Application of back propagation neural network to fault diagnosis of direct-drive wind turbine

The vibration signals of wind turbines are highly nonlinear and non-stationary due to wind turbine operation conditions that are very complicated. The signals will be more complex when a fault occurs. Aiming at these problems, a fault diagnosis method for direct-drive wind turbine is presented based on back propagation neural network (BPNN). The time-domain feature parameters of vibration signals in the horizontal and vertical direction are considered in the method. Five experiments of direct-drive wind turbine with normal, wind wheel mass imbalance, wind wheel aerodynamic imbalance, yaw and blade break are carried out in laboratory scale. Through analyzing the features of five conditions, the time-domain feature parameters in horizontal and vertical direction of the vibration signal are selected as the input samples of BPNN. By training, the BPNN model can be constructed between feature parameters and fault types. The validity of the BPNN model is verified using test samples. The results indicate that the proposed method has higher diagnostic accuracy. It can used in on-line fault diagnosis of direct-drive wind turbines.