Validation and Comparison of Two AOMA Approaches for the Ambient Vibration Testing of Long Suspension Bridges Under Strong Wind Loads

Several Automated Operational Modal Analysis (AOMA) approaches have been developed in the last few years. Among them, two recent data-driven, Machine Learning (ML) techniques have been selected for comparison. The first one uses a combination of two clustering algorithms (k-means with k equal to 2 and density-based clustering). The second one uses kernel density estimation (KDE) for the interpretation of the stabilization diagram and therefore the selection of stable pole alignments. Both algorithms have been previously validated on applications for Civil Engineering purposes. In this work, the two approaches are validated and directly compared for the first time for the Structural Health Monitoring of long suspension bridges. A compelling experimental case study is considered: the Hardanger Bridge dataset. This infrastructure crosses the Hardangerfjord in Norway, with a total length of 1380 m (1310 m in the central span alone, making it one of the longest suspension bridge spans in the world). The experimental data investigated here represent an interesting stress test for the capabilities of the two methods, since the bridge has several closely-spaced, very low frequency (less than 1 Hz) modes, due to its slenderness and flexibility. Not only that, the bridge is inserted in a harsh environment, subject to Atlantic storms and a complicated wind field around due to the local topography with steep mountains and wide fjords. Hence, this work also briefly assesses the potential applications of the proposed code for handling changing ambient conditions.