Statistical Approach to Model Track Dynamics Towards the Monitoring of Railway Turnouts

This paper proposes a method for the automatic generation of statistical models that describe the track stiffness in terms of the first and second track resonance frequencies, which are associated with the dynamic behavior of the ballast and rail pad layers. The method combines the empirical mode decomposition and a subspace identification method to estimate the track resonance frequencies from track vibrations induced during train passage.

The generalized extreme value distribution is found to be a robust descriptor of the estimates of the first and second track resonance frequencies across time and space. The method is demonstrated on track acceleration data collected over a period of two years at fixed locations along a turnout for three different types of trains, and models are built for the switch panel, the closure/crossing panel and a transition zone.

Further, it is shown that the degradation of track components is captured through the recursive generation of such statistical models, which can then become the basis for the development of condition monitoring systems.