关键词: 故障诊断, 数据降维, Schur分解, 正交邻域保持嵌入算法

Abstract: 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.

Key words: fault diagnosis, data dimension reduction, Schur decomposition, orthogonal neighborhood preserving embedding(ONPE) algorithm