Prediction of Railway Track Condition for Preventive Maintenance by Using a Data-Driven Approach

Abstract

For scheduling track maintenance, infrastructure managers perform inspections to assess the track condition. When the irregularities are higher than the threshold limits, a track has a defect that should be corrected to avoid future failures or traffic disruption. Scheduling maintenance actions contribute to reliability and availability but demand the prediction of the evolution of track degradation. In recent years, several degradation models have been defined to forecast geometrical evolution over time and/or tonnage, mainly for heavy rail systems. Nevertheless, most of those models have limitations when dealing with measured data collected in different time intervals as happens in reality. To overcome this problem, a data-driven model based on the logistic binary function is presented and validated with real inspection measurements. The results prove that the model has a 91.1% success rate, an excellent discrimination ability, and a high sensitivity, classifying correctly 84.1% of inspections in need of maintenance. The model also has high specificity as it classifies 94.5% of inspections with no demand of maintenance action. The model is easy to implement, which is also an advantage for the track asset management with guaranty of excellent sensitivity and discrimination.