Weld Condition Monitoring Using Expert Informed Extreme Value Analysis

On-board acceleration measurements bear significant potential for the early detection of damage to railway infrastructure components. Condition assessment forms a complex problem in this case, due to the mobile nature of the On Board monitoring (OBM) solution, and the uncertainties associated with the rail-wheel contact dynamics, a lack of knowledge on the excitation sources (track, rail and wheel irregularities, parameter and self excitation) and the variability of the environmental and operational conditions. Welds are, amongst critical railway components, an essential element where high response amplitudes can occur. The monitoring of welds is still largely based on human assessment via typically visual means. We propose an automated approach to weld condition diagnostics via use of Extreme Value Analysis; an outlier detection schema which allows the early detection of flaws. In a second assessment phase, these potentially damaged welds are then assessed by experts during in-office and on-site inspections. The evolution of these OBM-based weld condition indicators is then tracked over time, leading to early detection of damaged welds.