Machine learning methodology based on relative wavelet energy for identification of multiple out-of-round railway wheels

The term multi-damage encompasses situations wherein a structure or system experiences distinct types of damage or multiple instances of the same type of damage. Given the complexity inherent in real-world railway scenarios where several types of out-of-round (OOR) wheels may occur within one single train passage, this paper presents a new automated wavelet entropy-based methodology for multi-damage identification in railway vehicles. The proposed methodology consists on: (i) damage detection through wavelet entropy derived from acceleration responses, (ii) damage localization using the wavelet decomposition of the strain responses, and (iii) identification of the type of damage using a cluster analysis. This strategy is validated through a 3D numerical simulation of the dynamic response of the train-track interaction under distinct operational scenarios. The results show that it is possible to assess the type and quantity of damages existing in the train wheels based on a single passage and using an adequate wayside system.