Abstract: The ability to precisely control the phase behavior of polymer solutions is critical for designing advanced materials and optimizing manufacturing processes, from membrane formation to drug delivery systems. However, predicting this behavior remains a significant challenge, as classical thermodynamic models often lack quantitative accuracy without complex, system-specific parameterization. This talk will introduce a new framework that integrates machine learning with established physical principles to generate fast and accurate polymer-solvent phase diagrams.

We first introduce a series of data-driven models (e.g., XGBoost, Neural Networks, Gaussian Process Regression) capable of predicting cloud point temperature curves for a wide range of polymer-solvent chemistries. A comparison of molecular featurization strategies, including a novel method informed by Hansen solubility parameters, and generalization to new polymer-solvent chemistries is discussed. To establish the fundamental relationship between theoretical accuracy and theory-informed model performance, we present recent work on predicting small molecule solvation free energies. Leveraging these insights to address data scarcity in macromolecular systems, we incorporate thermodynamic constraints directly into a neural network model architecture. This physics-informed approach significantly improves the prediction of diverse phase behaviors, including upper critical solution temperature (UCST), lower critical solution temperature (LCST), and closed-loop miscibility gaps.

Furthermore, we demonstrate improved interpretability via analysis of critical point scaling. The framework is then extended to more complex macromolecular architectures, demonstrating that models trained on linear polymer data can successfully predict the phase behavior of star-shaped polymers and polymer-grafted nanoparticle solutions. By demonstrating how machine learning can be effectively guided by physical theory, this work provides a generalizable tool for accelerating the design and processing of next-generation polymer materials.