Generative Design of Thermoset Shape Memory Polymers Driven by Chemical Group: A Conditional Variational Autoencoder Approach

Abstract

The discovery of novel thermoset shape memory polymers (TSMPs) for additive manufacturing can be accelerated through the use of a deep-generative algorithm, minimizing the need for laborious traditional laboratory experiments. This study is the first to introduce an innovative approach that uses a deep generative learning model, namely the conditional variational autoencoder (CVAE), to discover novel TSMPs with lower glass transition temperature ($T_g$) and high recovery stress values ($E_r$). In this study, specific chemical groups, such as epoxy, amine, thiol, and vinyl, are integrated as constraints to generate novel TSMPs while preserving the essential reaction properties. To address the challenges posed by a small dataset, the CVAE model is used with graph-extracted features. Unlike previous studies focused on single-polymer systems, this research extends to two-monomer samples, discovering 22 novel TSMPs. This approach has practical implications in additive manufacturing, biomedical devices, aerospace, and robotics for the discovery of novel samples from limited data.