Skip to main content

Intelligent Characterization and Data-Driven Modeling of Granular Materials

Scientific and Technological Question: How to utilize machine learning and advanced experimental techniques to achieve precise characterization of microstructures and intelligent prediction of macroscopic mechanical behavior of granular materials?

Research Background​

Integrating artificial intelligence with experimental geomechanics, we are advancing granular material research from empirical description to data-driven quantitative prediction. This direction focuses on intelligent characterization, virtual generation, and mechanical performance prediction of particle morphology, providing high-quality data support and efficient analysis tools for computational geomechanics.

Intelligent Characterization and Data-Driven Modeling of Granular Materials
Intelligent Characterization and Data-Driven Modeling of Granular Materials

Conceptual schematic of intelligent characterization and data-driven modeling of granular materials (ClaudeBot generated)

Core Research Contents​

1. Deep Learning Image Reconstruction and Morphology Analysis​

Technical Breakthroughs:

  • CNN-based 3D Image Segmentation: Extracting particle morphology and pore structure from X-ray CT scans using convolutional neural networks and 3D image segmentation algorithms
  • Sub-voxel Level Analysis: Develop "machine learning image enhancement → level set segmentation" collaborative morphology reconstruction methodology, conquering sub-voxel level particle morphology analysis challenges
  • Extreme Environment Particle Characterization: Successfully applied to special granular material morphology digitalization including Martian regolith simulant and deep-sea coral sand

Application Outcomes: Established a full-process digital technology system for particle "morphology acquisition — morphology reconstruction — virtual generation".

2. Intelligent Virtual Particle Generation​

Technical Methods:

  • Spherical Harmonic Coefficient-Random Field Method: Establishing particle virtual generation methods combining spherical harmonic coefficients with random fields, revealing particle morphology spectrum evolution patterns
  • Diffusion Model-Driven Approach: Developing virtual porous granular material generation frameworks based on diffusion models, achieving intelligent generation of high-fidelity particle morphologies
  • Morphology Similarity: Generated virtual particles exhibit high morphology similarity with measured samples, breaking through limitations of physical experimental sample quantities

3. Data-Driven Constitutive Modeling​

Research Contents:

  • Morphology-Performance Correlation: Constructing machine learning mapping models between particle shape descriptors and macroscopic mechanical performance
  • Neural Network Contact Mechanics: Developing neural network-driven shape characterization and computational particle mechanics methods, achieving efficient contact detection and mechanical calculation based on SDF
  • Surrogate Model Acceleration: Establishing surrogate models for macroscopic responses of granular materials, enabling rapid parameter inversion and optimization design

4. Special Soils Research (e.g., Calcareous Sand)​

Research Objects:

  • Coral Sand: Establishing multiscale computational processes considering microscopic morphology effects for coral sand particles with intra-particle voids
  • Bio-cementation: Studying particle cementation mechanisms during Microbially Induced Calcite Precipitation (MICP) processes
  • Martian Regolith: Conducting DEM modeling of Martian regolith simulant particles, serving deep space exploration engineering

Key Findings:

  • Intra-particle void structures of coral sand significantly affect its compression and crushing behavior
  • Particle morphology irregularity is the key factor controlling mechanical response of calcareous sand

Research Significance​

Expanding applications of machine learning in computational mechanics, advancing granular material research from "empirical trial-and-error" toward quantitative data-driven prediction.

Representative Publications​

Deep Learning and Intelligent Computing​

  1. Li, C., Lai, Z.#, Huang, S., & Huang, L. (2026). Neural network-driven shape representation and computational particle mechanics via signed distance fields. Engineering Applications of Artificial Intelligence, 167, 113913. DOI | PDF

  2. Huang, S., Wang, P., Lai, Z.#, Yin, Z., Huang, L., & Xu, C. (2024). Machine-learning-enabled discrete element method: The extension to three dimensions and computational issues. Computer Methods in Applied Mechanics and Engineering, 432, 117445. DOI | PDF

  3. Lai, Z., Chen, Q., & Huang, L. (2021). Machine-learning-enabled discrete element method: Contact detection and resolution of irregular-shaped particles. International Journal for Numerical and Analytical Methods in Geomechanics, 46(1), 113-140. DOI | PDF

Virtual Particle Generation and Diffusion Models​

  1. Li, C., Huang, L., Lai, Z.#, Huang, S., & Lin, Y. (2026). A diffusion-based generative framework for virtual porous granular media generation. Powder Technology, 473, 122230. DOI | PDF

Particle Morphology Characterization and CT Image Reconstruction​

  1. Huang, S., Huang, L., Lai, Z.#, & Zhao, J. (2023). Morphology characterization and discrete element modeling of coral sand with intraparticle voids. Engineering Geology, 315, 107023. DOI | PDF

  2. Lai, Z., Chen, Q., & Huang, L. (2019). Reconstructing granular particles from X-ray computed tomography using the TWS machine learning tool and the level set method. Acta Geotechnica, 14(1), 1-18. DOI | PDF

  3. Lai, Z. & Huang, L. (2021). A polybezier-based particle model for the DEM modeling of granular media. Computers and Geotechnics, 134, 104052. DOI | PDF