Research On Fault Diagnosis Method Of Motor Bearing Based On Improved EMD

The large-scale,automation,and precision development of mechanical equipment including rotary electric machines has increasingly become a demand for actual industrial production.The high-speed operation of equipment and the increase in load have increased the incidence of local damage.Rolling bearing is the core component of the power conversion between the shaft and the shaft seat.The local damage type failure is an impact abnormal event that cannot be accurately predicted during operation.The occurrence of these events is accompanied by the occurrence of abnormal components in bearing nonlinear and non-stationary vibration signals.The appearance of anomalous components is an important basis for bearing fault diagnosis and has strong randomness.Traditional signal processing techniques are difficult to analyze accurately.Many factors make it difficult to analyze and process nonstationary signals.In this paper,through a large number of literatures and related researches,and based on the characteristics of bearing fault vibration signals,the Empirical Mode Decomposition(EMD)algorithm has been studied in depth.EMD not only has the self-adaptation of full data self-driven,but also can be very Good handling of non-stationary signals.However,its existing envelope fitting problem seriously affects its decomposition accuracy.Based on motor bearing fault diagnosis problem and the EMD algorithm and based on the vibration data of the bearings of the United States Case Western Reserve University and the measured rotating electrical machinery,the paper establishes a full-flow scheme for fault feature extraction and identification.The specific content is as follows:(1)The non-stationary signal has a strong unpredictability.Therefore,the enveloping overshoot and undershoot of the upper and lower envelope curves in the EMD decomposition will occur.This phenomenon seriously affects its decomposition accuracy and even leads to the failure of feature information extraction.For this phenomenon,based on the rational spline interpolation algorithm,this paper fully considers the changing trend of the two adjacent interpolation base points,and stipulates that the slope of rising or falling between two points,the initial value of the interpolation point,and the interpolation between any two adjacent points are satisfied.The conditions make the envelope curve closer to the actual signal shape and reflect the approximate feature information of the non-stationary signal.Next,the Hilbert marginal spectrum is obtained for the Intrinsic Mode Functions(IMFs)obtained by the decomposition,and the corresponding fault frequency is more accurately extracted therefrom.(2)The effect of fitting the upper and lower envelope curves in the EMD is to obtain the mean envelope curve.The error of the upper and lower envelope curves leads to the error of the mean curve.In this paper,the average value of adjacent extreme points is used to directly fit the mean curve,and the interpolation values between two adjacent interpolation points in the mean envelope interpolation are taken into consideration.The average envelope of theimproved algorithm(IEMD)not only finds more interpolated base points,but also more accurate interpolation estimates between two adjacent interpolated base points.The change of the mean envelope search method avoids the envelope error,so that the mean envelope preserves more vibration information similar to the iterative signal.IEMD has outstanding performance in suppressing endpoint effects,eliminating false modes and modal aliasing.Next,the corresponding fault frequency is more accurately extracted from the Hilbert marginal spectrum of the decomposed IMFs.(3)In order to extract more significant fault features and improve the accuracy of fault diagnosis,based on the essential IMFs of the original data,a local nonlinear embedding(LLE)algorithm is introduced to reduce the feature parameters of the signal to a reduced dimension.A more intuitive fault feature was introduced,and a support vector machine(SVM)algorithm was used to locate the fault.For the high-dimensional feature space constructed by the time domain and frequency domain feature parameters of the IEMD decomposed effective IMFs reconstructed signal,the LLE algorithm is used to reduce the dimension,and the selected low-dimensional sensitive feature parameter matrix is used as the training and test sample of the SVM.The results show that this recognition model guarantees the performance of SVM multi-rater classifier and improves the fault recognition rate.