Abstract

Polymeric membranes are widely used in membrane separation technologies due to their chemical inertness, cost-effectiveness, and high efficiency. Their tailorable structure and properties make them suitable for a wide range of applications. The performance of these membranes is influenced by polymer type, fabrication methods, microstructure, and specific applications. Consequently, optimizing these parameters remains a key focus in membrane design research.

This work aims to predict membrane microstructure formation, starting with an analysis of fundamental polar/non-polar solvent interactions in polymeric membrane formulations using molecular dynamics simulations.

Physics-based machine learning (ML) and explainable Artificial Intelligence (XAI) were applied to predict the liquid-liquid equilibrium (LLE) of this system. Additionally, this study introduced multimodal (MM) modeling for membrane structure prediction, utilizing a comprehensively curated membrane database. This research contributes to advancing membrane design through data-driven approaches.

Keywords: Machine learning (ML), Thermodynamics, Liquid-liquid equilibrium (LLE), Molecular Dynamics (MD), Multi-modal ML, Membranes, Structure prediction.