PSI - Issue 84

Alberto Barontini et al. / Procedia Structural Integrity 84 (2026) 352–359

357

Table 1. Optimal sensor placements according to the analysed approaches. Sensor A1 A2 A3 A4 A5 A6 DOF Node DOF Node DOF Node DOF Node DOF Node DOF Node S1 10 5y 8 4y 6 3y 6 3y 14 7y 10 5y S2 16 8y 16 8y 16 8y 16 8y 16 8y 20 10y S3 29 15x 26 13y 26 13y 22 11y 18 9y 21 11x S4 31 16x 31 16x 29 15x 29 15x 20 10y 26 13y

Fig. 3. Optimal sensor placements according to the analysed approaches.

(a)

(b)

(c)

Fig. 4. A3 approach: (a) global probability of selection; (b) probability of selection conditional on the presence of location 29; (b) sensor affinity.

Notably, based on their probabilities of selection, the four sensors identified through affinity with location 29 are also the most frequently selected across the 4000 samples. Nevertheless, locations 10, 20, and 22 also appear frequently, both in the global selection count and conditionally on the selection of location 29. These locations are identified as optimal by other approaches as well, confirming their relevance. In particular, it is worth noting that the results of method A6 identify locations 10 and 20 as alternatives to locations 6 and 18, respectively. Interestingly, the sensor affinity analysis also reveals additional recurring configurations, albeit with lower overall recurrence. A configuration similar to A3 is proposed by all other methods with minor variations in the instrumented nodes, with the exception of A1 and A5. Methods A1 and A2 do not account for sampling of the uncertain parameters; instead, they rely on a correlated covariance matrix to mitigate sensor clustering. Method A1 yields a comparatively less intuitive configuration, featuring two closely spaced vertical sensors and two horizontal sensors. It is worth noting that