Dams form critical components of civil infrastructure, as they play an essential role in flood control, hydroelectricity generation and water supply. The reliable and uninterrupted operation of dams throughout their designed lifespan bears regional and national impact. However, as any other infrastructure, dams are similarly vulnerable to ageing-related material degradation, extreme climatic events and overloading scenarios. In this context, for dam owners, operators and policy architects adequate maintenance and prediction of their structural condition become vital.

Data-driven Structural Health Monitoring (SHM) schemes are becoming particularly valuable for infrastructures that bear critical importance for modern societies. Historically, dam maintenance strategies have been founded on regular scheduled visual inspections. As the sensor technology and software-hardware interface advance, the opportunities for minimizing the shortcomings of the traditional approaches (e.g. manpower demand, inaccessibility, subjectivity and difficulty to predict) are quickly expanding. By exploiting sensory data from an instrumented dam structure, in addition to real-time structural diagnosis, Dam Health Monitoring (DHM) allows for development of a reliable prediction model.

This work focuses on testing the performance of a data-driven prognostic model based on the Polynomial Chaos Expansion (PCE) uncertainty quantification method. The model is applied on an experimental benchmark from a double curvature arch dam, located in the south of France and data collected from an instrumented double curvature arch concrete dam located in North Macedonia. Data handling, preparation of inputs, as well as specifics and characteristics of the developed models will be discussed. The presented results reveal the potential for robust real-time DHM and further integration into a multi-targeted diagnostic tool for operational infrastructure.