Model Migration Based on Subspace Separation for Development of a New Process Monitoring Model

Abstract

In this article, a model migration strategy based on subspace separation is proposed for process monitoring by taking advantage of common information between an old process and a new process. Firstly, a global basis vector is extracted and deemed to enclose the cross-set similar correlations. Then two different subspaces are separated from each other in the new dataset. The kernel principal component models are developed for the common and specific subspace respectively, and the monitoring is carried out in each subspace. The proposed strategy is illustrated with a simulated fed-batch penicillin fermentation. The results show that the strategy is effective.