Abnormal situation detection, three-way data and projection methods; robust data archiving and modeling for industrial applications

Three-way data collected from batch processes and from transitions of continuous processes (start ups, grade to grade transitions, re-starts) are dynamic in nature. The process variables in such processes are both auto correlated and cross correlated. Empirical models developed for the statistical process control of these processes should be capable of capturing the auto and cross correlation of the process variables. Data acquisition and storage should also be performed in a way that preserves these correlations. This paper addresses issues related to acquisition and compression of multivariate data and to modeling of three-way data using projection methods, such as principal component analysis (PCA) and partial least squares (PLS). Other issues such as trajectory alignment, direction of unfolding and modeling data collected from complicated multistage configurations are also discussed.