Prediction of 3D contact force chains using artificial neural networks

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

2 Scopus Citations
View graph of relations

Detail(s)

Original languageEnglish
Article number106444
Journal / PublicationEngineering Geology
Volume296
Online published10 Nov 2021
Publication statusPublished - Jan 2022

Abstract

Contact force chains (CFCs) in the heterogeneous granular materials are often considered to be structured physical systems that play a key role in their mechanical properties such as stiffness, strength, stability and flowability. In this context, quantitatively estimating the evolution of CFCs in a quasi-statically sheared granular system is essential for advancing our understanding of the mechanics of granular materials. In this paper, based on discrete element method (DEM) simulation data, an artificial neural network (ANN) is developed and applied to predict the anisotropy of the CFCs in two types of idealized granular materials with different initial relative densities undergoing triaxial shearing. Five features including particle size, coordination number and particle displacement (i.e., x-, y- and z-components of the particle displacement) at the particle-scale and the meso-scale each are used to train and test the established ANN model. The results of model prediction show that the 3D orientational distributions of the CFCs from the ANN predictions match very well the DEM simulation results during the whole shearing progress. It is found that for both dense and loose samples, the combined set of particle-scale and meso-scale features have a dominating influence on the CFC evolutions but the ANN model performs better in the CFCs estimation at the strain increment of 0–7% than at the strain increment of 7%–14%. The outcome of this study shows that machine learning is a promising tool for studying the complex mechanical behavior of granular materials.

Research Area(s)

  • Contact force chains, DEM, Granular material, Machine learning, Neural network