摘要
为深入分析晶变改性装备中反应釜充液氨工艺过程特性,同时也为了进行更精确、有效的过程监控,提出了一种分段线性多向主元分析(MPCA)方法,实现该间歇过程的建模及故障诊断。首先将三维建模数据降维展开并进行主元分析(PCA),得到采样点数据特征矩阵;然后基于间歇过程的多操作时段特性,根据特征矩阵相似度最大隶属原则,通过设定窗口长度可调的滑动聚类器,在系统运行轨迹上分别从数据变异的方向和幅值对采样点进行2次聚类划分,将间歇过程操作周期分割为一系列运行特性相同的子时段,最后利用各个子时段的线性MPCA模型逼近系统非线性运行过程,实现间歇生产过程建模及在线监测。通过反应釜充液氨工艺过程的实际应用,结果表明提出的建模方法能够对该间歇生产进行合理的、易于从过程特性角度解释的时段划分,所建模型具有可靠的监控性能,能及时准确检测出系统运行过程中的异常情况。
In order to in-depthly analyze the technological characteristic of the liquid ammonia filling process in the modification equipment and achieve more precise,effective process monitoring,this paper presents a piecewise linear multiway principal component analysis（ MPCA） method to realize the modeling and fault diagnosis of batch production process.First,the 3D modeling data are expanded with dimension reduction,principal component analysis is conducted,and the eigen matrix of the sampling point data is obtained.Based on the multiple operation period characteristic,according the maximum feature similarity principle of eigen matrix of PCA,the sliding clustering algorithm with adjustable window length is determined and used to cluster the sample points in the system trajectory from the direction and amplitude of data variability respectively;and the batch process operation period is divided into a series of transition substages and steady sub-stages with same operation characteristic.The piecewise linear MPCA model is used to approach the system nonlinear operation process,and the MPCA modeling and online monitoring of the batch process are achieved.The experiment results of ammonia reactor filling process show that this modeling method can implement the operation period division of batch process,which is reasonable and easy to be explained from the perspective of process characteristics;the proposed model has reliable monitoring performance and can timely and accurately detect the abnormal situations in the process of system operation.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2015年第2期271-279,共9页
Chinese Journal of Scientific Instrument
基金
国家科技支撑计划(2012BAF13B03)
教育部创新团队发展计划(IRT1220)资助项目
关键词
间歇过程
MPCA建模
滑动聚类
晶变改性
过程监测
batch process
MPCA modeling
sliding clustering
lattice distortion modification
process monitoring
作者简介
朱荷蕾,2010年于河南科技大学获得硕士学位，现为东华大学博士研究生，主要研究方向为自动化控制技术、纺织材料晶变改性装备研发及过程控制理论。E—mail：roioly@163．com孟婥（通讯作者），东华大学教授、博士生导师，主要从事机电一体化系统设计、高端纺织装备技术与系统研究。E—mail：mz@dhu．edu．cn
