Survival analysis and semi-Markov bridge deterioration modeling

All man-made structures begin to deteriorate almost immediately from the moment they are constructed. To manage any in-service facility effectively thus requires a thorough knowledge of its "health" condition or rather, its performance level over time. Providing a suitable measure for the performance and modeling its deterioration in time are two important issues to consider in any management system development. Existing bridge management systems (BMS) invariably use condition rating as a measure of bridge performance. Regression analysis and Markov chain theory are commonly used to model the bridge deterioration process.;Condition rating consists of a discrete, ordinal scale; with each numerical value representing a bridge condition. It is arbitrary in its definition and subjective in its evaluation. Despite its popularity, condition rating by itself is inadequate as a measure of bridge performance. On the other hand, regression and Markov chain models both have restrictive assumptions implicit in their respective formulations. Recent deterioration studies fall under either of these categories: (1) the search for an improved rating system for bridge condition; and (2) the investigation for alternative procedures to derive the transition matrices of the Markov chain. Regression analysis remains an indispensable statistical tool in many of these studies.;The study herein takes an altogether new direction from previous studies. First, it investigates the use of bridge reliability and load rating as bridge performance measures. Second, it explores use of the semi-Markov process in bridge deterioration modeling. A procedure was developed for converting the National Bridge Inventory (NBI) data into failure time data. Failure, was defined either in terms of condition rating or load rating. Survival analyses were applied to these data to obtain the reliability functions for bridges. Further, a procedure was developed to model the degradation of bridge condition or load-carrying capacity as a semi-Markov process. Applications of these degradation models in network and project level bridge management decisions were demonstrated. The results obtained from survival analysis were shown to be consistent with previous deterioration studies. The semi-Markov theory, being more general than the Markov theory, affords more flexibility in modeling bridge deterioration processes. Many of the procedures described in this dissertation are also applicable to other types of infrastructure such as road pavements, parking structures, pipelines and offshore structures.