Railway surface faults detection using dark field illumination and machine learning

Developing countries like Pakistan uses visual inspection for monitoring the health of railway tracks, which is hazardous as single negligence can result in a catastrophic outcome. Given the fact, that 70 % of railway accidents are caused by the lack of railway track condition monitoring. Therefore, this research focuses on the development of a realtime fault identification algorithm, which can diagnose track surface damages. The algorithm developed a binary classifier that detects the health of railway tracks using a novel frame design which is having dark field illumination algorithm. The accuracy achieved from the developed algorithm is over 90 % and it is validated on actual railway tracks, such as Kotri Junction, Pakistan Railways. Index Terms—Dark Field Illumination, surface faults, Real-time identification, Visual inspection, Optical sensor, Binary Classifier.