Abstract:The morphology of soil particles is crucial in determining its granular characteristics and assembly responses. However, how to introduce accurate and various morphologies of realistic particles in modeling can be challenging, as it often requires time-consuming and costly X-ray Computed Tomography (XRCT). This has led to two prevalent problems in modeling: morphological reconstruction and generation. For reconstruction, we develop a geometric-based Metaball-Imaging algorithm. This algorithm is capable of accurately reconstructing the complex morphologies of realistic particles, including those with concave voids, which cannot be easily represented using other shape descriptors such as the spherical harmonic function. It employs a two-step approach, capturing the main contour of the particles using a series of non-overlapping spheres and then refining surface-texture details through gradient search. Four types of soil particles, hundreds of samples, are applied for evaluations. The result shows good matches on key morphological indicators (i.e., volume, surface area, sphericity, circularity, corey-shape factor, nominal diameter and surface-equivalent-sphere diameter), confirming its reconstruction precision. For generation, we propose the Metaball Variational Autoencoder. Assisted by deep neural networks, this method can generate new 3D particles in Metaball form, while retaining coessential morphological features with parental particles. Additionally, this method allows for control over the generated shapes through an arithmetic pattern, enabling the generation of particles with specific shapes. Two sets of XRCT images different in sample number and geometric features are chosen as parental data. On each training set, one thousand particles are generated for validations. The generation fidelity is demonstrated through comparisons of morphologies and shape-feature distributions between generated and parental particles. Examples are also provided to demonstrate controllability on the generated shapes. With the Metaball-based simulation framework previously proposed by the authors, these methods can incorporate the real shape of particles into simulations, making it possible to simulate a large number of soil particles with varying shapes and behaviors. Together, they have the potential to provide valuable insights into the properties and behavior of actual soil particles.

Generation of 3D realistic geological particles using conditional generative adversarial network aided spherical harmonic analysis

A New Particle Generation Method for Arbitrary 2D Geometries in SPH Modeling

3D-Aware Image Synthesis Via Learning Structural and Textural Representations

SDF-based 3D Generative Adversarial Networks for Image and Geometry Generation

Randomly Generating Three-Dimensional Realistic Schistous Sand Particles Using Deep Learning: Variational Autoencoder Implementation

Synthesizing realistic sand assemblies with denoising diffusion in latent space

Synthesising microstructures of a partially frozen salty sand using voxel-based 3D generative adversarial networks

Characterization and Generation of 3D Realistic Geological Particles with Metaball Descriptor based on X-Ray Computed Tomography

A novel random generation method of two-dimensional particles based on the complex Fourier series

Generating Realistic Geology Conditioned on Physical Measurements with Generative Adversarial Networks

A systematic framework for DEM study of realistic gravel-sand mixture from particle recognition to macro- and micro-mechanical analysis

3DMaterialGAN: Learning 3D Shape Representation from Latent Space for Materials Science Applications

Generation of non-stationary stochastic fields using Generative Adversarial Networks

Three-dimensional reconstruction of granular porous media based on deep generative models.

Deep Learning-Based Reconstruction of 3D Morphology of Geomaterial Particles from Single-View 2D Images

Reconstruction and generation of 3D realistic soil particles with metaball descriptor

Predicting 3D particles shapes based on 2D images by using convolutional neural network

Parametrization and generation of geological models with generative adversarial networks

Stochastic reconstruction of shale combining multi-scale generators and discriminators with attention mechanisms

Intelligent Generation of Cross Sections Using a Conditional Generative Adversarial Network and Application to Regional 3D Geological Modeling

GSGAN: Adversarial Learning for Hierarchical Generation of 3D Gaussian Splats