面向转子故障特征提取的多尺度拉普拉斯特征映射方法

Multi-scale Laplace Feature Mapping for Rotor Fault Feature Extraction

融合多尺度分解理论和流形学习思想,提出了一种面向转子故障特征提取的多尺度拉普拉斯特征映射算法。首先对转子故障振动信号进行多尺度小波包分解,提取各独立频带信号的最优尺度小波熵,构建特征参量矩阵并估计其固有维数,然后通过拉普拉斯特征映射将特征参量数据嵌入到低维本征空间,得到故障的最敏感特征,最后融合决策实现故障的准确识别。实验表明,相对于主成分分析算法、局部线性嵌入算法和拉普拉斯特征映射算法,多尺度拉普拉斯特征映射方法提取的转子故障信号特征更容易识别。

Based on theory of multi-scale decomposition and manifold learning thought,a multiscale Laplasse feature map algorithm for fault feature extraction was proposed.Firstly,the multi-scale wavelet packet decomposition of the rotor fault vibration signals was carried out.The optimal scale wavelet entropy of each independent frequency band signals was extracted,and the characteristic parameter matrix was constructed and the intrinsic dimension was estimated.Then the characteristic parameters of data were embedded into a low dimensional eigenspace by Laplasse feature mapping to get the most sensitive feature of faults.Lastly,the accurate identification of faults was realized by the fusion decision.Experiments show that,compared with the principal component analysis,local linear embedding and Laplacian eigenmap algorithm,rotor fault feature signal extraction of multi-scale Laplasse feature mapping method is more easily identify.