Research On Fault Diagnosis Method Based On Empirical Mode Decomposition

With the development of modern industry and the improvement of machinery automation, anumber of large rotating machines are used more and more widely, how to ensure the safe andstable operation of the equipment has become an important issue, the fault diagnosis technologyof mechanical equipment has been paid more and more attention, which is the key technology offault signal processing. Fault diagnosis technology usually adopts time-frequency analysismethod, the traditional method has a certain effect in the processing of complex fault signal, butalso has limitations. Signal stability is the premise of traditional signal processing method, signalcharacteristics can be respectively reflected through a time domain or frequency domain, whilelocal and the overall characteristics of the signal in time domain and frequency domain can notbe taken into account, which limits the development of traditional time-frequency analysismethod, because it is not adaptive, and efficient analysis and further processing for complexsignal, in particular nonlinear nonstationary fault signal can’t be realized.Therefore, It is urgentto develop a new fault diagnosis technology, and fault analysis methods based on modern signalprocessing emerge as the times require. It has important significance for monitoring equipmentoperation state, guarantee of safety of machinery operation and prevention of accidents.This paper expounds several time-frequency analysis methods used in the field of faultdiagnosis, deeply studies the empirical mode decomposition method, spreads out discussionaccording to the problems of empirical mode decomposition algorithm, and puts forward thesolutions to these problems, the research contents of this paper are as follows:(1)The commonly used time-frequency analysis methods of machine fault signal are studied,such as short time Fourier transform, Wigner-Ville distribution, wavelet transform. Thecharacteristics and disadvantages of these methods are summarized. On the basis of this,Hilbert-Huang transform (HHT) as well as the essence of the empirical mode decomposition(EMD) have been introduced, and several problems of EMD algorithm which need to be solved,such as the mode mixing, end effect, and envelope fitting are pointed out.(2)A high frequency signal auxiliary method is put forward to inhibit mode mixing of EMDprocess. A known small high frequency sinusoidal signal is added to the original signal beforescreening in order to change the extreme value distribution of the original signal-the signalenvelope. In this way, which the abnormal events are no longer so obvious, and the signalenvelope becomes more natural, which can effectively suppress the mode mixing phenomenonand improve the overall efficiency of EMD. Compared with the traditional EMD method, theimproved method can effectively suppress problem of the mode mixing.(3) Linear extrapolation method is employed to process the signal at the end of extremevalue. Empirical mode decomposition describes upper and lower envelope of the signal throughextreme points, but the maximum and the minimum value of the signal at both ends of the border are hard to estimate, so the envelope has variables, which will produce a boundary error inempirical mode decomposition process. As the decomposition goes, boundary error willpropagate inward, thereby contaminating the internal data and causing decomposition results tobe unreasonable. Through the analysis of several typical methods of suppression of end effect,the linear extrapolation is introduced into EMD to obtain the extreme points of observationinterval boundary. This method is simple and can effectively suppress the end effect.(4)A new Envelop fitting method based on non-uniform rational B-splines curve isproposed. With this method, the node vector in the domain is obtained through chord lengthparameterization, using of backcalculation signal maxima and minima points, the controlpolygon of non-uniform rational B-splines curve is obtained. Then the node vector and controlpolygon are used to construct non-uniform rational B-splines curve which fits signal envelope.Using this method, accurate instantaneous average value can be obtained, thereby suppressingmeaningless signal fluctuations and avoiding problems such as overshoot, undershoot andincomplete envelope.(5)Using the improved EMD method proposed by this paper, the laboratory simulationsignal and the actual fault signals are analyzed respectively. Through the above analysis, it isproved that the improved EMD method is effective.