基于MSPCA的缸盖振动信号特征增强方法研究

Fault feature enhancement method for cylinder head vibration signal based on multiscale principal component analysis

针对发动机缸盖振动信号信噪比低的问题,提出了基于多尺度主元分析的故障特征增强方法。将缸盖振动信号小波包分解后,利用主成分分析对所有子带系数进行坐标变换,信号重构后再进行小波包分解,计算新坐标系下各子带的能量作为发动机故障的特征向量。仿真信号验证了本文所提算法对微弱冲击信号的增强能力,与支持向量机结合用于发动机十一种故障的诊断实例表明,故障分类准确率可达到98.76%。

For the engine cylinder head vibration signal contains merely the relatively weak fault information, an feature enhancement method based on muhiscale principal component analysis was proposed. A vibration signal was decomposed by wavelet package method and the principal component analysis was used for all sub-bands coordinate transformation. Then, the signal was reconstructed in the new coordinate system. The wavelet package method was used again to decompose the new signal and the energy of each sub-band becomes the feature vector of engine＇s fault. Simulated signals testify the effectiveness of the proposed method. The proposed algorithm combined with support vector machine has been used in the experiments for classification of eleven kinds of engine faults and the results show that the fault classification accuracy could reach 98.76%.