Cable-stayed bridge condition evaluations by data analysis methodologies based on structural health monitoring system

Date
2023
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
For transportation agencies or private bridge owners, damage detection is a key interest as they manage their structural assets. In the past, most damage was detected using periodic visual inspection, which is time-consuming, expensive, and subjective. The advent of Structural Health Monitoring (SHM) systems which can capture the real-time response of structures can help bridge owners evaluate their bridges for potential life-safety issues and help them mitigate economic loss. ☐ The goal of this study is to develop methodologies that can be used to evaluate bridge condition efficiently by monitoring bearing displacements, strains, and cable tensions. These explicit parameters can be easily understood and used by bridge owners. In so doing, the Indian River Inlet Bride (IRIB), and data collected from its SHM system, were used to evaluate the methodologies developed. Different characteristics of the IRIB were analyzed using machine learning algorithms and statistical modeling. Direct and indirect criteria that transportation agencies can easily interpret were generated and calculated. The effectiveness and sensitivity of the proposed methodologies were validated by damage simulation. The evaluations incorporated the monitoring of bearing displacements and strain peaks, as well as cable tension tracking under critical wind load conditions. The approached developed can be used to improve the structure inspection and evaluation process, thereby strengthening the owners structural management process. ☐ The dissertation research resulted in four papers written by the author. These papers will be presented in the distinct chapters of this dissertation. In Chapter 4, bearing displacements were predicted using ANN. The same algorithm is applied to deck strains in Chapter 6. Both of these chapters focus on real-time monitoring. In Chapter 5, statistical modeling was applied to strain data to evaluate bridge condition in long-term perspective, The final paper is pressented in Chapter 7 and it is focused on cable tension and natural frequencies monitoring. Various conclusions were made based on the research. The first conclusion made by analyzing bearing displacements is that they can be accurately predicted using distributed thermal loads on the bridge. There is a nonlinear relationship between the thermal loads and bearing displacements. The second conclusion is that the abnormal responses of the bridge can be detected by comparing Strain Threshold Indexes (STIs) with Decision Boundaries (DBs). Statistical analysis was applied to truck-induced strain peaks, and corresponding STIs and DBs were calculated. The truck-induced strain peaks were also analyzed by Artificial Neural Network (ANN). The trained ANNs represented the relationship between strain gages at different locations on the bridge. The foundation idea is that the relationship among their strain peaks should be stable when a truck passes over a bridge and the bridge condition does not change. Otherwise, the response of the bridge under abnormal conditions does not match with the prediction of the strain ANNs. A mismatch will cause large Prediction Errors (PEs), and an abnormal responses warning will be triggered. The final conclusion is that the cable tension estimation process is significantly influenced by wind load conditions. By identifying the general critical wind load conditions using Decision Trees (DTs), cable tensions can be estimated using just two hours of data. The computational load and storage requirements can be heavily reduced by limiting the data to two hours. ☐ In summary, bearing displacements, truck-induced strain peaks, and cable tensions were monitored closely using a variety of proposed methodologies, and each methodology has shown promise in adding to the way in which SHM data can be used for assessing bridge conditions.
Description
Keywords
Deep learning, Machine learning, Non destructive tests, Statistic modeling, Structural Heath Monitoring
Citation