This study aims to investigate the nonlinear coupled boundary-layer flow of a magnetohydrodynamic (MHD) non-Newtonian Eyring–Powell nanofluid over a porous cylinder using Physics-Informed Neural Networks (PINNs).
The governing nonlinear partial differential equations associated with momentum and heat transfer are transformed into a non-dimensionless form using appropriate transformations. The proposed PINNs framework incorporates governing physical laws directly into the learning process to achieve accurate and computationally efficient predictions of nonlinear heat transfer behavior. A deep PINN framework is then constructed in TensorFlow to solve the resulting equations while satisfying the imposed boundary conditions. The neural architecture consists of eight hidden layers with 156 neurons in each layer and is trained using a learning rate. The predictive performance of the developed PINN model is validated against the local non-similarity (LNS) method through comparisons.
The proposed PINN model demonstrates strong predictive accuracy and successfully captures the nonlinear transport behavior of MHD Eyring–Powell nanofluid flow. The numerical results exhibit excellent agreement with those obtained from the LNS method and established theoretical trends. The trained network accurately reproduces the thermal boundary conditions and heat transfer characteristics governed by the Prandtl number. Moreover, this study reveals that increasing the Eyring–Powell fluid material parameter significantly enhances the boundary layer thickness. This work also examines the effectiveness of PINNs in predicting heat transport characteristics and boundary layer behavior in comparison with the conventional LNS and finite difference method. This study confirming the robustness of the PINN framework as a reliable alternative for solving complex non-Newtonian transport phenomena in fluid flow and thermal engineering applications.
PINN-based modelling of MHD Eyring–Powell non-Newtonian boundary layer flow with heat transfer over horizontal cylindrical surfaces is still scarce in the literature.
