Bayesian Survival Analysis for US Concrete Highway Bridge Decks

Abstract

A leading factor in structural decline of highway bridges is the deterioration of concrete decks. Thus, a method to predict bridge deck performance is vital for transportation agencies to allocate future repair and rehabilitation funds. While service-life prediction tools are available, they rely on input parameters that are often difficult to obtain or estimate. This study estimates the relationships between concrete highway bridge deck performance and information readily available from the National Bridge Inventory (NBI), perhaps the single most comprehensive nationwide source of bridge information. As such, this paper takes full advantage of the NBI data using a scale of analysis exceeding that of previous studies. Using recent computational advances in Bayesian survival analysis, this paper models the factors affecting time-in-condition ratings (TICR)—defined as the time duration a bridge deck is assigned the same condition rating (CR) before it decreases—using over 150,000 bridge decks observed over 23 years. Because the dataset only spans 23 years of elapsed time and bridge deck deterioration takes place over years and sometimes decades, many observations of bridge deck CR only provide a censored view of how long a bridge deck may have been assigned a certain CR. Reasons for censorship include the following: (1) data is censored as its CR prior to 1992 is unknown; (2) data is censored as its rating after 2014 is unknown; (3) data is censored due to missing observations; and (4) data is censored due to an increase in CR from 1 year to the next, which is considered maintenance. Fortunately, the Bayesian approach provides a coherent method for handling censored observations while simultaneously providing meaningful estimates of parameter uncertainty. The results provide insight into the parameters driving concrete bridge deck deterioration and may help agencies with maintenance repair prioritization.