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

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Chemistry

Background:

• Artificial intelligence (AI) and machine learning (ML) are crucial for accelerating new material design.
• Polymer solubility is a key property for developing new formulations and processing techniques.
• Existing predictive models for polymer solubility are limited by insufficient experimental data.

Purpose of the Study:

• To develop an enhanced dataset for polymer solution behavior using Crystal16 turbidity measurements.
• To train a predictive model capable of forecasting polymer solution behavior across various conditions.
• To classify polymer/solvent pairs into three distinct solubility categories for greater predictive granularity.

Main Methods:

• Collected high-quality percent transmission data for diverse polymer solutions using Crystal16 turbidity measurements.
• Developed and trained an AI/ML model using the generated dataset to predict transmission data at multiple temperatures and concentrations.
• Classified polymer/solvent pairs based on predicted solubility, moving beyond binary solvent/nonsolvent classifications.

Main Results:

• Generated a comprehensive dataset of polymer solution behavior, including varied polymers, solvents, concentrations, and temperatures.
• Successfully trained a model that accurately predicts experimental transmission data.
• Achieved a three-category solubility classification, offering enhanced detail compared to prior binary models.

Conclusions:

• The developed dataset and predictive model significantly advance the capability to predict polymer solubility.
• The model's ability to handle multiple concentrations, temperatures, and partial solubility is valuable for industrial applications.
• This work provides a more granular and practical approach to solubility prediction for formulators and process designers.