Unsupervised Deep Learning for Structural Health Monitoring

In the last few decades, structural health monitoring has gained relevance in context of civil engineering, and much effort been made to automate process data acquisition analysis through use data-driven methods. Currently, main issues arising automated processing regard establishment a robust approach that covers all intermediate steps from output production interpretation. To overcome this limitation, we introduce dedicated artificial-intelligence-based for assessment conditions structures near-real time. The proposed is based on construction an unsupervised deep learning algorithm, with aim establishing reliable method anomaly detection acquired sensors positioned buildings. After preprocessing, are fed into various types artificial neural network autoencoders, which trained produce outputs as close possible inputs. We tested generated OpenSees numerical model railway bridge physical Historical Tower Ravenna (Italy). results show actually flags produced when damage scenarios activated coming damaged structure. also able reliably detect anomalous behaviors tower, preventing critical scenarios. Compared other state-of-the-art methods detection, shows very promising results.