A comprehensive data-driven approach to estimate track longitudinal level from inertial measurements

Abstract Infrastructure managers rely on diagnostic trains that periodically measure track geometry and vehicle accelerations to ensure the safety of the railway network. Their runs are scheduled depending on the line priority, in order to safely monitor the evolution of track defects. However, sudden and unpredictable defect growth may happen and be missed between successive runs. Therefore, condition monitoring systems have been installed on in-service vehicles. In fact, these trains run every day along the same line, so they can provide additional information useful for maintenance practices. When trains run along conventional lines, their speed significantly changes depending on the line characteristics, and vehicle accelerations strongly depend on speed. Therefore, monitoring systems that rely on vehicle accelerations should carefully take this effect into account. In this paper, a methodology to estimate the track longitudinal level using bogie accelerations from an in-service vehicle is presented. The recorded accelerations were double-integrated to account for the speed variation, and a model-based strategy was adopted to reduce the filtering action of the primary suspension. Data were recorded during a two-year monitoring campaign along an Italian railway line. The methodology allowed for the estimation of the longitudinal level along specific track sections, considering statistical measures like the peak value. A maximum error of 1 mm was found between the estimated values and those measured by the diagnostic train (considering a defect with magnitude of 7.5 mm). Therefore, the results showed that it is possible to estimate the peak longitudinal level between the two rails using one single vertical accelerometer installed on the bogie of an in-service vehicle. The results of this research may be used to support the current maintenance strategy with daily estimations of track longitudinal level. It should be noted that specific attention was given only to this type of track geometry parameter, since it often drives maintenance operations. In the future, the possibility to extend the methodology to the estimation of different type of defects, like cross-level and twist, could be considered.