Multi-Objective Optimization of Sustainable Epoxy Resin Systems through Bayesian Optimization and Machine Learning
Rodrigo Q. Albuquerque, Florian Rothenhäusler, Philipp Gröbel, Holger Ruckdäschel
Abstract
In this work, petroleum-based epoxy resins and curing agents are mixed with their biobased counterparts to create sustainable epoxy resin systems for resin transfer molding techniques. Multiobjective Bayesian optimization (BO) was employed to simultaneously maximize two mechanical and one thermal property of eight-component, biobased thermosets with as few as five additional experiments, enhancing sustainability by reducing resource-intensive trials. Machine learning (ML) models were used for property prediction based on the formulation composition. The LASSO model provided interpretable results, revealing relationships between specific components and target properties, besides exhibiting prediction accuracy of ca. 94%. This research highlights the potential of multiobjective BO in designing sustainable biobased epoxy resin systems and emphasizes the interpretability and predictive power of ML models in material formulation optimization, contributing to environmentally friendly and cost-effective material development.