带标志点的LTSA算法及其在轴承故障诊断中的应用  

Local tangent space alignment algorithm based on selecting landmark points and its application in fault diagnosis for rolling bearing

作　　者：杨庆[1] 陈桂明[1] 江良洲[1] 何庆飞[1] 

机构地区：[1]第二炮兵工程学院装备管理工程系,陕西西安710025

出　　处：《振动工程学报》2012年第6期732-738,共7页Journal of Vibration Engineering

摘　　要：针对非监督式流形学习算法面临的增量式学习问题,提出一种带标志点的增量式局部切空间排列算法。该方法在局部切空间排列算法的基础上,利用最小角度回归算法从原始训练样本中选取标志点,以选取的标志点和新增样本建立所有样本的全局坐标矩阵,利用原始样本低维嵌入坐标和全局坐标矩阵对新增样本的低维嵌入坐标进行估计,并采用全局坐标矩阵特征值迭代方法更新所有样本的低维嵌入坐标。滚动轴承4种不同状态振动数据样本的增量式识别结果表明,本方法在实现局部切空间排列算法增量式学习的基础上,保持了对滚动轴承不同状态样本较高的类别可分性测度。As for the incremental learning of the unsupervised manifold learning algorithms the paper provides an incremental local tangent space alignment （LTSA） algorithm based on selecting landmark points for the clustering of incremental points. Firstly, the least angle regression （LARS） algorithm was used to select landmark points from original training samples, and then using those selected points and new samples the global coordinate matrix for all samples could be constructed. The low- er-dimensional embedding coordinate of the incremental samples was estimated according to the global coordinate matrix and lower-dimensional embedding coordinate of the given training points. Then the lower-dimensional embedding coordinate of all samples was iteratively updated based on the eigenvalues of the global coordinate matrix. Experiments with the proposed method were carried out in four different cases of the vibration signal samples of rolling bearing. Results demonstrate the in- cremental LTSA algorithm based on selecting landmark points can preserve the class separability measure of the samples in different cases for rolling bearing.

关 键 词：局部切空间排列算法 最小角度回归算法 增量式学习 模式识别 滚动轴承 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]