Step-wise sequential phase partition (SSPP) algorithm based statistical modeling and online process monitoring

• We develop an automatic step-wise sequential phase partition (SSPP) algorithm for multiphase batch processes.
• We compare SSPP algorithm with clustering-based phase partition algorithm.
• The SSPP algorithm can overcome the problem of hard-partition and disorder in clustering-based division algorithm.
• SSPP based monitoring models provide better online fault detection performance than that of clustering-based monitoring models.

As batches operate at different statuses across different phases, it can be advantageous to partition the whole batch process into different phases and characterize them separately by multiple phase models. The conventional clustering-based division algorithm overlooks the time sequence of process phases and it is hard to capture the transition patterns between neighboring phases. In the present work, an automatic step-wise sequential phase partition (SSPP) algorithm is developed, which can capture the changes of process characteristics by checking their influences on monitoring system. Its theoretical support and the related statistical characteristics are analyzed. Using this algorithm, major phases are captured and also distinguished from transition patterns. The time-varying characteristics are thus described by different statistical models. For online application, the affiliation of each new sample can be realtime judged and its status can be checked by adopting the proper statistical model. Comparison is conducted between the proposed algorithm and clustering-based phase division algorithm. Comprehensive analyses are made regarding the influences of important parameters on monitoring performance. The proposed method is illustrated by a three-tank experimental system and an injection molding process which both present typical multiphase nature and transition characteristics.