Fault Diagnosis of Check Valve Based on WVD and NMF

Abstract

This paper proposes a check valve fault diagnosis method based on time-frequency images and Non-negative Matrix Factorization (NMF), which transforms the fault features extraction of time domain signals into fault features extraction of time-frequency images. Firstly, the vibration signals of the check valve are decomposed by Differential Empirical Mode Decomposition (DEMD), and the Intrinsic Mode Functions (IMFs) containing more feature information are selected to reconstruct the signals by correlation coefficient method. Secondly, Wigner-Ville Distribution (WVD) is used to analysis the reconstruct signals and obtain the time-frequency images, then NMF is applied to decompose the time-frequency image matrixes and get the feature matrix. Finally, the feature vectors are classified via the Support Vector Machine (SVM) which is optimized by Genetic Algorithm (GA) to complete the fault diagnosis of the high pressure diaphragm pump check valve. The method is validated using data from three operating states of the high pressure diaphragm pump check valve. The experimental result shows that the proposed method can effectively extract the fault features and identify fault types of the check valve. The average classification accuracy rate is up to 99.17%, which is higher than using the time domain and frequency domain features as input.