The feature extraction method based on quadratic wavelet packet energy entropy and t-SNE for bearing fault diagnosis

Rolling bearings are widely used in machinery and equipment, how to extract the feature and identify the fault of rolling bearings have become essential issues for ensuring the safe operation of rotation machinery. The fault signals of rolling bearings present nonlinear and non-smooth characteristics which introduce certain challenges to extracting the fault signal. To completely extract the features of signal, this study proposes a novel feature extraction method based on quadratic wavelet packet energy entropy (QWPEE) and t-distributed stochastic neighbor embedding (t-SNE) for bearing fault identification. Firstly, the vibration signals are divided into various node signals by wavelet packet decomposition (WPD). Next, the wavelet packet energy entropy (WPEE) of each node signal in the last layer is extracted to form the initial QWPEE feature vector. After that, the original QWPEE feature data are fused by the t-SNE method to obtain the final feature data set. Finally, the support vector machine (SVM) is employed to identify the states of the bearing fault. The experiments of bearing fault are created to ascertain the performance of the proposed methodology. The experimental outcomes demonstrate that the proposed methodology is efficacious in accurately identifying states of rolling bearing fault.