Road-Deterioration Detection using Road Vibration Data with Machine-Learning Approach

Recently, the maintenance and management of infrastructure, such as paved roads and bridges, at a low cost has become important. Although some measurement methods including the falling weight deflectometer test have been developed to assess the soundness of paved roads, it is difficult to measure the data in a constant manner, for instance, on a daily basis. Therefore, we present an approach as per which we install vibration sensors on paved roads and automatically detect the deterioration of the paved roads via the installed vibration sensor and a machine-learning technique. Deterioration detection techniques that exploit vibration sensors have been studied; however, those were limited to bridge monitoring. No studies for the vibration measurement of paved roads using fixed sensors have been conducted. Herein, we focus on the deterioration of paved roads, specifically, in the form of road cracks, and conduct vibration measurements that highlight the differences in the vibrations of roads with and without cracks. In this paper, we describe the vibration measurements of a paved road with and without cracks and propose a framework for detecting cracks. An anomaly detection technique is necessary for using our detection framework. In this paper, we also evaluate the detection performance using anomaly detection techniques - namely, one-class support vector machine, isolation forest, and local outlier factor - using the measured vibration data.