Fault diagnosis for hydraulic pump based on EEMD-KPCA and LVQ

Hydraulic pump is regarded as the heart of hydraulic system. Achieving the real-time fault diagnosis of hydraulic pump is of great importance for the maintenance of the entire system. An accurate fault clustering solution with self-adaptive signal processing is needed for extracting performance degradation information hidden in the nonlinear and non-stationary signals of hydraulic pumps. Therefore, a fault diagnosis approach based on ensemble empirical mode decomposition (EEMD), kernel principal component analysis (KPCA), and learning vector quantization (LVQ) network is proposed in this study. First, EEMD is employed to acquire more significant intrinsic mode functions (IMFs), thus overcoming the drawback of empirical mode decomposition, and further extracting the energy values of each IMF to form the feature vector. Second, KPCA, a nonlinear dimension reduction method, is used to remove redundancies of the extracted feature vector for high accuracy of fault diagnosis. Finally, LVQ is employed to classify faults based on the reduced feature vector. The efficiency and accuracy of the proposed method is validated by a case study based on the vibration dataset of a plunger pump