ENBIS-16 in Sheffield

11 – 15 September 2016; Sheffield Abstract submission: 20 March – 4 July 2016

Uncertainty Estimation of One-Step Ahead Railway Track Geometry Failure Predictions

13 September 2016, 15:40 – 16:00


Submitted by
Bjarne Bergquist
Bjarne Bergquist (Luleå University of Technology), Peter Söderholm (Swedish Transport Administration)
In this presentation, we exploit the prediction uncertainty of using recursively updated time series predictions of remaining useful life (RUL) for railway track. By RUL, we mean the time until dimensional properties of the track have degraded to a safety critical fault state, where actions such as speed restrictions and track alignment are required.

We use Kalman filters for predicting the evolvement of track property data. The uncertainty of the predictions relates to: measurement uncertainty (instrument imprecisions including measurement location uncertainty); uncertainty of substructure properties at measurements (e.g. because of spring thaw and frost heave), and irregular sampling frequency. Repeated sampling consisting of returning measurement trains at the same track section in a short time interval support point estimate uncertainties and tests for violation of the Gaussian assumptions. We also discuss the sampling intervals in terms of the model’s prediction abilities in light of practical concerns such as the difficulties to perform measurements during different yearly seasons.
View paper

Return to programme