Multiphase and Multimode Monitoring of Batch Processes Based on Density Peak Clustering and Just-in-time Learning

In this paper, a data-driven framework base on density peak clustering (DPC) and just-in-time learning (JITL) is developed to handle with multiphase and multimode monitoring problem of batch processes. To deal with batch-to-batch variations and non-Gaussian distributions of batch data, DPC is firstly used for phase and mode classification and identification. Due to the variety of output trajectories in the same phase and mode, JITL is used to extract similar trajectories as an advanced subdivision strategy to obtain sub-datasets with similar output trajectories. Thus, for each sub-phase in a certain sub-mode, local quality-relevant models are established to achieve an accurate modeling and monitoring scheme. Finally, Bayesian fusion is introduced as the ensemble strategy to determine the final probability of faulty conditions. For performance evaluation, a numerical example and a simulated fed-batch penicillin fermentation process are provided. The monitoring results show the effectiveness of the proposed method.