文献类型：期刊文献

中文题名：基于T-TSNPR的动态过程质量监控

英文题名：Dynamic Process Quality Monitoring Based on T-TSNPR

作者：吕铮[1];杨健[1];侍洪波[1];谭帅[1]

机构：[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室

年份：2019

卷号：45

期号：6

起止页码：946

中文期刊名：华东理工大学学报（自然科学版）

外文期刊名：Journal of East China University of Science and Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家自然科学基金（61703161,61673173）;中央高校基本科研业务费（222201714031）;中国博士后基金（2017M611472）

语种：中文

中文关键词：动态建模;邻域保持回归;质量监控;全投影回归

外文关键词：dynamic modeling;neighborhood preserving regression;quality monitoring;total projection regression

摘要：针对动态过程的质量监控,提出了一种全时间序列邻域保持回归(Total Time Series Neighborhood Preserving Regression,T-TSNPR)算法。首先,考虑到无关变量对构造特征空间的影响,对过程变量进行相关性分析,利用贡献度方法进行变量优化。在数据降维过程中考虑到数据间的时序相关性,T-TSNPR在一定长度的移动时间窗内进行邻域点挑选并构造目标函数,通过全投影回归提取出质量相关特征空间,并建立相应的T2统计量进行质量监控。最后,通过数值仿真和TE过程(Tennessee-Eastman process)仿真实验验证了T-TSNPR算法的有效性。
Quality monitoring has been a new research hot topic in recent years. The alarm should be issued only when the product quality is outside the normal range. However, the traditional neighborhood preserving embedding(NPE) may issue alarm to all faults, which inevitably results in lots of downtime and seriously affect the normal production. Aiming at the above problems, this paper proposes a total-time series neighborhood preserving regression(T-TSNPR) modeling approach. Firstly, by considering the influence of unrelated variables on feature extraction, the correlation analysis between process variables and quality variables is performed, and the contribution method is used to select the key variables. Secondly, in order to achieve the dimensionality reduction in dynamic process, the neighborhoods within certain timeframe are selected to construct the localized constraining relationship between neighborhoods such that the quality-related information can be extracted through total projection regression. Thirdly, Hotelling ’s T^2 statistic is established for online quality monitoring.Finally, a numerical example and the Tennessee-Eastman process are provided to verify the effectiveness of the TTSNPR algorithm.