Enhancing Railway Pantograph Carbon Strip Prognostics with Data Blending through a Time-Delay Neural Network Ensemble

Abstract

Energy supply for high-speed trains is mainly attained with
a high-voltage catenary (i.e., the source on the infrastructure)
in contact with a sliding pantograph (i.e., the drain on the
rolling-stock vehicle). The friction between these two elements
is minimised with a carbon strip that the pantograph
equips. In addition to erosion, this carbon strip is also subject
to abrasion due to the high current that flows from the catenary
to the train. Therefore, it is of utmost importance to keep
the degradation of the carbon material under control to guarantee
the reliability of the railway service. To attain this goal,
this article explores an accurate (i.e., uncertainty bounded)
predictive method based on a robust online non-linear multivariate
regression technique, considering some factors that
may have an impact on the degradation on the carbon strip,
such as the seasonal condition of the contact wire, which may
develop an especially critical ice build-up in the winter. The
proposed approach uses a neural ensemble to integrate all
these sources of potential utility with the carbon strip data,
which is convoluted in time with a set of spreading filters to
increase the overall robustness. Finally, the article evaluates
the effectiveness of this prognosis approach with a dataset of
pantograph carbon thickness measurements over a year at the
fleet level. The results of the analysis prove that it is definitely
possible to deploy a fine prediction, and thus yield a new avenue
for business improvement through the application of the
predictive maintenance approach to pantograph carbon strips.