Related Experiment Video
Updated: Dec 20, 2025
09:39
Proof-of-Concept for Gas-Entrapping Membranes Derived from Water-Loving SiO2/Si/SiO2 Wafers for Green Desalination
Published on: March 1, 2020
7.8K
Designing exceptional gas-separation polymer membranes using machine learning.
J Wesley Barnett1, Connor R Bilchak1, Yiwen Wang1
1Department of Chemical Engineering, Columbia University, New York, NY, USA.
Science Advances
|May 23, 2020
Summary
Machine learning accelerates polymer membrane discovery. This approach predicts gas separation performance, enabling the creation of novel materials that surpass current CO2/CH4 separation capabilities.
Area of Science:
- Materials Science
- Chemical Engineering
- Computational Chemistry
Background:
- Polymer membrane design traditionally relies on empirical methods, hindering the discovery of optimized materials for specific gas separations.
- The vast chemical space of polymers presents a significant challenge for traditional experimental screening.
Purpose of the Study:
- To develop and validate a machine learning (ML) approach for predicting polymer membrane performance in gas separation.
- To accelerate the discovery of novel polymer membranes with enhanced separation capabilities, particularly for CO2/CH4 mixtures.
Main Methods:
- A machine learning algorithm was trained using a topological, path-based hash of polymer repeating units.
- The model utilized experimental gas permeability data for six gases across approximately 700 polymers.
- Predictions were made for over 11,000 untested homopolymers.
Main Results:
- The ML model successfully predicted the gas-separation behavior of numerous polymers.
- Two synthesized membranes, identified as highly promising by the ML model, demonstrated superior CO2/CH4 separation performance, exceeding established upper bounds.
- The ML technique proved effective even with a limited dataset and no simulation data.
Conclusions:
- Machine learning offers an innovative and efficient strategy for exploring the polymer design space.
- This data-driven approach significantly advances the discovery of high-performance polymer membranes for gas separation applications.
- The developed ML technique can guide experimental efforts, reducing the need for exhaustive testing.
