Bearing Fault Diagnosis Based on Hilbert Marginal Spectrum and Supervised Locally Linear Embedding

A bearing vibration signal is nonlinear and nonstationary, with multiple components and multifractal properties. A bearing fault diagnosis method based on Hilbert marginal spectrum (HMS) and supervised locally linear embedding (SLLE) is proposed for the first time in this paper. HMS is introduced for feature extraction from faulty bearing vibration signals. Then SLLE is proposed for the dimensionality reduction of high-dimensional fault feature, which is more effective than other reducing dimension methods, such as principle component analysis (PCA), multi-dimensional scaling (MDS), and locally linear embedding (LLE). Finally, the support vector machine (SVM) is applied to achieve the bearing fault diagnosis according to the extracted feature vector. The results show that the proposed method improves the fault diagnostic and classification performance significantly.