PSI - Issue 44

Valentina Giglioni et al. / Procedia Structural Integrity 44 (2023) 1948–1955 Valentina Giglioni et al./ Structural Integrity Procedia 00 (2022) 000–000

1954

7

is applied on the pseudo-accelerations stemming from the FE model of the Z24 bridge. After the calibration of the model, several time-histories analysis are carried out to generate a sufficient number of data reproducing response accelerations during environmental and operational conditions. A realistic earthquake-induced damage scenario affecting both piers is simulated by reducing the concrete’s elastic modulus at the bottom of the elements and close to the connections with the deck. Four sensors are employed to extract response information in both longitudinal and transversal directions. All the generated acceleration sequences are fed into the autoencoder for being reconstructed and then, a specific damage index, based on the estimated reconstruction error, is properly selected to provide sequences labels for anomaly detection. Results remark the model capability in detecting abnormal conditions at the level of 2-minutes-length macro-sequences, yielding a significant minimization of false negative errors. Therefore, showing high performance, it can be assumed that the proposed technique is supposed to be able to detect different damage scenarios with a lower intensity, thereby representing a valid alternative to the classic OMA approaches and an effective tool for real-time post-earthquake assessment and for promptly identifying local damages with a limited number of sensors.

Fig. 3. Control charts obtained by analyzing 2-minutes-length macro-sequences for the four considered sensors.

Acknowledgements This work is supported by the funded project "Math4Bridges-Machine Learning and Transfer Learning approaches for Structural Health Monitoring and risk assessment of bridges" within the program "Fondo Ricerca di base di Ateneo, edizione 2020" of the University of Perugia.