基于Schur分解和正交邻域保持嵌入算法的故障数据集降维方法

Fault Data Set Dimension Reduction Method Based on Schur Decomposition and ONPE Algorithm

针对转子故障特征数据集降维问题,提出一种基于Schur分解和正交邻域保持嵌入算法的故障数据集降维方法——Schur-ONPE降维方法。该方法首先应用小波包分解提取不同频带内的能量以组成故障特征值集合,然后运用Schur分解和ONPE算法将高维特征集向低维投影,使降维后类内散度最小化及类间分离度最大化,最后将降维后得到的低维特征集输入K近邻分类器进行模式识别。通过双跨转子试验台的故障特征数据集进行验证,结果表明该方法能够有效地解决转子故障特征集的降维问题。

Aiming at dimension reduction of fault data set,a novel method in dimension reduction was proposed based on the combination of Schur decomposition and ONPE algorithm.Firstly wavelet packet decomposition was used to extract the fault signals of different frequency band energy features,then Schur decomposition and ONPE algorithm were used to project the high-dimensional data sets to lower dimensions.After the transformation,the considered pairwise samples within the same class were as close as possible,while those between classes were as far as possible.Finally,the lower dimension was collected and the K nearest neighbor classifier was input to recognize the different patterns.The fault characteristic data sets from a double span rotor test-rig were used to validate the proposed algorithm performances.The results show that this method may solve the problems of reducing the dimension of rotor fault features sets effectively.