Feature Extraction And Classification Recognition Of Rotor Faults Based On Ensemble Empirical Mode Decomposition

The reliability of machinery and equipment has become more and more prominent as China's industrialization process continues to expand and upgrade with the implementation of "Made in China 2025”.The operating state of the rotor,as the main transmission part of a common rotating machine,is directly related to the efficiency of the device user.The equipment needs to be accurately detected and diagnosed in the early stage of the fault to ensure the safety and reliability of the rotor.An adaptive fault diagnosis method for performing time-frequency analysis and multiple fault pattern identification methods are proposed.The purpose is to study the problem that the early mild fault characteristics of the rotor,which is the main transmission component of the rotating machine,are weak and difficult to diagnose.The experimental data were processed to solve specific diagnostic problems for feature extraction and classification and identification,which is two particularly important part in the rotor fault diagnosis process.Considering the merits of the adaptive decomposition signal of empirical mode decomposition,the empirical empirical mode decomposition method was chosen as the basis of feature extraction methods and classification recognition methods.On this basis,a wavelet-threshold noise reduction method based on weighted idea was proposed to preprocess the signal and reduced the interference factors in the feature extraction process;The kurtosis principle was used to screen the EEMD results to complete the reconstruction of the fault signal to increase the energy proportion of the rotor's early mild fault vibration.When analyzing the effect of the proposed feature extraction method,it was found that the effectiveness of the method did not reach expectations.Effective for all fault types,the feature extraction results of Hilbert-Huang transform analysis method were used as the contrast to illustrate the final effect of the proposed method;Finally,the combined auto-regressive model with mathematical statistical advantages and small sample training were highly accurate.The support vector machine method was used to classify and identify common faults of the rotor system.Through further experiments,five specific common rotor fault patterns are extracted and diagnosed by features.Compared with the Hilbert-Huang analysis method,it is found that the two fault methods generated by different vibration excitation methods cannot fully guarantee the effectiveness.The EEMD feature extraction method based on the kurtosis principle has better performance and physical meaning.The rotor fault classification and recognition method based on the support vector machine also verifies the 100% classification and recognition accuracy of the rotor fault classification method based on the support vector machine.