| With the energy structure optimization and adjustment of China, the scale of power grid gradually expands, and the proportion of hydroelectric energy increases, while the task of load frequency control for hydroelectric energy has become even more arduous, which has put forward more strict requirements for the security and stability of hydroelectric generator units (HGU). As the key equipment of the hydropower production process, HGU are becoming more and more large-scale and complex, which makes the coupling of different failure factors more obvious, and the mapping relation between faults and symptoms more complex. Therefore, the tradional fault diagnosis method based on the pattern recognition from single type of symtoms could hardly meet the requirments on the reliability of HGU fault diagnosis, and the extraction of multiple symptoms for the comprehensive failure detection of HGU has become urgently needed to guarantee the accuracy and reliability of fault diagnosis. In this paper, in order to solve the key scientific issues in engineering applications of fault diagnosis for HGU, the multi-symptom extraction of HGU faults is taken as the entry point, and the advanced signal analysis and image processing methods are introduce into the feature extraction process. The theory and engineering application of those methods deeply researched, and several modifications to improve the effectiveness of the fault symptoms identification are carried out according to the characteristics of HGU faults. Meanwhile, considering the possible local decision confliction in the diagnosis results of different symptoms, the symptom combination and decision fusion are applied into the information fusion among all the different symptoms. Then, the intelligent fault diagnosis system based on multi-symptoms fusion is established and applied to the fault diagnosis of HGU. The main contents and innovative achievements in this paper are as follows:(1) The mode mixing phenomenon in the Empirical Mode Decomposition (EMD) is deeply researched, and a novel Multi-differential EMD (MDEMD) is proposed, which used the differential operation to enhance the high-frequency component of the signal to improve the mode mixing. Firstly, multi-order differential signals were deduced and decomposed by EMD. Then, their energy distribution characteristics were extracted and utilized to construct the feature matrix for the fault diagnosis. Simulative and practical experiments were implemented respectively, and the results demonstrated that the proposed method is available to improve the mode mixing and promote the diagnosis precision of HGU faults.(2) Shaft orbit is a significant diagnosis criterion, and the main difficulty of shaft orbit identification is how to extract the shape features automatically and effectively. Therefore, a novel method named statistical fuzzy vector chain code (SFVCC) is proposed for the feature extraction of shaft orbit. It focuses on the rotating direction and angle information, and introduces fuzzy vectors to construct the chain code instead of the conventional shape codes. Furthermore, support vector machine (SVM) is utilized to identify various kinds of shaft orbits. Comparative experiments are implemented, the results reveal that, compared with previous methods, the proposed method can identify the shaft orbit more effectively and efficiently with satisfactory accuracy.(3) The defects of traditional feature extraction methods of the relation between vibration and other parameters in the HGU fault diagnosis are analyzed, and the relation curve is introduced to represent the relation characteristics of vibration. Taking the vibration-speed relation as example, the trend of the vibration change with rotating speed is researched. Considering the particularities of vibration-speed curve,1-D division is used in point selection instead of the sampling grid spacing, and fuzzy vectors are introduced to substitute the conventional shape codes to construct the SFVCC. Additionally, SVM is utilized to classify the extracted shape features and complete the automatic identification of vibration-speed curve. Through the simulation experiment the effectiveness of the proposed method is revealed. Furthermore, this method is applied in the research of the vibration analysis of the No.3unit in Ertan hydropower station, and the practicability is demonstrated.(4) Considering the lack of prior experience in HGU fault diagnosis, the non-supervised learning method is researched, and the fuzzy kernel clustering is introduced into the fault diagnosis process. In order to solve the problem of kernel parameter selecting and cluster center calculating, a novel electromagnetism-like artificial bee colony weighted kernel clustering (EAWKC) was proposed. Firstly, taking the influence of different symptoms, the data was weighted, and the clustering model was built based on kernel Xie-Beni clustering index. Then the electromagnetism-like artificial bee colony (ELABC) method was proposed and introduced to solve the objective function realizing the synchronized optimization of the clustering center, symptom weight and the kernel parameter. The classification accuracy of EAWKC was checked by three of the UCI testing data sets and the fault samples of HGU, and compared with the traditional method. The experimental results show that EAWKC has higher accuracy and can complete the fault diagnosis effectively.(5) Taking the possible local decision confliction in the multi-symptoms based fault diagnosis under consideration, a novel mixed decision fusion diagnostic strategy based on projected D-S evidence theory is proposed. First, the symptoms under the same identification frame are combined. Then, the diagnosis results of the symptoms under different frames are projected onto the integrated frame, where the decision fusion is completed. Based on the proposed diagnosis strategy, a fault diagnosis expert system of HGU is established, and applied to the fault diagnosis system of Songjianghe hydropower station, and its practicality and effectiveness can be demonstrated. |
PDF Full Text Request