Book Description

Bridges are critical components of highway networks, which provide mobility and economical vitality to a nation. Ensuring the safety and regular operation as well as accurate structural assessment of bridges is essential. Structural Identification (St-Id) can be utilized for better assessment of structures by integrating experimental and analytical technologies in support of decision-making. St-Id is defined as creating parametric or nonparametric models to characterize structural behavior based on structural health monitoring (SHM) data. In a recent study by the ASCE St-Id Committee, St-Id framework is given in six steps, including modeling, experimentation and ultimately decision making for estimating the performance and vulnerability of structural systems reliably through the improved simulations using monitoring data. In some St-Id applications, there can be challenges and considerations related to this six-step framework. For instance not all of the steps can be employed; thereby a subset of the six steps can be adapted for some cases based on the various limitations. In addition, each step has its own characteristics, challenges, and uncertainties due to the considerations such as time varying nature of civil structures, modeling and measurements. It is often discussed that even a calibrated model has limitations in fully representing an existing structure; therefore, a family of models may be well suited to represent the structure's response and performance in a probabilistic manner. The principle objective of this dissertation is to investigate nonparametric and parametric St-Id approaches by considering uncertainties coming from different sources to better assess the structural condition for decision making. In the first part of the dissertation, a nonparametric St-Id approach is employed without the use of an analytical model. The new methodology, which is successfully demonstrated on both lab and real-life structures, can identify and locate the damage by tracking correlation coefficients between strain time histories and can locate the damage from the generated correlation matrices of different strain time histories. This methodology is found to be load independent, computationally efficient, easy to use, especially for handling large amounts of monitoring data, and capable of identifying the effectiveness of the maintenance. In the second part, a parametric St-Id approach is introduced by developing a family of models using Monte Carlo simulations and finite element analyses to explore the uncertainty effects on performance predictions in terms of load rating and structural reliability. The family of models is developed from a parent model, which is calibrated using monitoring data. In this dissertation, the calibration is carried out using artificial neural networks (ANNs) and the approach and results are demonstrated on a laboratory structure and a real-life movable bridge, where predictive analyses are carried out for performance decrease due to deterioration, damage, and traffic increase over time. In addition, a long-span bridge is investigated using the same approach when the bridge is retrofitted. The family of models for these structures is employed to determine the component and system reliability, as well as the load rating, with a distribution that incorporates various uncertainties that were defined and characterized. It is observed that the uncertainties play a considerable role even when compared to calibrated model-based predictions for reliability and load rating, especially when the structure is complex, deteriorated and aged, and subjected to variable environmental and operational conditions. It is recommended that a family-of-models approach is suitable for structures that have less redundancy, high operational importance, are deteriorated, and are performing under close capacity and demand levels.