Physics-Informed Modular Digital Twin for Real-Time Prediction in Mechanized Tunnelling

The growing demand for tunnel excavation is directly linked to urbanization, with over 75% of Europe’s population already living in cities and this figure projected to exceed 80% by 2050. As urban areas expand, the need to extend infrastructure underground becomes critical. Subsurface development, however, introduces significant geotechnical risks, where even minor delays or failures can result in severe economic and societal consequences. Real-time predictive modelling tools are essential to ensure safety and resilience.Despite advances in modelling, predicting ground behaviour in real time remains a major challenge. Excavation involves coupled processes, soil displacement, pore pressure changes, and complex machine-ground interactions, that evolve dynamically. Traditional numerical methods are accurate but computationally intensive and lack real-time adaptability. Machine learning (ML) approaches are faster but often operate as opaque black boxes, requiring large datasets and offering limited generalizability.To address this gap, this project proposes the Physics-Informed Modular Excavation (PIMEX) framework: a real-time, modular system that integrates physics-based models with advanced ML. By embedding physical laws and constraints within neural network architectures such as PINNs, ConvLSTMs, and GNNs, PIMEX will enable interpretable, predictive simulations of excavation processes under sparse and noisy data.The framework will support real-time control, enhancing safety, efficiency, and decision-making in mechanized tunnelling. Validation will be performed through large-scale infrastructure datasets and international collaborations, ensuring the robustness and transferability of the approach. Ultimately, PIMEX aims to deliver a next-generation predictive system for underground excavation, bridging physics and data-driven methods to reduce risks, optimize operations, and contribute to safer, more cost-effective infrastructure delivery under evolving conditions.