Monitoring railway rail roughness using on-train vibration measurements

Abstract

The growth of roughness on the rails of railway tracks is a significant problem that leads to increased noise and vibration emissions that affect residents near the railway. This study is devoted to a novel method of measuring rail roughness to inform maintenance operations. Unlike existing methods, this method is potentially more cost-effective and can provide more frequent updates on the condition of a railway network. The novel method uses low-cost accelerometers fitted to the axle boxes of a train or rail vehicle that is in revenue service, avoiding the need for track access. These measure the vibration resulting from the rail roughness instead of the roughness itself; this vibration is filtered from the roughness by the dynamic behaviour of the vehicle-track system, and so signal processing techniques are required to calculate roughness from the measurements. This thesis proposes and evaluates three techniques aiming to solve some of the specific problems affecting the measurements. Firstly, a procedure is developed to derive the wavelength spectrum of rail roughness from axle-box acceleration (ABA) using a frequency-response function (FRF) that describes the dynamic behaviour of the vehicle-track system, accounting for variations in vehicle speed that otherwise affect ABA. The FRF is derived from the receptances of the track and vehicle, both of which are defined by analytical models. Secondly, track stiffness can vary with location and time, and this significantly affects ABA. To address this, a technique is devised to identify the track stiffness from ABA by fitting the FRF to a peak in the ABA spectrum that is associated with the P2 resonance of the vehicle-track system, in order to update the FRF for the track that the vehicle is on. Thirdly, the presence of wheel roughness can affect ABA where it exceeds the level of rail roughness. The use of a comb filter is proposed to remove the effect of wheel roughness from ABA. The techniques are evaluated by a combination of linear and non-linear simulations and on measurement data from the London Underground Victoria line. Together, the techniques are shown to determine the spectra of rail roughness from ABA with some overestimations by up to 9 dB but almost without underestimations. Overall, these techniques potentially enable an automated measurement system with axle-box accelerometers on revenue-service vehicles to reliably detect high levels of rail roughness across a railway network.