ABSTRACT
Deterioration models form an important component of bridge management systems (BMS) by predicting future maintenance, repair, and rehabilitation (MR&R) needs at the bridge and network level. Consequently, the efficacy of a BMS in optimally allocating MR&R budgets to ensure the preservation of bridge components and the safety of the traveling public is directly affected by the accuracy of the bridge deterioration models. Since the introduction of BMS frameworks in the early 1980s, approaches for deterioration modeling have continuously developed in complexity from the earliest purely deterministic methods. Currently, the most widely prevalent in US are the Markov chain based probabilistic approaches, which have also been incorporated in the AASHTOWare Pontis and Bridgit commercial BMS software adopted by many states.
The growth of the historical condition rating database has recently permitted duration-based probabilistic approaches to be investigated that can account for data censoring and have been found to be more accurate than Markov chain models. Under one such approach, multivariable proportional hazards regression techniques were employed to study the effect of explanatory factors on deterioration rates at individual condition ratings, although these studies were limited to small subsets of bridges, and did not produce any deterioration models. Under another approach, parametric Weibull survival functions developed at each condition rating were used to obtain the expected durations associated with each rating from which deterioration models were developed. In this prior work, the univariate survival analysis proposed at each condition rating is not capable of accounting for the individual effects of explanatory factors on the deterioration rate. Moreover, Weibull based models cannot model uni-modal distributions frequently found in infrastructure deterioration. For this reason, semi-parametric Cox proportional hazards regression has been employed in the current study as it does not restrict the shape of the distribution and also supports multivariate analysis.
The development and implementation of an automated software framework for this regression analysis is presented that analyzes the time-dependent effects of explanatory factors on deterioration rates over the life cycle of the structural components. Selected results pertaining to only deck components obtained from application to North Carolina’s statewide bridge database consisting of over 17,000 bridges spanning 35 years of historical inspection general condition ratings (GCR) are presented.
The variable influence of external factors over the service life of individual bridge components is quantified in terms of covariate specific hazard ratios at each condition rating, which is a unique aspect of the developed framework and a potentially significant improvement to conventional deterioration modeling approaches that rely on a priori bridge classification that is fixed over all condition rating states in the lifecycle. The weighted mean covariate hazard ratios calculated over all material-specific deck components are consistent and exhibit clear trends that are well supported by prior literature.
