基于集合经验模态分解样本熵和LIBSVM的离心风机故障诊断方法

A new fault diagnosis method for centrifugal fans based on ensemble empirical mode decomposition sample entropy and LIBSVM

针对离心风机故障信号非平稳、非线性的特征以及应用过程样本少等特点,提出基于集合经验模态分解(EEMD)样本熵和LIBSVM的离心风机故障诊断的新方法。首先采用总体平均经验模态将传感器信号分解成若干个平稳的本征模态函数(IMF)分量;其次通过相关系数准则对IMF分量进行选择去除虚假模式分量,对于筛选出的IMF分量分别计算样本熵并将其作为特征向量;最后将特征向量输入LIBSVM和BP神经网络进行模式识别。实验验证了该方法的可行性、有效性和优越性。

According to characteristics of the centrifugal fan fault signals, such as non-stationary,non-linear and sample scarcity during actual application process, a new method of ensemble empirical mode decomposition sample entropy （SE） and support vector machine （SVM） fan fault diagnosis was proposed. In this method,the axial displacement signal is decomposed into a series of stable intrinsic mode functions （IMF） by the EEMD algorithm firstly. Then,the phony mode components of the IMF are filtered and removed according to the correlation coefficient rule,and the entropy of the filtered IMF is calculate respectively and then be considered as the eigenveetors. Finally, the eigenvectors are put into the LIBSVM and BP neural network to carry out mode identification. At last, the feasibility, effectiveness and superiority of the above method are verified by experiments.