RESEARCH ON FAULT DIAGNOSIS WITH SMALL SAMPLE FOR PLANETARY GEAR SYSTEM WITH SEMI-SUPERVISED LEARNING AND DBN ALGORITHM

Abstract

Objected to fault diagnosis of planetary gearbox, the research and implementation of classification model on small sample with semi-supervised learning and DPN in this paper is carried out. Firstly, the acceleration sample data for four status of the planetary gearbox are obtained, which are including the normal, internal ring gear fault, sun gear fault and Coupling fault between planetary gear and bearing. And the feature vector is built with characteristic parameters such as average amplitude, kurtosis, root mean square, root square amplitude, form factor, crest factor and margin factor. Then the data is delt with CEEMD method for noise reduction and continually the parameters are computed. Then the vector is as input of DBN and Semi-supervised Learning algorithm to fault diagnosis for planetary gear system. Also the comparison competition are done by using DBN and SVM. The results show that under the small sample data, the method of CEEMD - DBN could be more effective under small sample data. The research could provide an effective method for the diagnosis and classification of small sampling projects.