Defect Prediction of Railway Wheel Flats based on Hilbert Transform and Wavelet Packet Decomposition
For efficient railway operation and maintenance, the demand for onboard monitoring systems is increasing with technological advances in high-speed trains. Wheel flats, one of the common defects, can be monitored in real-time through accelerometers mounted on each axle box so that the criteria of relevant standards are not exceeded. This study aims to identify the location and height of a single wheel flat based on non-stationary axle box acceleration (ABA) signals, which are generated through a train dynamics model with flexible wheelsets. The proposed feature extraction method is applied to extract the root mean square distribution of decomposed ABA signals on a balanced binary tree as orthogonal energy features using the Hilbert transform and wavelet packet decomposition. The neural network-based defect prediction model is created to define the relationship between input features and output labels. For insufficient input features, data augmentation is performed by the linear interpolation of existing features. The performance of defect prediction is evaluated in terms of the accuracy of detection and localization and improved by augmented input features and highly decomposed ABA signals. The results show that the trained neural network can predict the height and location of a single wheel flat from orthogonal energy features with high accuracy.