An application of dynamic programming to pavement management systems

Dynamic programming is employed in conjunction with a Markov chain probability-based prediction model to obtain minimum cost maintenance strategies over a given life-cycle analysis period. The output is taken in conjunction with a specified budget and run through a prioritization package to produce a prioritized list of sections with their recommended maintenance alternatives and cost for each year in which a budget is specified. Pavement sections are split into families on the basis of characteristics such as surface types, distress modes and traffic levels. Prediction curves are fitted using Markov chain theory to obtain transition probability values that define future performance in terms of PCI states. Various maintenance alternatives are considered, and the initial cost of applying each alternative in each state for each family is input as are the probability values associated with the predicted performance of the maintenance alternatives in the future. Any life-cycle period can be specified, and any interest rate can be used. The dynamic programming algorithm then takes these inputs and, simultaneously for every state in every family, outputs the desired maintenance alternatives that will minimize the total expected cost over a specified life-cycle subject to keeping all sections above pre-defined performance standards. The algorithm is very simple, extremely fast and produces guaranteed global optimal solutions. A simulation routine is used to estimate the variance associated with the mean cost estimates given by dynamic programming. These programs were rigorously tested using four databases. Changes in output as the input parameters were changed were analyzed. The outputs from the dynamic programming probabilistic approach were compared with results obtained from a deterministic strategy analysis, and were very similar. A prioritization program was developed to apply budget constraints and derive the best set of sections to be repaired within these constraints for each year specified. This program uses the outputs from dynamic programming and weighting factors related to specific section characteristics in conjunction with a heuristic approach to obtain the desired list of prioritized sections.