EMD-Based Intelligent Crack Detection in Freight Railway Axles

The failure of railway axles can lead to catastrophic accidents, with the human and economic consequences that this entails. The vibratory performance of a freight train bogie is studied in this work aiming to identify the defects induced in the wheelset. The defects are generated mechanically with four severity levels. The bogie is tested in a roller rig test bench and vibration signals are recorded from sensors placed in the axle boxes of the wheelset. These signals are decomposed into several sub-signals using the Empirical Mode Decomposition. Then, the spectral power of these sub-signals is used as input for a Feedforward Neural Network to classify the vibration signals according to the defect level. The results show that the trained network can accurately identify the presence or absence of wheelset defects and their severity.