The operating condition of internal combustion engine changes frequently. So its vibration sources are various and frequency spectrums of different source's responses overlap. There are all kinds of random noises during the operating condition, so the signals-to-noise ratio (SNR) is low, and Correlation analysis method and frequency spectrum analysis method are out of date when they are applied to fault detection and diagnosis of internal combustion engine based on its vibration signals analysis. A wavelet neural network is constructed. Wavelet analysis is used to de-noise and SNR is enhanced to a satisfactory degree. A newly defined characteristic vector derived from the de-noised signals is inputted into BP neural network and misfire fault is successfully detected.
展开▼