Multiwavelets Based Vibration Feature Extraction And Fault Diagnosis Methods For Hydro-turbine Generating Unit

With the development of national economy, power demand is growing dramatically in China. Hydropower is one of the most important components of energy resources, and its comprehensive development and utilization is becoming more and more important for sustainable development of our country. As the key equipment of hydropower station, hydro-turbine generating unit is becoming larger, more complicated, more integrated, more accuracy, and more automated. At the same time, the safety and stability problems of hydro-turbine generating unit are increasingly prominent. Vibration are very crucial characteristics related with running station of hydro-turbine generating unit, also are main fault types influencing running situation of hydro-turbine generating unit when it become larger than some figure. Therefore, carrying out research on technical of vibration fault diagnosis for hydro-turbine generating unit can guarantee hydro-turbine generating unit and power grid running safely and stably, increase utilization rate of hydro-turbine generating unit and avoid major economy losses and casualties.Researches on feature extraction and fault diagnosis methods are two important aspects in the vibration fault diagnosis process of hydro-turbine generating unit. Researchers have carried out comprehensive exploration and deep research on these areas and achieved a lot of fruits. However, the inherent characteristics of some traditional methods lead to the existence of shortages. Thus, to diagnosis the vibration fault of hydro-turbine generating unit fast and accurately, the existed methods need to be improved to exert their merits and suppress their shortages fully, besides, advanced technologies at home and abroad should be used and new feature extraction and fault diagnosis methods need to be explored and researched to seek more reasonable solution for vibration fault diagnosis of hydro-turbine generating unit.Wavelet analysis, which has overcome the shortcomings of Fourier transform method, can display features of signals in time and frequency domain. Multiwavelets analysis, as the extended form of wavelets analysis, not only preserves the advantages of wavelet, but also can have orthogonality, symmetry, compact support and higher vanishing moment simultaneously. Therefore, multiwavelets analysis has more wide application prospect than wavelet analysis. Because the advantages of wavelet and multiwavelets analysis, this paper has explored vibration feature extraction and fault diagnosis methods of hydro-turbine generating unit based on their theories by improving the existed methods and proposing or introducing new methods to diagnosis faults of hydro-turbine generating unit quickly and accurately and guarantee hydro-turbine generating unit running safely and stability. The specific contents of this paper are as follows:Theories of wavelet and multiwavelets analysis was elaborated and preprocessing methods of multiwavelets analysis was explored to lay a foundation for the research of denoising method and adaptive fault feature extraction method for vibration signals of hydro-turbine generating unit. Procedures of wavelet threshold denoising method and multiwavelets neighboring coefficients denosing method were analyzed detailedly and the denosing method of hydro-turbine generating unit based on multiwavelets neighboring coefficients denosing strategy was proposed. Taking the influences of different basis functions, different threshold functions and different multiwavelets preprocessing methods into consideration, the effects of this method was tested.A feature extraction method based on synthesis detection index for adaptive multiwavelets vibration fault diagnosis of hydro-turbine generating unit was proposed. The maximum value of the synthesis detection index was taken as the optimizing object function, and the optimal multiwavelets were searched from the library of adaptive multiwavelets by genetic algorithm. Then the optimal multiwavelets are used for extracting features from vibration signals. To prove the effectiveness of the proposed method, K-means classifier was used for analyzing the features extracted by the proposed feature extraction method. The results show that, by comparing with GHM multiwavelets and DB4feature extraction methods, the method proposed in this paper can improve the sensitivity of feature parameters and obtain higher fault recognition rate.Combining wavelet and neural network theories, a vibration fault diagnosis method for hydro-turbine generating unit based on ACO-initialized wavelet network was investigated. In this method, ACO algorithm was applied to initialize the parameters. Then, the obtained parameters were trained by gradient descent method. The experimental vibration fault diagnosis of hydro-turbine generating unit results show that the proposed method can overcome the shortcoming of traditional wavelet network in the aspect of sensitive to initial parameters, and has faster convergence speed and better generalization ability.Combining multiwavelets and neural network theories, a novel vibration fault diagnosis method for hydro-turbine generating unit based on radial multiwavenet was investigated. Multiscaling functions were used as the kernel functions of the proposed neural network. And the newly constructed neural network was used for vibration fault diagnosis of hydro-turbine generating unit to make the convergence speed to be faster and the generalization ability to be better, and to provide an effective solution to online vibration fault diagnosis for hydro-turbine generating unit.At last, the paper summed up the main contents systematically and pointed out the research emphasis of vibration fault diagnosis for hydro-turbine generating unit and the research direction in future in the area of wavelet and multiwavelets theories in applications to vibration fault diagnosis of hydro-turbine generating unit.