چكيده به لاتين
For continuous production of the electric power and avoiding additional costs in power plants, it is crucial to monitor the condition of equipment, especially the generators. Since there is no direct access to the rotating field part in brushless synchronous generators, fast and accurate fault diagnosis of this part is critical. In this thesis, according to the literature regarding fault diagnosis of brushless synchronous generators, a novel approach is proposed for probable fault detection of the brushless synchronous generator including diode open-circuit and short-circuit of the rectifier bridge, turn-to-turn fault and phase open-circuit of the exciter winding. Therefore, first, the parameter identification of the machine parameters is carried out to simulate and model the machine whose implementing is described in the appendices.
Then, numerical and analytical modeling of the machine is studied for different states of the generator, and the simulation is performed. Furthermore, the output terminal voltage of the synchronous generator is analyzed for different states of the machine. For fault detection, four types of signals are suggested in this thesis, including vibration, acoustic, thermal imaging, and terminal voltage which all are non-invasive for measurement. Hence, the proper type of signal processing among time-domain, frequency-domain, or time-frequency-domain is selected for each type of the signal. After the signal processing and extracting the features, feature selection approaches are employed to reduce the redundancy and improve the classification performance. For classifying the selected features, three classifiers are suggested including artificial neural network, k nearest neighbor, and support vector machine whose performances are compared with each other in each section to choose the most appropriate classifier for each signal. After selecting the features and the classifier, the online fault detection of the machine is carried out using the terminal voltage of the machine. Generally, the approach of the fault diagnosis is that in the first step, the state of the machine is diagnosed if it is faulty or healthy. In the next step, in the case of a faulty state, the type of the fault is detected, and also, in the case of diode open-circuit of short-circuit, the location of the defective diode is determined. Eventually, results show the accurate and proper performance of the proposed approach for fault detection of the field system.
