摘要
局部线性嵌入法(locally linear embedding,LLE)是一种典型的流形学习算法。在分析LLE算法的基本计算思路的基础上,提出了一种基于最佳分类效果的k和d综合参数选择方法。此方法综合考虑了故障类内和类间的离散度,并以此作为LLE算法特征压缩效果的评价依据。根据LLE算法的局部线性特征保持的基本特点,提出了一种增量式LLE算法用于柴油机机械故障特征压缩与诊断中。以平均子带能量法构造特征向量空间,子带数目的确定以同种故障类型特征参数间方差最小为准则。实验中,分别使用基于最佳参数选择的LLE算法、传统的主成分分析(principal component analysis,PCA)、增量式LLE算法对柴油机特征向量进行压缩,并对这三种算法的特征压缩结果运用K近邻算法(K-nearest neighborm,KNN)进行故障诊断与分类。结果表明基于最佳参数选择的LLE算法的诊断分类效果要优于传统的PCA方法,增量式LLE算法也取得良好的分类效果。实验表明,对LLE算法进行有关改进可以很好地应用到机械故障特征压缩与诊断中。
The locally linear embedding（ LLE） is a typical manifold learning method. The basic steps of the LLE method was analyzed,and a parameters selection method of the neighborhood factor k and the embedding dimension d based on the optimum comprehensive classification result were presented. The evaluation basis of the feature compression result of LLE method is the same and different fault types＇ discrete degree that considering in the parameters selection method. According to the basic characteristic of the local linear constant in the LLE method,an incremental LLE method was presented. And the method is applied to the feature compression and diagnosis of diesel mechanical faults. The eigenvector space is structured by the average sub-band energy method. And the number of the sub-band is determined by the method of feature parameters minimal variance in the same fault type.The diesel eigenvector space is compressed respectively by the parameters selection LLE method,the principal component analysis（ PCA）,and the incremental LLE method. And then the results of the feature compression are respectively diagnosed and classified by the K-nearest neighbor（ KNN） method. The result shows that the parameters selection LLE method is better than the PCA method,and the incremental LLE method has a good effect. The experiment shows that the improved LLE method can good apply to the feature compression and diagnosis of mechanical fault.
出处
《科学技术与工程》
北大核心
2016年第13期86-91,共6页
Science Technology and Engineering
基金
西安石油大学优秀硕士学位论文培育项目(2014yp130410)资助
关键词
改进LLE算法
机械故障诊断
特征压缩
子带能量
improved LLE method
mechanical fault diagnosis
feature compression
sub-band energy
作者简介
王江萍（1959-），教授。研究方向：机械系统传感测试理论及信息处理技术，机械故障诊断技术。E—mail：jpwang@xsyu．edu．cn。
崔锦（1990-），硕士研究生，研究方向：机械故障诊断技术。E—mail：cumin20082009@163．com。
