Research on Fault Classification of Wind Turbine Based on IMF Kurtosis and PSO-SOM-LVQ

In order to improve the efficiency of the mechanical fault diagnosis in the drive system of the wind power generator, and overcome the problems what exist in traditional vibration diagnosis technology. According to the principle of stator current induction for mechanical fault of drive system, the method of the Empirical Mode Decomposition (EMD) is used for decompose the Intrinsic Mode Function (IMF) components, and six groups of effective IMF component kurtosis are selected as the inputs characteristic quantity of the Self-Organizing Map (SOM) neural network, the data of the initial weights are obtained through optimizing SOM neural network by the Particle Swarm Optimization (PSO), the data is used as the parameter of the weight of the Learning Vector Quantization (LVQ) competition layer, a new competition layer and a linear output layer are constructed to initialize the LVQ neural network, the classification results those has optimal weight is got. The simulation experiments of the planetary gear box and the rolling bearing show that the trained neural network can classify the fault signals well with the IMF component kurtosis as the characteristic quantity, it can be used for the preliminary classification and diagnosis of mechanical faults in the drive system of wind turbines.