To enable resilient and sustainable underground infrastructure, current practice must evolve toward integrated Digital Twins (DTs) that support automated reconstruction, continuous condition assessment, and predictive analysis. Acting as a real-time single source of truth, DTs consolidate heterogeneous data into coherent digital representations of underground assets, enabling informed design decisions and proactive lifecycle management. We propose a DT reconstruction framework that emphasises automation and seamless coupling between digital and numerical models. The approach covers four core stages: automated component detection, digital model reconstruction, condition and defect mapping, and structural capacity assessment. Tunnel point clouds are converted into images and processed using the Segment Anything Model (SAM) for zero-shot segmentation, enabling reliable extraction of tunnel components without extensive labelled data [1]. Geometric features are then derived to reconstruct parametric digital models and support deformation and displacement monitoring. Detected defects are automatically embedded in the digital model, allowing quantitative assessment of their severity. Predictive assessment is achieved by linking the reconstructed DT to high-fidelity meshless numerical models, whose mesh-agnostic nature is well suited to automated DT workflows. Damage is incorporated through material stiffness degradation governed by prescribed damage evolution laws. Higher-order meshless formulations ensure accurate geometric representation while significantly reducing computational cost, enabling efficient structural analysis and capacity estimation [3]. Robustness is ensured through a computational strategy combining local sub-stepping with global error control. By integrating automated digital reconstruction with efficient meshless analysis, the proposed framework provides an accurate, scalable, and robust DT solution, advancing proactive and data-driven management of underground infrastructure.. REFERENCES [1] A. Kirillov, et al. "Segment anything." Proceedings of the IEEE/CVF International Conference on Computer Vision. 2023. [2] Y. Zehao, J. Ninic et al. "Sam-based instance segmentation models for the automation of structural damage detection." Advanced Engineering Informatics 62 (2024): 102826. [3] J. Ninić, et al. "BIM-to-IGA: A fully automatic design-through-analysis workflow for segmented tunnel linings." Advanced Engineering Informatics 46 (2020): 101