Improving the Estimation of Markov Transition Probabilities Using Mechanistic-Empirical Models

Abstract

In many current state-of-the-art bridge management systems, Markov models are used for both the prediction of deterioration and the determination of optimal intervention strategies. Although transition probabilities of Markov models are generally estimated using inspection data, it is not uncommon that there are situations where there are inadequate data available to estimate the transition probabilities. In this article, a methodology is proposed to estimate the transition probabilities from mechanistic-empirical models for reinforced concrete elements. The proposed methodology includes the estimation of the transition probabilities analytically when possible and when not through the use of Bayesian statistics, which requires the formulation of a likelihood function and the use of Markov Chain Monte Carlo simulations. In an example, the difference between the average condition predicted over a 100-year time period with a Markov model developed using the proposed methodology and the condition predicted using mechanistic-empirical models were found to be 54% of that when the state-of-the-art methodology, i.e., a methodology that estimates the transition probabilities using best fit curves based on yearly condition distributions, was used. The variation in accuracy of the Markov model as a function of the number of deterioration paths generated using the mechanistic-empirical models is also shown.