Non-Destructive Evaluation of Track Stability Using Doppler Lidar Systems

Abstract

An approach for non-destructive evaluation of revenue service track stability using Doppler Lidar measurements is provided. The Lidar system is installed onboard a track geometry car for non-contact measurement of the track lateral and vertical movement in response to the weight of the rolling wheel. The deployment of such velocity-based sensors could result in early and timely detection of track infrastructure stability issues, specifically any reduced vertical (tangent track) and lateral (curved track) stiffness, which could diminish the track’s dynamic stability margin. Two Lidar systems along with supporting instrumentation are installed in the cab of a track geometry car that is commonly used for evaluating and recording the condition of revenue service track. The recorded measurements include left and right rail lateral, vertical and longitudinal (down track) velocities, GPS position, and axle-mounted tachometer speed. Using a precise calibration sequence, the exact beam angles between the lenses and track are calculated from the Lidar measurements. The calibration constants are included in the calculations to ensure agreement between the left- and right-rail velocities, the track center-line speed, and speed measurements with onboard instruments (e.g., tachometer, GPS sensors). The recorded data are processed to ensure that the analyzed data represent the components of the measurements that are essential to the track condition. The GPS Velocity data and numerical interpolations are used to clean any Lidar drop-in and drop-outs where the signal drops below the noise floor for the measurements. In addition, the velocity data are scaled such that the left and right rails have the same forward velocities on a tangent track, as would be expected. To retrieve the lateral component of the Lidar signal, a high-pass filter is first applied to remove the effect of the longitudinal velocity on the recorded data. An unsupervised machine learning technique is developed to identify potentially unstable track segments using an automated data analysis process in Python. A visualization platform is also created to show the analyzed segments on the Google Map to accommodate any visual inspection or track maintenance. The study indicates promising results for early detection of track movement that could develop, in time, into track instability.