Research On Information Fusion For Vibration Faults Of Steam Turbine Shaft System

In this paper, the problem of information fusion and quantitative diagnosis of large-scale turbine generators were explored. The fusions of faulty signals were carried from time domain, frequency domain, and time-frequency domain respectively, and some research were applied on actual diagnosis system.At first, the fault simulation rotor test rig of turbine rotor shaft system was designed, and several typical faulty signals during speed rising were collected from this rotor test rig. This established the analysis foundation in this paper. To check the correctness of collected faulty signals, the Fourier ways, 3-D amplitude spectrum and Bode diagram was used to analyze these signals.At the second, the continuous wavelet transform scalogram was explored, and two features, wavelet gray moment and first-order wavelet gray moment vector, are proposed for fault classification of steam turbine rotor. The analysis indicates that first-order wavelet gray moment can reveal the time-frequency features of viberation signals well, and could be used to diagnose faults quantitatively. The effectiveness of the first-order wavelet gray moment vector is also demontrated by experimental data. Result show that the first-order wavelet gray moment vector is suitable to reflect the local information of scalogram, and would be a effective method of vibration signal analysis for fault diagnostics of rotating machinery.At the third, based on the method of information entropy, fusion research on four information entropy: singular spectrum entropy, power spectrum entropy, wavelet energy spectrum entropy and wavelet space state spectrum entropy were carried. Two methods were explored to fuse the four information entropy above, one is the Minimum Distance Classifier (MDC), another is Probability Neural Networks(PNN). Research shows that general MDC has poor classification ability on faults, and the improved MDC has better classification ability than general MDC, which has a good future in real-time application; With the advantages of Bayes classifier and neural networks, Probability Neural Networks have good classification ability of typical vibration faults of turbine, the accuracy of classification is far more than that of improved MDC, so it can be deduced that PNN is a practical fusion diagnosis method for typical fault identification of turbine rotor.The fourth, the fault classification ability by fractal dimension of vibration signal was researched. The fractal dimension of variation signal fuses the time domain information of signals well, which can evaluate irregularities of the time series'fluctuating on the baseline quantitatively, so it can be used as an index for fault identification. On the basis of computation research of correlative fractal dimension, two kinds of correlative dimension were calculated. One kind of correlative dimension was calculated from the time-serials of faulty signal directly, another kind of correlative dimension was calculated from the high frequency reconstruction signal of original time-serials. the analysis shows that correlative dimension can be used as a quantitative index for fault diagnosis, and the correlative dimension calculated from the high frequency reconstruction signal has better identification ability than that from time-serials, and is worthy of theoretical and applicable research.At last, application of the wavelet transform scalogram and the first-order wavelet gray moment were researched and used in an actual diagnosis system. The program realization of wavelet algorithm was studied. The wavelet transform scalogram and the first-order wavelet gray moment were designed as an independent wavelet diagnosis module, and was applied on a provincial long-distance networked steam turbine group monitoring and fault diagnose system. As auxiliary module of fault diagnosis, the wavelet diagnosis module has played on its role in the long-distance diagnosis system.