Automatic Rail Track Surface Anomaly Detection with Smartphone Based Monitoring System

Abstract

Railroad companies spend a significant portion of their revenue on track inspections to maintain safety and maximize operational efficiency. Deviations from the designed track geometry over time could lead to poor ride quality and possible derailments. The existing approaches to track inspections are expensive and laborious. The use of low-cost sensors aboard revenue service trains to screen the infrastructure for track irregularities could improve the cost-efficiency of track inspections by targeting the available resources to high-risk locations. Unevenness of rail track running surfaces cause dynamic forces generated at the wheel/rail contacts which in return results the vibration of the car. This study focuses on detecting track unevenness by associating its influence on vibration with it. A comparative analysis is carried out on unevenness response prediction and the accuracy of detecting such track surface unevenness is analyzed with the ground truth location collected by the railroad track inspectors. The main finding of the study were 1) the unevenness event estimation error are within 15 matters with one run for one phone based system and 2) the three-phone based track surface anomaly detection system can improve its forecasting accuracy to 5 meters and to 3 meters with two traversals.