Basis for an improved prediction model and source model for railway noise in Nord2000
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Sound and vibration (MPSOV), MSc
Publicerad
2024
Författare
Ratay, Vincent
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
The prediction models Nord2000 and CNOSSOS-EU are used to assess railway noise. The
source models of these prediction models are simplifications of the complex sound radiation
of trains. Therefore, the exposure from these source models close to the train is not accurate.
This thesis aims to improve the vertical sound power distribution over equivalent sources, as
well as the prediction of the overall sound power radiation from trains in Nord2000, based on
the speed and the type of train.
A linear equation system, which describes the propagation path between equivalent sources
on a train and several microphones, is used to streamline the calculation method in
Nord2000. For a more accurate prediction of the overall sound power radiation, based
on the speed and type of train, a linear regression model, a neural network, and a decision
tree model are built, trained and evaluated with a dataset of measured train pass-bys.
The determined linear equation system of the propagation path, in combination with existing
measurement data, is also solved to find the vertical sound power distribution over
equivalent sources.
The prediction models are evaluated using a test dataset, that is split from the original
dataset and not used in the training phase of the models. Comparison against the predictions
with the Nord2000 model shows a potential for improvement of the current prediction
model. The linear regression model yields reliable predictions, while the neural network
and decision tree are more affected by outliers in the data and, therefore, result in a worse
overall prediction. The method to find vertical sound power distribution only yields results
at low frequencies. Generally, more sound power contribution from the sources, higher on
the train, can be seen around 40 Hz and around 200 Hz.
While the prediction of the sound radiation from a train, based on the speed and type
of train, can be improved with the new linear regression model in this thesis, it was
not possible to incorporate the rail and wheel roughness into the models, as the wheel
roughness could not be derived from the available data. As the roughness of both the rail
and wheel is the root cause of rolling noise, the performance of the models might not be
limited by the amount of data, but the absence of the determining factor to predict rolling
noise, the roughness of the rail and wheel.
The results from the method to find a vertical sound power distribution on a train are
contrary to the results according to the literature. Finding an appropriate vertical sound
power distribution with measured sound exposure levels at only two microphones might
not be at all possible, as transfer paths between sources and microphones are too similar.
Finding an accurate vertical sound power distribution might require array measurements.
These measurements have to be carefully evaluated, as they can underestimate the sound
power contribution from the rail when beam-forming is only carried out in the normal
direction to the track
Beskrivning
Ämne/nyckelord
Railway Noise, Prediction Model, Source Model, Nord2000