PSI - Issue 84

Ana Avramova et al. / Procedia Structural Integrity 84 (2026) 560–568

562

(a)

(b)

Fig. 2 Typical cross-sections of (a) VI03 bridge and (b) VI04 bridge (dimensions in cm).

Fig. 3 Sensors’ layout adopted in the continuous dynamic monitoring of VI04 (top) and VI03 (bottom) viaducts.

Ambient vibration tests were performed before the installation of the dynamic monitoring system. During those tests, carried out in early March 2023, the sensor layout subsequently adopted in the monitoring was defined: as shown in Fig. 3, 28 MEMS accelerometers were installed on the top of the lower flanges of each bridge. 3. Software tools The acceleration response is continuously acquired at a sampling frequency of 200 Hz in hourly datasets. Each raw data file is pre-processed, and the preliminary processing includes signal detrending, computation of peak and RMS values, application of a low-pass filter with a cutoff frequency of 10 Hz, down-sampling the time series to 20 Hz, and organising a database suitable for modal identification. Modal parameter estimation (MPE) and subsequent modal tracking (MT) were performed through the MATLAB based software package DyMoND (Dynamic Monitoring and Novelty Detection, Gentile et al., 2024). The MPE involves the detection of physical modes in the stabilization diagram generated by the covariance-driven Stochastic Subspace Identification (SSI-Cov) algorithm (Peeters & De Roeck, 1999). Firstly, the physical consistency of poles is investigated by checking the damping ratios and the complexity of modal components through the Mean Phase Collinearity (MPC) (Pappa et al., 1993). Subsequently, physically consistent poles—i.e., the poles exhibiting similar frequencies and mode shapes—are clustered together. Once the modes have been identified within each hourly dataset, the time evolution of modal properties is generated after the comparison between a reference set of modal parameters and the currently extracted ones. The OMA-based Structural Health Monitoring (SHM) metrics mainly include natural frequencies, the MAC correlation (Allemang & Brown, 1982) between the current mode shapes and the reference ones, and Mean Phase Collinearity (MPC) (Pappa et al., 1993), quantifying the complexity of current mode shapes. To detect structural anomalies, the evolution of natural frequencies should be carefully investigated, with a particular focus on isolating and mitigating the influence of environmental and operational variability (EOV). An effective approach for addressing this challenge involves the application of techniques such as the Principal Component Analysis (PCA) (Jolliffe, 1986), which enables the identification of long-term relationships among non stationary modal parameters. If the monitored structure does not change, the PCA-based regression model— established during an appropriate training period— still holds for the newly identified natural frequencies. The