Identification and Design of Better Diamine-Hardened Epoxy-Based Thermoset Shape Memory Polymers: Simulation and Machine Learning

Abstract

An approach for designing thermoset shape memory polymers (TSMPs) with improved shape memory properties through the integration of molecular dynamics (MD) simulation, machine learning (ML), and chemical intuition is presented. We identified key molecular features correlated with desired shape memory properties, and used MD simulations to create an initial data set of TSMPs consisting of commercially available and manually designed monomers. Our prediction set was prepared by employing four different approaches for modifying existing monomers based on chemical intuition and insights gleaned from the literature. We trained our ML model on the initial data set, used it to identify the most promising candidates, evaluated their properties, and added them to our initial data set. To further speed up the process, we identified the most promising candidate after a few cycles and modified its structure to obtain a variant with better properties. Our approach, which capitalizes on the synergy between computational methodologies and human expertise to enable efficient exploration of vast chemical space, resulted in the design of a monomer exhibiting more than 60% increase in the desired recovery stress compared to the highest experimentally validated one.