In the industrial setting, polymer membranes are frequently employed to separate gases like CO2 from flue gas and methane from natural gas. Over the years, scientists have studied different polymers to increase their permeability and usefulness, but up until now, they have not been able to rapidly and effectively test them all.
A novel approach to using machine learning (ML) to test and discover new polymer membranes has been discovered by the University of Connecticut Assistant Professor of Mechanical Engineering Ying Li, University of Connecticut Centennial Professor of Chemical and Biomolecular Engineering Jeff McCutcheon, University of Connecticut researchers Lei Tao and Jinlong He, and researcher Jason Yang from the California Institute of Technology.
This methodology was recently published in Science Advances.
After a thorough investigation, the study team remarked, “In the decades of technological development in the membrane science field, design of new membrane materials has been, and remains, a largely trial-and-error process, guided by experience and intuition. Current approaches generally involve tuning chemical groups to increase affinity and solubility towards the desired gas or incorporating greater free volume to increase overall diffusivity.”
Computational models can be used to forecast membrane performance instead of time-consuming trials. However, they either lack precision due to the streamlined approximations or are excessively costly. The researchers created a precise strategy for applying ML techniques to find novel, high-performing polymers to overcome this deficiency.
The researchers linked membrane chemistry to performance using deep learning on a limited dataset using numerous fingerprint characteristics and fixed chemical descriptors.
The study discovered that deep neural networks performed well due to the usage of ensembling, which integrates prediction from numerous models. RF (Random Forest) models are often considered to perform well on small data sets.
The scientists also discovered that the ML model could find thousands of polymers whose performance was projected to be better than the Robeson upper bound, a measure used to quantify the trade-off between permeability and selectivity for polymer gas-separation membranes.
The industry would also be able to undertake gas separations with a greater throughput while retaining a high level of selectivity, thanks to newly found polymers with ultrahigh permeability.
The researchers further stated, “Ultimately, we provide the membrane design community with many novel high-performance polymer candidates and key chemical features to consider when designing their molecular structures. Lessons from the workflow demonstrated in this study can likely serve as a guide for other materials discovery and design tasks, such as polymer membranes for desalination and water treatment, high-temperature fuel cells, and catalysis.”
They concluded, “With the continual improvement of ML techniques and an increase in computing power, we expect that ML-assisted design frameworks will only gain popularity and deliver increasingly substantial results in materials discovery for a wide range of applications.”
The Air Force Office of Scientific Research (FA9550-20-1-0183; program manager: M.-J. Pan), the National Science Foundation (CMMI-1934829 and CAREER Award CMMI-2046751), 3M’s Non-Tenured Faculty Award, and the National Alliance for Water Innovation (NAWI), under Funding Opportunity Announcement Number DE-FOA-0001905 of the U.S. Department of Energy, provided funding in full or in part for this project.
Yang, J., et al. (2022) Machine learning enables interpretable discovery of innovative polymers for gas separation membranes. Science Advances. doi:10.1126/sciadv.abn9545.