Predicting the Phase Behavior of Ternary Polymer Solutions Consisting of Two Polymers in a Common Solvent Using Autonomous Experimentation and Machine Learning

Knowledge of the phase behavior of multi-component polymer solutions is essential for many technologies, ranging from efficient synthesis to formulation of inks for additive manufacturing. Although the phase behavior has been extensively studied experimentally and theoretically, accurate prediction remains elusive. Recently, machine learning was shown to successfully predict upper, lower, and closed binary polymer solution behavior using previously published co-existence data. Here, we combine automated experimentation and machine learning to expand prediction to the phase behavior of ternary polymer solutions. AI algorithms are used to guide subsequent experiments based on past data to efficiently develop polymer 1-polymer 2-solvent ternary co-existence curves. The resulting data is combined with previously published results, and used to train a machine learning model (e.g. neural network) for generalized prediction of ternary polymer 1-polymer 2-solvent systems.