The Research For Fault Diagnosis Of The Steam Turbine Based On The Simulated Vibration Signal

In the process of the electricity production, steam turbine is an important device, it plays an increasingly important role in the energy production of our country, and the safety and reliability of its operation take more and more attention of us. Because of its complex structure and working principle, in addition of the particularity of working environment, the incidence of its failure is high, steam turbine's fault diagnosis has become an important aspect of the application of the rotating machinery fault diagnosis.Steam turbine system failure is often reflected in the vibration signals by inspecting the vibration signals of the steam turbine, the realization of the diagnosis of mechanical failure is an effective method. Traditional Fourier transform analysis is the basis of the vibration signal detection, it can achieve not only a linear spectrum analysis, but also the key to the power spectrum estimated. Fourier transform is essentially observed in the frequency domain, however, the vibration of mechanical fault signal is often a singular non-stationary signal. Therefore, Fourier transform for the steam turbine unit can do little.The theory of wavelet analysis came into being, providing a high feasibility solution for the amplification of a part of the mathematical functions. In recent years, wavelet technique, developed in the wavelet analysis theory, has been widely used in thermal equipment in regard to fault diagnosis. Considering the research status of the current wavelet transform analysis, selecting the artificial neural network based on the theory developed bionics efficiency of the modeling, optimization methods. By ever-changing the strength of the connection to the processing unit through the pre-fixed number of "neurons (Neuron)" does the "training" for the network and to optimize network, realizing the eventual objective for application.Paper summarizes the common failure vibration characteristics of the steam turbine rotor. Using the time-domain analysis,time-frequency analysis and combined with neural network analysis to the non-stationary vibration signal of high-power gas turbine rotor, do the fault diagnosis. Through the analysis and simulation of the several common fault vibration signal, verify the effectiveness of the algorithm, comparing with the traditional methods such as Fourier Transform, realizes the fault feature's extraction and fault diagnosis. Study shows that, use of wavelet packet's decomposition and reconstruction can approach to fault extraction and analysis of signals, has good ability to the localization in time domain and frequency domain, also can focus on the arbitrary details of the signal, and has a strong ability to identify the signal's mutation. Extract some of the band features which reflect the fault features , not only reduces the dimension of feature vector, but also inhibited the interference of the noise effectively ,so that the complexity of the BP network to reduce, the convergence rate speed up ,accurate results can be get. Through the analysis, we achieve the purposes of fault classification and fault detection.This study shows that, in the fault diagnosis for thermal equipment, using the combination of wavelet packet analysis and BP neural network etc, improves the limitations of traditional methods, enriches and develops the methods of fault diagnosis and researches ways for thermal device, has good theoretical and practical feasibility.Finally, put out the brief discussion about direction of further research.