Pavement maintenance is one of the major issues of public agencies. Insufficient investment or inefficient maintenance strategies lead to high economic expenses in the long term. Under budgetary restrictions, the optimal allocation of resources becomes a crucial aspect. Two traditional approaches (sequential and holistic) and four classes of optimization methods (selection based on ranking, mathematical optimization, near optimization, and other methods) have been applied to solve this problem. They vary in the number of alternatives considered and how the selection process is performed. Therefore, a previous understanding of the problem is mandatory to identify the most suitable approach and method for a particular network. This study aims to assist highway agencies, researchers, and practitioners on when and how to apply available methods based on a comparative analysis of the current state of the practice. Holistic approach tackles the problem considering the overall network condition, while the sequential approach is easier to implement and understand, but may lead to solutions far from optimal. Scenarios defining the suitability of these approaches are defined. Finally, an iterative approach gathering the advantages of traditional approaches is proposed and applied in a case study. The proposed approach considers the overall network condition in a simpler and more intuitive manner than the holistic approach.
Pavement management systems (PMS) should assist agencies in the decision making process about which sections of a pavement network should be preserved, maintained, and/or rehabilitated
Modules in a PMS used to evaluate the suitability of maintenance programs at the network level.
Once input data and management system modules are defined, the optimal design of maintenance programs is not straightforward. Indeed, it presents
Previous works in other research areas (i.e., bridge management and construction engineering) have analyzed the use of existing optimization methods in the decision making process [
The objective of this study is to recommend the most suitable approaches and optimization methods for the design of maintenance programs under different scenarios. Based on an analysis of the current state of the practice, this study proposes an iterative approach that gathers the advantages of traditional approaches (sequential and holistic) by considering the overall network condition in a simpler and more intuitive manner than with a holistic approach.
The study is part of a threeyear project developed in Chile by the Pontificia Universidad Católica de Chile (PUC) and named Fondef D09I1018 “Investigación y Desarrollo de Soluciones para la Gestión de Pavimentos Urbanos en Chile” (Research and Development of Solutions for Urban Pavement Management in Chile). The project is being partnered and advised by the Centre for Pavement and Transportation Technology (CPATT) of the University of Waterloo, Canada. The overall project resulted as a cooperative initiative of the PUC and funding partners to accomplish the current and future needs of urban pavements and provide effective management tools to assist agencies that manage urban networks in decision making. Even though the project is being developed in Chile, the expected outcomes, such as technical tools and the resulting Urban Pavement Management System, may be adapted and adopted in other countries for urban pavement management.
To achieve the proposed objective, a fourstep research method is proposed:
review of existing optimization methods applied to pavement management including existing applications in available PMS under traditional approaches (sequential and holistic);
comparative analysis of reviewed approaches and optimization methods, identifying their advantages and limitations; based on this analysis, recommendations of the most suitable approach and optimization method to implement in future PMS are driven considering different scenarios;
proposal of an iterative approach gathering the advantages of traditional approaches;
application of the proposed iterative approach in an illustrative case study and comparison to traditional approaches.
As shown in Table
Reviewed optimization methods consider either sequential or holistic approach.
Sequential approach  Holistic approach  

Optimization method  Treatment 
Section  
Selection based 
Judgment  A20  
Pavement condition  A19  
Economic analysis  A20  A8, A20  


Mathematical 
Linear and nonlinear programming  A12  A1  A6, A13 
Integer programming  A16  A17  A9, A22  
Dynamic programming  A8  A7, A11  A23  


Near optimization 
Incremental benefit/cost analysis  A18  A2, A18  
Local search heuristics  A5, A21  
Evolutionary algorithms  A11  A3, A7  A4, A14  


Other optimization methods  Neural networks  A10  
Fuzzy logic  A15 
Number of alternatives and type of approach considered in reviewed applications.
Code  Author  Problem  Approach  Reference  




Alternatives  Sequential  Holistic  
A1  AmadorJiménez and Mrawira  3  —  30 

x  [  
A2  Chamorro  39  4  10 

x  [  
A3  Chan et al.  500  —  —  500  x  [  
A4  Chootinan et al.  35  4  10 

x  [  
A5  Chou and Le  1  15  15 

x  [  
A6  De La Garza et al.  5  9  15 

x  [  
A7  Farhan and Fwa  150  4  1 

x  [  
A8  Feighan et al.  14  5  5–15 

x  [  
A9  Ferreira et al.  27  6  4 

x  [  
A10  Fwa and Chan  128  —  —  128  x  [  
A11  Fwa and Farhan  150  4  1 

x  [  
A12  Gao and Zhang  —  4  5 

x  [  
A13  Gao et al.  3  4  10 

x  [  
A14  Meneses and Ferreira  32  7  20 

x  [  
A15  Moazami et al.  131  —  —  131  x  [  
A16  Ng et al.  —  4  5–10 

x  [  
A17  Odoki and Kerali  Integer program.  100  16  5  x  x  [  
A18  Increm. benefit cost  400  17  12  x  x  [  
A19  Reddy and Veeraragavan  52  —  —  52  x  [  
A20  Shah et al.  21  4  10 

x  [  
A21  Tsunokawa et al.  —  5  20 

x  [  
A22  Wang et al.  10  5  5 

x  [  
A23  Yoo and GarciaDiaz  40  4  7 

x  [ 
Selection based on ranking is performed by enlisting and rating alternatives based on an indicator. This indicator can be based on judgment, pavement condition, or economic analysis.
When based on judgment, agencies determine from an expert panel a criterion to rate and rank alternatives. Shah et al. [
Selection based on pavement condition ranks sections to treat considering either a Single or a Composite Condition Index. Single Condition Index is normally based on roughness or structural index, whereas Composite Condition Index often considers pavement condition and functional classification. A Composite Condition Index considering pavement surface distresses, traffic information, and expert opinion is used by Reddy and Veeraragavan [
Ranking based on economic analysis allows a rational comparison among alternatives because it considers costs and benefits. This method was used by Shah et al. [
Mathematical optimization methods select alternatives maximizing or minimizing an objective function while satisfying some constraints. Objective functions commonly considered are maintenance costs, vehicle operating costs, and effectiveness, among others [
Linear and nonlinear programming seek optimal solutions using continuous variables. The main difference is that the former considers linear functions correlated with time, while the latter may consider curvilinear dependency [
Integer programming simplifies the analysis by considering two variables: a do nothing alternative or to do something. Applications are found using both sequential [
Dynamic programming is used in situations that require a number of sequential decisions. This optimization method starts at the desired final solution and works backwards to find the optimal value of variables. Dynamic programming has been applied using holistic [
Near optimization methods, also called heuristic methods, give solutions that are close approximations to those derived from mathematical optimization. These optimization methods start with an initial solution and look for better solutions within the constraints. They differ in how they search for better solutions: incremental benefit/cost analysis, local search heuristics, and evolutionary algorithms.
Incremental benefit/cost analyzes the benefits gained by selecting alternatives with higher costs. This optimization method is often referred to as the efficiency frontier. This frontier is defined in a plot of benefit against cost and gathers the alternatives with higher benefits given a certain cost. Incremental benefit/cost analysis is implemented in HDM4 PMS under a sequential approach for selecting the maintenance strategy and the sections to treat. However, HDM4 limits this application to a maximum of 400 sections, 17 treatment alternatives, and a 12year analysis period [
Local search heuristics start with random initial solution and explore the solution inference space seeking for better feasible solutions. Different local search heuristic can be implemented: gradient search, threshold acceptance, simulated annealing, and so forth. These heuristics have been applied under a sequential approach to optimize a road section treatment strategy: Tsunokawa et al. [
Evolutionary algorithms (EA) mimic the natural evolution guided by learning and adaptation. Among EA, genetic algorithms are one of the most applicable optimization methods in infrastructure management [
This section gathers optimization methods that assist decision making in managing pavement maintenance at the network level but they cannot be categorized in the above groups as neural networks and fuzzy logic.
Neural networks are able to learn from examples, enabling these systems to make generalizations and simulate decisions. Fwa and Chan [
Fuzzy logic systems incorporate imprecise qualitative data in the decision making. Moazami et al. [
This section analyzes the advantages and limitations of reviewed optimization methods and recommends the most suitable methods and approaches for future implementation in PMS under various scenarios. Finally, an iterative approach gathering the advantages of sequential and holistic approaches is proposed.
Selection based on ranking is easy to understand but it can only deal with a limited number of alternatives. Ranking based on judgment is the simplest method, but it may be subject to bias and inconsistency resulting in solutions far from optimal [
Mathematical optimization methods provide optimal solutions but they are not suited to deal with large networks. Indeed, mathematical optimization methods cannot handle large number of decision variables because this increases the complexity of the problem and requires long computing time.
In contrast, near optimization or heuristic methods provide simpler and more efficient solutions to large optimization problems. They are suitable to tackle the maintenance management at the network level leading to “good/near optimal” solutions [
Regarding other optimization methods, neural networks are useful to replicate a pattern and make generalizations. However, they do not guarantee the suitability of the decision taken and they act as a “black box,” being not possible to easily extract the path followed to explain a solution. Finally, fuzzy logic enable introducing rules from experience or intuition but it has no formal algorithms to learn from existing data [
Reviewed applications show a greater reliance on mathematical optimization and near optimization rather than on ranking (Table
Sequential approach simplifies the problem making it easier to understand than holistic approach because it defines first the treatment strategy and then selects the sections to treat. Nevertheless, sequential approach ignores the effect on the network as a whole. This may lead to recommending sequential approach for homogeneous or reduced networks in which the overall performance is less compromised by the section by section analysis. Regarding optimization methods, all the analyzed methods (ranking, optimization, and near optimization) have been used under the sequential approach. The recommendation on the optimization method to use would depend on the characteristics of a specific problem. In broad terms, selection based on judgment or condition should be avoided, as they may introduce bias and do not consider the effect of alternatives over time.
Holistic approach enables analyzing network maintenance alternatives as a whole, before any specific treatment strategy or section has been selected. However, this increases the complexity of the problem, making it necessary to use optimization and near optimization methods (Table
Gathering the advantages of sequential and holistic approaches, an iterative approach is proposed as shown in Figure
Decision making process of the proposed iterative approach.
The main difference between sequential and iterative approach is that the latter may select suboptimal treatment strategies for a certain section. On the contrary, reviewed applications considering a sequential approach only consider optimal solutions in the selection of treatment strategies. Therefore, iterative approach enables a deterioration of a solution at the section level if it leads to an improvement of the overall solution at the network level. As a result, the proposed iterative approach considers the overall network condition in a simpler and more intuitively manner than holistic approach.
Several of the reviewed optimization methods (ranking, optimization, and near optimization) may be used considering the proposed iterative approach. As the proposed approach considers two optimizations (treatment strategy and section selection), reviewed optimization methods may be combined. Indeed, iterations are also considered in the reviewed incremental benefit/cost analysis, as shown in the application of Videla and Gaete [
An illustrative case study is presented to compare the maintenance program obtained under traditional approaches (holistic and sequential) and the proposed iterative approach. The analyzed network, composed of five flexible pavements, is subject to both technical and budgetary restrictions. Each of the sections has a set of six possible rehabilitation treatments and a deterministic deterioration model adopted from Khurshid et al. [
Characteristics of sections considered in the case study.
Section  Type  Time since last 

1  Minimal SP with 102 mm ACO  15 
2  Minimal SP with saw and seal 102 mm ACO  20 
3  Intensive SP with 102 mm ACO  20 
4  Crack break and seat section with 102 mm ACO  25 
5  Crack break and seat section with 203 mm ACO  25 
The maintenance program seeks to maximize long term effectiveness (LTE) over a period of 25 years subject to budgetary restrictions. LTE of maintenance alternatives is assessed by the area bounded by the pavement performance curve (ABPC) and a threshold value (PSI ≥ 2, in this case study) (Figure
Long term effectiveness of a maintenance alternative.
This case study considers an available budget (in terms of total present worth cost, TPWC) 50% higher than the minimal cost solution that ensures a PSI greater than 2. This minimal cost solution (58 220 €) is taken as a base case to compare solutions obtained using different approaches. Although this case study considers an available budget higher than the minimal cost scenario, the proposed approach could deal with lower budgets. In fact, the ultimate goal of the proposed iterative approach is to assist pavement managers on the optimal design of maintenance programs subject to budgetary restrictions. Therefore, other budgetary scenarios could be similarly considered.
A local search heuristic based on simulated annealing was implemented on Matlab 12 in order to look for optimal solutions. Simulated annealing is based on the analogy of crystal formation from masses melted at high temperature and let to cool slowly [
Sequential and iterative approaches tackle the design of maintenance program by optimizing first the incremental cost effectiveness (IC
Optimal and suboptimal treatment strategies considered in the iterative approach.
Treatment 
Section 
Section 
Section 
Section 
Section 


IC ( 
IC/ 
IC ( 
IC/ 
IC ( 
IC/ 
IC ( 
IC/ 
IC ( 
IC/ 

Optimal  40 773  0.88  14 802  3.84  20 343  1.47  43 183  0.94  19 970  1.18 
Suboptimal 1  53 220  0.71  33 251  2.38  43 470  1.36  55 349  0.73  26 725  0.88 
Suboptimal 2  11 248  0.67  32 136  2.29  54 550  1.10  68 002  0.60  39 339  0.60 
Considering that the available budget is 50% higher than the minimal cost solution, there is an additional budget of 29 110 €. With this budgetary restriction for improving the network from the minimal cost scenario, sequential approach will solve the optimization problem by only treating Section
Treatment strategies for the different sections of the network under different approaches.
Sequential  Holistic  Iterative  

Section 
MC  MC  Suboptimal 2 
Section 
Optimal  Holistic optimal  Optimal 
Section 
MC  MC  MC 
Section 
MC  MC  MC 
Section 
MC  MC  MC 
Total present worth of solutions under different approaches.
Iterative approach, in contrast, enables the selection of suboptimal solutions at the section level looking for an increase in overall performance (Figure
Average PSI of the network under different approaches.
Finally, holistic approach selects a maintenance program based on minimum cost solution except of Section
From this numerical application it can be concluded that the proposed iterative approach leads to more efficient solutions than sequential approach while considering the overall network condition in a simpler and more intuitive manner than holistic approach.
From the literature review, two approaches are identified in the design of maintenance programs at the network level: holistic and sequential approach. The former tackles the problem as a whole, before any specific section or treatments are defined, dealing with the
Different optimization methods can be applied in the design of the maintenance programs at the network level considering either sequential or holistic approach: selection based on ranking, mathematical optimization, near optimization, or heuristic methods and other optimization methods. From the revision of these optimization methods and their applications the following can be concluded.
Ranking systems are easy to understand, but they can only be used when the number of alternatives is limited and they often ignore future needs.
Mathematical optimization methods provide optimal solutions, but they require long computing time. They may not be feasible for a large network with long period of analysis.
Near optimization methods give near optimal solutions with less computational effort than mathematical optimization methods. They can handle large number of decision variables and are suitable to solve combinatorial optimization problems.
Other optimization methods, such as neural networks and fuzzy logic can replicate a pattern, but they do not guarantee the suitability of the decision taken.
Based on the advantages and limitations of the reviewed optimization methods and their applications under holistic and sequential approaches, several recommendations can be driven for future implementation in PMS.
Sequential approach is easy to understand but it fails to consider the effect on the network as a whole. It may be recommended for the analysis of homogeneous or reduced networks, in which the overall performance of the network is less compromised by the section by section analysis.
Holistic approach analyzes network maintenance alternatives as a whole, before any specific treatment strategy or section has been selected. However, this increases the complexity of the problem, making it necessary to use optimization and near optimization methods. Reviewed applications of mathematical optimization methods using holistic approach show a trend of simplifying the problem or limiting the number of variables (sections, treatments, and/or analysis period). Meanwhile, near optimization methods are able to solve the problem under a holistic approach with no sacrificing of its complexity. Therefore, holistic approach using near optimization methods may be recommended when dealing with large networks.
Finally, an iterative approach is proposed and applied to an illustrative case study. This approach gathers the advantages of sequential and holistic approaches leading to more intuitive and effective design of maintenance programs at the network level. Based on a sequential structure, the proposed iterative approach includes iterations between the selection of treatment strategies and sections to treat looking for a more holistic view of the problem. In this iteration process, the proposed iterative approach may select suboptimal treatment strategies for a certain section if it leads to an improvement of the overall solution at the network level. As a result, the proposed iterative approach considers the overall network condition in a simpler and more intuitive manner than holistic approach.
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors gratefully acknowledge members of the research group at the Pontificia Universidad Católica de Chile for their contributions and resources during the study. The research team acknowledges ConicytFondef/Decimoséptimo Concurso de Proyectos de Investigación y Desarrollo del Fondo de Fomento al Desarrollo Científico y Tecnológico, Fondef/Conicyt 2009 (D09I1018) for funding this project. Support of the associated institutions is also appreciated: Ministry of Housing and Urban Development (Ministerio de Vivienda y Urbanismo), Regional Government for Metropolitan Region (Gobierno Regional de la Región Metropolitana), Municipality of Santiago (Municipalidad de Santiago), and Municipality of Macul (Municipalidad de Macul). Funding over Santander Universidades (Becas Iberoamérica Jóvenes Profesionales e Investigadores, 2013) to support this work is sincerely appreciated.