PSI - Issue 62

Israel Alejandro Hernández-González et al. / Procedia Structural Integrity 62 (2024) 879–886 Hernández-González et al./ Structural Integrity Procedia 00 (2019) 000 – 000 5 corresponding terms are time-shifted to align them with the time sequence of the network's output. The loss function is minimized using the Adam adaptive gradient-descent backpropagation algorithm (Chen et al., 2016). 3. Case of Study: Méndez-Núñez Bridge 883

Fig. 2. (a) Views of the Méndez-Núñez Bridge. (b) Instrumentation system. (c) Plan, elevation, and sections views of the Granada Bridge with the sensors layout (dimensions in m). This case study investigates the effectiveness of the proposed MTL-DNN for modal identification applied to a real world in-operation bridge: the Méndez-Núñez Bridge. The bridge is located in the Spanish city of Granada in the Autonomous Community of Andalucía, between the municipalities of Jaén and Motril. It is a continuous five-spans 122.5 m long post-tensioned concrete bridge over the Avenida de Andalucía. The bridge was built in March 1989 by the Dirección General de Carreteras of the Province of Granada. This bridge consists of a continuous deck supported by 6 piers, the inner 4 corresponding to concrete columns with a section of 3.40 m x 1.50 m and founded on piles and pile caps. The end supports correspond to abutments consisting of reinforced concrete retaining walls. All the supports of the bridge are defined by means of elastomeric neoprene bearing pads of dimensions 0.90 m x 0.80 m x 0.15 m. For visual reference, Fig. 2 (a) shows two photographs of the current state of the structure. This bridge has a variable section along its entire length (Fig. 2 (c)), with three base typologies, at the abutments (Fig. 2 (i)), inner supports (Fig. 2 (ii)), and centre spans (Fig. 2 (iii)). Within the realm of a national R&D project, a permanent vibration-based SHM system was installed on September 27 th , 2023. The monitoring system comprises 10 uni-axial piezoelectric accelerometers model KB12VD (µ10% 10.0 V/g, broadband Resolution: 1 µg rms and ± 0.5 g pk) labelled with A1 to A10 as shown the Fig. 2 (c). Ambient vibrations are stored in separate data files containing 30-min-long records and sampling frequency of 200 Hz. The acceleration signals are recoded by a data acquisition system (DAQ) model cDAQ-9184 located on one of the support columns as shown the Fig. 2 (b). Temperature data are also recorded by four probes model Pt 1000/3850 placed on the bridge deck, and the humidity data by means of a hygrometer model AM2315 controlled using an independent Arduino Uno micro-controller as shown in Fig. 2 (b). Environmental data are collected with an acquisition frequency of 5 min. In the design of the network, the number of independent components to be determined is set equal to 14. To this aim, two different frequency broadbands were necessary, namely 0-11 Hz and 11-14 Hz. Figure 3 (a) furnishes the architecture of the designed MTL-DNN. The learning rate, the batch size, and the dropout rate are set to 0.001, 200 and 30%, respectively. To generate the training dataset, the independent modal components are extracted from the raw acceleration time series by standard SOBI applied to the two considered frequency broadbands independently. The raw acceleration responses and the estimated independent components serve as inputs and outputs to build the dataset, considering an input/output time lag equal to 10 time steps ( =10 ms). The dataset is divided into training