Research On Methods Of Vibration Signal Processing And Incipient Fault Identification For Large Rotating Machine

Against the background of green and low-carbon modern energy system,the safe and efficient utilization of clean energy is dramatically significant to accelerating energy structure adjustment and promoting ecological civilization construction.As the key apparatus in clean energy conversion,the maximization and complication of hydroelectric generator units and wind turbines have exacerbated the vibration problems and fault risks in the operating process,which raised higher demands for the vibration signal processing and incipient fault identification of system.Therefore,large rotating machines including hydroelectric generator units and wind turbines are taken as the research objects in this thesis.Then,according to refining the key scientific issues in incipient fault diagnosis for large rotating machine,the nonlinear dynamic characteristics and fault mechanism of the system under the coupled excitation of multiple fault sources are analyzed,and the intensive research on incipient fault identification of large rotating machine is carried out based on noise interference suppression and noise assisted analysis respectively.Meanwhile,a novel compound fault separation and feature extraction algorithm is proposed to detect the weaker characteristic signal in the incipiently compound fault of large rotating machine,and a performance assessment model for the key equipment of rotating machine is constructed to present the evolutionary process of initial failure.The main contents and innovative achievements in this thesis are listed as follows:(1)Aiming at the problems of misalignment,looseness and rub-impact in the oil supply system of tubular turbine,the dynamic model for system with coupled faults is established based on considering the time-varying property of nonlinear oil film force caused by the oil impurity.Simultaneously,with the variation of misalignment,oil impurity and radial stiffness of system,the nonlinear dynamic behaviors including period motion,quasi-periodic motion and chaos are investigated by numerical integral method.In addition,by virtue of time domain waveforms,frequency spectrum diagrams,axis orbits,poincare maps and bifurcation diagrams,the nonlinear dynamic responses are analyzed to reveal the fault mechanism of system under the coupled excitation of multiple fault sources.(2)Due to the fact that incipient fault identification of large rotating machine is usually polluted by heavy background noise,the research on slight feature signal detection based on noise interference suppression is conducted in this thesis.On the one hand,the influence of noise intensity on optimal intrinsic mode functions reconstruction effect in traditional empirical mode decomposition denoising methd is analyzed,and a novel denoising algorithm using similarity measure between probability density functions of different intrinsic mode functions and reconstructed signals is proposed to extract incipient fault feature in low SNR condition.On the other hand,in order to solve the problem of large amplitude noise reservation in traditional empirical mode decomposition denoising methd,a probabilistic entropy empirical mode decomposition thresholding algorithm for periodic fault signal enhancement in rotating machine is presented in this thesis.In this method,the entropy threshold is introduced instead of the threshold applied to N sampling points of each intrinsic mode function directly,and a multi-scale thresholding based on quantile statistics for providing probability indexes is contructed to make the amplitudes reducing in a smooth way.Model simulation,experiment and engineering application demonstrate that the proposed methods are effective in the noise reduction and weak feature signal detection in large rotating machine.(3)Bssides carrying on the research about noise reduction to identify initial failure described above,the stochastic resonance method is employed to amplify slight fault feature of large rotating machine by utilizing noise signal.According to the qualitative and quantitative analysis of periodic fault signal enhancement effect under different noise intensity,the occurrence condition of stochastic resonance phenomenon in Duffing oscillator system is derivated,and the variation tendency of fault characteristic spectrum amplitude versus noise intensity is discussed in various conditions.Moreover,for the sake of the energy maximization transfer from noise to fault feature signal,the principle for signal selection based on permutation entropy is constructed and a partly Duffing oscillator system stochastic resonance algorithm is proposed in this thesis.Experiment results demonstrate that the proposed method achieves a remarkable effect on the incipient fault identification of large rotating machine.(4)Since the compound fault features in the initial failure of wind turbine are coupled with each other,it is unfeasible to extract the weaker fault feature via the traditional vibration analysis.Aiming at this problem,an incipiently compound fault diagnosis method based on continuous spectral kurtosis deconvolution is presented.Firstly,the phenomenon of multiple spectral kurtosis maxima in fast kurtogram of compound fault signals is analyzed.Then,the band-pass FIR filter is established for deconvolution to obtain the worse fault characteristic signal,and the band-stop FIR filter is developed for filtering significant resonance spectrum aroused by the worse fault to restrain the influence on extracting weaker feature in compound fault.Finally,the feasibility of the proposed method is validated by the typical model simulation and engineering application.The results indicate that the algorithm not only effectively realizes the separation of compound fault signals,but also successfully extracts the weaker feature in the incipiently compound fault of large rotating machine.(5)Considering that the performance of key equipment has a significant effect on vibration characteristics and stability of large rotating machine,the trend analysis and condition evaluation of equipment degradation based on the theory of logistic regression are carried out from the point of probability.In order to enhance generalization ability and robustness of performance assessment model constructed by massive historical data,the hierarchical K-means clustering algorithm is introduced to discretize the independent variables of logistic regression model.Engineering application exhibits that the established performance assessment model effectively presents the evolutionary process of equipment failure,and also has instruction significance in incipient fault diagnosis of large rotating machine.