Sparse discriminant manifold projections for bearing fault diagnosis

Abstract

The monitored vibration signal of bearing is usually nonlinear and nonstationary, and may be corrupted by background noise. Thus, it is very difficult to accurately extract sensitive and reliable characteristics information from the vibration signal to diagnose bearing health conditions. This paper proposes a novel bearing fault diagnosis method based on sparse discriminant manifold projections (SDMP). The SDMP was developed based on sparsity preserving projections, and sparse manifold clustering and embedding. The SDMP can effectively extract the meaningful low-dimensional intrinsic features that hidden in a high-dimensional feature dataset. After dimensionality reduction with the SDMP, the least squares support vector machine (LS-SVM) is utilized to classify the different low-dimensional feature data for fault recognition. The effectiveness and superiorities of the proposed method are demonstrated through several comparative experiments with other three manifold learning methods. The experimental results validate that the SDMP is more effective than the other three manifold learning methods for implementation bearing fault diagnosis, and it is more robust when deal with noise interference signal.