Files
Abstract
Evaluation of live load effects has a profound influence on the design, maintenance, and rehabilitation of bridges in the United States. This dissertation investigates three main topics related to live loads. First, statistical methods are used to evaluate high-impact and low-probability bridge overloading events. This study proposes a cognitive approach to evaluating live load factors because stakeholders discount the probability of observing overweight vehicles based on the prospect theory. It quantifies the likelihood of observing extreme weights on bridges from various Weigh-In-Motion (WIM) sites in Georgia using the Extreme Value Theory and examines the dependency of live load factors on the choice of a threshold or an extreme percentile. Subsequently, the process of predicting maximum live load factors is validated using another state’s data. It is concluded that a live load factor is affected by a shape parameter and numerically quantifiable for each site, and near-term live load factors are more salient for preparing bridges for high-risk low-probability overloading events. In a subsequent study, National Bridge Inventory (NBI) data is used to gain better understanding of the correlation between bridge capacity and bridge load ratings by means of a supervised machine learning model. Lastly, a data-driven reliability analysis is conducted by employing WIM and NBI data, and their reliability is quantified using real-life traffic data, in order to mitigate an adverse selection problem when maintaining and replacing bridges. It is concluded that the capacity of bridges is correlated with bridge load ratings, and a WIM data-driven reliability analysis can assist in minimizing information asymmetry and adverse selection problems in bridge safety. The last study analyzes anomalies in vehicle weight data using the Jensen-Shannon divergence method. Additionally, a comparative time series analysis utilizing machine learning and deep learning algorithms is performed to search for a pattern for a time-history of Class 9 vehicle weights, establish a control dataset, and ultimately overcome dynamic inconsistency that may exist when making a data-driven decision for transportation assets.