Skip to Main Content
Making on-specification products is a primary goal, and also a challenge in chemical batch process operation. Due to the uncertainty of raw materials and instability of operating conditions, it may not produce the desired on-spec final product. It would be helpful if one can predict the product quality during each operation, so that one can make adjustments to process conditions in order to make on-spec product. This paper addresses the issue of real-time prediction of final product quality during a batch operation. First, a data-driven modeling approach is presented. This multimodel approach uses available process information up to the current points to capture their time-varying relationships with the final product quality during the course of operation, so that the prognosis of product quality can be obtained in real-time. Then, due to its data-driven nature, the focus is given on how to make the models robust in order to eliminate the effect of noise, especially, outliers in the data. A model-based outlier detection method is presented. The proposed approach is applied to a generic chemical batch case study, with its prediction performance being evaluated.