基于核形态成分分析的齿轮箱复合故障诊断研究

形态成分分析（MCA）是最新提出的一种基于稀疏表示的信号和图像分解(分离)方法，该方法的主要思想是利用信号组成成分的形态差异性（可以由不同的字典稀疏表示）进行分离。结合核函数把基于MCA的线性盲分离方法拓展到非线性混叠情况，给出了一种非线性混叠信号盲分离算法。该算法通过非线性映射将混叠信号投影到高维特征空间，将样本空间的非线性混叠问题转化成高维特征空间的线性混叠问题，然后应用MCA算法对高维特征空间中的混叠信号进行分离。通过对齿轮齿根裂纹、轴承内圈、外圈复合故障的实验信号的分析，表明该方法能有效地分离出齿轮箱的复合故障。

Compound fault diagnosis for gearbox based on kernel morphological component analysis

Morphological component analysis(MCA) is a novel decomposition method based on sparse representation of signals and images.A nonlinear blind source separation algorithm was proposed by extending the linear blind source separation algorithm based on MCA to the nonlinear domain.The mixing signals were first mapped to high dimensional feature space,the nonlinear mixing problem was converted into linear mixing problem,and the MCA algorithm was then applied to linear mixing signals in the feature space.The experiment results show the efficiency of the proposed algorithm.