Keywords: anomaly detection, track geometry degradation, vehicle-track interaction, maintenance decision basis, track geometry assessment.

The assessment of track geometry degradation plays an important role for maintenance planning. State-of-the-art methods for assessment are commonly based on maximum and standard deviation values. A problem of analysing track degradation only by this pure geometry features is that possible occurrences of 'hidden anomalies' are missed. Hidden anomalies are defined in this paper as special track geometry shape patterns that cause unexpected vehicle responses under certain operating conditions. This paper describes a method to detect hidden anomalies in track geometry degradation by consideration of vehicle responses. In the first step, Multi Body Dynamics (MBD) simulation scenarios are carried out for different vehicle types, vehicle speeds, vehicle loading conditions and different wheel/rail friction conditions. The excitations are a combination of the real track layout and a set of historical track geometry irregularities. Virtual sensors measure wheel/rail forces as well as axle box, bogie and carbody accelerations. In the next step, time series data of each virtual sensor obtained by the simulation results are used to assess the 'current' response of the 'current' measured track geometry. Therefore, the track is divided into 200 m sections and the maximum values as well as the standard deviation values of the vehicle response forces and accelerations are calculated. For each virtual sensor, a linear regression model based on the historical responses is calculated. To assess the 'current track geometry', each vehicle response value (resulting of the current track geometry irregularities) is compared with its 95% prediction interval. A value outside the prediction interval means that the vehicle response value is not expected due to the historical track geometry degradation. For an overall assessment of the 'current track geometry', the expectation values of all sensors are statistically combined to a final expectation value for a hidden anomaly. To compare the proposed method with state-of-the-art methods, artificial anomalies are superposed to the real measured track geometry. In almost every section, these hidden anomalies cannot be detected by analysing only the pure geometry degradation behaviour. In contrast, the proposed method of this paper shows unexpected values for almost every section. This comparison emphasizes the high potential of the proposed method to detect hidden anomalies in track geometry measurement data.

download the full-text of this paper (PDF, 326 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description