2019 CSCE Annual Conference - Laval (Greater Montreal)

2019 CSCE Annual Conference - Laval (Greater Montreal) Conference


Title
Development of a Modified Pavement Condition Index for Provincial Highways in Ontario Using Machine Learning Technologies

Return to Session

Author(s)
Dr. Guangyuan Zhao, University of Waterloo
Ms. Ju Huyan, University of Waterloo (Presenter)
Dr. Susan Tighe, CPATT - University of Waterloo
Dr. Wei Li, Chang'an University
Abstract

The evaluation of pavement condition is an integral part of Pavement Management System (PMS). The pavement condition is typically expressed as a numerical index, whereby the schedule of pavement maintenance and rehabilitation can be triggered, the extent of pavement repair work and associated cost can be estimated, and various pavements can be compared so that the optimum network management strategy can be determined. In Ontario, pavement condition is evaluated in terms of two interrelated performance measures: ride quality and distress manifestation. The ride quality is denoted by Ride Condition Index (RCI) based on International Roughness Index (IRI), whereas distress manifestation is represented by Distress Manifestation Index (DMI). Currently, an Automatic Road Analyzer (ARAN) has been utilized to collect data in ride quality and distress manifestation by the Ministry of Transportation of Ontario (MTO) to calculate Pavement Condition Index (PCI) in a simple linear model as the overall measurement of pavement condition. Meanwhile, Pavement Condition Rating (PCR), a subjective overall performance measure, is also assigned based on the experienced inspection crew’s perception in riding comfort and observed surface distresses. With the two overall performance measures (PCI and PCR) serving the same purpose but limited resources in terms of labor and time were found available for PCR, there is a need to evaluate the correlation between PCI and PCR and whether one can replace the other. According to the research conducted in automatic pavement condition performance assessment of Ontario highways, significant differences have been observed between PCI and PCR, which might impose strong impact on the following M&R strategy prioritizations. Facing this problem, this research aims to develop a modified-Pavement Condition Index (m-PCI) for Ontario provincial highways with different pavement types (Asphalt Concrete, Composite, Portland Concrete, and Surface Treated pavements) by utilizing the state-of-the-art machine learning technologies. Firstly, correlation analysis was conducted between PCI and PCR using the collected dataset containing more than 10,000 observations from year 2010 to 2014. The Pearson correlation coefficient was found varying between 0.4 and 0.75 among different pavement types. Next, statistical analysis and machine learning approaches (Neural Networks, XGBoost methods) were employed simultaneously for the performance modeling of different pavement types and comparative experiments, thereby proposing a novel m-PCI for more accurate pavement performance evaluation. Finally, the effectiveness of m-PCI was validated by the test dataset (20% of the total dataset) and the usage guidelines were provided for a more dynamic and reliable automatic PMS.