基于改进VMD和深度置信网络的风机易损部件故障预警

考虑到风电机组运行时监测到的轴承、齿轮等易损部件的振动信号早期故障特征微弱且难以提取,提出了基于变分模态分解的风机易损部件故障特征提取方法,并采用深度置信网络对故障进行预警。为克服变分模态分解参数选取对特征提取效果的影响,基于各分量的相关系数确定分解个数,并采用粒子群算法来优化惩罚因子,将改进的变分模态分解用于振动信号进行分析处理;在此基础上,进一步提取各分量的排列熵和均方根值并将其构成的高维特征向量作为深度置信网络的输入,建立早期故障诊断模型;选取风机传动故障诊断实验平台早期故障数据和某风电机组的现场信号进行故障诊断分析。结果表明,该方法能准确稳定地提取风机易损部件故障信号的微弱特征,并进行故障有效识别,提高了风机易损部件故障预警的准确性。

Fault detection of vulnerable units of wind turbine based on improved VMD and DBN

Considering that the early fault characteristics of the vibration signals of the vulnerable components such as bearings and gears monitored during the operation of wind turbines are weak and difficult to extract, a fault feature extraction method based on VMD was proposed. The deep belief network was used to troubleshoot the faults. In order to overcome the influence of the parameters of the variational mode decomposition on the feature extraction, the number of decompositions was determined based on the correlation coefficients of each component, and the particle swarm optimization algorithm was used to optimize the penalty factor. The improved variational mode decomposition was applied to the vibration signals analysis and processing. Based on this, the permutation entropy and rms value of each modal component were further extracted and the high-dimensional eigenvectors formed by them were used as the input of the deep beilef network to establish an early fault diagnosis model. Finally, fault diagnosis and analysis of wind turbine drive fault diagnosis experimental platform early fault data and an offshore wind turbine site signal were carried out. The results show that the method can extract the weak features of fault signals of fan vulnerable components more accurately and steadily.