The structure-property relationship of granular materials with different friction coefficients: Insight from machine learning

When granular materials are subjected to mechanical disturbance, dynamic heterogeneity can be observed. Although it has long been considered that dynamic heterogeneity is related to structure in material science, there were still few studies focusing on the structure-property of granular materials. In this study, we simulate conventional triaxial tests of polydisperse spheres using discrete element method. Different friction coefficients were used to represent different microscopic contact modes in the simulation. The machine learning (ML) model for predicting the plastic deformation of granular materials is successfully developed from particles' local structural information using the eXtreme Gradient Boosting algorithm. Besides, we focus on how the structural indicators of granular materials affect the multiple physical and mechanical properties. We further observed and explained the variation of ML predictive power in granular systems with different friction coefficients. Overall, our study presents a more intensive and innovative insight into the structure-property relationship of granular materials.