Drive-by inference of railway track longitudinal profile using accelerometer readings taken by in-service vehicles

Accurate knowledge of the longitudinal profile of railway track is essential to support maintenance planning by track asset managers. The dynamic response of a train is largely dependent on the longitudinal profile of the... [ view full abstract ]

Accurate knowledge of the longitudinal profile of railway track is essential to support maintenance planning by track asset managers. The dynamic response of a train is largely dependent on the longitudinal profile of the railway track it crosses. This dynamic response can potentially be used to determine that profile. Cross Entropy optimisation is used to infer railway track longitudinal profile elevations through analysis of measured vehicle bogie accelerations with added uncertainty in vehicle and track properties. A numerical analysis is presented in this paper using a 2 dimensional half car vehicle and a finite element 3-layer track model implemented in Matlab. A population of track longitudinal profiles is generated through a random mechanism. A vehicle track interaction with randomly generated uncertainty in vehicle and track properties is carried out for each longitudinal profile in the population. The bogie acceleration signal produced for each profile is compared to the measured signal. The best fitting bogie accelerations are used to gather an elite set of rail longitudinal profiles. This elite set is used to generate an improved population of estimates for the next iteration. Once a convergence criterion is met the profile generating an acceleration signal that best fits the measured bogie acceleration signal is kept as the inferred longitudinal rail profile. This paper reports the results of the numerical simulations.