Abnormal data detection for structural health monitoring: State-of-the-art review

Abstract

Structural health monitoring (SHM) is widely used to monitor and assess the condition and performance of engineering structures such as, buildings, bridges, dams, and tunnels. Owing to sensor defects, data acquisition errors, and environmental interference, abnormal data are often collected and stored in monitoring systems. The abnormal data in this study are essentially different from so-called “abnormal state data,” which result from structural physical damage or performance degradation. Abnormal data are totally related to the external interference rather than changes in the inherent structural features. However, abnormal data can significantly affect the performance assessment of engineering structures. It is imperative to detect and remove abnormal data from measurements to avoid misjudging structural performance in SHM. This paper summarizes abnormal data detection in the SHM field and discusses relevant challenges. Moreover, background knowledge regarding abnormal data detection is introduced. Abnormal data detection methods are then classified into statistical probability methods, predictive models, and computer vision methods. The advantages, disadvantages, and scope of each method are investigated. An example of detecting abnormal monitoring data for a cable-stayed bridge is introduced. In addition, the issues of existing studies are summarized, and future study interests are discussed.