| Bridge behavior is used as an objective, data-driven indicator of the performance of bridges. A framework with which bridge behavior can be identified and learned is presented, and a method of long-term damage identification using the expected bridge behavior is introduced. At the Powder Mill Bridge (PMB) in Barre, Massachusetts, strains at each strain gage location are recorded during operational traffic events. Bridge behavior is defined as each sensor location's range of expected peak strain during a traffic event based on all other measured strains at the time at which it experiences its peak strain. Artificial neural networks (ANNs) are trained with operational data in a bootstrapping scheme to generate a probabilistic model of bridge behavior. When tested against new data, the ANN-learned model of bridge behavior is validated for a variety of traffic events with unknown loading conditions.;Structural damage is one way that bridge behavior, an indicator of performance, of a bridge can change. Damage scenarios are simulated in a finite element model (FEM) which is calibrated to PMB truck load test data. The effects of damage are extracted from FEM truck runs and applied to operational data to assess the capability of the proposed damage identification method through a series of trials. It is effective at detecting damage, with no Type I and no Type II errors when using a Wilcoxon rank-sum test of an appropriate significance level. Damage is effectively localized for two out of three damage scenarios. |
