A data-driven approach to interfacial polymerization exploiting machine learning for predicting thin-film composite...

Area of Science:

• Materials Science
• Chemical Engineering
• Polymer Science

Background:

• Polymeric thin-film membranes are crucial for liquid separation, offering reduced industrial waste and energy use.
• Current limitations in monomer diversity restrict the development of new membranes.
• A data-driven approach is needed to expand the chemical space for membrane materials.

Purpose of the Study:

• To develop a divide and conquer strategy for interfacial polymerization membrane development.
• To create a large, open-access dataset of interfacial polymerization reactions.
• To enable data-driven prediction of thin-film formation from monomer properties.

Main Methods:

• Compiled a dataset of 18 organic and 73 water-phase monomers, performing 1246 interfacial reactions.
• Analyzed membrane properties using Atomic Force Microscopy (AFM) and optical microscopy.
• Trained five machine learning models using molecular structures and Density Functional Theory (DFT) calculations.

Main Results:

• Demonstrated that film formation can be predicted directly from monomer characteristics.
• Established an unprecedentedly large and open-access dataset for membrane development.
• Identified key parameters influencing thin-film formation in interfacial polymerization.

Conclusions:

• The proposed data-driven approach facilitates the development of novel thin-film membranes.
• Focusing on thin-film formation, rather than just performance, offers a new perspective in membrane research.
• This work paves the way for accelerated and more efficient membrane discovery.