AUTOMATIC ON-LINE FAULT DETECTIONALGORITHM FOR HYDRAULIC TURBINES

Power plants are one of the most important parts of electric networks. The most important parts in the power plants are turbines, generators and power transformers. These devices are very expensive and their healthy is vitally important. Therefore diagnosis and monitoring systems for preventing catastrophic faults and also for detecting the on-going damages are really necessary and valuable. According to the expense of maintenance of turbines and transformers and sometimes spending a lot of time to inspect them inside, it is necessary to predict the fault in them before happening and without inspecting inside (nonintrusive). With on-line predictive maintenance we would be able to decrease the cost of maintenance and surly improve the performance, and also detecting the incipient faults and finally forecasting the remained life age of these devices. According to my research, the best methods for predicting faults in power transformers are on-line DGA, on-line SFRA and on-line PD method. Also one of the best on-line methods for diagnosis in turbines is vibration analysis. By mounting some acceleration transducers on the bearings of turbines, we can measure the vibration of the system and after that by doing some post processing; we can decide that the turbine’s condition is normal or no. Also sometimes we can discover the origin of the vibration, for example unbalance, misalignment and so on. In this thesis, first of all I describe the main methods that I have used for writing the fault detection algorithm and then I apply the fault detection algorithm on RTR simulation model in MATLAB environment to check the validity of algorithm. After that I apply the written algorithm on a real test-bench called “5assi” that it is a rotor with 3 active magnetic bearings. I measure the vibration signals of 5assi by using some displacement sensors and I check the performance of the algorithm in many cases includes normal condition and unbalance condition with different levels of unbalance. The results show that the algorithm is able to detect the unbalance problem and warn operators perfectly. At the last step I apply the algorithm on a real time monitoring system on a turbine in the power plant “Pont St. Martin” in the province “Aosta” in the north of Italy to check the performance of the algorithm in a real situation. We use eight accelerometers to measure the vibration signals and finally we generate one index for each sensor that shows the condition of the correspondent signal. With trending these indexes during the time and according to some defined alarm and trip levels, we would be able to detect almost any kind of faults in turbines before happening and warn the operators in a proper time. With using this on-line monitoring system, we would be able to save a lot of money and time and surly we can increase the performance of the power plants.