Experimental and numerical investigation of the possibilities for the structural health monitoring of railway axles based on acceleration measurements

In this article, numerical and experimental investigations are carried out to assess the possible use of vibration measurements to identify the presence of a fatigue crack in railway axles, detecting components of axle vibration occurring at frequencies that are integer multiples of the axle’s frequency of revolution (N×Rev components). A model of a cracked axle is defined using the Timoshenko beam finite elements incorporating an equivalent beam element having cross-sectional area and moments which are periodically changing with the angular position of the axle turn to reflect the crack breathing mechanics. Two levels of validation are considered for this model: first, verification with numerical results from a solid model of railway cracked axle (solid model was built before using the solid finite element model) and, second, validation against experimental results obtained from full-scale tests performed on cracked axles using a rotating bending fatigue test rig available at the labs of Politecnico di Milano. To consider the effect of disturbances arising from train–track interaction, track irregularity and wheel out-of-roundness, a multi-body model of a complete railway vehicle is defined, incorporating one axle modelled as a flexible cracked axle. This model is used to evaluate the N×Rev components of axle vibration occurring in the presence of axle cracks having different positions along the axle and different depths, in combination with different sources of disturbance. The results show that the 2×Rev and 3×Rev components of the horizontal axle-box acceleration are well correlated to the size of the crack and are almost insensitive to the effect of all disturbances considered in this study. Hence, they can be used for the continuous monitoring of axel integrity. Finally, a criterion for crack detection is defined and applied to the experimental and numerical results, showing the possibility of detecting cracks with sizes in the order of 8% of the total section area.