Uncertainty aware digital twins for structural health monitoring of civil infrastructures

This contribution synthesises the main outcomes of three recent research projects – IRIS, TECNODIGIT and AID-STRU – all centred on the development of digital twins for the monitoring and management of civil infrastructures, with a specific focus on the treatment of uncertainties. Across these projects, digital twins are conceived as hybrid, continuously updated models combining physics-based representations, data-driven components and multi-source sensing (fixed and mobile sensors, robotic inspection, Earth observation). The paper first classifies the main sources of uncertainty that affect digital-twin-based structural health monitoring: modelling assumptions (geometry, boundary conditions, material laws), measurement noise, environmental and operational variability, and incomplete or biased information in the historical record. The different strategies adopted in the three projects are then discussed, including Bayesian model updating, stochastic simulation of environmental loads, interval and sensitivity analyses, and the use of surrogate models to propagate uncertainty efficiently in near real time. Selected case studies – involving large building complexes, transport infrastructures and heritage structures – are used to show how uncertainty quantification improves damage detection, anomaly classification and the robustness of performance indicators. Particular attention is paid to decision-oriented metrics, such as confidence intervals on predicted responses, probability of detection and false alarm rates. The contribution concludes by outlining a set of methodological guidelines for integrating uncertainty treatment into the full life cycle of digital twins, from design and calibration to operation and maintenance planning.