Analysis of the Influence of Image Processing, Feature Selection, and Decision Tree Classification on Noise Separation of Electric Vehicle Powertrains

Journal Article
10-07-01-0002
ISSN: 2380-2162, e-ISSN: 2380-2170
Published September 23, 2022 by SAE International in United States
Analysis of the Influence of Image Processing, Feature Selection, and
                    Decision Tree Classification on Noise Separation of Electric Vehicle
                    Powertrains
Sector:
Citation: Fröhlingsdorf, K., Dreßen, M., Pischinger, S., Steffens, C. et al., "Analysis of the Influence of Image Processing, Feature Selection, and Decision Tree Classification on Noise Separation of Electric Vehicle Powertrains," SAE Int. J. Veh. Dyn., Stab., and NVH 7(1):23-33, 2023, https://doi.org/10.4271/10-07-01-0002.
Language: English

Abstract:

With the increasing electrification of road vehicle powertrains, new NVH challenges are emerging. Due to the temporary deactivation of the internal combustion engine or even its complete elimination, the interior sound is primarily determined by noise shares of the electrical components and shares of tire and wind noise. In particular, the high-frequency tonal noise components induced by the electric powertrain are often dominant. These components usually have a negative impact on the perceived sound quality of the total vehicle. Major noise-causing components of the electric powertrain are the electric machine and the transmission. Since acoustic optimization is a complex and time-consuming process, this article develops and presents a novel method for the automated separation of the noise shares of the electric machine and the transmission gears in the vehicle interior. This method is based on image processing using Hough transformation for the detection of tonal order lines in order spectrograms. The detected lines are then allocated to their emitting components by classification via a decision tree whose features are selected by sequential feature selection. Performed variations of feature selections and classifications show that the proposed method performs best. The entire separation process is designed to work even without the knowledge of component parameters such as the number of pole pairs of the electric machine and the number of teeth of the gears in the transmission. The classification accuracy of electric machine and gear orders in the vehicle interior is about 86%.