A neural network–integrated computational scheme for thermal transport predictions in MHD-driven viscous ternary flow through convergent-divergent systems

Darvesh A.; Collantes Santisteban L.J.; El-Rahman M.A.; Al Garellah H.; Jastaneyah Z.; Khan S. · Aerospace Science and Technology (2026) · DOI: 10.1016/j.ast.2026.112687

Resumen

Convergent-divergent channels offer a desirable design option to enhance the efficiency of flow control and thrust generation in the current state of the art aerospace propulsion systems, and thermal transport processes. This paper introduces a new artificial intelligence based computational framework to simulate the viscous ternary flow driven by the magnetohydrodynamics (MHD) in non-uniform convergent systems. The equations governing momentum, heat and mass transfers are formulated, simplified with the help of similarity transformations, and numerically solved with the bvp4c scheme. This is then followed by a by an artificial neural networks (ANN) scheme with the multi hidden layers to achieve improved predictive accuracy and allow the generation of high-fidelity simulations over a wide parameter space at low computational costs. The hybrid model has outstanding accuracy represented by a mean square error of 1.92×10−11 and regression coefficient of value equal to1. The results have shown that flow velocity rises in the convergent section and therefore the flow can attain sonic conditions in the throat which is critical in generating thrust. Stronger electromagnetic effects increase velocity, and large Reynolds numbers and wide opening angles decreases velocity and elevate the change of recirculation and boundary layers separation in the divergent section. Brownian diffusion have a significant effect on the temperature and concentration distributions, which have direct impact on thermal gradients and species transport in high-pressure combustion chambers and nozzle cooling systems. The convergent part is essential to the transfer of momentum and thrust to the best, whereas the divergent part needs a cautious design to avoid the recirculation-based losses and avoid the flow instabilities that often impair the structural integrity. These AI-based understandings would offer a strong, data driven base to the development of a next-generation, energy-efficient propulsion nozzle, such that is could permit the increase of thrust output, thermal performance, longer component existence, and the utilization of less fuel in both jet engines and energy-saving marine propulsion.

Instituciones: Universidad Nacional Pedro Ruiz Gallo