Study On Vibration Fault Of Turbine Shafts Based On Non-linear Theory

The thesis devoted to turbo-generator unit condition monitoring and accessing in theory and application. Based on non-linear theory, the thesis explored the representative vibration faults by both theoretical and experimental methods systematically. The paper studied the characteristics of rub-impact phenomenon between rotor and stator in the power plant. Introducing chaotic theory, the paper discussed its feature deeply and developed a valid criterion to identify rub-impact fault between rotor and stator. Especially for fault of rotor filled with oil, which is difficult to be diagnosed, the thesis applied the exclusive approach to abstract the reliable judgements from examples in the field. Moreover, the characteristics of representative non-linear vibration faults such as crack and oil-whirl were researched experimentally in detail. Thus established vibration database can be applied by diagnostic system. Because of vibration faults corresponding to the thermal factors, the present diagnosis study considered non-linear coupling effects caused by various factors. Combining thermal symptoms with vibration spectrum, a new approach is developed for fault diagnosis based on multi-symptom, which overcome the limitation of only applying spectrum characteristics. Also the thesis improved the diagnosis model of fuzzy-neuron networks, established the composite sub-nets based on radial basis function. Thus the ability of the system to classify the faults was enhanced obviously. Besides, the paper put forward the concept contribution factor to describe the different function of different symptoms in different areas. The current fault diagnosis system can also avoid the common defect of symptom redundancy or symptom scarcity. At present, the greatest obstacle to restrict the development of self-learning for diagnosis is lack of applied knowledge. The correlating fault samples can be classified by calculating he measurement coefficient?between two signals. With the traversal technique of composite sub-nets, the thesis put forward a new method to acquire the standard samples of novel faults. By the improved diagnosis model, the system can identify the novel faults exactly. The acquired knowledge samples of novel faults can be considered as the knowledge source to realize self-learning function in diagnostic process and added to diagnosis system. Using the phase space reconstruction method of chaotic theory, the thesis built the non-linear predicting model, which is used to forecast the vibration condition and analyze the condition trend. Without any hypothesis, the model can ensure the studied signal does not distort. It is shown that the current method is better than the traditional means. Moreover, by exploring the fractal feature of vibration series systematically, the paper introduced the correlation dimension as the standard to evaluate the predicted ability of time series. Some examples such as vibration, temperature and vacuum of condenser were taken for forecasting and got well agreement with real signals. The paper is the primary research for predictive maintenance theoretically. The system collected various input signals separately of two sets of turbo- generator units, with the minimum alteration of measuring components in the field. The thesis established the distributed condition monitoring and fault diagnosis system, which was applied in Tianjin First Power Plant. The system can improve the operation and manag...