Jixie chuandong (Jan 2017)
Fault Identification based on Improved Supervised Orthogonal Neighborhood Preserving Embedding
Abstract
The Orthogonal neighborhood preserving embedding（ ONPE） is an unsupervised feature dimension reduction method and only use global neighborhood parameter,when it is used to high- dimension fault feature for feature dimension reduction,it is incapacity of using sample class label information and adaptive adjust neighborhood parameter while the space distribution of samples changed. Aiming at the problems above,a fault identification method based on improved supervised orthogonal neighborhood preserving embedding（ IS-ONPE） for feature dimension reduction is proposed. In IS- ONPE,the distance between different points is adjusted by utilizing class label information,thereby a new distance matrix is formed and the neighborhood is constructed through this new distance matrix,at the same time,the neighborhood parameter are adaptive adjusted according to local cluster coefficient. With the low- dimensional feature as inputs of support vector machine（ SVM） for identifying fault types. The experiment results of gear fault identification indicate that the proposed method can identification gear fault in high accuracy,it has some superiority.
