Batch Process Monitoring Based on Multisubspace Multiway Principal Component Analysis and Time-Series Bayesian Inference

Abstract

Multiway principal component analysis (MPCA), which is a dimensionality reduction method for process variables, has been widely used to monitor batch and fed-batch processes. However, three main factors affect the performance of MPCA monitoring: The future status of the online batch has to be predicted, the discarded principal components with small variance might contain useful information, and self-correlation and industrial noise exist in process data. Thus, a new batch process monitoring method based on multisubspace multiway principal component analysis and time-series Bayesian inference through a moving window is developed. The feasibility and effectiveness of the proposed batch process monitoring method is demonstrated using a numerical process and the fed-batch penicillin fermentation process, and its performance is compared with that of the MPCA. The results show that the proposed method is more accurate in detecting different types of batch process faults.