Issue 78

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

should be considered response-based rather than independently validated. To ensure data integrity in an operational environment, the installation was inspected before and after flights with attention to bond integrity, chafing, connector security, and the absence of interference with moving assemblies (e.g., gear articulation, tires, brake components).

Figure 2: (a) FBGs (S1-S4) location; (b) data acquisition system installed on the rear seat.

Data import, integrity screening, and quality controls Raw wavelength time series were imported in native form without resampling. Because operational recordings can include intermittent dropouts, saturation artifacts, and isolated discontinuities, a channel-wise integrity screening and quality control procedure was applied prior to conversion and spectral analysis. The screening was executed sequentially as follows. (i) Range and desaturation checks. Samples were flagged as invalid when wavelength values were non-physical or indicative of interrogator desaturation. Conservative bounds were adopted by excluding data at or below 0 nm and values outside the 1500–1600 nm band, which covers the expected operating region for the installed gratings. (ii) Isolated spikes were attenuated by applying a short median operator, with a width of 3–5 samples, to the first difference of the wavelength series. This step was intended to reduce impulsive acquisition artifacts while preserving genuine high dynamic event responses. (iii) Gap handling. Short gaps (up to 10 samples) were interpolated to preserve continuity for subsequent windowed spectral analysis. Longer outages were retained as exclusions to avoid contaminating spectra with extended interpolated segments. (iv) Frame-level quality gates were applied before time–frequency feature extraction. For each analysis frame, a time-domain signal-to-noise ratio was defined as         , , 10 , 0 RMS SNR 20log RMS j m j m j x n x n          (1) where   , j m x n is the de-meaned, detrended, and high-pass filtered microstrain segment of sensor (j) in the current frame (m), and   , 0 j x n is the corresponding high-pass filtered microstrain segment extracted from the initial quiet interval used to estimate the baseline noise floor. The operator ሺ∙ሻ denotes the root-mean-square value. Frames with ൏ 10 dB were excluded from ridge tracking and descriptor computation. Frames containing more than 5% interpolated samples were also discarded. This quality-gating step was treated as part of the measurement workflow rather than as optional post processing. Its purpose was to prevent ridge extraction and feature computation from being driven by repaired segments, residual artifacts, or near-baseline intervals with insufficient strain energy. The resulting time–frequency descriptors were therefore restricted to frames in which the response remained measurable and interpretable under field-acquired conditions.

523