Predicting the Wear Rate and Thickness of Train Contact Wire Using Data-Driven Modelling

Predictive maintenance of railroads is essential to prevent costly disruptions. A critical aspect of this maintenance is ensuring the integrity of the pantograph-catenary system, where copper alloy wires experience continuous friction and wear. The degradation rate and condition of these wires are vital factors in planning maintenance activities. Current wear rate prediction methods are largely theoretical and often inaccurate, overlooking essential contextual details. Additionally, wire condition data frequently show inaccuracies and inconsistencies in spatial and temporal resolution, complicating the feasibility of using data-driven approaches. This research investigates a data-driven framework to accurately predict wear rates, emphasizing data processing and optimized data use. A dataset spanning nine years of Dutch railway infrastructure measurements is used, employing various machine learning techniques to determine the most effective approach. Findings indicate that, in 95% of cases, average wire thickness can be predicted with a precision of ±0.12 mm over a four-year period. This study advances the field by proposing a framework that addresses measurement errors, a common challenge in sensor-based assessments, making data-driven maintenance a more reliable option.