Binary Segmentation for Structural Condition Classification Using Structural Health Monitoring Data

Abstract

Structural health monitoring (SHM) is the process of conducting structural condition diagnosis and prognosis based on appropriate analyses of in situ measurement data. Direct assessment of structural condition using time series response measurements can be classified as a type of statistical pattern recognition, in which structural condition is evaluated by comparing the statistical features of current data with those of baseline data. The philosophy behind this approach is that the time series response acquired under different structural conditions presents different statistical characteristics. As a consequence, the key step in structural condition classification is to detect the points at which the statistical properties of a time series response change; this is referred to as change-point analysis. The present study proposes the use of a computationally efficient binary segmentation (BS) approach for change-point detection in order to classify and assess structural health condition. The proposed approach, which falls into the category of data-driven diagnosis, does not require knowledge about the structure and is appealing for attaining an automated SHM system. The practicality and effectiveness are illustrated through real-world monitoring data acquired from a cable-stayed bridge and a high-speed train, both of which experienced structural damage/degradation over their service lives.