基于统计模式分析的多变量连续过程故障检测  

Study of fault-detection method in continuous process based on statistical features analysis

作　　者：逄玉俊[1] 李娜[1] 李元[1] 张成[1] 

机构地区：[1]沈阳化工大学信息工程学院,沈阳110142

出　　处：《计算机应用研究》2015年第7期2060-2064,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61034006;61174119;60774070);辽宁省教育厅科学研究项目(L2012139);辽宁省博士启动基金项目(20131089)

摘　　要：针对一些批处理过程中,如同步批轨迹处理和多峰分布等问题,提出了一种基于统计模量(statistics pattern analysis,SPA)分析连续过程的故障诊断方法。FD-SPA和MPCA的显著差别是前者的监测对象是批次变量的统计特征,而后者监控过程变量。MPCA通过分析过程变量的方差—协方差进行故障检测,在SPA中,既要统计过程变量的均值与方差,又要统计过程变量间的协方差结构、偏度、峭度、自相关和互相关性。提出了一种基于滑动窗口的统计模量方法监测非线性的连续过程,使故障检测的准确性与可靠性得到提高。通过在TE过程中与传统的MPCA和KNN方法对比,验证了此方法的有效性。This paper proposed a new method to monitor continuous processes based on the statistics pattern analysis （SPA） framework. It developed the SPA framework to address some challenges associated with batch process monitoring, such as unsynchronized batch trajectories and muhimodal distribution. The major difference between the multiway principal component analysis （MPCA） and SPA was that MPCA monitored process variables while SPA monitored the statistics of process variables. In other words, MPCA examined the variance-covariance of the process variables to perform fault detection while SPA examined the variance- covariance of the process variable statistics （e. g. , mean, variance, autoeorrelation, cross-correlation, etc. ）. It proposed a window-based SPA method to address the challenges associated with nonlinear continuous processes. It could enhance the accuracy in the fault detection. For the application in Tennessee Eastman Continuous process, it also shows better performances than conventional MPCA and KNN.

关 键 词：连续过程 统计模量 多向主元分析 故障检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]