Multivariable Proportional Hazards-Based Probabilistic Model for Bridge Deterioration Forecasting

Abstract

Duration-based deterioration modeling approaches have been investigated over the past two decades with a view toward utilizing their prognostic capabilities in accurately identifying maintenance needs and improving asset management under constrained budgets. However, despite significant advances in asset management systems during this time, implementation of such approaches for probabilistic infrastructure deterioration modeling has been limited, resulting in underutilization of their predictive potential. In this paper, a comprehensive framework based on a combination of the Cox proportional hazards method and Markovian theory is presented to develop multivariable deterioration models that probabilistically incorporate the effects of explanatory factors on deterioration over the complete life cycle of a bridge component. Both stationary (time-independent) and nonstationary (time-dependent) transition probability approaches are introduced and compared. Sample results from implementation of this framework on North Carolina’s statewide bridge inspection database containing 35 years of data for more than 17,000 bridges are discussed. The predictive fidelity of the developed models is analyzed relative to the actually recorded condition ratings to demonstrate the effectiveness of these models in accurately forecasting deterioration.