基于CMMFDE与多传感器信息融合的旋转机械故障诊断研究  

Rotating machinery fault diagnosis based on CMMFDE and multi-sensor information fusion

作　　者：程志平 王潞红[3] 欧斌 吴军良 CHENG Zhiping;WANG Luhong;OU Bin;WU Junliang(School of Computer and Information Engineering,Jiangxi Normal University,Nanchang 330022,China;School of Artificial Intelligence,Nanchang Jiaotong Institute,Nanchang 330100,China;Department of Mechanical and Electronic Engineering,Changzhi Vocational and Technical College,Changzhi 046000,China)

机构地区：[1]江西师范大学计算机信息工程学院,江西南昌330022 [2]南昌交通学院人工智能学院,江西南昌330100 [3]长治职业技术学院机械电子工程系,山西长治046000

出　　处：《机电工程》2024年第5期807-816,共10页Journal of Mechanical & Electrical Engineering

基　　金：江西省教育厅科学技术研究项目(218414)。

摘　　要：采用单一传感器采集的振动信号难以准确描述旋转机械动态特性,导致提取的故障特征无法准确辨识旋转机械故障。针对这一缺陷,提出了一种基于复合多元多尺度波动散布熵(CMMFDE)、多传感器信息融合和哈里斯鹰算法优化极限学习机(HHO-ELM)的旋转机械故障诊断方法。首先,引入复合多元粗粒化处理,提出了CMMFDE方法,避免了传统单变量分析方法只能处理单一通道振动信号而导致特征的表征性能不足的缺陷,增强了故障特征的表征性能;随后,利用布置在旋转机械不同部位的传感器收集了多种类型的信号,组成混合多通道信号,并进行了CMMFDE分析,构建了故障特征;最后,采用HHO对极限学习机的参数进行了自适应优化,并对特征样本进行了训练和测试,完成了旋转机械的故障识别工作;利用齿轮箱、离心泵两种典型的旋转机械数据集进行了实验分析。研究结果表明:该方法对多个通道的信号进行分析时,所获得的准确率达到了100%和98%,优于对单个通道信号进行分析时获得的准确率,同时CMMFDE方法的准确率和特征提取时间均优于精细复合多元多尺度熵(RCMMSE)、精细复合多元多尺度模糊熵(RCMMFE)、精细复合多元多尺度排列熵(RCMMPE)、多元多尺度波动散布熵(MMFDE)。In view of the defect that the vibration signal collected by a single sensor is difficult to accurately describe the dynamic characteristics of rotating machinery,which lead to the inability to accurately identify the fault characteristics of rotating machinery.A fault diagnosis method for rotating machinery based on composite multivariate multiscale fluctuation dispersion entropy(CMMFDE),multi-sensor information fusion and Harris hawk algorithm optimized extreme learning machine(HHO-ELM)was proposed.Firstly,the composite multivariate coarse-grained processing was introduced and the CMMFDE method was proposed,the defect of traditional univariate analysis methods was avoided,it could only handle single channel vibration signals,resulting in insufficient feature characterization performance,and the representation performance of fault features was enhanced.Then,sensors arranged in different parts of the rotating machinery were used to collect various types of signals to compose mixed multi-channel signals,and CMMFDE analysis was carried out to construct fault characteristics.Finally,HHO was used to optimize the parameters of the extreme learning machine,and the feature samples were trained and tested to complete the fault identification of the rotating machine.Two typical rotating machinery data sets of gear box and centrifugal pump were used for experimental analysis.The research results show that the accuracy obtained when analyzing signals from multiple channels reaches 100%and 98%,which is superior to the accuracy obtained when analyzing single channel signals.At the same time,the recognition accuracy and feature extraction time of the CMMFDE method are superior to the refined composite multivariate multiscale sample entropy(RCMMSE),refined composite multivariate multiscale fuzzy entropy(RCMMFE),refined composite multivariate multiscale permutation entropy(RCMMPE),and multivariate multiscale fluctuation dispersion entropy(MMFDE).

关 键 词：旋转机械 故障诊断 齿轮箱 离心泵 复合多元多尺度波动散布熵 哈里斯鹰优化极限学习机 

分 类 号：TH132.41[机械工程—机械制造及自动化] TH17[自动化与计算机技术—检测技术与自动化装置] TP212[自动化与计算机技术—控制科学与工程]