Centrifugal pump fault diagnosis based on MEEMD-PE Time-frequency information entropy and Random forest

Abstract

In the process of fault diagnosis of centrifugal pump, according to the characteristics of large amount of information, non-stationary and nonlinear of vibration signal, a fault diagnosis method based on Modified Ensemble Empirical Mode Decomposition- Permutation Entropy (MEEMD-PE) time-frequency information entropy and Random forest is proposed in this paper. First, the intrinsic mode functions (IMFs) component from high frequency to low frequency is obtained by MEEMD-PE method, and the IMFs with noise components are determined by the permutation entropy, These IMFs are regarded as pseudo components and removed. The main remaining IMFs, which contain important fault information are retained; Second, the short-time Fourier transform is performed on a series of IMFs. Then the time-frequency matrix containing the fault feature information is obtained. In addition, entropy of time-frequency matrixis also calculated byinformation entropy, which regarded as feature vector. Meanwhile, the feature vector is removed redundant feature information by principal component analysis method. At the same time, wavelet entropy feature extraction method is used to compare MEEMD-PE time-frequency information entropy. Finally, the fault feature matrix after dimensionality reduction is classified by random forest. The experimental results show that the method can effectively diagnose the centrifugal pump.