The calibration challenge when inferring longitudinal track profile from the inertial response of an in-service train

Abstract

An Irish Rail intercity train was instrumented for a period of one month with inertial sensors. In this paper, a novel calibration algorithm is proposed to determine, with reasonable accuracy, vehicle model parameters from the measured vehicle response data. Frequency domain decomposition (FDD) is used to find the dominant frequencies in the captured data. Randomly chosen 2 km data segments are chosen from a number of datasets, thereby averaging out the effects of variations in track longitudinal profile, track stiffness, signal noise and other unknowns. The remaining dominant peaks are taken to be vehicle frequencies. An optimization technique known as Cross Entropy is used to find vehicle mass and stiffness properties that best match modal vehicle eigenfrequencies identified in the frequency analysis. Finally, the calibrated vehicle is run over a measured track profile and the resulting model output is compared to measured data to validate the results.