Answers are grounded in this paper and cite their sources.
Deformation monitoring is critical for risk management in underground tunnel excavations, yet existing point cloud methods often struggle with complex irregular tunnel surface geometries and manual parameter selection. This study proposes NIFCyl, a novel unsupervised neural implicit learning framework for full-field tunnel deformation quantification. It learns a Neural Implicit Function to approximate tunnel surface geometry and generate scale-invariant normal fields from point clouds, coupled with a cylinder-based module for robust spatial averaging. In contrast to previous deep learning approaches, NIFCyl requires no ground-truth labels and eliminates the reliance on geometric fitting of local regions or specific surface roughness assumptions. Comparing it with the benchmark method-M3C2 algorithm on the synthetic deformation scenarios demonstrates that NIFCyl achieves superior accuracy (R 2 = 0.963, RMSE = 0.009 m) and a five-fold increase in computational efficiency. Experimental field tests in an underground metal mine further confirm that NIFCyl outperforms M3C2 on curved surfaces by 6% in mean absolute error. A case study in an active mining tunnel reveals that NIFCyl successfully captures real-world deformations caused by scaling work. Systemic uncertainty analysis further demonstrates its robust noise control ability, establishing it as an accurate, efficient, and automated digital tool for 3D full-field deformation monitoring in complex underground tunnels.
No takes yet. Share an insight, caveat, or question.
Han et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: