Stochastic performance and maintenance optimization models for pavement infrastructure management

Highway infrastructure, includingroads/pavements, contributes significantly to a country’s economic growth,quality of life improvement, and negative environmental impacts. Hence, highwayagencies strive to make efficient and effective use of their limited funding tomaintain their pavement infrastructure in good structural and functionalconditions. This necessitates predicting pavement performance and schedulingmaintenance interventions accurately and reliably by using appropriateperformance modeling and maintenance optimization methodologies, whileconsidering the impact of influential variables and the uncertainty inherent inpavement condition data. Despite the enormous research effortstoward stochastic pavement performance modeling and maintenance optimization,several research gaps still exist. Prior research has not provided a synthesisof Markovian models and their associated methodologies that could assistresearchers and highway agencies in selecting the Markov methodology that isappropriate for use with the data available to the agency. In addition, pastMarkovian pavement performance models did not adequately account for themarginal effects of the preventive maintenance (PM) treatments due to the lackof historical PM data, resulting in potentially unreliable models. The primarycomponents of a Markov model are the transition probability matrix, number ofcondition states (NCS), and length of duty cycle (LDC). Previous Markovian pavement performancemodels were developed using NCS and LDC based on data availability, pavementcondition indicator and data collection frequency. However, the selection ofNCS and LDC should also be based on producing pavement performance models withhigh levels of prediction accuracy. Prior stochastic pavement maintenanceoptimization models account for the uncertainty of the budget allocated topavement preservation at the network level. Nevertheless, variables such aspavement condition deterioration and improvement that are also associated withuncertainty, were not included in stochastic optimization models due to theexpected large size of the optimization problem.The overarching goal of this dissertationis to contribute to filling these research gaps with a view to improvingpavement management systems, helping to predict probabilistic pavementperformance and schedule pavement preventive maintenance accurately andreliably. This study reviews Markovian pavement performance models usingvarious Markov methodologies and transition probabilities estimation methods,presents a critical analysis of the different aspects of Markovian models asapplied in the literature, reveals gaps in knowledge, and offers suggestionsfor bridging those gaps. This dissertation develops a decision tree which couldbe used by researchers and highway agencies to select appropriate Markovmethodologies to model pavement performance under different conditions of dataavailability. The lack of consideration of pavement PM impacts intoprobabilistic pavement performance models due to absence of historical PM datamay result in erroneous and often biased pavement condition predictions,leading to non-optimal pavement maintenance decisions. Hence, this researchintroduces and validates a hybrid approach to incorporate the impact of PM intoprobabilistic pavement performance models when historical PM data are limitedor absent. The types of PM treatments and their times of application areestimated using two approaches: (1) Analysis of the state of practice ofpavement maintenance through literature and expert surveys, and (2) Detectionof PM times from probabilistic pavement performance curves. Using a newlydeveloped optimization algorithm, the estimated times and types of PMtreatments are integrated into pavement condition data. A non-homogeneousMarkovian pavement performance model is developed by estimating the transitionprobabilities of pavement condition using the ordered-probit method. Thedeveloped hybrid approach and performance models are validated using cross-validationwith out-of-sample data and through surveys of subject mat er experts inpavement engineering and management. The results show that the hybrid approachand models developed can predict probabilistic pavement condition incorporatingPM effects with an accuracy of 87%.The key Markov chain methodologies,namely, homogeneous, staged-homogeneous, non-homogeneous, semi- and hiddenMarkov, have been used to develop stochastic pavement performance models. Thisdissertation hypothesizes that the NCS and LDC significantly influence theprediction accuracy of Markov models and that the nature of such influencevaries across the different Markov methodologies. As such, this study developsand compares the Markovian pavement performance models using empirical data andinvestigates the sensitivity of Markovian model prediction accuracy to the NCSand LDC. The results indicate that the semi-Markov is generally statisticallysuperior to the homogeneous and staged-homogeneous Markov (except in a fewcases of NCS and LDC combinations) and that Markovian model prediction accuracyis significantly sensitive to the NCS and LDC: an increase in NCS improves theprediction accuracy until a certain NCS threshold after which the accuracydecreases, plausibly due to data overfitting. In addition, an increase in LDCimproves the prediction accuracy when the NCS is small.Scheduling pavementmaintenance at road network level without considering the uncertainty ofpavement condition deterioration and improvement over the long-term (typically,pavement design life) likely results in mistiming maintenance applications andless optimal decisions. Hence, this dissertation develops stochastic pavementmaintenance optimization models that account for the uncertainty of pavementcondition deterioration and improvement as well as the budget constraint. Theobjectives of the stochastic optimization models are to minimize the overalldeterioration of road network condition while minimizing the total maintenancecost of the road network over a 20-year planning horizon (typical pavementdesign life). Multi-objective Genetic Algorithm (MOGA) is used because of itsrobust search capabilities, which lead to global optimal solutions. In order toreduce the number of combinations of solutions of stochastic MOGA models, threeapproaches are proposed and applied: (1) using PM treatments that are mostcommonly used by highway agencies, (2) clustering pavement sections based ontheir ages, and (3) creating a filtering constraint that applies a rest periodafter treatment applications. The results of the stochastic MOGA models showthat the Pareto optimal solutions change significantly when the uncertainty ofpavement condition deterioration and improvement is included.