Statistical analysis and online monitoring for handling multiphase batch processes with varying durations

In the present work, statistical analysis and online monitoring is presented for handling uneven-length multiphase batch processes. Firstly, the irregular batches are classified into different uneven-length groups according to the changes of underlying characteristics. Then multi-source measurement data can be dealt with, each corresponding to one operation mode. The basic principle is that over different uneven-length groups, despite the uneven-length operation patterns, there are both similarity and dissimilarity to a certain extent among their underlying correlations. By an adequate decomposition, two different subspaces are separated, modeling the group-common and specific information respectively. Their corresponding confidence regions are constructed by searching similar patterns respectively. Accordingly, the online monitoring system is set up, which can track different types of variations closely. This analysis adds a detailed insight into the inherent nature of uneven-length multiphase batch processes. Its feasibility and performance are illustrated by a typical practical case with uneven cycles.

► We model and analyze uneven-length multiphase batch processes.
► We examine the changes of underlying characteristics of uneven-length processes and classify the irregular batches into different uneven-length groups.
► We separate the group-common and specific underlying correlations among different uneven-length groups.
► Close confidence intervals are enclosed to improve monitoring performance.
► During online application, the current phase and uneven-group affiliation can be correctly identified realtime.