Machine learning discovers ‘hidden-gem’ materials for heat-free gas separation

Chemical separation, including gas separation, is a common process that is required for manufacturing and research. It accounts for a whopping 15% of U.S. energy consumption and produces millions of tons of carbon emissions.

Separating gases by passing them through membranes could be an efficient, environmentally friendly alternative to current methods—if only the right materials could be found to make them.

Applying a graph-based machine learning approach, a team of chemical and mechanical engineers and computer scientists at the University of Notre Dame have discovered, synthesized and tested polymer membranes that can separate gases up to 6.7 times more effectively than previously synthesized membranes.

Their results have been published in Cell Reports Physical Science.

“What determines the membrane’s performance is the material’s microscopic porosity,” said Agboola Suleiman, doctoral student in the lab of Ruilan Guo, the Frank M. Freimann Collegiate Professor of Engineering.

“The ideal membrane material strikes a balance between selectivity and permeability—permeable enough to let gases in, but selective enough to keep some out,” said Suleiman, who is co-author on the paper.

To identify this Goldilocks material, the team used graph neural networks (GNN), a type of machine learning particularly well-suited to representing a material’s molecular structure as well as its relationship with other molecules. After being trained on datasets, GNN identified two polymers that had the right properties to outperform previously synthesized membranes.

“Our machine learning algorithms led us to materials that had previously only been used for electronics applications,” said Tengfei Luo, the Dorini Family Professor for Energy Studies, associate chair of the Department of Aerospace and Mechanical Engineering and co-author on the paper. “Then we synthesized and tested these materials in the lab, verifying their high performance in separating gases. It was like finding hidden gems.”

Synthesizing polymers can be costly and time-consuming, so the data available about their molecular structure and chemical properties are scarce and incomplete.

However, algorithmic innovations devised by co-authors and computer scientists Meng Jiang and his doctoral student Gang Liu solved this problem.

“By using machine learning techniques, we were able to augment and improve our data,” said Jiaxin Xu, doctoral student in Luo’s lab and co-author on the paper. “The graph-based model, enriched with information about each material’s molecular properties, allowed us not only to predict the best membrane materials but also to explain why they’re the best.”

The team’s top-performing polymers may be used to create membranes capable of separating several gas pairs, which are critical for industrial applications.