改进主元分析方法及数据重构在工业系统中的故障诊断研究  

Research on fault diagnosis of industrial process based on improved PCA method and data reconstruction

作　　者：杜海莲[1] 苗诗瑜 杜文霞[1] 吕锋[1] Du Hailian;Miao Shiyu;Du Wenxia;Lv Feng(College of Career Technology,Hebei Normal University,Shijiazhuang 050024,China;School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]河北师范大学职业技术学院,河北石家庄050024 [2]北京交通大学电气工程学院,北京100044

出　　处：《南京理工大学学报》2019年第1期72-77,85,共7页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61673160;60974063;61175059);河北省自然科学基金(F2014205115);河北省教育厅课题(ZD2016053;QN2018087)

摘　　要：为了更加准确地对复杂工业生产系统进行故障判断,使生产系统更加稳定地运行,采用了改进的主元分析(Principal component analysis,PCA)方法及数据重构对工业过程进行故障诊断研究。采集工业系统正常和故障状态时的数据,将传统的PCA算法中平方预测误差(Squared prediction error,SPE)统计量分成两个,分别为主元显著关联的检测残差变量(Principal-component-related variable residual,PVR)和一般变量残差(Common variable residual,CVR)对系统进行故障判断。为了使系统在检测出故障之后尽量减少故障数据对系统的影响,又进一步应用数据重构方法,将故障数据重构成正常数据,并采用有效度指标进行验证。在故障发生的过程中对故障部分进行检修和排除,把生产系统受到故障的影响降到最低。改进的PCA方法和数据重构方法运用田纳西—伊斯曼过程的数据验证,使故障的检测结果更加准确,保证了生产系统的正常运行行。Not only in order to determine the fault more accurately in the industrial system,but also in order to make the production system operation more stable,the improved principal component analysis method and data reconstruction method is used in the industrial process. The data of the normal and fault state of industrial system are collected,the SPE statistics of the traditional principal component analysis is divided into principal-component-related variable residual(PVR)and common variable residua(CVR),which are used to diagnose the system. In order to minimize the impact of the failure data on the system after detecting the failure,the data reconstruction method is further applied . The failure data are reconstituted into normal data,and the validity index is used to verify. When the fault happenes,the fault is repaired and excluded,and the failure impact on the production system is minimized. In order to verify the diagnosis method,the method is applied to the data of the Tennessee-Eastman system,the detection result of the fault is more precise,and the normal production system is ensured to work.

关 键 词：主元分析 故障分析 故障重构 主元显著关联的检测残差变量 一般变量残差 生产安全 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]