PSI - Issue 84

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

227

to discard non-physical solutions and were subsequently grouped via hierarchical clustering, using a cut-off distance of 0.030 and a minimum cluster size of 8–12 stable poles (depending on the setup). Global modal properties for each AVT were then obtained by clustering the per-setup results across setups using DBSCAN. The identified natural frequencies for AVT-Acc-1 and AVT-Acc-2 are reported in Table 1. As shown, both AVTs consistently captured three vibration modes, whose corresponding mode shapes are presented in Fig. 4.

Table 1. Natural frequencies identified across AVTs and FEM predictions. Mode # AVT-Acc-1 AVT-Acc-2 AVT-SG-1 AVT-SG-2

Mean (AVTs)

FEM 2.681 5.458 5.462

1 2 3

2.808 4.116 5.397

2.817 4.144 5.437

2.820

2.778 4.041 5.348

2.806 4.100 5.406

-

5.440

Fig. 4. Identified mode shapes: (a) first bending X, (b) second bending X, and (c) first bending Y.

Analogously, the pre-processed signals from SGs were used for modal identification, restricting the analysis to frequencies up to 10 Hz. COV-SSI algorithm was configured with time lags of 3.457 s (AVT-SG-1) and 3.238 s (AVT-SG-2), and model orders ranging from 2 to 100 were considered. For each model order, the estimated poles were screened using standard hard and soft criteria and subsequently grouped via hierarchical clustering, using a cut off distance of 0.030 and a minimum cluster size of five stable poles for AVT-SG-1, and 0.025 and six stable poles for AVT-SG-2. Figures 5(a) and 5(b) show the corresponding stabilization diagrams, and the identified natural frequencies are summarized in Table 1. As shown, AVT-SG-1 enabled the identification of two of the target modes, whereas AVT-SG-2 captured all three modes, consistently with the conventional accelerometer-based OMA results. Fig. 5. Stabilization diagram of the modal poles obtained from the COV-SSI for (a) AVT-SG-1, and (b) AVT-SG-2. The identified clusters of stable poles are highlighted with red dashed lines.

The comparison across the different AVTs indicates that the identified frequencies are consistent between tests. The largest discrepancies with respect to the mean value occur in AVT-SG-2, which systematically yields slightly lower values, with frequency variations ranging from -0.99% to -1.45% relative to the mean value. These differences can be partly attributed to the thermal variability during the monitoring campaigns. Table 2 reports an estimate of the