风力机故障诊断神经网络特征参数确定方法

为了诊断风力发电机组的故障,在搭建故障诊断神经网络平台时,选择合适的输入层特征参数搭建小波神经网络以达到网络训练时的稳定收敛.通过对风力发电机组故障诊断神经网络系统输入层特征参数的选择研究.发现风力发电机的齿轮箱、转子、叶片为独具代表性的易故障部件.分别对3个典型故障部件的一般故障类型和其产生机理进行了分析,得出齿轮箱的频率特性可以用来表征其故障类型,不同的转子故障会对应于不同的轴心轨迹,而叶片的故障诊断则可以运用声发射系统.根据分析的结果,提出了输入层特征参数的确定方法.齿轮箱按照其故障的时-频特性来确定输入层特征参数;转子利用其轴心轨迹能够反映故障类型的这一特性,来确定输入层特征参数;而风机的叶片则是通过“声发射系统”测量叶片表面性能时产生的特性数据作为输入层的特征参数.该方法可为风电机组故障诊断神经网络的建立提供参考.

Method for determining neural network characteristic parameters in fault diagnosis system for wind turbines

Neural network has increasingly been used in a fault diagnosis system for wind turbines. The choice of input layer characteristic parameters plays an important role in solving the unstable problem in convergence during network training. First of all, the selection of input layer characteristic parameters of neural network was studied in a fault diagnosis system for wind turbines. It was identified that a wind turbine has three typical components, such as gear box, rotor and blade, in which a fault can occur frequently. Then the fault type and its mechanism were analyzed. It was shown that the frequency characteristics of the gear box can be used to characterize its fault types, a fault in the rotor can be related to an axis orbit;however the fault diagnosis of blade needs an acoustic emission system.Based on these ana-lytical results, several kinds of methods for determining the input layer characteristic parameters were proposed at last. For the gear box, the input layer characteristic parameters can be determined by means of time-frequency characteristics of a fault; the parameters for rotor faults can be reflected by its axis orbit, the parameters for the blade faults can be decided by the characteristic data, which are generated on the blade surfaces and can be detected with an acoustic emission system. This approach can provide a reference for neural network establishment of a fault diagnosis system for wind turbine units.