PSI - Issue 78

Marco Pirrò et al. / Procedia Structural Integrity 78 (2026) 1641–1648

1648

(as illustrated in Fig. 5). For example, Table 2 allows to accurately identifying the location of damage DMG1, which is introduced near sensor A3. As the extent of the damage increases (i.e., DMG2), the impact is primarily around channel A3, while in the DMG3 scenario, the area near channel A4 is more affected, thereby confirming the method’s effectiveness in pinpointing damage. 6. Conclusions This study presents a novel approach to anomaly detection that leverages continuous dynamic monitoring and a suitably configured SAE network. The method processes acceleration time series captured simultaneously by all available sensors to train the neural network. When structural behavior begins to diverge from the baseline established during training, the SAE's capacity to faithfully reconstruct the input signals declines, which may serve as a trigger for anomaly alerts. In contrast to traditional OMA-based methods, this SAE-driven approach offers several key benefits: (a) it enables the localization of damage; (b) it functions effectively with shorter time segments of input data; and (c) it inherently accounts for EOV in the dynamic responses without needing an explicit minimization step. The data gathered from the old ADA steel truss bridge in Japan were used to evaluate the proposed method and illustrate its effectiveness. Prior to the bridge’s decommissioning, artificial damage scenarios were introduced and are considered in the analysis. These experiments were conducted under constant temperature conditions, with the damage varying in both severity and location, including cases where the structure was partially restored to its undamaged condition. Compared to earlier research based on OMA, showing limitations in adopting modal parameters for detecting all the inflicted damages, the findings here confirm that the method accurately identifies the damage scenarios in terms of their presence, location, and degree of stiffness reduction. References Al-Ghalib, AA., 2022. Structural damage detection of old ADA steel truss bridge using vibration data.Structural Control and Health Monitoring 29(11): e3098. Borlenghi, P., Gentile, C., Pirrò, M., 2023. Continuous dynamic monitoring and automated Modal Identification of an arch bridge. In: Rizzo, P., Milazzo, A. (Ed.). European Workshop on Structural Health Monitoring, Springer, Cham. Chang, K.C., Kim, C.W., 2016. Modal-parameter identification and vibration-based damage detection of a damaged steel truss bridge. Engineering Structures 122, 156 – 173. Farrar, C., Worden, K., 2007. An introduction to structural health monitoring. Philosophical Transactions of the Royal Society A 365(1851), 303 – 315. Finotti, R., Gentile, C., Barbosa, F., Cury, A., 2022. Structural novelty detection based on sparse Autoencoders and control charts. Structural Engineering and Mechanics 81(5), 647 – 664. Goodfellow, I., Bengio, Y., Courville, A., 2016. Deep Learning. MIT Press, USA. Kim, C., Zhang, F., Chang, K., McGetrick, P., Goi, Y., 2021. Ambient and vehicle-induced vibration data of a steel truss bridge subject to artificial damage. Journal of Bridge Engineering 26 (7). Kingmaand, D., Ba, J., 2015. Adam: A method for stochastic optimization. In: Proc. 3rd International Conference on Learning Representations (ICLR), San Diego, CA, USA. Magalh ã es, F., Cunha, À., Caetano, E., 2012. Vibration based structural health monitoring of an arch bridge: from automated OMA to damage detection. Mechanical Systems and Signal Processing 28, 212 – 228. Ng, A., 2011. Sparse auto-encoder. CS294A Lecture Notes 72: 1 – 19. Pathirage, C., Li, J., Li, L., Hao, H., Liu, W., 2018. Application of deep autoencoder model for structural condition monitoring. Journal of Systems Engineering and Electronics 29(4), 873 – 880. Pirrò, M., Gentile, C., 2025. Detection and localization of anomalies in bridges using accelerometer data and sparse autoencoders. Developments in the Built Environment 23, 100715. Singh, D., Singh, B., 2020. Investigating the impact of data normalization on classification performance. Applied Soft Computing 97: 105524. Wang, Z., Cha, Y., 2021. Unsupervised deep learning approach using a deep auto-encoder with a one-class support vector machine to detect damage. Structural Health Monitoring 20(1), 406 – 425. Yang, L., Shami, A., 2020. On Hyperparameter Optimization of Machine Learning Algorithms: Theory and Practice. Neurocomputing 415.