Automated operational modal analysis of an end-supported pontoon bridge using covariance-driven stochastic subspace identification and a density-based hierarchical clustering algorithm
Knut Andreas Kvåle, Ole Øiseth
Abstract
The dynamic response and external excitation sources of an end-supported pontoon bridge have been monitored and recorded over a seven-year-long period, to enable quantification of the uncertainties of response prediction and modelling, and to learn about its dynamic behaviour. By automatically performing modal analysis to the bulk of the recordings, the correlation between environmental parameters and resulting modal parameters can be investigated, which is useful for understanding the bridge’s behaviour. Herein, a procedure to automatically establish modal estimates is suggested. The poles resulting from a parametric method for modal analysis are analyzed by applying a density-based hierarchical clustering algorithm often applied as an exploratory step to machine learning, abbreviated to and known as HDBSCAN. The suggested method is independent of the operational modal analysis method applied and will be relevant for all parametric methods where stabilization charts are traditionally needed, even though the covariance-driven stochastic subspace identification algorithm is used herein.