Detection of singular contact wire wear patterns at span scale by dimensionality reduction

Over the electrified rail network, the wear of the contact wire is caused by a variety of physical phenomena (e.g., mechanical, electrical, thermal). The monitoring of the thickness of the wire is mandatory to prevent catenary incidents and to plan relevant maintenance operations. However, even if wear is generally unavoidable, in certain cases the overhead contact line (OCL) can end up in a configuration favourable to wear, accelerating the phenomenon. In these cases, wear may be mitigated by appropriate OCL adjustments. This paper presents an approach to distinguish common wear, assumed to be inevitable and therefore not requiring specific adjustment, from the singular wear, which may be mitigated by appropriate adjustment of the OCL. The scale of interest is the span at which similar wear profiles are expected to be observed. We propose a two-step approach consisting, first, in a dimensionality reduction of the contact wire thickness measurements at the span scale, then in the analysis of the residuals to classify the wear pattern between common and singular. Two dimensionality reduction methods are assessed: the principal component analysis and the non-negative principal component analysis. Such a method aims to help to better handle maintenance and design operations. It allows to focus monitoring and maintenance efforts on locations classified as singular,