Research on compound fault pattern recognition of rotor system based on grid search VMD parameters combined with RCMDE-Relief-F-GRNN

Aiming at the difficulty of feature extraction of rotor vibration signals under strong noise, a fault diagnosis method based on mesh search variational mode decomposition (VMD) parameters combined with fine composite multiscale spread entropy (RCMDE-Relieve-F) is proposed. Based on time-domain energy entropy, kurtosis and Pearson correlation coefficient, a new index grid was formed to search VMD to decompose the optimal K of the original signal, a value was reconstructed, RCMDE in the reconstructed signal was extracted as the characteristic value, and input was filtered into the generalized regression neural network (GRNN) for training and fault pattern recognition using Relief-F dimensionality reduction. The effectiveness of VMD noise reduction parameter selection was verified by numerical simulation and comparison with other decomposition methods. Moreover, the fault simulation experiment results showed that compared with other algorithm models, the mesh search-optimized VMD combined with the RCMDE-Relief-F-GRNN method could effectively filter noise with accuracy of 96 %.