Predicting orientation-dependent plastic susceptibility from static structure in amorphous solids via deep learning

Abstract

It has been a long-standing materials science challenge to establish structure-property relations in amorphous solids. Here we introduce a rotationally non-invariant local structure representation that enables different predictions for different loading orientations, which is found essential for high-fidelity prediction of the propensity for stress-driven shear transformations. This novel structure representation, when combined with convolutional neural network (CNN), a powerful deep learning algorithm, leads to unprecedented accuracy for identifying atoms with high propensity for shear transformations (i.e., plastic susceptibility), solely from the static structure in both two- and three-dimensional model glasses. The data-driven models trained on samples at one composition and a given processing history are found transferrable to glass samples with different processing histories or at different compositions in the same alloy system. Our analysis of the new structure representation also provides valuable insight into key atomic packing features that influence the local mechanical response and its anisotropy in glasses.

Predicting the failure of two-dimensional silica glasses

“…Modern machine learning methods provide a promising alternative pathway for the systematic development of structure-based predictions 3 , as has been shown in the context of molecular properties 4 – 6 , density functional theory force fields 7 – 9 , governing equations for dynamical systems and flow 10 – 12 and dislocation models for crystal plasticity 13 – 15 . Applications to glasses have been so far restricted to idealized models, so-called Lennard-Jones glasses, that were analyzed with support vector machines (SVM) 16 – 18 , graph neural networks (GNN) 19 and deep learning 20 , 21 . Predictions based on deep learning methods are becoming increasingly accurate but they are also hard to interpret.…”

Predicting Fracture Propensity in Amorphous Alumina from Its Static Structure Using Machine Learning

“…Although these theories have proposed the existence of defects in the glass structure, they are unfortunately phenomenological and fail to provide a precise definition of these defects, thus hindering their identification from first principles . Recently, several data-driven models have been proposed to forecast the propensity of particles or atoms to rearrange. − However, due to the structural complexity of amorphous materials, the correlations between the structure and fracture mechanism in oxide glasses remain unknown.…”

“…Some structural descriptors such as local density, free volume, or bond orientational order can be correlated with flow defects, but are insufficient for fully predicting glass fracture . As a promising alternative path, advances within machine learning have made it possible to predict nonintuitive structural descriptors using algorithms such as support vector machine (SVM), graph neural network, and convolutional neural network . Among these, the nonintuitive structural metric termed “softness” derived from SVM is found to be strongly correlated with the dynamics of specific atoms only based on their local structural environments.…”

Machine-learning integrated glassy defect from an intricate configurational-thermodynamic-dynamic space

“…As such, the IGD could be regarded as a quantitative parameter to identify defect in glass, in analogy to the Burgers' vector in crystals. Note that not all high-IGD regions should experience plastic rearrangement, which is a stochastic phenomenon and sensitive to the loading protocol and thermal fluctuation [41,62]. It is the anisotropic interaction between the glassy defect and the high-dimensional loading direction that cooperatively determines if a glassy defect would lead to real shear transformation.…”