Ellahi’s PINN approach embedded with nonsimilarity features for MHD Casson fluid over a porous stretching cylinder under thermal radiation

Purpose This study aims to introduce a local nonsimilarity feature–embedded physics-informed neural network (PINN) framework to model the thermal boundary layer flow of a non-Newtonian Casson fluid under the combined effects of viscous dissipation, magnetic field and radiative heat transfer over a porous stretching cylinder, while accounting for buoyancy effects. Design/methodology/approach The governing nonlinear partial differential equations are transformed into a nonsimilar form through appropriate transformations, and the solutions are approximated using a PINN trained with the Adam optimizer in Python. The network architecture consists of eight hidden layers with 64 neurons each. The accuracy of the proposed model is validated against the local nonsimilarity method, and its performance is further tested on independent data sets. The results show excellent agreement between the PINN predictions and reference solutions, confirming strong generalization capability and consistency with established theoretical models. Findings The analysis reveals that the Casson fluid parameter plays a crucial role in influencing both the rheological behavior and thermal characteristics of the flow. The findings demonstrate the efficiency and versatility of the PINN framework in solving nonlinear, high-dimensional boundary-layer problems, offering a robust and data-efficient alternative to conventional numerical methods for non-Newtonian fluid dynamics. Originality/value To the best of the authors’ knowledge, this study is among the first to embed local nonsimilarity formulations within a PINN for Casson fluid flow over a porous stretching cylinder, improving accuracy, stability and generalization.