Railway rolling noise separation and wheel and rail roughness estimation from pass by measurement: field tests and theoretical modelling

A desirable capability for those seeking to understand, and manage, railway rolling noise is the ability to separate the contributions from the wheel and track from revenue service train pass-by events. Relevant aspects of these contributions are the contributions to either: the excitation mechanism, driven by contacting surface roughness; or, to the noise radiated by the various elements of the vehicle and track. Efficient methods for both rolling noise separation, and wheel and rail roughness estimation, have been proposed, including a method known as VTN for noise separation and Janssens‘s Method for roughness estimation. Each relies on simplified models. The validity and acceptability for use have not been adequately tested. Field tests were conducted in order to test the validity of these methods. The results show that, using VTN as currently implemented, estimation errors relative to measured values, for overall noise level, of 2.3 dB(A). In contrast, predictions of the noise spectra result in errors of up to 20 dB at a certain frequencies. For roughness estimation, Janssens‘s method returns roughness values with errors of 10dB for some wavelengths for the total wheel and rail effective roughness. In an attempt to improve the accuracy of these methods, more accurate parameter estimation models have been developed. These are based on the novel use of: the Semi-Analytical Finite Element (SAFE) method for the rail vibration model; and the Wave-number Boundary Element (WBE) method for the rail radiation model. The SAFE rail model has been developed to accommodate multiple layers of rail support. Rail vibration and radiation behaviours have been investigated utilising these rail vibration and radiation models. The results show that 1) practical models for determining the key parameters for rolling noise and roughness prediction, incorporating more accurate assumptions for rail and rail support behaviour can be developed; 2) below 1 kHz, the assumed rail support has a significant effect on both the calculated rail vibration response and dispersion relations as well as the calculated rail radiation power; 3) the results also show that the assumed rail support has a limited effect on the rail radiation ratio, under a vertical excitation, and on the calculated rail radiation directivity, under either vertical or lateral excitation; 4) the Timoshenko beam-based rail model adequately predicts vertical rail vibration response for the rail foot centre but inadequately predictions of the rail lateral response, should accurate predictions be required for particular frequencies, as is the case for the separation of wheel and rail rolling noise from a total noise measurement; and, 5) the analytical line source model is inadequate for the prediction of the rail radiation directivity factor above 1 kHz, and the rail radiation ratio given lateral excitation. The parameters obtained by using the SAFE and WBE rail models developed have been applied to both the VTN rolling noise separation method and Janssens‘s roughness estimation method. Field tests undertaken show that 1) for those cases analysed, the application of the new models results in accurate prediction of rail radiation i.e. elimination of the over-estimation typical of the existing models, and 2) calculated wheel and rail total effective roughness correlation can be improved, with the total effective roughness from direct wheel and rail roughness measurements. In addition to providing more accurate parameters for the VTN and Janssens‘s methods, the models developed in this work can also be used to improve the accuracy of rolling noise prediction software as well as to investigate the design of rail dampers and low profile noise barriers. Validation of these results using more extensive field trials is required in order to confirm these results. In addition, although the rail models developed allow the inclusion of multiple layers of support with different characteristics, they only accommodate the assumption of a continuous support. A more accurate model would allow discrete supports so representing the actual support provided by the sleepers. The effect of this limitation is restricted to specific frequencies