摘要：AbstractBatch and semi-batch processes are often characterized with strong nonlinearity. Most of the existing kernel-based methods proposed for nonlinear batch process monitoring have not considered the nonlinear characteristic, sequential phase division and uneven problem simultaneously. In this article, a novel similarity index is first defined on the basis of the kernel technique to describe the similarity of nonlinear characteristic in the high feature space. Then, the pseudo time-slice is constructed for each sample by searching samples within a range resembling to each time using the k-nearest neighbor (kNN) rule, which can effectively tackle the uneven problem without trajectory synchronization. A novel automatic sequential phase division procedure is proposed by analysing the nonlinear similarity between the local models derived from the pseudo time-slice and the representative model of each phase. For online application, the affiliation of each new sample is real-time judged to determine the proper phase model and fault status can be distinguished from phase shift event. To illustrate the feasibility and effectiveness, the proposed algorithm is applied to fed-batch penicillin fermentation process.
