基于张量分解的间歇过程故障诊断方法  

A fault diagnosis method based on tensor factorization for batch processes

作　　者：张晓玲[1] 曹玉苹[2] 邓晓刚[2] ZHANG Xiao-ling;CAO Yu-ping;DENG Xiao-gang(Shengli College,China University of Petroleum,Dongying 257097,China;College of Control Science and Engineering,China University of Petroleum,Qingdao 266580,China)

机构地区：[1]中国石油大学胜利学院,山东东营257097 [2]中国石油大学(华东)控制科学与工程学院,山东青岛266580

出　　处：《高校化学工程学报》2020年第1期190-199,共10页Journal of Chemical Engineering of Chinese Universities

基　　金：山东省自然科学基金(ZR2016FQ21);山东省高等学校科技计划项目(J18KA359);山东省重点研发计划项目(2018GGX101025);中国石油大学胜利学院春晖计划重点项目(KY2017002);中央高校基本科研业务费专项资金(17CX02054)

摘　　要：针对间歇过程的三维数据特点和常出现的渐变故障,提出一种基于张量分解的故障诊断方法:累加和的张量主元分析(summed tensor principal component analysis,STPCA)。该方法先结合累积和控制图(CUSUM)对三维样本数据进行累加处理,累积叠加历史信息,然后利用张量分解思想直接对三维数据进行TPCA分解得到投影矩阵U和V,避免传统方法在展开成二维数据过程中破坏原有数据结构问题,最后构造监测统计量,求取置信限建立故障诊断模型。在盘尼西林发酵仿真实验中,将多向主元分析(MPCA)和基于张量分解的TPCA、STPCA方法比较,得出结论:针对过程的跳变故障,TPCA方法检测故障准确有效,对于渐变故障,基于STPCA的过程监控方法故障检测性能更突出。Considering three-way batch dataset and gradual faults in batch processes,a fault diagnosis method named summed tensor principal component analysis(STPCA)was proposed based on tensor factorization.This method first utilized the cumulative sum(CUSUM)control chart method to sum several preprocessed time series observations at each time point and accumulate historical information.Tensor principal component analysis(TPCA)was then applied to obtain two projection matrices U and V by tensor factorization.This approach did not unfold the three-way data to two-way in order to avoid destructing the original three-way data structure.Monitoring statistics were constructed and their confidence limits were generated to build the fault diagnosis model.In the simulation experiments,fed-batch penicillin fermentation benchmark was used to compare the monitoring performances of multi-way PCA(MPCA),TPCA and STPCA methods.The results show that the TPCA method can precisely and effectively detect the step faults,while the monitoring performance by STPCA is better for gradual faults in batch processes.

关 键 词：间歇过程 张量分解 主元分析 累加和 渐变故障 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]