Data-Assisted Prediction of Deterioration of Reinforced Concrete Bridges Using Physics-Based Models

Abstract

Bridge management systems are primarily employed to predict the future condition state of deteriorating bridges and plan for the maintenance and repair actions required for them. The deterioration models utilized in the current bridge management systems have been developed using Markov processes, the accuracy of which depends on the transition probabilities generated based on a combination of inspection data and expert elicitation. However, there are notable limitations in both sources of information, including the absence of long-term inspection recordings and the uncertainty in expert opinions. To address such limitations, the main motivation of this research investigation was to incorporate the physics-based models developed for the study of deterioration into the current bridge management systems. Thus, a set of physics-based models were used to generate the transition probabilities required for the Markov processes. The transition probabilities were then trained by the bridge inspection data obtained from eight counties of the state of Florida. The training was performed with (1) bridges in a specific county, (2) bridges in a specific county with a specific environmental condition, and (3) bridges in the entire eight counties with a specific environmental condition and age range. The results revealed that the physics-based models trained with a fraction of inspection data (from a subset of bridges) can predict the condition state of other similar bridges, for which no data was available, with adequate accuracy. The outcome of this study is expected to enhance the capabilities of bridge management systems without requiring a significant investment to overhaul the current bridge inspection process.