Estimation of the Flory-Huggins interaction parameter of polymer-solvent mixtures using machine learning

The Flory-Huggins interaction parameter chi for polymer-solvent mixtures captures the nature of interactions and provides insights on solubility. chi is usually estimated using experimental or (empirical) computational methods, which may be expensive, time-consuming or inaccurate. Here, we built a machine learning (ML) model to instantly predict temperature-dependent chi for a given polymer-solvent pair. The ML model was trained using 1586 experimental polymer-solvent datapoints, and a hierarchical polymer and solvent fingerprinting scheme. Extensive testing has been performed to verify the accuracy and generality of this model. This work demonstrates an ML model that can progressively be improved as new data emerges.

Automated summary

This study presents a machine learning model that can instantly predict the temperature-dependent Flory-Huggins interaction parameter for polymer-solvent mixtures. The model has been trained using a large dataset of experimental data and demonstrates high accuracy and generality.