A model updating approach of multivariate statistical process monitoring

Multivariate statistical process control based on conventional principal component analysis (PCA) has been used widely in practice. The slow and normal changes in the processes often lead to false alarm since the conventional PCA algorithm is static. In this paper, we proposed a model updating approach of multivariate statistical process monitoring. By the proposed approach, the PCA model which presents the norm operation condition has been remodeled every N samples. Those remodeling data are chosen by quality information and engineer experience. Furthermore, the method of calculating the updating interval has been discussed. Finally, this model updating approach has been evaluated by a mathematic example and CSTR process simulation. The results show the effectiveness of this method.