Search

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023; Aachen : RWTH Aachen University 1 Online-Ressource : Illustrationen, Diagramme (2023). = Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2023, Vehicle interior noise is one of the most important auditory quality characteristics in the automotive sector. In this context individually perceptible noises are disturbing, if they are undesirable. Disturbing noises can imply either an insufficient product quality or even product defects. In this thesis, methods are developed to extract and classify disturbing noise components from monaural sound pressure measurements in the vehicle interior as well as from the an echoic engine test bench. The focus of this work are powertrain components that emit noise with tonal and impulsive noise characteristics. Image recognition methods are applied to frequency spectrograms and frequency order spectrograms to detect and extract tonal noises. The interpretation of tonal noises as lines in the image domain in combination with the adaptive segmentation of the spectrogram represents an important innovation to the state of the art. The extraction of impulsive noise components is achieved by different blind source separation methods. The subsequent grouping of individual impulsive noises is based on temporal correlation. The most important feature for the classification of tonal noise components is the temporal progression of the engine order. Its analysis enables the detection of slip, grouping of tonal noises and based on this the calculation of possible gear ratios. In this work, it is additionally elicited that the integer and the range of values of the engine order are important features for the classification of the considered components. The classification of impulsive noise, onthe other hand, is based on the frequency spectrum and sound pressure level descriptive features. In addition to the global frequency centroid, which describes accurately the noise phenomenon (knocking or ticking), local resonance excitations are detected and described quantitatively by local center frequencies and by its frequency widths. These features allowa description of the component specific noise characteristics. The sound pressure level aswell as the similarity of successive pulses further support the automated classification. The classification is achieved using support vector machines. The accuracy of the classificationof separated impulsive noises in the vehicle interior is 88% and 73% for anechoic enginetest bench accordingly. The accuracy is based on the respective test data. All developed methods have been implemented in a software tool with an intuitive user interface that allows non-acoustic experts to apply the sound separation and classification techniques to monaural vehicle measurements., Published by RWTH Aachen University, Aachen