Researchers develop physics-informed neural networks that combine machine learning with thermodynamic principles to create digital twins for monitoring distillation columns in real-time.

Physics-Informed Neural Network (PINN) framework embeds fundamental thermodynamic constraints directly into the neural network, including vapor-liquid equilibrium and mass/energy balances

Model trained on 961 timestamped measurements spanning 8 hours of transient operation data generated in Aspen HYSYS simulation software

Digital twin technology enables industrial process monitoring, control, and optimization by combining machine learning with high-fidelity physics-based modeling

Designed for dynamic, tray-wise modeling of binary distillation columns operating under transient (non-steady-state) conditions using McCabe-Thiele methodology