Abstract:For the engine cylinder head vibration signal contains the relatively weak fault information, an feature enhancement method based on multiscale principal component analysis is proposed. First, vibration signal is decomposed by wavelet package and principal component analysis is used for all sub-bands coordinate transformation. Then, a signal is reconstructed in the new coordinate system. We use wavelet package to decompose the new signal and the energy of each sub-band is the feature vector of engine’s fault. Simulated signal testify the effectiveness of the proposed method. Experimnet used the proposed algorithm combined with support vector machine for eleven kinds fault of engine show that the fault classification accuracy could reach 98.76%.