The surface condition of roadways has direct consequences on a wide range of processes related to the transportation technology, quality of road facilities, road safety, and traffic noise emissions. Methods developed for detection of road surface condition are crucial for maintenance and rehabilitation plans, also relevant for driving environment detection for autonomous transportation systems and e-mobility solutions. In this paper, the clustering of the tire-road noise emission features is proposed to detect the condition of the wheel tracks regions during naturalistic driving events. This acoustic-based methodology was applied in urban areas under nonstop real-life traffic conditions. Using the proposed method, it was possible to identify at least two groups of surface status on the inspected routes over the wheel-path interaction zone. The detection rate on urban zone reaches 75% for renewed lanes and 72% for distressed lanes.
Keywords: data clustering; pavement condition; road surface; tire-road noise; unsupervised machine learning.