نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
This study investigates misfire detection in a four-cylinder, four-stroke gasoline engine by combining acoustic signal processing techniques with artificial neural networks. Misfires were induced by sequentially cutting fuel injection for each cylinder while the engine operated at a constant speed of 760rpm, and the resulting acoustic signals were captured under controlled test-cell conditions. Fast Fourier Transform (FFT), Mel-Frequency Cepstral Coefficients (MFCC), and Short-Time Fourier Transform (STFT) techniques were applied for feature extraction. The findings indicate that the artificial neural network (ANN) and the one-dimensional convolutional neural network (1D-CNN), when using FFT-derived features, achieved accuracies of 98.40% and 99.36%, respectively. Additionally, the two-dimensional convolutional neural network (2D-CNN), utilizing features extracted from the STFT, achieved an accuracy of 99.71%. These results demonstrate that the proposed methods—particularly the 2D-CNN perform exceptionally well in distinguishing between healthy and misfiring engine states, proving to be an effective approach for the real-time monitoring and fault diagnosis of gasoline engines.
کلیدواژهها English
