Research On Vibration Signal Analysis And Diagnosis Method Of Marine Rotor

As a power plant for large machinery,steam turbines are widely used in high-power machinery and equipment.The rotor is one of the main parts of steam turbines.The high-speed and heavy-duty working environment makes it very easy for malfunction.In this thesis,a large-scale rotating machinery test rig that simulates the actual working conditions of a steam turbine is used to carry out research on vibration signal analysis and diagnosis methods to realize early fault monitoring of steam turbines,so as to effectively avoid machine damage and reduce economic losses.The main research contents of the thesis are as follows:(1)The vibration mechanism and basic vibration characteristics of the rotor are analyzed.The common faults of the turbine rotor and the type of axial trajectory are described in detail.The causes,waveform characteristics and spectral characteristics of various faults are analyzed.(2)Purification of rotor axis trajectory by modern signal processing method is analyzed.Aiming at the characteristics of multi-and non-stationary signals of rotor vibration signals,considering singular value decomposition is a kind of nonlinear filtering.Its denoising has the characteristics of no delay and zero phase shift.The singular value decomposition is used to decompose the rotor vibration signal.The singular value difference spectrum is used to select the characteristic singular value to perform SVD reconstruction,thereby eliminating the random noise in the signal,restoring the true fault information of the rotor and realizing the purification of the axial trajectory.Harmonic wavelet packet transform can infinitely subdivide all frequency bands of the vibration signal,so this method is also applied to the axial trajectory purification.According to the actual working condition of the steam turbine,a large bearing-rotor vibration test rig was built.The rotor displacement signal was used to study the axial trajectory purification.In the two sets of experiments,it was found that the shape of the trajectory of the SVD and the harmonic wavelet packet transformation were typical.The outer 8 characters and the petal shape correspond to the misalignment and oil film oscillation failure.However,compared with the purification result of SVD,the harmonicwavelet packet is more clear and smooth,especially the petal-like axis trajectory.The purification effect of SVD is much better than the harmonic wavelet packet.(3)In the aspect of feature extraction of axial trajectory,the invariant moment and Fourier descriptor are used to extract the feature of the axial trajectory.It is found that the graphical representation ability of the two methods is limited,and the feature distinguishing between different axial trajectories is small.It is impossible to accurately describe the graphical features of the axis trajectory.For this reason,the original invariant moment is improved.The contour of the axis trajectory is extracted by the Sobel operator,the shape geometric feature of the contour and the invariant moment are constructed to form the combined moment,which improves the validity of the graphical feature.(4)In the aspect of automatic identification of axis trajectory,BP neural network,support vector machine and random forest are used to classify the axis trajectory.The five types of measured axial trajectories after SVD purification were used as sample sets.The invariant moments of the axis trajectory,the Fourier descriptors and the combined moments are input as feature vectors to the above three classifier models for classification experiments.Experiments show that the classification of the axis trajectory based on random forest has high accuracy,and the graphical representation ability of the combined moment is stronger than the invariant moment.It shows that the research methods discussed in this thesis can achieve better results.