Expert evaluation system for pothole defect detection

• This study introduces an unsupervised approach for assessing pothole severity using vibration data. • The methodology incorporates IoT-enabled accelerometers facilitating scalable pavement monitoring. • A detailed performance comparison with a traditional wavelet-based changepoint algorithm is conducted. • The proposed system is validated using a full-scale study detecting pavement anomalies under realistic traffic conditions. • The adaptability of the proposed method is verified experimentally across varying test scenarios. The rapid deterioration of transportation infrastructure, accelerated by extreme weather events and increasing traffic loads, poses significant challenges for roadway maintenance. Potholes, a common form of pavement distress, not only compromise road safety but also increase vehicle maintenance costs and disrupt economic productivity. To address this issue, this study presents a mobile Internet-of-Things (IoT)-based pavement monitoring system for the automated detection and evaluation of potholes. The proposed system can detect and estimate pothole size based on real-time vibration data collected from unmanned ground vehicles (UGVs) by combining IoT-enabled accelerometers and a novel unsupervised threshold-based methodology. This paper introduces a scalable and cost-efficient framework that integrates IoT data acquisition technology with advanced pavement monitoring algorithms, providing municipalities and infrastructure managers with an automated solution for identifying and prioritizing road repairs. The proposed system was validated through multiple field trials and a full-scale study, where it accurately identified potholes of varying sizes across different road conditions. The threshold-based proposed approach is compared with a wavelet-based changepoint detection algorithm to demonstrate its versatility in delivering reliable and robust results even in adverse environmental conditions. The proposed method is validated using nine potholes of varying sizes which are successfully identified and estimated without any user intervention.