PSI - Issue 64

1371 3

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

Fig. 1. Convolutional autoencoder.

A convolutional autoencoder is a type of autoencoder that uses convolutional layers in the encoder and decoder. Fig. 2 shows the design architecture of an autoencoder for extracting bridge damage-sensitive features from the impedance spectrum. Convolutional autoencoders consist of two major components: the encoder and the decoder. During the encoding process, a compressed representation of the impedance spectrum is learnt, while the decoder takes this compressed representation and reconstructs those inputs at the output layer. The compression is achieved by applying convolutional layers to extract features from the input and downsampling them to reduce the dimensionality. The decoding process is just the reverse of this, where upsampling layers are used to increase the dimensionality and convolutions are applied to reconstruct the image. The final output of the decoder is the reconstructed spectrum, which is supposed to be as close as possible to the original input spectrum. The autoencoder was built with Pytorch, and the Adam optimization algorithm was used to update network weights. 3. Experimental set-up In this work, the experimental tests were performed on a reinforced concrete beam strengthened with a CFRP strip following the NSM method (Fig. 2). The material properties of the concrete, the reinforcement steel and the CFRP were the following: a) Concrete: fc = 30 MPa, Ec = 26 GPa, fct = 3 MPa; b) Steel: fy =500 MPa, Es=210 GPa; c) CFRP: ffu = 2500 MPa, Ef = 170 GPa. Four PZT sensors were directly bonded on the internal CFRP strip while other four PZT sensors were attached to the external tensile face of concrete (Fig. 3). Fig.4 shows a photo of the experimental set-up as well as a detail of the sensors bonded to the concrete face.

Fig. 2. Experimental beam (Units in mm).