Anomaly detection in bridge structural health monitoring via 1D-LBP and statistical feature fusion

Over the past few years, numerous bridges have been equipped with structural health monitoring (SHM) systems to continuously monitor critical structural parameters, enabling early detection of potential issues and timely maintenance. However, the monitoring data frequently contain anomalies due to various interferences during the acquisition and transmission processes. To address this, this paper proposes a robust and efficient anomaly detection and classification method. The method extracts one-dimensional local binary pattern (1D-LBP) features and time-domain statistical features from the monitoring data, fusing them into a comprehensive feature representation. These fused features are then input into an extreme gradient boosting (XG-Boost) classifier for anomaly detection. Additionally, a 1D-LBP feature simplification method is introduced to enhance detection efficiency. The effectiveness of the proposed method was validated using monitoring data from a long-span cablestayed bridge SHM system. Experimental results demonstrate the excellent performance of the method in terms of detection accuracy and efficiency.