AFT–Weibull Modeling of Bridge Deck Deterioration with Optimized Censoring

Bridge management systems traditionally rely on deterioration models that adopt deterministic assumptions and are often vulnerable to inspector subjectivity, encoding errors, and limited interpretability. These issues are further amplified by censoring in the National Bridge Inventory data, where inspection records are available only at discrete intervals. As a result, the entire duration of a bridge’s condition ratings is rarely observed, making it difficult to accurately estimate how long a bridge remains in each condition. This leads to inaccurate deterioration predictions and forces agencies into reactive maintenance strategies, thereby increasing life cycle costs and compromising reliability. To address these challenges, this paper presents a novel framework for deterioration modeling that explicitly accounts for high levels of data censoring. The study evaluates statistical models, including the Weibull and accelerated failure time (AFT)–Weibull models, and introduces an enhanced optimization algorithm to better accommodate censored observations. Results demonstrate that the proposed approach yields more accurate estimates of condition durations, which enables more reliable deterioration forecasting. This advancement supports proactive maintenance planning and more efficient allocation of infrastructure preservation resources. The proposed approach is applied to bridge inspection rating data from the New York State DOT, demonstrating its practical effectiveness.