Estimation of contact forces of granular materials under uniaxial compression based on a machine learning model

This paper presents a graph neural network model to estimate the contact forces of granular materials under uniaxial compression. We show that the maximum normal contact force of each particle of a narrowly graded granular assembly at the end of a loading increment can be estimated using the particle kinematics and inter-particle contact kinematics of the assembly during the increment, which are based on knowledge obtained from the uniaxial compression of other granular systems with different initial microstructures (i.e., grain locations and contact distribution). The model is trained using data generated by 3D discrete element modelling (DEM) of granular assemblies with different initial microstructures under uniaxial compression. Model predictions of normalized particle maximum normal contact forces of a typical granular assembly under uniaxial compression are presented. They are found to be well consistent with those from DEM simulations.

Automated summary

This paper presents a graph neural network model to estimate the contact forces of granular materials under uniaxial compression. The model combines particle kinematics and inter-particle contact kinematics with the compression behavior of granular systems with different initial microstructures to estimate the maximum normal contact force of each particle in the assembly.