Gear incipient fault prognosis using Density-adjustable Spectral Clustering and Transductive SVM

Abstract

A novel method is presented in this paper, which using Density-adjustable Spectral Clustering and Transductive Support Vector Machine, called DSTSVM, to accomplish feature extraction and fault detection. Firstly, the features are extracted via Density-adjustable Spectral clustering, and the Kernel function of TSVM (Transductive Support Vector Machine) is also constructed. Then the TSVM is trained by gradient descent learning and applied in gear failure detection. Gear fault experiments were conducted on an automobile transmission tests platform, and the proposed method were compared with those using TSVM, CKSVM (Cluster Kernel). Experiments results indicated that the proposed approach can reflect the data structure well, and has high classification accuracy with few labeled data.