase in local heat transmission rates and wall thermal gradients. The researchers explained this deceleration by noting that base liquid density is lower than the density associated with suspended nanoparticles, which increases inertial resistance and causes the moving fluid layer to slow down. Concurrently, elevated nanoparticle loading enhances effective thermal conductivity, facilitating more rapid thermal energy transfer away from the bounding surface despite diminished bulk flow velocity.
The investigation demonstrated that thermo-diffusion and diffusion-thermo cross-effects exert substantial control over boundary layer structures. Increasing the Dufour number produced pronounced elevations in both local fluid temperature and thermal boundary layer thickness, as concentration gradients contributed directly to internal thermal energy generation and intensified convective dynamics within the thermal layer. In parallel, increasing the Soret number enhanced molar mass diffusivity, leading to substantial increases in nanoparticle concentration distributions and elevated Sherwood numbers. When Soret and Dufour influences were intensified simultaneously, overall rates of mass and heat transfer across the sheet rose markedly, establishing that cross-diffusion mechanisms play a decisive role in regulating simultaneous energy and species transport in hybrid nanoliquid systems.
Analysis of biological profiles revealed that the spatial distribution of motile microorganisms is governed by the interplay between advective cell transport and diffusive spreading. As the Peclet number increased, the density profile of motile microorganisms declined significantly while the rate of advective transport rose sharply, reflecting the sweep of cells downstream by dominant fluid motion. Similarly, increasing the bio-convection Schmidt number and the oxytactic microbe Schmidt number caused steady reductions in microbial concentration profiles. These findings confirmed that higher Schmidt numbers, which reflect lower cellular or oxygen diffusion relative to fluid momentum, constrain the penetration depth of oxytactic and gyrotactic bacteria across the boundary layer.
The numerical and neural network predictions were validated against previously published benchmark studies by adjusting auxiliary model parameters to zero, demonstrating close quantitative agreement with established Nusselt number literature values across multiple Prandtl numbers. Contextualizing their results within technological applications, the authors discussed how oxytactic microorganisms assist wastewater treatment and pollutant degradation by migrating toward oxygen-rich zones, while gyrotactic microbes enhance fluid mixing in catalytic bioreactors, chip-based bio-microsystems, and biosensors.
Subject of Research: Bayesian neural network modeling of radiative hybrid nanofluid bioconvection
Article Title: Bayesian optimized neural network modeling for radiative flow of MHD hybrid nanoliquid with oxytactic microbes for biomedical applications
Article References: Abbas, A., A. Ebad, S., Choukaier, D., Alghawli, A. S. A., Alsarari, F., Alisherov, F., & Abdulhamid, G. H. (2026). Bayesian optimized neural network modeling for radiative flow of MHD hybrid nanoliquid with oxytactic microbes for biomedical applications. Discover Nano, 21(1), Article 581. https://doi.org/10.1186/s11671-026-05065-9
Image Credits: AI Generated
DOI: 10.1186/s11671-026-05065-9
Keywords: hybrid nanofluid, carbon nanotubes, oxytactic microbes, gyrotactic microorganisms, Bayesian regularization, artificial neural networks, bioconvection, magnetohydrodynamics, Joule heating, viscous dissipation, Soret effect, Dufour effect, Darcy-Forchheimer flow, Peclet number, thermal conductivity, Hamilton-Crosser model, mean squared error, shooting technique
News Source: Cassandra Pierce. (November 14, 2026). Bayesian Neural Network Models Radiative Hybrid Nanofluid Flow with Oxytactic Microbes. Scienmag.



