排列熵与核极限学习机在滚动轴承故障诊断中的应用

摘要

针对极限学习机(Extreme Learning Machine,ELM)隐含层节点数需要人为设定,致使滚动轴承故障分类模型精度低、稳定性差,提出基于排列熵(Permutation Entropy,PE)与核极限学习机(Kernel Extreme Learning Machine,K-ELM)的滚动轴承故障诊断方法.首先,将测得信号经集合经验模态分解(Ensemble Empirical Mode Decomposition,EEMD)处理后得到一系列IMF本征模态函数,并提取各分量的排列熵PE值组成高维特征向量集;其次,在利用高斯核函数的内积来表达ELM输出函数,从而自适应确定隐含层节点数;最后,将所得高维特征向量集作为K-ELM算法的输入建立核函数极限学习机滚动轴承故障分类模型,进行滚动轴承不同故障状态的分类辨识.实验结果表明:K-ELM滚动轴承故障分类模型比SVM、ELM故障分类模型具有更高的精度、更强的稳定性.%The number of nodes in the hidden layer of the extreme learning machine needs to be artificially set and the fault classification model of the rolling bearing is of low accuracy and poor stability, rolling bear-ing fault diagnosis method based on permutation entropy and nuclear kernel extreme learning machine. First, the measured signal by set of empirical mode decomposition treated by a series of IMF the intrinsic mode functions and extraction of various components of the permutation entropy PE value high dimensional feature vector set. Second, in the inner product by Gauss kernel function to express the ELM output function to a-daptively determine the number of the hidden layer nodes;After that, the high dimension feature vector set is used as the input of the K-ELM algorithm to establish the kernel function limit learning machine rolling bearing fault classification model, and the classification and identification of different fault states of rolling bearings are carried out. The experimental results show that the K-ELM rolling bearing fault classification model is better than ELM, and the SVM fault classification model has higher accuracy and stronger stability.