Data Unfolding PCA Modelling and Monitoring of Multiphase Batch Processes

Abstract Multiple phases are inherently characteristics of most batch processes in industrial manufacturing. In order to monitor this kind of batch process more accurately and efficiently, different multivariate statistical methods have been proposed recently. In this work, a new online monitoring approach is proposed with an efficient regularized batch-wise unfolding modeling approach. The approach computes the average correlation structure of the L-length moving window, while capturing the time varying characteristics of the batch process. Based on principal component analysis modeling, the control limits of SPE and T2 statistics are then calculated for each phase PCA model for online monitoring. The approach needs fewer observations in the construction of batch data than traditional batch-wise unfolding monitoring methods. The L consecutive time periods are effectively averaged over a window, which results in a smoother PCA model especially for the transition period between phases. It makes the approach more applicable to online monitoring of the multiphase batch processes. Application examples show the advantages of the proposed method on online monitoring of multiphase batch processes.