Issue 78

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

Dataset B (Fig. 7). The first two event candidates occur in the first half of the record and are followed by partial recovery; the third event candidate is associated with a stronger transition after which the response exhibits more persistent activity and reduced reconvergence compared with Dataset A. Under a strictly response-based interpretation, this behavior is compatible with a transition toward a more sustained loaded-response condition after Event 3, but it cannot be uniquely attributed to a specific operational action without synchronized external references. In both datasets, a slow low-frequency component is visible in the baseline-normalized time histories. This component may include thermal contributions, slow strain-transfer effects, and quasi-static mechanical response. For the spectral tracking reported below, this component was reduced by detrending and high-pass filtering; therefore, the subsequent interpretation focuses on transient and mid frequency descriptors rather than on long-duration or quasi-static strain trends. Time–frequency maps and comparison of frequency families across datasets Time–frequency representations reveal spectral “families”, namely persistent narrowband components whose prominence changes around Events 1–3, consistent with the event-driven changes observed in the time domain. Spectrograms were computed using (i) a conventional Hann-window STFT and (ii) a Thomson multitaper estimate, under matched window length and hop. Figures 8 and 9 present multitaper spectrograms for sensors S3 and S4 for Dataset A and Dataset B, respectively. Across both datasets, the maps show persistent bands within the analysis range, with transient broadband energy increases during the event-centered windows and clearer emergence or strengthening of narrowband components during high-activity phases. The key comparative point is that several frequency families are observed in both representative datasets, supporting recurrence of sensor-dependent spectral content within the analyzed records rather than statistically validated repeatability over a population of landings. Figures 10 and 11 show the corresponding Hann STFT spectrograms for the same sensors and parameters. Compared with the multitaper estimate, the single-taper Hann STFT shows greater background fluctuation when the same frame length and hop are used. As a result, frame-wise identification of the dominant peak becomes less stable, particularly when nearby frequency components have similar amplitudes. It should be noted that the maps in Figs. 8–11 were computed using the main configuration, L=2048 samples and H=512 samples, corresponding to a frame duration of 2.048 s. Consequently, abrupt changes occurring around a high-energy event transition may be temporally smeared within a single estimate, because pre-event and post-event responses can contribute to the same local spectrum. Accordingly, these maps are more appropriately interpreted as descriptors of evolving spectral families within event-centered windows, rather than as instantaneous modal identifications or independently validated touchdown markers. When finer localization of brief transitions is required, the shorter configuration, L=1024 and H=256, provides improved time localization, although at reduced frequency resolution. In Figs. 8–11, the shaded bands mark the FBG-derived event centered windows defined from the multichannel energy proxy. These overlays are shown for visual reference only and do not alter the underlying time–frequency estimates or constitute external validation of the detected events.

Figure 8: Dataset A: Multitaper time–frequency maps (DPSS; NW=3, K=5) for sensors (a) S3 and (b) S4. Spectrograms computed with a Thomson multitaper estimate using the same frame length and hop as the Hann STFT (L=2048 samples, H=512 samples). Narrowband frequency families are visible within the analyzed record and change in prominence around the FBG-derived event-centered windows.

531