Research On Fault Diagnosis Method Of Rotating Machinery Based On Vibration Signal Processing

Rotating machinery is widely used in industrial production.The large rotating machinery,such as steam turbine generating unit,has feature of long-span,complex structure,and changeable conditions.It is advantageous to take control measures to avoid the occurrence of serious or vicious accidents via monitoring equipment working state and fault diagnosis timely.Rolling bearings are important support parts for small and medium sized rotating machinery.It has important significance to explore the effective method for diagnosing bearing faults.Aiming at several typical faults diagnosis questions,such as rotor rub-impact,oil whirl,oil whip,bearing fault diagnosis and pattern recognition,the paper has done some research work from the view of vibration signal analysis and processing.The main contents are as follows:The basic theory and cardinal principle of variational mode decomposition(VMD)has been analyzed.VMD has band-pass filter characteristics which is similar to wavelet packet decomposition and good multi-component signal decomposition ability.The influence of the number of components and the penalty factor on the effect of VMD is discussed.Combining with the advantages of fast speed and high resolution of Teager energy operator,A new Teager-VMD time-frequency analysis method is proposed.It determines the component number by judging the correlation between each component and the original signal through mutual information criterion.The analysis results of rotor experiment data show that this method can clearly express the time-frequency characteristics of non-stationary signal,can accurately diagnose the slight rubbing and serious rubbing,can clearly monitor the whole process of the occurrence and development of oil whirl,and can accurately analyze complicated frequency components of oil whip.The comparison results to HHT show that the method is superior to HHT in monitoring and analyzing rotor system faults.The research results show that the rotor oil whip fault frequency has not only the working frequency and oscillation frequency,but also the sum and difference frequency components.This conclusion has important reference value to determine the oil whip accurately.Although VMD use the variational pattern to decompose,it still has endpoint effect.In order to meet the demand of accurate analysis,waveform matching extension method is used to modify VMD.The simulation and experimental data analysis results show that the endpoint effect is improved obviously.Rolling bearing is an important and easy wearing part for rotating machinery.The early bearing fault is weak and influenced by environmental noise,so it is often difficult to diagnose accurately.In order to solve this problem,a method based on minimum entropy deconvolution(MED)and VMD is proposed.MED can highlight the impact of fault component in vibration signal and reduce the influence of noise.Then,the denoised signal by MED is decomposed by VMD to isolate the fault impact component.Bearing weak fault simulation signal and bearing whole life cycle test signal analysis results verify the effectiveness of the proposed method.Comparing to MED,the maximum correlation kurtosis deconvolution(MCKD)algorithm is based on correlated kurtosis criterion.Focusing on continuous fault pulse component,MCKD has better ability to detect and enhance fault impact.An adaptive MCKD method is proposed to diagnose bearing early fault based on variable step search arithmetic after stating MCKD principle.Bearing whole life cycle vibration signal analysis shows that this method can earlier detect bearing fault.The gathered signal of rolling bearing has nonlinear and non-stationary characteristics because of equipment structure and transmission path.Phase space reconstruction can excavate deep information of the nonlinear time series.A bearing fault blind source separation mthod based on phase space reconstruction and stationary subspace analysis(SSA)is proposed.Firstly,the bearing vibration signal is promoted to high dimensional space by phase space reconstruction.Then,SSA is used to separate non-stationary source signals which include the fault impact component.Finally,spectral kurtosis band-pass filter is used to enhance the impact component.The simulation signal and experimental signal analysis results show the method can effectively extract fault impact component of bearing and realize the blind source separation of bearing fault.In order to identify the running state of rolling bearing for rotating machinery,a pattern recognition method based on VMD,improved multi-scale permutation entropy(IMPE)and probabilistic neural network(PNN)is proposed.The fault impact component is firstly isolated from vibration signal by VMD.Then,the different scales characteristics of fault impact component is excavated by IMPE algorithm and is structured to feature vector.The feature vector is input PNN classifier for training and testing to distinguish different damage types and different degree of bearing.