Fault feature extraction of rotating machinery based on EWT and a weighted multi neighborhood rough set

In the use of attribute reduction with a neighborhood rough set (NRS), the neighborhood radius was needed to be adjusted for several times iteratively. And it was not determined automatically. In order to solve this inconvenience, a feature selection method based on weighted multi neighborhood rough set (WMNRS) was proposed. Combined with the method of empirical wavelet transform (EWT) in rotating machinery, a fault feature extraction method for rotating machinery was proposed. Firstly, the vibration signal of rotating machinery with nonlinear and strong noise was reconstructed with a group of EWT' optimal modal component selected by correlation, and a high dimensional original feature set was constructed with time domain characteristics of the reconstructed signal. Then, a feature subset was obtained from the original feature dataset by NRSin different neighborhood radius. Last, the probability of occurrence for each feature in the attribute reduction with multiple neighborhood rough sets was counted as feature weight, which was weighted with feature value as sensitive feature set. A characteristic of this method was that it can extract feature automatically in neighborhood rough sets, and the extracted features were more distinguishable. A rotor experiment shows that this method can extract the characteristics of vibration signals effectively, and the fault types of the rotor can be identified correctly according to feature vectors. It provides a theoretical base for solving the classification problem of nonlinear, strong noise, and high-dimensional fault dataset. © 2019, Editorial Office of Journal of Vibration and Shock. All right reserved.