Research On The Dynamic Characteristics And Fault Diagnosis For Gear Drive System Of Wind Turbine

High failure rate is currently one of the main reasons why the average availability of wind turbines is low, which results in the loss of generated energy from wind farms and increases in maintenance cost, makes a rise of wind farm operation cost. Therefore, reducing the failure rate of wind turbines, increasing the generated energy and decreasing the maintenance cost have become important ways of improving the operation economic benefits of wind farms. At present many wind turbines are manufactured with gear drive systems, and the gearbox has become the component with one of the higher failure rate in this kind of wind turbines. Making study of the dynamics characteristics of wind turbine gear drive system, analyzing the characteristics and influence for system fault and researching the fault diagnosis of system are enormously theoretically meaningful for understanding the operation condition of wind turbines, timely detecting and accurate locating of wind turbine faults, future operation condition predicting, maintenance planning in wind farms and increasing the operation reliability of wind turbine.This dissertation takes the gear drive system of wind turbines as research object, and particularly discusses the dynamics characteristics of systems, the dynamics characteristics in the condition of gear failure of system and the strategies for failure diagnosis of system. The main contribution of the dissertation can be summarized and highlighted as follows:1. According to Newton's second law, centralized parameter method is applied to build the torsional vibration model of wind turbine gear drive system and deep analysis of natural characteristics of system is made. Sensitivity analysis method is applied to analyze the influence of system parameters to the natural frequency.2. Based on comprehensive consideration of time-varying meshing stiffness excitation of spur gears and helical gears and error excitation of wind turbine gear drive system, the vibration response of system is calculated by numerical analysis and power spectral is obtained.3. The gear failure models of gear drive system of wind turbines are built. The analysis of dynamic characteristics has been made upon gears with fault, and the study of influences induced by different types of failure in gears to the dynamic characteristics of the whole gear drive system is made, which can be used to provide effective analyzing method and diagnosis basis to the analysis, detection and diagnosis for wind turbine gear drive system.4. A flexible modelling method of wind turbine gear drive system is proposed. It adds virtual rigid body and uses flexible connectors of ADAMS to simulate the time varying meshing stiffness of gears; and discretizes the shafts to simulate the torsional deformation of shafts. The modelling method this dissertation proposes has made the virtual prototype more similar to the physical prototype, with clear physical meaning. The simulation results show that the meshing stiffness of gears and torsional deformation of shafes have large influence on the vibration response characteristics of system.5. A method based on BP neural networks trained by improved particle swarm optimization (PSO) algorithm was proposed in this dissertation for fault diagnosis of wind turbine gearbox. The rule of square sum of error in within-class distance of cluster analysis and the concept of crossover operator in genetic algorithm are introduced to ensure the diversity of particle swarm from initial process to the late search, and the risk of particle swarm's low efficiency and local optimum is reduced. By improving the PSO, the optimization of network weights and bias of BP is realized and the risk of BP neural network's low efficiency and local optimum is reduced. Meanwhile the training efficiency of neural network is improved and convergence speed of the network is faster. Aiming at the vibration signals of wind turbine gearbox being uncertainty, nonstationarity and complexity, the power spectral entropy, wavelet entropy, kurtosis, skewness, correlation dimension and box dimension are extracted as fault features. After the analysis of actual data from wind farms, it is verified that this algorithm has fast convergence speed and the results of judgment is correct.