The predictive power of track dynamic response for monitoring ballast degradation in turnouts

This paper develops a novel method for the prediction of the level of degradation of the ballast layer in railway turnouts by establishing a statistical mapping between the train-induced vertical track acceleration and the vertical track geometry. This is performed by exploiting the full-scale data provided by a loaded geometry car and the track-side sensing system over a two-year period. Irregularities in the vertical track geometry are analyzed and correlated with the level of ballast degradation by means of fractal analysis. Partial least squares regression is then employed to identify the components of the power spectral density of vertical track acceleration, which provided more information about the ballast degradation. The proposed method has a remarkable predictive power of the ballast degradation as evaluated both in terms of accuracy and consistency of the prediction. The reported results, based on the data from four different locations of the turnout before and after a ballast tamping event, show a mean absolute prediction error between 2% and 8% and a minimum correlation greater than 90%. The proposed method enables the continuous monitoring of the track substructure components by exploiting the vertical track acceleration as a supplementary source for the prediction of ballast degradation.