Gear Fault Detection Based on Best Feature Selection by Particle Swarm Optimization

In this paper, a new method is presented for gear fault detection. The vibrational signals of gearbox set are collected in three conditions: normal, chipped and worn teeth. These signals are adaptively decomposed into a number of intrinsic mode functions (IMFs) by the empirical mode decomposition (EMD). Since, all of the IMFs drived from the EMD are not appropriate for fault detection, the cross-corrolation concept is used to select all most apptoptiate IMFs. Then, feature matrix corresponding to each condition, is extracted using statistical functions. “One-against-one” support vector machine (SVM-OAO) is utilized to classify the faults. Since, all of the extracted features are not suitable for fault detection and SVM has parameters to be set, the particle swarm optimization (PSO) is used to select the best feature and detect optimal parameters of SVM. Objective function in this paper is accuracy of the SVM classifier in predicting of gearbox condition. Obtained results show that the selected features in this method and optimized SVM have the excellent ability to classify the faults.