PSI - Issue 64

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

1373

5

PZT3

PZT4

PZT7

PZT8

Fig. 5. Measured vs reconstructed EMI spectrums for the undamaged beam.

Secondly, Fig. 6 compares the measured (original) and reconstructed EMI spectrums for two different damage stages, one with minor damage and the other with more severe damage. The same sensors used in Fig. 5 are used in this case. As shown in Fig. 6, the proposed deep autoencoder reconstructs the EMI spectrum with some inaccuracies, which is a symptom of the existence of damage and the capacity of the autoencoder to identify it, since the proposed autoencoder model was trained using only the dataset from the undamaged strengthened beam. That means it is not able to predict accurately an unknown dataset from the damaged beam. The loss function computed between the measured and reconstructed EMI spectrums might be used as damage-sensitive feature. As expected, it is clear that the loss function is higher for severe damage than for minor damage. In the same way, damage affects more to the internal sensors (7 and 8) than to the external sensors (3 and 4).

PZT3 – Minor damage

PZT4 – Minor damage