Abstract Summary
Automated operational modal analysis (AOMA) of civil engineering structures and suspension bridges based on vibrational monitoring data is now regularly performed to extract the modal features of these structures, on which structural health monitoring is performed and the state of the infrastructure is inferred. AOMA algorithms have evolved since their serious beginnings around 2005 improving both in performance and in automation. However, in the authors’ opinion based on extensive benchmarking work of multiple AOMA algorithms, none has reached a level of automation and performance which can be considered sufficient and reliable enough for high quality SHM work. Many AOMA algorithms meant for bridge applications (Magalhaes 2009, Reynders 2012, Zhang 2014, Neu 2017, and Yang 2019) are based around two clustering steps using first k-means and then hierarchical clustering, using slightly different parameters and/or workflows to achieve their results. This work proposes an improved automatic OMA based around the framework proposed by the above-mentioned algorithms and their strong points. The k-means spurious pole clearing step is replaced by a gaussian mixture model clustering method, more adapted to the data structure at hand. Hierarchical clustering is maintained and fully automated with data-based cut-off distance calculation. Physical modes are then separated from their mathematical counterparts in an automated way. Comparison and testing of the algorithm is done using real-world full-scale bridge monitoring data.
