Integrated Multiple-Defect Detection and Evaluation of Rail Wheel Tread Images using Convolutional Neural Networks

The wheel-rail interface is regarded as the most important factor for dynamic behaviour of a railway vehicle, affecting safety service, passenger comfort, and life wheelset asset. degradation wheels in contact with rail visibly manifest on their treads form defects such indentations, flats, cavities, etc. To guarantee reliable service maximise availability rolling-stock assets, these need to be constantly periodically monitored severity evolves. This inspection task usually conducted manually at fleet level therefore it takes lot human resources. In order add value this maintenance activity, article presents an automatic Deep Learning method jointly detect classify wheel tread based smartphone pictures taken by team. architecture approach framework Convolutional Neural Networks, which applied different tasks diagnosis process including location defect area within image, prediction size, identification type. With information determined, maintenancecriteria rules can ultimately obtain actionable results. presented neural has been evaluated set collected over course nearly two years, concluding that reliably automate condition half current workload thus reduce lead time take action, significantly reducing engineering hours verification validation. Overall, creates platform or significant progress automated predictive rolling stock wheelsets.