Fault Detection, Identification and Isolation of South Pars Gas Turbine Using a Combined Method Based on the Data Mining Techniques, k-means, PCA and SVM

In this paper, fault detection, identification and isolation of gas turbines has been investigated. At first, by using k-means algorithm, dimension of primary data is reduced and then with the implementation of principal component analysis (PCA), the knowledge hidden in the data of normal operating conditions of gas turbine, is extracted and faults in the gas turbine have been detected. Then, in the next step, by applying support vector machine (SVM), the detected faults are isolated. Using the combination of data mining techniques and utilizing strong points of these techniques are highlighted points of this paper. Two real systems, GE gas turbine MS6001 and Nuovo Pignone Gas turbine MS5002C, which are located in power generation unit and gas station in second refinery of south Pars are considered. Based on scientific and empirical knowledge, signals are selected and required devices for recording them is implemented on gas turbines by authors. The results of the proposed method are included in the paper.
Keywords