Sparsity discriminant preserving projection for machinery fault diagnosis

Abstract

Extracting features of vibration signals is a key technology in machinery state monitoring and fault diagnosis. Sparsity preserving projection (SPP) is a promising feature extraction algorithm for machinery fault diagnosis, but it is characterized by low efficiency and deficient discrimination information. For this reason, we propose a sparsity discriminant preserving projection (SDPP) algorithm for machinery fault diagnosis. The SDPP is developed on basis of SPP and locality-constrained linear coding (LLC). The low-dimensional features are extracted from the time-frequency representations of original signals by the SDPP. And the least squares support vector machine (LS-SVM) is utilized to recognize the working states of machinery. Three comparative experiments are employed to verify the effectiveness and superiorities of the SDPP. The experimental results show that the SDPP outperforms the other several dimension reduction methods for machinery fault diagnosis. It also demonstrates potential practical applications of the method for recognizing mechanical fault patterns.