Machine Learning Methodology for Identification of Multiple Out-of-Round Railway Wheels using Data from Wayside Monitoring Systems

This research presents a new automated diagnosis methodology for out-of-round multi-damage wheels that addresses the damage detection and localization, using only acceleration and strain data measured on the railway track.The methodology is based on wavelet relative energy and comprises two stages: i) detect damage through the wavelet entropy derived from vertical acceleration responses and ii) localize damage by mathematically processing wavelet decomposition and using strain responses to determine the specific axle location of the detected damaged wheel.The proposed methodology is numerically validated for two different types of out-of-round damage in railway vehicles, such as polygonal wheels and wheel flats, and for a five-car freight train with different damage combinations and localizations.