PSI - Issue 78

Akshay Rai et al. / Procedia Structural Integrity 78 (2026) 891–898

898

Table 1: Comparison of metrics for reconstructed dominant-SV-based anomaly detection (RBAD).

threshold Number of anomalies detected

Test damaged samples (total 122)

Test healthy samples (total 168)

Features

Accu racy

Sensi tivity

Speci ficity

F1 score

MCC

Percentile = 0.95

1. Reconstruc tionError

0.025

10

122

0.9655 1.00

0.9405 0.9606 0.9323

≈

2. OGSR 3. Cosine similarity

30.49

12

122

0.9586 1.00

0.9286 0.9531 0.9195

2.346

5

122

0.9828 1.00

0.9606 0.9849 0.9654

4.Wasser stein Distance

0.0921

19

122

0.9345 1.00

0.8869 0.9278 0.8760

5. KL Divergence 6. JS Divergence

38.79

6

122

0.9793 1.00

0.9643 0.9760 0.9587

1.566

14

122

0.9577 1.00

0.9167 0.9457 0.9086

environmental variations, minimising false positives during assessments. The framework was validated using real world data from the 14th-century Consoli Palace in Gubbio, Italy, particularly focusing on anomalies related to the May 15, 2021, earthquake. It demonstrates rapid deployment capabilities for immediate assessments after sudden events, allowing for timely anomaly detection. To evaluate CAE performance, six temperature-compensated spec tral metrics were developed, incorporating thermal history, frequency-domain errors, and statistical divergence measures for e ff ective reconstruction evaluation. The approach shows fast convergence to high accuracy and low false positive rates, crucial for real-time data analysis, supported by a strong F1-score for damaged samples. In summary, this framework o ff ers a robust, scalable, and interpretable solution for real-time SHM, providing essen tial early warning capabilities to protect historically significant masonry structures.

Acknowledgements

This study was supported by (i) FABRE – “Research consortium for the evaluation and monitoring of struc tures” in the FABRE-ANAS 2021-2026 program; (ii) AIDMIX - Artificial Intelligence for Decision Making; and (iii) MAT4BRIDGES - Machine Learning approaches for Structural Health Monitoring, both are funded by the University of Perugia. The opinions expressed in the paper do not necessarily reflect those of the funding agency.

References

[1] Moallemi, A., Burrello, A., Brunelli, D. and Benini, L., 2021, May. Model-based vs. data-driven approaches for anomaly detection in structural health monitoring: A case study. In 2021 IEEE International instrumentation and measurement technology conference (I2MTC) (pp. 1-6). IEEE. [2] Entezami, A., Sarmadi, H. and Behkamal, B., 2024. Short-term damage alarming with limited vibration data in bridge structures: A fully non-parametric machine learning technique. Measurement, 235, p.114935. [3] Giglioni, V., Venanzi, I., Poggioni, V., Milani, A. and Ubertini, F., 2023. Autoencoders for unsupervised real-time bridge health assessment. Computer-Aided Civil and Infrastructure Engineering, 38(8), pp.959-974. [4] Kita, A., Cavalagli, N. and Ubertini, F., 2019. Temperature e ff ects on static and dynamic behavior of Consoli Palace in Gubbio, Italy. Mechan ical Systems and Signal Processing, 120, pp.180-202. [5] Pellegrini, D., Barontini, A., Mendes, N. and Lourenc¸o, P.B., 2024. Dynamic Response of Masonry Structures to Temperature Variations: Experimental Investigation of a Brick Masonry Wall. Sensors, 24(23), p.7573. [6] Borlenghi, P., Saisi, A. and Gentile, C., 2024. Vibration monitoring of masonry bridges to assess damage under changing temperature. Devel opments in the Built Environment, 20, p.100555. [7] Falchi, F., Girardi, M., Gurioli, G., Messina, N., Padovani, C. and Pellegrini, D., 2024. Deep learning and structural health monitoring: Temporal Fusion Transformers for anomaly detection in masonry towers. Mechanical Systems and Signal Processing, 215, p.111382. [8] Carrara, F., Falchi, F., Girardi, M., Messina, N., Padovani, C. and Pellegrini, D., 2022. Deep learning for structural health monitoring: An application to heritage structures. arXiv preprint arXiv:2211.10351. [9] Katsigiannis, S., Seyedzadeh, S., Agapiou, A. and Ramzan, N., 2023. Deep learning for crack detection on masonry fac¸ades using limited data and transfer learning. Journal of Building Engineering, 76, p.107105. [10] Ibrahim, Y., Nagy, B. and Benedek, C., 2020. Deep learning-based masonry wall image analysis. Remote Sensing, 12(23), p.3918. [11] Zhu, Y., Ni, Y.Q., Jin, H., Inaudi, D. and Laory, I., 2019. A temperature-driven MPCA method for structural anomaly detection. Engineering Structures, 190, pp.447-458.