(287e) Machine Learning Inspired Polyimide Gas Separation Membranes: A Leap Beyond Conventional Aromatic Counterpart

The quest for polymer gas separation membranes with exceptional performance remains a daunting challenge in membrane science, often involving lots of extensive trial and error and leading to long membrane material development cycles. Recent endeavors have turned to machine learning (ML) models to expedite this Edisonian process. However, the budding potential of ML models is to accelerate the discovery and development of innovative polymeric membrane materials for gas separation. We leverage a graph-augmented ML method that is interpretable with explainable polymer substructures to discover “hidden gems” from existing polymers that outperform the permeability-selectivity upper bounds, as well as to guide the design of innovative high-performance polymers. With the aid of the improved ML technique, we developed a series of novel partially alicyclic polyimides (PAPIs) for gas separation, demonstrating excellent separation performance for key gas pairs such as H₂/CH₄, H₂/N₂, CO₂/CH₄, and an exceptionally high selectivity of 15 for O₂/N₂. Incorporating alicyclic units in polyimides contributes to processibility, increased membrane flexibility, and free volume, facilitating enhanced gas transport while maintaining robust mechanical, thermal, and chemical stability. Importantly, the incorporation of alicyclic diamines with non-fluorinated dianhydrides into the polyimide backbone mitigates environmental and health concerns. Specifically, two aromatic PIs identified by ML algorithms and four innovative partially alicyclic polyimides inspired by ML predictions were synthesized and comprehensively characterized in terms of structure and gas transport properties. Our experiments successfully validated the ML model predictions within the degree of uncertainty. This talk will focus on these high-performance polyimides' synthesis, characterization, and gas transport properties, emphasizing the fundamental structure-property relationships. The ML-aided design of PAPIs with unique features substantiated by experimental validation demonstrates significant progress in the quest for high-performance and eco-friendly membrane-based gas separation polymers. Additionally, the scalability of these materials suggests a feasible transition from laboratory to industry.