Structural Health Monitoring Strategies Using Traditional Sensors and Computer Vision

Abstract

The vibration-based condition assessment of structures is the predominant method in structural health monitoring. The condition assessment of structures can be determined through the response of structures (i.e., peak displacement and acceleration), or through change characterization (i.e., system and damage identification). This dissertation presents three improved strategies for structural health monitoring using traditional sensors and computer vision. One strategy uses data fusion of acceleration and strain to estimate the displacement of building structures subjected to nonstationary wind load. In particular, this study presents two methods (data fusion A and B) that can accurately estimate both components of the displacement–the pseudo-static and the dynamic components. The two methods are validated numerically using a 20-story structure and experimentally using a small-scale 6-story structure. The second strategy is based on a computer vision method for system identification using consumer-level cameras and small structural motions. The Kanade-Lucas-Tomasi (KLT) and the Phase-Based Motion Processing (PBMP) methods are adopted in the proposed method. The method is validated experimentally using two small-scale steel structures: a 6-story building and a single-span truss bridge. The third strategy relies on the use of computer vision in damage identification by means of the Damage Locating Vector (DLV) method. This study also investigated the impact of using aliased modes in damage identification. The small-scale truss bridge was used for numerical and experimental evaluations of computer vision in system and damage identifications.