Research On Fault Diagnosis And Prognosis Method And Its Application For Rotating Machinery

The study on rotating machinery fault diagnosis and prediction techniques is of extremely important significance for assuring the operation safety of mechanical equipment, most rotating machinery are non-stationary signals, and slight faults are likely to be "submerged" by noise, especially in the case of intensive background noise; furthermore, rotating machinery may frequently develop combination faults at different positions in different forms and to different extents, and the interference, interaction and coupling between faults may create challenges for fault diagnosis. Hence, it’s a difficult point in machinery fault diagnosis to identify slight fault in intensive background noise. Time-frequency analysis method is applicable for rotating machinery fault diagnosis since it extracts the local information of time domain and frequency domain of vibration signal synchronously. However, it’s an urgent task to develop new rotating machinery fault diagnosis and other time-frequency analysis methods are defective to varying extents. With respect to above-noted issues, this paper, with rotating machinery as the subject investigated, studies such time-frequency methods as morphology filter, local mean decomposition, multidimensional empirical mode decomposition and noise-aided multidimensional empirical mode decomposition, as well as their application in incipient fault diagnosis for rotating machinery so as to offer new effective means for the identification and prediction of rotating machinery fault diagnosis and fault evolution state. The primary coverage of the paper is as follows:1. A roller bearing fault diagnosis technique based on local mean decomposition and morphology filter is proposed to minimize the effect of noise on signal on local mean decomposition result. Design and set up a railway wagon wheel set roller bearing test system and analyze bearing fault vibration signal; simulation signal and test signal demonstrated the effectiveness of this technique. A genetic algorithm-based adaptive morphological filtering algorithm was proposed against the shortcoming of morphology filter that "manual selection of parameters is needed"; according to the analytical result of simulation signal and test data, adaptive morphological filtering method is significantly effective for signal noise suppression and impact feature extraction.2. An improved MEMD-based incipient rotating machinery fault diagnosis technique was proposed against the shortcoming of EMD, LMD and other traditional methods that "multi-channel vibration signals of rotating machinery can’t be processed" and against the problems in feature extraction for early slight fault and combination fault of rotating machinery. By this method, multi-channel vibration signal is adaptively decomposed into a series of multivariate IMF components, while kurtosis criterion, mutual information and correlation coefficient are employed in the selection of IMF components so as to further eliminate the effect of adverse noise and pseudo components. According to the result of comparison between the proposed method and the EEMD and EMD methods performed based on simulation signal and rotating machinery fault signal, the improved MEMD method is superior and effective in respect of the accuracy and robustness etc. of multi-channel signal decomposition, thereby offering a new concept and means for multi-channel vibration information fusion.3. NAMEMD is a new nonlinear signal adaption time-frequency decomposition method that overcomes such problems as mode mixing normally observed in MEMD and EEMD, but researches show that NAMEMD is not in a position to completely suppress mode mixing frequently observed in MEMD, and the IMF obtained has to be processed due to the existence of mode mixing. The improved NAMEMD method was proposed to suppress mode mixing in decomposition performed by NAMEMD method. Permutation entropy-based randomness test technique is used for timely detection of exceptional signal and noise signal, while residual signals are subjected to NAMEMD decomposition so as to suppress the mode mixing normally observed in NAMEMD method; on this basis, the rotating machinery fault diagnosis method based on improved NAMEMD morphology and Teager energy operator demodulation was proposed against the weak rotating machinery fault signal in very noisy environments and the difficulty in extraction of fault characteristics. According to the result of comparison between the proposed method and the EEMD and NAMEMD methods performed with simulation signal and rotating machinery fault signal, the improved NAMEMD method could avoid mode mixing resulting from white noise-induced time-frequency characteristic difference during integrated averaging of EEMD; as compared with EEMD, the decomposition result exhibits more accurate IMF spectral distribution and more favorable noise reduction, and is more accurate. The proposed method is superior to EEMD and NAMEMD in terms of suppressing mode mixing, enhancing noise reduction and improving decomposition accuracy, and the result demonstrated its effectiveness and superiority.4. The noise-aided multidimensional empirical mode decomposition and permutation entropy-based intelligent bearing fault diagnosis method was proposed based on analysis of the rationale of sample entropy and permutation entropy to address the non-linearity and non-stationary nature of bear vibration signal and the difficulty in acquisition of substantive fault samples. Effective and accurate classification of roller bearing fault type could be achieved by subjecting vibration signal to NAMEMD decomposition, calculating the permutation entropy of the first 5 significant IMF components, and entering the permutation entropy value as eigenvector into the trained support vector machines classifier.5. The improved NAMEMD and permutation entropy-based roller bearing fault evolution state identification method was proposed through the combination of adaptive decomposition of NAMEMD and the nonlinear dynamic parameters signal complexity-based permutation entropy theory so as to overcome the difficulty in the setup of rotating machinery performance deterioration identification indicators. Furthermore, the method proposed overcomes the complexity of roller bearing failure mechanism and the difficulty in characterization of failure features in a very noisy environment, and establishes the connection between roller bearing vibration signal and deterioration state. The findings in simulation test and full life test on roller bearing showed that the performance degradation indicators established could favorably reflect the degradation trend of bearing. This enables the effective identification of roller bearing life cycle state, exhibits strong robustness, and offers a new effective means of deterioration state identification.6. Probing into optimized support vector machine-based roller bearing performance degradation trend prediction method for purpose of roller bearing degradation state trend prediction. Optimize the internal parameters of support vector machine using genetic algorithm, build optimized parameters-based deterioration trend prediction model, accurately predict the performance degradation trend, and evaluate the trend of variation of bearing state within a future period of time so as to improve the operation safety and stability of mechanical equipment. The full life test on bearing demonstrated the accuracy and effectiveness of proposed method, as well as its high prediction accuracy and robustness; this method guarantees the accuracy of roller bearing state prediction, and is of important guiding significance for engineering practice.