Machine learning computes the elastic response of cross-linked polymers

Abstract

Synthetic polymers with a cross-linked molecular structure similar to that of natural rubber have vastly pervaded daily life and highly specialized realms. Efficient and accurate quantification of their elastic response is crucial for current applications and next-generation elastomer design. Here, we introduce a generalized machine learning (ML) model that fingerprints the most pervasive physicochemical variables that intrinsically govern the formation of the cross-linked molecular structure, topology, and reaction kinetics. These physicochemical variables are also experimentally determinable. We then show that these descriptors can efficiently compute the elastic response of this class of elastomers. The robustness and accuracy of ML computations of elastic response are compared against the experimental stress–strain profiles of three important members of this class, namely (poly)dimethyl siloxane (PDMS), carboxylated acrylonitrile butadiene rubber (XNBR) and styrene-butadiene rubber (SBR), which are synthesized in our laboratory. We vary the physicochemical variables related to the descriptors used in the ML model and the chemical species during experiments, making ML computations-experimental comparison rigorous.