Using a 2.5D boundary element model to predict the sound distribution on train external surfaces due to rolling noise

Abstract

[EN] In order to be able to predict train interior noise, it is first important to calculate the external sound pressure distribution on the floor, sidewalls and roof. This can then be combined with the transmission loss of the train panels to determine the interior noise. Traditional techniques such as the finite element and boundary element (FE/BE) methods in three dimensions (3D) can achieve this result but are computationally very expensive. In this paper, a wavenumber-domain boundary element (2.5D BE) approach is instead adopted to predict the propagation of rolling noise from the wheels, rails and sleepers to the train external surfaces. In the 2.5D models, only the cross-section of the vehicle is represented by using boundary elements, while the third direction is considered in terms of a spectrum of wavenumbers. The rail is treated directly in the wavenumber domain but, to include the wheel, a method of representing point sources in a 2.5D approach is developed. An inverse Fourier transform is applied to obtain the spatial distribution of the sound pressure on the train surfaces. The validity of this approach has been verified by comparison with experimental data. The 2.5D BE method was first used to predict the sound distribution on a 1:5 scale train surfaces due to a point source below the vehicle, and later it was used to predict the sound pressure on a full-scale metro vehicle due to a loudspeaker. Comparisons of predictions with measurements on the scale model and on the metro vehicle showed good agreements. For a point source below the vehicle, the sound pressure levels on the train floor were found to be around 20 dB higher than on the sides, and the sound pressure on the train roof was negligible. The 2.5D BE method was also used to predict the sound pressure on the metro vehicle surfaces in running operation, in which the predicted sound pressure levels on the train external surfaces agreed with measurements to within 3 dB and similar trends were found in terms of spectra and longitudinal distribution of pressure.