工业过程故障诊断的FDKICA-PCA方法  

Fault Diagnosis of Industrial Process Based on FDKICA-PCA

作　　者：张静[1] 朱菲菲 刘佳兴 王江涛 ZHANG Jing;ZHU Fei-fei;LIU Jia-xing;WANG Jiang-tao(.College of Automation,Harbin University of Science of Technology,Harbin 150080,China;School of Computer Science and Software Engineering,East China Normal University,Shanghai 200062,China)

机构地区：[1]哈尔滨理工大学自动化学院,黑龙江哈尔滨150080 [2]华东师范大学计算机科学与软件工程学院,上海200062

出　　处：《哈尔滨理工大学学报》2018年第6期88-93,共6页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金(51377037)

摘　　要：由于故障诊断中忽略生产过程中自相关与滞后相关的动态特性,核独立成分分析&主成分分析(KICA-PCA)方法缺少可用的变量贡献分析,对微小故障和渐变故障检测效果很差,因此提出基于小波包滤波的动态核独立成分分析&主成分分析(FDKICA-PCA)的故障诊断方法。该方法将小波包滤波理论与AR模型预测数据特性融入到KICA-PCA中,进而提取过程变量自相关、滞后相关的特征信息。文中采用KICA-PCA算法提取过程变量的独立成分与主成分以确定3个监控指标T2、SPE、I2的控制限,利用非线性贡献图进行故障诊断,并通过田纳西过程仿真结果验证了FDKICA-PCA方法的优越性。Because the dynamic characteristics of autocorrelation and lag correlation in production process are neglected in fault diagnosis,Kernel Independent Component Analysis-Principal Component Analysis(KICA-PCA)is very poor in detecting small and gradual faults because of lacking available variable contribution analysis.In this paper,a dynamic kernel independent component analysis(KICA-PCA)fault diagnosis method based on wavelet packet filtering is proposed.This method integrates wavelet packet filtering theory and AR model prediction data characteristics into KICA-PCA to extract the feature information of process variable autocorrelation and lag-related.In this paper,KICA-PCA algorithm is used to extract the independent components and principal components of process variables to determine the control limits of three monitoring indicators T2,SPE,I2.Nonlinear contribution graph is used for fault diagnosis,and the advantage of FDKICA-PCA method is verified by simulation results of Tennessee process.

关 键 词：故障诊断 小波包 主成分分析 核独立成分分析 AR模型 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]