Forecast of Glass Transition Zone of Thermoset Polymers Using a Multiscale Machine Learning Approach

Many previous studies have used machine learning (ML) techniques to predict a single, precise glass transition temperature (Tg) for polymers, focusing narrowly on a specific point rather than on the progressive glass transition process. In contrast, our paper introduces a novel ML approach that predicts the change of the storage modulus as a function of temperature within the entire glass transition zone for thermoset polymers, thus offering a more comprehensive prediction of this phase transition. This method differentiates itself by using features across three scalesmicroscopic, mesoscopic, and macroscopicas inputs to develop a multiscale fingerprinting technique. We have identified four key essential features for capturing the complete modulus change with temperature accurately. To model the glass transition zone effectively, our approach utilized three supervised learning techniques: support vector regression (SVR), artificial neural network (ANN), and Gaussian process (GP) models. After evaluating four common indices, we selected the ANN model as our primary tool due to its superior performance. We also detailed the mathematical methods underlying the models to elucidate the performance differences. To demonstrate the efficacy of our model, we applied it to predict the glass transition zone curves for three classes of new polymers involving four polymers and experimentally validated them, confirming that it basically captured the essential characteristics of the experimental curves. Thus, we believe our approach not only advances ML applications in polymer design but also serves as a valuable tool for future innovations in material science, significantly enhancing how we manipulate polymer properties.