Image-informed numerical modelling of particulate systems with irregular grains

Abstract

Granular materials are everywhere around us. Their omnipresence makes our interaction with them on a daily basis a certainty, and yet our understanding of their mechanical behaviour is far from complete. Regarding geotechnical applications, most natural granular materials, such as silts, sands, gravels and ballast, feature irregular particle shapes, a fact that makes their mechanical behaviour all the more complex across scales, from micro to meso and macro. A multitude of experimental and numerical studies have demonstrated the importance of particle morphology in the shear strength of particulate materials, although rarely demonstrating a direct link or mechanisms of causality between them. This is mainly due to the high complexity of the problem but also partially due to the lack of intelligible and accessible tools to quantify the morphology of three-dimensional irregular particles. This thesis aims to contribute to the current state-of-art studying the characterisation of granular materials by providing analytical and numerical tools for shape characterisation. Regarding analytical tools, this thesis attempts a critical review of existing indices to characterise and classify particle form, while introducing a new set of indices. Regarding numerical tools, this thesis provides novel software solutions for automatic particle shape characterisation and for the generation of image-informed numerical models. These open-source tools are meant to shed light on the inherent subjectivity of performing shape characterisation on a practical level. Regarding the generation of numerical models based on imaging data, algorithmic implementations are offered to create simplified polyhedra and multi-sphere particles at user-defined fidelity levels of resolution, the morphology of which can also be characterised and compared to that of the original fidelity level. Combining the produced analytical and numerical tools, this thesis demonstrates a seamless workflow between particle imaging data and numerical modelling, using the discrete element method and non-spherical particles. This workflow is utilised to develop a methodology for the generation of Representative Element Volumes (REVs) of non-spherical particles, which represent the polydispersity of both particle size and shape, aiming to link quantitative morphology characterisation at the particle scale and mechanical characterisation at the level of a representative assembly of particles. The methodology is then applied to systematically generate REVs of railway ballast using image-informed multi-sphere particles of various levels of simulation fidelity, allowing for a parametric study of the effect of several modelling parameters on the shear strength of the material.