Discover and Understand Microporous Polymers for Size-sieving Separation Membranes using Active Learning

Membrane-based separation technologies have great promise to dramatically drive down the energy, carbon, and water intensity of traditional thermally-driven separation processes such as distillation. The creation of novel membrane materials with tailorable yet predictable structure and properties holds the key to providing low energy solutions to some of the nation’s most challenging and important separations, such as in clean energy industries (e.g., hydrogen gas purification) and environmental remediation (e.g., carbon capture).

However, the development cycles of such materials are usually exceptionally long due to the requisite trial-and-error strategy employed. This project aims to combine simulations, machine learning, and experimental studies to accelerate the development of high-performance polymer gas separation membranes. The knowledge gained from this project will enable a more rational strategy for the design of advanced materials for energy-efficient separations and provide potentially revolutionary solutions to the grand materials challenges in the membrane separation field.

This project will also strengthen the multi-disciplinary collaboration for incorporating machine learning into polymeric membrane materials. In addition to advancing knowledge and technology, the research will be integrated with graduate and undergraduate student education and training opportunities and through local K-12 student and teacher outreach programs.

The overarching goal of this project is to accelerate the discovery and enrich the fundamental understanding of highly permeable and selective polymer gas separation membranes using an active learning scheme that synergistically combines molecular simulations, machine learning, and experiments. The scope of this project will include (1) establishing a standardized polymer database by combining existing database, open literature, and high-throughput molecular simulations; (2) employing transfer learning, molecular simulations, and experiments to develop accurate surrogate models that map out chemistry-property relations for polymer gas separation membranes; (3) establishing a Bayesian Optimization framework to guide the iteration of transfer learning and experimental discoveries; and (4) using classification together with detailed molecular simulation and experimental study to understand molecular features impacting polymer free volume architecture and gas transport properties.

The Materials Informatics-based active learning approach to be established in this project will be a valuable strategy for the field concerning polymer separation membranes, and it can be readily extended to design polymers with other desirable properties beyond gas separation, which can save the cost and time traditionally required for new material development in general.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.