3D Reconstruction of Geomaterial Microstructure Based on Differentiable Descriptors

Zijie He^1,2Yu Ning^3*Yutao Zhang⁴Qingfu Huang³Yuang Zhou^1,2Qingxiang Meng^1,2*

• ¹Key Laboratory of Ministry of Education for Geomechanics and Embankment Engineering, Hohai University, Nanjing, China
• ²Hohai University Geotechnical Research Institute, Nanjing, China
• ³Kunming Engineering Co. Ltd., Power Construction Corporation, Kunming, China
• ⁴University of California Berkeley, Berkeley, United States

The microstructure of geomaterials plays a crucial role in determining their physical and mechanical properties. The complex mechanical behavior of certain coarse-grained geomaterials significantly affects their resistance to deformation, making an accurate characterization of their internal structure essential. However, obtaining core samples of such materials is often costly and labor-intensive, whereas acquiring two-dimensional (2D) structural information is more feasible. This study presents a microstructure reconstruction technique based on a differentiable optimization framework, wherein the reconstruction process minimizes the error of a given descriptor while considering its derivative. The proposed method enables reconstruction using either a single 2D slice or three orthogonal 2D slices. Furthermore, the effectiveness of the 2D-to-3D reconstruction is validated through actual computed tomography (CT) scans of coarse-grained geomaterial samples collected from the Xinjiang region.