基于数据差异的多变量统计过程控制

Multivariate Statistical Process Control Based on Dissimilarity of Process Data

研究数据差异(DISSIM)方法在多变量统计过程控制中的应用.为检测过程数据分布变化来监视操作条件变化,并且定量地评价两组数据之间的差异,DISSIM方法定义一种差异指标D.计算D时运用时间窗口,时间窗口对指标D具有平滑作用,与MSPC的统计指标T2或Q相比,D变化平稳.针对多变量自回归过程和TE过程,分别应用MSPC和DISSIM两种方法做仿真试验.仿真结果表明:与MSPC相比,适当选择时间窗口时,DISSIM善于检测过程中小的、缓慢的变化,而且能够检测操作条件的变化,明显改善了监视性能.

This paper studies the application of DISSIM method in multivariate statistics process control. In order to detect a change of operating condition by monitoring a distribution of process data and quantitatively evaluate the difference between two data sets, a dissimilarity index is introduced. A time window is used when the index D is computed. The time window has a smoothing effect. Compared with T^2 or Q, the variation of D is smooth. This paper applies MSPC and DISSIM to simulate for multivariate AR process and TE process. Simulations results demonstrate： Compared with MSPC, DISSIM is good at detecting small and slow changes when a time-window size is appropriately selected. In addition, DISSIM can detect a change of operating condition, so it improves the monitoring performance evidently.