基于测地线距离的核主元分析在齿轮泵故障识别中应用

针对传统的高斯径向基核函数中采用欧氏距离计算方法难以完全反映非线性振动数据样本点与点之间位置关系的问题，提出了改进的核主元分析方法.在高斯径向基核函数中使用测地线距离代替欧氏距离，建立基于样本类内散度和类间距的评价函数，运用遗传算法优化测地线距离中邻近点参数k以及高斯径向基核函数中参数σ.对采集的齿轮泵不同状态的振动数据进行经验模态分解，从分解的各阶本征模态分量和残余分量中提取10个无量纲参数构成原始特征参数集；运用优化后的核函数对原始特征参数集进行核主元分析.实验结果表明，改进的核主元分析方法取得了较好的识别效果.

Application of Kernel Principal Component Analysis Based on Geodesic Distance in Pattern Recognition of Gear Pump

As traditional Gauss radial basis kernel which adopts the method of calculating Euclidean distances can not describe completely the relationship between the nonlinear vibration sample data,an improved kernel principal component analytic method was proposed.The method substitutes geodesic distance for Euclidean distance in Gauss radial basis kernel,establishes an appraisement function based on scatter of the sort and space between different sorts,and then uses genetic algorithm(GA) to optimize the k paramete...