PSI - Issue 78

Ivan Roselli et al. / Procedia Structural Integrity 78 (2026) 128–136

135

For each DI cluster a centroid and standard deviation bars of ORSR and MSE were calculated. The distance of each DI cluster centroid from the undamaged case (DI centroid distance) was calculated as shown in Fig. 5. The achieved values of DI centroid distance were remarkable correlated with DI values. The found regression line (Fig. 6) has a slope of 3.9, intercept of 0.8 and very high coefficient of determination (R 2 = 0.9445). Moreover, the found correlation was very interesting and particularly meaningful for moderate and severe damage cases (DI in the range 0.5-0.9). Conversely, the correlation is little significant for low damage scenarios (at DI values lower than 0.5 we have only one point). It is worth noting that at DI = 0 the DI centroid distance is zero by definition, so the regression curve should have a very small intercept. 4. Conclusions The present study demonstrates the promising potential of utilizing the proposed AI procedure based on the CVAE technique to analyze ambient vibration data for quantifying the level of damage in structures, in alternative to methods based on modal analysis. The results were validated through experiments conducted on shaking table, in which data of white-noise vibration (representing approximation of ambient vibration in outdoor environment) of a reinforced concrete frame were analyzed. The following conclusions can be drawn from the experimental results achieved in this work:  The proposed AI procedure provided a good accuracy in the reconstruction of vibration time history signals of the training dataset (representing the structure in undamaged conditions, DI = 0).  At DI values higher than zero, the reconstruction of the vibration time history signals was less satisfactory (test dataset at different levels of damage), as expected.  The higher the DI value, the less satisfactory is the achieved reconstruction, which essentially leads to higher ORSR and MSE values. Consequently, the cluster of the undamaged case (DI = 0) tends to cover an area closer to the origin of the axes in the MSE-ORSR plane. While clusters of higher DI values tend to be positioned gradually further from the origin. This confirms the potentiality of the MSE-ORSR plane to quantify damage, not only to discriminate undamaged from damaged cases.  The DI centroid distance proved to be a good parameter to quantify the level of damage: the linear regression with true DI values gave a slope of 3.9, intercept of 0.8 and very high coefficient of determination (R 2 > 0.9). Ahmed, T., Longo, L., 2020. Examining the Size of the Latent Space of Convolutional Variational Autoencoders Trained With Spectral Topographic Maps of EEG Frequency Bands. IEEE Access 10, 99, 107575-17586. DOI: 10.1109/ACCESS.2022.3212777. Azzara, R.M., De Roeck, G., Girardi, M., Padovani, C., Pellegrini, D., Reynders, E., 2018, The influence of environmental parameters on the dynamic behaviour of the San Frediano bell tower in Lucca. Engineering Structures 156, 175-187. DOI: 10.1016/j.engstruct.2017.10.045 Cataldo, A., Roselli, I., Fioriti, V., Saitta F., Colucci, A., Tati, A., Ponzo, F. C., Ditommaso, R., Mennuti, C., Marzani, A., 2023. Advanced video based processing for low-cost damage assessment of buildings under seismic loading in shaking table tests. Sensors 23(11), 5303. DOI: 10.3390/s23115303 Cha, Y.J., Ali, R., Lewis, J., Büyük ӧ ztürk, O., 2024. Deep learning-based structural health monitoring. Automation in Construction 161, 105328. DOI: 10.1016/j.autcon.2024.105328. Cross, E.J., Gibson, S.J., Jones, M.R., Pitchforth, D.J., Zhang, S., Rogers, T.J., 2022. Physics-informed machine learning for structural health monitoring. In: Cury, A., Ribeiro, D., Ubertini, F. and Todd, M.D., (eds.) Structural Health Monitoring Based on Data Science Techniques. Structural Integrity (21). Springer Cham, 347-367. DOI: 10.1007/978-3-030-81716-9_17. De Angelis, A., Bilotta, A., Pecce, M.R., Pollastro, A., Prevete, R., 2024. Dynamic identification methods and artificial intelligence algorithms for damage detection of masonry in-fills. J Civil Struct Health Monit 14, 1383–1402. DOI: 10.1007/s13349-024-00790-0. De Canio, G., Andersen, P., Roselli, I., Mongelli, M., Esposito, E., 2011. Displacement Based approach for a robust operational modal analysis. Proc. of the Society for Experimental Mechanics Series 2011(6), 29th IMAC a Conference on Structural Dynamics, 31 January – 3 February, Jacksonville, Florida, USA, 187-195. DOI: 10.1007/978-1-4419-9507-0_19. De Canio, G., Mongelli, M., Roselli, I., 2013. 3D Motion capture application to seismic tests at ENEA Casaccia Research Center: 3DVision system and DySCo virtual lab. WIT Transactions on The Built Environment 134, 803-814. DOI: 10.2495/SAFE130711. References