| dc.description.abstract | Pavements play a vital role in the transportation infrastructure in the United States. Pavement performance modeling is an essential step in pavement design and management. Recently, several software and tools have been developed to help to design a pavement at the project level. Pavement mechanistic-empirical (ME) design is one of the AAHSHTOWare Design software built to design new and rehabilitated pavements with flexible, rigid, and composite structures. The nationally calibrated performance models in Pavement ME do not well represent the construction and materials specifications, traffic, and climate conditions specific to each state and cannot precisely reflect the pavement performance. On the other hand, at the network level, the pavement performance should be monitored regularly, and maintenance and rehabilitation (M&R) treatments should be planned to keep the pavement in good condition. An acceptable treatment policy maximizes the service life and returns the benefits of the constructed pavement. The goal of this research is to enhance designing models for the flexible pavement of Oklahoma and develop a new M&R decision process using the surface roughness and structural capacity of the pavement section.
The nationally calibrated models show an improper prediction performance and a significant bias, which asserts the necessity of local calibration. Local calibration of Pavement Mechanistic-Empirical (ME) software improved the pavement performance prediction models and optimized the performance models for the pavement network of Oklahoma. The locally calibrated coefficients for distress and IRI models were determined for the Oklahoma pavement system. As a result, the error in performance prediction models was reduced through the calibration process. The distress and IRI models show that the calibrated coefficients improve Pavement ME predictions and the design of flexible pavements in Oklahoma.
The second objective of this research was developing a new maintenance and rehabilitation decision process which considers the stochasticity of the pavement performance prediction and suggests the optimized maintenance activities for the given section by implementing a newly developed predictive model. The developed M&R decision method is a Markov Decision Process that employs IRI from a newly developed IRI prediction model and structural number from historical data. The IRI prediction model predicts the IRI with high accuracy by having the structural number, road class, climate condition, traffic load, and subgrade and structural information. Several advanced machine learning techniques were investigated, and the best model was implemented in the MDP M&R method. This model considers the M&R activities from pavement history, which affects the pavement deterioration rate, and suggests an M&R Policy for the given pavement system. By improving predictions and developing effective maintenance decision policies, machine learning algorithms can optimize maintenance and rehabilitation interventions and reduce maintenance costs. | |
