Methodology for Probabilistic Modeling of Highway Bridge Infrastructure Condition: Accounting for Improvement Effectiveness and Incorporating Random Effects

Abstract

The peculiar nature of bridge infrastructure condition data persistently poses challenges in predicting bridge component deterioration that necessitate the continued investigation of probabilistic modeling techniques. These challenges include the uncertainty that characterizes bridge condition data due to the inherent random nature of deterioration factors and the existence of other variables that are not typically measured (unobserved factors responsible for deterioration), the panel nature of the data and its consequent observation-specific correlation and heterogeneity bias, and the lack of knowledge of the type and nature of past interventions. To these ends, this paper introduces a novel probabilistic modeling methodology intended to enhance the reliability of condition prediction by defining and quantifying the types of interventions and incorporating newly introduced explanatory variables to capture the effect of these intervention types on the future deterioration of bridge components. Unlike the current probabilistic techniques, the methodology proposed in this paper duly accounts for the maintenance and intervention history of in-service bridge components in the condition modeling process to reflect the real-world nondecay deterioration of the components. This paper demonstrates how the developed methodology can be implemented to predict the probability that a bridge component is in a given condition state at any given year. The marginal effects analysis provides bridge engineers with additional information about the individual strengths of the influential factors of bridge deterioration. The developed models can be useful for various agency functions, including the monitoring of bridge component performance and predicting their remaining service lives based on the bridge design and operational and environmental attributes.