Research On Local Mean Decomposition Method And Its Application To Rotating Machinery Fault Diagnosis

The research on rotating machinery fault diagnosis technology has importantpractical significance. Extracting the fault feature is the key in machinery faultdiagnosis. However, the rotating machinery vibration signals usually have thecharacteristics including non-stationary, nonlinearity and low signal to noise ratio, sothat the contained state characteristics information of mechanical equipment will notbe reflected directly. Therefore, it is always a hot topic of the research on how toapply the appropriate signal processing and analysis method to extract the statecharacteristics information from vibration signals. The time-frequency analysismethods are considered to be the most effective means in analyzing the non-stationaryand nonlinear signals at present, but the commonly used time-frequency analysismethods such as Short-time Fourier transform (STFT), Wigner distribution (WD),Wavelet transform (WT) and Hilbert-Huang transform (HHT) have their certainlimitations. Recently, a new adaptive time-frequency analysis method namely Localmean decomposition (LMD) is proposed. It exhibits the advantages by both the theoryand the applications in some fields in comparison to other time-frequency analysismethods. Thus, in view of the problem of extracting fault features in machinery faultdiagnosis, and under the support from National Natural Science Foundation, thisdissertation makes a systematic and exhaustive study on LMD theory and itsapplication to rotating machinery fault diagnosis. The major innovations are asfollows:1. The theory of LMD method is studied. The problems including the lengthselection of moving average, criterion for pure frequency modulation (FM) signal,end effect, mode mixing and computational efficiency of the algorithm are solved.(1) Due to the problem of the length selection of moving average in LMDalgorithm, an automatic method of determining the length of the moving average isproposed. This method considers both the integrity of information after smoothingand the computational efficiency of algorithm, and thereby the smoothing effect isensured.(2) According to the characteristic that PF components have the orthogonality,the orthogonality criterion used as the iterative termination condition of the pure FMsignal is proposed. It is validated that the PFs are determined by the orthogonalitycriterion not only satisfy the orthogonality requirement but also reflect the physical information of the original signal.(3) On the basis of analyzing the cause of end effect in LMD method, aself-adaptive waveform matching extending method is proposed to extend the signal.This method fully considers both the inherent laws of signal and the changing trend ofsignal edges, so it makes the extension more reasonable and has the adaptability. Theanalytical results from simulated signal and experimental signal indicate that thismethod can transfer the distortion caused by end effect to the outside of signal, andthus the end effect of LMD can be restrained effectively.(4) The decomposition results of LMD may generate mode mixing phenomenon.In view of this problem, an improved method based on noise-assisted analysis namelyEnsemble local mean decomposition (ELMD) is proposed. The analytical results fromsimulated signal and experimental signal indicate the ELMD method can improve themode mixing phenomenon of the original LMD method effectively.(5) In order to improve the computational efficiency of LMD algorithm andreduce the calculating time, the LMD method based on rational spline function isproposed by replacing the moving average smoothing with the rational spline functioninterpolation. The analytical results from simulated signal and experimental signalindicate that the computational efficiency of this method is improved significantly incomparison to the original LMD method.2. Application of LMD method to rotating machinery fault diagnosis is studied,and the corresponding fault diagnosis methods are proposed.(1) The instantaneous frequency spectrum method based on LMD is proposed.This method can extract the frequency modulation characteristics from the gearvibration signals, so it can judge the work conditions of gear. The analytical resultsfrom gear experiment signal and actual vibration signal of fan gearbox indicate theinstantaneous frequency spectrum method based on LMD can be applied to gear faultdiagnosis effectively.(2) Seeing that the energy of gear fault vibration signal usually exhibits theperiodic shock characteristics on the time-frequency plane, the local energy spectrummethod based on LMD is proposed. The analytical results from gear experiment signaland actual vibration signal of fan gearbox indicate that this method can reflect thechange of the energy of gear vibration signal with the time and frequency clearly.(3) In view of the disadvantages of the traditional envelope analysis method inthe rolling bearing fault diagnosis, the envelope analysis method based on LMD andspectrum kurtosis (SK) is proposed. This method firstly adopts LMD to separate the frequency components and preliminary denoise of the rolling bearing vibration signal,and then selects the optimal parameters of band-pass filter for extracting the faultimpact components by SK. The analytical results from rolling bearing simulationsignal and experimental signal indicate that this method can extract the fault featureinformation of rolling bearing effectively.(4) The pattern spectrum and pattern spectrum entropy method based on LMD isproposed and applied to rotor system fault diagnosis. This method firstly applies LMDto decompose the original rotor vibration signal, and then calculates the patternspectrum and pattern spectrum entropy for PF components which contain the faultfeatures. The analytical results indicate that the fault features of rotor vibration signalcan be extracted accurately from pattern spectrum, and all types of faults of the rotorsystem can be distinguished perfectly by pattern spectrum entropy.(5) The morphological fractal dimension method based on LMD is proposed andapplied to rolling bearing fault diagnosis. The analytical results from rolling bearingexperimental signal indicate that this method can clearly judge the working conditionsand fault types of rolling bearing.(6) For the fault features extraction of rotating machinery under variable rotatingspeed conditions, the envelope order spectrum method based on LMD is proposed. Byanalyzing rolling bearing and gear experimental signals under variable speedconditions, the effectiveness of the method is validated.