Research On Non-stationary Signal Time-frequency Analysis And Diagnosis Method Of Rotor Fault

With the development of modern units toward high parameters,large capacity and the application of new technologies such as high and low double-axle arrangement,some unpredictable faults are easy to occur in the operation of units.whether the rotor system can operate efficiently and steadily is related to the safe production and economic benefits of power plants,it is necessary to monitor the working status of the unit and carry out fault diagnosis.from the perspective of non-stationary signals time-frequency analysis and fault diagnosis of rotor vibration fault,this paper conducts in-depth research on the diagnosis and pattern recognition of several typical faults in rotor system operation.The main contents are as follows:Firstly,the Ensemble empirical mode decomposition(EEMD)signal time-frequency analysis method is studied.by observing and analyzing the simulated signal IMF component and HHT analysis,it is verified that the decomposition performance of EEMD algorithm with Gaussian white noise is better than EMD,and it also shows that it is difficult to eradicate the modal aliasing and endpoint effect defects of EMD and EEMD.Secondly,the rotor fault diagnosis method with improved intrinsic time scale decomposition(ITD),correlation coefficient,singular value decomposition(SVD)and least squares support vector machine(LSSVM)is studied.firstly,an improved ITD algorithm is introduced for the problem of ITD waveform distortion.The simulation results show that the improved ITD algorithm can effectively solve the waveform distortion problem.then the correlation coefficient method is used to remove the false components in the improved ITD algorithm decomposition PRC component,and the SVD feature vector values of the effective components with large correlation coefficients are extracted,and then input LSSVM for fault diagnosis.which improve the fault diagnosis efficiency.finally,the experimental signals of rotor unbalance,rotor misalignment,dynamic and static rubbing,and oil film whirl are collected on the simulated rotor test rig.the above methods are used to analyze and diagnose the measured signals.the results show that the method can effectively extract fault features and identify them effectively.Finally,the rotor fault diagnosis method based on variable mode decomposition(VMD),relative entropy(Re),cloud model and optimized LSSVM is studied.firstly,the improved fruit fly optimization algorithm is used to optimize the LSSVM hyperparameters,which improves the machine learning performance of LSSVM.the relative entropy is used to measure the similarity between each component and the original signal,which effectively avoids over-decomposition.the least relative entropy component is selected to extract fault feature using reverse cloud generator,and then input LSSVM to classify and recognize.for the on-line monitoring and fault diagnosis of the steam turbine group,the real-time data processing can be effectively reduced.finally,the method is verified by dynamic and static rubbing and oil film eddy experiment signals,and compared with the EMD-Re cloud model and EEMD-Re cloud model identification results.the results show that the VMD-Re cloud model can accurately diagnose the rotor fault.and it has advantages compared to EMD and EEMD diagnostic methods.