Threshold PCA-SSD: a novel Steady State Detection algorithm, F.F. Noelle, T.M.Louw, L. Auret, and S.M. Bradshaw

Modern industrial processes undergo frequent process state shifts due to changes in demand, feedstock or maintenance operations. Multimodal continuous process monitoring is therefore required for effective plant operation. Steady state analysis is a key component in effective multimodal monitoring. Various issues still remain in determining the starting and ending times of transition states. High dimensionality of process data is a characteristic that makes it difficult to select variables to consider when conducting steady state analysis. In order to address these problems, steady state detection on the principal component space is proposed. First, data undergoes principle component analysis to reduce dimensionality, then existing multivariate steady state detection algorithms are applied on the principle component scores. A novel algorithm, threshold PCA-SSD, is proposed and compared to an existing technique for steady state detection. The algorithms are investigated on high dimensional synthetic data Keywords: Multimode process, steady state detection, principle component analysis