Online prediction of end-of-batch product quality using phase-specific PLS models

Abstract

Batch processes are commonly operated in multiple phases; phase-based approaches are intuitively well suited for batch process monitoring and quality prediction. In this paper, we therefore propose a phase-based approach to online prediction of end-of-batch quality. Our framework integrates dynamic feature synchronization and dynnamic time warping (DTW) with Dynamic Partial Least Square (DPLS). The Singular Point concept [1] is applied to divide process trajectories into phases with similar dynamics. The challenge of unequal phase lengths is solved by using DTW to equalize the phase trajectories from different runs. A DPLS model [2] is subsequently built between each phase and the final quality variables. Using a well-known simulation, PenSim, we show that this phase-based DPLS (PDPLS) outperforms original DPLS in both monitoring as well as online prediction of end-of-batch quality. © 2006 Elsevier B.V. All rights reserved.