Long-term structural health monitoring of long-span suspension bridges and anomaly detection using statistical indicators

Abstract

Some research has been conducted on structural health monitoring (SHM) utilizing raw data to discover structural behavior changes, anomalies, or damage assessment. However, few of them employed statistical indicators represented as input data. Long-term monitoring has several advantages (before/after event comparison, detecting/tracking abrupt changes or aging effects), while the data collection process may have disadvantages (data storing issues, impractical computationally expensive analyses). Thus, the current work proposes two unsupervised machine learning techniques for detecting anomalies using statistical indicators extracted from long-term raw dynamic measurements recorded from a three-dimensional accelerometer installed on the midspan of the Osman Gazi Bridge built in Turkey. For this purpose, one-class Support Vector Machine (SVM) and Local Outlier Factor (LOF) are used as machine learning algorithms to detect anomalous instances. Comparison between these two machine learning revealed similar anomalies. The acquired results support the development of computational methods for assessing structural anomalies based on statistical indicators of acceleration measurements. © 2024 Elsevier B.V., All rights reserved.