Physical insights into stress–strain process of polymers under tensile deformation via machine learning

ABSTRACT Strain localization is a ubiquitous phenomenon of soft matters subjected to strain. Polymeric materials are a very important class of soft materials and widely used nowadays. Polymers have unique strain localization behavior such as crazing during tensile deformation. How to understand the mechanism at the molecular level of strain localization in polymeric materials has become an important topic in material science. In this work, tensile deformation process of polymers both under a melt state and a glassy state are investigated in MD simulations using a generic coarse-grained model. We use a machine learning technique, i.e., support vector machine (SVM) algorithm, to understand the local molecular structure and the dynamical properties during tensile deformation. By defining “softness” from the SVM model, we investigate the stress–strain behavior of both ductile polymer above glass transition temperature and brittle polymer glass during tensile deformation. We demonstrated that the softness can be used to predict physical properties efficiently; the softness provides deep physical insights into the non-equilibrium stress–strain process. We also find that the Hookean behavior of polymer glasses is mostly contributed by the hard regions of the system, and the elastic limit is quantitatively discussed as well.