PSI - Issue 84

Francesco Mariani et al. / Procedia Structural Integrity 84 (2026) 773–780

778

the system dynamics and enable effective filtering of environmental effects and anomalous measurements. One week of data was used to train the statistical model, while the remaining portion of the dataset was employed for the testing phase to evaluate the false positive rate and the model’s capability to detect real structural variations. Figure 5(a) illustrates the tracking of natural frequencies over the monitoring period from October 23rd to November 1st. The frequency trends exhibit a clear shift on October 29 th , indicating a notable structural change, corresponding to the barrier removal. Before this point, frequencies remain relatively stable with minor fluctuations, while after the event a significant drop is observed across all tracked modes, suggesting a reduction in stiffness. Figure 5(b) presents the results of an outlier detection analysis, where the T 2 scores remain low and stable before the barriers removal but show a sharp increase immediately after the marked time. This sudden rise in outlier scores aligns with the frequency shifts observed (summarized in Table 3) and serves as confirmation of abnormal structural behavior indicative of configuration change.

Fig. 4. Operational Modal Analysis results in the configuration no.5: PSD with identified modal frequencies, MAC matrix, and first five identified mode shapes.

Table 3. Comparison between stage 4 and 5 indentified frequencies. Stage 4 Mode 1 Mode 2 Mode 3

Mode 4 Bending

Mode 5 Lateral 22.105

Shape type Bending

Lateral

Torsional

f (Hz)

6.003

8.856

13.713

17.857