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

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.