Gearbox fault diagnosis based on adaptive variational modal decomposition

Aiming at the problem that the vibration signals collected in the fault diagnosis of aviation gear⁃ boxes contain complex noise interference and redundant components，a gearbox fault diagnosis method based on adaptive variational modal decomposition（AVMD）is proposed. Firstly，the adaptive selection of the K value in the variational modal decomposition（VMD）is accomplished using the comprehensive evaluation index. By set⁃ ting the thresholds of correlation coefficient and energy entropy，the components that are simultaneously larger than the thresholds are filtered to be reconstructed as the components that contain the main energy and are more similar to the original signal. In this way，noise reduction and feature enhancement of the signal are realized. Sec⁃ ondly，the RCMDE is utilized to extract features from the noise-canceled signal. The nonlinear features reflecting the complexity of the vibration signal at different time scales are fully extracted to form the feature vector. Finally，the extracted features are identified using Kernel Extreme Learning Machine（KELM）optimized by Particle Swarm Algorithm（PSO）. The model is experimentally validated to have an average accuracy of 95.04% over ten tests. And compared with other feature extraction and pattern recognition methods，the proposed method has high⁃ er diagnostic accuracy. It provides a new method for the fault diagnosis of aviation gearboxes. © 2024 Journal of Propulsion Technology. All rights reserved.