Research On Adaptive Fault Feature Extraction Of Impulse Signal Based On EEMD And MED

The change in vibration signal is an instinctive response to mechanical systems environment. Meanwhile, the vibration signal carries lots of information about equipment status information, and can reflect the deterioration process of the mechanical system, therefore, it is widely used in the condition monitoring and fault diagnosis technique. Feature extraction is the critical technique for fault diagnosis. In order to extract the fault feature information from non-stationary and nonlinear impulse signal, two adaptive feature extraction methods of ensemble empirical mode decomposition(EEMD) method and minimum entropy deconvolution(MED) are researched deeply. The research contents are as following:Over enveloping problem may appeared in the mean curve construction process of EEMD. Therefore, in this paper, the extreme redundancy or pseudo extreme phenomena is found. The local extreme mean curve structure method is proposed to solved the over enveloping problem. Then, the improved method of EEMD is proposed by replace the original mean curve structure method with the new extreme mean curve structure method. Through comparative analysis of the original EEMD,the effectiveness to eliminate mode mixing of the novel technique are verified.The method to select the sensitive IMFs has become an important issue in fault diagnosis applications of EEMD. In this paper, a new binary interval criterion for selecting the sensitive IMFs is put forward. It also defines the concept of‘subordinate’ and ‘un-subordinate’ in the course of implementing of the proposed criterion. In practical applications, the band interval of maximum value of spectral kurtosis is select as the feature interval, since kurtosis value is very sensitive to the impulse signal. Then, the sensitive IMFs is selected by combining the feature interval and binary interval criterion. Simulation and experimental result demonstrate the effectiveness of the proposed sensitive IMFs selected method.The sensitive IMFs is the convolution calculation results of the pulse signal andthe transfer function. In order to extract the pulse signal form the convolution results,this paper introduces the MED to the EEMD sensitive IMFs. It analyzes the influencing factors in the pulse signal extraction of MED. Then, a new enhancement method of pulse signal extraction is proposed based on EEMD-MED and mathematical morphology. The simulation experiment and case study indicate that the proposed method can enhance the extraction effect of the pulse signal effectively.A large number of high-order harmonic components may emerged when the traditional envelope demodulation method is applied to the MED pulse signal.Therefore, in this paper, a new demodulation ideas is proposed by convolution calculation between the pulse signal and Gaussian kernel first, and then the calculation results is further processing by Fourier transform analysis. In practical applications, the window function used in short-time Fourier transform is Gaussian kernel, and the calculation is consistent with the above demodulation idea. Therefore,a ridge demodulation analysis method based on short-time Fourier transform is proposed for enhanced MED pulse signal. Compared with the EEMD envelope demodulation method and spectral kurtosis envelope demodulation method in rolling bearing fault diagnoses, the validity of the proposed method is verified.