EEMD结合概率神经网络的风力机轴承故障诊断研究

Research on Fault Diagnosis for Wind Turbine Based on EEMD and PNN

滚动轴承是风力发电机传动装置中的非常关键的零部件之一.当其发生故障时,采集到的信号大多是非平稳和非线性的,传统的时域和频域分析不能准确分析这些信号的特征.提出使用总体平均经验模式分解(Ensemble Empirical Mode Decomposition,EEMD)进行滚动轴承故障特征提取和使用概率神经网络进行故障特征识别的研究方法.首先使用EEMD算法对实验采集到的不同故障状态的原始信号进行分解,得到每个故障类型的本征模态函数(Intrinsic Mode Function,IMF),利用相关系数法过滤掉不重要的IMF分量.然后算出每个类型故障的IMF分量的能量值和占总能量值的能量比,把能量比当作故障特征向量元素,构造出每个类型的故障特征向量.最后把不同的故障类型和对应的特征向量使用概率神经网络经(Probabilistic Neural Network,PNN)进行识别,得到识别结果,并把结果同使用极限学习机的识别结果进行对比,经验证该方法具有较高的识别正确率.

Rolling bearings are one of the most critical components in wind turbines.Most of the signals collected when the fault occurs are non-stationary and nonlinear.The traditional time-domain and frequency-domain analysis can not accurately analyze the characteristics of these signals.The research methods of fault feature extraction of rolling bearing and identification of fault features using probabilistic neural network are proposed by using Ensemble Empirical Mode Decomposition(EEMD).Firstly,EEMD is used to decompose the different fault signals collected in the experiment,The intrinsic mode function(IMF)of each fault type is obtained,and the unimportant IMF components are filtered out by using the correlation coefficient method,and then the energy values of the IMF components and the total energy values of each type of fault are calculated and use the energy ratio as the fault feature vector element to construct the fault feature vector of each type.Finally,different fault types and corresponding vectors are identified by using probabilistic neural network(PNN)The results are compared with the results of the extreme learning machine.The results show that the method has high recognition accuracy.