Abstract:Intelligent tunnel engineering requires accurate and comprehensive digital twin models to represent complex geological environments. The digital twin model of tunnel geological environment has multi-level and diversified in-depth applications such as problem diagnosis, risk assessment, trend prediction, emergency response, etc. Since tunnel construction is a long-term continuous dynamic spatial–temporal progress, the tunnel face constantly advances during excavations. Thus, the geological conditions of the construction surface and surrounding areas are continuously revealed and updated, which causes the digital twin model to have a very high update frequency of the geological information in tunnel engineering. The existing methods fail to represent and update the geological environment comprehensively and in real-time, especially for those long deep tunnels with the complicated topology of geological structures and the heterogeneous distribution of internal properties. Due to the digital twin application having a high-level requirement of completeness, accuracy, and timeliness, 3D modeling of tunnel geological environments has remained challenging within digital twin application contexts. This study aims to deliver a solution that efficiently represents and updates the geological structures and internal non-uniform properties of tunnel engineering. Since dense voxels can represent inner intricate heterogeneous property information, sparse voxels can efficiently represent complex structural information. Meanwhile, voxels using Volumetric Dynamic B + trees (VDB) data structure are easy to integrate and update. However, according to the characteristics of long strip distribution of tunnel geological environments, directly using the VDB will still cause an efficiency problem. Therefore, this study proposes an efficient linear segmentation-based multi-level voxel representation method for the prolonged deep tunnel geological environment using the VDB data structure to support dynamic updates. The typical Mountain Tunnel of China is employed for experimental analysis to validate the proposed method, and four representative data with different modalities are integrated into the digital twin model. The results demonstrate that spatial efficiency augments 28.49% after segmentation, and data access with O(1) time complexity supports efficiently dynamic updates.

ViRFD: a Virtual-Realistic Fused Dataset for Rock Size Analysis in TBM Construction.

Multi Mask Learning of Stone Segmentation for Auto-Monitoring System in TBM Construction

Self-intersection Attention Pooling Based Classification for Rock Recognition

Rock segmentation visual system for assisting driving in TBM construction

Benchmarking Large-Scale Multi-View 3D Reconstruction Using Realistic Synthetic Images

Rock fragmentation indexes reflecting rock mass quality based on real-time data of TBM tunnelling

Prediction of geological conditions for a tunnel boring machine using big operational data

Deep learning-based pixel-level rock fragment recognition during tunnel excavation using instance segmentation model

A data driven real-time perception method of rock condition in TBM construction

Automated rock mass condition assessment during TBM tunnel excavation using deep learning

RockGPT: reconstructing three-dimensional digital rocks from single two-dimensional slice with deep learning

Segmentation of X-Ray Images of Rocks Using Deep Learning

RockFormer: A U-Shaped Transformer Network for Martian Rock Segmentation

Vision-based size distribution analysis of rock fragments using multi-modal deep learning and interactive annotation

Multi-level voxel representations for digital twin models of tunnel geological environment

Zero-Shot Digital Rock Image Segmentation with a Fine-Tuned Segment Anything Model

A real-time intelligent classification model using machine learning for tunnel surrounding rock and its application

Accurate Rock-Mass Extraction From Terrestrial Laser Point Clouds via Multiscale and Multiview Convolutional Feature Representation

Intelligent Algorithm for Rock Core RQD Based on Object Detection and Image Segmentation to Suppress Noise and Vibration

RockGPT: Reconstructing three-dimensional digital rocks from single two-dimensional slice from the perspective of video generation

Targeted Rock Slope Assessment Using Voxels and Object-Oriented Classification