Classification of internal combustion vehicle engines according to fuel based on generated sound

[M. M. Milivojčević]

VI, 137 listova

Biografija autora: list. 137 Bibliografija: list. 124-136

Acoustics

Noise pollution and exhaust emissions represent significant environmental challenges, with vehicle classification based on the type of fuel contributing to the reduction of air pollution and improvement of urban space management. This research develops an original system for acquiring engine sounds of passenger vehicles under real-world conditions, enabling autonomous vehicle detection, minimal environmental impact, and compatibility with future IoT systems. Key challenges included optimizing the microphone position to avoid obstructing vehicle passage, as well as selecting the engine operating mode that carries the most relevant acoustic information. The system was tested on 350 vehicles, forming a database of 475 sound samples, along with the implementation of a method for automatic extraction of acoustically relevant operating modes. Two groups of audio features suitable for distinguishing fuel types were identified: spectrogram representations (mel-spectrograms, gammatonegrams) and psychoacoustic characteristics (loudness, roughness, sharpness, crest factor). Classification was carried out using deep learning methods (CNN) and unsupervised learning (SOM), with CNN achieving an F1 score of up to 97%, and SOM reaching 96.7% after eliminating technically faulty vehicles. The contribution of this research lies in the development of an innovative acquisition system, the identification of key acoustic features, and the demonstration of machine and deep learning applications in engine sound analysis. The created sound sample database represents a valuable resource for the scientific community and future research in noise analysis, fuel type recognition, and fault detection. The results confirmed that it is possible to successfully classify vehicles based on fuel type using sound, opening opportunities for traffic monitoring, noise reduction, and the development of standards for assessing engine sound quality.

Klasifikacija vozila, motori sa unutrašnjim sagorevanjem, analiza zvuka, akvizicija akustičkih signala, spektralna analiza, psihoakustička obeležja, duboko učenje, mašinsko učenje

Vehicle classification, Internal combustion engines, Sound analysis, Acoustic signal acquisition, Spectral analysis, Psychoacoustic features, Deep learning, Machine learning

(534.87+534.836)004.85

T 121

Tekst