Micro- and macroscopic aspects of the intermittent behaviors of granular materials related by graph neural network

Abstract

Slowly sheared granular materials deform via intermittent slip avalanches, characterized by repetitive cycles of elastic loading interrupted by sudden stress drops. Although many mesoscopic models have been proposed to describe the relations between generic features in the macroscopic flow and microscopic plasticity, quantitative relations between microscopic plasticity and macroscopic stress drop remain elusive. We simulate the simple shearing of a three-dimensional dense granular assembly using the discrete element method. Microscopic plasticity is more spatially concentrated and tends to generate larger active clusters with the increase of the magnitude of stress drop. The initialization, development, and aggregation of microscopic plasticity constitute the microscopic origin of macroscopic slip avalanches. We then utilize the graph convolution neural network to establish a quantitative relation between microscopic plasticity and macroscopic stress drop. Further analysis on the number of convolution layers reveals that the spatial extent of microscopic plasticity determines the magnitude of macroscopic stress drop.