Abstract
A new monitoring method of batch processes using relationships between dynamic time warping (DTW) and external variables is proposed. DTW has been used as a preprocessing tool for batch process monitoring based on multiway principal component analysis (MPCA) in order to align batch data from various batch lengths by expanding and/or shrinking time axis of data trajectories flexibly. Although MPCA has experienced many successes, the fault detection ability is limited if a fault affects only the progress speed of the process and doesn't change the correlation structure between variables. In this study, another usage of DTW is proposed to overcome this difficulty. If the time warped information of DTW is correlated with external variables of the process such as temperature, initial concentration of materials and so on, changes of input-output relations between external variables and time warped information become good indicator of fault occurrences. The proposed method is applied to a fed-batch penicillin fermentation process simulator to show its effectiveness.
