Abstract

Abstract

Machine learning for homogeneous grouping of pavements.
Kanan Mukhtarli

Rapid pavement deterioration is a major problem in areas with harsh weather conditions or high traffic loading. Despite many studies focused on the pavement management systems, there is not, to the date, a robust method explaining how to process large amounts of pavement data to create homogeneous groups for rehabilitation-related decision making. This thesis employs machine learning to develop an approach capable of partitioning pavement data with a close response to casual factors like traffic and weather conditions and considering its performance through international roughness index and deflections. Two different methods: K-means and Self Organizing Maps (SOM) clustering techniques were tested to understand the correlation between daily factors and pavements deterioration. The goodness of clustering was tested using extrinsic and intrinsic evaluation methods. It was concluded from the results that SOM clustering provided better results as it relies on a soft clustering method where one point can represent two clusters at the same time. Moreover, it became obvious from the methodology that including the previous year’s data has very little to no effect on homogeneous groups. Techniques discussed and developed in this study can help road asset managers with decision making for the maintenance and rehabilitation of pavement. Moreover, future researchers can use the results of this study to further develop the idea of building decision support systems for pavement rehabilitation.