Transition Phase Approach for Statistical Model in Multiphase Process

Abstract Online monitoring of batch reactor is essential to maintain the performance against batch-to-batch variations. A major challenge in using a statistical model for monitoring is to guarantee the data from different batches have similar dynamic characteristics at the same time points. By detecting singular points, uneven batch data are synchronized and the DPLS model of each singular episode is used for process monitoring. A new phase division method is proposed by introducing transition phase at a phase boundary to improve estimation performance around the transition region.