Rail corrugation detection using one-dimensional convolution neural network and data-driven method

Abstract

Rail corrugation is a very common wear phenomenon occurring on rail surface, especially on sharp curves, and is one of the main excitation sources of noise and vibration during railway transportation. Timely monitoring of rail corrugation is of great benefit to make scheduled maintenance and save maintenance costs. This paper proposes a novel method combining deep learning and data-driven fusion algorithms to detect rail corrugation on metro lines. First, a one-dimensional convolutional neural network (1DCNN) is constructed to intelligently identify the state of rail corrugation and classify its wavelength. Then, a vehicle–track coupling dynamics model considering the flexibilities of the wheelsets and track structure is established to simulate the dynamic response of the axle box under the excitation of rail corrugation. Next, the depth characteristics of rail corrugation can be calculated using the Kriging surrogate model (KSM) and particle swarm optimization (PSO) algorithms. Finally, a series of field tests are carried out and the feasibility of the proposed method has been verified based on the field measurement data. The results demonstrate that the 1DCNN–KSM–PSO data-driven method can efficiently and quantitatively detect the severity of rail corrugation with a relative error less than 15% and an average error of 6.75%.