Stochastic-based pavement rehabilitation model at the network level with prediction uncertainty considerations

Abstract

An optimum network-level pavement rehabilitation model has been developed for generating a long-term rehabilitation schedule comprised of a specified number of annual rehabilitation cycles. The optimum model deploys the discrete-time Markov model to predict the performances of both original and rehabilitated pavements wherein the pavement improvement rates are incorporated into the transition probability matrix. The model implements continuous cyclic improvements in the long-term performance curve compared to the traditionally assumed vertical improvements. A Markov chain with (m) condition states can incorporate (m-1) rehabilitation treatments with an expected improvement outcome being the upgrade to condition state (1), the state with best pavement condition. The optimum model deploys an effective decision-making policy that maximises the long-term performance while minimising rehabilitation cost. The optimum model can be solved using exhaustive search with functional evaluations. The sample results obtained for a pavement network comprised of (12) highways indicated the efficiency of proposed model in yielding practical long-term rehabilitation schedules. The sample results also provided the minimal annual budget required to progressively remove the ‘very poor' pavements that greatly affect the life-cycle cost. Furthermore, investigation of prediction uncertainty resulted in a relatively mild impact when considering lower and upper-limit performance values using 95% confidence level.