PSI - Issue 64

Ricardo Perera et al. / Procedia Structural Integrity 64 (2024) 1369–1375 Author name / Structural Integrity Procedia 00 (2019) 000–000

1374 6

PZT7 – Minor damage

PZT8 – Minor damage

PZT3 – Severe damage

PZT4 – Severe damage

PZT7 – Severe damage

PZT8 – Severe damage

Fig. 6. Measured vs reconstructed EMI spectrums for the damaged beam.

Any discrepancy over a threshold between the original input spectrum to the autoencoder and the output spectrum is a symptom of damage since the autoencoder is only trained for undamaged signals. As the spectra are captured from PZT sensors and variations of these spectra are identified with stiffness variations of the analyzed structure, these changes can be identified with structural damage. Furthermore, as PZT sensors operate in the high frequency range, they are able to identify minor damage such as the one originated previous the the debonding of the FRP strengthening. 4. Conclusions A deep learning framework based on convolutional autoencoders has been proposed to identify the presence of anomalies in FRP strengthened RC elements from EMI measurements. This framework is particularly useful for structural health monitoring in near-real-time. The proposed methodology has been validated using EMI data of experimental tests carried out on a NSM-FRP strengthened beam subjected to different levels of damage. Only the training dataset from the undamaged beam was used for training the autoencoder, while the unknown testing dataset from undamaged and damaged stages was used for testing the trained autoencoder. From the analysis of the results, it can be concluded that the method detects successfully anomalies due to minor damage and severe damage for this kind of strengthened structures. The loss