Updating Bridge Deck Condition Transition Probabilities as New Inspection Data Are Collected: Methodology and Empirical Evaluation

A Bayesian updating method is proposed to estimate a Markov chain based concrete deck deterioration model in a manner that combines condition data collected over two consecutive inspections and the deterioration information available prior to the collection of these data. A dataset of bridge deck condition assessments based on AASHTO condition state definitions collected by a state infrastructure agency spanning two years is used to evaluate the performance of this method. Training and validation datasets are selected from the original dataset where the former is used for estimation and the latter for prediction and evaluation. Single period transition probabilities are estimated using Bayesian updating, where prior deterioration information is combined with the condition data, and maximum likelihood estimation where only the collected condition data over two consecutive inspections are used. The evaluation is based on measuring the degree of similarity between reported condition states and those predicted based on the estimated transition probabilities using the two estimation methods. While updating transition probabilities as new data are collected is found to be advantageous for many cases, this advantage is highly dependent on the extent to which the training dataset is representative of the deterioration nature of the bridge decks for which condition is to be predicted. The less representative the training dataset, the more value is derived from Bayesian updating based predictions where prior deterioration information is considered.