Batch process monitoring based on fuzzy segmentation of multivariate time-series

Abstract

This paper proposes a novel batch process monitoring method called adjoined time series principal component analysis (AdTsPCA). In this method, a modified GG clustering is used for phase identification and data segmentation and multiple time-ordered overlapping PCA models are constructed from the data segments. The PCA models are then used for statistical process monitoring. The key characteristic of AdTsPCA is that additional information contained in the order of PCA models allows for additional diagnosis by the comparison of known process phase and suspected abnormal situation. The proposed AdTsPCA is applied to an industrial penicillin fermentation process to illustrate the effectiveness of the method. AdTsPCA is able to detect faults in the process and significantly reduces the number of false positive errors in the process monitoring.