Self-tuning final product quality control of batch processes using kernel latent variable model

Abstract

A mid-course correction (MCC) strategy based self-tuning final product quality control of batch processes is presented. The method employs KPLS model developed using batch-wise unfolding data set to capture the relationship between the process variables and final quality. The estimators for the future unknown trajectories are accomplished using statistical latent variable missing data imputation method based on multi-PCA models. Then the optimal control problem is formulated such that the solution is constrained to lie in the kernel latent variable space of the model defined by historical batch data set, and heuristic rule is used for weighting factor to balance the control objective and score magnitude. Finally, SQP is implemented to solve the constraint optimization problem. Application to a simulated cobalt oxalate synthesis process demonstrates that the proposed modeling and quality control strategy can improve process performance.