Machine learning aided rail corrugation monitoring for railway track maintenance

Urban rail transit is a critical infrastructure system thatsupports urban economic and social development. It has a significant mass transportation capacity while enables environmental benefits. Public transport is a way to resolve large-scale urban road traffic problems and contributestowards sustainable development. However, with the operations of railway vehicles on curves, unbalanced and undulated wears often appear on rails, especially on the low rail. This rail surface defect, so-called ‘rail corrugation’, directly affects the service life of rolling stocks and track components. The high-frequency vibrations caused by train-track interaction over rail corrugations also impair passenger ride comfort and generate excessive noises. In severe cases, the defects may even endanger the safe passage of a railway vehicle. In practice, rail corrugation has brought huge challenges to the reliable operations and maintenance of railway networks. With the continuous expansion of railway lines and the increasing traffic demands, any existing rail corrugation test method is not enough to meet the actual needs of track maintainers to promptly identify and mitigate rail surface defects. Therefore, this investigation aims to establish a new technique to prognose and classify rail corrugations efficiently and effectively. This study adopts D-track dynamic simulation package to obtain over thousands of vibration data in the form of axle box accelerations from train-track interactions under different conditions. Neural network models have been developed to recognize the rail corrugations and then classify their severity to aid the planning and prioritization of rail track maintenance activities. The models have been trained and tested using the vibration data, achieving the accuracy of over 90%. The optimal model has then been highlighted. The investigation has demonstrated the potential of the neural network to detect and classify rail corrugations, which can be used practically for curved track condition monitoring and maintenance planning.