Research On Condition Monitoring And Fault Prediction Based On The Wind Turbines

With the development of science and technology, great changes have taken place in the electric power industry. Nuclear power, wind power, hydro-power and solar power generation is becoming more and more popular than traditional thermal power. The tendency of wind power is gratifying, but the following problem of stable operation and equipment condition monitoring is tricky. The fault rate of wind turbine is particularly high, because the wind farm is built in the remote areas whose temperature difference is large and wind power is not stable. Especially when the fault occurs in the gear box,generator and other large parts. Due to the difficulty of maintenance, a large number of power is lost. And the distance between the fans is long, it is not easy to repair as well.So it is important to find a method to realize real-time dynamic monitoring and fault prediction.Firstly, a comprehensive investigation of the wind farm is conducted to analysis the faults of the wind turbine. Then the data comes from the SCADA(Supervisory Control and Data Acquisition, SCADA) is used to build the MSET(Multivariate State Estimation Technique, MSET) model and Neural network model.The process memory matrix is based on the normal operation conditions. Through the MSET modeling, the estimated temperature of the gear box bearing is calculated. So does the residual value of the estimated value and observed value. By choosing the appropriate sliding window, the residual value of the bearing temperature is obtained.The mean value of the residuals can directly reflect the change trend of the temperature of the gear box bearing. In this paper, the change trend of the mean value of the residual is monitored in real time. So it can be judged in real time whether the change of temperature of gear box bearing is abnormal. Once the residuals exceed the threshold value, it shows that the change trend is abnormal. And the system will send out warning.Another monitoring method is the time series method of wavelet BP neural network.When the gear box is working correctly, the wavelet BP neural network model can well cover the normal operation of the equipment, and the model’s estimation accuracy is very high. When the gear box works abnormally, its dynamic characteristics change. At this time, the prediction of the temperature can be significantly different as well as the residual. so it is easy to achieve the purpose of early warning by BP neural network.Through the verification and comparison of MSET method and wavelet BP neural network method, it is found that the training process of MSET is simple, the physical meaning is clear, and the accuracy is high. On the other hand, the neural network method is more time-consuming and it is not suitable for on-line prediction, moreover it is easy to fall into local minimum and the early warning accuracy is greatly reduced. In addition,SCADA system alarm mode mostly adopts the limit alarm, but the limit always covers a wide range. It can not predict unless the faults become urgent. It results in the failure to locate the trouble and to track its development trend timely and accurately. The method proposed in this paper based on MSET model and wavelet BP neural network can track the operation state in real time. So the potential failure can be found in time by the method. So does the early warning of failure.