Papers by Author: Mei Jun Zhang

Abstract: Aims at the difficult to measure and influence serious in hydraulic impact fault, the intelligent diagnosis of hydraulic impact fault is proposed with improved EEMD and SVM in this paper. Three states of normal and sudden stop and suddenly reversing shock are set up on the hydraulic experimental bench. The improve EEMD is proposed by EEMD noise reduction, SVM extension signal, cubic spline interpolation improvement, and related pseudo components eliminating. The intelligent fault diagnosis in hydraulic system is researched by the improve EEMD to extract the IMF energy as feature vector, and the SVM training classification. Because distinguish and clear between normal state samples with the two impact fault samples, classification results are very right under a linear, polynomial kernel function, or a RBF, sigmoid and precomputed kernel function.

Abstract: Though the Support Vector Regression Machines (SVRM) is considered to be an effective method for time series prediction, its performance is greatly influenced by its parameters. In order to improve the rationality of parameter setting, the influences of the parameters (the number of support vectors N_S and the prediction length N_E) and signal characteristics on the SVRM performance were discussed. The results proved that the existence of confliction between prediction accuracy and prediction efficiency, and SVRM may inappropriate to long-term prediction, and N_S should be greater than a threshold which depends on the signal characteristics for an accurate prediction result. The research results may provide a theoretical basis for the improvments of SVRM algorithm.

Abstract: In order to perform the bearing intelligent fault diagnosis,combined improved EEMD with SVM respectively applied to the binary classification identification of bearing normal and ball fault, normal and inner circle fault,normal and outer ring fault in this paper.Improve EEMD decomposed 9d normalized energy for characteristic vector,the SVM binary classification and recognition of bearings normal and ball fault, normal and inner circle fault, normal and outer ring fault is researched.Compared to the SVM classification accuracy using different kernel functions that is linear kernel function, polynomial kernel function, RBF kernel function and Sigmoid kernel function.In the same parameters,SVM classification accuracy based on linear kernel function and polynomial kernel function is a hundred percent.Bearing normal and ball fault,normal and inner circle fault,normal and outer ring fault is completely correct apart.And there are the classification errors based on RBF kernel function and Sigmoid kernel functions.

Abstract: Intelligent diagnosis is the development direction of mechahnical condition monitoring and fault diagnosis.Conbined improved EEMD with SVM in fault intelligent diagnosis is researched in this paper.To bearing normal and fault as an example,impove EEMD decomposed 9D normalized energy for characteristic vector applied to the binary classification and identification.Compared to the SVM classification accuracy using different kernel functions that is linear,polynomial,RBF and Sigmoid kernel function.In the same parameters,SVM classification accuracy based on linear and polynomial kernel function is a hundred percent.Bearing normal and fault two kinds of state is completely correct apart. And the normal and fault state of the binary classification and identification using RBF and Sigmoid kernel function appeared wtong points.

Abstract: The trend of the measured signal can not only reflect the influence of the external environment, and also reflect the performance of the machine itself mutations. Therefore, removing and extracting tendency item is the necessary link in signal pretreatment.In order to eliminate endpoint effect and modal aliasing phenomenon in EMD and EEMD, based on EEMD,improved EEMD is put forward and the improved EEMD in the application of the signal trend analysis is researched in this paper.In the measured signals to join in a ramp signals,With the improved EEMD decomposition extracted residual items,and the residual items with the original slope signal similarity analysis,the similarity is 0.975.compared to EMD extracted residual items similarity 0.898, EEMD extracted residual items similarity 0.961,the improved EEMD extracted residual item can more accurately reflect the trend of signal.

Abstract: In order to extract effectively detection signals in the noise background for non-stationary signal.On the basis of EEMD, improved EEMD is put forward, the improve EEMD threshold noise reduction is researched in this paper.The simulation signal compared the noise reduction effect of the wavelet,EMD,EEMD,and the improved EEMD. The improved EEMD threshold noise reduction have the best noise reduction result , the highest signal-to-noise ratio, the smallest standard deviation error.After the improved EEMD threshold noise reduction , the measurement signal time domain waveform smooth. More high frequency noise was obviously reduced in Hilbert time- frequency spectrum. Signal-to-noise ratio significantly improve, and signal characteristics are very clear.

Abstract: In order to solve the endpoint effect and modal aliasing phenomenon in EMD and EEMD,Improved EEMD is put forward, and the application in signal singularity detection is researched in this paper. The improved EEMD will signal drops down into a series of different IMF to highlight the different local characteristics of original data, and then calculate Hilbert marginal spectrum and time-frequency spectrum to determine the frequency of these mutations and mutations position. To compared with FT, STFT, WVD,WT, EMD and EEMD etc, No cross-terms and no false IMF components are produced in the Hilbert time-frequency spectrum of the improved EEMD. Different frequency components and frequency mutations position are clearly distinguished at the same time. The Hilbert time-frequency spectrum of the improved EEMD has more superior detection signal singularity ability.

Abstract: In order to correct the endpoint effect and modal aliasing phenomenon in EMD and EEMD, improved EEMD is put forward in this paper. on the basis of the cause of the endpoint effect,the improvement measures are proposed. The pulse interference and noise pollution is suppressed by threshold noise reduction.The overshoot and undershoot phenomenon is controlled by improving 3-spline interpolation to fit envelope. The endpoint effects is lessened by signal SVM prolongation.Compared with EMD and EEMD, not only the IMF false component is reduced, the modal aliasing avoided, and effectively the endpoint effect restrained, the distortion problem in the signal decomposition produces corrected in the results of simulation signal and the measured signal by the improved EEMD in this paper.