Physics-Informed Neural Network Method for the Thermal Optimization of Turbulence and Heat Transfer in a Shell-and-Tube Heat Exchanger

Sánchez-De-La-Hoz J.; Troncoso-Palacio A.; Marenco J.F.; Sanmartin-Mendoza P.; Marcelo-Aldana D.; Archbold-Taylor G. · C3 2025 - IEEE Colombian Caribbean Conference (2025) · DOI: 10.1109/c366505.2025.11340452

Resumen

Shell-and-tube heat exchangers are critical components in industrial thermal systems where energy efficiency and performance depend on turbulence modeling and heat transfer efficiency. Conventional optimization relies on Computational Fluid Dynamics (CFD) using Reynolds-Averaged Navier-Stokes (RANS) models such as k-ϵ and k-ω SST, which, although accurate, demand high computational resources and often struggle to generalize turbulent flow behavior. In this work, a PhysicsInformed Neural Network (PINN) framework is proposed for the thermal optimization of turbulence and heat transfer in a FUNKE A100 shell-and-tube heat exchanger. The system was modeled in SolidWorks and simulated with ANSYS Fluent under steady-state turbulent flow conditions, supported by mesh independence analyses to ensure numerical reliability. The k-ϵ and k-ω SST turbulence models were validated against manufacturer data, with results indicating that k-ω SST provided a more accurate prediction of near-wall gradients (4.7% deviation) compared to ϵ(7.9 %). Heat transfer coefficients reached 812 W / m2 K under optimal flow conditions, while shell-side pressure drops were quantified at 21.4 kPa. These CFD datasets were subsequently used to train and evaluate the PINN model, which successfully reproduced thermal-hydraulic behaviors with a mean relative error below 3.5%, while reducing computational cost by approximately 85% compared to traditional CFD. The integration of physics-based deep learning and governing conservation laws demonstrates that PINNs can deliver accurate, fast, and scalable predictions of turbulent heat transfer, enabling real-time optimization and offering a robust alternative to CFD for the design and operation of industrial heat exchangers.

Instituciones: Universidad de Piura