Development of Empirical and Mechanistic Empirical Performance models at Project and Network levels
Performance prediction models are a vital component in pavement management systems (PMS). Along with decision trees, prediction models are used to set priorities for maintenance and rehabilitation planning, and ultimately for budget allocations at the network level. Reliable and accurate prediction of pavement deterioration over time helps transportation agencies accurately predict future spending and save significant amounts of money. Within a PMS, raw performance data is often converted into aggregated performance indices, such as the Riding Comfort Index (RCI), to quantify the road’s roughness, or the Distress Surface Index (SDI), to quantify accumulated pavement distress. Technology has evolved rapidly in the last two decades, making data collection for pavement conditions (i.e. roughness and distress data) more feasible for transportation agencies. However, transportation agencies, especially at the municipal level, only maintain condition data to evaluate the present pavement status. Only limited attempts have so far been made to develop or enhance existing deterioration models in pavement management systems, using periodically collected condition data over time. A well-maintained historical database of pavement condition measurements and performance indices can be a useful source for the development of performance prediction models. In some cases, however, the database may contain incomplete data and insufficient information to develop reliable performance models. In addition to inconsistency in the historical performance data, the age of the pavement or the date of the last maintenance/ rehabilitation treatment may not be available to develop the pavement performance over time. The goal of this research is to develop enhanced empirical performance models capable of capturing the unpredictable and indeterminate nature of pavement deterioration behavior. This research provides a methodology to develop empirical models in the absence of the construction and/or rehabilitation dates. The models developed in this research use limited available historical data, and examine different parameters, such as pavement thickness, traffic pattern, and subgrade condition. Parameters such as the date of pavement construction and the age of the pavement are also incorporated into the proposed models, and are constrained by local experience and engineering judgment. A linear programming optimization technique is employed to develop the empirical models presented in this research. The approach demonstrated in this research can also be expanded to account for additional parameters, and can easily be adapted to match the needs of different agencies based on their local experience. In addition, the current research develops a second set of deterioration models based on mechanistic-empirical principles. Models incorporated into the mechanistic-empirical design guide are locally calibrated. A genetic algorithm optimization technique is employed to guide the calibration process, in order to determine the coefficients that best represent pavement performance over time. The two sets of performance models developed in this research are compared at both the project and network level of analysis. A decision-making framework is implemented to incorporate the two sets of models, and a comprehensive life cycle cost analysis is carried out to compare design alternatives in the project level analysis. The two model sets are also evaluated at the network level analysis using a municipal pavement management system. Two budget scenarios are executed, based on the developed performance models, and a comparison between network performance and budget spending is presented. Finally, a summary and current research contribution to the pavement industry will be presented, along with recommendations for future research.