Development of Microtomography-based Full-field Particle Tracking Methods for Complex Granular Soils

Abstract

Granular soils are ubiquitous, including both the terrestrial soils such as sand, gravel, and clay, as well as the marine soils such as coral sand, marine sand, and volcanic deposits. Among the various factors affecting the mechanical response of granular soils, particle breakage stands out as a particularly significant aspect. Particle breakage of granular soils under static or dynamic loading substantially impacts their macroscopic behaviors, leading to various geological disasters and geotechnical issues, such as rock avalanches, landslides, and surface subsidence, etc.

The aim of this study is to experimentally investigate the particle-scale mechanical behavior of granular soils and to quantitatively analyze its evolution using X-ray micro-tomography combined with a particle tracking method. For the experimental program, a medium-sized triaxial apparatus was designed to enable in-situ testing in conjunction with the X-ray micro-tomography facility. The test material consists of completely decomposed granite (CDG), a relatively low-quality residual soil collected from Hong Kong. A cylindrical specimen with dimensions of 25 × 50 mm (diameter × height) was prepared, and both consolidated drained (CD) and consolidated undrained (CU) triaxial tests were conducted with a spatial resolution of 9 μm. A series of images were acquired, and advanced image processing techniques were employed to identify and extract individual particles from the raw CT images, enabling the determination of particle attributes such as volume, equivalent diameter, morphological characteristics, and inter-particle contacts. Regarding the particle tracking method, it can be categorized into three main aspects: (1) to identify and track morphologically simple sand particles that undergo zero or moderate fragmentation; (2) to investigate the fragmentation patterns of sand particles by developing a global tracking method capable of capturing highly fragmented particles with complex morphologies; and (3) to propose a fragmentation modeling framework for the numerical simulation of crushable particles during continuous breakage in triaxial tests.

This study was conducted in four phases. In the first phase, a medium-scale in situ triaxial test on CD and CU specimens of CDG particles was conducted at the Shanghai Synchrotron Radiation Facility (SSRF). During the test, segmented scanning was performed using X-ray micro-tomography to non-destructively capture microscale information of CDG particles throughout the triaxial shearing process. Subsequently, advanced image processing techniques were employed to denoise the raw CT images and to accurately identify and extract individual particle information. Finally, a series of particle-based quantitative methods were applied to investigate the evolution of particle-scale characteristics. In this phase, the blurred raw CT images were significantly enhanced by a novel filtering technique, providing a solid basis for subsequent particle segmentation and extraction. In addition, microcracks on CDG particles were successfully identified and extracted, serving as important reference features in the segmentation process. In the second phase, a novel pattern recognition method based on a neural network, PointConv, was developed to identify and track intact Leighton Buzzard sand (LBS) particles in a miniature triaxial sample. By integrating this approach with the PointNetLK network, both intact and fragmented LBS particles were successfully tracked throughout the triaxial test. In this phase, nearly all intact particles, as well as those that underwent moderate degree of breakage, were successfully tracked. The results indicate that particle rotation evolves markedly with increasing strain, primarily driven by local interactions such as friction and interlocking. Moreover, particle breakage leads to the formation of numerous small, irregular, and angular fragments, which are predominantly concentrated along the central axis of the specimen. In the next phase, the three-dimensional descriptor SHOT was employed to describe and encode the local morphological features of highly decomposed granite (HDG) particles. By leveraging the similarity in local shape features and geometric consistency, 10 HDG particles and their corresponding fragments were successfully matched in a short time. Based on this, a full-field tracking method named SHOT++ was proposed by combining the SHOT descriptor with the ICP-RANSAC algorithm. This approach facilitated a comprehensive investigation of particle fragmentation behaviors, including the evolution of the particle breakage ratio during shearing, the evolution and spatial distribution of various particle failure modes, and the correlation between coordination number and fragmentation patterns, etc. In the last phase, the crushing behavior observed in experiments was reproduced using the discrete element method (DEM). Specifically, the particle crushing events were modelled by replacing the particles that are identified to crush upon satisfying a prescribed particle crushing criterion with a group of fragments. The fragmentation pattern was governed by the coordination number of the crushed particle, which determined both the number and volume of the generated fragments.

Date of Award	24 Nov 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Jianfeng Jeff WANG (Supervisor)