Utilizing transition information in online quality prediction of multiphase batch processes

Abstract

In most previous researches, the data information of the transitions from one phase to another has been ignored, although it may be critical to product quality. In this paper, based on the cycle repeatability of batch processes, a repeatability factor is defined to divide the process into different steady phases and transition periods. Then an online quality prediction algorithm is proposed. In each steady phase where the process nature keeps similar, a representative phase model is developed; while in the transition period, a local modeling strategy is utilized for just-in-time modeling of each time slice, based on which only those batches having high similarities with the current batch are selected for modeling. Furthermore, a series of cumulative regression models are constructed for combining the results from different phases and transitions, which could further improve the prediction performance. Two examples are provided to demonstrate the feasibility and efficiency of the proposed method.