PSI - Issue 62

Valentina Giglioni et al. / Procedia Structural Integrity 62 (2024) 887–894 Giglioni et al. / Structural Integrity Procedia 00 (2019) 000–000

893

7

corresponding mode shapes. The second reason is related to the influence of temperature variability on modal properties. Because of the clear relationship visible in Fig. 5a, damage clusters appear quite close to normal data, increasing thus the risk to be masked. Therefore, a careful removal of environmental factors may be useful to improve the reliability of the results and facilitate anomaly detection.

Fig. 6. DA results: transfer of M2 and M4 labels (a) and M1 and M3 labels (b) across bridge spans. The illustrated confusion matrices are computed by considering Span 1 and Span 2 as source and target domains, and vice versa. 4. Conclusions PBSHM enhances the capability to leverage information on structural health-state and therefore enable diagnostic capabilities across a population of similar structures. This recently-developed theory can be considered as an attractive solution towards the implementation of multi-asset SHM. Following a TL-based framework, the present paper suggests a new application of DA, where the attention is focussed on damage identification via knowledge transfer between different bridge spans. By taking advantage of the available damage labels for one span, the goal is to identify the same damage in different regions of the monitored bridge. According to the method, natural frequencies are transformed via DA, combining SA and kernel-based approaches, so that common health-state clusters can be identified within a latent space where the information of source and target domains are merged. Given that bridge