Dynamic monitoring and sparse autoencoder networks to assess the health condition of a historic iron bridge

The use of dynamic monitoring methods integrated with sparse Auto-Encoder (SAE) networks is exemplified in the paper with the aim of detecting and localizing structural anomalies. The methodology herein adopted is based on the definition of a single SAE network, trained with the signals simultaneously collected from several sensors. After a concise background on SAE networks and the description of the proposed strategy of data analysis, the effectiveness and accuracy of the SAE methodology of damage assessment is demonstrated using accelerations data collected in the continuous monitoring of the historic San Michele bridge. The investigated infrastructure is an iron arch bridge that links the small towns of Paderno and Calusco d’Adda, about 50 km from Milan. The bridge, built in 1889 to complete one of the first Italian railway lines, is still used as a combined road and railway bridge. During the dynamic monitoring performed in 2012-2013, the acceleration data associated with ambient excitation and road traffic were used for automated OMA and allowed to detect structural performance anomalies and changes in the dynamic characteristics of the infrastructure, conceivably related to the progress of the damage due to corrosion. In this paper, short duration time series induced by the road traffic are used to train the SAE network, to detect and localize the structural changes and comparison with the OMA results is presented as well.