Date of Award
Program or Major
Master of Science
Erin Santini Bell
The purpose of this research is to establish a protocol for using strain gauge data to characterize the undamaged state of a bridge for structural health monitoring. The Powder Mill Bridge (PMB), which has been instrumented with strain gauges since its opening in 2009, is used as a case study. and the strain gauges used in this study are located at 27 different locations throughout the bridge. Artificial neural networks (ANNs) and linear regression are presented as methodologies for characterizing the relationship between the strains at each of the stations on the bridge. One linear regression analysis was performed as well as 60 different ANN trials, each with unique parameters regarding the architecture and training algorithm of the ANNs.
The linear regression model and all 60 ANN trials were able to predict the strain at each of the 27 stations with an average error of less than 5%. A calibrated finite element model was used to simulate damage in the PMB for three damage scenarios: fascia girder corrosion, girder fracture, and deck delamination. The models trained using the linear regression and the ANN methods were able to detect damage in all scenarios with damage being localized in many cases.
Based on the results from the linear regression and ANN analyses, a recommended protocol for integrating a data-driven model into a bridge structural health monitoring system is outlined with the goal of providing a bridge owner with real-time, objective information regarding the state of the bridge.
Kaspar, Kathryn Elaine, "A Protocol for Using Long-Term Structural Health Monitoring Data to Detect and Localize Damage in Bridges" (2018). Master's Theses and Capstones. 1257.