Probabilistic service life prediction model for concrete bridge decks
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Bridge deck deterioration is the leading cause of repair, rehabilitation or
replacement of bridge superstructures (Li et al 2001). According to the National
Bridge Inventory (NBI) in 2011, there are currently 605,102 bridges in the United
States of which 361,986 have cast-in-place concrete decks and 53,286 have prestressed
concrete decks. It is of great interest, therefore, to be able to reliably predict
the durability of concrete bridge decks based on initial and time-dependent input
parameters such as design, construction, environmental, site, and service parameters;
and thus forecast and prevent failure.
In this study, a probabilistic model framework using Markov Models was
developed that can be used to predict the service-life of concrete bridge decks based
on the deck condition ratings recorded in the NBI database. This study also explores
the correlations between several parameters and their effect in the bridge deterioration
process. For instance, design loads, average daily traffic (ADT), environmental
factors as established by location were all considered in the parameter analysis.
NBI data for fourteen different states was analyzed and used for the parameter
analysis and the prediction model. It was concluded that, perhaps because of the
qualitative nature and possible subjectivity of the deck condition ratings, no clear
correlations were found between this rating and the selected parameters.
Additionally, no significant environmental or geographical trends were observed.
Therefore, the prediction model was developed by taking the condition ratings per
state and creating a Markov chain based solely on this parameter. This model is
capable of predicting the probability of future deck ratings given the current condition of the deck. The Markov chain model is also capable of predicting steady state
probabilities of the deck ratings for each of the fourteen states analyzed. These
fourteen models provide with some insight on how deck deterioration compares from
one state to another. Moreover, these models and parameter analysis provide the
basis from which a more complex prediction model based on observed data and NBI
condition ratings can be developed as more information becomes available.
Additionally, other more refined probabilistic methods, Bayesian Networks
(BN), were explored and are suggested for future work.