基于LTSA与K-最近邻分类器的故障诊断

发布时间：2018-04-05 23:07

本文选题：局部切空间排列　切入点：K-最近邻分类器　出处：《振动与冲击》2017年11期

【摘要】：针对局部切空间排列算法(LTSA)的效果受近邻数k值影响较大的缺点,提出基于聚类准则的LTSA与K-最近邻分类器的故障诊断模型。基于振动信号的时域特征构建高维特征矩阵;对高维矩阵进行标准化预处理,依据聚类准则确定局部切空间排列中的最佳近邻数k,运用LTSA提取高维矩阵的低维特征向量;将提取的低维特征向量利用K-最近邻分类器进行故障模式识别。采用轴承诊断实验系统进行验证,结果表明,基于聚类准则的优化方法可有效地克服近邻数k选择的盲目性,提高了局部切空间的降维精度和故障模式识别正确率,其在轴承时域特征维数约简方面,效果优于主成分分析(PCA)与拉普拉斯特征映射(LE),适用于轴承故障诊断。
[Abstract]:Aiming at the disadvantage that the effect of local tangent space permutation algorithm (LTSA) is greatly influenced by the value of nearest neighbor k, a fault diagnosis model of LTSA and K-nearest neighbor classifier based on clustering criterion is proposed.The high dimensional feature matrix is constructed based on the time domain feature of vibration signal, the preprocessing of the high dimensional matrix is carried out, the best nearest neighbor number in local tangent space arrangement is determined according to the clustering criterion, and the low dimensional eigenvector of the high dimensional matrix is extracted by using LTSA.The extracted low dimensional feature vectors are used for fault pattern recognition using K-nearest neighbor classifier.The results show that the optimization method based on clustering criterion can effectively overcome the blindness of nearest neighbor number k selection and improve the dimension reduction accuracy of local tangent space and the correct rate of fault pattern recognition.It is superior to the principal component analysis (PCA) and Laplace feature mapping in reducing the dimension of bearing features in time domain, and is suitable for bearing fault diagnosis.
【作者单位】：北京化工大学机电工程学院;
【基金】：国家自然科学基金(51675035;51375037)
【分类号】：TH133.3

【相似文献】