基于EEMD-KECA的风电机组滚动轴承故障诊断

WIND TURBINE ROLLING BEARINGS FAULT DIAGNOSIS BASED ON EEMD-KECA

针对传统频域诊断算法不能充分挖掘出非线性、非平稳信号内部本质信息的问题,提出基于聚合经验模态分解(EEMD)的复合特征提取和基于核熵成分分析(KECA)的故障自动诊断算法。该方法首先采用EEMD将原始信号分解成若干特征模态函数(IMF),计算IMF能量和信号的能量熵构建复合特征向量并作为KECA的输入,之后建立KECA非线性分类器并引入一种新的监测统计量——散度测度统计量,实现故障的实时监测与自动诊断。采用KECA可实现根据熵值大小进行特征分类,具有较强的非线性处理能力,且不同特征信息之间呈现出显著的角度差异,易于分类。最后通过实际风电机组滚动轴承应用实例对算法进行验证,结果表明该方法可有效提取信号中的故障特征,实现对滚动轴承的故障诊断,相比神经网络分类方法具有更高的识别率。

The working condition of wind turbine rolling bearings is always complex, therefore acquired vibration signals are nonlinear, non-stationary. Most traditional algorithms based on the frequency domain cannot fully extract intrinsic information of signals. A new method for fault diagnosis was proposed polymerization based on empirical mode decomposition （EEMD） and kernel entropy component analysis （KECA）. Through the EEMD raw signal is decomposed into several intrinsic mode function （IMF） , calculation of IMF energy and signal energy entropy to construct feature vectors as the input of the KECA table, KECA classifier is built on a fault monitoring and identification. According to the size of the entropy feature extraction using KECA, the maximum extent retained the features of the signal and strong ability of nonlinear processing, which can realize fault classification and recognition more effectively. Finally, the results of experimental analysis showed that the proposed method can effectively extract sensitive features, also demonstrated that the diagnosis accuracy of the proposed model based on EEMD-KECA that is better than that based on neural network and wavelet energy entropy methods.