Measurements and predictions of wheel / rail vibration using a 1/5th scale rig

The scale rig, which was donated by London Underground Ltd, has been modified to incorporate a new track bed that resembles a mainline concrete sleeper track. Rail lengths can be interchanged to allow measurements of vibration due to different railhead profiles. Several practical methods have been devised. For example, a new method of the measurement of wheel and railhead surface profiles has been developed to improve the quality of these inputs to the prediction models. Another method is presented, for the situation of a surface roughness input, which determines the decay of vibration along the length of the track. This has been formed from measurements of rolling vibration by estimating the rail acceleration at the point of wheel / rail contact and comparing this with the average acceleration over a length of track. The estimate of rail acceleration at the point of contact has also been used to estimate the effect of wheel / rail contact patch filtering by comparing this (the acceleration at the point of contact) with a prediction of rail vibration. The performance of two types of prediction models are considered: existing linear and non-linear wheel / rail contact models are investigated, adapted, and improved for application to the 1/5 scale rig. These existing prediction models have been investigated in the form of transfer functions (displacement divided by roughness) to demonstrate the effect of various contact stiffness values whilst the wheel remains in contact with the rail. It has been found that low contact stiffness attenuates the transfer function from the high frequencies. The transfer functions have also been expressed as poles and zeros and were found to remain stable for a wide range of contact force (or stiffness) values. The outputs from the prediction models have been compared with measurements of rail vibration on the scale rig for surface roughness, step-joint, dipped rail, and simulated wheel flat profiles. A method of splitting the input to the prediction models has been implemented to allow separate analysis of the surface roughness and discontinuity parts of the input. Good agreement is shown between the measurements and predictions for the majority of the railhead surface profiles considered.