Assessment of Structures Using Dynamic Monitoring and Auto-Encoders: Application to Baixo Sabor Dam

An increasing attention has been recently payed to Artificial Intelligence (AI) techniques in Structural Health Monitoring (SHM) of Civil Engineering structures. When compared to traditional OMA-based techniques, AI exhibits the advantage of relatively low computational efforts, not requiring any system identification, as well as the implicit modeling of environmental and operational variability (EOV). Within the context of vibration-based SHM, the present paper proposes a procedure based on Auto-Encoder (AE) network for detection and localization of structural changes In more details, the data simultaneously collected by all available data channels are used to define an AE network in a training period during which, as usual, the monitored system is supposed to be in healthy condition under normal EOVs. Once trained, the network is used to reconstruct the newly collected data, with the mean reconstruction error (between the measured and the reconstructed signals) increasing as soon as the monitored system departs from healthy condition. The application of the presented procedure is exemplified to data measured in Baixo Sabor dam (Portugal) under varying temperature, water level of the reservoir and operating conditions of the turbine.