基于MEWMA-PCA的微小故障检测方法研究及其应用

Research and Application of Small Shifts Detection Method Based on MEWMA-PCA

摘要: 针对化工生成过程中的微小故障检测问题，提出一种新的多变量统计过程监测方法．把传统的单变量指数加权滑动平均（Exponent Weighted Moving Average，EWMA）扩展为多变量EWMA，并与主元分析（Principal Component Analysis，PCA）方法相结合，构成新的多变量（Multivariate EWMA-PCA，MEWMA-PCA）方法．重新构造统计量T_MEWMA-PCA²和Q_MEWMA-PCA，并建立其对应的统计限．详细分析了各个统计量的统计性能指标及其影响因素．Tennessee Eastman（TE）过程的仿真研究说明提出的方法是可行的，并有效地改进了该过程微小故障的检测效果，从而更好地保证了过程运行的安全性、稳定性．

Abstract: For the detection problem of small shifts in chemical industry,a new multivariate statistical process monitoring method is proposed.The method extends the conventional EWMA(Exponent Weighted Moving Average) to MEWMA(Multivariate EWMA),and combines it with PCA to form a new MEWMA-PCA.Then two statistics(T_MEWMA-PCA² and Q_MEWMA-PCA) and their corresponding statistical limits are built.The statistical performance indices and their influential ingredients are detailedly analyzed.A case study of the Tennessee Eastman(TE) process shows that the proposed method is feasible and efficient,and the process monitoring performance is evidently improved,hence enhancing the reliability and stability of the process.