Bridging Experimental Research and Process Design via Machine Learning: AI-Enhanced Simulation of Azeotropic Distillation Processes in a Polyamide Recycle Process

Abstract

The integration of Artificial Intelligence with process engineering offers new opportunities for optimizing complex industrial operations. In this study, Artificial Intelligence-assisted models were developed to enhance the simulation and design of azeotropic distillation applied to solvent recovery (organic acid mixture dehydration) in a polyamide recycling process. Experimental data from a pilot scale distillation column were complemented with process simulations performed in AVEVA™ PRO/II process simulation software. Two neural network architectures, Kolmogorov-Arnold Network and Sequential Neural Network, were trained using both experimental and simulated data to construct a Digital Twin of the distillation unit. Model predictions were compared against experimental and simulated results, evaluating the fitting of the generated data in terms of solvent recovery efficiency and energy consumption. Results highlight the potential of hybrid Artificial Intelligence, simulation and experimental approaches to accelerate process optimization, reduce experimental effort, and improve prediction accuracy for the investigated system.