A DBSCAN-based automated operational modal analysis algorithm for bridge monitoring. (28th February 2023)

• Record Type: Journal Article
• Authors: Civera, Marco; Sibille, Luigi; Zanotti Fragonara, Luca; Ceravolo, Rosario
• Is Part Of: Measurement. Volume 208(2023)
• Publication Date: 2023-02-28
• Subjects: Automated operational modal analysis -- Machine learning -- System identification -- Signal processing -- Bridge monitoring -- DBSCAN; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2023.112451
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: An algorithm for the Automated Operational Modal Analysis (AOMA) is presented. The algorithm applies density-based clustering and is proposed for bridge monitoring. A novel, data-driven procedure is proposed for the automatic setting of its parameters. The AOMA method is validated on experimental acquisitions from the well-known Z24 case study. The proposal is benchmarked against the conventional (hierarchical clustering-based) approach. Its computational efficiency, capabilities for damage assessment, and robustness to changing environmental conditions are all assessed. Abstract: Advanced data analysis techniques are of paramount importance for the Structural Health Monitoring (SHM) of civil buildings and infrastructures. In particular, Automated Operational Modal Analysis (AOMA) algorithms are necessary for the output-only monitoring of such massive and large structures. The unsupervised estimation of their modal parameters from ambient vibrations enables assessing their integrity efficiently and continuously. This is particularly important for reinforced concrete (RC) bridges, which need constant maintenance. In this context, the classic cluster-based, multi-stage approach is effective in cleaning the stabilisation diagram and discerning stable and unstable modes. However, due to the shortcomings of binary classification with (k = 2)-means clustering, the labelling between 'possibly physical' and 'certainly spurious' modes may not be completely reliable. The procedure described here applies Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to bypass this limitation. This allows, among other advantages, to automatically detect and remove outliers, differently from the traditional techniques. The algorithm is fully automated, including the data-driven setting of DBSCAN parameters. Its viability is tested here on a real, full-scale case study, the Z24 road bridge dataset.