Machine learning-based epoxy resin property prediction

Epoxy resins have been utilized across various industries due to their superior mechanical and chemical properties. However, discovering the optimal design of epoxy resins is challenging because of the large chemical space of polymer systems. In this study, we adopted a data-driven approach to develop an effective prediction system for epoxy resin. In particular, we constructed a database of 789 epoxy resins, encompassing four key properties: density, coefficient of thermal expansion, glass transition temperature, and Young's modulus, obtained through molecular dynamics simulations. We devised descriptors that effectively represent epoxy resins. Ultimately, a machine learning model was trained, successfully predicting properties with reasonable accuracy. Our predictive model is a generalized model that was verified across various types of epoxy resins, making it applicable to all kinds of epoxy and hardener combinations. This achievement enables large-scale screening over numerous polymers, accelerating the discovery process. Further, we conducted an in-depth analysis of the important features that have a high impact on the epoxy resin. This provides valuable insights into the structure-property relationship which can guide researchers in designing new epoxy resins. This study accelerates the discovery of epoxy resins by effectively predicting their properties and unraveling intricate structure-property relationships, based on molecular simulation data and machine learning techniques.