Different highway agencies use different maintenance strategies. Our research indicates that each agency should compute its own cost estimates of maintenance on the basis of: • the measure of pavement performance that it uses to trigger maintenance (and accordingly, the appropriate value of the axle load power which is not necessarily equal to four), • and the actual maintenance activities and policies that it uses (which are not necessarily thick overlays applied at regular time intervals).

If agencies were to move toward more realistic cost estimates, they would be able to better forecast maintenance costs and plan their maintenance programs accordingly. They also could establish fees that would be more equitable across truck sizes and potentially even raise additional needed dollars.

A recursive non-linear model was developed for the prediction of pavement performance as a function of traffic characteristics, pavement structural properties and environmental conditions. The model highlights some of the advantages of relaxing the linear restriction that is usually placed on the specification form of pavement performance models. First, a functional form that better represents the physical deterioration process can be used. Second, the estimated parameters are unbiased, owing to a proper specification and the use of sound statistical techniques. Finally, the standard error of the prediction is reduced by half that of the equivalent existing linear model. This improved accuracy has important economic implications in the context of pavement management.

The model developed as part of this research enables the determination of an unbiased exponent of the so-called power law and of the equivalent loads for different axle configurations. The estimated exponent confirms the value of 4.2 traditionally used. However, it should be noted that this exponent is only to be used for determining damage in terms of serviceability. On the other hand, equivalent loads estimated for different axle configurations tend to differ from traditionally used values, especially in the case of single axles with single wheels.

The report attempts to deliver a proof of concept that optimal planning and management strategies can be formulated through applying robust optimization methodology such that limited resources could be allocated more rationally, and reliability of a highway network improved more efficiently.

The report focuses on two applications: the network design problem under demand uncertainty, and fleet allocation for freeway services patrols. Their corresponding decision-makings are formulated as several optimization models. By solving these models robust optimal strategies can be obtained. Numerical examples and simulation tests are presented to demonstrate the validity and usefulness of the proposed models.

The report has proved that through applying robust optimization methodology, robust optimal improvement strategies can be obtained to improve substantially the capacity of a highway system against high-consequence scenarios incurred by fluctuations in travel demand or irregular incidents.

The objective of transportation infrastructure management is to provide optimal maintenance, rehabilitation and replacement (MR&R) policies for a system of facilities over a planning horizon. While most approaches in the literature have studied it as a finite resource allocation problem, the presence of an underlying network configuration has been largely ignored. The recognition of the network configuration introduces several challenges, as well as opportunities, for system-level MR&R decision-making, which cannot be adequately handled by the existing decision-making frameworks.

This dissertation focuses on furthering the development of Markov decision process (MDP)-based system-level MR&R decision-making frameworks. In particular, two problems of interest are identified. The first problem concerns itself with identifying an optimal system-level optimization approach for solving budget allocation problems. The second problem of interest involves moving beyond traditional budget allocation problems to incorporate network considerations into system-level decision-making.

In the first part of the dissertation, a revised MDP-based optimization framework is proposed for solving the budget allocation problem. The framework, referred to as simultaneous network optimization (SNO), combines the salient features of the different MDP-based optimization approaches in infrastructure management literature, and provides optimal facility-specific MR&R policies for budget allocation problems. The proposed methodology is then compared with the other state-of-the-art MDP methodologies using a parametric study involving varying system sizes. The results of the study indicate the SNO outperforms the other MDP-based optimization frameworks.

In the second part of the dissertation, it is argued that while SNO is optimal for solving budget allocation problems, it can produce sub-optimal policies upon introducing network constraints. Consequently, the use of an approximated dynamic programming (ADP) framework is motivated to solve system-level MR&R decision-making problems involving network constraints. ADP facilitates the modeling of complex problem formulations by overcoming the curse of dimensionality associated with traditional dynamic programming frameworks.

To assess the suitability of ADP for system-level infrastructure management, two scenarios involving network considerations are investigated. In the first scenario, an approximate dynamic programming framework is proposed, wherein capacity losses due to construction activities are subjected to an agency-defined network capacity threshold. A parametric study is conducted on a stylized network configuration to infer the impact of network-based constraints on the decision-making process. The results indicate that ADP performs better than SNO when the network capacity constraints are binding on the decision-making process.

In the second scenario, the impact of introducing economies of scale (EOS) within budget allocation problems is investigated. Herein, incorporating network considerations leads to economic interdependence, wherein potential cost savings can be achieved by combining MR&R activities across adjacent road sections. Using parametric case studies, it is observed that the performances of ADP and SNO are comparable, with ADP improving upon the results of SNO under low budget and high EOS settings.

In conclusion, the findings from this dissertation indicate that ADP is a robust modeling framework for MDP-based infrastructure management problems. While previous research illustrates the use of ADP in solving system-level budget allocation problems, it is shown here that ADP is more relevant for modeling problems involving complex inter-facility dynamics. In particular, ADP is most beneficial in scenarios wherein finding optimal policies using analytical frameworks is not feasible.

Predicting pavement performance under the combined action of traffic and the environment provides valuable information to a highway agency. The estimation of the time at which the pavement conditions will fall below an acceptable level (failure) is essential to program maintenance and rehabilitation works and for budgetary purposes. However, the failure time of a pavement is a highly variable event; terminal conditions will be reached at different times at various locations along a homogeneous pavement section. A common problem in modeling event duration is caused by unobserved failure events in a typical data set. Data collection surveys are usually of limited length. Thus, some pavement sections will have already failed by the day the survey starts, others will reach terminal conditions during the survey period, while others will only fail after the survey is concluded. If only the failure events observed during the survey are included in the statistical analysis (disregarding the information on the after and before events), the model developed will suffer from truncation bias. If the censoring of the failure events is not accounted for properly, the model may suffer from censoring bias.

In this paper, an analysis of the data collected during the AASHO Road Test is presented. The analysis is based on the use of probabilistic duration modeling techniques. Duration models enable the stochastic nature of pavement failure time to be evaluated as well as censored data to be incorporated in the statistical estimation of the model parameters. The results, based on sound statistical principles, show that the failure times predicted with the model match the observed pavement failure data better than the original AASHO equation.

This dissertation presents a methodology for the joint optimization of a variety of pavement construction and management activities for segment and system problems under multiple budget constraints. The objective of pavement management is to minimize the total discounted life time costs for the agency and the highway users by finding optimal policies. The scope of the dissertation is focused on continuous time and continuous state formulations of pavement condition. We use a history-dependent pavement deterioration model to account for the influence of history on the deterioration rate.

Three topics, representing different aspects of the problem are covered in the dissertation. In the first part, the subject is the joint optimization of pavement design, maintenance and rehabilitation (M&R) strategies for the segment-level problem. A combination of analytical and numerical tools is proposed to solve the problem. In the second part of the dissertation, we present a methodology for the joint optimization of pavement maintenance, rehabilitation and reconstruction (MR&R) activities for the segment-level problem. The majority of existing Pavement Management Systems (PMS) do not optimize reconstruction jointly with maintenance and rehabilitation policies. We show that not accounting for reconstruction in maintenance and rehabilitation planning results in suboptimal policies for pavements undergoing cumulative damage in the underlying layers (base, sub-base or subgrade). We propose dynamic programming solutions using an augmented state which includes current surface condition and age. In the third part, we propose a methodology for the joint optimization of rehabilitation and reconstruction activities for heterogeneous pavement systems under multiple budget constraints. Within a bottom-up solution approach, Genetic Algorithm (GA) is adopted. The complexity of the algorithm is polynomial in the size of the system and the policy-related parameters.

In 2012, over 25% of the bridges in the United States were rated as structurally deficient or functionally obsolete. Moreover, 35% of bridges are serving beyond their theoretical design lifespan and the number has been projected to increase over the next decade. The imperative needs of improving the overall condition of the bridge system has been impeded by the shortage of funding available for bridge repairs and maintenance. In 2006 the gap between Federal Highway Administration's (FHWA) estimates to eliminate the bridge maintenance backlog and the actual appropriations to bridges for repairs and maintenance from the Highway Bridge Program was $43.4 Billion. In 2009, the gap increased to $65.7 Billion. Such conflict has made effective bridge management more critical than ever.

In bridge management, agencies collect bridge condition data and develop deterioration models that predict the bridges' future conditions and associated costs, based on which maintenance, rehabilitation and reconstruction (MR&R) decisions are made. It is therefore critical to have accurate deterioration models. However, limited availability of data and incomplete understanding of the deterioration process result in inaccurate models, which lead to sub-optimal MR&R decisions and significant cost increases.

To address the inaccuracy stemming from limited bridge condition data, researchers have proposed Adaptive Control (AC) methods that update the deterioration models successively as new data become available. The underlying belief is that agencies can obtain more accurate deterioration models through updating and subsequently improve their MR&R decisions and achieve cost savings. State-of-the-art bridge management systems, such as Pontis, use a class of AC procedures known as Certainty Equivalent Control (CEC). The procedure used in Pontis updates the transition probabilities (i.e., the parameters of the component deterioration models) after each condition survey, and uses the updated probabilities in subsequent planning of MR&R decisions. Unfortunately, CEC does not necessarily lead to more accurate models, or guarantee savings in system costs; in other words, updating of the type in Pontis is not necessarily beneficial.

In the present dissertation, an AC method, Open-Loop Feedback Control (OLFC), is proposed for system-level bridge management. The performance of OLFC and the Pontis CEC is tested in a numerical study and empirical results show that OLFC has superior performance with respect to two criteria. In terms of improvement in model accuracy, the Pontis CEC yields systematic bias in model parameter estimates and therefore does not improve model accuracy. In all testing scenarios, the resulting deterioration models lead to faster deterioration than the true models. OLFC, on the other hand, results in consistent convergence to the true models in all testing scenarios and improves model accuracy. When evaluated by system costs, the Pontis CEC consistently results in higher system costs than the no-updating scenario. The increases are on the order of $180 Million at the level of the State of California. To the contrary, updating with OLFC consistently achieves system costs savings compared to the no-updating scenario, and results in system costs that do not differ significantly from the system costs when true models are used for MR&R decision-making.

In addition, a computationally tractable optimization routine is developed for MR&R decision-making. The routine ensures strict conformity to system budget constraints and achieves satisfactory computational efficiency even given high levels of heterogeneity in bridge systems.