Assessing the structural health of strategic infrastructures such as concrete dams is of paramount importance for clean energy production and flood control in the current climate change context. Monitoring aimed at early detection of possible local failures is particularly important for long-standing and aging facilities situated in Alpine regions. A specific sensor network, usually installed in the dam body, collects a large amount of data related to the structural response to external actions, consisting mainly of seasonal variation in temperature and water level. Additional measurements could be made by drone-mounted equipment. The acquired information can be processed by different approaches, such as statistical models or machine learning tools that are trained to detect anomalous trends [1]. However, despite the tremendous improvement in the analysis tools available in recent years, some limitations are hard to overcome. First, the amount of data available on damaged dams is scarce and difficult to transfer from one situation to another. In fact, dams are resilient structures and almost all represent unique prototypes due to their peculiar geometry and environmental conditions. This problem can be partly overcome by the consideration of a digital twin of the investigated structure to reproduce the expected response under the conditions predicted to be most critical. However, the measurable quantities may show limited sensitivity to those parameters that allow proper calibration of the model and detection of local failures in real cases. This contribution illustrates and discusses these aspects by referring to some specific example. [1] G. Bolzon, A. Frigerio, M. Hajjar, C. Nogara, E. Zappa, Structural health assessment of existing dams based on non-destructive testing, physics-based models and machine learning tools, NDT&E International 2025; 150:103271.