Research On Methods Of Fault Diagnosis, Prognostics And Condition Assessment For Hydroelectric Generator Units

With developing of the energy structure of China, the scale of wind power and photovoltaic power generator gradually expand, hydroelectric generator units (HGU) undertakes the more and more task of load frequency control, which will result in the more complex working condition for HGU, it causes more likely to occur failure and unplanned outage. In order to solve the scientific issues in engineering applications of fault diagnosis for HGU, the signal process and condition analysis of fault is taken as the entry point in this paper, we extract the time-frequency features and operating mode features, the both side features from the integrated feature which can improve the accuracy of fault features. The main contents and innovative achievements in our paper are as follows:(1) Due to the vibration signal of HGU extant non-stationary, nonlinear and strong background noise, traditional signal process technology don't work well to extract the weak fault signs. Research work will be fuse the vibration signal from different channels, and independent component analysis (ICA) is employed to extract firstly the statistically independent components (ICs) which represent exciting signals of the main vibration sources, it effectively eliminate or reduce the mode mixing. Then, correlation analysis is utilized to eliminate non periodic noise of ICs, and ICs are adaptively decomposed by the empirical mode decomposition (EMD). Finally, the fault signals are produced by reconstructing the intrinsic mode function (IMFs) with the same fault frequency. Test and analysis results indicate that the proposed method is more suits to extract the faulty feature of non-stationary vibration of HGU in strong background noise, which is especially fit for analyzing and processing the weak signal and singular signal.(2) Current fault method mostly based on the time-frequency characteristics of the vibration signal for HGU, and ignored the effect of the working condition on the time-frequency characteristics of the vibration signal, which dramatically reduces the reliability of fault diagnosis. For the sake of those problems, we introduce the statistical process control (SPC) technology into fault diagnosis of HGU, and integrate time-frequency characteristics of the vibration signal with working condition parameters for modeling for each of all condition processes for HGU. Meanwhile, a novel process diagnosing technology based on KICA-PCA is proposed. This method improves the reliability of fault diagnosis for HGU. Test and analysis results indicate that the proposed method has a better performance and efficiency compared with PCA and ICA-PCA method.(3) Nonlinear relationships between vibration variables and condition parameters are revealed through digging into the long-term condition data with mutual information algorithm and fuzzy set theory. The main relevant condition parameters of each vibration variable are extracted by calculating the contribution degree of condition parameters. On this basis, the vibration forecasting model of HGU is built on the main relevant condition parameters. In addition, mathematical statistics and fuzzy system theory are introduced into the state assessment for pressure pulsation of HGU, fuzzy sets about the health condition of pressure pulsation are defined, and the deteriorative function of pressure pulsation is defined by historical data statistics method. On this basis, we proposed a new method of state assessment for pressure pulsation of HGU, which achieves more accuracy of assessment. Test and analysis results indicate that the proposed method's results are consonant with the practice. The proposed method offers an effective approach for state assessment of pressure pulsation of HGU.(4) It is hard to build the nonlinear analytic model of the fault occurrence and propagation, which bring more obstacles for fault prognostics of HGU. Based on former research on fault prediction methods for rotating machinery, the combination of fuzzy inference system and RBF neural network, a novel generalized dynamic fuzzy neural network with constraint (CEGD-FNN) is proposed for fault prognostics of HGU in this paper. In this method, time-frequency characteristics of the vibration signal and condition correlation characteristics are regarded as the input variables, fuzzy rule reasoning and the learning ability of neural network are used for dynamic prediction of the impending fault. Test and analysis results indicate that the proposed method has high precision and good generalization ability. In addition, we research and explore the relationship between failure probabilities and maintain time limit, and propose mapping function between them. The map function bridges the fault prognostics to maintain decision, which is of important meaning to implement the HGU condition maintenance.(5) The current existing fault diagnosis system lack the function to effectively organize and manage data, the expert experience knowledge and fault cases of users, designers, producers and researchers are difficult to integrate and disseminate, those knowledge resources are not underutilized, and do not exploit the their value. Thus, we proposed a kind of remote fault diagnosis and estate assessment system built on service-oriented architecture (SOA). It can integrate experience knowledge and fault cases of the different users in different places by providing an open platform with networking transition, knowledgeable services and shared resources, constantly enrich and complement the experience knowledge database, and furtherly improve performance of the fault system. The remote fault diagnosis and estate assessment system have been successfully launched in Guizhou Wujiang River Dongfeng Station, and has been applied on the hydropower generator units.