Machine Learning-Assisted Design of Thin-Film Composite Membranes for Solvent Recovery

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Organic solvents are vital industrial chemicals, necessitating efficient recovery for environmental and economic reasons.
• Organic solvent nanofiltration (OSN) is an energy-efficient technology for solvent recovery, but membrane development relies on inefficient trial-and-error methods.
• Developing advanced membranes is crucial for sustainable manufacturing and environmental protection.

Purpose of the Study:

• To develop a machine learning (ML) approach for the rational design of new thin-film composite membranes for organic solvent nanofiltration (OSN).
• To predict the performance of novel OSN membranes using ML models trained on monomer and property data.
• To identify high-performing membrane candidates for efficient solvent recovery.

Main Methods:

• Utilized gradient boosting regression to train ML models on featurized monomer, membrane, solvent, and solute properties.
• Designed and predicted the OSN performance of 167 new membranes using ML models for common organic solvents.
• Experimentally synthesized and tested a novel membrane to validate ML predictions.

Main Results:

• ML models accurately predicted OSN performance, including solvent permeance and solute rejection.
• Identified novel membrane candidates exhibiting superior methanol permeance compared to existing membranes.
• Discovered that nitrogen-containing heterocyclic monomers enhance membrane microporosity and permeance.

Conclusions:

• The developed ML approach enables a bottom-up strategy for the rational design of high-performance OSN membranes.
• This data-driven methodology significantly accelerates the discovery of new materials for solvent recovery.
• The approach holds potential for designing membranes for various other technologically important applications beyond solvent recovery.