A Spectral-Based Clustering for Structural Health Monitoring of the Sydney Harbour Bridge

This paper presents the results of a large scale Structural Health Monitoring application on the Sydney Harbour Bridge in Australia. This bridge has many structural components, and our work focuses on a subset of 800 jack arches under the traffic lane 7. Our goal is to identify which of these jack arches (if any) respond differently to the traffic input, due to potential structural damages or instrumentation issues. We propose a novel non-model-based method to achieve this objective, using a spectrum-driven feature based on the Spectral Moments (SMs) from measured responses from the jack arches. SMs contain information from the entire frequency range, thus subtle differences between the normal signals and distorted ones could be identified. Our method then applies a modified k-means−− clustering algorithm to these features, followed by a selection mechanism on the clustering results to identify jack arches with abnormal responses. We performed an extensive evaluation of the proposed method using real data from the bridge. This evaluation included a control component, where the approach successfully detected jack arches with already known damage or issues. It also included a test component, which applied the method to a large set of nodes over a month of data to detect any potential anomaly. The detected anomalies turned out to have indeed system issues after further investigations.