Research On State Trend Prediction And Fault Diagnosis Methods For Turbo-generator Unit

In this paper, methods of state trend prediction and fault diagnosis have been researched for turbo-generator unit. First, the common typical fault principle and feature are analyzed. Second, aiming at the practical signal is easy to be interrupted in acquisition field, a pre-processing method is proposed based on the adaptive structure element of generalized morphology filtering. Third, due to the state characteristic parameters have regularity, the equal dimension dynamic combination model is built to predict the vibration peak-to-peak values and characteristic values which are abstracted by harmonic window decomposition method which is no need for layer decomposition. Finally, due to the shape of rotor center's orbit could express rotor's fault directly, a fault diagnosis method by purifying rotor center's orbit is proposed based on the harmonic window decomposition method which is no need for layer decomposition.Chapter 1 gives a comprehensive description about the research work's background and significance, the current research status of state trend prediction and fault diagnosis skill of turbo-generator unit at home and abroad. Some problems existed in this area are pointed out and the main research content is given.Chapter 2 qualitatively analyzes the common typical faults of turbo-generator unit, studies the fault principle and feature. Then the vibration criterion for rotating machinery used in power plant is introduced, it supplies basis for latter chapters.Chapter 3 puts forward a pre-processing method based on the adaptive structure element of generalized morphology filtering. With this method, there is no need to know the spectrum feature of original signal when de-noising. By using a small structure and a big one in morphology processing, the noise interferences will be eliminated. The detailed principle and construction are given, and then the effectiveness of this method has been proved by simulation and practical data processing.Chapter 4 builds an equal dimension dynamic combination model on the base of state trend prediction task and predictable analysis. Then the predictive model is built for vibration peak-to-peak values of turbo-generator unit. The predictive precision has been verified by simulation and practical results.Chapter 5 proposes a method to extract vibration characteristic values on the base of the harmonic window decomposition method which is no need for layer decomposition. Using this method to extract vibration characteristic values which express operation status, then the equal dimension dynamic combination model is built for these characteristic values. It is useful to master the state changing regularity and find the symptom of forepart fault.Chapter 6 puts forward a fault diagnosis method by purifying rotor center's orbit based on the harmonic window decomposition method which is no need for layer decomposition, whose shape could express rotor's fault directly. By recognizing frequency components of practical signal, the main frequency components are abstracted by harmonic window decomposition method, and then the purified rotor center's orbit is reconstructed. In the end, the fault diagnosis will be done according to rotor center's orbit of the typical fault.The last chapter summaries the main research conclusion and key innovative points, and then further research work is put forward.