Issue 78

C. Vendittozzi et alii, Fracture and Structural Integrity, 78 (2026) 519-540; DOI: 10.3221/IGF-ESIS.78.30

remain dominated by low-end components, with ridge medians clustered near 18.1–18.6 Hz in all three event-centered windows and relatively small dispersion, particularly for S2. Sensor S3 retains a mid-frequency ridge family near 150.4 Hz in Event 1 and 149.9 Hz in Event 2, but with broad ridge dispersion, with IQR values of approximately 132.8 Hz and 132.7 Hz, respectively. This broad dispersion indicates switching or competition between multiple spectral families within the same event-centered windows. In Event 1, the distinction between ridge-based and band-limited descriptors is especially important: although the ridge median lies near 150 Hz, the RMS-dominant band is 350–500 Hz, indicating that the integrated spectral contribution in the high-frequency interval is larger than that associated with the ridge median alone. Sensor S4 shows the clearest high-frequency persistence in Dataset B. The ridge median remains near 427.7 Hz in Event 1 and 426.8 Hz in Event 2, with narrow IQR values of approximately 4.2 Hz and 2.4 Hz, respectively, while ridge persistence in the 350–500 Hz RMS-dominant band reaches 100% in both windows. This indicates a stable high-frequency ridge family during the earlier event-centered windows. The third event in Dataset B departs from this pattern. In that window, the ridge medians of S3 and S4 shift toward the low end of the analyzed band, approximately 18.6 Hz, and sensor S4 exhibits a very broad ridge dispersion, with IQR approximately 415.0 Hz. This behavior indicates stronger switching between competing spectral contributions within the same event-centered window and a less stationary post-event response than that observed in Dataset A. Taken together, the quantitative descriptors show that the two representative datasets do not exhibit identical trajectories or identical event responses. Rather, they show recurrence of comparable sensor-dependent frequency regions, combined with changes in persistence, dispersion, and relative spectral weighting across event-centered windows. In both datasets, the same broad structure is recovered: low-end dominant content in S1–S2, a mid-frequency family near 150 Hz in S3, and a high-frequency family near 427–429 Hz in S4, especially in the earlier event windows. These observations should be interpreted as case-study evidence of recurrent spectral organization within the analyzed recordings, not as population level repeatability or validated modal identification. To facilitate interpretation of the ridge-based and band-limited descriptors, the main recurrent frequency families and their sensor dependence are summarized in Table 4.

Most evident families

Shared across sensors or sensor-specific

Sensor

Most evident in

Behavior

Possible interpretation

Mostly shared with S2

Datasets A and B, all events Datasets A and B, all events Dataset A Events 1– 2; Dataset B Events 1–2 Dataset A Event 1; Dataset B Events 1– 2

Persistent low-end family Persistent low-end family Mid-frequency recurring family, variable persistence High-frequency stable family, strong persistence

Global/low-order response component Global/low-order response component Local structural participation near sensor location / load-path sensitivity Localized high-frequency response, sensor-position dependent

S1

~10–20 Hz

S2

~10–20 Hz

Mostly shared with S1

Partly sensor-specific, occasionally shared weakly

S3

~150 Hz

S4

~427–429 Hz

Strongly sensor-specific

Table 4: Summary of recurrent frequency families and sensor dependence within the two representative datasets. The proposed physical interpretations are qualitative and require future validation using synchronized operational references and/or independent modal information. As shown in Table 4, the overall response is structured into a limited number of recurrent frequency regions, some of which are shared across sensors whereas others remain more strongly sensor-dependent. The association of these regions with global or local structural participation should therefore be considered physically plausible, but not definitive. Per-event power spectra as compact event-level spectral summaries (Datasets A and B) To complement the time-resolved spectrograms, per-event power spectra, namely periodograms, were computed over each FBG-derived event-centered window. For numerical stability and consistent frequency resolution, the window duration was capped at 60 s, thereby preventing excessively long FFTs. These spectra provide compact frequency-domain summaries for each event-centered window and facilitate comparison among Events 1–3 within the same acquisition and between the two representative datasets. Figures 13 and 14 report the per-event power spectra for sensor S3, used here as a representative example, for Dataset A and Dataset B, respectively. Within each dataset, the persistence of characteristic peaks or bands across event-centered windows indicates recurrence of similar spectral content within the analyzed acquisition. This

535