Modelling the Deterioration of Bridge Decks Based on Semi-Markov Decision Process

Abstract

Deterioration models represent a very important pillar for the effective use of bridge management systems (BMS's). This article presents a probabilistic time-based model that predicts the condition ratings of the concrete bridge decks along their service life. The deterioration process of the concrete bridge decks is modeled using a semi-Markov decision process. The sojourn time of each condition state is fitted to a certain probability distribution based on some goodness of fit tests. The parameters of the probability density functions are obtained using maximum likelihood estimation. The cumulative density functions are defined based on Latin hypercube sampling. Finally, a comparison is conducted between the Markov Chain, semi-Markov chain, Weibull and gamma distributions to select the most accurate prediction model. Results indicate that the semi-Markov model outperformed the other models in terms of three performance indicators are: root-mean square error (RMSE), mean absolute error (MAE), chi-squared statistic (x2).