Identification of Gear Fault Signal Based on Adaptive EMD Feature Extraction and Optimal GA-LSSVM Classification Model

Abstract

The paper proposed a new efficient diagnostic method for the various fault statuses of gear. This method identifies the gear fault exactly by an optimal classification model based on our relevant features. Firstly, the empirical mode decomposition (EMD) method is used to decompose a non-linear and non-stationary vibration signal into intrinsic mode functions (IMFs). The statistical characteristics of the first several IMFs are calculated to form a feature-matrix. Secondly, the meaningful singular values are exploited by the singular value decomposition (SVD) method, which represent as an important feature vector of the original vibration signal, named EMD-SVD. The obtained EMD-SVD feature is the necessary input to serve the classifier model. Finally, a least square support vector machine model (LSSVM) is optimally trained in the EMD-SVD feature, learning to identify the respective fault pattern with the various status of gear. In the formation of the optimal classifier model, the genetic algorithm is used to find out the best parameter pair of the LSSVM classifier model, commonly called GA-LSSVM. We experimented on gear vibration dataset, and the achieved results demonstrated that the proposed diagnostic method was effective in terms of identification accuracy in the short time.