Keywords: condition monitoring, drive-by monitoring, exploratory data analysis, anomaly detection, repeatability analysis, instrumented railway vehicle.

This paper presents an exploratory data analysis workflow that establishes baseline, expected time-series behaviour for instrumented railway vehicles by accounting for track location and context, such as turnouts, bridges, and curve transitions. A multi-sensor dataset spanning six months is analysed, focusing on bogie-frame vertical accelerations and secondary-suspension displacement estimates referenced to track position. A novel data quality score is applied to remove intervals with low supply voltage or physical impossibilities, filtering the dataset to retain only locations with sufficient passes for repeatability analysis. The study demonstrates that turnout regions exhibit significantly wider distributions and more frequent large responses compared to regular track segments, necessitating separate analytical treatment. The primary achievement is the generation of repeatability maps with contextual overlays. These maps facilitate systematic inspection and establish testable expectations for subsequent defect detection models in scenarios where validated defect labels are sparse or unavailable.

download the full-text of this paper (PDF, 2263 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description