Abstract

The dependence of elastomers' behavior on loading conditions, e.g., strain rate or temperature, has been the subject of interest in recent decades, and numerous phenomenological models have captured it successfully. However, given the complexity of the compound property relation, so far, a few physics-based models can take into account the micromechanics of hyperelasticity have been produced. Advancing our recent machine-learned constitutive model for elastomers[1], we developed data-infused knowledge-driven machine learned surrogate functions that describe the quasi-static response of polymer batches in crosslinked elastomers. Based on the above work and a new generation of machine learning algorithms known as conditional neural network (CondNN), the final engine can describe constitutive behaviour considering the effects of loading conditions, such as temperature and strain rate, as well as the effects of compound morphology such as filler percentage and crosslink density. Following our previous work, we implemented knowledge from polymer science, statistical physics, machine learning, and continuum mechanics principles to reduce the 3D stress-strain mapping space into a 1D space. The proposed order reduction significantly reduces the training cost by minimizing the search space. Our developed engine can consider the roles of strain rate, temperature, and filler percentage in different deformation states and enjoys a high training speed and accuracy even in complicated loading scenarios. Our engine can be generalized to soft robotics, soft digital materials (DMs), hydrogels, and adhesives and also can be used in material design.

This content is only available via PDF.
You do not currently have access to this content.