Keywords: pavement, maintenance, management, expert system, knowledge-based system, decision making.

The aim of this work is to develop a knowledge-based system that can represent human expertise on making pavement maintenance decisions over an extensive road network. The development of such a system can reveal the importance of each parameter involved in the decision making process, provide decision support to maintenance agencies, and facilitate further attempts to develop an integrated highway management system.

Previous research efforts reported in the literature have led to the development of a few expert systems focusing mainly on specific aspects of pavement maintenance. A limitation of existing systems is that they involve only the major decision parameters (probably due to the difficulty of knowledge acquisition and coding in a multi-parameter decision making process).

The knowledge-based system that is presented in this work incorporates several parameters that are taken into account by maintenance engineers (such as distress type, severity and extent, expected deterioration rate, road functional classification and traffic loads, pavement age, soil type, and environmental conditions) and consists of three decision modules, one for setting maintenance priorities, another for determining feasible treatments for each road section, and a third for maintenance planning and resource allocation at a network level.

The first module records all parameters which are considered in the decision making process along with their relative weights. Each parameter takes a number of values or qualitative assignments. A road section is assigned a grade for each parameter. Following a multi-criteria analysis, a total grade is computed for each section, which indicates the maintenance priority. Parameter weights and section grading are at this stage set by experience and checked by a trial and error technique. The objective of such operation was to reach an acceptable level of coincidence regarding priority ranking between the system output and the actual expert decisions.

The second module investigates all applicable treatment considering the specific conditions of each section. A rule-based expert system has been developed to represent the expert's decision making under each possible condition. The system employs forward reasoning as an inference engine and may be depicted in the form of a decision tree. Besides identifying applicable treatments, the system also sorts these treatments in each case according to their effectiveness-cost ratio. Effectiveness is considered in terms of the expected lifetime of each treatment until another treatment will be necessary.

The resource allocation module aims to select the sections that will be treated and the most appropriate maintenance action for each one based on the output of the previous modules and considering any budget constraints. In particular, starting from the section with the highest maintenance priority, the best cost-effective treatment among all applicable ones is proposed. The process continues with the following sections in the priority list until the available budget is allocated. If the number of sections that are treated is less than desirable, within the available budget, alternative less costly treatments may be selected so that more sections are maintained. An optimal decision is made based on a multi-objective linear programming formulation. In particular, the conflicting objectives are to maximise the number of sections that are maintained (considering also the priority list) and to apply treatments with as high effectiveness-cost ratio as possible (such treatments are typically the most expensive ones). The two objectives are accordingly weighted to cope with user requirements and constraints.

The methodology has been evaluated with actual and simulated data (generated through a Monte Carlo technique). Evaluation results indicate that it can provide a valuable tool for assisting maintenance decisions.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £105 +P&P)