FAULT DIAGNOSIS BASED ON SEMI-SUPERVISED ORTHOGONAL NEIGHBORHOOD ADAPTIVE LOCALITY PRESERVING PROJECTIONS

Aiming at the problem that accuracy of orthogonal locality preserving projections（OLPP） for fault diagnosis is not high enough,a fault diagnosis method based on semi-supervised neighborhood adaptive orthogonal locality preserving projections（SSNA-OLPP） for dimension reduction is proposed.In this method,fault features that can represent the fault state is firstly constructed based on local characteristic-scale decomposition（LCD） and time-frequency domain feature.And then,the SSNA-OLPP is used to compress the high-dimension feature into low-dimension feature which has better discrimination.Finally,the low-dimension feature are input support vector machine（SVM） to identification fault.SSNA-OLPP can adaptive adjust the neighborhood with the guidance of local cluster coefficient,at the same time,information of some labeled samples are also used to adjust the weight matrix among all samples in the original characteristic space,as a result,better fault diagnosis accuracy can achieved.The experiment results of rolling bearing fault diagnosis verified the effectiveness of the method.