Experimental comparison of automatic operational modal analysis algorithms for application to long-span road bridges

Abstract

Operational modal analysis (OMA) of long-span road bridges from vibration measurements is a topic of interest due to its potential applications within structural health monitoring. Algorithms for the automation of OMA (AOMA) have been proposed since around 2005, relying on machine learning to automate previously manual tasks. The fundamental principles of AOMA are explained in the theory of this work, as are the functioning principles of the clustering algorithms employed by most AOMA algorithms. A performance comparison of six AOMA algorithms (Magalhaes 2009, Reynders 2012, Zhang 2014, Neu 2017, Yang 2019, Kvåle 2020) is provided using real-world data from the Hardanger suspension bridge. To the authors’ best knowledge, it is the first comparison of AOMA algorithms for bridges. While Reynders 2012 is shown to be the only fully automated algorithm, Magalhaes 2009, Zhang 2014, and Kvåle 2020 are the algorithms with the highest successful detection rate. Neu 2017 and Yang 2019, however, make the least detection errors, respectively, the least false detections and the least duplicate detections. The variability amongst datasets is shown not to impact the algorithms’ comparison. An outright recommendation on which algorithm to use is impossible due to the multitude of potential use cases, but Neu 2017 seems to provide the best performance compromise for the tested case.