Automated and model-free bridge damage indicators with simultaneous multi-parameter modal anomaly detection

Tran, Thanh T. X. and Ozer, Ekin (2020) Automated and model-free bridge damage indicators with simultaneous multi-parameter modal anomaly detection. Sensors, 20 (17). 4752. ISSN 1424-8220 (

[thumbnail of Tran-Ozer-Sensors-2020-Automated-and-model-free-bridge-damage-indicators-with-simultaneous]
Text. Filename: Tran_Ozer_Sensors_2020_Automated_and_model_free_bridge_damage_indicators_with_simultaneous.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo

Download (15MB)| Preview


This paper pursues a simultaneous modal parameter anomaly detection paradigm to structural damage identification inferred from vibration-based structural health monitoring (SHM) sensors, e.g., accelerometers. System Realization Using Information Matrix (SRIM) method is performed in short duration sweeping time windows for identification of state matrices, and then, modal parameters with enhanced automation. Stable modal poles collected from stability diagrams are clustered and fed into the Gaussian distribution-based anomaly detection platform. Different anomaly thresholds are examined both on frequency and damping ratio terms taking two testbed bridge structures as application means, and simplistic Boolean Operators are performed to merge univariate anomalies. The first bridge is a reinforced concrete bridge subjected to incremental damage through a series of seismic shake table experiments conducted at the University of Nevada, Reno. The second bridge is a steel arch structure at Columbia University Morningside Campus, which reflects no damage throughout the measurements, unlike the first one. Two large-scale implementations indicate the realistic performance of automated modal analysis and anomaly recognition with minimal human intervention in terms of parameter extraction and learning supervision. Anomaly detection performance, presented in this paper, shows variation according to the designated thresholds, and hence, the information retrieval metrics being considered. The methodology is well-fitted to SHM problems which require sole data-driven, scalable, and fully autonomous perspectives.