Fault Diagnosis Research On Wind Turbine Based On The Running Data In Field

As the supply of fossil energy becomes tight and environmental degradation, renewable energy is becoming more and more important and attaching great importance to the development of wind power. At present, industry of wind power has also gotten rapid development in the world. Due to bad working conditions, there are frequent failures on wind turbine. With the increase of installed capacity of wind power, the cost of running and maintenance on wind turbine is also increasing. The major parts of large wind turbine such as blade, gearbox, generator located in top of the tower, once one part is faulty, there will need large mechanical equipments such as crane to maintenance, which leads to high cost of maintenance and low output power. According to statistics, the wind turbine of the machine design life for 20 years, its running and maintenance cost can reach 10% to 15% in the total income. At present, maintenance strategies that mostly wind farms are using are “regular maintenance” and “accident maintenance”, but for the abnormal not happened in the period of scheduled maintenance will lead to serious accident. Thus, the system of supervisory control and data acquisition in wind power has gotten rapid developments. The alarm mode of SCADA system is based on the boundaries of a parameter. Usually, the threshold of alarm is broad. Only when the fault deteriorates to a certain degree, the alarm system starts to work. So, the alarm system of SCADA will not locate and track the fault in turbine timely and accurately.In this article, collecting running data of relevant variables from the SCADA system, after processing, using improved BP neural network and SVM to establish the model of wind generator. By observing the trend of residuals between the actual output power and forecasting output power of wind generator to judge whether the running state of the wind turbine is normal. Analysis of SCADA system in wind turbine, monitoring parameters can be divided into two classes: parameters of large range in value and dependent variables, parameters of small range in value and independent variables. Based on this, for parameters of large range in value and dependent variables using MSET(Multivariate State Estimation Technique) method to establish fault warning model, for parameters of small range in value and independent variables, using SCADA system to monitor the running state of wind turbine. In this paper, using real spindle bearing failure of wind turbine in a wind farm to verify the improved BP neural network model. For mechanical parameters of generator, establishing MSET model, by adding appropriate interval deviation on generator front bearing temperature to test the effectiveness of the MSET method. For parameters of small range in value and independent variables, there analyzing the monitoring principle of the SCADA system.When there is a warning on above model, using hierarchical model to select the characteristics variables of subsystems. Making use of degradation degree to mark the characteristics variables. Determine the weight of characteristics variables based on entropy and historical risk. Using fuzzy evaluation method to determine the most likely fault on turbine.