A Semi-supervised Constraints Propagation Based Method for Fault Diagnosis

Abstract

Fault detection and identification could minimize unexpected degradation of system and further avoid dangerous situation. Due to the rapid development of sensor technology as well as the Internet, exponential data could be collected, resulting in that data-driven based fault diagnosis method receives increasing attention. However, most works often learned low dimensional representations so that they couldn't preserve the real local geometric structure of original data. This might degrade fault diagnosis capabilities. In this paper, a novel semi-supervised constraints propagation based approach for fault diagnosis was proposed. The key point was to spread the linking information of supervised data to its neighbors via constraints propagation. Accordingly, the propagated similarity matrix could correctly reflect the structure of the samples. Further, with the aid of propagated matrix, sample indexes were learned via singular value decomposition and support vector machine were utilized to identify the type of faults. The effectiveness of the proposed methods was demonstrated through the experimental results, compared with other popular fault diagnosis methods.