改进滑动粗粒化和集成波动色散熵的故障诊断方法  

Fault diagnosis using improved sliding coarsening and integrated fluctuation-based dispersion entropy

作　　者：穆凌霞 田璐 冯楠 汪红鑫 张建 吴世海 刘丁[1,3] MU Lingxia;TIAN Lu;FENG Nan;WANG Hongxin;ZHANG Jian;WU Shihai;LIU Ding(School of Automation and Information Engineering,Xi’an University of Technology,Xi’an 710048,China;School of Intelligence Science and Technology,University of Science and Technology Beijing,Beijing 100083,China;Crystal Growth Equipment and Systems Integration National Engineering Research Center,Xi’an University of Technology,Xi’an 710048,China)

机构地区：[1]西安理工大学自动化与信息工程学院,陕西西安710048 [2]北京科技大学智能科学与技术学院,北京100083 [3]西安理工大学晶体生长设备及系统集成国家地方联合工程研究中心,陕西西安710048

出　　处：《智能系统学报》2025年第2期363-375,共13页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(62373299,62127809);陕西省重点研发计划项目(2024GX-YBXM-093);中国博士后科学基金项目(2022MD723834);陕西省科协青年人才托举计划项目(20210114).

摘　　要：在多尺度波动色散熵中,多尺度粗粒化会忽略重构子序列部分相邻点之间的信息,并且子序列长度随着尺度因子的增大长度减小,这种特征提取的方法不利于故障分类。为了解决这个问题,本文提出了一种n次滑动粗粒化的方法,在确定的比例因子下,利用n次滑动方法保留了每点之间的信息,保证重构后的序列与重构前的序列长度一致。针对波动色散熵中映射技术过于单一的问题,本文利用集成波动色散熵对重构后的序列进行特征提取,使得熵计算更加准确。用西储大学等轴承数据集对算法进行了验证,所提方法的故障诊断精确度显著提高。In multiscale fluctuation-based dispersion entropy,multiscale coarse graining loses the information between adjacent points in the reconstructed subsequences.Additionally,the length decreases as the scale factor increases,and the features extracted through this coarse-grained method are not conducive to fault classification.To address this problem,this paper proposes a method of n steps sliding.This method ensures that the information between points is preserved under the given scale factor,maintaining the length of the reconstructed sequence to be consistent with the original sequence.Aiming at the problem that the mapping technology in the fluctuation dispersion entropy is too simple,integrated dispersion entropy is used to extract features from the reconstructed sequence,enhancing the accuracy of entropy calculations.The algorithm is verified using bearing datasets from Case Western Reserve University and other institutions,the proposed method notably improves fault diagnosis accuracy.

关 键 词：滑动粗粒化 序列重构 故障诊断 故障分类 集成波动色散熵 滚动轴承 振动信号 特征提取 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]