Feature extraction and fault diagnosis of gearbox based on ICEEMDAN, MPE, RF and SVM

Feature extraction and fault diagnosis of gearbox based on ICEEMDAN, MPE, RF and SVM

Blog Article

To solve the challenges related to non-stationary vibration signals in gearboxes, i.e.difficult feature extraction, high redundancy of feature vectors and low fault identification rate, this paper proposed a method of feature extraction and fault diagnosis of gearboxes based on the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), multi-scale permutation entropy (MPE), random forest (RF) feature importance ranking and support vector machine (SVM).Firstly, the vibration Dark of the Moon signals of gears in various fault states were decomposed into a series of intrinsic mode functions (IMF) with different frequency distributions by ICEEMDAN.Then, the MPE values of the IMFs were calculated to Wellies obtain the nonlinear dynamic features of non-stationary signals in time-frequency distribution.

Finally, the importance of such features was evaluated by the RF algorithm, and the sensitive features with high importance were selected to form the optimal feature subset as the input to SVM for fault pattern recognition.The experimental results show that this method with strong feature extraction and representation ability and as high as 99.79% recognition rate on average under different operating conditions is more robust in multi-operating conditions and small sample data sets, compared with other methods.