基于小波包散布熵-mRMR特征选取与HHO-KELM的轴承故障诊断方法  

Bearing Fault Diagnosis Method Based onWavelet Packet Dispersion Entropy-mRMR Feature Selection and HHO-KELM

作　　者：宋明瑞 郭佑民[1] 刘运航 郭啸 SONG Mingrui;GUO Youmin;LIU Yunhang;GUO Xiao(Institute of Mechanical and Electrical Technology,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学机电技术研究所,兰州730070

出　　处：《噪声与振动控制》2023年第5期154-160,共7页Noise and Vibration Control

摘　　要：针对3层小波包分解(Wavelet Packet Decomposition,WPD)忽略了1和2层分解信号以及核极限学习机(Kernel Extreme Learning Machine,KELM)参数选择困难的问题,提出一种基于小波包散布熵-mRMR特征选取与HHO-KELM的轴承故障诊断方法。该方法首先对小波包分解中1-3层的14个小波包散布熵(Dispersion Entropy,DE)应用最大相关最小冗余算法(max-relevance and min-redundancy,mRMR)进行特征排序,确定最佳向量维度;然后应用哈里斯鹰优化算法(Harris Hawks Optimization,HHO)实现对KELM参数的优化;最后将最佳维度的小波包散布熵输入到经HHO优化的KELM中进行故障识别。实验结果表明,将mRMR特征选取功能和HHO-KELM聚类功能进行有效结合,可实现故障诊断过程中对分解信号的充分利用,与将只用到第3层分解信号的小波包散布熵输入到KELM的故障分类方法相比,识别准确率提高11.38%。In view of the problem that the 3-layer wavelet packet decomposition(WPD)ignores the decomposition signals of the 1st and 2nd layers and the difficulty of selecting the parameters of the Kernel Extreme Learning Machine(KELM),the bearing fault diagnosis method based on wavelet packet dispersion entropy-mRMR feature selection and HHOKELM is proposed.First of all,the max-relevance and min-redundancy(mRMR)are applied to the 14 wavelet packet dispersion entropies(DE)in wavelet packet decomposition for characteristic arrangement to determine the optimal vector dimension.Then,Harris Hawks Optimization(HHO)is applied to optimize the KELM parameters.Finally,the optimal dimension of the wavelet packet dispersion entropy is input to the HHO optimized KELM for fault identification.Experimental results show that the effective combination of mRMR feature selection function and HHO-KELM clustering function can realize the full use of the decomposition signal in the fault diagnosis process,and the accuracy rate is improved by 11.38%compared with the fault classification method in which only the third layer of wavelet packet dispersion entropy is applied to input to KELM.

关 键 词：故障诊断 滚动轴承 小波包散布熵 最大相关最小冗余 特征选取 核极限学习机 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP206.3[自动化与计算机技术—检测技术与自动化装置]