Use of Flexible Wheelset Model, Comb Filter and Track Identification to Derive Rail Roughness from Axle-Box Acceleration in the Presence of Wheel Roughness

Abstract

Axle-box accelerometers fitted to in-service vehicles are a cost-effective way of monitoring rail roughness that can supplant the use of manual trolleys and dedicated measurement trains. Unlike the latter, accelerometers measure the vibration caused by roughness rather than the roughness itself, and so require processing to derive the levels of roughness. This is challenging due to variations in track stiffness, vibration coupling between wheels, and the presence of wheel roughness. This paper describes the derivation of rail roughness from axle-box acceleration (ABA) using a frequency-domain technique based on a vehicle-track model comprising multiple flexible wheelsets on two rails. The model is continuously updated by identifying variations in track stiffness from the P2 resonance peak in the ABA measurements. Furthermore, the effect of wheel roughness is removed from ABA by using a comb filter. The new methods are evaluated using measurements of ABA and rail roughness from the London Underground. The results demonstrate the derivation of narrow-band roughness spectra within 10 dB of conventional measurements across most of the 40–1000 mm wavelength range, even with variations in track stiffness.