Multiphase calibration modeling and quality interpretation by priority sorting

Having realized the multiplicity of underlying characteristics and their different effects on quality interpretation in multiphase processes, how to utilize them efficiently is a challenging problem. In this paper, the concept of priority sorting is proposed and an improved calibration modeling and quality analysis strategy is designed for multiphase batch processes. Over multiple phases, the predictor latent variables (LVs) are automatically sorted by checking their contributions to qualities. A most quality-informative predictor LV plane is then identified by collecting those LVs with higher priority. Priority regression system is built based on this LV plane. The basic consideration is that quality-critical LVs, input as predictors, can reveal a more reliable statistical regression relationship than lower-priority LVs so that they should be given precedence during quality description. The advantage is mainly that the underlying phase information can be explored more efficiently, the quality-critical information can be made better use of and thus improved quality prediction performance can be achieved. The feasibility and performance of the proposed method are illustrated with a typical multiphase batch process.

► A priority sorting based regression modeling strategy is proposed.
► An automatic priority identification algorithm is designed.
► Priority regression system is built based on predictor LVs with higher priority.
► The quality-critical information can be made better use of for quality prediction.
► Improvement is illustrated with a typical multiphase batch process.