Railway Track Geometry Degradation Modelling and Prediction for Maintenance Decision Support

This paper describes methods for predicting track geometry degradation to provide support for maintenance planning. The track geometry data recorded by the Track Recording Car (TRC) and the maintenance work orders were used for degradation modelling. The degradation indicator considered in this study was the deviation of the longitudinal level from the design value sampled every 100 m-long track segment. A Wiener process model was built to model the time-evolution of the geometry condition indicator for a track segment and the parameters were estimated via standard maximum likelihood estimation techniques. It was found that the proposed degradation model provided good fit to the data, except in some extreme cases where the TRC data exhibits large changes in the degradation indicator between two runs. Nevertheless, even in these cases the Wiener process model “responds” to this lack of confidence about the slope in an intuitively appealing manner: the variance on the slope is increased to reflect a lack of confidence in the degradation rate. Finally, the utility of the model in maintenance decision making is demonstrated by predicting the expected number of different maintenance interventions and policies.