Fast and accurate modeling of thermoset fracture by active learning quantum-chemical bond scission

A molecular understanding of thermoset fracture is crucial for enhancing performance and durability across applications. However, achieving accurate atomistic modeling of thermoset fracture remains computationally prohibitive due to the high cost associated with quantum mechanical methods for describing bond breaking. In this work, we introduce an active learning (AL) framework for our recently developed machine-learning based adaptable bond topology (MLABT) model that uses datasets generated via density functional theory (DFT) calculations that are both minimalistic and informative. Employing MLABT integrated with AL and DFT, we explore fracture behavior in highly crosslinked thermosets, assessing the variations in fracture behavior induced by system temperature, temperature fluctuations, strain rate, cooling rate, and degree of crosslinking. Notably, we discover that while fracture is minimally affected by temperature, it is strongly influenced by strain rate, suggesting the absence of the time-temperature superposition in thermoset plasticity. Furthermore, while the structural disparities introduced by different network annealing rates influence the elastic properties, they are inconsequential for thermoset fracture. In contrast, network topology emerges as the dominant determinant of fracture, influencing both the ultimate strain and stress. The integration of MLABT with the AL framework paves the way for efficient and DFT-accurate modeling of thermoset fracture, providing an affordable and accurate approach for calculating polymer network fracture across chemical space.