The use of an artificial neural network for assessing tone perception in electric powertrain noise, vibration and harshness

Abstract

The transition from internal combustion engines to electric powertrains brings new challenges for the Noise, Vibration, and Harshness (NVH) analysis of these vehicles. The tonal nature of the electromagnetic excitations and of the gear meshing mechanism are reflected in the radiated noise of electric powertrains, often leading drivers and passengers to rate the noise from electric vehicles with an increased nuisance even if they are quieter than internal combustion driven powertrains. In this paper, a flexible multi-body dynamics model is developed to calculate the vibration and forces transmitted from the bearings to the housing of an electric powertrain. Acceleration, force and sound spectra data are used to train an artificial neural network to assess the prominence of tones in the noise based on the results of the structural simulation. The results show it is possible to identify psychoacoustic metrics from the multibody dynamics simulation alone. With this new approach, it is feasible to quickly investigate how changes in the powertrain will affect the tonal perception of the noise without the need of new acoustic simulations and experiments. For the tonal perception analysis, the Prominence Ratio is used as a metric. This framework of combining multibody dynamics simulation with initial acoustic data and neural networks can be also applied to different NVH metrics as appropriate.