Identifying strain-dependent structural defects in granular materials from the hidden structure-plasticity relationship

Abstract

When subjected to mechanical loads, granular materials exhibit heterogeneous local plastic responses. Although there is a general consensus that such plastic heterogeneity is closely correlated to the structure configuration, the extent and manner in which the initial particle structure determines the plastic dynamics remain ill-understood due to the complex local environment of granular materials. In this study, we present an investigation of the structural origin of shear-induced plastic heterogeneity through a series of numerical triaxial shear tests. A machine learning (ML) framework is employed to address the hidden relationship between the initial structure configuration and plasticity at different loading stages. It is confirmed that there are some pre-existing structural defects independent of loading protocol which is the origin of small strain plastic rearrangements and can be identified with outstanding accuracy by our trained ML models. We then show that the dominant structural features determining plasticity depend on strain, wherein the plastic activity at small strains is governed by contact number but becomes dominated by local packing fraction at large strains. By exploring the temporal correlation of the two structural features, we finally prove this transition results from the differences in their memory effects.