Research On Fault Diagnosis Of Turbine Generator Based On Principal Component Analysis And Support Vector Machine

Turbine generator is an important equipment of power production. Due to the complexity of the equipments and the particularity of the operating environments, the failure rate of turbine-generator units is pretty high, meanwhile the consequences are very great. Therefore, fault diagnosis of turbine-generator has been an important aspect of fault diagnosis technology.This paper makes an in-depth analysis and research of the architecture and the diagnostics of vibration fault diagnosis system for turbine. The paper mainly works on the common problems in the process of fault diagnosis, such as: lack of fault samples; parameters for diagnosis involves a wide range and have a complex level; fault signal contains noise; failure The expression of symptoms is not clear, hierarchical classification is very fuzzy, poor correlation , including the following:1. Support Vector Machines (SVM) is a new learning machine based on statistical learning theory, its biggest advantage is that it can fit in small samples of the fault classification decisions. Application of SVM in turbine fault diagnosis, can improve the accuracy of diagnosis and has a good theoretical and practical value.2. Based on the advantages of support vector machines, this paper introduced the cognitive science into the pre-process of information, using the geometric method build the model of cognitive ability, reducing the impact of noise while increasing the separability of the data.3. Principal component analysis was used for feature extraction, in the case of the smallest loss of information, using fewer variables to replace the original high-dimensional variables, eliminating irrelevant information, not only reduces the amount of computation but also improves the accuracy of classification.4. According to the typical vibration faults of turbine generator, a classification model was established on the basis of the best features. Simulation results show that the proposed diagnostic method has a high accuracy and a strong robustness, what's more, it also has a good value for promotion and application .