基于小波包域主成分分析的缸盖声压特征增强方法研究

Feature Enhancement Method for Cylinder Head Acoustic Pressure Based on Wavelet Package-Principal Component Analysis Method

针对发动机缸盖声压中故障特征信号较为微弱的问题,在主成分分析坐标变换思想的基础上,提出了基于小波包域主成分分析的缸盖声压特征增强方法。在缸盖声压信号高通滤波后进行小波包分解,对各子带的小波系数建立主成分分析(principal component analysis,PCA)模型,将缸盖声压信号变换到PCA坐标系下,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故障特征向量。仿真信号验证了小波包域主成分分析对微弱冲击信号的增强能力。新方法与支持向量机结合用于发动机11种工况的诊断实例表明:故障分类准确率达到97.53%。

For extraction of the relatively weak fault information contained in engine cylinder head noise signals, basing on coordinate transformation theory of principal component analysis （PCA）, a signal enhancement method by wavelet package-principal component analysis was proposed. Filtrated by a high-pass filter, the cylinder head noise signals were decomposed by wavelet package, and principal component analysis was used for each sub-band to transfer the original coordinate system. Then, signals were reconstructed in the PCA coordinate system and decomposed again using wavelet package method and the energy of each sub-band was considered as the feature vector of engine fault. The simulation signals obtained demonstrate that weak shock signals were effectively enhanced by the wavelet package-principal component analysis method. Practical example using proposed method combined with the support vector machine to detect engine faults in 11 operation conditions shows that the fault classification accuracy reaches 97.53%.