Machine Learning Models for Predicting Polymer Solubility in Solvents across Concentrations and Temperatures

Description

Mona Amrihesari, Joseph Kern, Hilary Present, Sofia Moreno Briceno, Rampi Ramprasad, and Blair Brettmann predict whether polymers dissolve in this two thousand twenty-four paper. How well a polymer dissolves governs its mixing and use, but past datasets were too small and coarse to capture how strength and heat matter. The team built a dataset from Crystal16 cloudiness measures, recording percent transmission for many polymer-solvent pairs across strengths and heat. That signal tells clear solutions from cloudy, partly dissolved ones.

They trained a model predicting the transmission data across conditions. From those predictions, pairs sort into three grades of dissolving, finer than the binary labels of earlier models. Adding many strengths, heat levels, and partly soluble cases extends prediction toward what mixers and designers need. The work shows machine learning can pull out graded behavior, not just class labels, when base measures are dense enough.

Catalog record

Source: https://europepmc.org/articles/PMC11684024