新しい統計的プロセス管理手法の化学プロセスへの適用

抄録

In order to improve the performance of multivariate statistical process control (MSPC), two advanced methods, known as moving principal component analysis (MPCA) and DISSIM, have been proposed. In MPCA, changes in the direction of each principal component or changes in the subspace spanned by several principal components are monitored. DISSIM is based on the idea that a change of operating condition can be detected by monitoring a distribution of process data, which reflects the corresponding operating condition. The dissimilarity index was introduced for quantitatively evaluating the difference between two data sets. In the present work, the advanced methods, i.e. MPCA and DISSIM, and the conventional method based on principal component analysis were compared with their applications to the Tennessee Eastman process. The monitoring performance of MPCA is strongly affected by selection of principal components and time-window size. In particular, selection of principal components is crucial for effective functioning of MPCA. For a large process, monitoring a subspace spanned by several principal components can achieve better performance than monitoring each principal component. On the other hand, only a time-window size is a design parameter of DISSIM. From this viewpoint, DISSIM seems more practical than MPCA. The application results have clearly shown the superiority of MPCA and DISSIM over the conventional method. The advantage of MPCA and DISSIM comes from the fact that those methods focus on changes in the distribution of process data. Furthermore, a contribution of each process variable to the dessimilarity index is introduced for identifying the variables that contribute significantly to an out-of-control value of the index, and then the effectiveness of the proposed contribution is evaluated.