Abstract Summary
In this work, a method is proposed to detect and assess discontinuous/ fractured rail by analyzing the axle-box acceleration. The method uses a combination of unsupervised-machine learning algorithm and time-frequency analysis to detect the defect. In the previous work, fishplate rail joints modeling and impact loading induced by the track discontinuities was analyzed [1-2]. Most of the past work has been reported on fishplate joints, and not much work is found related to the broken welded rail joint. Hence, a study is performed to detect the rail discontinuities using axle-box acceleration. A multibody vehicle-track model is used to generate the acceleration data. The multibody vehicle-track model is developed in SIMPACK. The vehicle model consists of a coach, two bogies, and four axles. Linear spring and damper system is used to model the primary and secondary suspension of the vehicle. The equivalent stiffness of the track along the length of the track is calculated and imported into the SIMPACK model. A finite element-based Euler-Bernoulli beam model is used to calculate the equivalent vertical stiffness of the rail and its support. Two cantilever beams are used to model the two pieces of the fractured rail. The overhanging ends of the cantilever beams face each other. Sleepers/rail fasteners are modeled using equispaced springs that support the overhanging portion of the rail. These equispaced springs have stiffness equal to the combined stiffness of the railpad, sleeper, and ballast. Track vertical irregularity of levels six and five is modeled. These track irregularities are generated from the power spectral density function obtained by the Federal Railway Administration (FRA) of America. Results are obtained for different vehicle speeds, axle loads, and different overhanging lengths. To detect the defect, the axle-box acceleration passes through two stages. In the first stage, a clustering algorithm is applied to detect the location of the defect. Statistical features are calculated for the axle-box acceleration. The feature selection is done by the principal component analysis (PCA). In the clustering algorithm dominating feature is used to separate the defect signal from the rest of the data. The clustering algorithm works very well in separating the defect signal from the rest of the part irrespective of vehicle speed, axle-load, and different overhanging portions of rail. After locating the defect, in the second stage, the continuous wavelet transform method is applied to the data to detect the severity of the defect. The continuous wavelet transform efficiently classifies the severity of the defect in terms of the frequency content in the response. [1] Koro, K., Abe, K., Ishida, M. and Suzuki, T., 2004. Timoshenko beam finite element for vehicle—track vibration analysis and its application to jointed railway track. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, 218(2), pp.159-172. [2] Steenbergen, M.J., 2006. Modelling of wheels and rail discontinuities in dynamic wheel–rail contact analysis. Vehicle System Dynamics, 44(10), pp.763-787.
