Analysis of heterogeneity in infrastructure condition assessment models

Condition states of civil infrastructure such as pavements and bridges are usually indexed on discrete scales, and Markov chain models are often used to estimate transition rates from one state to another, as well as sojourn times in any given state. Such models tend to assume homogenous transition rates between states, even though the models should account for heterogeneity to improve the accuracy of the predicted outcomes. Therefore, the objective of this study is to evaluate variations in model predictions when transition rates between condition states vary. Building on existing research, this paper considers a Monte Carlo simulation of a non-homogeneous Markov chain model for condition prediction of infrastructure systems. The method considers two scenarios (i) when transition rates are regressed according to a Weibull probability density function, and (ii) when transition rates are sampled as random values from interval estimates. For each scenario, multiple cases are further investigated to quantitatively describe variations in model predictions generated from variable transition rates. For the first scenario, simulations demonstrated that the accuracy of the model predictions is sensitive to the shape parameter for the Weibull distribution. Similarly, for the second scenario simulations illustrated that the interval estimates when chosen correctly can be used to establish confidence intervals for the transition rate parameter. Methods developed in this study can be used to test the accuracy of performance predictions for infrastructure systems, thus improving a decision-maker\u27s confidence in using the models to optimize infrastructure maintenance schedules. © 2012 ASCE