PSI - Issue 84

Antonio S. López-Cuervo et al. / Procedia Structural Integrity 84 (2026) 223–230

226

were 13 min (AVT-SG-1) and 16 min (AVT-SG-2), with sampling frequencies of 78.97 Hz and 84.30 Hz, respectively. The strain signals were pre-processed through detrending, spike removal, and a 4th-order Butterworth band-pass filter (2–30 Hz), analogously to the accelerometer case. Figures 2(c) and 2(d) show the resulting filtered time series for a representative channel in AVT-SG-1 and AVT-SG-2, respectively. 2.3. Sensor placement Sensor locations were determined using the Effective Independence (EfI) algorithm, first introduced by Kammer (1991). EfI is a greedy optimal sensor placement algorithm that selects locations among candidates to maximize the linear independence of target mode shapes. Given the modal matrix restricted to the current candidate set, EfI builds the Fisher Information Matrix = ⊤ and derives a per-DOF contribution index from its eigen decomposition. At each iteration, the DOF with the lowest contribution is discarded, and the procedure continues until sensor locations remain. For the accelerometer-based AVTs, a two-setup configuration was used for each campaign. Reference sensor positions were first optimized (three references for AVT-Acc-1 and two references for AVT-Acc-2) using FEM predicted target modes, and the remaining roving sensors were subsequently optimized in a second stage. This procedure was automated using the OSP-SAP software (López-Cuervo et al., 2025). Fig. 3(a) shows the resulting layout for the nine accelerometers used in AVT-Acc-1, whereas Fig. 3(b) shows the corresponding layout for the eight accelerometers in AVT-Acc-2. For the strain-based AVTs, two placement strategies were adopted. In AVT-SG-1, EfI was applied in its standard form to select eight strain-gauge locations (Fig. 3(c)). In AVT-SG-2, the candidate set was restricted to DOFs with high modal strain for the FEM target modes, and EfI was subsequently applied over this reduced set.

Fig. 3. Sensor positions: roving and reference accelerometers for (a) AVT-Acc-1 and (b) AVT-Acc-2, and strain gauges positions in (c) AVT-SG 1, and (d) AVT-SG-2. For strain gauges, the arrows indicate the surface perpendicular to which the gauge is installed.

3. Results and discussion 3.1. Dynamic identification

Modal identification for each accelerometer-based AVT was conducted in MOVA software (García-Macías and Ubertini, 2020) using the covariance-driven stochastic subspace identification (COV-SSI) algorithm applied to the pre-processed signals. A separate identification was performed for each setup, and the analysis was restricted to frequencies up to 10 Hz. COV-SSI was configured with time lags between 1.285 s and 2.570 s and model orders ranging from 2 to 140. For each model order, the estimated poles were first filtered using standard hard and soft criteria