PSI - Issue 84

Laura Dieci et al. / Procedia Structural Integrity 84 (2026) 591–598

593

The observed discrepancies in element dimensions, along with the bridge aging and repairs, make calibration of numerical models a valuable tool for understanding its actual structural behavior.

(a)

(b)

(c)

Fig. 1: “Ponte delle Grazie” - (a) side view 2024; (b) aerial view from Google Earth; (c) longitudinal profile and section.

3. Experimental tests This section provides an overview of the experimental investigations carried out on the bridge, including both dynamic and static tests. The dynamic campaign enabled the identification of the main modal parameters, such as natural frequencies and mode shapes, while the static load test provided additional information on the structural response through measured deflections. Together, these datasets form the experimental basis for validating and updating the numerical model. 3.1. Dynamic Identification Dynamic measurements were performed under ambient excitation conditions to characterize the vibration response of the bridge prior to the installation of the permanent monitoring system. Three sensing technologies were employed: a network of 14 uniaxial piezoelectric accelerometers; five tri-axial MEMS accelerometers of the same type later used for permanent monitoring; four Fiber Bragg Grating (FBG) optical accelerometers in different configurations. The sensors were installed at midspan, quarter-span, and additional symmetric locations along the deck to ensure observability of the dominant global and local modes. Data were collected over time windows of at least 1500 seconds, with sampling frequencies of 200 Hz for piezoelectric and MEMS sensors and 1000 Hz for the FBG system. Pre processing included signal demeaning and band-pass filtering in the range 0.5-40 Hz to isolate the modal content. Modal parameters were extracted using two standard operational modal analysis techniques: the Enhanced Frequency Domain Decomposition (Brincker et al. (2001)) and the covariance-driven Stochastic Subspace Identification (Peeters and De Roeck (1999)). Both methods yielded consistent results across the three sensing systems (Scocciolini et al. (2025)). The first bending mode was identified at approximately 4.4 Hz, followed by a torsional mode around 9.3 Hz and a local bending mode near 16 Hz. Across all technologies, discrepancies in natural frequencies remained below 2%, and the corresponding mode shapes exhibited MAC values above 0.80, indicating strong agreement. The natural modes identified from piezoelectric sensor measurements were adopted as the reference dataset for the subsequent calibration of the numerical model and are shown in Fig. 2.