Detection and localization of anomalies in bridges using accelerometer data and sparse auto-encoders

The paper focuses on the use of dynamic monitoring and sparse Auto-Encoder (SAE) networks for the detection and the localization of structural anomalies/damages. Unlike previous contributions in the literature, a single SAE is herein defined by simultaneously using the responses acquired at all instrumented points. Once trained using responses collected under healthy/normal condition, the SAE network is expected to accurately reconstruct new data as long as the structure remains in healthy state; however, if structural changes occur, the reconstruction error – measured as the difference between the actual and reconstructed signals – will increase, indicating a deviation from the normal condition. Moreover, the increase in the reconstruction error is conceivably more significant when the reconstructed signal refers to the neighborhood of damage, so that localization of critical regions is attained as well. The accuracy and reliability of the proposed methodology is exemplified using data collected on two real bridges. • The use of sparse Auto-Encoder (SAE) networks to detect and localize structural anomalies is investigated. • A procedure is presented to train a single network by using the dynamic responses simultaneously acquired by all channels. • The proposed methodology is validated through data collected on two real bridges. • Structural changes have been correctly detected and localized under both constant and changing environmental conditions.