A Novel Fault Identification Methd based on Fast Ensemble Empirical Mode Decomposition and Improved weighted K-Nearest Neighbor Algorithm

Centrifugal pump is one of the most important equipment in modern industry, but it is also very prone to failure. Vibration detection is a very important part of fault identification, but the interference of noise often makes the result not accurate enough. In view of this, a novel fault diagnosis method based on fast ensemble empirical mode decomposition (FEEMD) and improved weighted K-nearest neighbor is presented. Firstly, the multi-scale sample entropy (MSE) is calculated on the basis of FEEMD decomposition, then the MSE, the time domain and frequency domain feature parameters are combined to form the feature parameter matrix. Secondly, the extreme forest feature selection method is used for feature selection, and the weights of feature parameters are calculated by the improved Relief-F algorithm. Finally, K-nearest neighbor algorithm with Gaussian distance weighting is used to classify the samples. Results show that our proposed method can well realize the fault diagnosis comparing with other methods.