Risk-Informed Support Vector Machine Regression Model for Component Replacement—A Case Study of Railway Flange Lubricator

Abstract

The railway-rolling stock wheel flange lubricator protects the wheels and railhead by lubricating their contacts. Failed or missing flange lubricators can lead to excessive wheel wear, wheel flats, wheel cracks, rolling contact fatigue, rail damage, and derailment accidents. In extreme cases, missing or worn flange lubricators due to nonlinear rail conditions may lead to fire hazards, particularly in underground rail infrastructure. In addition, the location of lubricators present accessibility issues and prolong the diagnosis of failure. This study therefore proposes an adaptive risk-based support vector regression (SVR) machine with a Gaussian kernel function that can accurately and proactively predict the wear loss of flange lubricators from a small data set. While most flange lubricators fail owing to wear loss, others fail owing to premature failure modes such as cracks and fatigue. The risk-informed feature evaluates failure rates associated with failures other than wear loss to support a balanced determination of the optimised replacement frequency. The proposed model was applied and validated as a case study for the London underground train. The findings showed that the optimised maintenance inspection of the flange lubricator, as a balance between safety and organisational resource constraints, was an average of every 4000 km between train operations.